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);
1593 lu_context_exit(&th->th_ctx);
1594 lu_context_fini(&th->th_ctx);
1598 #ifndef HAVE_SB_START_WRITE
1599 # define sb_start_write(sb) do {} while (0)
1600 # define sb_end_write(sb) do {} while (0)
1603 static struct thandle *osd_trans_create(const struct lu_env *env,
1604 struct dt_device *d)
1606 struct osd_thread_info *oti = osd_oti_get(env);
1607 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1608 struct osd_thandle *oh;
1613 CERROR("%s: someone try to start transaction under "
1614 "readonly mode, should be disabled.\n",
1615 osd_name(osd_dt_dev(d)));
1617 RETURN(ERR_PTR(-EROFS));
1620 /* on pending IO in this thread should left from prev. request */
1621 LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1623 sb_start_write(osd_sb(osd_dt_dev(d)));
1625 OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1627 oh->ot_quota_trans = &oti->oti_quota_trans;
1628 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1632 th->th_tags = LCT_TX_HANDLE;
1634 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1635 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1638 memset(oti->oti_declare_ops, 0,
1639 sizeof(oti->oti_declare_ops));
1640 memset(oti->oti_declare_ops_cred, 0,
1641 sizeof(oti->oti_declare_ops_cred));
1642 memset(oti->oti_declare_ops_used, 0,
1643 sizeof(oti->oti_declare_ops_used));
1645 sb_end_write(osd_sb(osd_dt_dev(d)));
1646 th = ERR_PTR(-ENOMEM);
1651 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1653 struct osd_thread_info *oti = osd_oti_get(env);
1654 struct osd_thandle *oh;
1656 oh = container_of0(th, struct osd_thandle, ot_super);
1657 LASSERT(oh != NULL);
1659 CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n",
1660 oti->oti_declare_ops[OSD_OT_CREATE],
1661 oti->oti_declare_ops_cred[OSD_OT_CREATE],
1662 oti->oti_declare_ops_used[OSD_OT_CREATE],
1663 oti->oti_declare_ops[OSD_OT_DESTROY],
1664 oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1665 oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1666 CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1667 oti->oti_declare_ops[OSD_OT_ATTR_SET],
1668 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1669 oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1670 oti->oti_declare_ops[OSD_OT_XATTR_SET],
1671 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1672 oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1673 CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1674 oti->oti_declare_ops[OSD_OT_WRITE],
1675 oti->oti_declare_ops_cred[OSD_OT_WRITE],
1676 oti->oti_declare_ops_used[OSD_OT_WRITE],
1677 oti->oti_declare_ops[OSD_OT_PUNCH],
1678 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1679 oti->oti_declare_ops_used[OSD_OT_PUNCH],
1680 oti->oti_declare_ops[OSD_OT_QUOTA],
1681 oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1682 oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1683 CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n",
1684 oti->oti_declare_ops[OSD_OT_INSERT],
1685 oti->oti_declare_ops_cred[OSD_OT_INSERT],
1686 oti->oti_declare_ops_used[OSD_OT_INSERT],
1687 oti->oti_declare_ops[OSD_OT_DELETE],
1688 oti->oti_declare_ops_cred[OSD_OT_DELETE],
1689 oti->oti_declare_ops_used[OSD_OT_DELETE]);
1690 CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1691 oti->oti_declare_ops[OSD_OT_REF_ADD],
1692 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1693 oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1694 oti->oti_declare_ops[OSD_OT_REF_DEL],
1695 oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1696 oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1700 * Concurrency: shouldn't matter.
1702 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1705 struct osd_thread_info *oti = osd_oti_get(env);
1706 struct osd_device *dev = osd_dt_dev(d);
1708 struct osd_thandle *oh;
1713 LASSERT(current->journal_info == NULL);
1715 oh = container_of0(th, struct osd_thandle, ot_super);
1716 LASSERT(oh != NULL);
1717 LASSERT(oh->ot_handle == NULL);
1719 rc = dt_txn_hook_start(env, d, th);
1723 if (unlikely(osd_param_is_not_sane(dev, th))) {
1724 static unsigned long last_printed;
1725 static int last_credits;
1727 /* don't make noise on a tiny testing systems
1728 * actual credits misuse will be caught anyway */
1729 if (last_credits != oh->ot_credits &&
1730 time_after(jiffies, last_printed +
1731 msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1732 osd_transaction_size(dev) > 512) {
1733 CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1734 oh->ot_credits, osd_transaction_size(dev));
1735 osd_trans_dump_creds(env, th);
1736 libcfs_debug_dumpstack(NULL);
1737 last_credits = oh->ot_credits;
1738 last_printed = jiffies;
1740 /* XXX Limit the credits to 'max_transaction_buffers', and
1741 * let the underlying filesystem to catch the error if
1742 * we really need so many credits.
1744 * This should be removed when we can calculate the
1745 * credits precisely. */
1746 oh->ot_credits = osd_transaction_size(dev);
1747 } else if (ldiskfs_track_declares_assert != 0) {
1748 /* reserve few credits to prevent an assertion in JBD
1749 * our debugging mechanism will be able to detected
1750 * overuse. this can help to debug single-update
1752 oh->ot_credits += 10;
1753 if (unlikely(osd_param_is_not_sane(dev, th)))
1754 oh->ot_credits = osd_transaction_size(dev);
1758 * XXX temporary stuff. Some abstraction layer should
1761 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1765 LASSERT(oti->oti_txns == 0);
1766 lu_context_init(&th->th_ctx, th->th_tags);
1767 lu_context_enter(&th->th_ctx);
1769 lu_device_get(&d->dd_lu_dev);
1770 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1781 static int osd_seq_exists(const struct lu_env *env,
1782 struct osd_device *osd, u64 seq)
1784 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1785 struct seq_server_site *ss = osd_seq_site(osd);
1789 LASSERT(ss != NULL);
1790 LASSERT(ss->ss_server_fld != NULL);
1792 rc = osd_fld_lookup(env, osd, seq, range);
1795 CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1796 osd_name(osd), seq, rc);
1800 RETURN(ss->ss_node_id == range->lsr_index);
1803 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1805 struct dt_txn_commit_cb *dcb;
1806 struct dt_txn_commit_cb *tmp;
1808 /* call per-transaction stop callbacks if any */
1809 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1811 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1812 "commit callback entry: magic=%x name='%s'\n",
1813 dcb->dcb_magic, dcb->dcb_name);
1814 list_del_init(&dcb->dcb_linkage);
1815 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1820 * Concurrency: shouldn't matter.
1822 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1825 int rc = 0, remove_agents = 0;
1826 struct osd_thandle *oh;
1827 struct osd_thread_info *oti = osd_oti_get(env);
1828 struct osd_iobuf *iobuf = &oti->oti_iobuf;
1829 struct osd_device *osd = osd_dt_dev(th->th_dev);
1830 struct qsd_instance *qsd = osd->od_quota_slave;
1831 struct lquota_trans *qtrans;
1834 oh = container_of0(th, struct osd_thandle, ot_super);
1836 /* reset OI cache for safety */
1837 oti->oti_ins_cache_used = 0;
1839 remove_agents = oh->ot_remove_agents;
1841 qtrans = oh->ot_quota_trans;
1842 oh->ot_quota_trans = NULL;
1844 if (oh->ot_handle != NULL) {
1846 handle_t *hdl = oh->ot_handle;
1849 * add commit callback
1850 * notice we don't do this in osd_trans_start()
1851 * as underlying transaction can change during truncate
1853 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1856 LASSERT(oti->oti_txns == 1);
1859 rc = dt_txn_hook_stop(env, th);
1861 CERROR("%s: failed in transaction hook: rc = %d\n",
1864 osd_trans_stop_cb(oh, rc);
1865 /* hook functions might modify th_sync */
1866 hdl->h_sync = th->th_sync;
1868 oh->ot_handle = NULL;
1869 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1871 CERROR("%s: failed to stop transaction: rc = %d\n",
1872 osd_name(osd), rc2);
1876 osd_trans_stop_cb(oh, th->th_result);
1880 /* inform the quota slave device that the transaction is stopping */
1881 qsd_op_end(env, qsd, qtrans);
1883 /* as we want IO to journal and data IO be concurrent, we don't block
1884 * awaiting data IO completion in osd_do_bio(), instead we wait here
1885 * once transaction is submitted to the journal. all reqular requests
1886 * don't do direct IO (except read/write), thus this wait_event becomes
1889 * IMPORTANT: we have to wait till any IO submited by the thread is
1890 * completed otherwise iobuf may be corrupted by different request
1892 wait_event(iobuf->dr_wait,
1893 atomic_read(&iobuf->dr_numreqs) == 0);
1894 osd_fini_iobuf(osd, iobuf);
1896 rc = iobuf->dr_error;
1898 if (unlikely(remove_agents != 0))
1899 osd_process_scheduled_agent_removals(env, osd);
1901 sb_end_write(osd_sb(osd));
1906 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1908 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1911 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1912 LASSERT(&dcb->dcb_func != NULL);
1913 if (dcb->dcb_flags & DCB_TRANS_STOP)
1914 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1916 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1922 * Called just before object is freed. Releases all resources except for
1923 * object itself (that is released by osd_object_free()).
1925 * Concurrency: no concurrent access is possible that late in object
1928 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1930 struct osd_object *obj = osd_obj(l);
1931 struct inode *inode = obj->oo_inode;
1933 LINVRNT(osd_invariant(obj));
1936 * If object is unlinked remove fid->ino mapping from object index.
1939 osd_index_fini(obj);
1940 if (inode != NULL) {
1941 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
1942 qid_t uid = i_uid_read(inode);
1943 qid_t gid = i_gid_read(inode);
1946 obj->oo_inode = NULL;
1949 struct osd_thread_info *info = osd_oti_get(env);
1950 struct lquota_id_info *qi = &info->oti_qi;
1952 /* Release granted quota to master if necessary */
1953 qi->lqi_id.qid_uid = uid;
1954 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1956 qi->lqi_id.qid_uid = gid;
1957 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1959 qi->lqi_id.qid_uid = i_projid_read(inode);
1960 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1966 * Concurrency: ->loo_object_release() is called under site spin-lock.
1968 static void osd_object_release(const struct lu_env *env,
1969 struct lu_object *l)
1971 struct osd_object *o = osd_obj(l);
1972 /* nobody should be releasing a non-destroyed object with nlink=0
1973 * the API allows this, but ldiskfs doesn't like and then report
1974 * this inode as deleted */
1975 if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1980 * Concurrency: shouldn't matter.
1982 static int osd_object_print(const struct lu_env *env, void *cookie,
1983 lu_printer_t p, const struct lu_object *l)
1985 struct osd_object *o = osd_obj(l);
1986 struct iam_descr *d;
1988 if (o->oo_dir != NULL)
1989 d = o->oo_dir->od_container.ic_descr;
1992 return (*p)(env, cookie,
1993 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1995 o->oo_inode ? o->oo_inode->i_ino : 0UL,
1996 o->oo_inode ? o->oo_inode->i_generation : 0,
1997 d ? d->id_ops->id_name : "plain");
2001 * Concurrency: shouldn't matter.
2003 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2004 struct obd_statfs *sfs)
2006 struct osd_device *osd = osd_dt_dev(d);
2007 struct super_block *sb = osd_sb(osd);
2008 struct kstatfs *ksfs;
2012 if (unlikely(osd->od_mnt == NULL))
2013 return -EINPROGRESS;
2015 /* osd_lproc.c call this without env, allocate ksfs for that case */
2016 if (unlikely(env == NULL)) {
2017 OBD_ALLOC_PTR(ksfs);
2021 ksfs = &osd_oti_get(env)->oti_ksfs;
2024 result = sb->s_op->statfs(sb->s_root, ksfs);
2028 statfs_pack(sfs, ksfs);
2029 if (unlikely(sb->s_flags & MS_RDONLY))
2030 sfs->os_state |= OS_STATE_READONLY;
2031 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2032 LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2033 sfs->os_maxbytes = sb->s_maxbytes;
2035 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2038 * Reserve some space so to avoid fragmenting the filesystem too much.
2039 * Fragmentation not only impacts performance, but can also increase
2040 * metadata overhead significantly, causing grant calculation to be
2043 * Reserve 0.78% of total space, at least 8MB for small filesystems.
2045 CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2046 reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2047 if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2048 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2050 sfs->os_blocks -= reserved;
2051 sfs->os_bfree -= min(reserved, sfs->os_bfree);
2052 sfs->os_bavail -= min(reserved, sfs->os_bavail);
2055 if (unlikely(env == NULL))
2061 * Estimate space needed for file creations. We assume the largest filename
2062 * which is 2^64 - 1, hence a filename of 20 chars.
2063 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2065 #ifdef __LDISKFS_DIR_REC_LEN
2066 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2068 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2072 * Concurrency: doesn't access mutable data.
2074 static void osd_conf_get(const struct lu_env *env,
2075 const struct dt_device *dev,
2076 struct dt_device_param *param)
2078 struct super_block *sb = osd_sb(osd_dt_dev(dev));
2082 * XXX should be taken from not-yet-existing fs abstraction layer.
2084 param->ddp_max_name_len = LDISKFS_NAME_LEN;
2085 param->ddp_max_nlink = LDISKFS_LINK_MAX;
2086 param->ddp_symlink_max = sb->s_blocksize;
2087 param->ddp_mount_type = LDD_MT_LDISKFS;
2088 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2089 param->ddp_maxbytes = sb->s_maxbytes;
2091 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2092 /* inode are statically allocated, so per-inode space consumption
2093 * is the space consumed by the directory entry */
2094 param->ddp_inodespace = PER_OBJ_USAGE;
2095 /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2096 * = 128MB) which is unlikely to be hit in real life. Report a smaller
2097 * maximum length to not under count the actual number of extents
2098 * needed for writing a file. */
2099 param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2100 /* worst-case extent insertion metadata overhead */
2101 param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2102 param->ddp_mntopts = 0;
2103 if (test_opt(sb, XATTR_USER))
2104 param->ddp_mntopts |= MNTOPT_USERXATTR;
2105 if (test_opt(sb, POSIX_ACL))
2106 param->ddp_mntopts |= MNTOPT_ACL;
2108 /* LOD might calculate the max stripe count based on max_ea_size,
2109 * so we need take account in the overhead as well,
2110 * xattr_header + magic + xattr_entry_head */
2111 ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2112 LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2114 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2115 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2116 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2120 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2124 * Concurrency: shouldn't matter.
2126 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2130 CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2132 rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2134 CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2141 * Start commit for OSD device.
2143 * An implementation of dt_commit_async method for OSD device.
2144 * Asychronously starts underlayng fs sync and thereby a transaction
2147 * \param env environment
2148 * \param d dt device
2150 * \see dt_device_operations
2152 static int osd_commit_async(const struct lu_env *env,
2153 struct dt_device *d)
2155 struct super_block *s = osd_sb(osd_dt_dev(d));
2158 CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2159 RETURN(s->s_op->sync_fs(s, 0));
2162 /* Our own copy of the set readonly functions if present, or NU if not. */
2163 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2164 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2165 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2168 * Concurrency: shouldn't matter.
2170 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2172 struct super_block *sb = osd_sb(osd_dt_dev(d));
2173 struct block_device *dev = sb->s_bdev;
2174 int rc = -EOPNOTSUPP;
2177 if (priv_dev_set_rdonly) {
2178 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2181 CERROR("*** setting %s read-only ***\n",
2182 osd_dt_dev(d)->od_svname);
2184 if (sb->s_op->freeze_fs) {
2185 rc = sb->s_op->freeze_fs(sb);
2190 if (jdev && (jdev != dev)) {
2191 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2193 priv_dev_set_rdonly(jdev);
2195 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2196 priv_dev_set_rdonly(dev);
2198 if (sb->s_op->unfreeze_fs)
2199 sb->s_op->unfreeze_fs(sb);
2204 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2205 osd_dt_dev(d)->od_svname, (long)dev, rc);
2211 * Note: we do not count into QUOTA here.
2212 * If we mount with --data_journal we may need more.
2214 const int osd_dto_credits_noquota[DTO_NR] = {
2217 * INDEX_EXTRA_TRANS_BLOCKS(8) +
2218 * SINGLEDATA_TRANS_BLOCKS(8)
2219 * XXX Note: maybe iam need more, since iam have more level than
2222 [DTO_INDEX_INSERT] = 16,
2225 * just modify a single entry, probably merge few within a block
2227 [DTO_INDEX_DELETE] = 1,
2231 [DTO_INDEX_UPDATE] = 16,
2233 * 4(inode, inode bits, groups, GDT)
2234 * notice: OI updates are counted separately with DTO_INDEX_INSERT
2236 [DTO_OBJECT_CREATE] = 4,
2238 * 4(inode, inode bits, groups, GDT)
2239 * notice: OI updates are counted separately with DTO_INDEX_DELETE
2241 [DTO_OBJECT_DELETE] = 4,
2243 * Attr set credits (inode)
2245 [DTO_ATTR_SET_BASE] = 1,
2247 * Xattr set. The same as xattr of EXT3.
2248 * DATA_TRANS_BLOCKS(14)
2249 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2250 * are also counted in. Do not know why?
2252 [DTO_XATTR_SET] = 14,
2254 * credits for inode change during write.
2256 [DTO_WRITE_BASE] = 3,
2258 * credits for single block write.
2260 [DTO_WRITE_BLOCK] = 14,
2262 * Attr set credits for chown.
2263 * This is extra credits for setattr, and it is null without quota
2265 [DTO_ATTR_SET_CHOWN] = 0
2268 static const struct dt_device_operations osd_dt_ops = {
2269 .dt_root_get = osd_root_get,
2270 .dt_statfs = osd_statfs,
2271 .dt_trans_create = osd_trans_create,
2272 .dt_trans_start = osd_trans_start,
2273 .dt_trans_stop = osd_trans_stop,
2274 .dt_trans_cb_add = osd_trans_cb_add,
2275 .dt_conf_get = osd_conf_get,
2276 .dt_sync = osd_sync,
2278 .dt_commit_async = osd_commit_async,
2281 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2284 struct osd_object *obj = osd_dt_obj(dt);
2285 struct osd_thread_info *oti = osd_oti_get(env);
2287 LINVRNT(osd_invariant(obj));
2289 LASSERT(obj->oo_owner != env);
2290 down_read_nested(&obj->oo_sem, role);
2292 LASSERT(obj->oo_owner == NULL);
2296 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2299 struct osd_object *obj = osd_dt_obj(dt);
2300 struct osd_thread_info *oti = osd_oti_get(env);
2302 LINVRNT(osd_invariant(obj));
2304 LASSERT(obj->oo_owner != env);
2305 down_write_nested(&obj->oo_sem, role);
2307 LASSERT(obj->oo_owner == NULL);
2308 obj->oo_owner = env;
2312 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2314 struct osd_object *obj = osd_dt_obj(dt);
2315 struct osd_thread_info *oti = osd_oti_get(env);
2317 LINVRNT(osd_invariant(obj));
2319 LASSERT(oti->oti_r_locks > 0);
2321 up_read(&obj->oo_sem);
2324 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2326 struct osd_object *obj = osd_dt_obj(dt);
2327 struct osd_thread_info *oti = osd_oti_get(env);
2329 LINVRNT(osd_invariant(obj));
2331 LASSERT(obj->oo_owner == env);
2332 LASSERT(oti->oti_w_locks > 0);
2334 obj->oo_owner = NULL;
2335 up_write(&obj->oo_sem);
2338 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2340 struct osd_object *obj = osd_dt_obj(dt);
2342 LINVRNT(osd_invariant(obj));
2344 return obj->oo_owner == env;
2347 static struct timespec *osd_inode_time(const struct lu_env *env,
2348 struct inode *inode, __u64 seconds)
2350 struct osd_thread_info *oti = osd_oti_get(env);
2351 struct timespec *t = &oti->oti_time;
2353 t->tv_sec = seconds;
2355 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2359 static void osd_inode_getattr(const struct lu_env *env,
2360 struct inode *inode, struct lu_attr *attr)
2362 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2363 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2364 LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2365 LA_BLKSIZE | LA_TYPE;
2367 attr->la_atime = LTIME_S(inode->i_atime);
2368 attr->la_mtime = LTIME_S(inode->i_mtime);
2369 attr->la_ctime = LTIME_S(inode->i_ctime);
2370 attr->la_mode = inode->i_mode;
2371 attr->la_size = i_size_read(inode);
2372 attr->la_blocks = inode->i_blocks;
2373 attr->la_uid = i_uid_read(inode);
2374 attr->la_gid = i_gid_read(inode);
2375 attr->la_projid = i_projid_read(inode);
2376 attr->la_flags = ll_inode_to_ext_flags(inode->i_flags);
2377 attr->la_nlink = inode->i_nlink;
2378 attr->la_rdev = inode->i_rdev;
2379 attr->la_blksize = 1 << inode->i_blkbits;
2380 attr->la_blkbits = inode->i_blkbits;
2382 * Ext4 did not transfer inherit flags from raw inode
2383 * to inode flags, and ext4 internally test raw inode
2384 * @i_flags directly. Instead of patching ext4, we do it here.
2386 if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2387 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2390 static int osd_attr_get(const struct lu_env *env,
2391 struct dt_object *dt,
2392 struct lu_attr *attr)
2394 struct osd_object *obj = osd_dt_obj(dt);
2396 if (unlikely(!dt_object_exists(dt)))
2398 if (unlikely(obj->oo_destroyed))
2401 LASSERT(!dt_object_remote(dt));
2402 LINVRNT(osd_invariant(obj));
2404 spin_lock(&obj->oo_guard);
2405 osd_inode_getattr(env, obj->oo_inode, attr);
2406 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2407 attr->la_flags |= LUSTRE_ORPHAN_FL;
2408 spin_unlock(&obj->oo_guard);
2413 static int osd_declare_attr_qid(const struct lu_env *env,
2414 struct osd_object *obj,
2415 struct osd_thandle *oh, long long bspace,
2416 qid_t old_id, qid_t new_id, bool enforce,
2420 struct osd_thread_info *info = osd_oti_get(env);
2421 struct lquota_id_info *qi = &info->oti_qi;
2423 qi->lqi_type = type;
2424 /* inode accounting */
2425 qi->lqi_is_blk = false;
2427 /* one more inode for the new id ... */
2428 qi->lqi_id.qid_uid = new_id;
2430 /* Reserve credits for the new id */
2431 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2432 if (rc == -EDQUOT || rc == -EINPROGRESS)
2437 /* and one less inode for the current id */
2438 qi->lqi_id.qid_uid = old_id;
2440 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2441 if (rc == -EDQUOT || rc == -EINPROGRESS)
2446 /* block accounting */
2447 qi->lqi_is_blk = true;
2449 /* more blocks for the new id ... */
2450 qi->lqi_id.qid_uid = new_id;
2451 qi->lqi_space = bspace;
2453 * Credits for the new uid has been reserved, re-use "obj"
2454 * to save credit reservation.
2456 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2457 if (rc == -EDQUOT || rc == -EINPROGRESS)
2462 /* and finally less blocks for the current uid */
2463 qi->lqi_id.qid_uid = old_id;
2464 qi->lqi_space = -bspace;
2465 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2466 if (rc == -EDQUOT || rc == -EINPROGRESS)
2472 static int osd_declare_attr_set(const struct lu_env *env,
2473 struct dt_object *dt,
2474 const struct lu_attr *attr,
2475 struct thandle *handle)
2477 struct osd_thandle *oh;
2478 struct osd_object *obj;
2486 LASSERT(dt != NULL);
2487 LASSERT(handle != NULL);
2489 obj = osd_dt_obj(dt);
2490 LASSERT(osd_invariant(obj));
2492 oh = container_of0(handle, struct osd_thandle, ot_super);
2493 LASSERT(oh->ot_handle == NULL);
2495 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2496 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2498 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2499 osd_dto_credits_noquota[DTO_XATTR_SET]);
2501 if (attr == NULL || obj->oo_inode == NULL)
2504 bspace = obj->oo_inode->i_blocks;
2505 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2506 bspace = toqb(bspace);
2508 /* Changing ownership is always preformed by super user, it should not
2511 * We still need to call the osd_declare_qid() to calculate the journal
2512 * credits for updating quota accounting files and to trigger quota
2513 * space adjustment once the operation is completed.*/
2514 if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2516 uid = i_uid_read(obj->oo_inode);
2517 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2518 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2519 attr->la_uid, enforce, USRQUOTA);
2523 gid = i_gid_read(obj->oo_inode);
2524 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2525 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2526 i_gid_read(obj->oo_inode),
2527 attr->la_gid, enforce, GRPQUOTA);
2532 #ifdef HAVE_PROJECT_QUOTA
2533 if (attr->la_valid & LA_PROJID) {
2534 __u32 projid = i_projid_read(obj->oo_inode);
2535 enforce = (attr->la_valid & LA_PROJID) &&
2536 (attr->la_projid != projid);
2537 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2538 (qid_t)projid, (qid_t)attr->la_projid,
2547 static int osd_inode_setattr(const struct lu_env *env,
2548 struct inode *inode, const struct lu_attr *attr)
2550 __u64 bits = attr->la_valid;
2552 /* Only allow set size for regular file */
2553 if (!S_ISREG(inode->i_mode))
2554 bits &= ~(LA_SIZE | LA_BLOCKS);
2559 if (bits & LA_ATIME)
2560 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
2561 if (bits & LA_CTIME)
2562 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
2563 if (bits & LA_MTIME)
2564 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
2565 if (bits & LA_SIZE) {
2566 spin_lock(&inode->i_lock);
2567 LDISKFS_I(inode)->i_disksize = attr->la_size;
2568 i_size_write(inode, attr->la_size);
2569 spin_unlock(&inode->i_lock);
2572 /* OSD should not change "i_blocks" which is used by quota.
2573 * "i_blocks" should be changed by ldiskfs only. */
2575 inode->i_mode = (inode->i_mode & S_IFMT) |
2576 (attr->la_mode & ~S_IFMT);
2578 i_uid_write(inode, attr->la_uid);
2580 i_gid_write(inode, attr->la_gid);
2581 if (bits & LA_PROJID)
2582 i_projid_write(inode, attr->la_projid);
2583 if (bits & LA_NLINK)
2584 set_nlink(inode, attr->la_nlink);
2586 inode->i_rdev = attr->la_rdev;
2588 if (bits & LA_FLAGS) {
2589 /* always keep S_NOCMTIME */
2590 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2593 * Ext4 did not transfer inherit flags from
2594 * @inode->i_flags to raw inode i_flags when writing
2595 * flags, we do it explictly here.
2597 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2598 LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2603 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2607 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2608 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2611 ll_vfs_dq_init(inode);
2613 if (attr->la_valid & LA_UID)
2614 iattr.ia_valid |= ATTR_UID;
2615 if (attr->la_valid & LA_GID)
2616 iattr.ia_valid |= ATTR_GID;
2617 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2618 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2620 rc = ll_vfs_dq_transfer(inode, &iattr);
2622 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2623 "enforcement enabled on the ldiskfs "
2624 "filesystem?\n", inode->i_sb->s_id, rc);
2629 /* Handle project id transfer here properly */
2630 if (attr->la_valid & LA_PROJID &&
2631 attr->la_projid != i_projid_read(inode)) {
2632 #ifdef HAVE_PROJECT_QUOTA
2633 rc = ldiskfs_transfer_project(inode, attr->la_projid);
2638 CERROR("%s: quota transfer failed: rc = %d. Is project "
2639 "enforcement enabled on the ldiskfs "
2640 "filesystem?\n", inode->i_sb->s_id, rc);
2647 static int osd_attr_set(const struct lu_env *env,
2648 struct dt_object *dt,
2649 const struct lu_attr *attr,
2650 struct thandle *handle)
2652 struct osd_object *obj = osd_dt_obj(dt);
2653 struct inode *inode;
2656 if (!dt_object_exists(dt))
2659 LASSERT(handle != NULL);
2660 LASSERT(!dt_object_remote(dt));
2661 LASSERT(osd_invariant(obj));
2663 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2665 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2666 struct osd_thread_info *oti = osd_oti_get(env);
2667 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2668 struct lu_fid *fid1 = &oti->oti_fid;
2669 struct osd_inode_id *id = &oti->oti_id;
2670 struct iam_path_descr *ipd;
2671 struct iam_container *bag;
2672 struct osd_thandle *oh;
2675 fid_cpu_to_be(fid1, fid0);
2676 memset(id, 1, sizeof(*id));
2677 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2678 fid0)->oi_dir.od_container;
2679 ipd = osd_idx_ipd_get(env, bag);
2680 if (unlikely(ipd == NULL))
2683 oh = container_of0(handle, struct osd_thandle, ot_super);
2684 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2685 (const struct iam_rec *)id, ipd);
2686 osd_ipd_put(env, bag, ipd);
2687 return(rc > 0 ? 0 : rc);
2690 inode = obj->oo_inode;
2692 rc = osd_quota_transfer(inode, attr);
2696 spin_lock(&obj->oo_guard);
2697 rc = osd_inode_setattr(env, inode, attr);
2698 spin_unlock(&obj->oo_guard);
2702 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2704 if (!(attr->la_valid & LA_FLAGS))
2707 /* Let's check if there are extra flags need to be set into LMA */
2708 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2709 struct osd_thread_info *info = osd_oti_get(env);
2710 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2712 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2713 &info->oti_ost_attrs);
2717 lma->lma_incompat |=
2718 lustre_to_lma_flags(attr->la_flags);
2719 lustre_lma_swab(lma);
2720 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2721 lma, sizeof(*lma), XATTR_REPLACE);
2723 struct osd_device *osd = osd_obj2dev(obj);
2725 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2726 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2727 lma->lma_incompat, rc);
2730 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2732 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2735 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2740 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2741 struct osd_object *obj,
2742 const char *name, const int namelen)
2744 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2747 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2748 umode_t mode, struct dt_allocation_hint *hint,
2752 struct osd_device *osd = osd_obj2dev(obj);
2753 struct osd_thandle *oth;
2754 struct dt_object *parent = NULL;
2755 struct inode *inode;
2757 LINVRNT(osd_invariant(obj));
2758 LASSERT(obj->oo_inode == NULL);
2759 LASSERT(obj->oo_hl_head == NULL);
2761 if (S_ISDIR(mode) && ldiskfs_pdo) {
2762 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2763 if (obj->oo_hl_head == NULL)
2767 oth = container_of(th, struct osd_thandle, ot_super);
2768 LASSERT(oth->ot_handle->h_transaction != NULL);
2770 if (hint != NULL && hint->dah_parent != NULL &&
2771 !dt_object_remote(hint->dah_parent))
2772 parent = hint->dah_parent;
2774 inode = ldiskfs_create_inode(oth->ot_handle,
2775 parent ? osd_dt_obj(parent)->oo_inode :
2776 osd_sb(osd)->s_root->d_inode,
2778 if (!IS_ERR(inode)) {
2779 /* Do not update file c/mtime in ldiskfs. */
2780 inode->i_flags |= S_NOCMTIME;
2782 /* For new created object, it must be consistent,
2783 * and it is unnecessary to scrub against it. */
2784 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2786 obj->oo_inode = inode;
2789 if (obj->oo_hl_head != NULL) {
2790 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2791 obj->oo_hl_head = NULL;
2793 result = PTR_ERR(inode);
2795 LINVRNT(osd_invariant(obj));
2803 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2804 struct lu_attr *attr,
2805 struct dt_allocation_hint *hint,
2806 struct dt_object_format *dof,
2810 struct osd_thandle *oth;
2811 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2813 LASSERT(S_ISDIR(attr->la_mode));
2815 oth = container_of(th, struct osd_thandle, ot_super);
2816 LASSERT(oth->ot_handle->h_transaction != NULL);
2817 result = osd_mkfile(info, obj, mode, hint, th);
2822 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2823 struct lu_attr *attr,
2824 struct dt_allocation_hint *hint,
2825 struct dt_object_format *dof,
2829 struct osd_thandle *oth;
2830 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2832 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2834 LASSERT(S_ISREG(attr->la_mode));
2836 oth = container_of(th, struct osd_thandle, ot_super);
2837 LASSERT(oth->ot_handle->h_transaction != NULL);
2839 result = osd_mkfile(info, obj, mode, hint, th);
2841 LASSERT(obj->oo_inode != NULL);
2842 if (feat->dif_flags & DT_IND_VARKEY)
2843 result = iam_lvar_create(obj->oo_inode,
2844 feat->dif_keysize_max,
2846 feat->dif_recsize_max,
2849 result = iam_lfix_create(obj->oo_inode,
2850 feat->dif_keysize_max,
2852 feat->dif_recsize_max,
2859 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2860 struct lu_attr *attr,
2861 struct dt_allocation_hint *hint,
2862 struct dt_object_format *dof,
2865 LASSERT(S_ISREG(attr->la_mode));
2866 return osd_mkfile(info, obj, (attr->la_mode &
2867 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2870 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2871 struct lu_attr *attr,
2872 struct dt_allocation_hint *hint,
2873 struct dt_object_format *dof,
2876 LASSERT(S_ISLNK(attr->la_mode));
2877 return osd_mkfile(info, obj, (attr->la_mode &
2878 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2881 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2882 struct lu_attr *attr,
2883 struct dt_allocation_hint *hint,
2884 struct dt_object_format *dof,
2887 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2890 LINVRNT(osd_invariant(obj));
2891 LASSERT(obj->oo_inode == NULL);
2892 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2893 S_ISFIFO(mode) || S_ISSOCK(mode));
2895 result = osd_mkfile(info, obj, mode, hint, th);
2897 LASSERT(obj->oo_inode != NULL);
2899 * This inode should be marked dirty for i_rdev. Currently
2900 * that is done in the osd_attr_init().
2902 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2905 LINVRNT(osd_invariant(obj));
2909 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2911 struct dt_allocation_hint *hint,
2912 struct dt_object_format *dof,
2915 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2917 osd_obj_type_f result;
2933 result = osd_mk_index;
2944 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2945 struct dt_object *parent, struct dt_object *child,
2950 ah->dah_parent = parent;
2951 ah->dah_mode = child_mode;
2953 if (parent != NULL && !dt_object_remote(parent)) {
2954 /* will help to find FID->ino at dt_insert("..") */
2955 struct osd_object *pobj = osd_dt_obj(parent);
2956 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2960 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2961 struct lu_attr *attr, struct dt_object_format *dof)
2963 struct inode *inode = obj->oo_inode;
2964 __u64 valid = attr->la_valid;
2967 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2969 if (dof->dof_type != DFT_NODE)
2970 attr->la_valid &= ~LA_RDEV;
2971 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2972 attr->la_valid &= ~LA_ATIME;
2973 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2974 attr->la_valid &= ~LA_CTIME;
2975 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2976 attr->la_valid &= ~LA_MTIME;
2978 result = osd_quota_transfer(inode, attr);
2982 if (attr->la_valid != 0) {
2983 result = osd_inode_setattr(info->oti_env, inode, attr);
2985 * The osd_inode_setattr() should always succeed here. The
2986 * only error that could be returned is EDQUOT when we are
2987 * trying to change the UID or GID of the inode. However, this
2988 * should not happen since quota enforcement is no longer
2989 * enabled on ldiskfs (lquota takes care of it).
2991 LASSERTF(result == 0, "%d\n", result);
2992 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2995 attr->la_valid = valid;
2999 * Helper function for osd_create()
3001 * \retval 0, on success
3003 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3004 struct lu_attr *attr, struct dt_allocation_hint *hint,
3005 struct dt_object_format *dof, struct thandle *th)
3010 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3012 /* we drop umask so that permissions we pass are not affected */
3013 umask = current->fs->umask;
3014 current->fs->umask = 0;
3016 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3018 if (likely(obj->oo_inode != NULL)) {
3019 LASSERT(obj->oo_inode->i_state & I_NEW);
3021 /* Unlock the inode before attr initialization to avoid
3022 * unnecessary dqget operations. LU-6378 */
3023 unlock_new_inode(obj->oo_inode);
3026 if (likely(result == 0)) {
3027 osd_attr_init(info, obj, attr, dof);
3028 osd_object_init0(obj);
3031 /* restore previous umask value */
3032 current->fs->umask = umask;
3034 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3040 * Helper function for osd_create()
3042 * \retval 0, on success
3044 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3045 const struct lu_fid *fid, struct thandle *th)
3047 struct osd_thread_info *info = osd_oti_get(env);
3048 struct osd_inode_id *id = &info->oti_id;
3049 struct osd_device *osd = osd_obj2dev(obj);
3050 struct osd_thandle *oh;
3053 LASSERT(obj->oo_inode != NULL);
3055 oh = container_of0(th, struct osd_thandle, ot_super);
3056 LASSERT(oh->ot_handle);
3057 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3059 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3060 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3061 OI_CHECK_FLD, NULL);
3062 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3067 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3068 u64 seq, struct lu_seq_range *range)
3070 struct seq_server_site *ss = osd_seq_site(osd);
3072 if (fid_seq_is_idif(seq)) {
3073 fld_range_set_ost(range);
3074 range->lsr_index = idif_ost_idx(seq);
3078 if (!fid_seq_in_fldb(seq)) {
3079 fld_range_set_mdt(range);
3081 /* FIXME: If ss is NULL, it suppose not get lsr_index
3083 range->lsr_index = ss->ss_node_id;
3087 LASSERT(ss != NULL);
3088 fld_range_set_any(range);
3089 /* OSD will only do local fld lookup */
3090 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3093 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3094 struct lu_attr *attr,
3095 struct dt_allocation_hint *hint,
3096 struct dt_object_format *dof,
3097 struct thandle *handle)
3099 struct osd_thandle *oh;
3103 LASSERT(handle != NULL);
3105 oh = container_of0(handle, struct osd_thandle, ot_super);
3106 LASSERT(oh->ot_handle == NULL);
3108 /* EA object consumes more credits than regular object: osd_mk_index
3109 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3110 * leaf_node, could involves the block, block bitmap, groups, GDT
3111 * change for each block, so add 4 * 2 credits in that case. */
3112 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3113 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3114 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3115 /* Reuse idle OI block may cause additional one OI block
3117 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3118 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3123 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3124 attr->la_projid, 1, oh, osd_dt_obj(dt),
3125 NULL, OSD_QID_INODE);
3129 /* will help to find FID->ino mapping at dt_insert() */
3130 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3137 * Called to destroy on-disk representation of the object
3139 * Concurrency: must be locked
3141 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3144 struct osd_object *obj = osd_dt_obj(dt);
3145 struct inode *inode = obj->oo_inode;
3146 struct osd_thandle *oh;
3153 oh = container_of0(th, struct osd_thandle, ot_super);
3154 LASSERT(oh->ot_handle == NULL);
3156 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3157 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3158 /* Recycle idle OI leaf may cause additional three OI blocks
3160 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3161 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3162 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3163 /* one less inode */
3164 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3165 i_projid_read(inode), -1, oh, obj, NULL,
3169 /* data to be truncated */
3170 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3171 i_projid_read(inode), 0, oh, obj, NULL,
3176 /* will help to find FID->ino when this object is being
3177 * added to PENDING/ */
3178 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3183 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3186 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3187 struct osd_object *obj = osd_dt_obj(dt);
3188 struct inode *inode = obj->oo_inode;
3189 struct osd_device *osd = osd_obj2dev(obj);
3190 struct osd_thandle *oh;
3194 oh = container_of0(th, struct osd_thandle, ot_super);
3195 LASSERT(oh->ot_handle);
3197 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3199 if (unlikely(fid_is_acct(fid)))
3202 if (S_ISDIR(inode->i_mode)) {
3203 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3204 inode->i_nlink == 2);
3205 /* it will check/delete the inode from remote parent,
3206 * how to optimize it? unlink performance impaction XXX */
3207 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3209 CERROR("%s: delete inode "DFID": rc = %d\n",
3210 osd_name(osd), PFID(fid), result);
3212 spin_lock(&obj->oo_guard);
3214 spin_unlock(&obj->oo_guard);
3215 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3218 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3220 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3222 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3223 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3224 oh->ot_handle, OI_CHECK_FLD);
3226 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3227 /* XXX: add to ext3 orphan list */
3228 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3230 /* not needed in the cache anymore */
3231 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3232 obj->oo_destroyed = 1;
3238 * Put the fid into lustre_mdt_attrs, and then place the structure
3239 * inode's ea. This fid should not be altered during the life time
3242 * \retval +ve, on success
3243 * \retval -ve, on error
3245 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3247 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3248 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3250 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3251 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3255 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3258 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3261 lustre_loa_init(loa, fid, compat, incompat);
3262 lustre_loa_swab(loa, false);
3264 /* For the OST device with 256 bytes inode size by default,
3265 * the PFID EA will be stored together with LMA EA to avoid
3266 * performance trouble. Otherwise the PFID EA can be stored
3267 * independently. LU-8998 */
3268 if ((compat & LMAC_FID_ON_OST) &&
3269 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3270 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3271 sizeof(*loa), XATTR_CREATE);
3273 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3274 sizeof(*lma), XATTR_CREATE);
3275 /* LMA may already exist, but we need to check that all the
3276 * desired compat/incompat flags have been added. */
3277 if (unlikely(rc == -EEXIST)) {
3278 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3279 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3283 if (rc < sizeof(*lma))
3286 lustre_loa_swab(loa, true);
3287 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3288 ((compat == 0 && incompat == 0) ||
3289 (!(~lma->lma_compat & compat) &&
3290 !(~lma->lma_incompat & incompat))))
3293 lma->lma_self_fid = *fid;
3294 lma->lma_compat |= compat;
3295 lma->lma_incompat |= incompat;
3296 if (rc == sizeof(*lma)) {
3297 lustre_lma_swab(lma);
3298 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3299 sizeof(*lma), XATTR_REPLACE);
3301 lustre_loa_swab(loa, false);
3302 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3303 sizeof(*loa), XATTR_REPLACE);
3311 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3312 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3313 * To have compatilibility with 1.8 ldiskfs driver we need to have
3314 * magic number at start of fid data.
3315 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3318 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3319 const struct lu_fid *fid)
3321 if (!fid_is_namespace_visible(fid) ||
3322 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3323 param->edp_magic = 0;
3327 param->edp_magic = LDISKFS_LUFID_MAGIC;
3328 param->edp_len = sizeof(struct lu_fid) + 1;
3329 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3333 * Try to read the fid from inode ea into dt_rec.
3335 * \param fid object fid.
3337 * \retval 0 on success
3339 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3340 __u32 ino, struct lu_fid *fid,
3341 struct osd_inode_id *id)
3343 struct osd_thread_info *info = osd_oti_get(env);
3344 struct inode *inode;
3347 osd_id_gen(id, ino, OSD_OII_NOGEN);
3348 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3350 RETURN(PTR_ERR(inode));
3356 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3358 struct inode *parent_dir,
3359 const struct lu_fid *dot_fid,
3360 const struct lu_fid *dot_dot_fid,
3361 struct osd_thandle *oth)
3363 struct ldiskfs_dentry_param *dot_ldp;
3364 struct ldiskfs_dentry_param *dot_dot_ldp;
3365 __u32 saved_nlink = dir->i_nlink;
3368 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3369 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3371 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3372 dot_ldp->edp_magic = 0;
3374 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3375 dir, dot_ldp, dot_dot_ldp);
3376 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3377 * the subseqent ref_add() will increase the dir->i_nlink
3378 * as 3. That is incorrect for new created directory.
3380 * It looks like hack, because we want to make the OSD API
3381 * to be order-independent for new created directory object
3382 * between dt_insert(..) and ref_add() operations.
3384 * Here, we only restore the in-RAM dir-inode's nlink attr,
3385 * becuase if the nlink attr is not 2, then there will be
3386 * ref_add() called following the dt_insert(..), such call
3387 * will make both the in-RAM and on-disk dir-inode's nlink
3388 * attr to be set as 2. LU-7447 */
3389 set_nlink(dir, saved_nlink);
3394 * Create an local agent inode for remote entry
3396 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3397 struct osd_device *osd,
3398 struct osd_object *pobj,
3399 const struct lu_fid *fid,
3403 struct osd_thread_info *info = osd_oti_get(env);
3404 struct inode *local;
3405 struct osd_thandle *oh;
3410 oh = container_of(th, struct osd_thandle, ot_super);
3411 LASSERT(oh->ot_handle->h_transaction != NULL);
3413 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3414 if (IS_ERR(local)) {
3415 CERROR("%s: create local error %d\n", osd_name(osd),
3416 (int)PTR_ERR(local));
3420 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3421 * correct gid on remote file, not agent here */
3422 local->i_gid = current_fsgid();
3423 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3424 unlock_new_inode(local);
3426 /* Agent inode should not have project ID */
3427 #ifdef HAVE_PROJECT_QUOTA
3428 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3429 i_projid_read(pobj->oo_inode) != 0) {
3430 rc = ldiskfs_transfer_project(local, 0);
3432 CERROR("%s: quota transfer failed: rc = %d. Is project "
3433 "quota enforcement enabled on the ldiskfs "
3434 "filesystem?\n", local->i_sb->s_id, rc);
3435 RETURN(ERR_PTR(rc));
3439 /* Set special LMA flag for local agent inode */
3440 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3442 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3443 osd_name(osd), PFID(fid), rc);
3444 RETURN(ERR_PTR(rc));
3450 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3451 lu_object_fid(&pobj->oo_dt.do_lu),
3454 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3455 osd_name(osd), PFID(fid), rc);
3456 RETURN(ERR_PTR(rc));
3463 * when direntry is deleted, we have to take care of possible agent inode
3464 * referenced by that. unfortunately we can't do this at that point:
3465 * iget() within a running transaction leads to deadlock and we better do
3466 * not call that every delete declaration to save performance. so we put
3467 * a potention agent inode on a list and process that once the transaction
3468 * is over. Notice it's not any worse in terms of real orphans as regular
3469 * object destroy doesn't put inodes on the on-disk orphan list. this should
3470 * be addressed separately
3472 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3473 struct osd_thandle *oh,
3476 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3477 struct osd_obj_orphan *oor;
3484 oor->oor_env = (struct lu_env *)env;
3485 spin_lock(&osd->od_osfs_lock);
3486 list_add(&oor->oor_list, &osd->od_orphan_list);
3487 spin_unlock(&osd->od_osfs_lock);
3489 oh->ot_remove_agents = 1;
3495 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3496 struct osd_device *osd)
3498 struct osd_thread_info *info = osd_oti_get(env);
3499 struct osd_obj_orphan *oor, *tmp;
3500 struct osd_inode_id id;
3501 struct list_head list;
3502 struct inode *inode;
3507 INIT_LIST_HEAD(&list);
3509 spin_lock(&osd->od_osfs_lock);
3510 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3511 if (oor->oor_env == env) {
3512 list_del(&oor->oor_list);
3513 list_add(&oor->oor_list, &list);
3516 spin_unlock(&osd->od_osfs_lock);
3518 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3522 list_del(&oor->oor_list);
3525 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3526 inode = osd_iget_fid(info, osd, &id, &fid);
3530 if (!osd_remote_fid(env, osd, &fid)) {
3535 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3537 mark_inode_dirty(inode);
3538 ldiskfs_journal_stop(jh);
3546 * OSD layer object create function for OST objects (b=11826).
3548 * The FID is inserted into inode xattr here.
3550 * \retval 0, on success
3551 * \retval -ve, on error
3553 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3554 struct lu_attr *attr, struct dt_allocation_hint *hint,
3555 struct dt_object_format *dof, struct thandle *th)
3557 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3558 struct osd_object *obj = osd_dt_obj(dt);
3559 struct osd_thread_info *info = osd_oti_get(env);
3560 int result, on_ost = 0;
3564 if (dt_object_exists(dt))
3567 LINVRNT(osd_invariant(obj));
3568 LASSERT(!dt_object_remote(dt));
3569 LASSERT(osd_is_write_locked(env, obj));
3570 LASSERT(th != NULL);
3572 if (unlikely(fid_is_acct(fid)))
3573 /* Quota files can't be created from the kernel any more,
3574 * 'tune2fs -O quota' will take care of creating them */
3577 result = __osd_create(info, obj, attr, hint, dof, th);
3579 if (fid_is_idif(fid) &&
3580 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3581 struct lu_fid *tfid = &info->oti_fid;
3582 struct ost_id *oi = &info->oti_ostid;
3584 fid_to_ostid(fid, oi);
3585 ostid_to_fid(tfid, oi, 0);
3587 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3588 LMAC_FID_ON_OST, 0);
3590 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3592 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3593 on_ost ? LMAC_FID_ON_OST : 0,
3596 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3597 obj->oo_dt.do_body_ops = &osd_body_ops;
3601 result = __osd_oi_insert(env, obj, fid, th);
3603 /* a small optimization - dt_insert() isn't usually applied
3604 * to OST objects, so we don't need to cache OI mapping for
3606 if (result == 0 && on_ost == 0) {
3607 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3608 result = osd_idc_find_and_init(env, osd, obj);
3609 LASSERT(result == 0);
3612 LASSERT(ergo(result == 0,
3613 dt_object_exists(dt) && !dt_object_remote(dt)));
3614 LINVRNT(osd_invariant(obj));
3618 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3619 struct thandle *handle)
3621 struct osd_thandle *oh;
3623 /* it's possible that object doesn't exist yet */
3624 LASSERT(handle != NULL);
3626 oh = container_of0(handle, struct osd_thandle, ot_super);
3627 LASSERT(oh->ot_handle == NULL);
3629 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3630 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3636 * Concurrency: @dt is write locked.
3638 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3641 struct osd_object *obj = osd_dt_obj(dt);
3642 struct inode *inode = obj->oo_inode;
3643 struct osd_thandle *oh;
3646 if (!dt_object_exists(dt) || obj->oo_destroyed)
3649 LINVRNT(osd_invariant(obj));
3650 LASSERT(!dt_object_remote(dt));
3651 LASSERT(osd_is_write_locked(env, obj));
3652 LASSERT(th != NULL);
3654 oh = container_of0(th, struct osd_thandle, ot_super);
3655 LASSERT(oh->ot_handle != NULL);
3657 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3659 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3660 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3662 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3663 * (65000) subdirectories by storing "1" in i_nlink if the link count
3664 * would otherwise overflow. Directory tranversal tools understand
3665 * that (st_nlink == 1) indicates that the filesystem dose not track
3666 * hard links count on the directory, and will not abort subdirectory
3667 * scanning early once (st_nlink - 2) subdirs have been found.
3669 * This also has to properly handle the case of inodes with nlink == 0
3670 * in case they are being linked into the PENDING directory
3672 spin_lock(&obj->oo_guard);
3673 if (unlikely(inode->i_nlink == 0))
3674 /* inc_nlink from 0 may cause WARN_ON */
3675 set_nlink(inode, 1);
3677 ldiskfs_inc_count(oh->ot_handle, inode);
3678 if (!S_ISDIR(inode->i_mode))
3679 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3681 spin_unlock(&obj->oo_guard);
3683 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3684 LINVRNT(osd_invariant(obj));
3686 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3691 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3692 struct thandle *handle)
3694 struct osd_thandle *oh;
3696 if (!dt_object_exists(dt))
3699 LASSERT(!dt_object_remote(dt));
3700 LASSERT(handle != NULL);
3702 oh = container_of0(handle, struct osd_thandle, ot_super);
3703 LASSERT(oh->ot_handle == NULL);
3705 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3706 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3712 * Concurrency: @dt is write locked.
3714 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3717 struct osd_object *obj = osd_dt_obj(dt);
3718 struct inode *inode = obj->oo_inode;
3719 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3720 struct osd_thandle *oh;
3722 if (!dt_object_exists(dt))
3725 LINVRNT(osd_invariant(obj));
3726 LASSERT(!dt_object_remote(dt));
3727 LASSERT(osd_is_write_locked(env, obj));
3728 LASSERT(th != NULL);
3730 oh = container_of0(th, struct osd_thandle, ot_super);
3731 LASSERT(oh->ot_handle != NULL);
3733 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3735 spin_lock(&obj->oo_guard);
3736 /* That can be result of upgrade from old Lustre version and
3737 * applied only to local files. Just skip this ref_del call.
3738 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3739 if (inode->i_nlink == 0) {
3740 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3741 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3742 ", maybe an upgraded file? (LU-3915)\n",
3743 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3744 spin_unlock(&obj->oo_guard);
3748 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3749 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3751 ldiskfs_dec_count(oh->ot_handle, inode);
3752 spin_unlock(&obj->oo_guard);
3754 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3755 LINVRNT(osd_invariant(obj));
3757 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3763 * Concurrency: @dt is read locked.
3765 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3766 struct lu_buf *buf, const char *name)
3768 struct osd_object *obj = osd_dt_obj(dt);
3769 struct inode *inode = obj->oo_inode;
3770 struct osd_thread_info *info = osd_oti_get(env);
3771 struct dentry *dentry = &info->oti_obj_dentry;
3772 bool cache_xattr = false;
3777 /* version get is not real XATTR but uses xattr API */
3778 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3779 dt_obj_version_t *ver = buf->lb_buf;
3781 /* for version we are just using xattr API but change inode
3783 if (buf->lb_len == 0)
3784 return sizeof(dt_obj_version_t);
3786 if (buf->lb_len < sizeof(dt_obj_version_t))
3789 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3790 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3792 *ver = LDISKFS_I(inode)->i_fs_version;
3794 return sizeof(dt_obj_version_t);
3797 if (!dt_object_exists(dt))
3800 LASSERT(!dt_object_remote(dt));
3801 LASSERT(inode->i_op != NULL);
3802 LASSERT(inode->i_op->getxattr != NULL);
3804 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3805 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3809 rc = osd_oxc_get(obj, name, buf);
3814 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3815 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3816 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3817 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3818 struct filter_fid *ff;
3819 struct ost_layout *ol;
3821 if (!osd_dev(dt->do_lu.lo_dev)->od_is_ost)
3824 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3828 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3834 if (buf->lb_len == 0 || !buf->lb_buf)
3837 if (buf->lb_len < rc)
3841 ol = &ff->ff_layout;
3842 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3843 PFID_STRIPE_IDX_BITS);
3844 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3845 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3846 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3847 if (lma->lma_compat & LMAC_COMP_INFO) {
3848 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3849 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3850 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3852 ol->ol_comp_start = 0;
3853 ol->ol_comp_end = 0;
3860 if (rc == -ENOENT || rc == -ENODATA)
3861 osd_oxc_add(obj, name, NULL, 0);
3862 else if (rc > 0 && buf->lb_buf != NULL)
3863 osd_oxc_add(obj, name, buf->lb_buf, rc);
3869 static int osd_declare_xattr_set(const struct lu_env *env,
3870 struct dt_object *dt,
3871 const struct lu_buf *buf, const char *name,
3872 int fl, struct thandle *handle)
3874 struct osd_thandle *oh;
3876 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3878 LASSERT(handle != NULL);
3880 oh = container_of0(handle, struct osd_thandle, ot_super);
3881 LASSERT(oh->ot_handle == NULL);
3883 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3884 /* For non-upgrading case, the LMA is set first and
3885 * usually fit inode. But for upgrade case, the LMA
3886 * may be in another separated EA block. */
3887 if (dt_object_exists(dt)) {
3888 if (fl == LU_XATTR_REPLACE)
3893 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3895 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3896 /* We may need to delete the old PFID EA. */
3897 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3898 if (fl == LU_XATTR_REPLACE)
3905 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3910 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3911 /* learn xattr size from osd_xattr_get if
3912 attribute has not been read yet */
3913 buflen = __osd_xattr_get(
3914 osd_dt_obj(dt)->oo_inode,
3915 &osd_oti_get(env)->oti_obj_dentry,
3920 buflen = buf->lb_len;
3923 if (buflen > sb->s_blocksize) {
3924 credits += osd_calc_bkmap_credits(
3926 (buflen + sb->s_blocksize - 1) >>
3927 sb->s_blocksize_bits);
3931 * xattr set may involve inode quota change, reserve credits for
3932 * dquot_initialize()
3934 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3937 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3943 * Concurrency: @dt is write locked.
3945 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3946 const struct lu_buf *buf, const char *name, int fl,
3947 struct thandle *handle)
3949 struct osd_object *obj = osd_dt_obj(dt);
3950 struct inode *inode = obj->oo_inode;
3951 struct osd_thread_info *info = osd_oti_get(env);
3952 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3953 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3962 /* version set is not real XATTR */
3963 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3964 dt_obj_version_t *version = buf->lb_buf;
3966 /* for version we are just using xattr API but change inode
3968 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3970 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3971 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3973 LDISKFS_I(inode)->i_fs_version = *version;
3974 /* Version is set after all inode operations are finished,
3975 * so we should mark it dirty here */
3976 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3981 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3982 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3985 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3986 if (fl & LU_XATTR_REPLACE)
3987 fs_flags |= XATTR_REPLACE;
3989 if (fl & LU_XATTR_CREATE)
3990 fs_flags |= XATTR_CREATE;
3992 /* For the OST device with 256 bytes inode size by default,
3993 * the PFID EA will be stored together with LMA EA to avoid
3994 * performance trouble. Otherwise the PFID EA can be stored
3995 * independently. LU-8998 */
3996 if (strcmp(name, XATTR_NAME_FID) == 0 &&
3997 LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
3998 struct dentry *dentry = &info->oti_obj_dentry;
3999 struct filter_fid *ff;
4000 struct ost_layout *ol;
4003 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4006 ol = &ff->ff_layout;
4007 /* Old client does not send stripe information, store
4008 * the PFID EA on disk directly. */
4009 if (buf->lb_len == sizeof(struct lu_fid) ||
4010 ol->ol_stripe_size == 0) {
4011 len = sizeof(struct lu_fid);
4015 if (buf->lb_len != sizeof(*ff))
4018 rc = osd_get_lma(info, inode, dentry, loa);
4019 if (unlikely(rc == -ENODATA)) {
4020 /* Usually for upgarding from old device */
4021 lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4022 LMAC_FID_ON_OST, 0);
4030 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4031 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4032 PFID_STRIPE_IDX_BITS;
4033 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4034 lma->lma_compat |= LMAC_STRIPE_INFO;
4035 if (ol->ol_comp_id != 0) {
4036 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4037 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4038 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4039 lma->lma_compat |= LMAC_COMP_INFO;
4042 lustre_loa_swab(loa, false);
4044 /* Remove old PFID EA entry firstly. */
4045 ll_vfs_dq_init(inode);
4046 rc = inode->i_op->removexattr(dentry, name);
4047 if (rc && rc != -ENODATA)
4050 /* Store the PFID EA inside the LMA EA. */
4051 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4055 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4056 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4060 lma->lma_incompat |= LMAI_STRIPED;
4061 lustre_lma_swab(lma);
4062 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4063 sizeof(*lma), XATTR_REPLACE);
4069 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4070 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4073 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4074 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4075 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4081 * Concurrency: @dt is read locked.
4083 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4084 const struct lu_buf *buf)
4086 struct osd_object *obj = osd_dt_obj(dt);
4087 struct inode *inode = obj->oo_inode;
4088 struct osd_thread_info *info = osd_oti_get(env);
4089 struct dentry *dentry = &info->oti_obj_dentry;
4091 if (!dt_object_exists(dt))
4094 LASSERT(!dt_object_remote(dt));
4095 LASSERT(inode->i_op != NULL);
4096 LASSERT(inode->i_op->listxattr != NULL);
4098 dentry->d_inode = inode;
4099 dentry->d_sb = inode->i_sb;
4100 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4103 static int osd_declare_xattr_del(const struct lu_env *env,
4104 struct dt_object *dt, const char *name,
4105 struct thandle *handle)
4107 struct osd_thandle *oh;
4108 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4110 LASSERT(!dt_object_remote(dt));
4111 LASSERT(handle != NULL);
4113 oh = container_of0(handle, struct osd_thandle, ot_super);
4114 LASSERT(oh->ot_handle == NULL);
4116 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4117 osd_dto_credits_noquota[DTO_XATTR_SET]);
4119 * xattr del may involve inode quota change, reserve credits for
4120 * dquot_initialize()
4122 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4128 * Concurrency: @dt is write locked.
4130 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4131 const char *name, struct thandle *handle)
4133 struct osd_object *obj = osd_dt_obj(dt);
4134 struct inode *inode = obj->oo_inode;
4135 struct osd_thread_info *info = osd_oti_get(env);
4136 struct dentry *dentry = &info->oti_obj_dentry;
4139 if (!dt_object_exists(dt))
4142 LASSERT(!dt_object_remote(dt));
4143 LASSERT(inode->i_op != NULL);
4144 LASSERT(inode->i_op->removexattr != NULL);
4145 LASSERT(handle != NULL);
4147 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4149 ll_vfs_dq_init(inode);
4150 dentry->d_inode = inode;
4151 dentry->d_sb = inode->i_sb;
4152 rc = inode->i_op->removexattr(dentry, name);
4153 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4154 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4156 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4158 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4159 &info->oti_ost_attrs);
4161 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4166 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4167 lustre_lma_swab(lma);
4168 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4169 sizeof(*lma), XATTR_REPLACE);
4174 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4177 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4178 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4179 osd_oxc_del(obj, name);
4184 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4185 __u64 start, __u64 end)
4187 struct osd_object *obj = osd_dt_obj(dt);
4188 struct inode *inode = obj->oo_inode;
4189 struct osd_thread_info *info = osd_oti_get(env);
4190 struct dentry *dentry = &info->oti_obj_dentry;
4191 struct file *file = &info->oti_file;
4196 dentry->d_inode = inode;
4197 dentry->d_sb = inode->i_sb;
4198 file->f_path.dentry = dentry;
4199 file->f_mapping = inode->i_mapping;
4200 file->f_op = inode->i_fop;
4201 set_file_inode(file, inode);
4203 rc = ll_vfs_fsync_range(file, start, end, 0);
4208 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4217 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4218 const struct dt_index_features *feat)
4220 struct iam_descr *descr;
4222 if (osd_object_is_root(o))
4223 return feat == &dt_directory_features;
4225 LASSERT(o->oo_dir != NULL);
4227 descr = o->oo_dir->od_container.ic_descr;
4228 if (feat == &dt_directory_features) {
4229 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4235 feat->dif_keysize_min <= descr->id_key_size &&
4236 descr->id_key_size <= feat->dif_keysize_max &&
4237 feat->dif_recsize_min <= descr->id_rec_size &&
4238 descr->id_rec_size <= feat->dif_recsize_max &&
4239 !(feat->dif_flags & (DT_IND_VARKEY |
4240 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4241 ergo(feat->dif_flags & DT_IND_UPDATE,
4242 1 /* XXX check that object (and file system) is
4247 static int osd_iam_container_init(const struct lu_env *env,
4248 struct osd_object *obj,
4249 struct osd_directory *dir)
4251 struct iam_container *bag = &dir->od_container;
4254 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4258 result = iam_container_setup(bag);
4260 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4262 iam_container_fini(bag);
4269 * Concurrency: no external locking is necessary.
4271 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4272 const struct dt_index_features *feat)
4276 struct osd_object *obj = osd_dt_obj(dt);
4278 LINVRNT(osd_invariant(obj));
4280 if (osd_object_is_root(obj)) {
4281 dt->do_index_ops = &osd_index_ea_ops;
4283 } else if (feat == &dt_directory_features) {
4284 dt->do_index_ops = &osd_index_ea_ops;
4285 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4290 } else if (unlikely(feat == &dt_otable_features)) {
4291 dt->do_index_ops = &osd_otable_ops;
4293 } else if (unlikely(feat == &dt_acct_features)) {
4294 dt->do_index_ops = &osd_acct_index_ops;
4297 } else if (!osd_has_index(obj)) {
4298 struct osd_directory *dir;
4303 spin_lock(&obj->oo_guard);
4304 if (obj->oo_dir == NULL)
4308 * Concurrent thread allocated container data.
4311 spin_unlock(&obj->oo_guard);
4313 * Now, that we have container data, serialize its
4316 down_write(&obj->oo_ext_idx_sem);
4318 * recheck under lock.
4320 if (!osd_has_index(obj))
4321 result = osd_iam_container_init(env, obj,
4325 up_write(&obj->oo_ext_idx_sem);
4333 if (result == 0 && skip_iam == 0) {
4334 if (!osd_iam_index_probe(env, obj, feat))
4337 LINVRNT(osd_invariant(obj));
4342 static int osd_otable_it_attr_get(const struct lu_env *env,
4343 struct dt_object *dt,
4344 struct lu_attr *attr)
4350 static const struct dt_object_operations osd_obj_ops = {
4351 .do_read_lock = osd_read_lock,
4352 .do_write_lock = osd_write_lock,
4353 .do_read_unlock = osd_read_unlock,
4354 .do_write_unlock = osd_write_unlock,
4355 .do_write_locked = osd_write_locked,
4356 .do_attr_get = osd_attr_get,
4357 .do_declare_attr_set = osd_declare_attr_set,
4358 .do_attr_set = osd_attr_set,
4359 .do_ah_init = osd_ah_init,
4360 .do_declare_create = osd_declare_create,
4361 .do_create = osd_create,
4362 .do_declare_destroy = osd_declare_destroy,
4363 .do_destroy = osd_destroy,
4364 .do_index_try = osd_index_try,
4365 .do_declare_ref_add = osd_declare_ref_add,
4366 .do_ref_add = osd_ref_add,
4367 .do_declare_ref_del = osd_declare_ref_del,
4368 .do_ref_del = osd_ref_del,
4369 .do_xattr_get = osd_xattr_get,
4370 .do_declare_xattr_set = osd_declare_xattr_set,
4371 .do_xattr_set = osd_xattr_set,
4372 .do_declare_xattr_del = osd_declare_xattr_del,
4373 .do_xattr_del = osd_xattr_del,
4374 .do_xattr_list = osd_xattr_list,
4375 .do_object_sync = osd_object_sync,
4376 .do_invalidate = osd_invalidate,
4379 static const struct dt_object_operations osd_obj_otable_it_ops = {
4380 .do_attr_get = osd_otable_it_attr_get,
4381 .do_index_try = osd_index_try,
4384 static int osd_index_declare_iam_delete(const struct lu_env *env,
4385 struct dt_object *dt,
4386 const struct dt_key *key,
4387 struct thandle *handle)
4389 struct osd_thandle *oh;
4391 oh = container_of0(handle, struct osd_thandle, ot_super);
4392 LASSERT(oh->ot_handle == NULL);
4394 /* Recycle may cause additional three blocks to be changed. */
4395 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4396 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4402 * delete a (key, value) pair from index \a dt specified by \a key
4404 * \param dt osd index object
4405 * \param key key for index
4406 * \param rec record reference
4407 * \param handle transaction handler
4410 * \retval -ve failure
4412 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4413 const struct dt_key *key,
4414 struct thandle *handle)
4416 struct osd_thread_info *oti = osd_oti_get(env);
4417 struct osd_object *obj = osd_dt_obj(dt);
4418 struct osd_thandle *oh;
4419 struct iam_path_descr *ipd;
4420 struct iam_container *bag = &obj->oo_dir->od_container;
4424 if (!dt_object_exists(dt))
4427 LINVRNT(osd_invariant(obj));
4428 LASSERT(!dt_object_remote(dt));
4429 LASSERT(bag->ic_object == obj->oo_inode);
4430 LASSERT(handle != NULL);
4432 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4434 ipd = osd_idx_ipd_get(env, bag);
4435 if (unlikely(ipd == NULL))
4438 oh = container_of0(handle, struct osd_thandle, ot_super);
4439 LASSERT(oh->ot_handle != NULL);
4440 LASSERT(oh->ot_handle->h_transaction != NULL);
4442 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4443 /* swab quota uid/gid provided by caller */
4444 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4445 key = (const struct dt_key *)&oti->oti_quota_id;
4448 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4449 osd_ipd_put(env, bag, ipd);
4450 LINVRNT(osd_invariant(obj));
4451 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4455 static int osd_index_declare_ea_delete(const struct lu_env *env,
4456 struct dt_object *dt,
4457 const struct dt_key *key,
4458 struct thandle *handle)
4460 struct osd_thandle *oh;
4461 struct inode *inode;
4465 LASSERT(!dt_object_remote(dt));
4466 LASSERT(handle != NULL);
4468 oh = container_of0(handle, struct osd_thandle, ot_super);
4469 LASSERT(oh->ot_handle == NULL);
4471 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4472 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4473 /* '..' to a remote object has a local representative */
4474 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4475 /* to reset LMAI_REMOTE_PARENT */
4478 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4480 inode = osd_dt_obj(dt)->oo_inode;
4484 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4485 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4490 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4493 struct osd_fid_pack *rec;
4496 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4497 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4498 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4499 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4505 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4506 const struct lu_fid *fid)
4508 struct seq_server_site *ss = osd_seq_site(osd);
4511 /* FID seqs not in FLDB, must be local seq */
4512 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4515 /* If FLD is not being initialized yet, it only happens during the
4516 * initialization, likely during mgs initialization, and we assume
4517 * this is local FID. */
4518 if (ss == NULL || ss->ss_server_fld == NULL)
4521 /* Only check the local FLDB here */
4522 if (osd_seq_exists(env, osd, fid_seq(fid)))
4529 * Index delete function for interoperability mode (b11826).
4530 * It will remove the directory entry added by osd_index_ea_insert().
4531 * This entry is needed to maintain name->fid mapping.
4533 * \param key, key i.e. file entry to be deleted
4535 * \retval 0, on success
4536 * \retval -ve, on error
4538 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4539 const struct dt_key *key, struct thandle *handle)
4541 struct osd_object *obj = osd_dt_obj(dt);
4542 struct inode *dir = obj->oo_inode;
4543 struct dentry *dentry;
4544 struct osd_thandle *oh;
4545 struct ldiskfs_dir_entry_2 *de = NULL;
4546 struct buffer_head *bh;
4547 struct htree_lock *hlock = NULL;
4548 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4549 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4553 if (!dt_object_exists(dt))
4556 LINVRNT(osd_invariant(obj));
4557 LASSERT(!dt_object_remote(dt));
4558 LASSERT(handle != NULL);
4560 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4562 oh = container_of(handle, struct osd_thandle, ot_super);
4563 LASSERT(oh->ot_handle != NULL);
4564 LASSERT(oh->ot_handle->h_transaction != NULL);
4566 ll_vfs_dq_init(dir);
4567 dentry = osd_child_dentry_get(env, obj,
4568 (char *)key, strlen((char *)key));
4570 if (obj->oo_hl_head != NULL) {
4571 hlock = osd_oti_get(env)->oti_hlock;
4572 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4573 dir, LDISKFS_HLOCK_DEL);
4575 down_write(&obj->oo_ext_idx_sem);
4578 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4580 /* If this is not the ".." entry, it might be a remote DNE
4581 * entry and we need to check if the FID is for a remote
4582 * MDT. If the FID is not in the directory entry (e.g.
4583 * upgraded 1.8 filesystem without dirdata enabled) then
4584 * we need to get the FID from the LMA. For a remote directory
4585 * there HAS to be an LMA, it cannot be an IGIF inode in this
4588 * Delete the entry before the agent inode in order to
4589 * simplify error handling. At worst an error after deleting
4590 * the entry first might leak the agent inode afterward. The
4591 * reverse would need filesystem abort in case of error deleting
4592 * the entry after the agent had been removed, or leave a
4593 * dangling entry pointing at a random inode. */
4594 if (strcmp((char *)key, dotdot) != 0) {
4595 LASSERT(de != NULL);
4596 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4597 if (rc == -ENODATA) {
4598 /* can't get FID, postpone to the end of the
4599 * transaction when iget() is safe */
4600 osd_schedule_agent_inode_removal(env, oh,
4601 le32_to_cpu(de->inode));
4602 } else if (rc == 0 &&
4603 unlikely(osd_remote_fid(env, osd, fid))) {
4604 osd_schedule_agent_inode_removal(env, oh,
4605 le32_to_cpu(de->inode));
4608 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4614 ldiskfs_htree_unlock(hlock);
4616 up_write(&obj->oo_ext_idx_sem);
4621 /* For inode on the remote MDT, .. will point to
4622 * /Agent directory, Check whether it needs to delete
4623 * from agent directory */
4624 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4627 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4630 /* Sigh, the entry has been deleted, and
4631 * it is not easy to revert it back, so
4632 * let's keep this error private, and let
4633 * LFSCK fix it. XXX */
4634 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4635 osd_name(osd), PFID(fid), ret);
4638 LASSERT(osd_invariant(obj));
4639 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4644 * Lookup index for \a key and copy record to \a rec.
4646 * \param dt osd index object
4647 * \param key key for index
4648 * \param rec record reference
4650 * \retval +ve success : exact mach
4651 * \retval 0 return record with key not greater than \a key
4652 * \retval -ve failure
4654 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4655 struct dt_rec *rec, const struct dt_key *key)
4657 struct osd_object *obj = osd_dt_obj(dt);
4658 struct iam_path_descr *ipd;
4659 struct iam_container *bag = &obj->oo_dir->od_container;
4660 struct osd_thread_info *oti = osd_oti_get(env);
4661 struct iam_iterator *it = &oti->oti_idx_it;
4662 struct iam_rec *iam_rec;
4666 if (!dt_object_exists(dt))
4669 LASSERT(osd_invariant(obj));
4670 LASSERT(!dt_object_remote(dt));
4671 LASSERT(bag->ic_object == obj->oo_inode);
4673 ipd = osd_idx_ipd_get(env, bag);
4677 /* got ipd now we can start iterator. */
4678 iam_it_init(it, bag, 0, ipd);
4680 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4681 /* swab quota uid/gid provided by caller */
4682 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4683 key = (const struct dt_key *)&oti->oti_quota_id;
4686 rc = iam_it_get(it, (struct iam_key *)key);
4688 if (S_ISDIR(obj->oo_inode->i_mode))
4689 iam_rec = (struct iam_rec *)oti->oti_ldp;
4691 iam_rec = (struct iam_rec *) rec;
4693 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4695 if (S_ISDIR(obj->oo_inode->i_mode))
4696 osd_fid_unpack((struct lu_fid *) rec,
4697 (struct osd_fid_pack *)iam_rec);
4698 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4699 osd_quota_unpack(obj, rec);
4704 osd_ipd_put(env, bag, ipd);
4706 LINVRNT(osd_invariant(obj));
4711 static int osd_index_declare_iam_insert(const struct lu_env *env,
4712 struct dt_object *dt,
4713 const struct dt_rec *rec,
4714 const struct dt_key *key,
4715 struct thandle *handle)
4717 struct osd_thandle *oh;
4719 LASSERT(handle != NULL);
4721 oh = container_of0(handle, struct osd_thandle, ot_super);
4722 LASSERT(oh->ot_handle == NULL);
4724 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4725 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4731 * Inserts (key, value) pair in \a dt index object.
4733 * \param dt osd index object
4734 * \param key key for index
4735 * \param rec record reference
4736 * \param th transaction handler
4739 * \retval -ve failure
4741 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4742 const struct dt_rec *rec,
4743 const struct dt_key *key, struct thandle *th,
4746 struct osd_object *obj = osd_dt_obj(dt);
4747 struct iam_path_descr *ipd;
4748 struct osd_thandle *oh;
4749 struct iam_container *bag;
4750 struct osd_thread_info *oti = osd_oti_get(env);
4751 struct iam_rec *iam_rec;
4755 if (!dt_object_exists(dt))
4758 LINVRNT(osd_invariant(obj));
4759 LASSERT(!dt_object_remote(dt));
4761 bag = &obj->oo_dir->od_container;
4762 LASSERT(bag->ic_object == obj->oo_inode);
4763 LASSERT(th != NULL);
4765 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4767 ipd = osd_idx_ipd_get(env, bag);
4768 if (unlikely(ipd == NULL))
4771 oh = container_of0(th, struct osd_thandle, ot_super);
4772 LASSERT(oh->ot_handle != NULL);
4773 LASSERT(oh->ot_handle->h_transaction != NULL);
4774 if (S_ISDIR(obj->oo_inode->i_mode)) {
4775 iam_rec = (struct iam_rec *)oti->oti_ldp;
4776 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4777 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4778 /* pack quota uid/gid */
4779 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4780 key = (const struct dt_key *)&oti->oti_quota_id;
4781 /* pack quota record */
4782 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4783 iam_rec = (struct iam_rec *)rec;
4785 iam_rec = (struct iam_rec *)rec;
4788 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4790 osd_ipd_put(env, bag, ipd);
4791 LINVRNT(osd_invariant(obj));
4792 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4797 * Calls ldiskfs_add_entry() to add directory entry
4798 * into the directory. This is required for
4799 * interoperability mode (b11826)
4801 * \retval 0, on success
4802 * \retval -ve, on error
4804 static int __osd_ea_add_rec(struct osd_thread_info *info,
4805 struct osd_object *pobj, struct inode *cinode,
4806 const char *name, const struct lu_fid *fid,
4807 struct htree_lock *hlock, struct thandle *th)
4809 struct ldiskfs_dentry_param *ldp;
4810 struct dentry *child;
4811 struct osd_thandle *oth;
4814 oth = container_of(th, struct osd_thandle, ot_super);
4815 LASSERT(oth->ot_handle != NULL);
4816 LASSERT(oth->ot_handle->h_transaction != NULL);
4817 LASSERT(pobj->oo_inode);
4819 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4820 if (unlikely(pobj->oo_inode ==
4821 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4824 osd_get_ldiskfs_dirent_param(ldp, fid);
4825 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4826 child->d_fsdata = (void *)ldp;
4827 ll_vfs_dq_init(pobj->oo_inode);
4828 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4829 child, cinode, hlock);
4830 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4831 struct ldiskfs_dir_entry_2 *de;
4832 struct buffer_head *bh;
4835 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4838 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4841 if (S_ISDIR(cinode->i_mode))
4842 de->file_type = LDISKFS_DIRENT_LUFID |
4843 LDISKFS_FT_REG_FILE;
4845 de->file_type = LDISKFS_DIRENT_LUFID |
4847 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4858 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4859 * into the directory.Also sets flags into osd object to
4860 * indicate dot and dotdot are created. This is required for
4861 * interoperability mode (b11826)
4863 * \param dir directory for dot and dotdot fixup.
4864 * \param obj child object for linking
4866 * \retval 0, on success
4867 * \retval -ve, on error
4869 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4870 struct osd_object *dir,
4871 struct inode *parent_dir, const char *name,
4872 const struct lu_fid *dot_fid,
4873 const struct lu_fid *dot_dot_fid,
4876 struct inode *inode = dir->oo_inode;
4877 struct osd_thandle *oth;
4880 oth = container_of(th, struct osd_thandle, ot_super);
4881 LASSERT(oth->ot_handle->h_transaction != NULL);
4882 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4884 if (strcmp(name, dot) == 0) {
4885 if (dir->oo_compat_dot_created) {
4888 LASSERT(inode->i_ino == parent_dir->i_ino);
4889 dir->oo_compat_dot_created = 1;
4892 } else if (strcmp(name, dotdot) == 0) {
4893 if (!dir->oo_compat_dot_created)
4895 /* in case of rename, dotdot is already created */
4896 if (dir->oo_compat_dotdot_created) {
4897 return __osd_ea_add_rec(info, dir, parent_dir, name,
4898 dot_dot_fid, NULL, th);
4901 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4902 struct lu_fid tfid = *dot_dot_fid;
4905 result = osd_add_dot_dotdot_internal(info,
4906 dir->oo_inode, parent_dir, dot_fid,
4909 result = osd_add_dot_dotdot_internal(info,
4910 dir->oo_inode, parent_dir, dot_fid,
4915 dir->oo_compat_dotdot_created = 1;
4923 * It will call the appropriate osd_add* function and return the
4924 * value, return by respective functions.
4926 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4927 struct inode *cinode, const char *name,
4928 const struct lu_fid *fid, struct thandle *th)
4930 struct osd_thread_info *info = osd_oti_get(env);
4931 struct htree_lock *hlock;
4934 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4936 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4938 if (hlock != NULL) {
4939 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4942 down_write(&pobj->oo_ext_idx_sem);
4945 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4946 lu_object_fid(&pobj->oo_dt.do_lu),
4949 if (hlock != NULL) {
4950 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4951 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4953 down_write(&pobj->oo_ext_idx_sem);
4956 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4957 struct lu_fid *tfid = &info->oti_fid;
4961 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4964 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4969 ldiskfs_htree_unlock(hlock);
4971 up_write(&pobj->oo_ext_idx_sem);
4977 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4978 struct osd_idmap_cache *oic)
4980 struct osd_scrub *scrub = &dev->od_scrub;
4981 struct lu_fid *fid = &oic->oic_fid;
4982 struct osd_inode_id *id = &oic->oic_lid;
4983 struct inode *inode = NULL;
4989 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4992 if (scrub->os_pos_current > id->oii_ino)
4996 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4997 if (rc == -ENOENT) {
4998 __u32 gen = id->oii_gen;
5004 inode = osd_iget(oti, dev, id);
5005 /* The inode has been removed (by race maybe). */
5006 if (IS_ERR(inode)) {
5007 rc = PTR_ERR(inode);
5009 RETURN(rc == -ESTALE ? -ENOENT : rc);
5012 /* The OI mapping is lost. */
5013 if (gen != OSD_OII_NOGEN)
5016 /* The inode may has been reused by others, we do not know,
5017 * leave it to be handled by subsequent osd_fid_lookup(). */
5019 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5026 if (thread_is_running(&scrub->os_thread)) {
5027 if (inode == NULL) {
5028 inode = osd_iget(oti, dev, id);
5029 /* The inode has been removed (by race maybe). */
5030 if (IS_ERR(inode)) {
5031 rc = PTR_ERR(inode);
5033 RETURN(rc == -ESTALE ? -ENOENT : rc);
5037 rc = osd_oii_insert(dev, oic, insert);
5038 /* There is race condition between osd_oi_lookup and OI scrub.
5039 * The OI scrub finished just after osd_oi_lookup() failure.
5040 * Under such case, it is unnecessary to trigger OI scrub again,
5041 * but try to call osd_oi_lookup() again. */
5042 if (unlikely(rc == -EAGAIN))
5045 if (!S_ISDIR(inode->i_mode))
5048 rc = osd_check_lmv(oti, dev, inode, oic);
5053 if (!dev->od_noscrub && ++once == 1) {
5054 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5056 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5057 "%s: trigger partial OI scrub for RPC inconsistency "
5058 "checking FID "DFID": rc = %d\n",
5059 osd_dev2name(dev), PFID(fid), rc);
5060 if (rc == 0 || rc == -EALREADY)
5073 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5074 struct osd_device *dev,
5075 struct osd_idmap_cache *oic,
5076 struct lu_fid *fid, __u32 ino)
5078 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5079 struct inode *inode;
5082 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5083 inode = osd_iget(oti, dev, &oic->oic_lid);
5084 if (IS_ERR(inode)) {
5085 fid_zero(&oic->oic_fid);
5086 return PTR_ERR(inode);
5089 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5092 fid_zero(&oic->oic_fid);
5094 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5098 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5099 struct osd_inode_id *id, const struct lu_fid *fid)
5101 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5102 id->oii_ino, id->oii_gen, info);
5103 info->oti_cache.oic_lid = *id;
5104 info->oti_cache.oic_fid = *fid;
5105 info->oti_cache.oic_dev = osd;
5109 * Get parent FID from the linkEA.
5111 * For a directory which parent resides on remote MDT, to satisfy the
5112 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5113 * the other hand, to make the lookup(..) on the directory can return
5114 * the real parent FID, we append the real parent FID after its ".."
5115 * name entry in the /REMOTE_PARENT_DIR.
5117 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5118 * backup. So after the restore, we cannot get the right parent FID from
5119 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5120 * we have stored the real parent FID in the directory object's linkEA,
5121 * we can parse the linkEA for the real parent FID.
5123 * \param[in] env pointer to the thread context
5124 * \param[in] obj pointer to the object to be handled
5125 * \param[out]fid pointer to the buffer to hold the parent FID
5127 * \retval 0 for getting the real parent FID successfully
5128 * \retval negative error number on failure
5130 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5131 struct osd_object *obj,
5134 struct osd_thread_info *oti = osd_oti_get(env);
5135 struct lu_buf *buf = &oti->oti_big_buf;
5136 struct dentry *dentry = &oti->oti_obj_dentry;
5137 struct inode *inode = obj->oo_inode;
5138 struct linkea_data ldata = { NULL };
5143 if (!S_ISDIR(inode->i_mode))
5147 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5148 buf->lb_buf, buf->lb_len);
5149 if (rc == -ERANGE) {
5150 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5153 lu_buf_realloc(buf, rc);
5154 if (buf->lb_buf == NULL)
5161 if (unlikely(rc == 0))
5167 if (unlikely(buf->lb_buf == NULL)) {
5168 lu_buf_realloc(buf, rc);
5169 if (buf->lb_buf == NULL)
5176 rc = linkea_init_with_rec(&ldata);
5178 linkea_first_entry(&ldata);
5179 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5185 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5186 struct inode *inode,
5187 const struct lu_fid *pfid,
5188 const char *name, const int namelen)
5190 struct osd_thread_info *oti = osd_oti_get(env);
5191 struct lu_buf *buf = &oti->oti_big_buf;
5192 struct dentry *dentry = &oti->oti_obj_dentry;
5193 struct linkea_data ldata = { NULL };
5194 struct lu_name cname = { .ln_name = name,
5195 .ln_namelen = namelen };
5200 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5201 buf->lb_buf, buf->lb_len);
5203 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5208 if (unlikely(rc == 0))
5211 if (buf->lb_len < rc) {
5212 lu_buf_realloc(buf, rc);
5213 if (buf->lb_buf == NULL)
5220 rc = linkea_init_with_rec(&ldata);
5222 rc = linkea_links_find(&ldata, &cname, pfid);
5228 * Calls ->lookup() to find dentry. From dentry get inode and
5229 * read inode's ea to get fid. This is required for interoperability
5232 * \retval 0, on success
5233 * \retval -ve, on error
5235 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5236 struct dt_rec *rec, const struct dt_key *key)
5238 struct inode *dir = obj->oo_inode;
5239 struct dentry *dentry;
5240 struct ldiskfs_dir_entry_2 *de;
5241 struct buffer_head *bh;
5242 struct lu_fid *fid = (struct lu_fid *) rec;
5243 struct htree_lock *hlock = NULL;
5248 LASSERT(dir->i_op != NULL);
5249 LASSERT(dir->i_op->lookup != NULL);
5251 dentry = osd_child_dentry_get(env, obj,
5252 (char *)key, strlen((char *)key));
5254 if (obj->oo_hl_head != NULL) {
5255 hlock = osd_oti_get(env)->oti_hlock;
5256 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5257 dir, LDISKFS_HLOCK_LOOKUP);
5259 down_read(&obj->oo_ext_idx_sem);
5262 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5264 struct osd_thread_info *oti = osd_oti_get(env);
5265 struct osd_inode_id *id = &oti->oti_id;
5266 struct osd_idmap_cache *oic = &oti->oti_cache;
5267 struct osd_device *dev = osd_obj2dev(obj);
5269 ino = le32_to_cpu(de->inode);
5270 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5272 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5276 rc = osd_get_fid_from_dentry(de, rec);
5278 /* done with de, release bh */
5281 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5282 const char *name = (const char *)key;
5284 /* If the parent is on remote MDT, and there
5285 * is no FID-in-dirent, then we have to get
5286 * the parent FID from the linkEA. */
5287 if (likely(strlen(name) == 2 &&
5288 name[0] == '.' && name[1] == '.'))
5289 rc = osd_get_pfid_from_linkea(env, obj,
5292 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5295 osd_id_gen(id, ino, OSD_OII_NOGEN);
5298 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5299 fid_zero(&oic->oic_fid);
5304 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5305 rc = osd_consistency_check(oti, dev, oic);
5307 fid_zero(&oic->oic_fid);
5316 ldiskfs_htree_unlock(hlock);
5318 up_read(&obj->oo_ext_idx_sem);
5322 static int osd_index_declare_ea_insert(const struct lu_env *env,
5323 struct dt_object *dt,
5324 const struct dt_rec *rec,
5325 const struct dt_key *key,
5326 struct thandle *handle)
5328 struct osd_thandle *oh;
5329 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5330 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5331 const struct lu_fid *fid = rec1->rec_fid;
5332 int credits, rc = 0;
5333 struct osd_idmap_cache *idc;
5336 LASSERT(!dt_object_remote(dt));
5337 LASSERT(handle != NULL);
5338 LASSERT(fid != NULL);
5339 LASSERT(rec1->rec_type != 0);
5341 oh = container_of0(handle, struct osd_thandle, ot_super);
5342 LASSERT(oh->ot_handle == NULL);
5344 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5346 /* we can't call iget() while a transactions is running
5347 * (this can lead to a deadlock), but we need to know
5348 * inum and object type. so we find this information at
5349 * declaration and cache in per-thread info */
5350 idc = osd_idc_find_or_init(env, osd, fid);
5352 RETURN(PTR_ERR(idc));
5353 if (idc->oic_remote) {
5354 /* a reference to remote inode is represented by an
5355 * agent inode which we have to create */
5356 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5357 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5360 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5362 if (osd_dt_obj(dt)->oo_inode != NULL) {
5363 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5365 /* We ignore block quota on meta pool (MDTs), so needn't
5366 * calculate how many blocks will be consumed by this index
5368 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5370 i_projid_read(inode), 0,
5371 oh, osd_dt_obj(dt), NULL,
5376 #ifdef HAVE_PROJECT_QUOTA
5377 /* Reserve credits for local agent inode to transfer
5378 * to 0, quota enforcement is ignored in this case.
5380 if (idc->oic_remote &&
5381 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5382 i_projid_read(inode) != 0)
5383 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5384 0, i_projid_read(inode),
5385 0, false, PRJQUOTA);
5394 * Index add function for interoperability mode (b11826).
5395 * It will add the directory entry.This entry is needed to
5396 * maintain name->fid mapping.
5398 * \param key it is key i.e. file entry to be inserted
5399 * \param rec it is value of given key i.e. fid
5401 * \retval 0, on success
5402 * \retval -ve, on error
5404 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5405 const struct dt_rec *rec,
5406 const struct dt_key *key, struct thandle *th,
5409 struct osd_object *obj = osd_dt_obj(dt);
5410 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5411 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5412 const struct lu_fid *fid = rec1->rec_fid;
5413 const char *name = (const char *)key;
5414 struct osd_thread_info *oti = osd_oti_get(env);
5415 struct inode *child_inode = NULL;
5416 struct osd_idmap_cache *idc;
5420 if (!dt_object_exists(dt))
5423 LASSERT(osd_invariant(obj));
5424 LASSERT(!dt_object_remote(dt));
5425 LASSERT(th != NULL);
5427 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5429 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5431 idc = osd_idc_find(env, osd, fid);
5432 if (unlikely(idc == NULL)) {
5433 /* this dt_insert() wasn't declared properly, so
5434 * FID is missing in OI cache. we better do not
5435 * lookup FID in FLDB/OI and don't risk to deadlock,
5436 * but in some special cases (lfsck testing, etc)
5437 * it's much simpler than fixing a caller */
5438 CERROR("%s: "DFID" wasn't declared for insert\n",
5439 osd_name(osd), PFID(fid));
5441 idc = osd_idc_find_or_init(env, osd, fid);
5443 RETURN(PTR_ERR(idc));
5446 if (idc->oic_remote) {
5447 /* Insert remote entry */
5448 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5449 struct osd_mdobj_map *omm = osd->od_mdt_map;
5450 struct osd_thandle *oh;
5452 /* If parent on remote MDT, we need put this object
5454 oh = container_of(th, typeof(*oh), ot_super);
5455 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5457 CERROR("%s: add "DFID" error: rc = %d\n",
5459 PFID(lu_object_fid(&dt->do_lu)), rc);
5463 child_inode = igrab(omm->omm_remote_parent->d_inode);
5465 child_inode = osd_create_local_agent_inode(env, osd,
5466 obj, fid, rec1->rec_type & S_IFMT, th);
5467 if (IS_ERR(child_inode))
5468 RETURN(PTR_ERR(child_inode));
5471 /* Insert local entry */
5472 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5473 /* for a reason OI cache wasn't filled properly */
5474 CERROR("%s: OIC for "DFID" isn't filled\n",
5475 osd_name(osd), PFID(fid));
5478 child_inode = oti->oti_inode;
5479 if (unlikely(child_inode == NULL)) {
5480 struct ldiskfs_inode_info *lii;
5484 child_inode = oti->oti_inode = &lii->vfs_inode;
5486 child_inode->i_sb = osd_sb(osd);
5487 child_inode->i_ino = idc->oic_lid.oii_ino;
5488 child_inode->i_mode = rec1->rec_type & S_IFMT;
5491 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5493 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5494 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5496 if (child_inode && child_inode != oti->oti_inode)
5498 LASSERT(osd_invariant(obj));
5499 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5504 * Initialize osd Iterator for given osd index object.
5506 * \param dt osd index object
5509 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5510 struct dt_object *dt,
5513 struct osd_it_iam *it;
5514 struct osd_object *obj = osd_dt_obj(dt);
5515 struct lu_object *lo = &dt->do_lu;
5516 struct iam_path_descr *ipd;
5517 struct iam_container *bag = &obj->oo_dir->od_container;
5519 if (!dt_object_exists(dt))
5520 return ERR_PTR(-ENOENT);
5524 return ERR_PTR(-ENOMEM);
5526 ipd = osd_it_ipd_get(env, bag);
5527 if (likely(ipd != NULL)) {
5531 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5532 return (struct dt_it *)it;
5535 return ERR_PTR(-ENOMEM);
5540 * free given Iterator.
5542 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5544 struct osd_it_iam *it = (struct osd_it_iam *)di;
5545 struct osd_object *obj = it->oi_obj;
5547 iam_it_fini(&it->oi_it);
5548 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5549 osd_object_put(env, obj);
5554 * Move Iterator to record specified by \a key
5556 * \param di osd iterator
5557 * \param key key for index
5559 * \retval +ve di points to record with least key not larger than key
5560 * \retval 0 di points to exact matched key
5561 * \retval -ve failure
5564 static int osd_it_iam_get(const struct lu_env *env,
5565 struct dt_it *di, const struct dt_key *key)
5567 struct osd_thread_info *oti = osd_oti_get(env);
5568 struct osd_it_iam *it = (struct osd_it_iam *)di;
5570 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5571 /* swab quota uid/gid */
5572 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5573 key = (struct dt_key *)&oti->oti_quota_id;
5576 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5582 * \param di osd iterator
5584 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5586 struct osd_it_iam *it = (struct osd_it_iam *)di;
5588 iam_it_put(&it->oi_it);
5592 * Move iterator by one record
5594 * \param di osd iterator
5596 * \retval +1 end of container reached
5598 * \retval -ve failure
5601 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5603 struct osd_it_iam *it = (struct osd_it_iam *)di;
5605 return iam_it_next(&it->oi_it);
5609 * Return pointer to the key under iterator.
5612 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5613 const struct dt_it *di)
5615 struct osd_thread_info *oti = osd_oti_get(env);
5616 struct osd_it_iam *it = (struct osd_it_iam *)di;
5617 struct osd_object *obj = it->oi_obj;
5620 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5622 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5623 /* swab quota uid/gid */
5624 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5625 key = (struct dt_key *)&oti->oti_quota_id;
5632 * Return size of key under iterator (in bytes)
5635 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5637 struct osd_it_iam *it = (struct osd_it_iam *)di;
5639 return iam_it_key_size(&it->oi_it);
5643 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5645 /* check if file type is required */
5646 if (ent->lde_attrs & LUDA_TYPE) {
5647 struct luda_type *lt;
5648 int align = sizeof(*lt) - 1;
5650 len = (len + align) & ~align;
5651 lt = (struct luda_type *)(ent->lde_name + len);
5652 lt->lt_type = cpu_to_le16(DTTOIF(type));
5655 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5659 * build lu direct from backend fs dirent.
5663 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5664 char *name, __u16 namelen, __u16 type, __u32 attr)
5666 ent->lde_attrs = attr | LUDA_FID;
5667 fid_cpu_to_le(&ent->lde_fid, fid);
5669 ent->lde_hash = cpu_to_le64(offset);
5670 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5672 strncpy(ent->lde_name, name, namelen);
5673 ent->lde_name[namelen] = '\0';
5674 ent->lde_namelen = cpu_to_le16(namelen);
5676 /* append lustre attributes */
5677 osd_it_append_attrs(ent, namelen, type);
5681 * Return pointer to the record under iterator.
5683 static int osd_it_iam_rec(const struct lu_env *env,
5684 const struct dt_it *di,
5685 struct dt_rec *dtrec, __u32 attr)
5687 struct osd_it_iam *it = (struct osd_it_iam *)di;
5688 struct osd_thread_info *info = osd_oti_get(env);
5691 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5692 const struct osd_fid_pack *rec;
5693 struct lu_fid *fid = &info->oti_fid;
5694 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5700 name = (char *)iam_it_key_get(&it->oi_it);
5702 RETURN(PTR_ERR(name));
5704 namelen = iam_it_key_size(&it->oi_it);
5706 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5708 RETURN(PTR_ERR(rec));
5710 rc = osd_fid_unpack(fid, rec);
5714 hash = iam_it_store(&it->oi_it);
5716 /* IAM does not store object type in IAM index (dir) */
5717 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5719 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5720 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5721 (struct iam_rec *)dtrec);
5722 osd_quota_unpack(it->oi_obj, dtrec);
5724 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5725 (struct iam_rec *)dtrec);
5732 * Returns cookie for current Iterator position.
5734 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5736 struct osd_it_iam *it = (struct osd_it_iam *)di;
5738 return iam_it_store(&it->oi_it);
5742 * Restore iterator from cookie.
5744 * \param di osd iterator
5745 * \param hash Iterator location cookie
5747 * \retval +ve di points to record with least key not larger than key.
5748 * \retval 0 di points to exact matched key
5749 * \retval -ve failure
5752 static int osd_it_iam_load(const struct lu_env *env,
5753 const struct dt_it *di, __u64 hash)
5755 struct osd_it_iam *it = (struct osd_it_iam *)di;
5757 return iam_it_load(&it->oi_it, hash);
5760 static const struct dt_index_operations osd_index_iam_ops = {
5761 .dio_lookup = osd_index_iam_lookup,
5762 .dio_declare_insert = osd_index_declare_iam_insert,
5763 .dio_insert = osd_index_iam_insert,
5764 .dio_declare_delete = osd_index_declare_iam_delete,
5765 .dio_delete = osd_index_iam_delete,
5767 .init = osd_it_iam_init,
5768 .fini = osd_it_iam_fini,
5769 .get = osd_it_iam_get,
5770 .put = osd_it_iam_put,
5771 .next = osd_it_iam_next,
5772 .key = osd_it_iam_key,
5773 .key_size = osd_it_iam_key_size,
5774 .rec = osd_it_iam_rec,
5775 .store = osd_it_iam_store,
5776 .load = osd_it_iam_load
5782 * Creates or initializes iterator context.
5784 * \retval struct osd_it_ea, iterator structure on success
5787 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5788 struct dt_object *dt,
5791 struct osd_object *obj = osd_dt_obj(dt);
5792 struct osd_thread_info *info = osd_oti_get(env);
5793 struct osd_it_ea *oie;
5795 struct lu_object *lo = &dt->do_lu;
5796 struct dentry *obj_dentry;
5799 if (!dt_object_exists(dt) || obj->oo_destroyed)
5800 RETURN(ERR_PTR(-ENOENT));
5802 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5804 RETURN(ERR_PTR(-ENOMEM));
5805 obj_dentry = &oie->oie_dentry;
5807 obj_dentry->d_inode = obj->oo_inode;
5808 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5809 obj_dentry->d_name.hash = 0;
5811 oie->oie_rd_dirent = 0;
5812 oie->oie_it_dirent = 0;
5813 oie->oie_dirent = NULL;
5814 if (unlikely(!info->oti_it_ea_buf_used)) {
5815 oie->oie_buf = info->oti_it_ea_buf;
5816 info->oti_it_ea_buf_used = 1;
5818 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5819 if (oie->oie_buf == NULL)
5820 RETURN(ERR_PTR(-ENOMEM));
5824 file = &oie->oie_file;
5826 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5827 if (attr & LUDA_64BITHASH)
5828 file->f_mode = FMODE_64BITHASH;
5830 file->f_mode = FMODE_32BITHASH;
5831 file->f_path.dentry = obj_dentry;
5832 file->f_mapping = obj->oo_inode->i_mapping;
5833 file->f_op = obj->oo_inode->i_fop;
5834 set_file_inode(file, obj->oo_inode);
5837 RETURN((struct dt_it *) oie);
5841 * Destroy or finishes iterator context.
5843 * \param di iterator structure to be destroyed
5845 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5847 struct osd_thread_info *info = osd_oti_get(env);
5848 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5849 struct osd_object *obj = oie->oie_obj;
5850 struct inode *inode = obj->oo_inode;
5853 oie->oie_file.f_op->release(inode, &oie->oie_file);
5854 osd_object_put(env, obj);
5855 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5856 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5858 info->oti_it_ea_buf_used = 0;
5859 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5864 * It position the iterator at given key, so that next lookup continues from
5865 * that key Or it is similar to dio_it->load() but based on a key,
5866 * rather than file position.
5868 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5871 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5873 static int osd_it_ea_get(const struct lu_env *env,
5874 struct dt_it *di, const struct dt_key *key)
5876 struct osd_it_ea *it = (struct osd_it_ea *)di;
5879 LASSERT(((const char *)key)[0] == '\0');
5880 it->oie_file.f_pos = 0;
5881 it->oie_rd_dirent = 0;
5882 it->oie_it_dirent = 0;
5883 it->oie_dirent = NULL;
5891 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5895 struct osd_filldir_cbs {
5896 #ifdef HAVE_DIR_CONTEXT
5897 struct dir_context ctx;
5899 struct osd_it_ea *it;
5902 * It is called internally by ->readdir(). It fills the
5903 * iterator's in-memory data structure with required
5904 * information i.e. name, namelen, rec_size etc.
5906 * \param buf in which information to be filled in.
5907 * \param name name of the file in given dir
5909 * \retval 0 on success
5910 * \retval 1 on buffer full
5912 #ifdef HAVE_FILLDIR_USE_CTX
5913 static int osd_ldiskfs_filldir(struct dir_context *buf,
5915 static int osd_ldiskfs_filldir(void *buf,
5917 const char *name, int namelen,
5918 loff_t offset, __u64 ino, unsigned d_type)
5920 struct osd_it_ea *it =
5921 ((struct osd_filldir_cbs *)buf)->it;
5922 struct osd_object *obj = it->oie_obj;
5923 struct osd_it_ea_dirent *ent = it->oie_dirent;
5924 struct lu_fid *fid = &ent->oied_fid;
5925 struct osd_fid_pack *rec;
5928 /* this should never happen */
5929 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5930 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5934 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5938 /* "." is just the object itself. */
5939 if (namelen == 1 && name[0] == '.') {
5940 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5941 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5942 rec = (struct osd_fid_pack*) (name + namelen + 1);
5943 if (osd_fid_unpack(fid, rec) != 0)
5948 d_type &= ~LDISKFS_DIRENT_LUFID;
5950 /* NOT export local root. */
5951 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5952 ino = obj->oo_inode->i_ino;
5953 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5956 ent->oied_ino = ino;
5957 ent->oied_off = offset;
5958 ent->oied_namelen = namelen;
5959 ent->oied_type = d_type;
5961 memcpy(ent->oied_name, name, namelen);
5963 it->oie_rd_dirent++;
5964 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5969 * Calls ->readdir() to load a directory entry at a time
5970 * and stored it in iterator's in-memory data structure.
5972 * \param di iterator's in memory structure
5974 * \retval 0 on success
5975 * \retval -ve on error
5976 * \retval +1 reach the end of entry
5978 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5979 const struct dt_it *di)
5981 struct osd_it_ea *it = (struct osd_it_ea *)di;
5982 struct osd_object *obj = it->oie_obj;
5983 struct inode *inode = obj->oo_inode;
5984 struct htree_lock *hlock = NULL;
5985 struct file *filp = &it->oie_file;
5987 struct osd_filldir_cbs buf = {
5988 #ifdef HAVE_DIR_CONTEXT
5989 .ctx.actor = osd_ldiskfs_filldir,
5995 it->oie_dirent = it->oie_buf;
5996 it->oie_rd_dirent = 0;
5998 if (obj->oo_hl_head != NULL) {
5999 hlock = osd_oti_get(env)->oti_hlock;
6000 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6001 inode, LDISKFS_HLOCK_READDIR);
6003 down_read(&obj->oo_ext_idx_sem);
6006 #ifdef HAVE_DIR_CONTEXT
6007 buf.ctx.pos = filp->f_pos;
6008 rc = inode->i_fop->iterate(filp, &buf.ctx);
6009 filp->f_pos = buf.ctx.pos;
6011 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6015 ldiskfs_htree_unlock(hlock);
6017 up_read(&obj->oo_ext_idx_sem);
6019 if (it->oie_rd_dirent == 0) {
6020 /*If it does not get any dirent, it means it has been reached
6021 *to the end of the dir */
6022 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6026 it->oie_dirent = it->oie_buf;
6027 it->oie_it_dirent = 1;
6034 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6035 * to load a directory entry at a time and stored it in
6036 * iterator's in-memory data structure.
6038 * \param di iterator's in memory structure
6040 * \retval +ve iterator reached to end
6041 * \retval 0 iterator not reached to end
6042 * \retval -ve on error
6044 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6046 struct osd_it_ea *it = (struct osd_it_ea *)di;
6051 if (it->oie_it_dirent < it->oie_rd_dirent) {
6053 (void *) it->oie_dirent +
6054 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6055 it->oie_dirent->oied_namelen);
6056 it->oie_it_dirent++;
6059 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6062 rc = osd_ldiskfs_it_fill(env, di);
6069 * Returns the key at current position from iterator's in memory structure.
6071 * \param di iterator's in memory structure
6073 * \retval key i.e. struct dt_key on success
6075 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6076 const struct dt_it *di)
6078 struct osd_it_ea *it = (struct osd_it_ea *)di;
6080 return (struct dt_key *)it->oie_dirent->oied_name;
6084 * Returns the key's size at current position from iterator's in memory structure.
6086 * \param di iterator's in memory structure
6088 * \retval key_size i.e. struct dt_key on success
6090 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6092 struct osd_it_ea *it = (struct osd_it_ea *)di;
6094 return it->oie_dirent->oied_namelen;
6097 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6099 if (LDISKFS_DIR_REC_LEN(de) >=
6100 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6107 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6108 unsigned blocksize, bool dotdot)
6111 return osd_dotdot_has_space(de);
6113 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6114 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6121 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6122 handle_t *jh, struct dentry *dentry,
6123 const struct lu_fid *fid, struct buffer_head *bh,
6124 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6127 struct inode *dir = dentry->d_parent->d_inode;
6128 struct inode *inode = dentry->d_inode;
6129 struct osd_fid_pack *rec;
6130 struct ldiskfs_dentry_param *ldp;
6131 int namelen = dentry->d_name.len;
6133 struct osd_thread_info *info = osd_oti_get(env);
6136 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6137 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6140 /* There is enough space to hold the FID-in-dirent. */
6141 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6142 rc = ldiskfs_journal_get_write_access(jh, bh);
6146 de->name[namelen] = 0;
6147 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6148 rec->fp_len = sizeof(struct lu_fid) + 1;
6149 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6150 de->file_type |= LDISKFS_DIRENT_LUFID;
6151 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6158 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6162 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6163 osd_get_ldiskfs_dirent_param(ldp, fid);
6164 dentry->d_fsdata = (void *)ldp;
6165 ll_vfs_dq_init(dir);
6166 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6167 /* It is too bad, we cannot reinsert the name entry back.
6168 * That means we lose it! */
6170 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6171 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6172 osd_ino2name(inode),
6173 dir->i_ino, dir->i_generation, namelen,
6174 dentry->d_name.name, PFID(fid), rc);
6180 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6181 struct osd_it_ea *it, struct lu_fid *fid,
6182 struct osd_inode_id *id, __u32 *attr)
6184 struct osd_thread_info *info = osd_oti_get(env);
6185 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6186 struct osd_device *dev = osd_obj2dev(obj);
6187 struct super_block *sb = osd_sb(dev);
6188 const char *devname = osd_name(dev);
6189 struct osd_it_ea_dirent *ent = it->oie_dirent;
6190 struct inode *dir = obj->oo_inode;
6191 struct htree_lock *hlock = NULL;
6192 struct buffer_head *bh = NULL;
6193 handle_t *jh = NULL;
6194 struct ldiskfs_dir_entry_2 *de;
6195 struct dentry *dentry;
6196 struct inode *inode;
6197 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6200 bool dotdot = false;
6204 if (ent->oied_name[0] == '.') {
6205 if (ent->oied_namelen == 1)
6208 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6212 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6213 inode = osd_iget(info, dev, id);
6214 if (IS_ERR(inode)) {
6215 rc = PTR_ERR(inode);
6216 if (rc == -ENOENT || rc == -ESTALE) {
6217 /* Maybe dangling name entry, or
6218 * corrupted directory entry. */
6219 *attr |= LUDA_UNKNOWN;
6222 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6223 "check_repair, dir = %lu/%u, name = %.*s, "
6224 "ino = %llu, rc = %d\n",
6225 devname, dir->i_ino, dir->i_generation,
6226 ent->oied_namelen, ent->oied_name,
6233 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6235 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6236 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6241 /* We need to ensure that the name entry is still valid.
6242 * Because it may be removed or renamed by other already.
6244 * The unlink or rename operation will start journal before PDO lock,
6245 * so to avoid deadlock, here we need to start journal handle before
6246 * related PDO lock also. But because we do not know whether there
6247 * will be something to be repaired before PDO lock, we just start
6248 * journal without conditions.
6250 * We may need to remove the name entry firstly, then insert back.
6251 * One credit is for user quota file update.
6252 * One credit is for group quota file update.
6253 * Two credits are for dirty inode. */
6254 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6255 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6257 if (dev->od_dirent_journal != 0) {
6260 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6263 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6264 "check_repair, dir = %lu/%u, credits = %d, "
6265 "name = %.*s, ino = %llu: rc = %d\n",
6266 devname, dir->i_ino, dir->i_generation, credits,
6267 ent->oied_namelen, ent->oied_name,
6270 GOTO(out_inode, rc);
6273 if (obj->oo_hl_head != NULL) {
6274 hlock = osd_oti_get(env)->oti_hlock;
6275 /* "0" means exclusive lock for the whole directory.
6276 * We need to prevent others access such name entry
6277 * during the delete + insert. Neither HLOCK_ADD nor
6278 * HLOCK_DEL cannot guarantee the atomicity. */
6279 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6281 down_write(&obj->oo_ext_idx_sem);
6284 if (obj->oo_hl_head != NULL) {
6285 hlock = osd_oti_get(env)->oti_hlock;
6286 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6287 LDISKFS_HLOCK_LOOKUP);
6289 down_read(&obj->oo_ext_idx_sem);
6293 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6294 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6295 *attr |= LUDA_IGNORE;
6300 /* For dotdot entry, if there is not enough space to hold the
6301 * FID-in-dirent, just keep them there. It only happens when the
6302 * device upgraded from 1.8 or restored from MDT file-level backup.
6303 * For the whole directory, only dotdot entry have no FID-in-dirent
6304 * and needs to get FID from LMA when readdir, it will not affect the
6305 * performance much. */
6306 if (dotdot && !osd_dotdot_has_space(de)) {
6307 *attr |= LUDA_UNKNOWN;
6313 if (lu_fid_eq(fid, &lma->lma_self_fid))
6316 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6317 struct lu_fid *tfid = &lma->lma_self_fid;
6319 if (likely(dotdot &&
6320 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6321 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6322 /* It must be REMOTE_PARENT_DIR and as the
6323 * 'dotdot' entry of remote directory */
6324 *attr |= LUDA_IGNORE;
6326 CDEBUG(D_LFSCK, "%s: expect remote agent "
6327 "parent directory, but got %.*s under "
6328 "dir = %lu/%u with the FID "DFID"\n",
6329 devname, ent->oied_namelen,
6330 ent->oied_name, dir->i_ino,
6331 dir->i_generation, PFID(tfid));
6333 *attr |= LUDA_UNKNOWN;
6340 if (!fid_is_zero(fid)) {
6341 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6343 if (rc == -ENOENT ||
6345 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6346 /* linkEA does not recognize the dirent entry,
6347 * it may because the dirent entry corruption
6348 * and points to other's inode. */
6349 CDEBUG(D_LFSCK, "%s: the target inode does not "
6350 "recognize the dirent, dir = %lu/%u, "
6351 " name = %.*s, ino = %llu, "
6352 DFID": rc = %d\n", devname, dir->i_ino,
6353 dir->i_generation, ent->oied_namelen,
6354 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6355 *attr |= LUDA_UNKNOWN;
6360 if (rc && rc != -ENODATA) {
6361 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6362 "dir = %lu/%u, name = %.*s, ino = %llu, "
6363 DFID": rc = %d\n", devname, dir->i_ino,
6364 dir->i_generation, ent->oied_namelen,
6365 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6366 *attr |= LUDA_UNKNOWN;
6373 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6374 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6375 if (*attr & LUDA_VERIFY_DRYRUN) {
6376 *fid = lma->lma_self_fid;
6377 *attr |= LUDA_REPAIR;
6384 dev->od_dirent_journal = 1;
6385 if (hlock != NULL) {
6386 ldiskfs_htree_unlock(hlock);
6389 up_read(&obj->oo_ext_idx_sem);
6395 *fid = lma->lma_self_fid;
6397 /* Update or append the FID-in-dirent. */
6398 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6399 bh, de, hlock, dotdot);
6401 *attr |= LUDA_REPAIR;
6403 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6404 "the dirent, dir = %lu/%u, name = %.*s, "
6405 "ino = %llu, "DFID": rc = %d\n",
6406 devname, dir->i_ino, dir->i_generation,
6407 ent->oied_namelen, ent->oied_name,
6408 ent->oied_ino, PFID(fid), rc);
6410 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6411 if (*attr & LUDA_VERIFY_DRYRUN) {
6412 if (fid_is_sane(fid)) {
6413 *attr |= LUDA_REPAIR;
6414 } else if (dev->od_index == 0) {
6415 lu_igif_build(fid, inode->i_ino,
6416 inode->i_generation);
6417 *attr |= LUDA_UPGRADE;
6425 dev->od_dirent_journal = 1;
6426 if (hlock != NULL) {
6427 ldiskfs_htree_unlock(hlock);
6430 up_read(&obj->oo_ext_idx_sem);
6437 if (unlikely(fid_is_sane(fid))) {
6438 /* FID-in-dirent exists, but FID-in-LMA is lost.
6439 * Trust the FID-in-dirent, and add FID-in-LMA. */
6440 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6442 *attr |= LUDA_REPAIR;
6444 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6445 "update dirent, dir = %lu/%u, "
6446 "name = %.*s, ino = %llu, "
6448 devname, dir->i_ino, dir->i_generation,
6449 ent->oied_namelen, ent->oied_name,
6450 ent->oied_ino, PFID(fid), rc);
6451 } else if (dev->od_index == 0) {
6452 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6453 /* It is probably IGIF object. Only aappend the
6454 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6455 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6456 bh, de, hlock, dotdot);
6458 *attr |= LUDA_UPGRADE;
6460 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6461 "after the dirent, dir = %lu/%u, "
6462 "name = %.*s, ino = %llu, "
6464 devname, dir->i_ino, dir->i_generation,
6465 ent->oied_namelen, ent->oied_name,
6466 ent->oied_ino, PFID(fid), rc);
6475 if (hlock != NULL) {
6476 ldiskfs_htree_unlock(hlock);
6478 if (dev->od_dirent_journal != 0)
6479 up_write(&obj->oo_ext_idx_sem);
6481 up_read(&obj->oo_ext_idx_sem);
6485 ldiskfs_journal_stop(jh);
6489 if (rc >= 0 && !dirty)
6490 dev->od_dirent_journal = 0;
6496 * Returns the value at current position from iterator's in memory structure.
6498 * \param di struct osd_it_ea, iterator's in memory structure
6499 * \param attr attr requested for dirent.
6500 * \param lde lustre dirent
6502 * \retval 0 no error and \param lde has correct lustre dirent.
6503 * \retval -ve on error
6505 static inline int osd_it_ea_rec(const struct lu_env *env,
6506 const struct dt_it *di,
6507 struct dt_rec *dtrec, __u32 attr)
6509 struct osd_it_ea *it = (struct osd_it_ea *)di;
6510 struct osd_object *obj = it->oie_obj;
6511 struct osd_device *dev = osd_obj2dev(obj);
6512 struct osd_thread_info *oti = osd_oti_get(env);
6513 struct osd_inode_id *id = &oti->oti_id;
6514 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6515 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6516 __u32 ino = it->oie_dirent->oied_ino;
6520 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6522 if (attr & LUDA_VERIFY) {
6523 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6524 attr |= LUDA_IGNORE;
6525 /* If the parent is on remote MDT, and there
6526 * is no FID-in-dirent, then we have to get
6527 * the parent FID from the linkEA. */
6528 if (!fid_is_sane(fid) &&
6529 it->oie_dirent->oied_namelen == 2 &&
6530 it->oie_dirent->oied_name[0] == '.' &&
6531 it->oie_dirent->oied_name[1] == '.')
6532 osd_get_pfid_from_linkea(env, obj, fid);
6534 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6538 if (!fid_is_sane(fid))
6539 attr |= LUDA_UNKNOWN;
6541 attr &= ~LU_DIRENT_ATTRS_MASK;
6542 if (!fid_is_sane(fid)) {
6543 bool is_dotdot = false;
6544 if (it->oie_dirent->oied_namelen == 2 &&
6545 it->oie_dirent->oied_name[0] == '.' &&
6546 it->oie_dirent->oied_name[1] == '.')
6548 /* If the parent is on remote MDT, and there
6549 * is no FID-in-dirent, then we have to get
6550 * the parent FID from the linkEA. */
6551 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6552 rc = osd_get_pfid_from_linkea(env, obj, fid);
6554 if (is_dotdot == false &&
6555 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6558 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6561 osd_id_gen(id, ino, OSD_OII_NOGEN);
6565 /* Pack the entry anyway, at least the offset is right. */
6566 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6567 it->oie_dirent->oied_name,
6568 it->oie_dirent->oied_namelen,
6569 it->oie_dirent->oied_type, attr);
6574 if (osd_remote_fid(env, dev, fid))
6577 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6578 osd_add_oi_cache(oti, dev, id, fid);
6580 RETURN(rc > 0 ? 0 : rc);
6584 * Returns the record size size at current position.
6586 * This function will return record(lu_dirent) size in bytes.
6588 * \param[in] env execution environment
6589 * \param[in] di iterator's in memory structure
6590 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6591 * calculate the lu_dirent size.
6593 * \retval record size(in bytes & in memory) of the current lu_dirent
6596 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6599 struct osd_it_ea *it = (struct osd_it_ea *)di;
6601 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6605 * Returns a cookie for current position of the iterator head, so that
6606 * user can use this cookie to load/start the iterator next time.
6608 * \param di iterator's in memory structure
6610 * \retval cookie for current position, on success
6612 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6614 struct osd_it_ea *it = (struct osd_it_ea *)di;
6616 return it->oie_dirent->oied_off;
6620 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6621 * to load a directory entry at a time and stored it i inn,
6622 * in iterator's in-memory data structure.
6624 * \param di struct osd_it_ea, iterator's in memory structure
6626 * \retval +ve on success
6627 * \retval -ve on error
6629 static int osd_it_ea_load(const struct lu_env *env,
6630 const struct dt_it *di, __u64 hash)
6632 struct osd_it_ea *it = (struct osd_it_ea *)di;
6636 it->oie_file.f_pos = hash;
6638 rc = osd_ldiskfs_it_fill(env, di);
6649 * Index lookup function for interoperability mode (b11826).
6651 * \param key, key i.e. file name to be searched
6653 * \retval +ve, on success
6654 * \retval -ve, on error
6656 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6657 struct dt_rec *rec, const struct dt_key *key)
6659 struct osd_object *obj = osd_dt_obj(dt);
6664 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6665 LINVRNT(osd_invariant(obj));
6667 rc = osd_ea_lookup_rec(env, obj, rec, key);
6674 * Index and Iterator operations for interoperability
6675 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6677 static const struct dt_index_operations osd_index_ea_ops = {
6678 .dio_lookup = osd_index_ea_lookup,
6679 .dio_declare_insert = osd_index_declare_ea_insert,
6680 .dio_insert = osd_index_ea_insert,
6681 .dio_declare_delete = osd_index_declare_ea_delete,
6682 .dio_delete = osd_index_ea_delete,
6684 .init = osd_it_ea_init,
6685 .fini = osd_it_ea_fini,
6686 .get = osd_it_ea_get,
6687 .put = osd_it_ea_put,
6688 .next = osd_it_ea_next,
6689 .key = osd_it_ea_key,
6690 .key_size = osd_it_ea_key_size,
6691 .rec = osd_it_ea_rec,
6692 .rec_size = osd_it_ea_rec_size,
6693 .store = osd_it_ea_store,
6694 .load = osd_it_ea_load
6698 static void *osd_key_init(const struct lu_context *ctx,
6699 struct lu_context_key *key)
6701 struct osd_thread_info *info;
6703 OBD_ALLOC_PTR(info);
6705 return ERR_PTR(-ENOMEM);
6707 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6708 if (info->oti_it_ea_buf == NULL)
6711 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6713 info->oti_hlock = ldiskfs_htree_lock_alloc();
6714 if (info->oti_hlock == NULL)
6720 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6723 return ERR_PTR(-ENOMEM);
6726 static void osd_key_fini(const struct lu_context *ctx,
6727 struct lu_context_key *key, void* data)
6729 struct osd_thread_info *info = data;
6730 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6731 struct osd_idmap_cache *idc = info->oti_ins_cache;
6733 if (info->oti_inode != NULL)
6735 if (info->oti_hlock != NULL)
6736 ldiskfs_htree_lock_free(info->oti_hlock);
6737 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6738 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6739 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6740 lu_buf_free(&info->oti_big_buf);
6742 LASSERT(info->oti_ins_cache_size > 0);
6743 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6744 info->oti_ins_cache = NULL;
6745 info->oti_ins_cache_size = 0;
6750 static void osd_key_exit(const struct lu_context *ctx,
6751 struct lu_context_key *key, void *data)
6753 struct osd_thread_info *info = data;
6755 LASSERT(info->oti_r_locks == 0);
6756 LASSERT(info->oti_w_locks == 0);
6757 LASSERT(info->oti_txns == 0);
6760 /* type constructor/destructor: osd_type_init, osd_type_fini */
6761 LU_TYPE_INIT_FINI(osd, &osd_key);
6763 struct lu_context_key osd_key = {
6764 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6765 .lct_init = osd_key_init,
6766 .lct_fini = osd_key_fini,
6767 .lct_exit = osd_key_exit
6771 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6772 const char *name, struct lu_device *next)
6774 struct osd_device *osd = osd_dev(d);
6776 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6777 >= sizeof(osd->od_svname))
6779 return osd_procfs_init(osd, name);
6782 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6784 struct seq_server_site *ss = osd_seq_site(osd);
6788 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6791 if (unlikely(ss == NULL))
6794 OBD_ALLOC_PTR(osd->od_cl_seq);
6795 if (osd->od_cl_seq == NULL)
6798 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6799 osd->od_svname, ss->ss_server_seq);
6801 OBD_FREE_PTR(osd->od_cl_seq);
6802 osd->od_cl_seq = NULL;
6806 if (ss->ss_node_id == 0) {
6807 /* If the OSD on the sequence controller(MDT0), then allocate
6808 * sequence here, otherwise allocate sequence after connected
6809 * to MDT0 (see mdt_register_lwp_callback()). */
6810 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6811 &osd->od_cl_seq->lcs_space, env);
6817 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6819 if (osd->od_cl_seq == NULL)
6822 seq_client_fini(osd->od_cl_seq);
6823 OBD_FREE_PTR(osd->od_cl_seq);
6824 osd->od_cl_seq = NULL;
6827 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6831 /* shutdown quota slave instance associated with the device */
6832 if (o->od_quota_slave != NULL) {
6833 qsd_fini(env, o->od_quota_slave);
6834 o->od_quota_slave = NULL;
6837 osd_fid_fini(env, o);
6842 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6846 if (o->od_mnt != NULL) {
6847 shrink_dcache_sb(osd_sb(o));
6848 osd_sync(env, &o->od_dt_dev);
6857 static int osd_mount(const struct lu_env *env,
6858 struct osd_device *o, struct lustre_cfg *cfg)
6860 const char *name = lustre_cfg_string(cfg, 0);
6861 const char *dev = lustre_cfg_string(cfg, 1);
6863 unsigned long page, s_flags, lmd_flags = 0;
6864 struct page *__page;
6865 struct file_system_type *type;
6866 char *options = NULL;
6868 struct osd_thread_info *info = osd_oti_get(env);
6869 struct lu_fid *fid = &info->oti_fid;
6870 struct inode *inode;
6871 int rc = 0, force_over_512tb = 0;
6874 if (o->od_mnt != NULL)
6877 if (strlen(dev) >= sizeof(o->od_mntdev))
6879 strcpy(o->od_mntdev, dev);
6881 str = lustre_cfg_string(cfg, 2);
6882 s_flags = simple_strtoul(str, NULL, 0);
6883 str = strstr(str, ":");
6885 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6886 opts = lustre_cfg_string(cfg, 3);
6888 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6889 CERROR("%s: device %s extents feature is not guaranteed to "
6890 "work on big-endian systems. Use \"bigendian_extents\" "
6891 "mount option to override.\n", name, dev);
6895 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6896 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6897 CWARN("force_over_128tb option is deprecated. "
6898 "Filesystems less than 512TB can be created without any "
6899 "force options. Use force_over_512tb option for "
6900 "filesystems greater than 512TB.\n");
6903 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6904 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6905 CWARN("force_over_256tb option is deprecated. "
6906 "Filesystems less than 512TB can be created without any "
6907 "force options. Use force_over_512tb option for "
6908 "filesystems greater than 512TB.\n");
6912 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6913 force_over_512tb = 1;
6915 __page = alloc_page(GFP_KERNEL);
6917 GOTO(out, rc = -ENOMEM);
6918 page = (unsigned long)page_address(__page);
6919 options = (char *)page;
6922 /* strip out the options for back compatiblity */
6923 static char *sout[] = {
6930 /* strip out option we processed in osd */
6931 "bigendian_extents",
6937 strcat(options, opts);
6938 for (rc = 0, str = options; sout[rc]; ) {
6939 char *op = strstr(str, sout[rc]);
6945 if (op == options || *(op - 1) == ',') {
6946 str = op + strlen(sout[rc]);
6947 if (*str == ',' || *str == '\0') {
6948 *str == ',' ? str++ : str;
6949 memmove(op, str, strlen(str) + 1);
6952 for (str = op; *str != ',' && *str != '\0'; str++)
6956 strncat(options, "user_xattr,acl", 14);
6959 /* Glom up mount options */
6960 if (*options != '\0')
6961 strcat(options, ",");
6962 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6964 type = get_fs_type("ldiskfs");
6966 CERROR("%s: cannot find ldiskfs module\n", name);
6967 GOTO(out, rc = -ENODEV);
6970 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6971 module_put(type->owner);
6973 if (IS_ERR(o->od_mnt)) {
6974 rc = PTR_ERR(o->od_mnt);
6976 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6980 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
6981 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
6982 force_over_512tb == 0) {
6983 CERROR("%s: device %s LDISKFS does not support filesystems "
6984 "greater than 512TB and can cause data corruption. "
6985 "Use \"force_over_512tb\" mount option to override.\n",
6987 GOTO(out_mnt, rc = -EINVAL);
6990 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
6991 if (priv_dev_set_rdonly) {
6992 priv_dev_set_rdonly(osd_sb(o)->s_bdev);
6993 o->od_dt_dev.dd_rdonly = 1;
6994 LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
6997 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7000 GOTO(out_mnt, rc = -EOPNOTSUPP);
7002 } else if (priv_dev_check_rdonly &&
7003 priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7004 CERROR("%s: underlying device %s is marked as "
7005 "read-only. Setup failed\n", name, dev);
7007 GOTO(out_mnt, rc = -EROFS);
7010 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7011 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7012 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7013 GOTO(out_mnt, rc = -EINVAL);
7016 #ifdef LDISKFS_MOUNT_DIRDATA
7017 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7018 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7019 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7020 else if (!o->od_is_ost)
7021 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7022 "enabling the dirdata feature. If you do not want to "
7023 "downgrade to Lustre-1.x again, you can enable it via "
7024 "'tune2fs -O dirdata device'\n", name, dev);
7026 inode = osd_sb(o)->s_root->d_inode;
7027 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7028 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7030 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7034 if (lmd_flags & LMD_FLG_NOSCRUB)
7045 __free_page(__page);
7050 static struct lu_device *osd_device_fini(const struct lu_env *env,
7051 struct lu_device *d)
7053 struct osd_device *o = osd_dev(d);
7056 osd_shutdown(env, o);
7058 osd_scrub_cleanup(env, o);
7059 osd_obj_map_fini(o);
7065 static int osd_device_init0(const struct lu_env *env,
7066 struct osd_device *o,
7067 struct lustre_cfg *cfg)
7069 struct lu_device *l = osd2lu_dev(o);
7070 struct osd_thread_info *info;
7074 /* if the module was re-loaded, env can loose its keys */
7075 rc = lu_env_refill((struct lu_env *) env);
7078 info = osd_oti_get(env);
7081 l->ld_ops = &osd_lu_ops;
7082 o->od_dt_dev.dd_ops = &osd_dt_ops;
7084 spin_lock_init(&o->od_osfs_lock);
7085 mutex_init(&o->od_otable_mutex);
7086 INIT_LIST_HEAD(&o->od_orphan_list);
7088 o->od_read_cache = 1;
7089 o->od_writethrough_cache = 1;
7090 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7092 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7093 sizeof(o->od_svname));
7094 if (cplen >= sizeof(o->od_svname)) {
7099 o->od_index = -1; /* -1 means index is invalid */
7100 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7101 if (rc == LDD_F_SV_TYPE_OST)
7104 o->od_full_scrub_ratio = OFSR_DEFAULT;
7105 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7106 rc = osd_mount(env, o, cfg);
7110 rc = osd_obj_map_init(env, o);
7114 rc = lu_site_init(&o->od_site, l);
7116 GOTO(out_compat, rc);
7117 o->od_site.ls_bottom_dev = l;
7119 rc = lu_site_init_finish(&o->od_site);
7123 INIT_LIST_HEAD(&o->od_ios_list);
7124 /* setup scrub, including OI files initialization */
7125 rc = osd_scrub_setup(env, o);
7129 rc = osd_procfs_init(o, o->od_svname);
7131 CERROR("%s: can't initialize procfs: rc = %d\n",
7133 GOTO(out_scrub, rc);
7136 LASSERT(l->ld_site->ls_linkage.next != NULL);
7137 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7139 /* initialize quota slave instance */
7140 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7142 if (IS_ERR(o->od_quota_slave)) {
7143 rc = PTR_ERR(o->od_quota_slave);
7144 o->od_quota_slave = NULL;
7145 GOTO(out_procfs, rc);
7153 osd_scrub_cleanup(env, o);
7155 lu_site_fini(&o->od_site);
7157 osd_obj_map_fini(o);
7164 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7165 struct lu_device_type *t,
7166 struct lustre_cfg *cfg)
7168 struct osd_device *o;
7173 return ERR_PTR(-ENOMEM);
7175 rc = dt_device_init(&o->od_dt_dev, t);
7177 /* Because the ctx might be revived in dt_device_init,
7178 * refill the env here */
7179 lu_env_refill((struct lu_env *)env);
7180 rc = osd_device_init0(env, o, cfg);
7182 dt_device_fini(&o->od_dt_dev);
7185 if (unlikely(rc != 0))
7188 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7191 static struct lu_device *osd_device_free(const struct lu_env *env,
7192 struct lu_device *d)
7194 struct osd_device *o = osd_dev(d);
7197 /* XXX: make osd top device in order to release reference */
7198 d->ld_site->ls_top_dev = d;
7199 lu_site_purge(env, d->ld_site, -1);
7200 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7201 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7202 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7204 lu_site_fini(&o->od_site);
7205 dt_device_fini(&o->od_dt_dev);
7210 static int osd_process_config(const struct lu_env *env,
7211 struct lu_device *d, struct lustre_cfg *cfg)
7213 struct osd_device *o = osd_dev(d);
7217 switch (cfg->lcfg_command) {
7219 rc = osd_mount(env, o, cfg);
7222 lu_dev_del_linkage(d->ld_site, d);
7223 rc = osd_shutdown(env, o);
7226 LASSERT(&o->od_dt_dev);
7227 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7228 cfg, &o->od_dt_dev);
7229 if (rc > 0 || rc == -ENOSYS) {
7230 rc = class_process_proc_param(PARAM_OST,
7231 lprocfs_osd_obd_vars,
7232 cfg, &o->od_dt_dev);
7244 static int osd_recovery_complete(const struct lu_env *env,
7245 struct lu_device *d)
7247 struct osd_device *osd = osd_dev(d);
7251 if (osd->od_quota_slave == NULL)
7254 /* start qsd instance on recovery completion, this notifies the quota
7255 * slave code that we are about to process new requests now */
7256 rc = qsd_start(env, osd->od_quota_slave);
7261 * we use exports to track all osd users
7263 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7264 struct obd_device *obd, struct obd_uuid *cluuid,
7265 struct obd_connect_data *data, void *localdata)
7267 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7268 struct lustre_handle conn;
7272 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7274 rc = class_connect(&conn, obd, cluuid);
7278 *exp = class_conn2export(&conn);
7280 spin_lock(&osd->od_osfs_lock);
7282 spin_unlock(&osd->od_osfs_lock);
7288 * once last export (we don't count self-export) disappeared
7289 * osd can be released
7291 static int osd_obd_disconnect(struct obd_export *exp)
7293 struct obd_device *obd = exp->exp_obd;
7294 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7295 int rc, release = 0;
7298 /* Only disconnect the underlying layers on the final disconnect. */
7299 spin_lock(&osd->od_osfs_lock);
7301 if (osd->od_connects == 0)
7303 spin_unlock(&osd->od_osfs_lock);
7305 rc = class_disconnect(exp); /* bz 9811 */
7307 if (rc == 0 && release)
7308 class_manual_cleanup(obd);
7312 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7313 struct lu_device *dev)
7315 struct osd_device *osd = osd_dev(dev);
7316 struct lr_server_data *lsd =
7317 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7321 if (osd->od_quota_slave != NULL) {
7322 /* set up quota slave objects */
7323 result = qsd_prepare(env, osd->od_quota_slave);
7328 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7329 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7330 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7331 osd->od_index_in_idif = 1;
7333 osd->od_index_in_idif = 0;
7334 result = osd_register_proc_index_in_idif(osd);
7339 osd->od_index_in_idif = 1;
7343 result = osd_fid_init(env, osd);
7348 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7349 struct lu_fid *fid, struct md_op_data *op_data)
7351 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7353 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7356 static const struct lu_object_operations osd_lu_obj_ops = {
7357 .loo_object_init = osd_object_init,
7358 .loo_object_delete = osd_object_delete,
7359 .loo_object_release = osd_object_release,
7360 .loo_object_free = osd_object_free,
7361 .loo_object_print = osd_object_print,
7362 .loo_object_invariant = osd_object_invariant
7365 const struct lu_device_operations osd_lu_ops = {
7366 .ldo_object_alloc = osd_object_alloc,
7367 .ldo_process_config = osd_process_config,
7368 .ldo_recovery_complete = osd_recovery_complete,
7369 .ldo_prepare = osd_prepare,
7372 static const struct lu_device_type_operations osd_device_type_ops = {
7373 .ldto_init = osd_type_init,
7374 .ldto_fini = osd_type_fini,
7376 .ldto_start = osd_type_start,
7377 .ldto_stop = osd_type_stop,
7379 .ldto_device_alloc = osd_device_alloc,
7380 .ldto_device_free = osd_device_free,
7382 .ldto_device_init = osd_device_init,
7383 .ldto_device_fini = osd_device_fini
7386 static struct lu_device_type osd_device_type = {
7387 .ldt_tags = LU_DEVICE_DT,
7388 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7389 .ldt_ops = &osd_device_type_ops,
7390 .ldt_ctx_tags = LCT_LOCAL,
7393 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7395 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7396 struct super_block *sb = osd_sb(osd);
7398 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7402 * lprocfs legacy support.
7404 static struct obd_ops osd_obd_device_ops = {
7405 .o_owner = THIS_MODULE,
7406 .o_connect = osd_obd_connect,
7407 .o_disconnect = osd_obd_disconnect,
7408 .o_fid_alloc = osd_fid_alloc,
7409 .o_health_check = osd_health_check,
7412 static int __init osd_init(void)
7416 LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7417 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7418 /* please, try to keep osd_thread_info smaller than a page */
7419 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7424 rc = lu_kmem_init(ldiskfs_caches);
7428 #ifdef CONFIG_KALLSYMS
7429 priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7430 priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7431 /* Clear readonly is unused at this time */
7432 /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7435 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7436 lprocfs_osd_module_vars,
7437 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7439 lu_kmem_fini(ldiskfs_caches);
7443 static void __exit osd_exit(void)
7445 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7446 lu_kmem_fini(ldiskfs_caches);
7449 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7450 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7451 MODULE_VERSION(LUSTRE_VERSION_STRING);
7452 MODULE_LICENSE("GPL");
7454 module_init(osd_init);
7455 module_exit(osd_exit);