4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_handler.c
34 * Top-level entry points into osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
37 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
40 #define DEBUG_SUBSYSTEM S_OSD
42 #include <linux/module.h>
43 #include <linux/user_namespace.h>
44 #ifdef HAVE_UIDGID_HEADER
45 # include <linux/uidgid.h>
48 /* prerequisite for linux/xattr.h */
49 #include <linux/types.h>
50 /* prerequisite for linux/xattr.h */
52 /* XATTR_{REPLACE,CREATE} */
53 #include <linux/xattr.h>
55 #include <ldiskfs/ldiskfs.h>
56 #include <ldiskfs/xattr.h>
57 #include <ldiskfs/ldiskfs_extents.h>
60 * struct OBD_{ALLOC,FREE}*()
63 #include <obd_support.h>
64 /* struct ptlrpc_thread */
65 #include <lustre_net.h>
66 #include <lustre_fid.h>
68 #include <uapi/linux/lustre/lustre_param.h>
70 #include "osd_internal.h"
71 #include "osd_dynlocks.h"
73 /* llo_* api support */
74 #include <md_object.h>
75 #include <lustre_quota.h>
77 #include <lustre_linkea.h>
79 #define PFID_STRIPE_IDX_BITS 16
80 #define PFID_STRIPE_COUNT_MASK ((1 << PFID_STRIPE_IDX_BITS) - 1)
83 module_param(ldiskfs_pdo, int, 0644);
84 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
86 int ldiskfs_track_declares_assert;
87 module_param(ldiskfs_track_declares_assert, int, 0644);
88 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
90 /* Slab to allocate dynlocks */
91 struct kmem_cache *dynlock_cachep;
93 /* Slab to allocate osd_it_ea */
94 struct kmem_cache *osd_itea_cachep;
96 static struct lu_kmem_descr ldiskfs_caches[] = {
98 .ckd_cache = &dynlock_cachep,
99 .ckd_name = "dynlock_cache",
100 .ckd_size = sizeof(struct dynlock_handle)
103 .ckd_cache = &osd_itea_cachep,
104 .ckd_name = "osd_itea_cache",
105 .ckd_size = sizeof(struct osd_it_ea)
112 static const char dot[] = ".";
113 static const char dotdot[] = "..";
115 static const struct lu_object_operations osd_lu_obj_ops;
116 static const struct dt_object_operations osd_obj_ops;
117 static const struct dt_object_operations osd_obj_otable_it_ops;
118 static const struct dt_index_operations osd_index_iam_ops;
119 static const struct dt_index_operations osd_index_ea_ops;
121 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
122 const struct lu_fid *fid);
123 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
124 struct osd_device *osd);
126 int osd_trans_declare_op2rb[] = {
127 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
128 [OSD_OT_PUNCH] = OSD_OT_MAX,
129 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
130 [OSD_OT_CREATE] = OSD_OT_DESTROY,
131 [OSD_OT_DESTROY] = OSD_OT_CREATE,
132 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
133 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
134 [OSD_OT_WRITE] = OSD_OT_WRITE,
135 [OSD_OT_INSERT] = OSD_OT_DELETE,
136 [OSD_OT_DELETE] = OSD_OT_INSERT,
137 [OSD_OT_QUOTA] = OSD_OT_MAX,
140 static int osd_has_index(const struct osd_object *obj)
142 return obj->oo_dt.do_index_ops != NULL;
145 static int osd_object_invariant(const struct lu_object *l)
147 return osd_invariant(osd_obj(l));
151 * Concurrency: doesn't matter
153 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
155 struct osd_thread_info *oti = osd_oti_get(env);
156 return oti->oti_w_locks > 0 && o->oo_owner == env;
160 * Concurrency: doesn't access mutable data
162 static int osd_root_get(const struct lu_env *env,
163 struct dt_device *dev, struct lu_fid *f)
165 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
170 * the following set of functions are used to maintain per-thread
171 * cache of FID->ino mapping. this mechanism is needed to resolve
172 * FID to inode at dt_insert() which in turn stores ino in the
173 * directory entries to keep ldiskfs compatible with ext[34].
174 * due to locking-originated restrictions we can't lookup ino
175 * using LU cache (deadlock is possible). lookup using OI is quite
176 * expensive. so instead we maintain this cache and methods like
177 * dt_create() fill it. so in the majority of cases dt_insert() is
178 * able to find needed mapping in lockless manner.
180 static struct osd_idmap_cache *
181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
182 const struct lu_fid *fid)
184 struct osd_thread_info *oti = osd_oti_get(env);
185 struct osd_idmap_cache *idc = oti->oti_ins_cache;
187 for (i = 0; i < oti->oti_ins_cache_used; i++) {
188 if (!lu_fid_eq(&idc[i].oic_fid, fid))
190 if (idc[i].oic_dev != osd)
199 static struct osd_idmap_cache *
200 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
201 const struct lu_fid *fid)
203 struct osd_thread_info *oti = osd_oti_get(env);
204 struct osd_idmap_cache *idc;
207 if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
208 i = oti->oti_ins_cache_size * 2;
210 i = OSD_INS_CACHE_SIZE;
211 OBD_ALLOC(idc, sizeof(*idc) * i);
213 return ERR_PTR(-ENOMEM);
214 if (oti->oti_ins_cache != NULL) {
215 memcpy(idc, oti->oti_ins_cache,
216 oti->oti_ins_cache_used * sizeof(*idc));
217 OBD_FREE(oti->oti_ins_cache,
218 oti->oti_ins_cache_used * sizeof(*idc));
220 oti->oti_ins_cache = idc;
221 oti->oti_ins_cache_size = i;
224 idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
227 idc->oic_lid.oii_ino = 0;
228 idc->oic_lid.oii_gen = 0;
235 * lookup mapping for the given fid in the cache, initialize a
236 * new one if not found. the initialization checks whether the
237 * object is local or remote. for local objects, OI is used to
238 * learn ino/generation. the function is used when the caller
239 * has no information about the object, e.g. at dt_insert().
241 static struct osd_idmap_cache *
242 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
243 const struct lu_fid *fid)
245 struct osd_idmap_cache *idc;
248 idc = osd_idc_find(env, osd, fid);
249 LASSERT(!IS_ERR(idc));
253 /* new mapping is needed */
254 idc = osd_idc_add(env, osd, fid);
259 rc = osd_remote_fid(env, osd, fid);
260 if (unlikely(rc < 0))
264 /* the object is local, lookup in OI */
265 /* XXX: probably cheaper to lookup in LU first? */
266 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
268 if (unlikely(rc < 0)) {
269 CERROR("can't lookup: rc = %d\n", rc);
273 /* the object is remote */
281 * lookup mapping for given FID and fill it from the given object.
282 * the object is lolcal by definition.
284 static int osd_idc_find_and_init(const struct lu_env *env,
285 struct osd_device *osd,
286 struct osd_object *obj)
288 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
289 struct osd_idmap_cache *idc;
291 idc = osd_idc_find(env, osd, fid);
292 LASSERT(!IS_ERR(idc));
294 if (obj->oo_inode == NULL)
296 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
297 LASSERT(idc->oic_lid.oii_ino == 0);
298 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
299 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
304 /* new mapping is needed */
305 idc = osd_idc_add(env, osd, fid);
309 if (obj->oo_inode != NULL) {
310 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
311 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
317 * OSD object methods.
321 * Concurrency: no concurrent access is possible that early in object
324 static struct lu_object *osd_object_alloc(const struct lu_env *env,
325 const struct lu_object_header *hdr,
328 struct osd_object *mo;
334 l = &mo->oo_dt.do_lu;
335 dt_object_init(&mo->oo_dt, NULL, d);
336 mo->oo_dt.do_ops = &osd_obj_ops;
337 l->lo_ops = &osd_lu_obj_ops;
338 init_rwsem(&mo->oo_sem);
339 init_rwsem(&mo->oo_ext_idx_sem);
340 spin_lock_init(&mo->oo_guard);
341 INIT_LIST_HEAD(&mo->oo_xattr_list);
348 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
349 struct dentry *dentry, struct lustre_ost_attrs *loa)
353 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
354 (void *)loa, sizeof(*loa));
356 struct lustre_mdt_attrs *lma = &loa->loa_lma;
358 if (rc < sizeof(*lma))
362 lustre_loa_swab(loa, true);
363 /* Check LMA compatibility */
364 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
365 CWARN("%s: unsupported incompat LMA feature(s) %#x "
366 "for fid = "DFID", ino = %lu\n",
368 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
369 PFID(&lma->lma_self_fid), inode->i_ino);
372 } else if (rc == 0) {
380 * retrieve object from backend ext fs.
382 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
383 struct osd_inode_id *id)
386 struct inode *inode = NULL;
388 /* if we look for an inode withing a running
389 * transaction, then we risk to deadlock */
390 /* osd_dirent_check_repair() breaks this */
391 /*LASSERT(current->journal_info == NULL);*/
393 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
395 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
396 id->oii_ino, PTR_ERR(inode));
397 } else if (id->oii_gen != OSD_OII_NOGEN &&
398 inode->i_generation != id->oii_gen) {
399 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
400 "i_generation = %u\n",
401 id->oii_ino, id->oii_gen, inode->i_generation);
403 inode = ERR_PTR(-ESTALE);
404 } else if (inode->i_nlink == 0) {
405 /* due to parallel readdir and unlink,
406 * we can have dead inode here. */
407 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
409 inode = ERR_PTR(-ESTALE);
410 } else if (is_bad_inode(inode)) {
411 CWARN("%s: bad inode: ino = %u\n",
412 osd_dev2name(dev), id->oii_ino);
414 inode = ERR_PTR(-ENOENT);
415 } else if ((rc = osd_attach_jinode(inode))) {
419 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
420 if (id->oii_gen == OSD_OII_NOGEN)
421 osd_id_gen(id, inode->i_ino, inode->i_generation);
423 /* Do not update file c/mtime in ldiskfs.
424 * NB: we don't have any lock to protect this because we don't
425 * have reference on osd_object now, but contention with
426 * another lookup + attr_set can't happen in the tiny window
427 * between if (...) and set S_NOCMTIME. */
428 if (!(inode->i_flags & S_NOCMTIME))
429 inode->i_flags |= S_NOCMTIME;
434 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
435 handle_t *handle, struct dentry *child,
436 struct inode *inode, struct htree_lock *hlock)
440 rc = __ldiskfs_add_entry(handle, child, inode, hlock);
441 if (rc == -ENOBUFS || rc == -ENOSPC) {
442 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
443 struct inode *parent = child->d_parent->d_inode;
444 struct lu_fid *fid = NULL;
446 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
448 fid = &loa->loa_lma.lma_self_fid;
449 } else if (rc2 == -ENODATA) {
450 if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
451 fid = &info->oti_fid3;
452 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
453 } else if (!osd->od_is_ost && osd->od_index == 0) {
454 fid = &info->oti_fid3;
455 lu_igif_build(fid, parent->i_ino,
456 parent->i_generation);
461 CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
462 "maximum entry limit\n",
463 osd_name(osd), parent->i_ino, PFID(fid),
464 rc == -ENOSPC ? "has reached" : "is approaching");
466 CWARN("%s: directory (inode: %lu, FID: unknown) %s "
467 "maximum entry limit\n",
468 osd_name(osd), parent->i_ino,
469 rc == -ENOSPC ? "has reached" : "is approaching");
471 /* ignore such error now */
480 static struct inode *
481 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
482 struct osd_inode_id *id, struct lu_fid *fid)
484 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
488 inode = osd_iget(info, dev, id);
492 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
494 *fid = loa->loa_lma.lma_self_fid;
495 } else if (rc == -ENODATA) {
496 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
497 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
499 lu_igif_build(fid, inode->i_ino, inode->i_generation);
507 static struct inode *osd_iget_check(struct osd_thread_info *info,
508 struct osd_device *dev,
509 const struct lu_fid *fid,
510 struct osd_inode_id *id,
517 /* The cached OI mapping is trustable. If we cannot locate the inode
518 * via the cached OI mapping, then return the failure to the caller
519 * directly without further OI checking. */
522 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
525 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
528 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
529 PFID(fid), id->oii_ino, rc);
533 if (is_bad_inode(inode)) {
538 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
539 PFID(fid), id->oii_ino);
543 if (id->oii_gen != OSD_OII_NOGEN &&
544 inode->i_generation != id->oii_gen) {
549 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
550 "oii_gen = %u, i_generation = %u\n", PFID(fid),
551 id->oii_ino, id->oii_gen, inode->i_generation);
555 if (inode->i_nlink == 0) {
560 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
561 PFID(fid), id->oii_ino);
565 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
569 __u32 saved_ino = id->oii_ino;
570 __u32 saved_gen = id->oii_gen;
573 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
575 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
576 /* XXX: There are four possible cases:
578 * Backup/restore caused the OI invalid.
580 * Someone unlinked the object but NOT removed
581 * the OI mapping, such as mount target device
582 * as ldiskfs, and modify something directly.
584 * Someone just removed the object between the
585 * former oi_lookup and the iget. It is normal.
586 * 4. Other failure cases.
588 * Generally, when the device is mounted, it will
589 * auto check whether the system is restored from
590 * file-level backup or not. We trust such detect
591 * to distinguish the 1st case from the 2nd case:
592 * if the OI files are consistent but may contain
593 * stale OI mappings because of case 2, if iget()
594 * returns -ENOENT or -ESTALE, then it should be
597 /* If the OI mapping was in OI file before the
598 * osd_iget_check(), but now, it is disappear,
599 * then it must be removed by race. That is a
600 * normal race case. */
603 /* It is the OI scrub updated the OI mapping by race.
604 * The new OI mapping must be valid. */
605 if (saved_ino != id->oii_ino ||
606 (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
615 if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
616 /* It still can be the case 2, but we cannot
617 * distinguish it from the case 1. So return
618 * -EREMCHG to block current operation until
619 * OI scrub rebuilt the OI mappings. */
627 if (inode->i_generation == id->oii_gen)
632 if (id->oii_gen == OSD_OII_NOGEN)
633 osd_id_gen(id, inode->i_ino, inode->i_generation);
635 /* Do not update file c/mtime in ldiskfs.
636 * NB: we don't have any lock to protect this because we don't
637 * have reference on osd_object now, but contention with
638 * another lookup + attr_set can't happen in the tiny window
639 * between if (...) and set S_NOCMTIME. */
640 if (!(inode->i_flags & S_NOCMTIME))
641 inode->i_flags |= S_NOCMTIME;
658 * \retval +v: new filter_fid, does not contain self-fid
659 * \retval 0: filter_fid_old, contains self-fid
660 * \retval -v: other failure cases
662 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
663 struct dentry *dentry, struct lu_fid *fid)
665 struct filter_fid_old *ff = &info->oti_ff;
666 struct ost_id *ostid = &info->oti_ostid;
669 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
670 if (rc == sizeof(*ff)) {
672 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
673 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
675 * XXX: use 0 as the index for compatibility, the caller will
676 * handle index related issues when necessary.
679 ostid_to_fid(fid, ostid, 0);
680 } else if (rc == sizeof(struct filter_fid)) {
682 } else if (rc >= 0) {
689 static int osd_lma_self_repair(struct osd_thread_info *info,
690 struct osd_device *osd, struct inode *inode,
691 const struct lu_fid *fid, __u32 compat)
696 LASSERT(current->journal_info == NULL);
698 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
699 osd_dto_credits_noquota[DTO_XATTR_SET]);
702 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
707 rc = osd_ea_fid_set(info, inode, fid, compat, 0);
709 CWARN("%s: cannot self repair the LMA: rc = %d\n",
711 ldiskfs_journal_stop(jh);
715 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
717 struct osd_thread_info *info = osd_oti_get(env);
718 struct osd_device *osd = osd_obj2dev(obj);
719 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
720 struct lustre_mdt_attrs *lma = &loa->loa_lma;
721 struct inode *inode = obj->oo_inode;
722 struct dentry *dentry = &info->oti_obj_dentry;
723 struct lu_fid *fid = NULL;
724 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
728 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
729 (void *)loa, sizeof(*loa));
730 if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
731 fid = &lma->lma_self_fid;
732 rc = osd_get_idif(info, inode, dentry, fid);
733 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
734 /* For the given OST-object, if it has neither LMA nor
735 * FID in XATTR_NAME_FID, then the given FID (which is
736 * contained in the @obj, from client RPC for locating
737 * the OST-object) is trusted. We use it to generate
739 osd_lma_self_repair(info, osd, inode, rfid,
750 lustre_lma_swab(lma);
751 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
752 (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
753 S_ISREG(inode->i_mode)))) {
754 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
755 "fid = "DFID", ino = %lu\n", osd_name(osd),
756 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
757 PFID(rfid), inode->i_ino);
760 fid = &lma->lma_self_fid;
764 if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
765 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
766 struct ost_id *oi = &info->oti_ostid;
767 struct lu_fid *fid1 = &info->oti_fid3;
768 __u32 idx = fid_idif_ost_idx(rfid);
770 /* For old IDIF, the OST index is not part of the IDIF,
771 * Means that different OSTs may have the same IDIFs.
772 * Under such case, we need to make some compatible
773 * check to make sure to trigger OI scrub properly. */
774 if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
775 /* Given @rfid is new, LMA is old. */
776 fid_to_ostid(fid, oi);
777 ostid_to_fid(fid1, oi, idx);
778 if (lu_fid_eq(fid1, rfid)) {
779 if (osd->od_index_in_idif)
780 osd_lma_self_repair(info, osd,
794 struct osd_check_lmv_buf {
795 #ifdef HAVE_DIR_CONTEXT
796 /* please keep it as first member */
797 struct dir_context ctx;
799 struct osd_thread_info *oclb_info;
800 struct osd_device *oclb_dev;
801 struct osd_idmap_cache *oclb_oic;
805 * It is called internally by ->readdir() to filter out the
806 * local slave object's FID of the striped directory.
808 * \retval 1 found the local slave's FID
809 * \retval 0 continue to check next item
810 * \retval -ve for failure
812 #ifdef HAVE_FILLDIR_USE_CTX
813 static int osd_stripe_dir_filldir(struct dir_context *buf,
815 static int osd_stripe_dir_filldir(void *buf,
817 const char *name, int namelen,
818 loff_t offset, __u64 ino, unsigned d_type)
820 struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
821 struct osd_thread_info *oti = oclb->oclb_info;
822 struct lu_fid *fid = &oti->oti_fid3;
823 struct osd_inode_id *id = &oti->oti_id3;
824 struct osd_device *dev = oclb->oclb_dev;
825 struct osd_idmap_cache *oic = oclb->oclb_oic;
833 sscanf(name + 1, SFID, RFID(fid));
834 if (!fid_is_sane(fid))
837 if (osd_remote_fid(oti->oti_env, dev, fid))
840 osd_id_gen(id, ino, OSD_OII_NOGEN);
841 inode = osd_iget(oti, dev, id);
843 return PTR_ERR(inode);
846 osd_add_oi_cache(oti, dev, id, fid);
850 rc = osd_oii_insert(dev, oic, true);
852 return rc == 0 ? 1 : rc;
855 /* When lookup item under striped directory, we need to locate the master
856 * MDT-object of the striped directory firstly, then the client will send
857 * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
858 * and the item's name. If the system is restored from MDT file level backup,
859 * then before the OI scrub completely built the OI files, the OI mappings of
860 * the master MDT-object and slave MDT-object may be invalid. Usually, it is
861 * not a problem for the master MDT-object. Because when locate the master
862 * MDT-object, we will do name based lookup (for the striped directory itself)
863 * firstly, during such process we can setup the correct OI mapping for the
864 * master MDT-object. But it will be trouble for the slave MDT-object. Because
865 * the client will not trigger name based lookup on the MDT to locate the slave
866 * MDT-object before locating item under the striped directory, then when
867 * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
868 * is invalid and does not know what the right OI mapping is, then the MDT has
869 * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
870 * the OI file, related OI mapping is unknown yet, please try again later. And
871 * then client will re-try the RPC again and again until related OI mapping has
872 * been updated. That is quite inefficient.
874 * To resolve above trouble, we will handle it as the following two cases:
876 * 1) The slave MDT-object and the master MDT-object are on different MDTs.
877 * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
878 * is linked under /REMOTE_PARENT_DIR with the name of its FID string.
879 * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
880 * directly. Please check osd_fid_lookup().
882 * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
883 * Under such case, during lookup the master MDT-object, we will lookup the
884 * slave MDT-object via readdir against the master MDT-object, because the
885 * slave MDT-objects information are stored as sub-directories with the name
886 * "${FID}:${index}". Then when find the local slave MDT-object, its OI
887 * mapping will be recorded. Then subsequent osd_fid_lookup() will know
888 * the correct OI mapping for the slave MDT-object. */
889 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
890 struct inode *inode, struct osd_idmap_cache *oic)
892 struct lu_buf *buf = &oti->oti_big_buf;
893 struct dentry *dentry = &oti->oti_obj_dentry;
894 struct file *filp = &oti->oti_file;
895 const struct file_operations *fops;
896 struct lmv_mds_md_v1 *lmv1;
897 struct osd_check_lmv_buf oclb = {
898 #ifdef HAVE_DIR_CONTEXT
899 .ctx.actor = osd_stripe_dir_filldir,
909 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
912 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
914 lu_buf_realloc(buf, rc);
915 if (buf->lb_buf == NULL)
916 GOTO(out, rc = -ENOMEM);
922 if (unlikely(rc == 0 || rc == -ENODATA))
928 if (unlikely(buf->lb_buf == NULL)) {
929 lu_buf_realloc(buf, rc);
930 if (buf->lb_buf == NULL)
931 GOTO(out, rc = -ENOMEM);
937 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
941 dentry->d_inode = inode;
942 dentry->d_sb = inode->i_sb;
944 filp->f_path.dentry = dentry;
945 filp->f_mode = FMODE_64BITHASH;
946 filp->f_mapping = inode->i_mapping;
948 filp->private_data = NULL;
949 set_file_inode(filp, inode);
951 #ifdef HAVE_DIR_CONTEXT
952 oclb.ctx.pos = filp->f_pos;
953 rc = fops->iterate(filp, &oclb.ctx);
954 filp->f_pos = oclb.ctx.pos;
956 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
958 fops->release(inode, filp);
962 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
963 DFID": rc = %d\n", osd_ino2name(inode),
964 inode->i_ino, inode->i_generation,
965 PFID(&oic->oic_fid), rc);
972 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
973 const struct lu_fid *fid,
974 const struct lu_object_conf *conf)
976 struct osd_thread_info *info;
977 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
978 struct osd_device *dev;
979 struct osd_idmap_cache *oic;
980 struct osd_inode_id *id;
981 struct inode *inode = NULL;
982 struct osd_scrub *scrub;
983 struct scrub_file *sf;
984 __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
991 bool updated = false;
994 LINVRNT(osd_invariant(obj));
995 LASSERT(obj->oo_inode == NULL);
996 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
999 scrub = &dev->od_scrub;
1000 sf = &scrub->os_file;
1001 info = osd_oti_get(env);
1003 oic = &info->oti_cache;
1005 if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1008 /* For the object is created as locking anchor, or for the object to
1009 * be created on disk. No need to osd_oi_lookup() at here because FID
1010 * shouldn't never be re-used, if it's really a duplicate FID from
1011 * unexpected reason, we should be able to detect it later by calling
1012 * do_create->osd_oi_insert(). */
1013 if (conf && conf->loc_flags & LOC_F_NEW)
1014 GOTO(out, result = 0);
1016 /* Search order: 1. per-thread cache. */
1017 if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1023 if (!list_empty(&scrub->os_inconsistent_items)) {
1024 /* Search order: 2. OI scrub pending list. */
1025 result = osd_oii_lookup(dev, fid, id);
1030 /* The OI mapping in the OI file can be updated by the OI scrub
1031 * when we locate the inode via FID. So it may be not trustable. */
1034 /* Search order: 3. OI files. */
1035 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1036 if (result == -ENOENT) {
1037 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1038 fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1039 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1041 GOTO(out, result = 0);
1050 obj->oo_inode = NULL;
1051 /* for later passes through checks, not true on first pass */
1052 if (!IS_ERR_OR_NULL(inode))
1055 inode = osd_iget_check(info, dev, fid, id, trusted);
1056 if (!IS_ERR(inode)) {
1057 obj->oo_inode = inode;
1065 result = PTR_ERR(inode);
1066 if (result == -ENOENT || result == -ESTALE)
1067 GOTO(out, result = 0);
1069 if (result != -EREMCHG)
1073 /* We still have chance to get the valid inode: for the
1074 * object which is referenced by remote name entry, the
1075 * object on the local MDT will be linked under the dir
1076 * of "/REMOTE_PARENT_DIR" with its FID string as name.
1078 * We do not know whether the object for the given FID
1079 * is referenced by some remote name entry or not, and
1080 * especially for DNE II, a multiple-linked object may
1081 * have many name entries reside on many MDTs.
1083 * To simplify the operation, OSD will not distinguish
1084 * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1085 * only happened for the RPC from other MDT during the
1086 * OI scrub, or for the client side RPC with FID only,
1087 * such as FID to path, or from old connected client. */
1088 if (!remote && !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1089 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1093 flags |= SS_AUTO_PARTIAL;
1094 flags &= ~SS_AUTO_FULL;
1099 if (thread_is_running(&scrub->os_thread)) {
1100 if (scrub->os_partial_scan && !scrub->os_in_join)
1103 if (IS_ERR_OR_NULL(inode) || result)
1104 GOTO(out, result = -EINPROGRESS);
1107 LASSERT(obj->oo_inode == inode);
1109 osd_add_oi_cache(info, dev, id, fid);
1110 osd_oii_insert(dev, oic, true);
1114 if (dev->od_noscrub) {
1116 GOTO(out, result = -EREMCHG);
1119 LASSERT(obj->oo_inode == inode);
1121 osd_add_oi_cache(info, dev, id, fid);
1126 rc1 = osd_scrub_start(dev, flags);
1127 LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1128 "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1129 if (rc1 && rc1 != -EALREADY)
1130 GOTO(out, result = -EREMCHG);
1132 if (IS_ERR_OR_NULL(inode) || result)
1133 GOTO(out, result = -EINPROGRESS);
1136 LASSERT(obj->oo_inode == inode);
1138 osd_add_oi_cache(info, dev, id, fid);
1139 osd_oii_insert(dev, oic, true);
1143 result = osd_check_lma(env, obj);
1147 LASSERTF(id->oii_ino == inode->i_ino &&
1148 id->oii_gen == inode->i_generation,
1149 "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1150 PFID(fid), id->oii_ino, id->oii_gen,
1151 inode->i_ino, inode->i_generation);
1153 saved_ino = inode->i_ino;
1154 saved_gen = inode->i_generation;
1156 if (unlikely(result == -ENODATA)) {
1157 /* If the OI scrub updated the OI mapping by race, it
1158 * must be valid. Trust the inode that has no LMA EA. */
1162 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1164 /* The OI mapping is still there, the inode is still
1165 * valid. It is just becaues the inode has no LMA EA. */
1166 if (saved_ino == id->oii_ino &&
1167 saved_gen == id->oii_gen)
1170 /* It is the OI scrub updated the OI mapping by race.
1171 * The new OI mapping must be valid. */
1177 /* "result == -ENOENT" means that the OI mappinghas been
1178 * removed by race, so the inode belongs to other object.
1180 * Others error can be returned directly. */
1181 if (result == -ENOENT) {
1184 obj->oo_inode = NULL;
1189 if (result != -EREMCHG)
1194 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1195 /* "result == -ENOENT" means the cached OI mapping has been removed
1196 * from the OI file by race, above inode belongs to other object. */
1197 if (result == -ENOENT) {
1200 obj->oo_inode = NULL;
1201 GOTO(out, result = 0);
1207 if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1212 /* It is the OI scrub updated the OI mapping by race.
1213 * The new OI mapping must be valid. */
1219 obj->oo_compat_dot_created = 1;
1220 obj->oo_compat_dotdot_created = 1;
1222 if (S_ISDIR(inode->i_mode) &&
1223 (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1224 osd_check_lmv(info, dev, inode, oic);
1226 result = osd_attach_jinode(inode);
1231 GOTO(out, result = 0);
1233 LASSERT(!obj->oo_hl_head);
1234 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1236 GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1239 if (result || !obj->oo_inode) {
1240 if (!IS_ERR_OR_NULL(inode))
1243 obj->oo_inode = NULL;
1245 fid_zero(&oic->oic_fid);
1248 LINVRNT(osd_invariant(obj));
1253 * Concurrency: shouldn't matter.
1255 static void osd_object_init0(struct osd_object *obj)
1257 LASSERT(obj->oo_inode != NULL);
1258 obj->oo_dt.do_body_ops = &osd_body_ops;
1259 obj->oo_dt.do_lu.lo_header->loh_attr |=
1260 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1264 * Concurrency: no concurrent access is possible that early in object
1267 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1268 const struct lu_object_conf *conf)
1270 struct osd_object *obj = osd_obj(l);
1273 LINVRNT(osd_invariant(obj));
1275 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1276 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1277 l->lo_header->loh_attr |= LOHA_EXISTS;
1281 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1282 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1283 if (result == 0 && obj->oo_inode != NULL) {
1284 struct osd_thread_info *oti = osd_oti_get(env);
1285 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1287 osd_object_init0(obj);
1288 result = osd_get_lma(oti, obj->oo_inode,
1289 &oti->oti_obj_dentry, loa);
1291 /* Convert LMAI flags to lustre LMA flags
1292 * and cache it to oo_lma_flags */
1294 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1295 } else if (result == -ENODATA) {
1300 LINVRNT(osd_invariant(obj));
1304 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1305 * The left part is for value, binary mode. */
1306 struct osd_xattr_entry {
1307 struct list_head oxe_list;
1311 struct rcu_head oxe_rcu;
1315 static int osd_oxc_get(struct osd_object *obj, const char *name,
1318 struct osd_xattr_entry *tmp;
1319 struct osd_xattr_entry *oxe = NULL;
1320 size_t namelen = strlen(name);
1325 list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1326 if (namelen == tmp->oxe_namelen &&
1327 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1334 GOTO(out, rc = -ENOENT);
1336 if (!oxe->oxe_exist)
1337 GOTO(out, rc = -ENODATA);
1340 rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1343 if (buf->lb_buf == NULL)
1346 if (buf->lb_len < rc)
1347 GOTO(out, rc = -ERANGE);
1349 memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1357 static void osd_oxc_free(struct rcu_head *head)
1359 struct osd_xattr_entry *oxe;
1361 oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1362 OBD_FREE(oxe, oxe->oxe_len);
1365 static void osd_oxc_add(struct osd_object *obj, const char *name,
1366 const char *buf, int buflen)
1368 struct osd_xattr_entry *oxe;
1369 struct osd_xattr_entry *old = NULL;
1370 struct osd_xattr_entry *tmp;
1371 size_t namelen = strlen(name);
1372 size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1374 OBD_ALLOC(oxe, len);
1378 INIT_LIST_HEAD(&oxe->oxe_list);
1380 oxe->oxe_namelen = namelen;
1381 memcpy(oxe->oxe_buf, name, namelen);
1383 LASSERT(buf != NULL);
1384 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1385 oxe->oxe_exist = true;
1387 oxe->oxe_exist = false;
1390 /* this should be rarely called, just remove old and add new */
1391 spin_lock(&obj->oo_guard);
1392 list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1393 if (namelen == tmp->oxe_namelen &&
1394 strncmp(name, tmp->oxe_buf, namelen) == 0) {
1400 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1401 call_rcu(&old->oxe_rcu, osd_oxc_free);
1403 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1405 spin_unlock(&obj->oo_guard);
1408 static void osd_oxc_del(struct osd_object *obj, const char *name)
1410 struct osd_xattr_entry *oxe;
1411 size_t namelen = strlen(name);
1413 spin_lock(&obj->oo_guard);
1414 list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1415 if (namelen == oxe->oxe_namelen &&
1416 strncmp(name, oxe->oxe_buf, namelen) == 0) {
1417 list_del_rcu(&oxe->oxe_list);
1418 call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1422 spin_unlock(&obj->oo_guard);
1425 static void osd_oxc_fini(struct osd_object *obj)
1427 struct osd_xattr_entry *oxe, *next;
1429 list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1430 list_del(&oxe->oxe_list);
1431 OBD_FREE(oxe, oxe->oxe_len);
1436 * Concurrency: no concurrent access is possible that late in object
1439 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1441 struct osd_object *obj = osd_obj(l);
1443 LINVRNT(osd_invariant(obj));
1446 dt_object_fini(&obj->oo_dt);
1447 if (obj->oo_hl_head != NULL)
1448 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1453 * Concurrency: no concurrent access is possible that late in object
1456 static void osd_index_fini(struct osd_object *o)
1458 struct iam_container *bag;
1460 if (o->oo_dir != NULL) {
1461 bag = &o->oo_dir->od_container;
1462 if (o->oo_inode != NULL) {
1463 if (bag->ic_object == o->oo_inode)
1464 iam_container_fini(bag);
1466 OBD_FREE_PTR(o->oo_dir);
1472 * Concurrency: no concurrent access is possible that late in object
1473 * life-cycle (for all existing callers, that is. New callers have to provide
1474 * their own locking.)
1476 static int osd_inode_unlinked(const struct inode *inode)
1478 return inode->i_nlink == 0;
1482 OSD_TXN_OI_DELETE_CREDITS = 20,
1483 OSD_TXN_INODE_DELETE_CREDITS = 20
1490 #if OSD_THANDLE_STATS
1492 * Set time when the handle is allocated
1494 static void osd_th_alloced(struct osd_thandle *oth)
1496 oth->oth_alloced = ktime_get();
1500 * Set time when the handle started
1502 static void osd_th_started(struct osd_thandle *oth)
1504 oth->oth_started = ktime_get();
1508 * Check whether the we deal with this handle for too long.
1510 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1511 ktime_t alloced, ktime_t started,
1514 ktime_t now = ktime_get();
1516 LASSERT(dev != NULL);
1518 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1519 ktime_us_delta(started, alloced));
1520 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1521 ktime_us_delta(closed, started));
1522 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1523 ktime_us_delta(now, closed));
1525 if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1526 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1527 oth, now, alloced, started, closed);
1528 libcfs_debug_dumpstack(NULL);
1532 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
1534 ktime_t __closed = ktime_get(); \
1535 ktime_t __alloced = oth->oth_alloced; \
1536 ktime_t __started = oth->oth_started; \
1539 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
1542 #else /* OSD_THANDLE_STATS */
1544 #define osd_th_alloced(h) do {} while(0)
1545 #define osd_th_started(h) do {} while(0)
1546 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
1548 #endif /* OSD_THANDLE_STATS */
1551 * Concurrency: doesn't access mutable data.
1553 static int osd_param_is_not_sane(const struct osd_device *dev,
1554 const struct thandle *th)
1556 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1558 return oh->ot_credits > osd_transaction_size(dev);
1562 * Concurrency: shouldn't matter.
1564 static void osd_trans_commit_cb(struct super_block *sb,
1565 struct ldiskfs_journal_cb_entry *jcb, int error)
1567 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1568 struct thandle *th = &oh->ot_super;
1569 struct lu_device *lud = &th->th_dev->dd_lu_dev;
1570 struct dt_txn_commit_cb *dcb, *tmp;
1572 LASSERT(oh->ot_handle == NULL);
1575 CERROR("transaction @0x%p commit error: %d\n", th, error);
1577 dt_txn_hook_commit(th);
1579 /* call per-transaction callbacks if any */
1580 list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1582 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1583 "commit callback entry: magic=%x name='%s'\n",
1584 dcb->dcb_magic, dcb->dcb_name);
1585 list_del_init(&dcb->dcb_linkage);
1586 dcb->dcb_func(NULL, th, dcb, error);
1589 lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1596 #ifndef HAVE_SB_START_WRITE
1597 # define sb_start_write(sb) do {} while (0)
1598 # define sb_end_write(sb) do {} while (0)
1601 static struct thandle *osd_trans_create(const struct lu_env *env,
1602 struct dt_device *d)
1604 struct osd_thread_info *oti = osd_oti_get(env);
1605 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1606 struct osd_thandle *oh;
1611 CERROR("%s: someone try to start transaction under "
1612 "readonly mode, should be disabled.\n",
1613 osd_name(osd_dt_dev(d)));
1615 RETURN(ERR_PTR(-EROFS));
1618 /* on pending IO in this thread should left from prev. request */
1619 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1621 sb_start_write(osd_sb(osd_dt_dev(d)));
1623 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1625 oh->ot_quota_trans = &oti->oti_quota_trans;
1626 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1631 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1632 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1635 memset(oti->oti_declare_ops, 0,
1636 sizeof(oti->oti_declare_ops));
1637 memset(oti->oti_declare_ops_cred, 0,
1638 sizeof(oti->oti_declare_ops_cred));
1639 memset(oti->oti_declare_ops_used, 0,
1640 sizeof(oti->oti_declare_ops_used));
1642 sb_end_write(osd_sb(osd_dt_dev(d)));
1643 th = ERR_PTR(-ENOMEM);
1648 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1650 struct osd_thread_info *oti = osd_oti_get(env);
1651 struct osd_thandle *oh;
1653 oh = container_of0(th, struct osd_thandle, ot_super);
1654 LASSERT(oh != NULL);
1656 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1657 oti->oti_declare_ops[OSD_OT_CREATE],
1658 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1659 oti->oti_declare_ops_used[OSD_OT_CREATE],
1660 oti->oti_declare_ops[OSD_OT_DESTROY],
1661 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1662 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1663 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1664 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1665 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1666 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1667 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1668 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1669 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1670 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1671 oti->oti_declare_ops[OSD_OT_WRITE],
1672 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1673 oti->oti_declare_ops_used[OSD_OT_WRITE],
1674 oti->oti_declare_ops[OSD_OT_PUNCH],
1675 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1676 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1677 oti->oti_declare_ops[OSD_OT_QUOTA],
1678 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1679 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1680 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1681 oti->oti_declare_ops[OSD_OT_INSERT],
1682 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1683 oti->oti_declare_ops_used[OSD_OT_INSERT],
1684 oti->oti_declare_ops[OSD_OT_DELETE],
1685 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1686 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1687 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1688 oti->oti_declare_ops[OSD_OT_REF_ADD],
1689 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1690 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1691 oti->oti_declare_ops[OSD_OT_REF_DEL],
1692 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1693 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1697 * Concurrency: shouldn't matter.
1699 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1702 struct osd_thread_info *oti = osd_oti_get(env);
1703 struct osd_device *dev = osd_dt_dev(d);
1705 struct osd_thandle *oh;
1710 LASSERT(current->journal_info == NULL);
1712 oh = container_of0(th, struct osd_thandle, ot_super);
1713 LASSERT(oh != NULL);
1714 LASSERT(oh->ot_handle == NULL);
1716 rc = dt_txn_hook_start(env, d, th);
1720 if (unlikely(osd_param_is_not_sane(dev, th))) {
1721 static unsigned long last_printed;
1722 static int last_credits;
1724 /* don't make noise on a tiny testing systems
1725 * actual credits misuse will be caught anyway */
1726 if (last_credits != oh->ot_credits &&
1727 time_after(jiffies, last_printed +
1728 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1729 osd_transaction_size(dev) > 512) {
1730 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1731 oh->ot_credits, osd_transaction_size(dev));
1732 osd_trans_dump_creds(env, th);
1733 libcfs_debug_dumpstack(NULL);
1734 last_credits = oh->ot_credits;
1735 last_printed = jiffies;
1737 /* XXX Limit the credits to 'max_transaction_buffers', and
1738 * let the underlying filesystem to catch the error if
1739 * we really need so many credits.
1741 * This should be removed when we can calculate the
1742 * credits precisely. */
1743 oh->ot_credits = osd_transaction_size(dev);
1744 } else if (ldiskfs_track_declares_assert != 0) {
1745 /* reserve few credits to prevent an assertion in JBD
1746 * our debugging mechanism will be able to detected
1747 * overuse. this can help to debug single-update
1749 oh->ot_credits += 10;
1750 if (unlikely(osd_param_is_not_sane(dev, th)))
1751 oh->ot_credits = osd_transaction_size(dev);
1755 * XXX temporary stuff. Some abstraction layer should
1758 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1762 LASSERT(oti->oti_txns == 0);
1764 lu_device_get(&d->dd_lu_dev);
1765 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1776 static int osd_seq_exists(const struct lu_env *env,
1777 struct osd_device *osd, u64 seq)
1779 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1780 struct seq_server_site *ss = osd_seq_site(osd);
1784 LASSERT(ss != NULL);
1785 LASSERT(ss->ss_server_fld != NULL);
1787 rc = osd_fld_lookup(env, osd, seq, range);
1790 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1791 osd_name(osd), seq, rc);
1795 RETURN(ss->ss_node_id == range->lsr_index);
1798 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1800 struct dt_txn_commit_cb *dcb;
1801 struct dt_txn_commit_cb *tmp;
1803 /* call per-transaction stop callbacks if any */
1804 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1806 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1807 "commit callback entry: magic=%x name='%s'\n",
1808 dcb->dcb_magic, dcb->dcb_name);
1809 list_del_init(&dcb->dcb_linkage);
1810 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1815 * Concurrency: shouldn't matter.
1817 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1820 int rc = 0, remove_agents = 0;
1821 struct osd_thandle *oh;
1822 struct osd_thread_info *oti = osd_oti_get(env);
1823 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1824 struct osd_device *osd = osd_dt_dev(th->th_dev);
1825 struct qsd_instance *qsd = osd->od_quota_slave;
1826 struct lquota_trans *qtrans;
1829 oh = container_of0(th, struct osd_thandle, ot_super);
1831 /* reset OI cache for safety */
1832 oti->oti_ins_cache_used = 0;
1834 remove_agents = oh->ot_remove_agents;
1836 qtrans = oh->ot_quota_trans;
1837 oh->ot_quota_trans = NULL;
1839 if (oh->ot_handle != NULL) {
1841 handle_t *hdl = oh->ot_handle;
1844 * add commit callback
1845 * notice we don't do this in osd_trans_start()
1846 * as underlying transaction can change during truncate
1848 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1851 LASSERT(oti->oti_txns == 1);
1854 rc = dt_txn_hook_stop(env, th);
1856 CERROR("%s: failed in transaction hook: rc = %d\n",
1859 osd_trans_stop_cb(oh, rc);
1860 /* hook functions might modify th_sync */
1861 hdl->h_sync = th->th_sync;
1863 oh->ot_handle = NULL;
1864 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1866 CERROR("%s: failed to stop transaction: rc = %d\n",
1867 osd_name(osd), rc2);
1871 osd_trans_stop_cb(oh, th->th_result);
1875 /* inform the quota slave device that the transaction is stopping */
1876 qsd_op_end(env, qsd, qtrans);
1878 /* as we want IO to journal and data IO be concurrent, we don't block
1879 * awaiting data IO completion in osd_do_bio(), instead we wait here
1880 * once transaction is submitted to the journal. all reqular requests
1881 * don't do direct IO (except read/write), thus this wait_event becomes
1884 * IMPORTANT: we have to wait till any IO submited by the thread is
1885 * completed otherwise iobuf may be corrupted by different request
1887 wait_event(iobuf->dr_wait,
1888 atomic_read(&iobuf->dr_numreqs) == 0);
1889 osd_fini_iobuf(osd, iobuf);
1891 rc = iobuf->dr_error;
1893 if (unlikely(remove_agents != 0))
1894 osd_process_scheduled_agent_removals(env, osd);
1896 sb_end_write(osd_sb(osd));
1901 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1903 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1906 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1907 LASSERT(&dcb->dcb_func != NULL);
1908 if (dcb->dcb_flags & DCB_TRANS_STOP)
1909 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1911 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1917 * Called just before object is freed. Releases all resources except for
1918 * object itself (that is released by osd_object_free()).
1920 * Concurrency: no concurrent access is possible that late in object
1923 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1925 struct osd_object *obj = osd_obj(l);
1926 struct inode *inode = obj->oo_inode;
1928 LINVRNT(osd_invariant(obj));
1931 * If object is unlinked remove fid->ino mapping from object index.
1934 osd_index_fini(obj);
1935 if (inode != NULL) {
1936 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1937 qid_t uid = i_uid_read(inode);
1938 qid_t gid = i_gid_read(inode);
1941 obj->oo_inode = NULL;
1944 struct osd_thread_info *info = osd_oti_get(env);
1945 struct lquota_id_info *qi = &info->oti_qi;
1947 /* Release granted quota to master if necessary */
1948 qi->lqi_id.qid_uid = uid;
1949 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1951 qi->lqi_id.qid_uid = gid;
1952 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1954 qi->lqi_id.qid_uid = i_projid_read(inode);
1955 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1961 * Concurrency: ->loo_object_release() is called under site spin-lock.
1963 static void osd_object_release(const struct lu_env *env,
1964 struct lu_object *l)
1966 struct osd_object *o = osd_obj(l);
1967 /* nobody should be releasing a non-destroyed object with nlink=0
1968 * the API allows this, but ldiskfs doesn't like and then report
1969 * this inode as deleted */
1970 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1975 * Concurrency: shouldn't matter.
1977 static int osd_object_print(const struct lu_env *env, void *cookie,
1978 lu_printer_t p, const struct lu_object *l)
1980 struct osd_object *o = osd_obj(l);
1981 struct iam_descr *d;
1983 if (o->oo_dir != NULL)
1984 d = o->oo_dir->od_container.ic_descr;
1987 return (*p)(env, cookie,
1988 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1990 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1991 o->oo_inode ? o->oo_inode->i_generation : 0,
1992 d ? d->id_ops->id_name : "plain");
1996 * Concurrency: shouldn't matter.
1998 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1999 struct obd_statfs *sfs)
2001 struct osd_device *osd = osd_dt_dev(d);
2002 struct super_block *sb = osd_sb(osd);
2003 struct kstatfs *ksfs;
2007 if (unlikely(osd->od_mnt == NULL))
2008 return -EINPROGRESS;
2010 /* osd_lproc.c call this without env, allocate ksfs for that case */
2011 if (unlikely(env == NULL)) {
2012 OBD_ALLOC_PTR(ksfs);
2016 ksfs = &osd_oti_get(env)->oti_ksfs;
2019 result = sb->s_op->statfs(sb->s_root, ksfs);
2023 statfs_pack(sfs, ksfs);
2024 if (unlikely(sb->s_flags & MS_RDONLY))
2025 sfs->os_state |= OS_STATE_READONLY;
2026 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2027 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2028 sfs->os_maxbytes = sb->s_maxbytes;
2030 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2033 * Reserve some space so to avoid fragmenting the filesystem too much.
2034 * Fragmentation not only impacts performance, but can also increase
2035 * metadata overhead significantly, causing grant calculation to be
2038 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2040 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2041 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2042 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2043 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2045 sfs->os_blocks -= reserved;
2046 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2047 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2050 if (unlikely(env == NULL))
2056 * Estimate space needed for file creations. We assume the largest filename
2057 * which is 2^64 - 1, hence a filename of 20 chars.
2058 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2060 #ifdef __LDISKFS_DIR_REC_LEN
2061 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2063 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2067 * Concurrency: doesn't access mutable data.
2069 static void osd_conf_get(const struct lu_env *env,
2070 const struct dt_device *dev,
2071 struct dt_device_param *param)
2073 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2077 * XXX should be taken from not-yet-existing fs abstraction layer.
2079 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2080 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2081 param->ddp_symlink_max = sb->s_blocksize;
2082 param->ddp_mount_type = LDD_MT_LDISKFS;
2083 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2084 param->ddp_maxbytes = sb->s_maxbytes;
2086 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2087 /* inode are statically allocated, so per-inode space consumption
2088 * is the space consumed by the directory entry */
2089 param->ddp_inodespace = PER_OBJ_USAGE;
2090 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2091 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2092 * maximum length to not under count the actual number of extents
2093 * needed for writing a file. */
2094 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2095 /* worst-case extent insertion metadata overhead */
2096 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2097 param->ddp_mntopts = 0;
2098 if (test_opt(sb, XATTR_USER))
2099 param->ddp_mntopts |= MNTOPT_USERXATTR;
2100 if (test_opt(sb, POSIX_ACL))
2101 param->ddp_mntopts |= MNTOPT_ACL;
2103 /* LOD might calculate the max stripe count based on max_ea_size,
2104 * so we need take account in the overhead as well,
2105 * xattr_header + magic + xattr_entry_head */
2106 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2107 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2109 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2110 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2111 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2115 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2119 * Concurrency: shouldn't matter.
2121 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2125 CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2127 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2129 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2136 * Start commit for OSD device.
2138 * An implementation of dt_commit_async method for OSD device.
2139 * Asychronously starts underlayng fs sync and thereby a transaction
2142 * \param env environment
2143 * \param d dt device
2145 * \see dt_device_operations
2147 static int osd_commit_async(const struct lu_env *env,
2148 struct dt_device *d)
2150 struct super_block *s = osd_sb(osd_dt_dev(d));
2153 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2154 RETURN(s->s_op->sync_fs(s, 0));
2157 /* Our own copy of the set readonly functions if present, or NU if not. */
2158 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2159 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2160 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2163 * Concurrency: shouldn't matter.
2165 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2167 struct super_block *sb = osd_sb(osd_dt_dev(d));
2168 struct block_device *dev = sb->s_bdev;
2169 int rc = -EOPNOTSUPP;
2172 if (priv_dev_set_rdonly) {
2173 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2176 CERROR("*** setting %s read-only ***\n",
2177 osd_dt_dev(d)->od_svname);
2179 if (sb->s_op->freeze_fs) {
2180 rc = sb->s_op->freeze_fs(sb);
2185 if (jdev && (jdev != dev)) {
2186 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2188 priv_dev_set_rdonly(jdev);
2190 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2191 priv_dev_set_rdonly(dev);
2193 if (sb->s_op->unfreeze_fs)
2194 sb->s_op->unfreeze_fs(sb);
2199 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2200 osd_dt_dev(d)->od_svname, (long)dev, rc);
2206 * Note: we do not count into QUOTA here.
2207 * If we mount with --data_journal we may need more.
2209 const int osd_dto_credits_noquota[DTO_NR] = {
2212 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2213 * SINGLEDATA_TRANS_BLOCKS(8)
2214 * XXX Note: maybe iam need more, since iam have more level than
2217 [DTO_INDEX_INSERT] = 16,
2220 * just modify a single entry, probably merge few within a block
2222 [DTO_INDEX_DELETE] = 1,
2226 [DTO_INDEX_UPDATE] = 16,
2228 * 4(inode, inode bits, groups, GDT)
2229 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2231 [DTO_OBJECT_CREATE] = 4,
2233 * 4(inode, inode bits, groups, GDT)
2234 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2236 [DTO_OBJECT_DELETE] = 4,
2238 * Attr set credits (inode)
2240 [DTO_ATTR_SET_BASE] = 1,
2242 * Xattr set. The same as xattr of EXT3.
2243 * DATA_TRANS_BLOCKS(14)
2244 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2245 * are also counted in. Do not know why?
2247 [DTO_XATTR_SET] = 14,
2249 * credits for inode change during write.
2251 [DTO_WRITE_BASE] = 3,
2253 * credits for single block write.
2255 [DTO_WRITE_BLOCK] = 14,
2257 * Attr set credits for chown.
2258 * This is extra credits for setattr, and it is null without quota
2260 [DTO_ATTR_SET_CHOWN] = 0
2263 static const struct dt_device_operations osd_dt_ops = {
2264 .dt_root_get = osd_root_get,
2265 .dt_statfs = osd_statfs,
2266 .dt_trans_create = osd_trans_create,
2267 .dt_trans_start = osd_trans_start,
2268 .dt_trans_stop = osd_trans_stop,
2269 .dt_trans_cb_add = osd_trans_cb_add,
2270 .dt_conf_get = osd_conf_get,
2271 .dt_sync = osd_sync,
2273 .dt_commit_async = osd_commit_async,
2276 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2279 struct osd_object *obj = osd_dt_obj(dt);
2280 struct osd_thread_info *oti = osd_oti_get(env);
2282 LINVRNT(osd_invariant(obj));
2284 LASSERT(obj->oo_owner != env);
2285 down_read_nested(&obj->oo_sem, role);
2287 LASSERT(obj->oo_owner == NULL);
2291 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2294 struct osd_object *obj = osd_dt_obj(dt);
2295 struct osd_thread_info *oti = osd_oti_get(env);
2297 LINVRNT(osd_invariant(obj));
2299 LASSERT(obj->oo_owner != env);
2300 down_write_nested(&obj->oo_sem, role);
2302 LASSERT(obj->oo_owner == NULL);
2303 obj->oo_owner = env;
2307 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2309 struct osd_object *obj = osd_dt_obj(dt);
2310 struct osd_thread_info *oti = osd_oti_get(env);
2312 LINVRNT(osd_invariant(obj));
2314 LASSERT(oti->oti_r_locks > 0);
2316 up_read(&obj->oo_sem);
2319 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2321 struct osd_object *obj = osd_dt_obj(dt);
2322 struct osd_thread_info *oti = osd_oti_get(env);
2324 LINVRNT(osd_invariant(obj));
2326 LASSERT(obj->oo_owner == env);
2327 LASSERT(oti->oti_w_locks > 0);
2329 obj->oo_owner = NULL;
2330 up_write(&obj->oo_sem);
2333 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2335 struct osd_object *obj = osd_dt_obj(dt);
2337 LINVRNT(osd_invariant(obj));
2339 return obj->oo_owner == env;
2342 static struct timespec *osd_inode_time(const struct lu_env *env,
2343 struct inode *inode, __u64 seconds)
2345 struct osd_thread_info *oti = osd_oti_get(env);
2346 struct timespec *t = &oti->oti_time;
2348 t->tv_sec = seconds;
2350 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2354 static void osd_inode_getattr(const struct lu_env *env,
2355 struct inode *inode, struct lu_attr *attr)
2357 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2358 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2359 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2360 LA_BLKSIZE | LA_TYPE;
2362 attr->la_atime = LTIME_S(inode->i_atime);
2363 attr->la_mtime = LTIME_S(inode->i_mtime);
2364 attr->la_ctime = LTIME_S(inode->i_ctime);
2365 attr->la_mode = inode->i_mode;
2366 attr->la_size = i_size_read(inode);
2367 attr->la_blocks = inode->i_blocks;
2368 attr->la_uid = i_uid_read(inode);
2369 attr->la_gid = i_gid_read(inode);
2370 attr->la_projid = i_projid_read(inode);
2371 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2372 attr->la_nlink = inode->i_nlink;
2373 attr->la_rdev = inode->i_rdev;
2374 attr->la_blksize = 1 << inode->i_blkbits;
2375 attr->la_blkbits = inode->i_blkbits;
2377 * Ext4 did not transfer inherit flags from raw inode
2378 * to inode flags, and ext4 internally test raw inode
2379 * @i_flags directly. Instead of patching ext4, we do it here.
2381 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2382 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2385 static int osd_attr_get(const struct lu_env *env,
2386 struct dt_object *dt,
2387 struct lu_attr *attr)
2389 struct osd_object *obj = osd_dt_obj(dt);
2391 if (unlikely(!dt_object_exists(dt)))
2393 if (unlikely(obj->oo_destroyed))
2396 LASSERT(!dt_object_remote(dt));
2397 LINVRNT(osd_invariant(obj));
2399 spin_lock(&obj->oo_guard);
2400 osd_inode_getattr(env, obj->oo_inode, attr);
2401 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2402 attr->la_flags |= LUSTRE_ORPHAN_FL;
2403 spin_unlock(&obj->oo_guard);
2408 static int osd_declare_attr_qid(const struct lu_env *env,
2409 struct osd_object *obj,
2410 struct osd_thandle *oh, long long bspace,
2411 qid_t old_id, qid_t new_id, bool enforce,
2415 struct osd_thread_info *info = osd_oti_get(env);
2416 struct lquota_id_info *qi = &info->oti_qi;
2418 qi->lqi_type = type;
2419 /* inode accounting */
2420 qi->lqi_is_blk = false;
2422 /* one more inode for the new id ... */
2423 qi->lqi_id.qid_uid = new_id;
2425 /* Reserve credits for the new id */
2426 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2427 if (rc == -EDQUOT || rc == -EINPROGRESS)
2432 /* and one less inode for the current id */
2433 qi->lqi_id.qid_uid = old_id;
2435 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2436 if (rc == -EDQUOT || rc == -EINPROGRESS)
2441 /* block accounting */
2442 qi->lqi_is_blk = true;
2444 /* more blocks for the new id ... */
2445 qi->lqi_id.qid_uid = new_id;
2446 qi->lqi_space = bspace;
2448 * Credits for the new uid has been reserved, re-use "obj"
2449 * to save credit reservation.
2451 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2452 if (rc == -EDQUOT || rc == -EINPROGRESS)
2457 /* and finally less blocks for the current uid */
2458 qi->lqi_id.qid_uid = old_id;
2459 qi->lqi_space = -bspace;
2460 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2461 if (rc == -EDQUOT || rc == -EINPROGRESS)
2467 static int osd_declare_attr_set(const struct lu_env *env,
2468 struct dt_object *dt,
2469 const struct lu_attr *attr,
2470 struct thandle *handle)
2472 struct osd_thandle *oh;
2473 struct osd_object *obj;
2481 LASSERT(dt != NULL);
2482 LASSERT(handle != NULL);
2484 obj = osd_dt_obj(dt);
2485 LASSERT(osd_invariant(obj));
2487 oh = container_of0(handle, struct osd_thandle, ot_super);
2488 LASSERT(oh->ot_handle == NULL);
2490 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2491 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2493 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2494 osd_dto_credits_noquota[DTO_XATTR_SET]);
2496 if (attr == NULL || obj->oo_inode == NULL)
2499 bspace = obj->oo_inode->i_blocks;
2500 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2501 bspace = toqb(bspace);
2503 /* Changing ownership is always preformed by super user, it should not
2506 * We still need to call the osd_declare_qid() to calculate the journal
2507 * credits for updating quota accounting files and to trigger quota
2508 * space adjustment once the operation is completed.*/
2509 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2511 uid = i_uid_read(obj->oo_inode);
2512 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2513 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2514 attr->la_uid, enforce, USRQUOTA);
2518 gid = i_gid_read(obj->oo_inode);
2519 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2520 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2521 i_gid_read(obj->oo_inode),
2522 attr->la_gid, enforce, GRPQUOTA);
2527 #ifdef HAVE_PROJECT_QUOTA
2528 if (attr->la_valid & LA_PROJID) {
2529 __u32 projid = i_projid_read(obj->oo_inode);
2530 enforce = (attr->la_valid & LA_PROJID) &&
2531 (attr->la_projid != projid);
2532 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2533 (qid_t)projid, (qid_t)attr->la_projid,
2542 static int osd_inode_setattr(const struct lu_env *env,
2543 struct inode *inode, const struct lu_attr *attr)
2545 __u64 bits = attr->la_valid;
2547 /* Only allow set size for regular file */
2548 if (!S_ISREG(inode->i_mode))
2549 bits &= ~(LA_SIZE | LA_BLOCKS);
2554 if (bits & LA_ATIME)
2555 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2556 if (bits & LA_CTIME)
2557 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2558 if (bits & LA_MTIME)
2559 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2560 if (bits & LA_SIZE) {
2561 spin_lock(&inode->i_lock);
2562 LDISKFS_I(inode)->i_disksize = attr->la_size;
2563 i_size_write(inode, attr->la_size);
2564 spin_unlock(&inode->i_lock);
2567 /* OSD should not change "i_blocks" which is used by quota.
2568 * "i_blocks" should be changed by ldiskfs only. */
2570 inode->i_mode = (inode->i_mode & S_IFMT) |
2571 (attr->la_mode & ~S_IFMT);
2573 i_uid_write(inode, attr->la_uid);
2575 i_gid_write(inode, attr->la_gid);
2576 if (bits & LA_PROJID)
2577 i_projid_write(inode, attr->la_projid);
2578 if (bits & LA_NLINK)
2579 set_nlink(inode, attr->la_nlink);
2581 inode->i_rdev = attr->la_rdev;
2583 if (bits & LA_FLAGS) {
2584 /* always keep S_NOCMTIME */
2585 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2588 * Ext4 did not transfer inherit flags from
2589 * @inode->i_flags to raw inode i_flags when writing
2590 * flags, we do it explictly here.
2592 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2593 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2595 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2600 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2604 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2605 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2608 ll_vfs_dq_init(inode);
2610 if (attr->la_valid & LA_UID)
2611 iattr.ia_valid |= ATTR_UID;
2612 if (attr->la_valid & LA_GID)
2613 iattr.ia_valid |= ATTR_GID;
2614 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2615 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2617 rc = ll_vfs_dq_transfer(inode, &iattr);
2619 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2620 "enforcement enabled on the ldiskfs "
2621 "filesystem?\n", inode->i_sb->s_id, rc);
2626 /* Handle project id transfer here properly */
2627 if (attr->la_valid & LA_PROJID &&
2628 attr->la_projid != i_projid_read(inode)) {
2629 #ifdef HAVE_PROJECT_QUOTA
2630 rc = ldiskfs_transfer_project(inode, attr->la_projid);
2635 CERROR("%s: quota transfer failed: rc = %d. Is project "
2636 "enforcement enabled on the ldiskfs "
2637 "filesystem?\n", inode->i_sb->s_id, rc);
2644 static int osd_attr_set(const struct lu_env *env,
2645 struct dt_object *dt,
2646 const struct lu_attr *attr,
2647 struct thandle *handle)
2649 struct osd_object *obj = osd_dt_obj(dt);
2650 struct inode *inode;
2653 if (!dt_object_exists(dt))
2656 LASSERT(handle != NULL);
2657 LASSERT(!dt_object_remote(dt));
2658 LASSERT(osd_invariant(obj));
2660 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2662 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2663 struct osd_thread_info *oti = osd_oti_get(env);
2664 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2665 struct lu_fid *fid1 = &oti->oti_fid;
2666 struct osd_inode_id *id = &oti->oti_id;
2667 struct iam_path_descr *ipd;
2668 struct iam_container *bag;
2669 struct osd_thandle *oh;
2672 fid_cpu_to_be(fid1, fid0);
2673 memset(id, 1, sizeof(*id));
2674 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2675 fid0)->oi_dir.od_container;
2676 ipd = osd_idx_ipd_get(env, bag);
2677 if (unlikely(ipd == NULL))
2680 oh = container_of0(handle, struct osd_thandle, ot_super);
2681 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2682 (const struct iam_rec *)id, ipd);
2683 osd_ipd_put(env, bag, ipd);
2684 return(rc > 0 ? 0 : rc);
2687 inode = obj->oo_inode;
2689 rc = osd_quota_transfer(inode, attr);
2693 spin_lock(&obj->oo_guard);
2694 rc = osd_inode_setattr(env, inode, attr);
2695 spin_unlock(&obj->oo_guard);
2699 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2701 if (!(attr->la_valid & LA_FLAGS))
2704 /* Let's check if there are extra flags need to be set into LMA */
2705 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2706 struct osd_thread_info *info = osd_oti_get(env);
2707 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2709 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2710 &info->oti_ost_attrs);
2714 lma->lma_incompat |=
2715 lustre_to_lma_flags(attr->la_flags);
2716 lustre_lma_swab(lma);
2717 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2718 lma, sizeof(*lma), XATTR_REPLACE);
2720 struct osd_device *osd = osd_obj2dev(obj);
2722 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2723 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2724 lma->lma_incompat, rc);
2727 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2729 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2732 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2737 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2738 struct osd_object *obj,
2739 const char *name, const int namelen)
2741 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2744 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2745 umode_t mode, struct dt_allocation_hint *hint,
2749 struct osd_device *osd = osd_obj2dev(obj);
2750 struct osd_thandle *oth;
2751 struct dt_object *parent = NULL;
2752 struct inode *inode;
2754 LINVRNT(osd_invariant(obj));
2755 LASSERT(obj->oo_inode == NULL);
2756 LASSERT(obj->oo_hl_head == NULL);
2758 if (S_ISDIR(mode) && ldiskfs_pdo) {
2759 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2760 if (obj->oo_hl_head == NULL)
2764 oth = container_of(th, struct osd_thandle, ot_super);
2765 LASSERT(oth->ot_handle->h_transaction != NULL);
2767 if (hint != NULL && hint->dah_parent != NULL &&
2768 !dt_object_remote(hint->dah_parent))
2769 parent = hint->dah_parent;
2771 inode = ldiskfs_create_inode(oth->ot_handle,
2772 parent ? osd_dt_obj(parent)->oo_inode :
2773 osd_sb(osd)->s_root->d_inode,
2775 if (!IS_ERR(inode)) {
2776 /* Do not update file c/mtime in ldiskfs. */
2777 inode->i_flags |= S_NOCMTIME;
2779 /* For new created object, it must be consistent,
2780 * and it is unnecessary to scrub against it. */
2781 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2783 obj->oo_inode = inode;
2786 if (obj->oo_hl_head != NULL) {
2787 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2788 obj->oo_hl_head = NULL;
2790 result = PTR_ERR(inode);
2792 LINVRNT(osd_invariant(obj));
2800 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2801 struct lu_attr *attr,
2802 struct dt_allocation_hint *hint,
2803 struct dt_object_format *dof,
2807 struct osd_thandle *oth;
2808 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2810 LASSERT(S_ISDIR(attr->la_mode));
2812 oth = container_of(th, struct osd_thandle, ot_super);
2813 LASSERT(oth->ot_handle->h_transaction != NULL);
2814 result = osd_mkfile(info, obj, mode, hint, th);
2819 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2820 struct lu_attr *attr,
2821 struct dt_allocation_hint *hint,
2822 struct dt_object_format *dof,
2826 struct osd_thandle *oth;
2827 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2829 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2831 LASSERT(S_ISREG(attr->la_mode));
2833 oth = container_of(th, struct osd_thandle, ot_super);
2834 LASSERT(oth->ot_handle->h_transaction != NULL);
2836 result = osd_mkfile(info, obj, mode, hint, th);
2838 LASSERT(obj->oo_inode != NULL);
2839 if (feat->dif_flags & DT_IND_VARKEY)
2840 result = iam_lvar_create(obj->oo_inode,
2841 feat->dif_keysize_max,
2843 feat->dif_recsize_max,
2846 result = iam_lfix_create(obj->oo_inode,
2847 feat->dif_keysize_max,
2849 feat->dif_recsize_max,
2856 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2857 struct lu_attr *attr,
2858 struct dt_allocation_hint *hint,
2859 struct dt_object_format *dof,
2862 LASSERT(S_ISREG(attr->la_mode));
2863 return osd_mkfile(info, obj, (attr->la_mode &
2864 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2867 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2868 struct lu_attr *attr,
2869 struct dt_allocation_hint *hint,
2870 struct dt_object_format *dof,
2873 LASSERT(S_ISLNK(attr->la_mode));
2874 return osd_mkfile(info, obj, (attr->la_mode &
2875 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2878 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2879 struct lu_attr *attr,
2880 struct dt_allocation_hint *hint,
2881 struct dt_object_format *dof,
2884 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2887 LINVRNT(osd_invariant(obj));
2888 LASSERT(obj->oo_inode == NULL);
2889 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2890 S_ISFIFO(mode) || S_ISSOCK(mode));
2892 result = osd_mkfile(info, obj, mode, hint, th);
2894 LASSERT(obj->oo_inode != NULL);
2896 * This inode should be marked dirty for i_rdev. Currently
2897 * that is done in the osd_attr_init().
2899 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2902 LINVRNT(osd_invariant(obj));
2906 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2908 struct dt_allocation_hint *hint,
2909 struct dt_object_format *dof,
2912 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2914 osd_obj_type_f result;
2930 result = osd_mk_index;
2941 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2942 struct dt_object *parent, struct dt_object *child,
2947 ah->dah_parent = parent;
2948 ah->dah_mode = child_mode;
2950 if (parent != NULL && !dt_object_remote(parent)) {
2951 /* will help to find FID->ino at dt_insert("..") */
2952 struct osd_object *pobj = osd_dt_obj(parent);
2953 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2957 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2958 struct lu_attr *attr, struct dt_object_format *dof)
2960 struct inode *inode = obj->oo_inode;
2961 __u64 valid = attr->la_valid;
2964 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2966 if (dof->dof_type != DFT_NODE)
2967 attr->la_valid &= ~LA_RDEV;
2968 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2969 attr->la_valid &= ~LA_ATIME;
2970 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2971 attr->la_valid &= ~LA_CTIME;
2972 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2973 attr->la_valid &= ~LA_MTIME;
2975 result = osd_quota_transfer(inode, attr);
2979 if (attr->la_valid != 0) {
2980 result = osd_inode_setattr(info->oti_env, inode, attr);
2982 * The osd_inode_setattr() should always succeed here. The
2983 * only error that could be returned is EDQUOT when we are
2984 * trying to change the UID or GID of the inode. However, this
2985 * should not happen since quota enforcement is no longer
2986 * enabled on ldiskfs (lquota takes care of it).
2988 LASSERTF(result == 0, "%d\n", result);
2989 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2992 attr->la_valid = valid;
2996 * Helper function for osd_create()
2998 * \retval 0, on success
3000 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3001 struct lu_attr *attr, struct dt_allocation_hint *hint,
3002 struct dt_object_format *dof, struct thandle *th)
3007 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3009 /* we drop umask so that permissions we pass are not affected */
3010 umask = current->fs->umask;
3011 current->fs->umask = 0;
3013 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3015 if (likely(obj->oo_inode != NULL)) {
3016 LASSERT(obj->oo_inode->i_state & I_NEW);
3018 /* Unlock the inode before attr initialization to avoid
3019 * unnecessary dqget operations. LU-6378 */
3020 unlock_new_inode(obj->oo_inode);
3023 if (likely(result == 0)) {
3024 osd_attr_init(info, obj, attr, dof);
3025 osd_object_init0(obj);
3028 /* restore previous umask value */
3029 current->fs->umask = umask;
3031 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3037 * Helper function for osd_create()
3039 * \retval 0, on success
3041 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3042 const struct lu_fid *fid, struct thandle *th)
3044 struct osd_thread_info *info = osd_oti_get(env);
3045 struct osd_inode_id *id = &info->oti_id;
3046 struct osd_device *osd = osd_obj2dev(obj);
3047 struct osd_thandle *oh;
3050 LASSERT(obj->oo_inode != NULL);
3052 oh = container_of0(th, struct osd_thandle, ot_super);
3053 LASSERT(oh->ot_handle);
3054 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3056 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3057 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3058 OI_CHECK_FLD, NULL);
3059 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3064 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3065 u64 seq, struct lu_seq_range *range)
3067 struct seq_server_site *ss = osd_seq_site(osd);
3069 if (fid_seq_is_idif(seq)) {
3070 fld_range_set_ost(range);
3071 range->lsr_index = idif_ost_idx(seq);
3075 if (!fid_seq_in_fldb(seq)) {
3076 fld_range_set_mdt(range);
3078 /* FIXME: If ss is NULL, it suppose not get lsr_index
3080 range->lsr_index = ss->ss_node_id;
3084 LASSERT(ss != NULL);
3085 fld_range_set_any(range);
3086 /* OSD will only do local fld lookup */
3087 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3090 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3091 struct lu_attr *attr,
3092 struct dt_allocation_hint *hint,
3093 struct dt_object_format *dof,
3094 struct thandle *handle)
3096 struct osd_thandle *oh;
3100 LASSERT(handle != NULL);
3102 oh = container_of0(handle, struct osd_thandle, ot_super);
3103 LASSERT(oh->ot_handle == NULL);
3105 /* EA object consumes more credits than regular object: osd_mk_index
3106 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3107 * leaf_node, could involves the block, block bitmap, groups, GDT
3108 * change for each block, so add 4 * 2 credits in that case. */
3109 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3110 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3111 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3112 /* Reuse idle OI block may cause additional one OI block
3114 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3115 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3120 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3121 attr->la_projid, 1, oh, osd_dt_obj(dt),
3122 NULL, OSD_QID_INODE);
3126 /* will help to find FID->ino mapping at dt_insert() */
3127 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3134 * Called to destroy on-disk representation of the object
3136 * Concurrency: must be locked
3138 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3141 struct osd_object *obj = osd_dt_obj(dt);
3142 struct inode *inode = obj->oo_inode;
3143 struct osd_thandle *oh;
3150 oh = container_of0(th, struct osd_thandle, ot_super);
3151 LASSERT(oh->ot_handle == NULL);
3153 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3154 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3155 /* Recycle idle OI leaf may cause additional three OI blocks
3157 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3158 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3159 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3160 /* one less inode */
3161 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3162 i_projid_read(inode), -1, oh, obj, NULL,
3166 /* data to be truncated */
3167 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3168 i_projid_read(inode), 0, oh, obj, NULL,
3173 /* will help to find FID->ino when this object is being
3174 * added to PENDING/ */
3175 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3180 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3183 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3184 struct osd_object *obj = osd_dt_obj(dt);
3185 struct inode *inode = obj->oo_inode;
3186 struct osd_device *osd = osd_obj2dev(obj);
3187 struct osd_thandle *oh;
3191 oh = container_of0(th, struct osd_thandle, ot_super);
3192 LASSERT(oh->ot_handle);
3194 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3196 if (unlikely(fid_is_acct(fid)))
3199 if (S_ISDIR(inode->i_mode)) {
3200 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3201 inode->i_nlink == 2);
3202 /* it will check/delete the inode from remote parent,
3203 * how to optimize it? unlink performance impaction XXX */
3204 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3206 CERROR("%s: delete inode "DFID": rc = %d\n",
3207 osd_name(osd), PFID(fid), result);
3209 spin_lock(&obj->oo_guard);
3211 spin_unlock(&obj->oo_guard);
3212 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3215 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3217 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3219 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3220 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3221 oh->ot_handle, OI_CHECK_FLD);
3223 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3224 /* XXX: add to ext3 orphan list */
3225 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3227 /* not needed in the cache anymore */
3228 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3229 obj->oo_destroyed = 1;
3235 * Put the fid into lustre_mdt_attrs, and then place the structure
3236 * inode's ea. This fid should not be altered during the life time
3239 * \retval +ve, on success
3240 * \retval -ve, on error
3242 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3244 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3245 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3247 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3248 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3252 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3255 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3258 lustre_loa_init(loa, fid, compat, incompat);
3259 lustre_loa_swab(loa, false);
3261 /* For the OST device with 256 bytes inode size by default,
3262 * the PFID EA will be stored together with LMA EA to avoid
3263 * performance trouble. Otherwise the PFID EA can be stored
3264 * independently. LU-8998 */
3265 if ((compat & LMAC_FID_ON_OST) &&
3266 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3267 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3268 sizeof(*loa), XATTR_CREATE);
3270 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3271 sizeof(*lma), XATTR_CREATE);
3272 /* LMA may already exist, but we need to check that all the
3273 * desired compat/incompat flags have been added. */
3274 if (unlikely(rc == -EEXIST)) {
3275 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3276 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3280 if (rc < sizeof(*lma))
3283 lustre_loa_swab(loa, true);
3284 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3285 ((compat == 0 && incompat == 0) ||
3286 (!(~lma->lma_compat & compat) &&
3287 !(~lma->lma_incompat & incompat))))
3290 lma->lma_self_fid = *fid;
3291 lma->lma_compat |= compat;
3292 lma->lma_incompat |= incompat;
3293 if (rc == sizeof(*lma)) {
3294 lustre_lma_swab(lma);
3295 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3296 sizeof(*lma), XATTR_REPLACE);
3298 lustre_loa_swab(loa, false);
3299 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3300 sizeof(*loa), XATTR_REPLACE);
3308 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3309 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3310 * To have compatilibility with 1.8 ldiskfs driver we need to have
3311 * magic number at start of fid data.
3312 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3315 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3316 const struct lu_fid *fid)
3318 if (!fid_is_namespace_visible(fid) ||
3319 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3320 param->edp_magic = 0;
3324 param->edp_magic = LDISKFS_LUFID_MAGIC;
3325 param->edp_len = sizeof(struct lu_fid) + 1;
3326 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3330 * Try to read the fid from inode ea into dt_rec.
3332 * \param fid object fid.
3334 * \retval 0 on success
3336 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3337 __u32 ino, struct lu_fid *fid,
3338 struct osd_inode_id *id)
3340 struct osd_thread_info *info = osd_oti_get(env);
3341 struct inode *inode;
3344 osd_id_gen(id, ino, OSD_OII_NOGEN);
3345 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3347 RETURN(PTR_ERR(inode));
3353 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3355 struct inode *parent_dir,
3356 const struct lu_fid *dot_fid,
3357 const struct lu_fid *dot_dot_fid,
3358 struct osd_thandle *oth)
3360 struct ldiskfs_dentry_param *dot_ldp;
3361 struct ldiskfs_dentry_param *dot_dot_ldp;
3362 __u32 saved_nlink = dir->i_nlink;
3365 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3366 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3368 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3369 dot_ldp->edp_magic = 0;
3371 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3372 dir, dot_ldp, dot_dot_ldp);
3373 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3374 * the subseqent ref_add() will increase the dir->i_nlink
3375 * as 3. That is incorrect for new created directory.
3377 * It looks like hack, because we want to make the OSD API
3378 * to be order-independent for new created directory object
3379 * between dt_insert(..) and ref_add() operations.
3381 * Here, we only restore the in-RAM dir-inode's nlink attr,
3382 * becuase if the nlink attr is not 2, then there will be
3383 * ref_add() called following the dt_insert(..), such call
3384 * will make both the in-RAM and on-disk dir-inode's nlink
3385 * attr to be set as 2. LU-7447 */
3386 set_nlink(dir, saved_nlink);
3391 * Create an local agent inode for remote entry
3393 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3394 struct osd_device *osd,
3395 struct osd_object *pobj,
3396 const struct lu_fid *fid,
3400 struct osd_thread_info *info = osd_oti_get(env);
3401 struct inode *local;
3402 struct osd_thandle *oh;
3407 oh = container_of(th, struct osd_thandle, ot_super);
3408 LASSERT(oh->ot_handle->h_transaction != NULL);
3410 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3411 if (IS_ERR(local)) {
3412 CERROR("%s: create local error %d\n", osd_name(osd),
3413 (int)PTR_ERR(local));
3417 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3418 * correct gid on remote file, not agent here */
3419 local->i_gid = current_fsgid();
3420 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3421 unlock_new_inode(local);
3423 /* Agent inode should not have project ID */
3424 #ifdef HAVE_PROJECT_QUOTA
3425 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3426 i_projid_read(pobj->oo_inode) != 0) {
3427 rc = ldiskfs_transfer_project(local, 0);
3429 CERROR("%s: quota transfer failed: rc = %d. Is project "
3430 "quota enforcement enabled on the ldiskfs "
3431 "filesystem?\n", local->i_sb->s_id, rc);
3432 RETURN(ERR_PTR(rc));
3436 /* Set special LMA flag for local agent inode */
3437 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3439 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3440 osd_name(osd), PFID(fid), rc);
3441 RETURN(ERR_PTR(rc));
3447 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3448 lu_object_fid(&pobj->oo_dt.do_lu),
3451 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3452 osd_name(osd), PFID(fid), rc);
3453 RETURN(ERR_PTR(rc));
3460 * when direntry is deleted, we have to take care of possible agent inode
3461 * referenced by that. unfortunately we can't do this at that point:
3462 * iget() within a running transaction leads to deadlock and we better do
3463 * not call that every delete declaration to save performance. so we put
3464 * a potention agent inode on a list and process that once the transaction
3465 * is over. Notice it's not any worse in terms of real orphans as regular
3466 * object destroy doesn't put inodes on the on-disk orphan list. this should
3467 * be addressed separately
3469 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3470 struct osd_thandle *oh,
3473 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3474 struct osd_obj_orphan *oor;
3481 oor->oor_env = (struct lu_env *)env;
3482 spin_lock(&osd->od_osfs_lock);
3483 list_add(&oor->oor_list, &osd->od_orphan_list);
3484 spin_unlock(&osd->od_osfs_lock);
3486 oh->ot_remove_agents = 1;
3492 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3493 struct osd_device *osd)
3495 struct osd_thread_info *info = osd_oti_get(env);
3496 struct osd_obj_orphan *oor, *tmp;
3497 struct osd_inode_id id;
3498 struct list_head list;
3499 struct inode *inode;
3504 INIT_LIST_HEAD(&list);
3506 spin_lock(&osd->od_osfs_lock);
3507 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3508 if (oor->oor_env == env) {
3509 list_del(&oor->oor_list);
3510 list_add(&oor->oor_list, &list);
3513 spin_unlock(&osd->od_osfs_lock);
3515 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3519 list_del(&oor->oor_list);
3522 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3523 inode = osd_iget_fid(info, osd, &id, &fid);
3527 if (!osd_remote_fid(env, osd, &fid)) {
3532 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3534 mark_inode_dirty(inode);
3535 ldiskfs_journal_stop(jh);
3543 * OSD layer object create function for OST objects (b=11826).
3545 * The FID is inserted into inode xattr here.
3547 * \retval 0, on success
3548 * \retval -ve, on error
3550 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3551 struct lu_attr *attr, struct dt_allocation_hint *hint,
3552 struct dt_object_format *dof, struct thandle *th)
3554 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3555 struct osd_object *obj = osd_dt_obj(dt);
3556 struct osd_thread_info *info = osd_oti_get(env);
3557 int result, on_ost = 0;
3561 if (dt_object_exists(dt))
3564 LINVRNT(osd_invariant(obj));
3565 LASSERT(!dt_object_remote(dt));
3566 LASSERT(osd_is_write_locked(env, obj));
3567 LASSERT(th != NULL);
3569 if (unlikely(fid_is_acct(fid)))
3570 /* Quota files can't be created from the kernel any more,
3571 * 'tune2fs -O quota' will take care of creating them */
3574 result = __osd_create(info, obj, attr, hint, dof, th);
3576 if (fid_is_idif(fid) &&
3577 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3578 struct lu_fid *tfid = &info->oti_fid;
3579 struct ost_id *oi = &info->oti_ostid;
3581 fid_to_ostid(fid, oi);
3582 ostid_to_fid(tfid, oi, 0);
3584 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3585 LMAC_FID_ON_OST, 0);
3587 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3589 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3590 on_ost ? LMAC_FID_ON_OST : 0,
3593 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3594 obj->oo_dt.do_body_ops = &osd_body_ops;
3598 result = __osd_oi_insert(env, obj, fid, th);
3600 /* a small optimization - dt_insert() isn't usually applied
3601 * to OST objects, so we don't need to cache OI mapping for
3603 if (result == 0 && on_ost == 0) {
3604 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3605 result = osd_idc_find_and_init(env, osd, obj);
3606 LASSERT(result == 0);
3609 LASSERT(ergo(result == 0,
3610 dt_object_exists(dt) && !dt_object_remote(dt)));
3611 LINVRNT(osd_invariant(obj));
3615 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3616 struct thandle *handle)
3618 struct osd_thandle *oh;
3620 /* it's possible that object doesn't exist yet */
3621 LASSERT(handle != NULL);
3623 oh = container_of0(handle, struct osd_thandle, ot_super);
3624 LASSERT(oh->ot_handle == NULL);
3626 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3627 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3633 * Concurrency: @dt is write locked.
3635 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3638 struct osd_object *obj = osd_dt_obj(dt);
3639 struct inode *inode = obj->oo_inode;
3640 struct osd_thandle *oh;
3643 if (!dt_object_exists(dt) || obj->oo_destroyed)
3646 LINVRNT(osd_invariant(obj));
3647 LASSERT(!dt_object_remote(dt));
3648 LASSERT(osd_is_write_locked(env, obj));
3649 LASSERT(th != NULL);
3651 oh = container_of0(th, struct osd_thandle, ot_super);
3652 LASSERT(oh->ot_handle != NULL);
3654 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3656 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3657 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3659 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3660 * (65000) subdirectories by storing "1" in i_nlink if the link count
3661 * would otherwise overflow. Directory tranversal tools understand
3662 * that (st_nlink == 1) indicates that the filesystem dose not track
3663 * hard links count on the directory, and will not abort subdirectory
3664 * scanning early once (st_nlink - 2) subdirs have been found.
3666 * This also has to properly handle the case of inodes with nlink == 0
3667 * in case they are being linked into the PENDING directory
3669 spin_lock(&obj->oo_guard);
3670 if (unlikely(inode->i_nlink == 0))
3671 /* inc_nlink from 0 may cause WARN_ON */
3672 set_nlink(inode, 1);
3674 ldiskfs_inc_count(oh->ot_handle, inode);
3675 if (!S_ISDIR(inode->i_mode))
3676 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3678 spin_unlock(&obj->oo_guard);
3680 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3681 LINVRNT(osd_invariant(obj));
3683 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3688 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3689 struct thandle *handle)
3691 struct osd_thandle *oh;
3693 if (!dt_object_exists(dt))
3696 LASSERT(!dt_object_remote(dt));
3697 LASSERT(handle != NULL);
3699 oh = container_of0(handle, struct osd_thandle, ot_super);
3700 LASSERT(oh->ot_handle == NULL);
3702 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3703 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3709 * Concurrency: @dt is write locked.
3711 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3714 struct osd_object *obj = osd_dt_obj(dt);
3715 struct inode *inode = obj->oo_inode;
3716 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3717 struct osd_thandle *oh;
3719 if (!dt_object_exists(dt))
3722 LINVRNT(osd_invariant(obj));
3723 LASSERT(!dt_object_remote(dt));
3724 LASSERT(osd_is_write_locked(env, obj));
3725 LASSERT(th != NULL);
3727 oh = container_of0(th, struct osd_thandle, ot_super);
3728 LASSERT(oh->ot_handle != NULL);
3730 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3732 spin_lock(&obj->oo_guard);
3733 /* That can be result of upgrade from old Lustre version and
3734 * applied only to local files. Just skip this ref_del call.
3735 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3736 if (inode->i_nlink == 0) {
3737 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3738 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3739 ", maybe an upgraded file? (LU-3915)\n",
3740 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3741 spin_unlock(&obj->oo_guard);
3745 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3746 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3748 ldiskfs_dec_count(oh->ot_handle, inode);
3749 spin_unlock(&obj->oo_guard);
3751 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3752 LINVRNT(osd_invariant(obj));
3754 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3760 * Concurrency: @dt is read locked.
3762 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3763 struct lu_buf *buf, const char *name)
3765 struct osd_object *obj = osd_dt_obj(dt);
3766 struct inode *inode = obj->oo_inode;
3767 struct osd_thread_info *info = osd_oti_get(env);
3768 struct dentry *dentry = &info->oti_obj_dentry;
3769 bool cache_xattr = false;
3774 /* version get is not real XATTR but uses xattr API */
3775 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3776 dt_obj_version_t *ver = buf->lb_buf;
3778 /* for version we are just using xattr API but change inode
3780 if (buf->lb_len == 0)
3781 return sizeof(dt_obj_version_t);
3783 if (buf->lb_len < sizeof(dt_obj_version_t))
3786 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3787 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3789 *ver = LDISKFS_I(inode)->i_fs_version;
3791 return sizeof(dt_obj_version_t);
3794 if (!dt_object_exists(dt))
3797 LASSERT(!dt_object_remote(dt));
3798 LASSERT(inode->i_op != NULL);
3799 LASSERT(inode->i_op->getxattr != NULL);
3801 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3802 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3806 rc = osd_oxc_get(obj, name, buf);
3811 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3812 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3813 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3814 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3815 struct filter_fid *ff;
3816 struct ost_layout *ol;
3818 if (!osd_dev(dt->do_lu.lo_dev)->od_is_ost)
3821 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3825 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3831 if (buf->lb_len == 0 || !buf->lb_buf)
3834 if (buf->lb_len < rc)
3838 ol = &ff->ff_layout;
3839 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3840 PFID_STRIPE_IDX_BITS);
3841 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3842 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3843 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3844 if (lma->lma_compat & LMAC_COMP_INFO) {
3845 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3846 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3847 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3849 ol->ol_comp_start = 0;
3850 ol->ol_comp_end = 0;
3857 if (rc == -ENOENT || rc == -ENODATA)
3858 osd_oxc_add(obj, name, NULL, 0);
3859 else if (rc > 0 && buf->lb_buf != NULL)
3860 osd_oxc_add(obj, name, buf->lb_buf, rc);
3866 static int osd_declare_xattr_set(const struct lu_env *env,
3867 struct dt_object *dt,
3868 const struct lu_buf *buf, const char *name,
3869 int fl, struct thandle *handle)
3871 struct osd_thandle *oh;
3873 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3875 LASSERT(handle != NULL);
3877 oh = container_of0(handle, struct osd_thandle, ot_super);
3878 LASSERT(oh->ot_handle == NULL);
3880 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3881 /* For non-upgrading case, the LMA is set first and
3882 * usually fit inode. But for upgrade case, the LMA
3883 * may be in another separated EA block. */
3884 if (dt_object_exists(dt)) {
3885 if (fl == LU_XATTR_REPLACE)
3890 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3892 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3893 /* We may need to delete the old PFID EA. */
3894 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3895 if (fl == LU_XATTR_REPLACE)
3902 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3907 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3908 /* learn xattr size from osd_xattr_get if
3909 attribute has not been read yet */
3910 buflen = __osd_xattr_get(
3911 osd_dt_obj(dt)->oo_inode,
3912 &osd_oti_get(env)->oti_obj_dentry,
3917 buflen = buf->lb_len;
3920 if (buflen > sb->s_blocksize) {
3921 credits += osd_calc_bkmap_credits(
3923 (buflen + sb->s_blocksize - 1) >>
3924 sb->s_blocksize_bits);
3928 * xattr set may involve inode quota change, reserve credits for
3929 * dquot_initialize()
3931 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3934 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3940 * Concurrency: @dt is write locked.
3942 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3943 const struct lu_buf *buf, const char *name, int fl,
3944 struct thandle *handle)
3946 struct osd_object *obj = osd_dt_obj(dt);
3947 struct inode *inode = obj->oo_inode;
3948 struct osd_thread_info *info = osd_oti_get(env);
3949 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3950 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3959 /* version set is not real XATTR */
3960 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3961 dt_obj_version_t *version = buf->lb_buf;
3963 /* for version we are just using xattr API but change inode
3965 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3967 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3968 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3970 LDISKFS_I(inode)->i_fs_version = *version;
3971 /* Version is set after all inode operations are finished,
3972 * so we should mark it dirty here */
3973 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3978 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3979 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3982 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3983 if (fl & LU_XATTR_REPLACE)
3984 fs_flags |= XATTR_REPLACE;
3986 if (fl & LU_XATTR_CREATE)
3987 fs_flags |= XATTR_CREATE;
3989 /* For the OST device with 256 bytes inode size by default,
3990 * the PFID EA will be stored together with LMA EA to avoid
3991 * performance trouble. Otherwise the PFID EA can be stored
3992 * independently. LU-8998 */
3993 if (strcmp(name, XATTR_NAME_FID) == 0 &&
3994 LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
3995 struct dentry *dentry = &info->oti_obj_dentry;
3996 struct filter_fid *ff;
3997 struct ost_layout *ol;
4000 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4003 ol = &ff->ff_layout;
4004 /* Old client does not send stripe information, store
4005 * the PFID EA on disk directly. */
4006 if (buf->lb_len == sizeof(struct lu_fid) ||
4007 ol->ol_stripe_size == 0) {
4008 len = sizeof(struct lu_fid);
4012 if (buf->lb_len != sizeof(*ff))
4015 rc = osd_get_lma(info, inode, dentry, loa);
4016 if (unlikely(rc == -ENODATA)) {
4017 /* Usually for upgarding from old device */
4018 lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4019 LMAC_FID_ON_OST, 0);
4027 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4028 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4029 PFID_STRIPE_IDX_BITS;
4030 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4031 lma->lma_compat |= LMAC_STRIPE_INFO;
4032 if (ol->ol_comp_id != 0) {
4033 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4034 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4035 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4036 lma->lma_compat |= LMAC_COMP_INFO;
4039 lustre_loa_swab(loa, false);
4041 /* Remove old PFID EA entry firstly. */
4042 ll_vfs_dq_init(inode);
4043 rc = inode->i_op->removexattr(dentry, name);
4044 if (rc && rc != -ENODATA)
4047 /* Store the PFID EA inside the LMA EA. */
4048 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4052 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4053 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4057 lma->lma_incompat |= LMAI_STRIPED;
4058 lustre_lma_swab(lma);
4059 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4060 sizeof(*lma), XATTR_REPLACE);
4066 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4067 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4070 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4071 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4072 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4078 * Concurrency: @dt is read locked.
4080 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4081 const struct lu_buf *buf)
4083 struct osd_object *obj = osd_dt_obj(dt);
4084 struct inode *inode = obj->oo_inode;
4085 struct osd_thread_info *info = osd_oti_get(env);
4086 struct dentry *dentry = &info->oti_obj_dentry;
4088 if (!dt_object_exists(dt))
4091 LASSERT(!dt_object_remote(dt));
4092 LASSERT(inode->i_op != NULL);
4093 LASSERT(inode->i_op->listxattr != NULL);
4095 dentry->d_inode = inode;
4096 dentry->d_sb = inode->i_sb;
4097 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4100 static int osd_declare_xattr_del(const struct lu_env *env,
4101 struct dt_object *dt, const char *name,
4102 struct thandle *handle)
4104 struct osd_thandle *oh;
4105 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4107 LASSERT(!dt_object_remote(dt));
4108 LASSERT(handle != NULL);
4110 oh = container_of0(handle, struct osd_thandle, ot_super);
4111 LASSERT(oh->ot_handle == NULL);
4113 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4114 osd_dto_credits_noquota[DTO_XATTR_SET]);
4116 * xattr del may involve inode quota change, reserve credits for
4117 * dquot_initialize()
4119 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4125 * Concurrency: @dt is write locked.
4127 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4128 const char *name, struct thandle *handle)
4130 struct osd_object *obj = osd_dt_obj(dt);
4131 struct inode *inode = obj->oo_inode;
4132 struct osd_thread_info *info = osd_oti_get(env);
4133 struct dentry *dentry = &info->oti_obj_dentry;
4136 if (!dt_object_exists(dt))
4139 LASSERT(!dt_object_remote(dt));
4140 LASSERT(inode->i_op != NULL);
4141 LASSERT(inode->i_op->removexattr != NULL);
4142 LASSERT(handle != NULL);
4144 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4146 ll_vfs_dq_init(inode);
4147 dentry->d_inode = inode;
4148 dentry->d_sb = inode->i_sb;
4149 rc = inode->i_op->removexattr(dentry, name);
4150 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4151 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4153 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4155 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4156 &info->oti_ost_attrs);
4158 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4163 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4164 lustre_lma_swab(lma);
4165 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4166 sizeof(*lma), XATTR_REPLACE);
4171 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4174 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4175 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4176 osd_oxc_del(obj, name);
4181 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4182 __u64 start, __u64 end)
4184 struct osd_object *obj = osd_dt_obj(dt);
4185 struct inode *inode = obj->oo_inode;
4186 struct osd_thread_info *info = osd_oti_get(env);
4187 struct dentry *dentry = &info->oti_obj_dentry;
4188 struct file *file = &info->oti_file;
4193 dentry->d_inode = inode;
4194 dentry->d_sb = inode->i_sb;
4195 file->f_path.dentry = dentry;
4196 file->f_mapping = inode->i_mapping;
4197 file->f_op = inode->i_fop;
4198 set_file_inode(file, inode);
4200 rc = ll_vfs_fsync_range(file, start, end, 0);
4205 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4214 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4215 const struct dt_index_features *feat)
4217 struct iam_descr *descr;
4219 if (osd_object_is_root(o))
4220 return feat == &dt_directory_features;
4222 LASSERT(o->oo_dir != NULL);
4224 descr = o->oo_dir->od_container.ic_descr;
4225 if (feat == &dt_directory_features) {
4226 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4232 feat->dif_keysize_min <= descr->id_key_size &&
4233 descr->id_key_size <= feat->dif_keysize_max &&
4234 feat->dif_recsize_min <= descr->id_rec_size &&
4235 descr->id_rec_size <= feat->dif_recsize_max &&
4236 !(feat->dif_flags & (DT_IND_VARKEY |
4237 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4238 ergo(feat->dif_flags & DT_IND_UPDATE,
4239 1 /* XXX check that object (and file system) is
4244 static int osd_iam_container_init(const struct lu_env *env,
4245 struct osd_object *obj,
4246 struct osd_directory *dir)
4248 struct iam_container *bag = &dir->od_container;
4251 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4255 result = iam_container_setup(bag);
4257 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4259 iam_container_fini(bag);
4266 * Concurrency: no external locking is necessary.
4268 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4269 const struct dt_index_features *feat)
4273 struct osd_object *obj = osd_dt_obj(dt);
4275 LINVRNT(osd_invariant(obj));
4277 if (osd_object_is_root(obj)) {
4278 dt->do_index_ops = &osd_index_ea_ops;
4280 } else if (feat == &dt_directory_features) {
4281 dt->do_index_ops = &osd_index_ea_ops;
4282 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4287 } else if (unlikely(feat == &dt_otable_features)) {
4288 dt->do_index_ops = &osd_otable_ops;
4290 } else if (unlikely(feat == &dt_acct_features)) {
4291 dt->do_index_ops = &osd_acct_index_ops;
4294 } else if (!osd_has_index(obj)) {
4295 struct osd_directory *dir;
4300 spin_lock(&obj->oo_guard);
4301 if (obj->oo_dir == NULL)
4305 * Concurrent thread allocated container data.
4308 spin_unlock(&obj->oo_guard);
4310 * Now, that we have container data, serialize its
4313 down_write(&obj->oo_ext_idx_sem);
4315 * recheck under lock.
4317 if (!osd_has_index(obj))
4318 result = osd_iam_container_init(env, obj,
4322 up_write(&obj->oo_ext_idx_sem);
4330 if (result == 0 && skip_iam == 0) {
4331 if (!osd_iam_index_probe(env, obj, feat))
4334 LINVRNT(osd_invariant(obj));
4339 static int osd_otable_it_attr_get(const struct lu_env *env,
4340 struct dt_object *dt,
4341 struct lu_attr *attr)
4347 static const struct dt_object_operations osd_obj_ops = {
4348 .do_read_lock = osd_read_lock,
4349 .do_write_lock = osd_write_lock,
4350 .do_read_unlock = osd_read_unlock,
4351 .do_write_unlock = osd_write_unlock,
4352 .do_write_locked = osd_write_locked,
4353 .do_attr_get = osd_attr_get,
4354 .do_declare_attr_set = osd_declare_attr_set,
4355 .do_attr_set = osd_attr_set,
4356 .do_ah_init = osd_ah_init,
4357 .do_declare_create = osd_declare_create,
4358 .do_create = osd_create,
4359 .do_declare_destroy = osd_declare_destroy,
4360 .do_destroy = osd_destroy,
4361 .do_index_try = osd_index_try,
4362 .do_declare_ref_add = osd_declare_ref_add,
4363 .do_ref_add = osd_ref_add,
4364 .do_declare_ref_del = osd_declare_ref_del,
4365 .do_ref_del = osd_ref_del,
4366 .do_xattr_get = osd_xattr_get,
4367 .do_declare_xattr_set = osd_declare_xattr_set,
4368 .do_xattr_set = osd_xattr_set,
4369 .do_declare_xattr_del = osd_declare_xattr_del,
4370 .do_xattr_del = osd_xattr_del,
4371 .do_xattr_list = osd_xattr_list,
4372 .do_object_sync = osd_object_sync,
4373 .do_invalidate = osd_invalidate,
4376 static const struct dt_object_operations osd_obj_otable_it_ops = {
4377 .do_attr_get = osd_otable_it_attr_get,
4378 .do_index_try = osd_index_try,
4381 static int osd_index_declare_iam_delete(const struct lu_env *env,
4382 struct dt_object *dt,
4383 const struct dt_key *key,
4384 struct thandle *handle)
4386 struct osd_thandle *oh;
4388 oh = container_of0(handle, struct osd_thandle, ot_super);
4389 LASSERT(oh->ot_handle == NULL);
4391 /* Recycle may cause additional three blocks to be changed. */
4392 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4393 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4399 * delete a (key, value) pair from index \a dt specified by \a key
4401 * \param dt osd index object
4402 * \param key key for index
4403 * \param rec record reference
4404 * \param handle transaction handler
4407 * \retval -ve failure
4409 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4410 const struct dt_key *key,
4411 struct thandle *handle)
4413 struct osd_thread_info *oti = osd_oti_get(env);
4414 struct osd_object *obj = osd_dt_obj(dt);
4415 struct osd_thandle *oh;
4416 struct iam_path_descr *ipd;
4417 struct iam_container *bag = &obj->oo_dir->od_container;
4421 if (!dt_object_exists(dt))
4424 LINVRNT(osd_invariant(obj));
4425 LASSERT(!dt_object_remote(dt));
4426 LASSERT(bag->ic_object == obj->oo_inode);
4427 LASSERT(handle != NULL);
4429 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4431 ipd = osd_idx_ipd_get(env, bag);
4432 if (unlikely(ipd == NULL))
4435 oh = container_of0(handle, struct osd_thandle, ot_super);
4436 LASSERT(oh->ot_handle != NULL);
4437 LASSERT(oh->ot_handle->h_transaction != NULL);
4439 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4440 /* swab quota uid/gid provided by caller */
4441 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4442 key = (const struct dt_key *)&oti->oti_quota_id;
4445 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4446 osd_ipd_put(env, bag, ipd);
4447 LINVRNT(osd_invariant(obj));
4448 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4452 static int osd_index_declare_ea_delete(const struct lu_env *env,
4453 struct dt_object *dt,
4454 const struct dt_key *key,
4455 struct thandle *handle)
4457 struct osd_thandle *oh;
4458 struct inode *inode;
4462 LASSERT(!dt_object_remote(dt));
4463 LASSERT(handle != NULL);
4465 oh = container_of0(handle, struct osd_thandle, ot_super);
4466 LASSERT(oh->ot_handle == NULL);
4468 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4469 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4470 /* '..' to a remote object has a local representative */
4471 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4472 /* to reset LMAI_REMOTE_PARENT */
4475 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4477 inode = osd_dt_obj(dt)->oo_inode;
4481 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4482 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4487 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4490 struct osd_fid_pack *rec;
4493 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4494 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4495 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4496 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4502 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4503 const struct lu_fid *fid)
4505 struct seq_server_site *ss = osd_seq_site(osd);
4508 /* FID seqs not in FLDB, must be local seq */
4509 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4512 /* If FLD is not being initialized yet, it only happens during the
4513 * initialization, likely during mgs initialization, and we assume
4514 * this is local FID. */
4515 if (ss == NULL || ss->ss_server_fld == NULL)
4518 /* Only check the local FLDB here */
4519 if (osd_seq_exists(env, osd, fid_seq(fid)))
4526 * Index delete function for interoperability mode (b11826).
4527 * It will remove the directory entry added by osd_index_ea_insert().
4528 * This entry is needed to maintain name->fid mapping.
4530 * \param key, key i.e. file entry to be deleted
4532 * \retval 0, on success
4533 * \retval -ve, on error
4535 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4536 const struct dt_key *key, struct thandle *handle)
4538 struct osd_object *obj = osd_dt_obj(dt);
4539 struct inode *dir = obj->oo_inode;
4540 struct dentry *dentry;
4541 struct osd_thandle *oh;
4542 struct ldiskfs_dir_entry_2 *de = NULL;
4543 struct buffer_head *bh;
4544 struct htree_lock *hlock = NULL;
4545 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4546 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4550 if (!dt_object_exists(dt))
4553 LINVRNT(osd_invariant(obj));
4554 LASSERT(!dt_object_remote(dt));
4555 LASSERT(handle != NULL);
4557 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4559 oh = container_of(handle, struct osd_thandle, ot_super);
4560 LASSERT(oh->ot_handle != NULL);
4561 LASSERT(oh->ot_handle->h_transaction != NULL);
4563 ll_vfs_dq_init(dir);
4564 dentry = osd_child_dentry_get(env, obj,
4565 (char *)key, strlen((char *)key));
4567 if (obj->oo_hl_head != NULL) {
4568 hlock = osd_oti_get(env)->oti_hlock;
4569 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4570 dir, LDISKFS_HLOCK_DEL);
4572 down_write(&obj->oo_ext_idx_sem);
4575 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4577 /* If this is not the ".." entry, it might be a remote DNE
4578 * entry and we need to check if the FID is for a remote
4579 * MDT. If the FID is not in the directory entry (e.g.
4580 * upgraded 1.8 filesystem without dirdata enabled) then
4581 * we need to get the FID from the LMA. For a remote directory
4582 * there HAS to be an LMA, it cannot be an IGIF inode in this
4585 * Delete the entry before the agent inode in order to
4586 * simplify error handling. At worst an error after deleting
4587 * the entry first might leak the agent inode afterward. The
4588 * reverse would need filesystem abort in case of error deleting
4589 * the entry after the agent had been removed, or leave a
4590 * dangling entry pointing at a random inode. */
4591 if (strcmp((char *)key, dotdot) != 0) {
4592 LASSERT(de != NULL);
4593 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4594 if (rc == -ENODATA) {
4595 /* can't get FID, postpone to the end of the
4596 * transaction when iget() is safe */
4597 osd_schedule_agent_inode_removal(env, oh,
4598 le32_to_cpu(de->inode));
4599 } else if (rc == 0 &&
4600 unlikely(osd_remote_fid(env, osd, fid))) {
4601 osd_schedule_agent_inode_removal(env, oh,
4602 le32_to_cpu(de->inode));
4605 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4611 ldiskfs_htree_unlock(hlock);
4613 up_write(&obj->oo_ext_idx_sem);
4618 /* For inode on the remote MDT, .. will point to
4619 * /Agent directory, Check whether it needs to delete
4620 * from agent directory */
4621 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4624 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4627 /* Sigh, the entry has been deleted, and
4628 * it is not easy to revert it back, so
4629 * let's keep this error private, and let
4630 * LFSCK fix it. XXX */
4631 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4632 osd_name(osd), PFID(fid), ret);
4635 LASSERT(osd_invariant(obj));
4636 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4641 * Lookup index for \a key and copy record to \a rec.
4643 * \param dt osd index object
4644 * \param key key for index
4645 * \param rec record reference
4647 * \retval +ve success : exact mach
4648 * \retval 0 return record with key not greater than \a key
4649 * \retval -ve failure
4651 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4652 struct dt_rec *rec, const struct dt_key *key)
4654 struct osd_object *obj = osd_dt_obj(dt);
4655 struct iam_path_descr *ipd;
4656 struct iam_container *bag = &obj->oo_dir->od_container;
4657 struct osd_thread_info *oti = osd_oti_get(env);
4658 struct iam_iterator *it = &oti->oti_idx_it;
4659 struct iam_rec *iam_rec;
4663 if (!dt_object_exists(dt))
4666 LASSERT(osd_invariant(obj));
4667 LASSERT(!dt_object_remote(dt));
4668 LASSERT(bag->ic_object == obj->oo_inode);
4670 ipd = osd_idx_ipd_get(env, bag);
4674 /* got ipd now we can start iterator. */
4675 iam_it_init(it, bag, 0, ipd);
4677 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4678 /* swab quota uid/gid provided by caller */
4679 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4680 key = (const struct dt_key *)&oti->oti_quota_id;
4683 rc = iam_it_get(it, (struct iam_key *)key);
4685 if (S_ISDIR(obj->oo_inode->i_mode))
4686 iam_rec = (struct iam_rec *)oti->oti_ldp;
4688 iam_rec = (struct iam_rec *) rec;
4690 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4692 if (S_ISDIR(obj->oo_inode->i_mode))
4693 osd_fid_unpack((struct lu_fid *) rec,
4694 (struct osd_fid_pack *)iam_rec);
4695 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4696 osd_quota_unpack(obj, rec);
4701 osd_ipd_put(env, bag, ipd);
4703 LINVRNT(osd_invariant(obj));
4708 static int osd_index_declare_iam_insert(const struct lu_env *env,
4709 struct dt_object *dt,
4710 const struct dt_rec *rec,
4711 const struct dt_key *key,
4712 struct thandle *handle)
4714 struct osd_thandle *oh;
4716 LASSERT(handle != NULL);
4718 oh = container_of0(handle, struct osd_thandle, ot_super);
4719 LASSERT(oh->ot_handle == NULL);
4721 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4722 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4728 * Inserts (key, value) pair in \a dt index object.
4730 * \param dt osd index object
4731 * \param key key for index
4732 * \param rec record reference
4733 * \param th transaction handler
4736 * \retval -ve failure
4738 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4739 const struct dt_rec *rec,
4740 const struct dt_key *key, struct thandle *th,
4743 struct osd_object *obj = osd_dt_obj(dt);
4744 struct iam_path_descr *ipd;
4745 struct osd_thandle *oh;
4746 struct iam_container *bag;
4747 struct osd_thread_info *oti = osd_oti_get(env);
4748 struct iam_rec *iam_rec;
4752 if (!dt_object_exists(dt))
4755 LINVRNT(osd_invariant(obj));
4756 LASSERT(!dt_object_remote(dt));
4758 bag = &obj->oo_dir->od_container;
4759 LASSERT(bag->ic_object == obj->oo_inode);
4760 LASSERT(th != NULL);
4762 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4764 ipd = osd_idx_ipd_get(env, bag);
4765 if (unlikely(ipd == NULL))
4768 oh = container_of0(th, struct osd_thandle, ot_super);
4769 LASSERT(oh->ot_handle != NULL);
4770 LASSERT(oh->ot_handle->h_transaction != NULL);
4771 if (S_ISDIR(obj->oo_inode->i_mode)) {
4772 iam_rec = (struct iam_rec *)oti->oti_ldp;
4773 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4774 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4775 /* pack quota uid/gid */
4776 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4777 key = (const struct dt_key *)&oti->oti_quota_id;
4778 /* pack quota record */
4779 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4780 iam_rec = (struct iam_rec *)rec;
4782 iam_rec = (struct iam_rec *)rec;
4785 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4787 osd_ipd_put(env, bag, ipd);
4788 LINVRNT(osd_invariant(obj));
4789 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4794 * Calls ldiskfs_add_entry() to add directory entry
4795 * into the directory. This is required for
4796 * interoperability mode (b11826)
4798 * \retval 0, on success
4799 * \retval -ve, on error
4801 static int __osd_ea_add_rec(struct osd_thread_info *info,
4802 struct osd_object *pobj, struct inode *cinode,
4803 const char *name, const struct lu_fid *fid,
4804 struct htree_lock *hlock, struct thandle *th)
4806 struct ldiskfs_dentry_param *ldp;
4807 struct dentry *child;
4808 struct osd_thandle *oth;
4811 oth = container_of(th, struct osd_thandle, ot_super);
4812 LASSERT(oth->ot_handle != NULL);
4813 LASSERT(oth->ot_handle->h_transaction != NULL);
4814 LASSERT(pobj->oo_inode);
4816 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4817 if (unlikely(pobj->oo_inode ==
4818 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4821 osd_get_ldiskfs_dirent_param(ldp, fid);
4822 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4823 child->d_fsdata = (void *)ldp;
4824 ll_vfs_dq_init(pobj->oo_inode);
4825 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4826 child, cinode, hlock);
4827 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4828 struct ldiskfs_dir_entry_2 *de;
4829 struct buffer_head *bh;
4832 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4835 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4838 if (S_ISDIR(cinode->i_mode))
4839 de->file_type = LDISKFS_DIRENT_LUFID |
4840 LDISKFS_FT_REG_FILE;
4842 de->file_type = LDISKFS_DIRENT_LUFID |
4844 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4855 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4856 * into the directory.Also sets flags into osd object to
4857 * indicate dot and dotdot are created. This is required for
4858 * interoperability mode (b11826)
4860 * \param dir directory for dot and dotdot fixup.
4861 * \param obj child object for linking
4863 * \retval 0, on success
4864 * \retval -ve, on error
4866 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4867 struct osd_object *dir,
4868 struct inode *parent_dir, const char *name,
4869 const struct lu_fid *dot_fid,
4870 const struct lu_fid *dot_dot_fid,
4873 struct inode *inode = dir->oo_inode;
4874 struct osd_thandle *oth;
4877 oth = container_of(th, struct osd_thandle, ot_super);
4878 LASSERT(oth->ot_handle->h_transaction != NULL);
4879 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4881 if (strcmp(name, dot) == 0) {
4882 if (dir->oo_compat_dot_created) {
4885 LASSERT(inode->i_ino == parent_dir->i_ino);
4886 dir->oo_compat_dot_created = 1;
4889 } else if (strcmp(name, dotdot) == 0) {
4890 if (!dir->oo_compat_dot_created)
4892 /* in case of rename, dotdot is already created */
4893 if (dir->oo_compat_dotdot_created) {
4894 return __osd_ea_add_rec(info, dir, parent_dir, name,
4895 dot_dot_fid, NULL, th);
4898 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4899 struct lu_fid tfid = *dot_dot_fid;
4902 result = osd_add_dot_dotdot_internal(info,
4903 dir->oo_inode, parent_dir, dot_fid,
4906 result = osd_add_dot_dotdot_internal(info,
4907 dir->oo_inode, parent_dir, dot_fid,
4912 dir->oo_compat_dotdot_created = 1;
4920 * It will call the appropriate osd_add* function and return the
4921 * value, return by respective functions.
4923 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4924 struct inode *cinode, const char *name,
4925 const struct lu_fid *fid, struct thandle *th)
4927 struct osd_thread_info *info = osd_oti_get(env);
4928 struct htree_lock *hlock;
4931 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4933 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4935 if (hlock != NULL) {
4936 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4939 down_write(&pobj->oo_ext_idx_sem);
4942 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4943 lu_object_fid(&pobj->oo_dt.do_lu),
4946 if (hlock != NULL) {
4947 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4948 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4950 down_write(&pobj->oo_ext_idx_sem);
4953 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4954 struct lu_fid *tfid = &info->oti_fid;
4958 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4961 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4966 ldiskfs_htree_unlock(hlock);
4968 up_write(&pobj->oo_ext_idx_sem);
4974 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4975 struct osd_idmap_cache *oic)
4977 struct osd_scrub *scrub = &dev->od_scrub;
4978 struct lu_fid *fid = &oic->oic_fid;
4979 struct osd_inode_id *id = &oic->oic_lid;
4980 struct inode *inode = NULL;
4986 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4989 if (scrub->os_pos_current > id->oii_ino)
4993 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4994 if (rc == -ENOENT) {
4995 __u32 gen = id->oii_gen;
5001 inode = osd_iget(oti, dev, id);
5002 /* The inode has been removed (by race maybe). */
5003 if (IS_ERR(inode)) {
5004 rc = PTR_ERR(inode);
5006 RETURN(rc == -ESTALE ? -ENOENT : rc);
5009 /* The OI mapping is lost. */
5010 if (gen != OSD_OII_NOGEN)
5013 /* The inode may has been reused by others, we do not know,
5014 * leave it to be handled by subsequent osd_fid_lookup(). */
5016 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5023 if (thread_is_running(&scrub->os_thread)) {
5024 if (inode == NULL) {
5025 inode = osd_iget(oti, dev, id);
5026 /* The inode has been removed (by race maybe). */
5027 if (IS_ERR(inode)) {
5028 rc = PTR_ERR(inode);
5030 RETURN(rc == -ESTALE ? -ENOENT : rc);
5034 rc = osd_oii_insert(dev, oic, insert);
5035 /* There is race condition between osd_oi_lookup and OI scrub.
5036 * The OI scrub finished just after osd_oi_lookup() failure.
5037 * Under such case, it is unnecessary to trigger OI scrub again,
5038 * but try to call osd_oi_lookup() again. */
5039 if (unlikely(rc == -EAGAIN))
5042 if (!S_ISDIR(inode->i_mode))
5045 rc = osd_check_lmv(oti, dev, inode, oic);
5050 if (!dev->od_noscrub && ++once == 1) {
5051 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5053 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5054 "%s: trigger partial OI scrub for RPC inconsistency "
5055 "checking FID "DFID": rc = %d\n",
5056 osd_dev2name(dev), PFID(fid), rc);
5057 if (rc == 0 || rc == -EALREADY)
5070 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5071 struct osd_device *dev,
5072 struct osd_idmap_cache *oic,
5073 struct lu_fid *fid, __u32 ino)
5075 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5076 struct inode *inode;
5079 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5080 inode = osd_iget(oti, dev, &oic->oic_lid);
5081 if (IS_ERR(inode)) {
5082 fid_zero(&oic->oic_fid);
5083 return PTR_ERR(inode);
5086 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5089 fid_zero(&oic->oic_fid);
5091 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5095 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5096 struct osd_inode_id *id, const struct lu_fid *fid)
5098 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5099 id->oii_ino, id->oii_gen, info);
5100 info->oti_cache.oic_lid = *id;
5101 info->oti_cache.oic_fid = *fid;
5102 info->oti_cache.oic_dev = osd;
5106 * Get parent FID from the linkEA.
5108 * For a directory which parent resides on remote MDT, to satisfy the
5109 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5110 * the other hand, to make the lookup(..) on the directory can return
5111 * the real parent FID, we append the real parent FID after its ".."
5112 * name entry in the /REMOTE_PARENT_DIR.
5114 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5115 * backup. So after the restore, we cannot get the right parent FID from
5116 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5117 * we have stored the real parent FID in the directory object's linkEA,
5118 * we can parse the linkEA for the real parent FID.
5120 * \param[in] env pointer to the thread context
5121 * \param[in] obj pointer to the object to be handled
5122 * \param[out]fid pointer to the buffer to hold the parent FID
5124 * \retval 0 for getting the real parent FID successfully
5125 * \retval negative error number on failure
5127 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5128 struct osd_object *obj,
5131 struct osd_thread_info *oti = osd_oti_get(env);
5132 struct lu_buf *buf = &oti->oti_big_buf;
5133 struct dentry *dentry = &oti->oti_obj_dentry;
5134 struct inode *inode = obj->oo_inode;
5135 struct linkea_data ldata = { NULL };
5140 if (!S_ISDIR(inode->i_mode))
5144 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5145 buf->lb_buf, buf->lb_len);
5146 if (rc == -ERANGE) {
5147 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5150 lu_buf_realloc(buf, rc);
5151 if (buf->lb_buf == NULL)
5158 if (unlikely(rc == 0))
5164 if (unlikely(buf->lb_buf == NULL)) {
5165 lu_buf_realloc(buf, rc);
5166 if (buf->lb_buf == NULL)
5173 rc = linkea_init_with_rec(&ldata);
5175 linkea_first_entry(&ldata);
5176 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5182 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5183 struct inode *inode,
5184 const struct lu_fid *pfid,
5185 const char *name, const int namelen)
5187 struct osd_thread_info *oti = osd_oti_get(env);
5188 struct lu_buf *buf = &oti->oti_big_buf;
5189 struct dentry *dentry = &oti->oti_obj_dentry;
5190 struct linkea_data ldata = { NULL };
5191 struct lu_name cname = { .ln_name = name,
5192 .ln_namelen = namelen };
5197 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5198 buf->lb_buf, buf->lb_len);
5200 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5205 if (unlikely(rc == 0))
5208 if (buf->lb_len < rc) {
5209 lu_buf_realloc(buf, rc);
5210 if (buf->lb_buf == NULL)
5217 rc = linkea_init_with_rec(&ldata);
5219 rc = linkea_links_find(&ldata, &cname, pfid);
5225 * Calls ->lookup() to find dentry. From dentry get inode and
5226 * read inode's ea to get fid. This is required for interoperability
5229 * \retval 0, on success
5230 * \retval -ve, on error
5232 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5233 struct dt_rec *rec, const struct dt_key *key)
5235 struct inode *dir = obj->oo_inode;
5236 struct dentry *dentry;
5237 struct ldiskfs_dir_entry_2 *de;
5238 struct buffer_head *bh;
5239 struct lu_fid *fid = (struct lu_fid *) rec;
5240 struct htree_lock *hlock = NULL;
5245 LASSERT(dir->i_op != NULL);
5246 LASSERT(dir->i_op->lookup != NULL);
5248 dentry = osd_child_dentry_get(env, obj,
5249 (char *)key, strlen((char *)key));
5251 if (obj->oo_hl_head != NULL) {
5252 hlock = osd_oti_get(env)->oti_hlock;
5253 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5254 dir, LDISKFS_HLOCK_LOOKUP);
5256 down_read(&obj->oo_ext_idx_sem);
5259 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5261 struct osd_thread_info *oti = osd_oti_get(env);
5262 struct osd_inode_id *id = &oti->oti_id;
5263 struct osd_idmap_cache *oic = &oti->oti_cache;
5264 struct osd_device *dev = osd_obj2dev(obj);
5266 ino = le32_to_cpu(de->inode);
5267 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5269 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5273 rc = osd_get_fid_from_dentry(de, rec);
5275 /* done with de, release bh */
5278 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5279 const char *name = (const char *)key;
5281 /* If the parent is on remote MDT, and there
5282 * is no FID-in-dirent, then we have to get
5283 * the parent FID from the linkEA. */
5284 if (likely(strlen(name) == 2 &&
5285 name[0] == '.' && name[1] == '.'))
5286 rc = osd_get_pfid_from_linkea(env, obj,
5289 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5292 osd_id_gen(id, ino, OSD_OII_NOGEN);
5295 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5296 fid_zero(&oic->oic_fid);
5301 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5302 rc = osd_consistency_check(oti, dev, oic);
5304 fid_zero(&oic->oic_fid);
5313 ldiskfs_htree_unlock(hlock);
5315 up_read(&obj->oo_ext_idx_sem);
5319 static int osd_index_declare_ea_insert(const struct lu_env *env,
5320 struct dt_object *dt,
5321 const struct dt_rec *rec,
5322 const struct dt_key *key,
5323 struct thandle *handle)
5325 struct osd_thandle *oh;
5326 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5327 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5328 const struct lu_fid *fid = rec1->rec_fid;
5329 int credits, rc = 0;
5330 struct osd_idmap_cache *idc;
5333 LASSERT(!dt_object_remote(dt));
5334 LASSERT(handle != NULL);
5335 LASSERT(fid != NULL);
5336 LASSERT(rec1->rec_type != 0);
5338 oh = container_of0(handle, struct osd_thandle, ot_super);
5339 LASSERT(oh->ot_handle == NULL);
5341 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5343 /* we can't call iget() while a transactions is running
5344 * (this can lead to a deadlock), but we need to know
5345 * inum and object type. so we find this information at
5346 * declaration and cache in per-thread info */
5347 idc = osd_idc_find_or_init(env, osd, fid);
5349 RETURN(PTR_ERR(idc));
5350 if (idc->oic_remote) {
5351 /* a reference to remote inode is represented by an
5352 * agent inode which we have to create */
5353 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5354 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5357 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5359 if (osd_dt_obj(dt)->oo_inode != NULL) {
5360 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5362 /* We ignore block quota on meta pool (MDTs), so needn't
5363 * calculate how many blocks will be consumed by this index
5365 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5367 i_projid_read(inode), 0,
5368 oh, osd_dt_obj(dt), NULL,
5373 #ifdef HAVE_PROJECT_QUOTA
5374 /* Reserve credits for local agent inode to transfer
5375 * to 0, quota enforcement is ignored in this case.
5377 if (idc->oic_remote &&
5378 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5379 i_projid_read(inode) != 0)
5380 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5381 0, i_projid_read(inode),
5382 0, false, PRJQUOTA);
5391 * Index add function for interoperability mode (b11826).
5392 * It will add the directory entry.This entry is needed to
5393 * maintain name->fid mapping.
5395 * \param key it is key i.e. file entry to be inserted
5396 * \param rec it is value of given key i.e. fid
5398 * \retval 0, on success
5399 * \retval -ve, on error
5401 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5402 const struct dt_rec *rec,
5403 const struct dt_key *key, struct thandle *th,
5406 struct osd_object *obj = osd_dt_obj(dt);
5407 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5408 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5409 const struct lu_fid *fid = rec1->rec_fid;
5410 const char *name = (const char *)key;
5411 struct osd_thread_info *oti = osd_oti_get(env);
5412 struct inode *child_inode = NULL;
5413 struct osd_idmap_cache *idc;
5417 if (!dt_object_exists(dt))
5420 LASSERT(osd_invariant(obj));
5421 LASSERT(!dt_object_remote(dt));
5422 LASSERT(th != NULL);
5424 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5426 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5428 idc = osd_idc_find(env, osd, fid);
5429 if (unlikely(idc == NULL)) {
5430 /* this dt_insert() wasn't declared properly, so
5431 * FID is missing in OI cache. we better do not
5432 * lookup FID in FLDB/OI and don't risk to deadlock,
5433 * but in some special cases (lfsck testing, etc)
5434 * it's much simpler than fixing a caller */
5435 CERROR("%s: "DFID" wasn't declared for insert\n",
5436 osd_name(osd), PFID(fid));
5438 idc = osd_idc_find_or_init(env, osd, fid);
5440 RETURN(PTR_ERR(idc));
5443 if (idc->oic_remote) {
5444 /* Insert remote entry */
5445 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5446 struct osd_mdobj_map *omm = osd->od_mdt_map;
5447 struct osd_thandle *oh;
5449 /* If parent on remote MDT, we need put this object
5451 oh = container_of(th, typeof(*oh), ot_super);
5452 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5454 CERROR("%s: add "DFID" error: rc = %d\n",
5456 PFID(lu_object_fid(&dt->do_lu)), rc);
5460 child_inode = igrab(omm->omm_remote_parent->d_inode);
5462 child_inode = osd_create_local_agent_inode(env, osd,
5463 obj, fid, rec1->rec_type & S_IFMT, th);
5464 if (IS_ERR(child_inode))
5465 RETURN(PTR_ERR(child_inode));
5468 /* Insert local entry */
5469 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5470 /* for a reason OI cache wasn't filled properly */
5471 CERROR("%s: OIC for "DFID" isn't filled\n",
5472 osd_name(osd), PFID(fid));
5475 child_inode = oti->oti_inode;
5476 if (unlikely(child_inode == NULL)) {
5477 struct ldiskfs_inode_info *lii;
5481 child_inode = oti->oti_inode = &lii->vfs_inode;
5483 child_inode->i_sb = osd_sb(osd);
5484 child_inode->i_ino = idc->oic_lid.oii_ino;
5485 child_inode->i_mode = rec1->rec_type & S_IFMT;
5488 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5490 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5491 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5493 if (child_inode && child_inode != oti->oti_inode)
5495 LASSERT(osd_invariant(obj));
5496 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5501 * Initialize osd Iterator for given osd index object.
5503 * \param dt osd index object
5506 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5507 struct dt_object *dt,
5510 struct osd_it_iam *it;
5511 struct osd_object *obj = osd_dt_obj(dt);
5512 struct lu_object *lo = &dt->do_lu;
5513 struct iam_path_descr *ipd;
5514 struct iam_container *bag = &obj->oo_dir->od_container;
5516 if (!dt_object_exists(dt))
5517 return ERR_PTR(-ENOENT);
5521 return ERR_PTR(-ENOMEM);
5523 ipd = osd_it_ipd_get(env, bag);
5524 if (likely(ipd != NULL)) {
5528 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5529 return (struct dt_it *)it;
5532 return ERR_PTR(-ENOMEM);
5537 * free given Iterator.
5539 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5541 struct osd_it_iam *it = (struct osd_it_iam *)di;
5542 struct osd_object *obj = it->oi_obj;
5544 iam_it_fini(&it->oi_it);
5545 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5546 osd_object_put(env, obj);
5551 * Move Iterator to record specified by \a key
5553 * \param di osd iterator
5554 * \param key key for index
5556 * \retval +ve di points to record with least key not larger than key
5557 * \retval 0 di points to exact matched key
5558 * \retval -ve failure
5561 static int osd_it_iam_get(const struct lu_env *env,
5562 struct dt_it *di, const struct dt_key *key)
5564 struct osd_thread_info *oti = osd_oti_get(env);
5565 struct osd_it_iam *it = (struct osd_it_iam *)di;
5567 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5568 /* swab quota uid/gid */
5569 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5570 key = (struct dt_key *)&oti->oti_quota_id;
5573 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5579 * \param di osd iterator
5581 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5583 struct osd_it_iam *it = (struct osd_it_iam *)di;
5585 iam_it_put(&it->oi_it);
5589 * Move iterator by one record
5591 * \param di osd iterator
5593 * \retval +1 end of container reached
5595 * \retval -ve failure
5598 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5600 struct osd_it_iam *it = (struct osd_it_iam *)di;
5602 return iam_it_next(&it->oi_it);
5606 * Return pointer to the key under iterator.
5609 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5610 const struct dt_it *di)
5612 struct osd_thread_info *oti = osd_oti_get(env);
5613 struct osd_it_iam *it = (struct osd_it_iam *)di;
5614 struct osd_object *obj = it->oi_obj;
5617 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5619 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5620 /* swab quota uid/gid */
5621 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5622 key = (struct dt_key *)&oti->oti_quota_id;
5629 * Return size of key under iterator (in bytes)
5632 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5634 struct osd_it_iam *it = (struct osd_it_iam *)di;
5636 return iam_it_key_size(&it->oi_it);
5640 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5642 /* check if file type is required */
5643 if (ent->lde_attrs & LUDA_TYPE) {
5644 struct luda_type *lt;
5645 int align = sizeof(*lt) - 1;
5647 len = (len + align) & ~align;
5648 lt = (struct luda_type *)(ent->lde_name + len);
5649 lt->lt_type = cpu_to_le16(DTTOIF(type));
5652 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5656 * build lu direct from backend fs dirent.
5660 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5661 char *name, __u16 namelen, __u16 type, __u32 attr)
5663 ent->lde_attrs = attr | LUDA_FID;
5664 fid_cpu_to_le(&ent->lde_fid, fid);
5666 ent->lde_hash = cpu_to_le64(offset);
5667 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5669 strncpy(ent->lde_name, name, namelen);
5670 ent->lde_name[namelen] = '\0';
5671 ent->lde_namelen = cpu_to_le16(namelen);
5673 /* append lustre attributes */
5674 osd_it_append_attrs(ent, namelen, type);
5678 * Return pointer to the record under iterator.
5680 static int osd_it_iam_rec(const struct lu_env *env,
5681 const struct dt_it *di,
5682 struct dt_rec *dtrec, __u32 attr)
5684 struct osd_it_iam *it = (struct osd_it_iam *)di;
5685 struct osd_thread_info *info = osd_oti_get(env);
5688 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5689 const struct osd_fid_pack *rec;
5690 struct lu_fid *fid = &info->oti_fid;
5691 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5697 name = (char *)iam_it_key_get(&it->oi_it);
5699 RETURN(PTR_ERR(name));
5701 namelen = iam_it_key_size(&it->oi_it);
5703 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5705 RETURN(PTR_ERR(rec));
5707 rc = osd_fid_unpack(fid, rec);
5711 hash = iam_it_store(&it->oi_it);
5713 /* IAM does not store object type in IAM index (dir) */
5714 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5716 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5717 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5718 (struct iam_rec *)dtrec);
5719 osd_quota_unpack(it->oi_obj, dtrec);
5721 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5722 (struct iam_rec *)dtrec);
5729 * Returns cookie for current Iterator position.
5731 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5733 struct osd_it_iam *it = (struct osd_it_iam *)di;
5735 return iam_it_store(&it->oi_it);
5739 * Restore iterator from cookie.
5741 * \param di osd iterator
5742 * \param hash Iterator location cookie
5744 * \retval +ve di points to record with least key not larger than key.
5745 * \retval 0 di points to exact matched key
5746 * \retval -ve failure
5749 static int osd_it_iam_load(const struct lu_env *env,
5750 const struct dt_it *di, __u64 hash)
5752 struct osd_it_iam *it = (struct osd_it_iam *)di;
5754 return iam_it_load(&it->oi_it, hash);
5757 static const struct dt_index_operations osd_index_iam_ops = {
5758 .dio_lookup = osd_index_iam_lookup,
5759 .dio_declare_insert = osd_index_declare_iam_insert,
5760 .dio_insert = osd_index_iam_insert,
5761 .dio_declare_delete = osd_index_declare_iam_delete,
5762 .dio_delete = osd_index_iam_delete,
5764 .init = osd_it_iam_init,
5765 .fini = osd_it_iam_fini,
5766 .get = osd_it_iam_get,
5767 .put = osd_it_iam_put,
5768 .next = osd_it_iam_next,
5769 .key = osd_it_iam_key,
5770 .key_size = osd_it_iam_key_size,
5771 .rec = osd_it_iam_rec,
5772 .store = osd_it_iam_store,
5773 .load = osd_it_iam_load
5779 * Creates or initializes iterator context.
5781 * \retval struct osd_it_ea, iterator structure on success
5784 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5785 struct dt_object *dt,
5788 struct osd_object *obj = osd_dt_obj(dt);
5789 struct osd_thread_info *info = osd_oti_get(env);
5790 struct osd_it_ea *oie;
5792 struct lu_object *lo = &dt->do_lu;
5793 struct dentry *obj_dentry;
5796 if (!dt_object_exists(dt) || obj->oo_destroyed)
5797 RETURN(ERR_PTR(-ENOENT));
5799 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5801 RETURN(ERR_PTR(-ENOMEM));
5802 obj_dentry = &oie->oie_dentry;
5804 obj_dentry->d_inode = obj->oo_inode;
5805 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5806 obj_dentry->d_name.hash = 0;
5808 oie->oie_rd_dirent = 0;
5809 oie->oie_it_dirent = 0;
5810 oie->oie_dirent = NULL;
5811 if (unlikely(!info->oti_it_ea_buf_used)) {
5812 oie->oie_buf = info->oti_it_ea_buf;
5813 info->oti_it_ea_buf_used = 1;
5815 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5816 if (oie->oie_buf == NULL)
5817 RETURN(ERR_PTR(-ENOMEM));
5821 file = &oie->oie_file;
5823 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5824 if (attr & LUDA_64BITHASH)
5825 file->f_mode = FMODE_64BITHASH;
5827 file->f_mode = FMODE_32BITHASH;
5828 file->f_path.dentry = obj_dentry;
5829 file->f_mapping = obj->oo_inode->i_mapping;
5830 file->f_op = obj->oo_inode->i_fop;
5831 set_file_inode(file, obj->oo_inode);
5834 RETURN((struct dt_it *) oie);
5838 * Destroy or finishes iterator context.
5840 * \param di iterator structure to be destroyed
5842 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5844 struct osd_thread_info *info = osd_oti_get(env);
5845 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5846 struct osd_object *obj = oie->oie_obj;
5847 struct inode *inode = obj->oo_inode;
5850 oie->oie_file.f_op->release(inode, &oie->oie_file);
5851 osd_object_put(env, obj);
5852 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5853 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5855 info->oti_it_ea_buf_used = 0;
5856 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5861 * It position the iterator at given key, so that next lookup continues from
5862 * that key Or it is similar to dio_it->load() but based on a key,
5863 * rather than file position.
5865 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5868 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5870 static int osd_it_ea_get(const struct lu_env *env,
5871 struct dt_it *di, const struct dt_key *key)
5873 struct osd_it_ea *it = (struct osd_it_ea *)di;
5876 LASSERT(((const char *)key)[0] == '\0');
5877 it->oie_file.f_pos = 0;
5878 it->oie_rd_dirent = 0;
5879 it->oie_it_dirent = 0;
5880 it->oie_dirent = NULL;
5888 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5892 struct osd_filldir_cbs {
5893 #ifdef HAVE_DIR_CONTEXT
5894 struct dir_context ctx;
5896 struct osd_it_ea *it;
5899 * It is called internally by ->readdir(). It fills the
5900 * iterator's in-memory data structure with required
5901 * information i.e. name, namelen, rec_size etc.
5903 * \param buf in which information to be filled in.
5904 * \param name name of the file in given dir
5906 * \retval 0 on success
5907 * \retval 1 on buffer full
5909 #ifdef HAVE_FILLDIR_USE_CTX
5910 static int osd_ldiskfs_filldir(struct dir_context *buf,
5912 static int osd_ldiskfs_filldir(void *buf,
5914 const char *name, int namelen,
5915 loff_t offset, __u64 ino, unsigned d_type)
5917 struct osd_it_ea *it =
5918 ((struct osd_filldir_cbs *)buf)->it;
5919 struct osd_object *obj = it->oie_obj;
5920 struct osd_it_ea_dirent *ent = it->oie_dirent;
5921 struct lu_fid *fid = &ent->oied_fid;
5922 struct osd_fid_pack *rec;
5925 /* this should never happen */
5926 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5927 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5931 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5935 /* "." is just the object itself. */
5936 if (namelen == 1 && name[0] == '.') {
5937 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5938 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5939 rec = (struct osd_fid_pack*) (name + namelen + 1);
5940 if (osd_fid_unpack(fid, rec) != 0)
5945 d_type &= ~LDISKFS_DIRENT_LUFID;
5947 /* NOT export local root. */
5948 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5949 ino = obj->oo_inode->i_ino;
5950 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5953 ent->oied_ino = ino;
5954 ent->oied_off = offset;
5955 ent->oied_namelen = namelen;
5956 ent->oied_type = d_type;
5958 memcpy(ent->oied_name, name, namelen);
5960 it->oie_rd_dirent++;
5961 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5966 * Calls ->readdir() to load a directory entry at a time
5967 * and stored it in iterator's in-memory data structure.
5969 * \param di iterator's in memory structure
5971 * \retval 0 on success
5972 * \retval -ve on error
5973 * \retval +1 reach the end of entry
5975 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5976 const struct dt_it *di)
5978 struct osd_it_ea *it = (struct osd_it_ea *)di;
5979 struct osd_object *obj = it->oie_obj;
5980 struct inode *inode = obj->oo_inode;
5981 struct htree_lock *hlock = NULL;
5982 struct file *filp = &it->oie_file;
5984 struct osd_filldir_cbs buf = {
5985 #ifdef HAVE_DIR_CONTEXT
5986 .ctx.actor = osd_ldiskfs_filldir,
5992 it->oie_dirent = it->oie_buf;
5993 it->oie_rd_dirent = 0;
5995 if (obj->oo_hl_head != NULL) {
5996 hlock = osd_oti_get(env)->oti_hlock;
5997 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5998 inode, LDISKFS_HLOCK_READDIR);
6000 down_read(&obj->oo_ext_idx_sem);
6003 #ifdef HAVE_DIR_CONTEXT
6004 buf.ctx.pos = filp->f_pos;
6005 rc = inode->i_fop->iterate(filp, &buf.ctx);
6006 filp->f_pos = buf.ctx.pos;
6008 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6012 ldiskfs_htree_unlock(hlock);
6014 up_read(&obj->oo_ext_idx_sem);
6016 if (it->oie_rd_dirent == 0) {
6017 /*If it does not get any dirent, it means it has been reached
6018 *to the end of the dir */
6019 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6023 it->oie_dirent = it->oie_buf;
6024 it->oie_it_dirent = 1;
6031 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6032 * to load a directory entry at a time and stored it in
6033 * iterator's in-memory data structure.
6035 * \param di iterator's in memory structure
6037 * \retval +ve iterator reached to end
6038 * \retval 0 iterator not reached to end
6039 * \retval -ve on error
6041 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6043 struct osd_it_ea *it = (struct osd_it_ea *)di;
6048 if (it->oie_it_dirent < it->oie_rd_dirent) {
6050 (void *) it->oie_dirent +
6051 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6052 it->oie_dirent->oied_namelen);
6053 it->oie_it_dirent++;
6056 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6059 rc = osd_ldiskfs_it_fill(env, di);
6066 * Returns the key at current position from iterator's in memory structure.
6068 * \param di iterator's in memory structure
6070 * \retval key i.e. struct dt_key on success
6072 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6073 const struct dt_it *di)
6075 struct osd_it_ea *it = (struct osd_it_ea *)di;
6077 return (struct dt_key *)it->oie_dirent->oied_name;
6081 * Returns the key's size at current position from iterator's in memory structure.
6083 * \param di iterator's in memory structure
6085 * \retval key_size i.e. struct dt_key on success
6087 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6089 struct osd_it_ea *it = (struct osd_it_ea *)di;
6091 return it->oie_dirent->oied_namelen;
6094 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6096 if (LDISKFS_DIR_REC_LEN(de) >=
6097 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6104 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6105 unsigned blocksize, bool dotdot)
6108 return osd_dotdot_has_space(de);
6110 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6111 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6118 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6119 handle_t *jh, struct dentry *dentry,
6120 const struct lu_fid *fid, struct buffer_head *bh,
6121 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6124 struct inode *dir = dentry->d_parent->d_inode;
6125 struct inode *inode = dentry->d_inode;
6126 struct osd_fid_pack *rec;
6127 struct ldiskfs_dentry_param *ldp;
6128 int namelen = dentry->d_name.len;
6130 struct osd_thread_info *info = osd_oti_get(env);
6133 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6134 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6137 /* There is enough space to hold the FID-in-dirent. */
6138 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6139 rc = ldiskfs_journal_get_write_access(jh, bh);
6143 de->name[namelen] = 0;
6144 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6145 rec->fp_len = sizeof(struct lu_fid) + 1;
6146 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6147 de->file_type |= LDISKFS_DIRENT_LUFID;
6148 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6155 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6159 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6160 osd_get_ldiskfs_dirent_param(ldp, fid);
6161 dentry->d_fsdata = (void *)ldp;
6162 ll_vfs_dq_init(dir);
6163 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6164 /* It is too bad, we cannot reinsert the name entry back.
6165 * That means we lose it! */
6167 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6168 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6169 osd_ino2name(inode),
6170 dir->i_ino, dir->i_generation, namelen,
6171 dentry->d_name.name, PFID(fid), rc);
6177 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6178 struct osd_it_ea *it, struct lu_fid *fid,
6179 struct osd_inode_id *id, __u32 *attr)
6181 struct osd_thread_info *info = osd_oti_get(env);
6182 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6183 struct osd_device *dev = osd_obj2dev(obj);
6184 struct super_block *sb = osd_sb(dev);
6185 const char *devname = osd_name(dev);
6186 struct osd_it_ea_dirent *ent = it->oie_dirent;
6187 struct inode *dir = obj->oo_inode;
6188 struct htree_lock *hlock = NULL;
6189 struct buffer_head *bh = NULL;
6190 handle_t *jh = NULL;
6191 struct ldiskfs_dir_entry_2 *de;
6192 struct dentry *dentry;
6193 struct inode *inode;
6194 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6197 bool dotdot = false;
6201 if (ent->oied_name[0] == '.') {
6202 if (ent->oied_namelen == 1)
6205 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6209 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6210 inode = osd_iget(info, dev, id);
6211 if (IS_ERR(inode)) {
6212 rc = PTR_ERR(inode);
6213 if (rc == -ENOENT || rc == -ESTALE) {
6214 /* Maybe dangling name entry, or
6215 * corrupted directory entry. */
6216 *attr |= LUDA_UNKNOWN;
6219 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6220 "check_repair, dir = %lu/%u, name = %.*s, "
6221 "ino = %llu, rc = %d\n",
6222 devname, dir->i_ino, dir->i_generation,
6223 ent->oied_namelen, ent->oied_name,
6230 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6232 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6233 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6238 /* We need to ensure that the name entry is still valid.
6239 * Because it may be removed or renamed by other already.
6241 * The unlink or rename operation will start journal before PDO lock,
6242 * so to avoid deadlock, here we need to start journal handle before
6243 * related PDO lock also. But because we do not know whether there
6244 * will be something to be repaired before PDO lock, we just start
6245 * journal without conditions.
6247 * We may need to remove the name entry firstly, then insert back.
6248 * One credit is for user quota file update.
6249 * One credit is for group quota file update.
6250 * Two credits are for dirty inode. */
6251 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6252 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6254 if (dev->od_dirent_journal != 0) {
6257 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6260 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6261 "check_repair, dir = %lu/%u, credits = %d, "
6262 "name = %.*s, ino = %llu: rc = %d\n",
6263 devname, dir->i_ino, dir->i_generation, credits,
6264 ent->oied_namelen, ent->oied_name,
6267 GOTO(out_inode, rc);
6270 if (obj->oo_hl_head != NULL) {
6271 hlock = osd_oti_get(env)->oti_hlock;
6272 /* "0" means exclusive lock for the whole directory.
6273 * We need to prevent others access such name entry
6274 * during the delete + insert. Neither HLOCK_ADD nor
6275 * HLOCK_DEL cannot guarantee the atomicity. */
6276 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6278 down_write(&obj->oo_ext_idx_sem);
6281 if (obj->oo_hl_head != NULL) {
6282 hlock = osd_oti_get(env)->oti_hlock;
6283 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6284 LDISKFS_HLOCK_LOOKUP);
6286 down_read(&obj->oo_ext_idx_sem);
6290 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6291 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6292 *attr |= LUDA_IGNORE;
6297 /* For dotdot entry, if there is not enough space to hold the
6298 * FID-in-dirent, just keep them there. It only happens when the
6299 * device upgraded from 1.8 or restored from MDT file-level backup.
6300 * For the whole directory, only dotdot entry have no FID-in-dirent
6301 * and needs to get FID from LMA when readdir, it will not affect the
6302 * performance much. */
6303 if (dotdot && !osd_dotdot_has_space(de)) {
6304 *attr |= LUDA_UNKNOWN;
6310 if (lu_fid_eq(fid, &lma->lma_self_fid))
6313 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6314 struct lu_fid *tfid = &lma->lma_self_fid;
6316 if (likely(dotdot &&
6317 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6318 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6319 /* It must be REMOTE_PARENT_DIR and as the
6320 * 'dotdot' entry of remote directory */
6321 *attr |= LUDA_IGNORE;
6323 CDEBUG(D_LFSCK, "%s: expect remote agent "
6324 "parent directory, but got %.*s under "
6325 "dir = %lu/%u with the FID "DFID"\n",
6326 devname, ent->oied_namelen,
6327 ent->oied_name, dir->i_ino,
6328 dir->i_generation, PFID(tfid));
6330 *attr |= LUDA_UNKNOWN;
6337 if (!fid_is_zero(fid)) {
6338 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6340 if (rc == -ENOENT ||
6342 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6343 /* linkEA does not recognize the dirent entry,
6344 * it may because the dirent entry corruption
6345 * and points to other's inode. */
6346 CDEBUG(D_LFSCK, "%s: the target inode does not "
6347 "recognize the dirent, dir = %lu/%u, "
6348 " name = %.*s, ino = %llu, "
6349 DFID": rc = %d\n", devname, dir->i_ino,
6350 dir->i_generation, ent->oied_namelen,
6351 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6352 *attr |= LUDA_UNKNOWN;
6357 if (rc && rc != -ENODATA) {
6358 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6359 "dir = %lu/%u, name = %.*s, ino = %llu, "
6360 DFID": rc = %d\n", devname, dir->i_ino,
6361 dir->i_generation, ent->oied_namelen,
6362 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6363 *attr |= LUDA_UNKNOWN;
6370 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6371 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6372 if (*attr & LUDA_VERIFY_DRYRUN) {
6373 *fid = lma->lma_self_fid;
6374 *attr |= LUDA_REPAIR;
6381 dev->od_dirent_journal = 1;
6382 if (hlock != NULL) {
6383 ldiskfs_htree_unlock(hlock);
6386 up_read(&obj->oo_ext_idx_sem);
6392 *fid = lma->lma_self_fid;
6394 /* Update or append the FID-in-dirent. */
6395 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6396 bh, de, hlock, dotdot);
6398 *attr |= LUDA_REPAIR;
6400 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6401 "the dirent, dir = %lu/%u, name = %.*s, "
6402 "ino = %llu, "DFID": rc = %d\n",
6403 devname, dir->i_ino, dir->i_generation,
6404 ent->oied_namelen, ent->oied_name,
6405 ent->oied_ino, PFID(fid), rc);
6407 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6408 if (*attr & LUDA_VERIFY_DRYRUN) {
6409 if (fid_is_sane(fid)) {
6410 *attr |= LUDA_REPAIR;
6411 } else if (dev->od_index == 0) {
6412 lu_igif_build(fid, inode->i_ino,
6413 inode->i_generation);
6414 *attr |= LUDA_UPGRADE;
6422 dev->od_dirent_journal = 1;
6423 if (hlock != NULL) {
6424 ldiskfs_htree_unlock(hlock);
6427 up_read(&obj->oo_ext_idx_sem);
6434 if (unlikely(fid_is_sane(fid))) {
6435 /* FID-in-dirent exists, but FID-in-LMA is lost.
6436 * Trust the FID-in-dirent, and add FID-in-LMA. */
6437 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6439 *attr |= LUDA_REPAIR;
6441 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6442 "update dirent, dir = %lu/%u, "
6443 "name = %.*s, ino = %llu, "
6445 devname, dir->i_ino, dir->i_generation,
6446 ent->oied_namelen, ent->oied_name,
6447 ent->oied_ino, PFID(fid), rc);
6448 } else if (dev->od_index == 0) {
6449 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6450 /* It is probably IGIF object. Only aappend the
6451 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6452 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6453 bh, de, hlock, dotdot);
6455 *attr |= LUDA_UPGRADE;
6457 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6458 "after the dirent, dir = %lu/%u, "
6459 "name = %.*s, ino = %llu, "
6461 devname, dir->i_ino, dir->i_generation,
6462 ent->oied_namelen, ent->oied_name,
6463 ent->oied_ino, PFID(fid), rc);
6472 if (hlock != NULL) {
6473 ldiskfs_htree_unlock(hlock);
6475 if (dev->od_dirent_journal != 0)
6476 up_write(&obj->oo_ext_idx_sem);
6478 up_read(&obj->oo_ext_idx_sem);
6482 ldiskfs_journal_stop(jh);
6486 if (rc >= 0 && !dirty)
6487 dev->od_dirent_journal = 0;
6493 * Returns the value at current position from iterator's in memory structure.
6495 * \param di struct osd_it_ea, iterator's in memory structure
6496 * \param attr attr requested for dirent.
6497 * \param lde lustre dirent
6499 * \retval 0 no error and \param lde has correct lustre dirent.
6500 * \retval -ve on error
6502 static inline int osd_it_ea_rec(const struct lu_env *env,
6503 const struct dt_it *di,
6504 struct dt_rec *dtrec, __u32 attr)
6506 struct osd_it_ea *it = (struct osd_it_ea *)di;
6507 struct osd_object *obj = it->oie_obj;
6508 struct osd_device *dev = osd_obj2dev(obj);
6509 struct osd_thread_info *oti = osd_oti_get(env);
6510 struct osd_inode_id *id = &oti->oti_id;
6511 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6512 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6513 __u32 ino = it->oie_dirent->oied_ino;
6517 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6519 if (attr & LUDA_VERIFY) {
6520 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6521 attr |= LUDA_IGNORE;
6522 /* If the parent is on remote MDT, and there
6523 * is no FID-in-dirent, then we have to get
6524 * the parent FID from the linkEA. */
6525 if (!fid_is_sane(fid) &&
6526 it->oie_dirent->oied_namelen == 2 &&
6527 it->oie_dirent->oied_name[0] == '.' &&
6528 it->oie_dirent->oied_name[1] == '.')
6529 osd_get_pfid_from_linkea(env, obj, fid);
6531 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6535 if (!fid_is_sane(fid))
6536 attr |= LUDA_UNKNOWN;
6538 attr &= ~LU_DIRENT_ATTRS_MASK;
6539 if (!fid_is_sane(fid)) {
6540 bool is_dotdot = false;
6541 if (it->oie_dirent->oied_namelen == 2 &&
6542 it->oie_dirent->oied_name[0] == '.' &&
6543 it->oie_dirent->oied_name[1] == '.')
6545 /* If the parent is on remote MDT, and there
6546 * is no FID-in-dirent, then we have to get
6547 * the parent FID from the linkEA. */
6548 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6549 rc = osd_get_pfid_from_linkea(env, obj, fid);
6551 if (is_dotdot == false &&
6552 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6555 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6558 osd_id_gen(id, ino, OSD_OII_NOGEN);
6562 /* Pack the entry anyway, at least the offset is right. */
6563 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6564 it->oie_dirent->oied_name,
6565 it->oie_dirent->oied_namelen,
6566 it->oie_dirent->oied_type, attr);
6571 if (osd_remote_fid(env, dev, fid))
6574 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6575 osd_add_oi_cache(oti, dev, id, fid);
6577 RETURN(rc > 0 ? 0 : rc);
6581 * Returns the record size size at current position.
6583 * This function will return record(lu_dirent) size in bytes.
6585 * \param[in] env execution environment
6586 * \param[in] di iterator's in memory structure
6587 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6588 * calculate the lu_dirent size.
6590 * \retval record size(in bytes & in memory) of the current lu_dirent
6593 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6596 struct osd_it_ea *it = (struct osd_it_ea *)di;
6598 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6602 * Returns a cookie for current position of the iterator head, so that
6603 * user can use this cookie to load/start the iterator next time.
6605 * \param di iterator's in memory structure
6607 * \retval cookie for current position, on success
6609 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6611 struct osd_it_ea *it = (struct osd_it_ea *)di;
6613 return it->oie_dirent->oied_off;
6617 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6618 * to load a directory entry at a time and stored it i inn,
6619 * in iterator's in-memory data structure.
6621 * \param di struct osd_it_ea, iterator's in memory structure
6623 * \retval +ve on success
6624 * \retval -ve on error
6626 static int osd_it_ea_load(const struct lu_env *env,
6627 const struct dt_it *di, __u64 hash)
6629 struct osd_it_ea *it = (struct osd_it_ea *)di;
6633 it->oie_file.f_pos = hash;
6635 rc = osd_ldiskfs_it_fill(env, di);
6646 * Index lookup function for interoperability mode (b11826).
6648 * \param key, key i.e. file name to be searched
6650 * \retval +ve, on success
6651 * \retval -ve, on error
6653 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6654 struct dt_rec *rec, const struct dt_key *key)
6656 struct osd_object *obj = osd_dt_obj(dt);
6661 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6662 LINVRNT(osd_invariant(obj));
6664 rc = osd_ea_lookup_rec(env, obj, rec, key);
6671 * Index and Iterator operations for interoperability
6672 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6674 static const struct dt_index_operations osd_index_ea_ops = {
6675 .dio_lookup = osd_index_ea_lookup,
6676 .dio_declare_insert = osd_index_declare_ea_insert,
6677 .dio_insert = osd_index_ea_insert,
6678 .dio_declare_delete = osd_index_declare_ea_delete,
6679 .dio_delete = osd_index_ea_delete,
6681 .init = osd_it_ea_init,
6682 .fini = osd_it_ea_fini,
6683 .get = osd_it_ea_get,
6684 .put = osd_it_ea_put,
6685 .next = osd_it_ea_next,
6686 .key = osd_it_ea_key,
6687 .key_size = osd_it_ea_key_size,
6688 .rec = osd_it_ea_rec,
6689 .rec_size = osd_it_ea_rec_size,
6690 .store = osd_it_ea_store,
6691 .load = osd_it_ea_load
6695 static void *osd_key_init(const struct lu_context *ctx,
6696 struct lu_context_key *key)
6698 struct osd_thread_info *info;
6700 OBD_ALLOC_PTR(info);
6702 return ERR_PTR(-ENOMEM);
6704 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6705 if (info->oti_it_ea_buf == NULL)
6708 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6710 info->oti_hlock = ldiskfs_htree_lock_alloc();
6711 if (info->oti_hlock == NULL)
6717 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6720 return ERR_PTR(-ENOMEM);
6723 static void osd_key_fini(const struct lu_context *ctx,
6724 struct lu_context_key *key, void* data)
6726 struct osd_thread_info *info = data;
6727 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6728 struct osd_idmap_cache *idc = info->oti_ins_cache;
6730 if (info->oti_inode != NULL)
6732 if (info->oti_hlock != NULL)
6733 ldiskfs_htree_lock_free(info->oti_hlock);
6734 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6735 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6736 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6737 lu_buf_free(&info->oti_big_buf);
6739 LASSERT(info->oti_ins_cache_size > 0);
6740 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6741 info->oti_ins_cache = NULL;
6742 info->oti_ins_cache_size = 0;
6747 static void osd_key_exit(const struct lu_context *ctx,
6748 struct lu_context_key *key, void *data)
6750 struct osd_thread_info *info = data;
6752 LASSERT(info->oti_r_locks == 0);
6753 LASSERT(info->oti_w_locks == 0);
6754 LASSERT(info->oti_txns == 0);
6757 /* type constructor/destructor: osd_type_init, osd_type_fini */
6758 LU_TYPE_INIT_FINI(osd, &osd_key);
6760 struct lu_context_key osd_key = {
6761 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6762 .lct_init = osd_key_init,
6763 .lct_fini = osd_key_fini,
6764 .lct_exit = osd_key_exit
6768 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6769 const char *name, struct lu_device *next)
6771 struct osd_device *osd = osd_dev(d);
6773 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6774 >= sizeof(osd->od_svname))
6776 return osd_procfs_init(osd, name);
6779 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6781 struct seq_server_site *ss = osd_seq_site(osd);
6785 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6788 if (unlikely(ss == NULL))
6791 OBD_ALLOC_PTR(osd->od_cl_seq);
6792 if (osd->od_cl_seq == NULL)
6795 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6796 osd->od_svname, ss->ss_server_seq);
6798 OBD_FREE_PTR(osd->od_cl_seq);
6799 osd->od_cl_seq = NULL;
6803 if (ss->ss_node_id == 0) {
6804 /* If the OSD on the sequence controller(MDT0), then allocate
6805 * sequence here, otherwise allocate sequence after connected
6806 * to MDT0 (see mdt_register_lwp_callback()). */
6807 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6808 &osd->od_cl_seq->lcs_space, env);
6814 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6816 if (osd->od_cl_seq == NULL)
6819 seq_client_fini(osd->od_cl_seq);
6820 OBD_FREE_PTR(osd->od_cl_seq);
6821 osd->od_cl_seq = NULL;
6824 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6828 /* shutdown quota slave instance associated with the device */
6829 if (o->od_quota_slave != NULL) {
6830 qsd_fini(env, o->od_quota_slave);
6831 o->od_quota_slave = NULL;
6834 osd_fid_fini(env, o);
6839 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6843 if (o->od_mnt != NULL) {
6844 shrink_dcache_sb(osd_sb(o));
6845 osd_sync(env, &o->od_dt_dev);
6854 static int osd_mount(const struct lu_env *env,
6855 struct osd_device *o, struct lustre_cfg *cfg)
6857 const char *name = lustre_cfg_string(cfg, 0);
6858 const char *dev = lustre_cfg_string(cfg, 1);
6860 unsigned long page, s_flags, lmd_flags = 0;
6861 struct page *__page;
6862 struct file_system_type *type;
6863 char *options = NULL;
6865 struct osd_thread_info *info = osd_oti_get(env);
6866 struct lu_fid *fid = &info->oti_fid;
6867 struct inode *inode;
6868 int rc = 0, force_over_512tb = 0;
6871 if (o->od_mnt != NULL)
6874 if (strlen(dev) >= sizeof(o->od_mntdev))
6876 strcpy(o->od_mntdev, dev);
6878 str = lustre_cfg_string(cfg, 2);
6879 s_flags = simple_strtoul(str, NULL, 0);
6880 str = strstr(str, ":");
6882 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6883 opts = lustre_cfg_string(cfg, 3);
6885 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6886 CERROR("%s: device %s extents feature is not guaranteed to "
6887 "work on big-endian systems. Use \"bigendian_extents\" "
6888 "mount option to override.\n", name, dev);
6892 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6893 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6894 CWARN("force_over_128tb option is deprecated. "
6895 "Filesystems less than 512TB can be created without any "
6896 "force options. Use force_over_512tb option for "
6897 "filesystems greater than 512TB.\n");
6900 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6901 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6902 CWARN("force_over_256tb option is deprecated. "
6903 "Filesystems less than 512TB can be created without any "
6904 "force options. Use force_over_512tb option for "
6905 "filesystems greater than 512TB.\n");
6909 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6910 force_over_512tb = 1;
6912 __page = alloc_page(GFP_KERNEL);
6914 GOTO(out, rc = -ENOMEM);
6915 page = (unsigned long)page_address(__page);
6916 options = (char *)page;
6919 /* strip out the options for back compatiblity */
6920 static char *sout[] = {
6927 /* strip out option we processed in osd */
6928 "bigendian_extents",
6934 strcat(options, opts);
6935 for (rc = 0, str = options; sout[rc]; ) {
6936 char *op = strstr(str, sout[rc]);
6942 if (op == options || *(op - 1) == ',') {
6943 str = op + strlen(sout[rc]);
6944 if (*str == ',' || *str == '\0') {
6945 *str == ',' ? str++ : str;
6946 memmove(op, str, strlen(str) + 1);
6949 for (str = op; *str != ',' && *str != '\0'; str++)
6953 strncat(options, "user_xattr,acl", 14);
6956 /* Glom up mount options */
6957 if (*options != '\0')
6958 strcat(options, ",");
6959 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6961 type = get_fs_type("ldiskfs");
6963 CERROR("%s: cannot find ldiskfs module\n", name);
6964 GOTO(out, rc = -ENODEV);
6967 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6968 module_put(type->owner);
6970 if (IS_ERR(o->od_mnt)) {
6971 rc = PTR_ERR(o->od_mnt);
6973 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6977 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
6978 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
6979 force_over_512tb == 0) {
6980 CERROR("%s: device %s LDISKFS does not support filesystems "
6981 "greater than 512TB and can cause data corruption. "
6982 "Use \"force_over_512tb\" mount option to override.\n",
6984 GOTO(out_mnt, rc = -EINVAL);
6987 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
6988 if (priv_dev_set_rdonly) {
6989 priv_dev_set_rdonly(osd_sb(o)->s_bdev);
6990 o->od_dt_dev.dd_rdonly = 1;
6991 LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
6994 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
6997 GOTO(out_mnt, rc = -EOPNOTSUPP);
6999 } else if (priv_dev_check_rdonly &&
7000 priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7001 CERROR("%s: underlying device %s is marked as "
7002 "read-only. Setup failed\n", name, dev);
7004 GOTO(out_mnt, rc = -EROFS);
7007 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7008 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7009 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7010 GOTO(out_mnt, rc = -EINVAL);
7013 #ifdef LDISKFS_MOUNT_DIRDATA
7014 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7015 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7016 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7017 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7018 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7019 "enabling the dirdata feature. If you do not want to "
7020 "downgrade to Lustre-1.x again, you can enable it via "
7021 "'tune2fs -O dirdata device'\n", name, dev);
7023 inode = osd_sb(o)->s_root->d_inode;
7024 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7025 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7027 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7031 if (lmd_flags & LMD_FLG_NOSCRUB)
7042 __free_page(__page);
7047 static struct lu_device *osd_device_fini(const struct lu_env *env,
7048 struct lu_device *d)
7050 struct osd_device *o = osd_dev(d);
7053 osd_shutdown(env, o);
7055 osd_scrub_cleanup(env, o);
7056 osd_obj_map_fini(o);
7062 static int osd_device_init0(const struct lu_env *env,
7063 struct osd_device *o,
7064 struct lustre_cfg *cfg)
7066 struct lu_device *l = osd2lu_dev(o);
7067 struct osd_thread_info *info;
7071 /* if the module was re-loaded, env can loose its keys */
7072 rc = lu_env_refill((struct lu_env *) env);
7075 info = osd_oti_get(env);
7078 l->ld_ops = &osd_lu_ops;
7079 o->od_dt_dev.dd_ops = &osd_dt_ops;
7081 spin_lock_init(&o->od_osfs_lock);
7082 mutex_init(&o->od_otable_mutex);
7083 INIT_LIST_HEAD(&o->od_orphan_list);
7085 o->od_read_cache = 1;
7086 o->od_writethrough_cache = 1;
7087 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7089 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7090 sizeof(o->od_svname));
7091 if (cplen >= sizeof(o->od_svname)) {
7096 o->od_index = -1; /* -1 means index is invalid */
7097 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7098 if (rc == LDD_F_SV_TYPE_OST)
7101 o->od_full_scrub_ratio = OFSR_DEFAULT;
7102 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7103 rc = osd_mount(env, o, cfg);
7107 rc = osd_obj_map_init(env, o);
7111 rc = lu_site_init(&o->od_site, l);
7113 GOTO(out_compat, rc);
7114 o->od_site.ls_bottom_dev = l;
7116 rc = lu_site_init_finish(&o->od_site);
7120 INIT_LIST_HEAD(&o->od_ios_list);
7121 /* setup scrub, including OI files initialization */
7122 rc = osd_scrub_setup(env, o);
7126 rc = osd_procfs_init(o, o->od_svname);
7128 CERROR("%s: can't initialize procfs: rc = %d\n",
7130 GOTO(out_scrub, rc);
7133 LASSERT(l->ld_site->ls_linkage.next != NULL);
7134 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7136 /* initialize quota slave instance */
7137 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7139 if (IS_ERR(o->od_quota_slave)) {
7140 rc = PTR_ERR(o->od_quota_slave);
7141 o->od_quota_slave = NULL;
7142 GOTO(out_procfs, rc);
7150 osd_scrub_cleanup(env, o);
7152 lu_site_fini(&o->od_site);
7154 osd_obj_map_fini(o);
7161 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7162 struct lu_device_type *t,
7163 struct lustre_cfg *cfg)
7165 struct osd_device *o;
7170 return ERR_PTR(-ENOMEM);
7172 rc = dt_device_init(&o->od_dt_dev, t);
7174 /* Because the ctx might be revived in dt_device_init,
7175 * refill the env here */
7176 lu_env_refill((struct lu_env *)env);
7177 rc = osd_device_init0(env, o, cfg);
7179 dt_device_fini(&o->od_dt_dev);
7182 if (unlikely(rc != 0))
7185 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7188 static struct lu_device *osd_device_free(const struct lu_env *env,
7189 struct lu_device *d)
7191 struct osd_device *o = osd_dev(d);
7194 /* XXX: make osd top device in order to release reference */
7195 d->ld_site->ls_top_dev = d;
7196 lu_site_purge(env, d->ld_site, -1);
7197 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7198 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7199 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7201 lu_site_fini(&o->od_site);
7202 dt_device_fini(&o->od_dt_dev);
7207 static int osd_process_config(const struct lu_env *env,
7208 struct lu_device *d, struct lustre_cfg *cfg)
7210 struct osd_device *o = osd_dev(d);
7214 switch (cfg->lcfg_command) {
7216 rc = osd_mount(env, o, cfg);
7219 lu_dev_del_linkage(d->ld_site, d);
7220 rc = osd_shutdown(env, o);
7223 LASSERT(&o->od_dt_dev);
7224 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7225 cfg, &o->od_dt_dev);
7226 if (rc > 0 || rc == -ENOSYS) {
7227 rc = class_process_proc_param(PARAM_OST,
7228 lprocfs_osd_obd_vars,
7229 cfg, &o->od_dt_dev);
7241 static int osd_recovery_complete(const struct lu_env *env,
7242 struct lu_device *d)
7244 struct osd_device *osd = osd_dev(d);
7248 if (osd->od_quota_slave == NULL)
7251 /* start qsd instance on recovery completion, this notifies the quota
7252 * slave code that we are about to process new requests now */
7253 rc = qsd_start(env, osd->od_quota_slave);
7258 * we use exports to track all osd users
7260 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7261 struct obd_device *obd, struct obd_uuid *cluuid,
7262 struct obd_connect_data *data, void *localdata)
7264 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7265 struct lustre_handle conn;
7269 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7271 rc = class_connect(&conn, obd, cluuid);
7275 *exp = class_conn2export(&conn);
7277 spin_lock(&osd->od_osfs_lock);
7279 spin_unlock(&osd->od_osfs_lock);
7285 * once last export (we don't count self-export) disappeared
7286 * osd can be released
7288 static int osd_obd_disconnect(struct obd_export *exp)
7290 struct obd_device *obd = exp->exp_obd;
7291 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7292 int rc, release = 0;
7295 /* Only disconnect the underlying layers on the final disconnect. */
7296 spin_lock(&osd->od_osfs_lock);
7298 if (osd->od_connects == 0)
7300 spin_unlock(&osd->od_osfs_lock);
7302 rc = class_disconnect(exp); /* bz 9811 */
7304 if (rc == 0 && release)
7305 class_manual_cleanup(obd);
7309 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7310 struct lu_device *dev)
7312 struct osd_device *osd = osd_dev(dev);
7313 struct lr_server_data *lsd =
7314 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7318 if (osd->od_quota_slave != NULL) {
7319 /* set up quota slave objects */
7320 result = qsd_prepare(env, osd->od_quota_slave);
7325 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7326 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7327 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7328 osd->od_index_in_idif = 1;
7330 osd->od_index_in_idif = 0;
7331 result = osd_register_proc_index_in_idif(osd);
7336 osd->od_index_in_idif = 1;
7340 result = osd_fid_init(env, osd);
7345 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7346 struct lu_fid *fid, struct md_op_data *op_data)
7348 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7350 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7353 static const struct lu_object_operations osd_lu_obj_ops = {
7354 .loo_object_init = osd_object_init,
7355 .loo_object_delete = osd_object_delete,
7356 .loo_object_release = osd_object_release,
7357 .loo_object_free = osd_object_free,
7358 .loo_object_print = osd_object_print,
7359 .loo_object_invariant = osd_object_invariant
7362 const struct lu_device_operations osd_lu_ops = {
7363 .ldo_object_alloc = osd_object_alloc,
7364 .ldo_process_config = osd_process_config,
7365 .ldo_recovery_complete = osd_recovery_complete,
7366 .ldo_prepare = osd_prepare,
7369 static const struct lu_device_type_operations osd_device_type_ops = {
7370 .ldto_init = osd_type_init,
7371 .ldto_fini = osd_type_fini,
7373 .ldto_start = osd_type_start,
7374 .ldto_stop = osd_type_stop,
7376 .ldto_device_alloc = osd_device_alloc,
7377 .ldto_device_free = osd_device_free,
7379 .ldto_device_init = osd_device_init,
7380 .ldto_device_fini = osd_device_fini
7383 static struct lu_device_type osd_device_type = {
7384 .ldt_tags = LU_DEVICE_DT,
7385 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7386 .ldt_ops = &osd_device_type_ops,
7387 .ldt_ctx_tags = LCT_LOCAL,
7390 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7392 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7393 struct super_block *sb = osd_sb(osd);
7395 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7399 * lprocfs legacy support.
7401 static struct obd_ops osd_obd_device_ops = {
7402 .o_owner = THIS_MODULE,
7403 .o_connect = osd_obd_connect,
7404 .o_disconnect = osd_obd_disconnect,
7405 .o_fid_alloc = osd_fid_alloc,
7406 .o_health_check = osd_health_check,
7409 static int __init osd_init(void)
7413 CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7414 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7415 /* please, try to keep osd_thread_info smaller than a page */
7416 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7421 rc = lu_kmem_init(ldiskfs_caches);
7425 #ifdef CONFIG_KALLSYMS
7426 priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7427 priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7428 /* Clear readonly is unused at this time */
7429 /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7432 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7433 lprocfs_osd_module_vars,
7434 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7436 lu_kmem_fini(ldiskfs_caches);
7440 static void __exit osd_exit(void)
7442 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7443 lu_kmem_fini(ldiskfs_caches);
7446 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7447 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7448 MODULE_VERSION(LUSTRE_VERSION_STRING);
7449 MODULE_LICENSE("GPL");
7451 module_init(osd_init);
7452 module_exit(osd_exit);