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;
2600 LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2605 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2609 if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2610 (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2613 ll_vfs_dq_init(inode);
2615 if (attr->la_valid & LA_UID)
2616 iattr.ia_valid |= ATTR_UID;
2617 if (attr->la_valid & LA_GID)
2618 iattr.ia_valid |= ATTR_GID;
2619 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2620 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2622 rc = ll_vfs_dq_transfer(inode, &iattr);
2624 CERROR("%s: quota transfer failed: rc = %d. Is quota "
2625 "enforcement enabled on the ldiskfs "
2626 "filesystem?\n", inode->i_sb->s_id, rc);
2631 /* Handle project id transfer here properly */
2632 if (attr->la_valid & LA_PROJID &&
2633 attr->la_projid != i_projid_read(inode)) {
2634 #ifdef HAVE_PROJECT_QUOTA
2635 rc = ldiskfs_transfer_project(inode, attr->la_projid);
2640 CERROR("%s: quota transfer failed: rc = %d. Is project "
2641 "enforcement enabled on the ldiskfs "
2642 "filesystem?\n", inode->i_sb->s_id, rc);
2649 static int osd_attr_set(const struct lu_env *env,
2650 struct dt_object *dt,
2651 const struct lu_attr *attr,
2652 struct thandle *handle)
2654 struct osd_object *obj = osd_dt_obj(dt);
2655 struct inode *inode;
2658 if (!dt_object_exists(dt))
2661 LASSERT(handle != NULL);
2662 LASSERT(!dt_object_remote(dt));
2663 LASSERT(osd_invariant(obj));
2665 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2667 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2668 struct osd_thread_info *oti = osd_oti_get(env);
2669 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2670 struct lu_fid *fid1 = &oti->oti_fid;
2671 struct osd_inode_id *id = &oti->oti_id;
2672 struct iam_path_descr *ipd;
2673 struct iam_container *bag;
2674 struct osd_thandle *oh;
2677 fid_cpu_to_be(fid1, fid0);
2678 memset(id, 1, sizeof(*id));
2679 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2680 fid0)->oi_dir.od_container;
2681 ipd = osd_idx_ipd_get(env, bag);
2682 if (unlikely(ipd == NULL))
2685 oh = container_of0(handle, struct osd_thandle, ot_super);
2686 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2687 (const struct iam_rec *)id, ipd);
2688 osd_ipd_put(env, bag, ipd);
2689 return(rc > 0 ? 0 : rc);
2692 inode = obj->oo_inode;
2694 rc = osd_quota_transfer(inode, attr);
2698 spin_lock(&obj->oo_guard);
2699 rc = osd_inode_setattr(env, inode, attr);
2700 spin_unlock(&obj->oo_guard);
2704 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2706 if (!(attr->la_valid & LA_FLAGS))
2709 /* Let's check if there are extra flags need to be set into LMA */
2710 if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2711 struct osd_thread_info *info = osd_oti_get(env);
2712 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2714 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2715 &info->oti_ost_attrs);
2719 lma->lma_incompat |=
2720 lustre_to_lma_flags(attr->la_flags);
2721 lustre_lma_swab(lma);
2722 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2723 lma, sizeof(*lma), XATTR_REPLACE);
2725 struct osd_device *osd = osd_obj2dev(obj);
2727 CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2728 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2729 lma->lma_incompat, rc);
2732 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2734 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2737 osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2742 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2743 struct osd_object *obj,
2744 const char *name, const int namelen)
2746 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2749 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2750 umode_t mode, struct dt_allocation_hint *hint,
2754 struct osd_device *osd = osd_obj2dev(obj);
2755 struct osd_thandle *oth;
2756 struct dt_object *parent = NULL;
2757 struct inode *inode;
2759 LINVRNT(osd_invariant(obj));
2760 LASSERT(obj->oo_inode == NULL);
2761 LASSERT(obj->oo_hl_head == NULL);
2763 if (S_ISDIR(mode) && ldiskfs_pdo) {
2764 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2765 if (obj->oo_hl_head == NULL)
2769 oth = container_of(th, struct osd_thandle, ot_super);
2770 LASSERT(oth->ot_handle->h_transaction != NULL);
2772 if (hint != NULL && hint->dah_parent != NULL &&
2773 !dt_object_remote(hint->dah_parent))
2774 parent = hint->dah_parent;
2776 inode = ldiskfs_create_inode(oth->ot_handle,
2777 parent ? osd_dt_obj(parent)->oo_inode :
2778 osd_sb(osd)->s_root->d_inode,
2780 if (!IS_ERR(inode)) {
2781 /* Do not update file c/mtime in ldiskfs. */
2782 inode->i_flags |= S_NOCMTIME;
2784 /* For new created object, it must be consistent,
2785 * and it is unnecessary to scrub against it. */
2786 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2788 obj->oo_inode = inode;
2791 if (obj->oo_hl_head != NULL) {
2792 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2793 obj->oo_hl_head = NULL;
2795 result = PTR_ERR(inode);
2797 LINVRNT(osd_invariant(obj));
2805 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2806 struct lu_attr *attr,
2807 struct dt_allocation_hint *hint,
2808 struct dt_object_format *dof,
2812 struct osd_thandle *oth;
2813 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2815 LASSERT(S_ISDIR(attr->la_mode));
2817 oth = container_of(th, struct osd_thandle, ot_super);
2818 LASSERT(oth->ot_handle->h_transaction != NULL);
2819 result = osd_mkfile(info, obj, mode, hint, th);
2824 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2825 struct lu_attr *attr,
2826 struct dt_allocation_hint *hint,
2827 struct dt_object_format *dof,
2831 struct osd_thandle *oth;
2832 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2834 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2836 LASSERT(S_ISREG(attr->la_mode));
2838 oth = container_of(th, struct osd_thandle, ot_super);
2839 LASSERT(oth->ot_handle->h_transaction != NULL);
2841 result = osd_mkfile(info, obj, mode, hint, th);
2843 LASSERT(obj->oo_inode != NULL);
2844 if (feat->dif_flags & DT_IND_VARKEY)
2845 result = iam_lvar_create(obj->oo_inode,
2846 feat->dif_keysize_max,
2848 feat->dif_recsize_max,
2851 result = iam_lfix_create(obj->oo_inode,
2852 feat->dif_keysize_max,
2854 feat->dif_recsize_max,
2861 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2862 struct lu_attr *attr,
2863 struct dt_allocation_hint *hint,
2864 struct dt_object_format *dof,
2867 LASSERT(S_ISREG(attr->la_mode));
2868 return osd_mkfile(info, obj, (attr->la_mode &
2869 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2872 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2873 struct lu_attr *attr,
2874 struct dt_allocation_hint *hint,
2875 struct dt_object_format *dof,
2878 LASSERT(S_ISLNK(attr->la_mode));
2879 return osd_mkfile(info, obj, (attr->la_mode &
2880 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2883 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2884 struct lu_attr *attr,
2885 struct dt_allocation_hint *hint,
2886 struct dt_object_format *dof,
2889 umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2892 LINVRNT(osd_invariant(obj));
2893 LASSERT(obj->oo_inode == NULL);
2894 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2895 S_ISFIFO(mode) || S_ISSOCK(mode));
2897 result = osd_mkfile(info, obj, mode, hint, th);
2899 LASSERT(obj->oo_inode != NULL);
2901 * This inode should be marked dirty for i_rdev. Currently
2902 * that is done in the osd_attr_init().
2904 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2907 LINVRNT(osd_invariant(obj));
2911 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2913 struct dt_allocation_hint *hint,
2914 struct dt_object_format *dof,
2917 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2919 osd_obj_type_f result;
2935 result = osd_mk_index;
2946 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2947 struct dt_object *parent, struct dt_object *child,
2952 ah->dah_parent = parent;
2953 ah->dah_mode = child_mode;
2955 if (parent != NULL && !dt_object_remote(parent)) {
2956 /* will help to find FID->ino at dt_insert("..") */
2957 struct osd_object *pobj = osd_dt_obj(parent);
2958 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2962 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2963 struct lu_attr *attr, struct dt_object_format *dof)
2965 struct inode *inode = obj->oo_inode;
2966 __u64 valid = attr->la_valid;
2969 attr->la_valid &= ~(LA_TYPE | LA_MODE);
2971 if (dof->dof_type != DFT_NODE)
2972 attr->la_valid &= ~LA_RDEV;
2973 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2974 attr->la_valid &= ~LA_ATIME;
2975 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2976 attr->la_valid &= ~LA_CTIME;
2977 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2978 attr->la_valid &= ~LA_MTIME;
2980 result = osd_quota_transfer(inode, attr);
2984 if (attr->la_valid != 0) {
2985 result = osd_inode_setattr(info->oti_env, inode, attr);
2987 * The osd_inode_setattr() should always succeed here. The
2988 * only error that could be returned is EDQUOT when we are
2989 * trying to change the UID or GID of the inode. However, this
2990 * should not happen since quota enforcement is no longer
2991 * enabled on ldiskfs (lquota takes care of it).
2993 LASSERTF(result == 0, "%d\n", result);
2994 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2997 attr->la_valid = valid;
3001 * Helper function for osd_create()
3003 * \retval 0, on success
3005 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3006 struct lu_attr *attr, struct dt_allocation_hint *hint,
3007 struct dt_object_format *dof, struct thandle *th)
3012 osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3014 /* we drop umask so that permissions we pass are not affected */
3015 umask = current->fs->umask;
3016 current->fs->umask = 0;
3018 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3020 if (likely(obj->oo_inode != NULL)) {
3021 LASSERT(obj->oo_inode->i_state & I_NEW);
3023 /* Unlock the inode before attr initialization to avoid
3024 * unnecessary dqget operations. LU-6378 */
3025 unlock_new_inode(obj->oo_inode);
3028 if (likely(result == 0)) {
3029 osd_attr_init(info, obj, attr, dof);
3030 osd_object_init0(obj);
3033 /* restore previous umask value */
3034 current->fs->umask = umask;
3036 osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3042 * Helper function for osd_create()
3044 * \retval 0, on success
3046 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3047 const struct lu_fid *fid, struct thandle *th)
3049 struct osd_thread_info *info = osd_oti_get(env);
3050 struct osd_inode_id *id = &info->oti_id;
3051 struct osd_device *osd = osd_obj2dev(obj);
3052 struct osd_thandle *oh;
3055 LASSERT(obj->oo_inode != NULL);
3057 oh = container_of0(th, struct osd_thandle, ot_super);
3058 LASSERT(oh->ot_handle);
3059 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3061 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3062 rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3063 OI_CHECK_FLD, NULL);
3064 osd_trans_exec_check(env, th, OSD_OT_INSERT);
3069 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3070 u64 seq, struct lu_seq_range *range)
3072 struct seq_server_site *ss = osd_seq_site(osd);
3074 if (fid_seq_is_idif(seq)) {
3075 fld_range_set_ost(range);
3076 range->lsr_index = idif_ost_idx(seq);
3080 if (!fid_seq_in_fldb(seq)) {
3081 fld_range_set_mdt(range);
3083 /* FIXME: If ss is NULL, it suppose not get lsr_index
3085 range->lsr_index = ss->ss_node_id;
3089 LASSERT(ss != NULL);
3090 fld_range_set_any(range);
3091 /* OSD will only do local fld lookup */
3092 return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3095 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3096 struct lu_attr *attr,
3097 struct dt_allocation_hint *hint,
3098 struct dt_object_format *dof,
3099 struct thandle *handle)
3101 struct osd_thandle *oh;
3105 LASSERT(handle != NULL);
3107 oh = container_of0(handle, struct osd_thandle, ot_super);
3108 LASSERT(oh->ot_handle == NULL);
3110 /* EA object consumes more credits than regular object: osd_mk_index
3111 * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3112 * leaf_node, could involves the block, block bitmap, groups, GDT
3113 * change for each block, so add 4 * 2 credits in that case. */
3114 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3115 osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3116 (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3117 /* Reuse idle OI block may cause additional one OI block
3119 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3120 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3125 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3126 attr->la_projid, 1, oh, osd_dt_obj(dt),
3127 NULL, OSD_QID_INODE);
3131 /* will help to find FID->ino mapping at dt_insert() */
3132 rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3139 * Called to destroy on-disk representation of the object
3141 * Concurrency: must be locked
3143 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3146 struct osd_object *obj = osd_dt_obj(dt);
3147 struct inode *inode = obj->oo_inode;
3148 struct osd_thandle *oh;
3155 oh = container_of0(th, struct osd_thandle, ot_super);
3156 LASSERT(oh->ot_handle == NULL);
3158 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3159 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3160 /* Recycle idle OI leaf may cause additional three OI blocks
3162 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3163 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3164 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3165 /* one less inode */
3166 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3167 i_projid_read(inode), -1, oh, obj, NULL,
3171 /* data to be truncated */
3172 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3173 i_projid_read(inode), 0, oh, obj, NULL,
3178 /* will help to find FID->ino when this object is being
3179 * added to PENDING/ */
3180 rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3185 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3188 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3189 struct osd_object *obj = osd_dt_obj(dt);
3190 struct inode *inode = obj->oo_inode;
3191 struct osd_device *osd = osd_obj2dev(obj);
3192 struct osd_thandle *oh;
3196 oh = container_of0(th, struct osd_thandle, ot_super);
3197 LASSERT(oh->ot_handle);
3199 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3201 if (unlikely(fid_is_acct(fid)))
3204 if (S_ISDIR(inode->i_mode)) {
3205 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3206 inode->i_nlink == 2);
3207 /* it will check/delete the inode from remote parent,
3208 * how to optimize it? unlink performance impaction XXX */
3209 result = osd_delete_from_remote_parent(env, osd, obj, oh);
3211 CERROR("%s: delete inode "DFID": rc = %d\n",
3212 osd_name(osd), PFID(fid), result);
3214 spin_lock(&obj->oo_guard);
3216 spin_unlock(&obj->oo_guard);
3217 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3220 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3222 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3224 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3225 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3226 oh->ot_handle, OI_CHECK_FLD);
3228 osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3229 /* XXX: add to ext3 orphan list */
3230 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3232 /* not needed in the cache anymore */
3233 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3234 obj->oo_destroyed = 1;
3240 * Put the fid into lustre_mdt_attrs, and then place the structure
3241 * inode's ea. This fid should not be altered during the life time
3244 * \retval +ve, on success
3245 * \retval -ve, on error
3247 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3249 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3250 const struct lu_fid *fid, __u32 compat, __u32 incompat)
3252 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3253 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3257 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3260 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3263 lustre_loa_init(loa, fid, compat, incompat);
3264 lustre_loa_swab(loa, false);
3266 /* For the OST device with 256 bytes inode size by default,
3267 * the PFID EA will be stored together with LMA EA to avoid
3268 * performance trouble. Otherwise the PFID EA can be stored
3269 * independently. LU-8998 */
3270 if ((compat & LMAC_FID_ON_OST) &&
3271 LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3272 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3273 sizeof(*loa), XATTR_CREATE);
3275 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3276 sizeof(*lma), XATTR_CREATE);
3277 /* LMA may already exist, but we need to check that all the
3278 * desired compat/incompat flags have been added. */
3279 if (unlikely(rc == -EEXIST)) {
3280 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3281 XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3285 if (rc < sizeof(*lma))
3288 lustre_loa_swab(loa, true);
3289 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3290 ((compat == 0 && incompat == 0) ||
3291 (!(~lma->lma_compat & compat) &&
3292 !(~lma->lma_incompat & incompat))))
3295 lma->lma_self_fid = *fid;
3296 lma->lma_compat |= compat;
3297 lma->lma_incompat |= incompat;
3298 if (rc == sizeof(*lma)) {
3299 lustre_lma_swab(lma);
3300 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3301 sizeof(*lma), XATTR_REPLACE);
3303 lustre_loa_swab(loa, false);
3304 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3305 sizeof(*loa), XATTR_REPLACE);
3313 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3314 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3315 * To have compatilibility with 1.8 ldiskfs driver we need to have
3316 * magic number at start of fid data.
3317 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3320 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3321 const struct lu_fid *fid)
3323 if (!fid_is_namespace_visible(fid) ||
3324 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3325 param->edp_magic = 0;
3329 param->edp_magic = LDISKFS_LUFID_MAGIC;
3330 param->edp_len = sizeof(struct lu_fid) + 1;
3331 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3335 * Try to read the fid from inode ea into dt_rec.
3337 * \param fid object fid.
3339 * \retval 0 on success
3341 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3342 __u32 ino, struct lu_fid *fid,
3343 struct osd_inode_id *id)
3345 struct osd_thread_info *info = osd_oti_get(env);
3346 struct inode *inode;
3349 osd_id_gen(id, ino, OSD_OII_NOGEN);
3350 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3352 RETURN(PTR_ERR(inode));
3358 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3360 struct inode *parent_dir,
3361 const struct lu_fid *dot_fid,
3362 const struct lu_fid *dot_dot_fid,
3363 struct osd_thandle *oth)
3365 struct ldiskfs_dentry_param *dot_ldp;
3366 struct ldiskfs_dentry_param *dot_dot_ldp;
3367 __u32 saved_nlink = dir->i_nlink;
3370 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3371 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3373 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3374 dot_ldp->edp_magic = 0;
3376 rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3377 dir, dot_ldp, dot_dot_ldp);
3378 /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3379 * the subseqent ref_add() will increase the dir->i_nlink
3380 * as 3. That is incorrect for new created directory.
3382 * It looks like hack, because we want to make the OSD API
3383 * to be order-independent for new created directory object
3384 * between dt_insert(..) and ref_add() operations.
3386 * Here, we only restore the in-RAM dir-inode's nlink attr,
3387 * becuase if the nlink attr is not 2, then there will be
3388 * ref_add() called following the dt_insert(..), such call
3389 * will make both the in-RAM and on-disk dir-inode's nlink
3390 * attr to be set as 2. LU-7447 */
3391 set_nlink(dir, saved_nlink);
3396 * Create an local agent inode for remote entry
3398 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3399 struct osd_device *osd,
3400 struct osd_object *pobj,
3401 const struct lu_fid *fid,
3405 struct osd_thread_info *info = osd_oti_get(env);
3406 struct inode *local;
3407 struct osd_thandle *oh;
3412 oh = container_of(th, struct osd_thandle, ot_super);
3413 LASSERT(oh->ot_handle->h_transaction != NULL);
3415 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3416 if (IS_ERR(local)) {
3417 CERROR("%s: create local error %d\n", osd_name(osd),
3418 (int)PTR_ERR(local));
3422 /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3423 * correct gid on remote file, not agent here */
3424 local->i_gid = current_fsgid();
3425 ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3426 unlock_new_inode(local);
3428 /* Agent inode should not have project ID */
3429 #ifdef HAVE_PROJECT_QUOTA
3430 if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3431 i_projid_read(pobj->oo_inode) != 0) {
3432 rc = ldiskfs_transfer_project(local, 0);
3434 CERROR("%s: quota transfer failed: rc = %d. Is project "
3435 "quota enforcement enabled on the ldiskfs "
3436 "filesystem?\n", local->i_sb->s_id, rc);
3437 RETURN(ERR_PTR(rc));
3441 /* Set special LMA flag for local agent inode */
3442 rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3444 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3445 osd_name(osd), PFID(fid), rc);
3446 RETURN(ERR_PTR(rc));
3452 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3453 lu_object_fid(&pobj->oo_dt.do_lu),
3456 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3457 osd_name(osd), PFID(fid), rc);
3458 RETURN(ERR_PTR(rc));
3465 * when direntry is deleted, we have to take care of possible agent inode
3466 * referenced by that. unfortunately we can't do this at that point:
3467 * iget() within a running transaction leads to deadlock and we better do
3468 * not call that every delete declaration to save performance. so we put
3469 * a potention agent inode on a list and process that once the transaction
3470 * is over. Notice it's not any worse in terms of real orphans as regular
3471 * object destroy doesn't put inodes on the on-disk orphan list. this should
3472 * be addressed separately
3474 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3475 struct osd_thandle *oh,
3478 struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3479 struct osd_obj_orphan *oor;
3486 oor->oor_env = (struct lu_env *)env;
3487 spin_lock(&osd->od_osfs_lock);
3488 list_add(&oor->oor_list, &osd->od_orphan_list);
3489 spin_unlock(&osd->od_osfs_lock);
3491 oh->ot_remove_agents = 1;
3497 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3498 struct osd_device *osd)
3500 struct osd_thread_info *info = osd_oti_get(env);
3501 struct osd_obj_orphan *oor, *tmp;
3502 struct osd_inode_id id;
3503 struct list_head list;
3504 struct inode *inode;
3509 INIT_LIST_HEAD(&list);
3511 spin_lock(&osd->od_osfs_lock);
3512 list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3513 if (oor->oor_env == env) {
3514 list_del(&oor->oor_list);
3515 list_add(&oor->oor_list, &list);
3518 spin_unlock(&osd->od_osfs_lock);
3520 list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3524 list_del(&oor->oor_list);
3527 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3528 inode = osd_iget_fid(info, osd, &id, &fid);
3532 if (!osd_remote_fid(env, osd, &fid)) {
3537 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3539 mark_inode_dirty(inode);
3540 ldiskfs_journal_stop(jh);
3548 * OSD layer object create function for OST objects (b=11826).
3550 * The FID is inserted into inode xattr here.
3552 * \retval 0, on success
3553 * \retval -ve, on error
3555 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3556 struct lu_attr *attr, struct dt_allocation_hint *hint,
3557 struct dt_object_format *dof, struct thandle *th)
3559 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3560 struct osd_object *obj = osd_dt_obj(dt);
3561 struct osd_thread_info *info = osd_oti_get(env);
3562 int result, on_ost = 0;
3566 if (dt_object_exists(dt))
3569 LINVRNT(osd_invariant(obj));
3570 LASSERT(!dt_object_remote(dt));
3571 LASSERT(osd_is_write_locked(env, obj));
3572 LASSERT(th != NULL);
3574 if (unlikely(fid_is_acct(fid)))
3575 /* Quota files can't be created from the kernel any more,
3576 * 'tune2fs -O quota' will take care of creating them */
3579 result = __osd_create(info, obj, attr, hint, dof, th);
3581 if (fid_is_idif(fid) &&
3582 !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3583 struct lu_fid *tfid = &info->oti_fid;
3584 struct ost_id *oi = &info->oti_ostid;
3586 fid_to_ostid(fid, oi);
3587 ostid_to_fid(tfid, oi, 0);
3589 result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3590 LMAC_FID_ON_OST, 0);
3592 on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3594 result = osd_ea_fid_set(info, obj->oo_inode, fid,
3595 on_ost ? LMAC_FID_ON_OST : 0,
3598 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3599 obj->oo_dt.do_body_ops = &osd_body_ops;
3603 result = __osd_oi_insert(env, obj, fid, th);
3605 /* a small optimization - dt_insert() isn't usually applied
3606 * to OST objects, so we don't need to cache OI mapping for
3608 if (result == 0 && on_ost == 0) {
3609 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3610 result = osd_idc_find_and_init(env, osd, obj);
3611 LASSERT(result == 0);
3614 LASSERT(ergo(result == 0,
3615 dt_object_exists(dt) && !dt_object_remote(dt)));
3616 LINVRNT(osd_invariant(obj));
3620 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3621 struct thandle *handle)
3623 struct osd_thandle *oh;
3625 /* it's possible that object doesn't exist yet */
3626 LASSERT(handle != NULL);
3628 oh = container_of0(handle, struct osd_thandle, ot_super);
3629 LASSERT(oh->ot_handle == NULL);
3631 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3632 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3638 * Concurrency: @dt is write locked.
3640 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3643 struct osd_object *obj = osd_dt_obj(dt);
3644 struct inode *inode = obj->oo_inode;
3645 struct osd_thandle *oh;
3648 if (!dt_object_exists(dt) || obj->oo_destroyed)
3651 LINVRNT(osd_invariant(obj));
3652 LASSERT(!dt_object_remote(dt));
3653 LASSERT(osd_is_write_locked(env, obj));
3654 LASSERT(th != NULL);
3656 oh = container_of0(th, struct osd_thandle, ot_super);
3657 LASSERT(oh->ot_handle != NULL);
3659 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3661 CDEBUG(D_INODE, DFID" increase nlink %d\n",
3662 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3664 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3665 * (65000) subdirectories by storing "1" in i_nlink if the link count
3666 * would otherwise overflow. Directory tranversal tools understand
3667 * that (st_nlink == 1) indicates that the filesystem dose not track
3668 * hard links count on the directory, and will not abort subdirectory
3669 * scanning early once (st_nlink - 2) subdirs have been found.
3671 * This also has to properly handle the case of inodes with nlink == 0
3672 * in case they are being linked into the PENDING directory
3674 spin_lock(&obj->oo_guard);
3675 if (unlikely(inode->i_nlink == 0))
3676 /* inc_nlink from 0 may cause WARN_ON */
3677 set_nlink(inode, 1);
3679 ldiskfs_inc_count(oh->ot_handle, inode);
3680 if (!S_ISDIR(inode->i_mode))
3681 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3683 spin_unlock(&obj->oo_guard);
3685 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3686 LINVRNT(osd_invariant(obj));
3688 osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3693 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3694 struct thandle *handle)
3696 struct osd_thandle *oh;
3698 if (!dt_object_exists(dt))
3701 LASSERT(!dt_object_remote(dt));
3702 LASSERT(handle != NULL);
3704 oh = container_of0(handle, struct osd_thandle, ot_super);
3705 LASSERT(oh->ot_handle == NULL);
3707 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3708 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3714 * Concurrency: @dt is write locked.
3716 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3719 struct osd_object *obj = osd_dt_obj(dt);
3720 struct inode *inode = obj->oo_inode;
3721 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3722 struct osd_thandle *oh;
3724 if (!dt_object_exists(dt))
3727 LINVRNT(osd_invariant(obj));
3728 LASSERT(!dt_object_remote(dt));
3729 LASSERT(osd_is_write_locked(env, obj));
3730 LASSERT(th != NULL);
3732 oh = container_of0(th, struct osd_thandle, ot_super);
3733 LASSERT(oh->ot_handle != NULL);
3735 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3737 spin_lock(&obj->oo_guard);
3738 /* That can be result of upgrade from old Lustre version and
3739 * applied only to local files. Just skip this ref_del call.
3740 * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3741 if (inode->i_nlink == 0) {
3742 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3743 D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3744 ", maybe an upgraded file? (LU-3915)\n",
3745 osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3746 spin_unlock(&obj->oo_guard);
3750 CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3751 PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3753 ldiskfs_dec_count(oh->ot_handle, inode);
3754 spin_unlock(&obj->oo_guard);
3756 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3757 LINVRNT(osd_invariant(obj));
3759 osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3765 * Concurrency: @dt is read locked.
3767 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3768 struct lu_buf *buf, const char *name)
3770 struct osd_object *obj = osd_dt_obj(dt);
3771 struct inode *inode = obj->oo_inode;
3772 struct osd_thread_info *info = osd_oti_get(env);
3773 struct dentry *dentry = &info->oti_obj_dentry;
3774 bool cache_xattr = false;
3779 /* version get is not real XATTR but uses xattr API */
3780 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3781 dt_obj_version_t *ver = buf->lb_buf;
3783 /* for version we are just using xattr API but change inode
3785 if (buf->lb_len == 0)
3786 return sizeof(dt_obj_version_t);
3788 if (buf->lb_len < sizeof(dt_obj_version_t))
3791 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3792 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3794 *ver = LDISKFS_I(inode)->i_fs_version;
3796 return sizeof(dt_obj_version_t);
3799 if (!dt_object_exists(dt))
3802 LASSERT(!dt_object_remote(dt));
3803 LASSERT(inode->i_op != NULL);
3804 LASSERT(inode->i_op->getxattr != NULL);
3806 if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3807 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3811 rc = osd_oxc_get(obj, name, buf);
3816 rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3817 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
3818 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3819 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3820 struct filter_fid *ff;
3821 struct ost_layout *ol;
3823 if (!osd_dev(dt->do_lu.lo_dev)->od_is_ost)
3826 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3830 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
3836 if (buf->lb_len == 0 || !buf->lb_buf)
3839 if (buf->lb_len < rc)
3843 ol = &ff->ff_layout;
3844 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3845 PFID_STRIPE_IDX_BITS);
3846 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3847 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3848 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3849 if (lma->lma_compat & LMAC_COMP_INFO) {
3850 ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3851 ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3852 ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3854 ol->ol_comp_start = 0;
3855 ol->ol_comp_end = 0;
3862 if (rc == -ENOENT || rc == -ENODATA)
3863 osd_oxc_add(obj, name, NULL, 0);
3864 else if (rc > 0 && buf->lb_buf != NULL)
3865 osd_oxc_add(obj, name, buf->lb_buf, rc);
3871 static int osd_declare_xattr_set(const struct lu_env *env,
3872 struct dt_object *dt,
3873 const struct lu_buf *buf, const char *name,
3874 int fl, struct thandle *handle)
3876 struct osd_thandle *oh;
3878 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3880 LASSERT(handle != NULL);
3882 oh = container_of0(handle, struct osd_thandle, ot_super);
3883 LASSERT(oh->ot_handle == NULL);
3885 if (strcmp(name, XATTR_NAME_LMA) == 0) {
3886 /* For non-upgrading case, the LMA is set first and
3887 * usually fit inode. But for upgrade case, the LMA
3888 * may be in another separated EA block. */
3889 if (dt_object_exists(dt)) {
3890 if (fl == LU_XATTR_REPLACE)
3895 } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3897 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3898 /* We may need to delete the old PFID EA. */
3899 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3900 if (fl == LU_XATTR_REPLACE)
3907 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3912 if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3913 /* learn xattr size from osd_xattr_get if
3914 attribute has not been read yet */
3915 buflen = __osd_xattr_get(
3916 osd_dt_obj(dt)->oo_inode,
3917 &osd_oti_get(env)->oti_obj_dentry,
3922 buflen = buf->lb_len;
3925 if (buflen > sb->s_blocksize) {
3926 credits += osd_calc_bkmap_credits(
3928 (buflen + sb->s_blocksize - 1) >>
3929 sb->s_blocksize_bits);
3933 * xattr set may involve inode quota change, reserve credits for
3934 * dquot_initialize()
3936 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3939 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3945 * Concurrency: @dt is write locked.
3947 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3948 const struct lu_buf *buf, const char *name, int fl,
3949 struct thandle *handle)
3951 struct osd_object *obj = osd_dt_obj(dt);
3952 struct inode *inode = obj->oo_inode;
3953 struct osd_thread_info *info = osd_oti_get(env);
3954 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3955 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3964 /* version set is not real XATTR */
3965 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3966 dt_obj_version_t *version = buf->lb_buf;
3968 /* for version we are just using xattr API but change inode
3970 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3972 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
3973 *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3975 LDISKFS_I(inode)->i_fs_version = *version;
3976 /* Version is set after all inode operations are finished,
3977 * so we should mark it dirty here */
3978 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3983 CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3984 PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3987 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3988 if (fl & LU_XATTR_REPLACE)
3989 fs_flags |= XATTR_REPLACE;
3991 if (fl & LU_XATTR_CREATE)
3992 fs_flags |= XATTR_CREATE;
3994 /* For the OST device with 256 bytes inode size by default,
3995 * the PFID EA will be stored together with LMA EA to avoid
3996 * performance trouble. Otherwise the PFID EA can be stored
3997 * independently. LU-8998 */
3998 if (strcmp(name, XATTR_NAME_FID) == 0 &&
3999 LDISKFS_INODE_SIZE(inode->i_sb) <= 256) {
4000 struct dentry *dentry = &info->oti_obj_dentry;
4001 struct filter_fid *ff;
4002 struct ost_layout *ol;
4005 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4008 ol = &ff->ff_layout;
4009 /* Old client does not send stripe information, store
4010 * the PFID EA on disk directly. */
4011 if (buf->lb_len == sizeof(struct lu_fid) ||
4012 ol->ol_stripe_size == 0) {
4013 len = sizeof(struct lu_fid);
4017 if (buf->lb_len != sizeof(*ff))
4020 rc = osd_get_lma(info, inode, dentry, loa);
4021 if (unlikely(rc == -ENODATA)) {
4022 /* Usually for upgarding from old device */
4023 lustre_loa_init(loa, lu_object_fid(&dt->do_lu),
4024 LMAC_FID_ON_OST, 0);
4032 fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4033 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4034 PFID_STRIPE_IDX_BITS;
4035 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4036 lma->lma_compat |= LMAC_STRIPE_INFO;
4037 if (ol->ol_comp_id != 0) {
4038 loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4039 loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4040 loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4041 lma->lma_compat |= LMAC_COMP_INFO;
4044 lustre_loa_swab(loa, false);
4046 /* Remove old PFID EA entry firstly. */
4047 ll_vfs_dq_init(inode);
4048 rc = inode->i_op->removexattr(dentry, name);
4049 if (rc && rc != -ENODATA)
4052 /* Store the PFID EA inside the LMA EA. */
4053 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
4057 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4058 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4062 lma->lma_incompat |= LMAI_STRIPED;
4063 lustre_lma_swab(lma);
4064 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4065 sizeof(*lma), XATTR_REPLACE);
4071 rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4072 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4075 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4076 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4077 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4083 * Concurrency: @dt is read locked.
4085 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4086 const struct lu_buf *buf)
4088 struct osd_object *obj = osd_dt_obj(dt);
4089 struct inode *inode = obj->oo_inode;
4090 struct osd_thread_info *info = osd_oti_get(env);
4091 struct dentry *dentry = &info->oti_obj_dentry;
4093 if (!dt_object_exists(dt))
4096 LASSERT(!dt_object_remote(dt));
4097 LASSERT(inode->i_op != NULL);
4098 LASSERT(inode->i_op->listxattr != NULL);
4100 dentry->d_inode = inode;
4101 dentry->d_sb = inode->i_sb;
4102 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4105 static int osd_declare_xattr_del(const struct lu_env *env,
4106 struct dt_object *dt, const char *name,
4107 struct thandle *handle)
4109 struct osd_thandle *oh;
4110 struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4112 LASSERT(!dt_object_remote(dt));
4113 LASSERT(handle != NULL);
4115 oh = container_of0(handle, struct osd_thandle, ot_super);
4116 LASSERT(oh->ot_handle == NULL);
4118 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4119 osd_dto_credits_noquota[DTO_XATTR_SET]);
4121 * xattr del may involve inode quota change, reserve credits for
4122 * dquot_initialize()
4124 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4130 * Concurrency: @dt is write locked.
4132 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4133 const char *name, struct thandle *handle)
4135 struct osd_object *obj = osd_dt_obj(dt);
4136 struct inode *inode = obj->oo_inode;
4137 struct osd_thread_info *info = osd_oti_get(env);
4138 struct dentry *dentry = &info->oti_obj_dentry;
4141 if (!dt_object_exists(dt))
4144 LASSERT(!dt_object_remote(dt));
4145 LASSERT(inode->i_op != NULL);
4146 LASSERT(inode->i_op->removexattr != NULL);
4147 LASSERT(handle != NULL);
4149 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4151 ll_vfs_dq_init(inode);
4152 dentry->d_inode = inode;
4153 dentry->d_sb = inode->i_sb;
4154 rc = inode->i_op->removexattr(dentry, name);
4155 if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) {
4156 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4158 LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost);
4160 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4161 &info->oti_ost_attrs);
4163 if (!(lma->lma_compat & LMAC_STRIPE_INFO)) {
4168 lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4169 lustre_lma_swab(lma);
4170 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4171 sizeof(*lma), XATTR_REPLACE);
4176 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4179 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4180 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4181 osd_oxc_del(obj, name);
4186 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4187 __u64 start, __u64 end)
4189 struct osd_object *obj = osd_dt_obj(dt);
4190 struct inode *inode = obj->oo_inode;
4191 struct osd_thread_info *info = osd_oti_get(env);
4192 struct dentry *dentry = &info->oti_obj_dentry;
4193 struct file *file = &info->oti_file;
4198 dentry->d_inode = inode;
4199 dentry->d_sb = inode->i_sb;
4200 file->f_path.dentry = dentry;
4201 file->f_mapping = inode->i_mapping;
4202 file->f_op = inode->i_fop;
4203 set_file_inode(file, inode);
4205 rc = ll_vfs_fsync_range(file, start, end, 0);
4210 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4219 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4220 const struct dt_index_features *feat)
4222 struct iam_descr *descr;
4224 if (osd_object_is_root(o))
4225 return feat == &dt_directory_features;
4227 LASSERT(o->oo_dir != NULL);
4229 descr = o->oo_dir->od_container.ic_descr;
4230 if (feat == &dt_directory_features) {
4231 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4237 feat->dif_keysize_min <= descr->id_key_size &&
4238 descr->id_key_size <= feat->dif_keysize_max &&
4239 feat->dif_recsize_min <= descr->id_rec_size &&
4240 descr->id_rec_size <= feat->dif_recsize_max &&
4241 !(feat->dif_flags & (DT_IND_VARKEY |
4242 DT_IND_VARREC | DT_IND_NONUNQ)) &&
4243 ergo(feat->dif_flags & DT_IND_UPDATE,
4244 1 /* XXX check that object (and file system) is
4249 static int osd_iam_container_init(const struct lu_env *env,
4250 struct osd_object *obj,
4251 struct osd_directory *dir)
4253 struct iam_container *bag = &dir->od_container;
4256 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4260 result = iam_container_setup(bag);
4262 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4264 iam_container_fini(bag);
4271 * Concurrency: no external locking is necessary.
4273 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4274 const struct dt_index_features *feat)
4278 struct osd_object *obj = osd_dt_obj(dt);
4280 LINVRNT(osd_invariant(obj));
4282 if (osd_object_is_root(obj)) {
4283 dt->do_index_ops = &osd_index_ea_ops;
4285 } else if (feat == &dt_directory_features) {
4286 dt->do_index_ops = &osd_index_ea_ops;
4287 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4292 } else if (unlikely(feat == &dt_otable_features)) {
4293 dt->do_index_ops = &osd_otable_ops;
4295 } else if (unlikely(feat == &dt_acct_features)) {
4296 dt->do_index_ops = &osd_acct_index_ops;
4299 } else if (!osd_has_index(obj)) {
4300 struct osd_directory *dir;
4305 spin_lock(&obj->oo_guard);
4306 if (obj->oo_dir == NULL)
4310 * Concurrent thread allocated container data.
4313 spin_unlock(&obj->oo_guard);
4315 * Now, that we have container data, serialize its
4318 down_write(&obj->oo_ext_idx_sem);
4320 * recheck under lock.
4322 if (!osd_has_index(obj))
4323 result = osd_iam_container_init(env, obj,
4327 up_write(&obj->oo_ext_idx_sem);
4335 if (result == 0 && skip_iam == 0) {
4336 if (!osd_iam_index_probe(env, obj, feat))
4339 LINVRNT(osd_invariant(obj));
4344 static int osd_otable_it_attr_get(const struct lu_env *env,
4345 struct dt_object *dt,
4346 struct lu_attr *attr)
4352 static const struct dt_object_operations osd_obj_ops = {
4353 .do_read_lock = osd_read_lock,
4354 .do_write_lock = osd_write_lock,
4355 .do_read_unlock = osd_read_unlock,
4356 .do_write_unlock = osd_write_unlock,
4357 .do_write_locked = osd_write_locked,
4358 .do_attr_get = osd_attr_get,
4359 .do_declare_attr_set = osd_declare_attr_set,
4360 .do_attr_set = osd_attr_set,
4361 .do_ah_init = osd_ah_init,
4362 .do_declare_create = osd_declare_create,
4363 .do_create = osd_create,
4364 .do_declare_destroy = osd_declare_destroy,
4365 .do_destroy = osd_destroy,
4366 .do_index_try = osd_index_try,
4367 .do_declare_ref_add = osd_declare_ref_add,
4368 .do_ref_add = osd_ref_add,
4369 .do_declare_ref_del = osd_declare_ref_del,
4370 .do_ref_del = osd_ref_del,
4371 .do_xattr_get = osd_xattr_get,
4372 .do_declare_xattr_set = osd_declare_xattr_set,
4373 .do_xattr_set = osd_xattr_set,
4374 .do_declare_xattr_del = osd_declare_xattr_del,
4375 .do_xattr_del = osd_xattr_del,
4376 .do_xattr_list = osd_xattr_list,
4377 .do_object_sync = osd_object_sync,
4378 .do_invalidate = osd_invalidate,
4381 static const struct dt_object_operations osd_obj_otable_it_ops = {
4382 .do_attr_get = osd_otable_it_attr_get,
4383 .do_index_try = osd_index_try,
4386 static int osd_index_declare_iam_delete(const struct lu_env *env,
4387 struct dt_object *dt,
4388 const struct dt_key *key,
4389 struct thandle *handle)
4391 struct osd_thandle *oh;
4393 oh = container_of0(handle, struct osd_thandle, ot_super);
4394 LASSERT(oh->ot_handle == NULL);
4396 /* Recycle may cause additional three blocks to be changed. */
4397 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4398 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4404 * delete a (key, value) pair from index \a dt specified by \a key
4406 * \param dt osd index object
4407 * \param key key for index
4408 * \param rec record reference
4409 * \param handle transaction handler
4412 * \retval -ve failure
4414 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4415 const struct dt_key *key,
4416 struct thandle *handle)
4418 struct osd_thread_info *oti = osd_oti_get(env);
4419 struct osd_object *obj = osd_dt_obj(dt);
4420 struct osd_thandle *oh;
4421 struct iam_path_descr *ipd;
4422 struct iam_container *bag = &obj->oo_dir->od_container;
4426 if (!dt_object_exists(dt))
4429 LINVRNT(osd_invariant(obj));
4430 LASSERT(!dt_object_remote(dt));
4431 LASSERT(bag->ic_object == obj->oo_inode);
4432 LASSERT(handle != NULL);
4434 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4436 ipd = osd_idx_ipd_get(env, bag);
4437 if (unlikely(ipd == NULL))
4440 oh = container_of0(handle, struct osd_thandle, ot_super);
4441 LASSERT(oh->ot_handle != NULL);
4442 LASSERT(oh->ot_handle->h_transaction != NULL);
4444 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4445 /* swab quota uid/gid provided by caller */
4446 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4447 key = (const struct dt_key *)&oti->oti_quota_id;
4450 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4451 osd_ipd_put(env, bag, ipd);
4452 LINVRNT(osd_invariant(obj));
4453 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4457 static int osd_index_declare_ea_delete(const struct lu_env *env,
4458 struct dt_object *dt,
4459 const struct dt_key *key,
4460 struct thandle *handle)
4462 struct osd_thandle *oh;
4463 struct inode *inode;
4467 LASSERT(!dt_object_remote(dt));
4468 LASSERT(handle != NULL);
4470 oh = container_of0(handle, struct osd_thandle, ot_super);
4471 LASSERT(oh->ot_handle == NULL);
4473 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4474 if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) {
4475 /* '..' to a remote object has a local representative */
4476 credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
4477 /* to reset LMAI_REMOTE_PARENT */
4480 osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4482 inode = osd_dt_obj(dt)->oo_inode;
4486 rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4487 i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4492 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4495 struct osd_fid_pack *rec;
4498 if (de->file_type & LDISKFS_DIRENT_LUFID) {
4499 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4500 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4501 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4507 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4508 const struct lu_fid *fid)
4510 struct seq_server_site *ss = osd_seq_site(osd);
4513 /* FID seqs not in FLDB, must be local seq */
4514 if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4517 /* If FLD is not being initialized yet, it only happens during the
4518 * initialization, likely during mgs initialization, and we assume
4519 * this is local FID. */
4520 if (ss == NULL || ss->ss_server_fld == NULL)
4523 /* Only check the local FLDB here */
4524 if (osd_seq_exists(env, osd, fid_seq(fid)))
4531 * Index delete function for interoperability mode (b11826).
4532 * It will remove the directory entry added by osd_index_ea_insert().
4533 * This entry is needed to maintain name->fid mapping.
4535 * \param key, key i.e. file entry to be deleted
4537 * \retval 0, on success
4538 * \retval -ve, on error
4540 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4541 const struct dt_key *key, struct thandle *handle)
4543 struct osd_object *obj = osd_dt_obj(dt);
4544 struct inode *dir = obj->oo_inode;
4545 struct dentry *dentry;
4546 struct osd_thandle *oh;
4547 struct ldiskfs_dir_entry_2 *de = NULL;
4548 struct buffer_head *bh;
4549 struct htree_lock *hlock = NULL;
4550 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
4551 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4555 if (!dt_object_exists(dt))
4558 LINVRNT(osd_invariant(obj));
4559 LASSERT(!dt_object_remote(dt));
4560 LASSERT(handle != NULL);
4562 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4564 oh = container_of(handle, struct osd_thandle, ot_super);
4565 LASSERT(oh->ot_handle != NULL);
4566 LASSERT(oh->ot_handle->h_transaction != NULL);
4568 ll_vfs_dq_init(dir);
4569 dentry = osd_child_dentry_get(env, obj,
4570 (char *)key, strlen((char *)key));
4572 if (obj->oo_hl_head != NULL) {
4573 hlock = osd_oti_get(env)->oti_hlock;
4574 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4575 dir, LDISKFS_HLOCK_DEL);
4577 down_write(&obj->oo_ext_idx_sem);
4580 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4582 /* If this is not the ".." entry, it might be a remote DNE
4583 * entry and we need to check if the FID is for a remote
4584 * MDT. If the FID is not in the directory entry (e.g.
4585 * upgraded 1.8 filesystem without dirdata enabled) then
4586 * we need to get the FID from the LMA. For a remote directory
4587 * there HAS to be an LMA, it cannot be an IGIF inode in this
4590 * Delete the entry before the agent inode in order to
4591 * simplify error handling. At worst an error after deleting
4592 * the entry first might leak the agent inode afterward. The
4593 * reverse would need filesystem abort in case of error deleting
4594 * the entry after the agent had been removed, or leave a
4595 * dangling entry pointing at a random inode. */
4596 if (strcmp((char *)key, dotdot) != 0) {
4597 LASSERT(de != NULL);
4598 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4599 if (rc == -ENODATA) {
4600 /* can't get FID, postpone to the end of the
4601 * transaction when iget() is safe */
4602 osd_schedule_agent_inode_removal(env, oh,
4603 le32_to_cpu(de->inode));
4604 } else if (rc == 0 &&
4605 unlikely(osd_remote_fid(env, osd, fid))) {
4606 osd_schedule_agent_inode_removal(env, oh,
4607 le32_to_cpu(de->inode));
4610 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4616 ldiskfs_htree_unlock(hlock);
4618 up_write(&obj->oo_ext_idx_sem);
4623 /* For inode on the remote MDT, .. will point to
4624 * /Agent directory, Check whether it needs to delete
4625 * from agent directory */
4626 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
4629 ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj),
4632 /* Sigh, the entry has been deleted, and
4633 * it is not easy to revert it back, so
4634 * let's keep this error private, and let
4635 * LFSCK fix it. XXX */
4636 CERROR("%s: delete remote parent "DFID": rc = %d\n",
4637 osd_name(osd), PFID(fid), ret);
4640 LASSERT(osd_invariant(obj));
4641 osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4646 * Lookup index for \a key and copy record to \a rec.
4648 * \param dt osd index object
4649 * \param key key for index
4650 * \param rec record reference
4652 * \retval +ve success : exact mach
4653 * \retval 0 return record with key not greater than \a key
4654 * \retval -ve failure
4656 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4657 struct dt_rec *rec, const struct dt_key *key)
4659 struct osd_object *obj = osd_dt_obj(dt);
4660 struct iam_path_descr *ipd;
4661 struct iam_container *bag = &obj->oo_dir->od_container;
4662 struct osd_thread_info *oti = osd_oti_get(env);
4663 struct iam_iterator *it = &oti->oti_idx_it;
4664 struct iam_rec *iam_rec;
4668 if (!dt_object_exists(dt))
4671 LASSERT(osd_invariant(obj));
4672 LASSERT(!dt_object_remote(dt));
4673 LASSERT(bag->ic_object == obj->oo_inode);
4675 ipd = osd_idx_ipd_get(env, bag);
4679 /* got ipd now we can start iterator. */
4680 iam_it_init(it, bag, 0, ipd);
4682 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4683 /* swab quota uid/gid provided by caller */
4684 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4685 key = (const struct dt_key *)&oti->oti_quota_id;
4688 rc = iam_it_get(it, (struct iam_key *)key);
4690 if (S_ISDIR(obj->oo_inode->i_mode))
4691 iam_rec = (struct iam_rec *)oti->oti_ldp;
4693 iam_rec = (struct iam_rec *) rec;
4695 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4697 if (S_ISDIR(obj->oo_inode->i_mode))
4698 osd_fid_unpack((struct lu_fid *) rec,
4699 (struct osd_fid_pack *)iam_rec);
4700 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4701 osd_quota_unpack(obj, rec);
4706 osd_ipd_put(env, bag, ipd);
4708 LINVRNT(osd_invariant(obj));
4713 static int osd_index_declare_iam_insert(const struct lu_env *env,
4714 struct dt_object *dt,
4715 const struct dt_rec *rec,
4716 const struct dt_key *key,
4717 struct thandle *handle)
4719 struct osd_thandle *oh;
4721 LASSERT(handle != NULL);
4723 oh = container_of0(handle, struct osd_thandle, ot_super);
4724 LASSERT(oh->ot_handle == NULL);
4726 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4727 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4733 * Inserts (key, value) pair in \a dt index object.
4735 * \param dt osd index object
4736 * \param key key for index
4737 * \param rec record reference
4738 * \param th transaction handler
4741 * \retval -ve failure
4743 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4744 const struct dt_rec *rec,
4745 const struct dt_key *key, struct thandle *th,
4748 struct osd_object *obj = osd_dt_obj(dt);
4749 struct iam_path_descr *ipd;
4750 struct osd_thandle *oh;
4751 struct iam_container *bag;
4752 struct osd_thread_info *oti = osd_oti_get(env);
4753 struct iam_rec *iam_rec;
4757 if (!dt_object_exists(dt))
4760 LINVRNT(osd_invariant(obj));
4761 LASSERT(!dt_object_remote(dt));
4763 bag = &obj->oo_dir->od_container;
4764 LASSERT(bag->ic_object == obj->oo_inode);
4765 LASSERT(th != NULL);
4767 osd_trans_exec_op(env, th, OSD_OT_INSERT);
4769 ipd = osd_idx_ipd_get(env, bag);
4770 if (unlikely(ipd == NULL))
4773 oh = container_of0(th, struct osd_thandle, ot_super);
4774 LASSERT(oh->ot_handle != NULL);
4775 LASSERT(oh->ot_handle->h_transaction != NULL);
4776 if (S_ISDIR(obj->oo_inode->i_mode)) {
4777 iam_rec = (struct iam_rec *)oti->oti_ldp;
4778 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4779 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4780 /* pack quota uid/gid */
4781 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4782 key = (const struct dt_key *)&oti->oti_quota_id;
4783 /* pack quota record */
4784 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4785 iam_rec = (struct iam_rec *)rec;
4787 iam_rec = (struct iam_rec *)rec;
4790 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4792 osd_ipd_put(env, bag, ipd);
4793 LINVRNT(osd_invariant(obj));
4794 osd_trans_exec_check(env, th, OSD_OT_INSERT);
4799 * Calls ldiskfs_add_entry() to add directory entry
4800 * into the directory. This is required for
4801 * interoperability mode (b11826)
4803 * \retval 0, on success
4804 * \retval -ve, on error
4806 static int __osd_ea_add_rec(struct osd_thread_info *info,
4807 struct osd_object *pobj, struct inode *cinode,
4808 const char *name, const struct lu_fid *fid,
4809 struct htree_lock *hlock, struct thandle *th)
4811 struct ldiskfs_dentry_param *ldp;
4812 struct dentry *child;
4813 struct osd_thandle *oth;
4816 oth = container_of(th, struct osd_thandle, ot_super);
4817 LASSERT(oth->ot_handle != NULL);
4818 LASSERT(oth->ot_handle->h_transaction != NULL);
4819 LASSERT(pobj->oo_inode);
4821 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4822 if (unlikely(pobj->oo_inode ==
4823 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4826 osd_get_ldiskfs_dirent_param(ldp, fid);
4827 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4828 child->d_fsdata = (void *)ldp;
4829 ll_vfs_dq_init(pobj->oo_inode);
4830 rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4831 child, cinode, hlock);
4832 if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4833 struct ldiskfs_dir_entry_2 *de;
4834 struct buffer_head *bh;
4837 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4840 rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4843 if (S_ISDIR(cinode->i_mode))
4844 de->file_type = LDISKFS_DIRENT_LUFID |
4845 LDISKFS_FT_REG_FILE;
4847 de->file_type = LDISKFS_DIRENT_LUFID |
4849 ldiskfs_handle_dirty_metadata(oth->ot_handle,
4860 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
4861 * into the directory.Also sets flags into osd object to
4862 * indicate dot and dotdot are created. This is required for
4863 * interoperability mode (b11826)
4865 * \param dir directory for dot and dotdot fixup.
4866 * \param obj child object for linking
4868 * \retval 0, on success
4869 * \retval -ve, on error
4871 static int osd_add_dot_dotdot(struct osd_thread_info *info,
4872 struct osd_object *dir,
4873 struct inode *parent_dir, const char *name,
4874 const struct lu_fid *dot_fid,
4875 const struct lu_fid *dot_dot_fid,
4878 struct inode *inode = dir->oo_inode;
4879 struct osd_thandle *oth;
4882 oth = container_of(th, struct osd_thandle, ot_super);
4883 LASSERT(oth->ot_handle->h_transaction != NULL);
4884 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4886 if (strcmp(name, dot) == 0) {
4887 if (dir->oo_compat_dot_created) {
4890 LASSERT(inode->i_ino == parent_dir->i_ino);
4891 dir->oo_compat_dot_created = 1;
4894 } else if (strcmp(name, dotdot) == 0) {
4895 if (!dir->oo_compat_dot_created)
4897 /* in case of rename, dotdot is already created */
4898 if (dir->oo_compat_dotdot_created) {
4899 return __osd_ea_add_rec(info, dir, parent_dir, name,
4900 dot_dot_fid, NULL, th);
4903 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4904 struct lu_fid tfid = *dot_dot_fid;
4907 result = osd_add_dot_dotdot_internal(info,
4908 dir->oo_inode, parent_dir, dot_fid,
4911 result = osd_add_dot_dotdot_internal(info,
4912 dir->oo_inode, parent_dir, dot_fid,
4917 dir->oo_compat_dotdot_created = 1;
4925 * It will call the appropriate osd_add* function and return the
4926 * value, return by respective functions.
4928 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4929 struct inode *cinode, const char *name,
4930 const struct lu_fid *fid, struct thandle *th)
4932 struct osd_thread_info *info = osd_oti_get(env);
4933 struct htree_lock *hlock;
4936 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4938 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4940 if (hlock != NULL) {
4941 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4944 down_write(&pobj->oo_ext_idx_sem);
4947 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4948 lu_object_fid(&pobj->oo_dt.do_lu),
4951 if (hlock != NULL) {
4952 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4953 pobj->oo_inode, LDISKFS_HLOCK_ADD);
4955 down_write(&pobj->oo_ext_idx_sem);
4958 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4959 struct lu_fid *tfid = &info->oti_fid;
4963 rc = __osd_ea_add_rec(info, pobj, cinode, name,
4966 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4971 ldiskfs_htree_unlock(hlock);
4973 up_write(&pobj->oo_ext_idx_sem);
4979 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4980 struct osd_idmap_cache *oic)
4982 struct osd_scrub *scrub = &dev->od_scrub;
4983 struct lu_fid *fid = &oic->oic_fid;
4984 struct osd_inode_id *id = &oic->oic_lid;
4985 struct inode *inode = NULL;
4991 if (!fid_is_norm(fid) && !fid_is_igif(fid))
4994 if (scrub->os_pos_current > id->oii_ino)
4998 rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
4999 if (rc == -ENOENT) {
5000 __u32 gen = id->oii_gen;
5006 inode = osd_iget(oti, dev, id);
5007 /* The inode has been removed (by race maybe). */
5008 if (IS_ERR(inode)) {
5009 rc = PTR_ERR(inode);
5011 RETURN(rc == -ESTALE ? -ENOENT : rc);
5014 /* The OI mapping is lost. */
5015 if (gen != OSD_OII_NOGEN)
5018 /* The inode may has been reused by others, we do not know,
5019 * leave it to be handled by subsequent osd_fid_lookup(). */
5021 } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5028 if (thread_is_running(&scrub->os_thread)) {
5029 if (inode == NULL) {
5030 inode = osd_iget(oti, dev, id);
5031 /* The inode has been removed (by race maybe). */
5032 if (IS_ERR(inode)) {
5033 rc = PTR_ERR(inode);
5035 RETURN(rc == -ESTALE ? -ENOENT : rc);
5039 rc = osd_oii_insert(dev, oic, insert);
5040 /* There is race condition between osd_oi_lookup and OI scrub.
5041 * The OI scrub finished just after osd_oi_lookup() failure.
5042 * Under such case, it is unnecessary to trigger OI scrub again,
5043 * but try to call osd_oi_lookup() again. */
5044 if (unlikely(rc == -EAGAIN))
5047 if (!S_ISDIR(inode->i_mode))
5050 rc = osd_check_lmv(oti, dev, inode, oic);
5055 if (!dev->od_noscrub && ++once == 1) {
5056 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
5058 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5059 "%s: trigger partial OI scrub for RPC inconsistency "
5060 "checking FID "DFID": rc = %d\n",
5061 osd_dev2name(dev), PFID(fid), rc);
5062 if (rc == 0 || rc == -EALREADY)
5075 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5076 struct osd_device *dev,
5077 struct osd_idmap_cache *oic,
5078 struct lu_fid *fid, __u32 ino)
5080 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5081 struct inode *inode;
5084 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5085 inode = osd_iget(oti, dev, &oic->oic_lid);
5086 if (IS_ERR(inode)) {
5087 fid_zero(&oic->oic_fid);
5088 return PTR_ERR(inode);
5091 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5094 fid_zero(&oic->oic_fid);
5096 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5100 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5101 struct osd_inode_id *id, const struct lu_fid *fid)
5103 CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5104 id->oii_ino, id->oii_gen, info);
5105 info->oti_cache.oic_lid = *id;
5106 info->oti_cache.oic_fid = *fid;
5107 info->oti_cache.oic_dev = osd;
5111 * Get parent FID from the linkEA.
5113 * For a directory which parent resides on remote MDT, to satisfy the
5114 * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5115 * the other hand, to make the lookup(..) on the directory can return
5116 * the real parent FID, we append the real parent FID after its ".."
5117 * name entry in the /REMOTE_PARENT_DIR.
5119 * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5120 * backup. So after the restore, we cannot get the right parent FID from
5121 * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5122 * we have stored the real parent FID in the directory object's linkEA,
5123 * we can parse the linkEA for the real parent FID.
5125 * \param[in] env pointer to the thread context
5126 * \param[in] obj pointer to the object to be handled
5127 * \param[out]fid pointer to the buffer to hold the parent FID
5129 * \retval 0 for getting the real parent FID successfully
5130 * \retval negative error number on failure
5132 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5133 struct osd_object *obj,
5136 struct osd_thread_info *oti = osd_oti_get(env);
5137 struct lu_buf *buf = &oti->oti_big_buf;
5138 struct dentry *dentry = &oti->oti_obj_dentry;
5139 struct inode *inode = obj->oo_inode;
5140 struct linkea_data ldata = { NULL };
5145 if (!S_ISDIR(inode->i_mode))
5149 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5150 buf->lb_buf, buf->lb_len);
5151 if (rc == -ERANGE) {
5152 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5155 lu_buf_realloc(buf, rc);
5156 if (buf->lb_buf == NULL)
5163 if (unlikely(rc == 0))
5169 if (unlikely(buf->lb_buf == NULL)) {
5170 lu_buf_realloc(buf, rc);
5171 if (buf->lb_buf == NULL)
5178 rc = linkea_init_with_rec(&ldata);
5180 linkea_first_entry(&ldata);
5181 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5187 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5188 struct inode *inode,
5189 const struct lu_fid *pfid,
5190 const char *name, const int namelen)
5192 struct osd_thread_info *oti = osd_oti_get(env);
5193 struct lu_buf *buf = &oti->oti_big_buf;
5194 struct dentry *dentry = &oti->oti_obj_dentry;
5195 struct linkea_data ldata = { NULL };
5196 struct lu_name cname = { .ln_name = name,
5197 .ln_namelen = namelen };
5202 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5203 buf->lb_buf, buf->lb_len);
5205 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5210 if (unlikely(rc == 0))
5213 if (buf->lb_len < rc) {
5214 lu_buf_realloc(buf, rc);
5215 if (buf->lb_buf == NULL)
5222 rc = linkea_init_with_rec(&ldata);
5224 rc = linkea_links_find(&ldata, &cname, pfid);
5230 * Calls ->lookup() to find dentry. From dentry get inode and
5231 * read inode's ea to get fid. This is required for interoperability
5234 * \retval 0, on success
5235 * \retval -ve, on error
5237 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5238 struct dt_rec *rec, const struct dt_key *key)
5240 struct inode *dir = obj->oo_inode;
5241 struct dentry *dentry;
5242 struct ldiskfs_dir_entry_2 *de;
5243 struct buffer_head *bh;
5244 struct lu_fid *fid = (struct lu_fid *) rec;
5245 struct htree_lock *hlock = NULL;
5250 LASSERT(dir->i_op != NULL);
5251 LASSERT(dir->i_op->lookup != NULL);
5253 dentry = osd_child_dentry_get(env, obj,
5254 (char *)key, strlen((char *)key));
5256 if (obj->oo_hl_head != NULL) {
5257 hlock = osd_oti_get(env)->oti_hlock;
5258 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5259 dir, LDISKFS_HLOCK_LOOKUP);
5261 down_read(&obj->oo_ext_idx_sem);
5264 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5266 struct osd_thread_info *oti = osd_oti_get(env);
5267 struct osd_inode_id *id = &oti->oti_id;
5268 struct osd_idmap_cache *oic = &oti->oti_cache;
5269 struct osd_device *dev = osd_obj2dev(obj);
5271 ino = le32_to_cpu(de->inode);
5272 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5274 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5278 rc = osd_get_fid_from_dentry(de, rec);
5280 /* done with de, release bh */
5283 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5284 const char *name = (const char *)key;
5286 /* If the parent is on remote MDT, and there
5287 * is no FID-in-dirent, then we have to get
5288 * the parent FID from the linkEA. */
5289 if (likely(strlen(name) == 2 &&
5290 name[0] == '.' && name[1] == '.'))
5291 rc = osd_get_pfid_from_linkea(env, obj,
5294 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5297 osd_id_gen(id, ino, OSD_OII_NOGEN);
5300 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5301 fid_zero(&oic->oic_fid);
5306 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5307 rc = osd_consistency_check(oti, dev, oic);
5309 fid_zero(&oic->oic_fid);
5318 ldiskfs_htree_unlock(hlock);
5320 up_read(&obj->oo_ext_idx_sem);
5324 static int osd_index_declare_ea_insert(const struct lu_env *env,
5325 struct dt_object *dt,
5326 const struct dt_rec *rec,
5327 const struct dt_key *key,
5328 struct thandle *handle)
5330 struct osd_thandle *oh;
5331 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5332 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5333 const struct lu_fid *fid = rec1->rec_fid;
5334 int credits, rc = 0;
5335 struct osd_idmap_cache *idc;
5338 LASSERT(!dt_object_remote(dt));
5339 LASSERT(handle != NULL);
5340 LASSERT(fid != NULL);
5341 LASSERT(rec1->rec_type != 0);
5343 oh = container_of0(handle, struct osd_thandle, ot_super);
5344 LASSERT(oh->ot_handle == NULL);
5346 credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5348 /* we can't call iget() while a transactions is running
5349 * (this can lead to a deadlock), but we need to know
5350 * inum and object type. so we find this information at
5351 * declaration and cache in per-thread info */
5352 idc = osd_idc_find_or_init(env, osd, fid);
5354 RETURN(PTR_ERR(idc));
5355 if (idc->oic_remote) {
5356 /* a reference to remote inode is represented by an
5357 * agent inode which we have to create */
5358 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5359 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5362 osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5364 if (osd_dt_obj(dt)->oo_inode != NULL) {
5365 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5367 /* We ignore block quota on meta pool (MDTs), so needn't
5368 * calculate how many blocks will be consumed by this index
5370 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5372 i_projid_read(inode), 0,
5373 oh, osd_dt_obj(dt), NULL,
5378 #ifdef HAVE_PROJECT_QUOTA
5379 /* Reserve credits for local agent inode to transfer
5380 * to 0, quota enforcement is ignored in this case.
5382 if (idc->oic_remote &&
5383 LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5384 i_projid_read(inode) != 0)
5385 rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5386 0, i_projid_read(inode),
5387 0, false, PRJQUOTA);
5396 * Index add function for interoperability mode (b11826).
5397 * It will add the directory entry.This entry is needed to
5398 * maintain name->fid mapping.
5400 * \param key it is key i.e. file entry to be inserted
5401 * \param rec it is value of given key i.e. fid
5403 * \retval 0, on success
5404 * \retval -ve, on error
5406 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5407 const struct dt_rec *rec,
5408 const struct dt_key *key, struct thandle *th,
5411 struct osd_object *obj = osd_dt_obj(dt);
5412 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5413 struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5414 const struct lu_fid *fid = rec1->rec_fid;
5415 const char *name = (const char *)key;
5416 struct osd_thread_info *oti = osd_oti_get(env);
5417 struct inode *child_inode = NULL;
5418 struct osd_idmap_cache *idc;
5422 if (!dt_object_exists(dt))
5425 LASSERT(osd_invariant(obj));
5426 LASSERT(!dt_object_remote(dt));
5427 LASSERT(th != NULL);
5429 osd_trans_exec_op(env, th, OSD_OT_INSERT);
5431 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5433 idc = osd_idc_find(env, osd, fid);
5434 if (unlikely(idc == NULL)) {
5435 /* this dt_insert() wasn't declared properly, so
5436 * FID is missing in OI cache. we better do not
5437 * lookup FID in FLDB/OI and don't risk to deadlock,
5438 * but in some special cases (lfsck testing, etc)
5439 * it's much simpler than fixing a caller */
5440 CERROR("%s: "DFID" wasn't declared for insert\n",
5441 osd_name(osd), PFID(fid));
5443 idc = osd_idc_find_or_init(env, osd, fid);
5445 RETURN(PTR_ERR(idc));
5448 if (idc->oic_remote) {
5449 /* Insert remote entry */
5450 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5451 struct osd_mdobj_map *omm = osd->od_mdt_map;
5452 struct osd_thandle *oh;
5454 /* If parent on remote MDT, we need put this object
5456 oh = container_of(th, typeof(*oh), ot_super);
5457 rc = osd_add_to_remote_parent(env, osd, obj, oh);
5459 CERROR("%s: add "DFID" error: rc = %d\n",
5461 PFID(lu_object_fid(&dt->do_lu)), rc);
5465 child_inode = igrab(omm->omm_remote_parent->d_inode);
5467 child_inode = osd_create_local_agent_inode(env, osd,
5468 obj, fid, rec1->rec_type & S_IFMT, th);
5469 if (IS_ERR(child_inode))
5470 RETURN(PTR_ERR(child_inode));
5473 /* Insert local entry */
5474 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5475 /* for a reason OI cache wasn't filled properly */
5476 CERROR("%s: OIC for "DFID" isn't filled\n",
5477 osd_name(osd), PFID(fid));
5480 child_inode = oti->oti_inode;
5481 if (unlikely(child_inode == NULL)) {
5482 struct ldiskfs_inode_info *lii;
5486 child_inode = oti->oti_inode = &lii->vfs_inode;
5488 child_inode->i_sb = osd_sb(osd);
5489 child_inode->i_ino = idc->oic_lid.oii_ino;
5490 child_inode->i_mode = rec1->rec_type & S_IFMT;
5493 rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5495 CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5496 obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5498 if (child_inode && child_inode != oti->oti_inode)
5500 LASSERT(osd_invariant(obj));
5501 osd_trans_exec_check(env, th, OSD_OT_INSERT);
5506 * Initialize osd Iterator for given osd index object.
5508 * \param dt osd index object
5511 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5512 struct dt_object *dt,
5515 struct osd_it_iam *it;
5516 struct osd_object *obj = osd_dt_obj(dt);
5517 struct lu_object *lo = &dt->do_lu;
5518 struct iam_path_descr *ipd;
5519 struct iam_container *bag = &obj->oo_dir->od_container;
5521 if (!dt_object_exists(dt))
5522 return ERR_PTR(-ENOENT);
5526 return ERR_PTR(-ENOMEM);
5528 ipd = osd_it_ipd_get(env, bag);
5529 if (likely(ipd != NULL)) {
5533 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5534 return (struct dt_it *)it;
5537 return ERR_PTR(-ENOMEM);
5542 * free given Iterator.
5544 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5546 struct osd_it_iam *it = (struct osd_it_iam *)di;
5547 struct osd_object *obj = it->oi_obj;
5549 iam_it_fini(&it->oi_it);
5550 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5551 osd_object_put(env, obj);
5556 * Move Iterator to record specified by \a key
5558 * \param di osd iterator
5559 * \param key key for index
5561 * \retval +ve di points to record with least key not larger than key
5562 * \retval 0 di points to exact matched key
5563 * \retval -ve failure
5566 static int osd_it_iam_get(const struct lu_env *env,
5567 struct dt_it *di, const struct dt_key *key)
5569 struct osd_thread_info *oti = osd_oti_get(env);
5570 struct osd_it_iam *it = (struct osd_it_iam *)di;
5572 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5573 /* swab quota uid/gid */
5574 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5575 key = (struct dt_key *)&oti->oti_quota_id;
5578 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5584 * \param di osd iterator
5586 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5588 struct osd_it_iam *it = (struct osd_it_iam *)di;
5590 iam_it_put(&it->oi_it);
5594 * Move iterator by one record
5596 * \param di osd iterator
5598 * \retval +1 end of container reached
5600 * \retval -ve failure
5603 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5605 struct osd_it_iam *it = (struct osd_it_iam *)di;
5607 return iam_it_next(&it->oi_it);
5611 * Return pointer to the key under iterator.
5614 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5615 const struct dt_it *di)
5617 struct osd_thread_info *oti = osd_oti_get(env);
5618 struct osd_it_iam *it = (struct osd_it_iam *)di;
5619 struct osd_object *obj = it->oi_obj;
5622 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5624 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5625 /* swab quota uid/gid */
5626 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5627 key = (struct dt_key *)&oti->oti_quota_id;
5634 * Return size of key under iterator (in bytes)
5637 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5639 struct osd_it_iam *it = (struct osd_it_iam *)di;
5641 return iam_it_key_size(&it->oi_it);
5645 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5647 /* check if file type is required */
5648 if (ent->lde_attrs & LUDA_TYPE) {
5649 struct luda_type *lt;
5650 int align = sizeof(*lt) - 1;
5652 len = (len + align) & ~align;
5653 lt = (struct luda_type *)(ent->lde_name + len);
5654 lt->lt_type = cpu_to_le16(DTTOIF(type));
5657 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5661 * build lu direct from backend fs dirent.
5665 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5666 char *name, __u16 namelen, __u16 type, __u32 attr)
5668 ent->lde_attrs = attr | LUDA_FID;
5669 fid_cpu_to_le(&ent->lde_fid, fid);
5671 ent->lde_hash = cpu_to_le64(offset);
5672 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5674 strncpy(ent->lde_name, name, namelen);
5675 ent->lde_name[namelen] = '\0';
5676 ent->lde_namelen = cpu_to_le16(namelen);
5678 /* append lustre attributes */
5679 osd_it_append_attrs(ent, namelen, type);
5683 * Return pointer to the record under iterator.
5685 static int osd_it_iam_rec(const struct lu_env *env,
5686 const struct dt_it *di,
5687 struct dt_rec *dtrec, __u32 attr)
5689 struct osd_it_iam *it = (struct osd_it_iam *)di;
5690 struct osd_thread_info *info = osd_oti_get(env);
5693 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5694 const struct osd_fid_pack *rec;
5695 struct lu_fid *fid = &info->oti_fid;
5696 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
5702 name = (char *)iam_it_key_get(&it->oi_it);
5704 RETURN(PTR_ERR(name));
5706 namelen = iam_it_key_size(&it->oi_it);
5708 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5710 RETURN(PTR_ERR(rec));
5712 rc = osd_fid_unpack(fid, rec);
5716 hash = iam_it_store(&it->oi_it);
5718 /* IAM does not store object type in IAM index (dir) */
5719 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5721 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5722 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5723 (struct iam_rec *)dtrec);
5724 osd_quota_unpack(it->oi_obj, dtrec);
5726 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5727 (struct iam_rec *)dtrec);
5734 * Returns cookie for current Iterator position.
5736 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5738 struct osd_it_iam *it = (struct osd_it_iam *)di;
5740 return iam_it_store(&it->oi_it);
5744 * Restore iterator from cookie.
5746 * \param di osd iterator
5747 * \param hash Iterator location cookie
5749 * \retval +ve di points to record with least key not larger than key.
5750 * \retval 0 di points to exact matched key
5751 * \retval -ve failure
5754 static int osd_it_iam_load(const struct lu_env *env,
5755 const struct dt_it *di, __u64 hash)
5757 struct osd_it_iam *it = (struct osd_it_iam *)di;
5759 return iam_it_load(&it->oi_it, hash);
5762 static const struct dt_index_operations osd_index_iam_ops = {
5763 .dio_lookup = osd_index_iam_lookup,
5764 .dio_declare_insert = osd_index_declare_iam_insert,
5765 .dio_insert = osd_index_iam_insert,
5766 .dio_declare_delete = osd_index_declare_iam_delete,
5767 .dio_delete = osd_index_iam_delete,
5769 .init = osd_it_iam_init,
5770 .fini = osd_it_iam_fini,
5771 .get = osd_it_iam_get,
5772 .put = osd_it_iam_put,
5773 .next = osd_it_iam_next,
5774 .key = osd_it_iam_key,
5775 .key_size = osd_it_iam_key_size,
5776 .rec = osd_it_iam_rec,
5777 .store = osd_it_iam_store,
5778 .load = osd_it_iam_load
5784 * Creates or initializes iterator context.
5786 * \retval struct osd_it_ea, iterator structure on success
5789 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5790 struct dt_object *dt,
5793 struct osd_object *obj = osd_dt_obj(dt);
5794 struct osd_thread_info *info = osd_oti_get(env);
5795 struct osd_it_ea *oie;
5797 struct lu_object *lo = &dt->do_lu;
5798 struct dentry *obj_dentry;
5801 if (!dt_object_exists(dt) || obj->oo_destroyed)
5802 RETURN(ERR_PTR(-ENOENT));
5804 OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5806 RETURN(ERR_PTR(-ENOMEM));
5807 obj_dentry = &oie->oie_dentry;
5809 obj_dentry->d_inode = obj->oo_inode;
5810 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5811 obj_dentry->d_name.hash = 0;
5813 oie->oie_rd_dirent = 0;
5814 oie->oie_it_dirent = 0;
5815 oie->oie_dirent = NULL;
5816 if (unlikely(!info->oti_it_ea_buf_used)) {
5817 oie->oie_buf = info->oti_it_ea_buf;
5818 info->oti_it_ea_buf_used = 1;
5820 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5821 if (oie->oie_buf == NULL)
5822 RETURN(ERR_PTR(-ENOMEM));
5826 file = &oie->oie_file;
5828 /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5829 if (attr & LUDA_64BITHASH)
5830 file->f_mode = FMODE_64BITHASH;
5832 file->f_mode = FMODE_32BITHASH;
5833 file->f_path.dentry = obj_dentry;
5834 file->f_mapping = obj->oo_inode->i_mapping;
5835 file->f_op = obj->oo_inode->i_fop;
5836 set_file_inode(file, obj->oo_inode);
5839 RETURN((struct dt_it *) oie);
5843 * Destroy or finishes iterator context.
5845 * \param di iterator structure to be destroyed
5847 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
5849 struct osd_thread_info *info = osd_oti_get(env);
5850 struct osd_it_ea *oie = (struct osd_it_ea *)di;
5851 struct osd_object *obj = oie->oie_obj;
5852 struct inode *inode = obj->oo_inode;
5855 oie->oie_file.f_op->release(inode, &oie->oie_file);
5856 osd_object_put(env, obj);
5857 if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
5858 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5860 info->oti_it_ea_buf_used = 0;
5861 OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
5866 * It position the iterator at given key, so that next lookup continues from
5867 * that key Or it is similar to dio_it->load() but based on a key,
5868 * rather than file position.
5870 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
5873 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
5875 static int osd_it_ea_get(const struct lu_env *env,
5876 struct dt_it *di, const struct dt_key *key)
5878 struct osd_it_ea *it = (struct osd_it_ea *)di;
5881 LASSERT(((const char *)key)[0] == '\0');
5882 it->oie_file.f_pos = 0;
5883 it->oie_rd_dirent = 0;
5884 it->oie_it_dirent = 0;
5885 it->oie_dirent = NULL;
5893 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
5897 struct osd_filldir_cbs {
5898 #ifdef HAVE_DIR_CONTEXT
5899 struct dir_context ctx;
5901 struct osd_it_ea *it;
5904 * It is called internally by ->readdir(). It fills the
5905 * iterator's in-memory data structure with required
5906 * information i.e. name, namelen, rec_size etc.
5908 * \param buf in which information to be filled in.
5909 * \param name name of the file in given dir
5911 * \retval 0 on success
5912 * \retval 1 on buffer full
5914 #ifdef HAVE_FILLDIR_USE_CTX
5915 static int osd_ldiskfs_filldir(struct dir_context *buf,
5917 static int osd_ldiskfs_filldir(void *buf,
5919 const char *name, int namelen,
5920 loff_t offset, __u64 ino, unsigned d_type)
5922 struct osd_it_ea *it =
5923 ((struct osd_filldir_cbs *)buf)->it;
5924 struct osd_object *obj = it->oie_obj;
5925 struct osd_it_ea_dirent *ent = it->oie_dirent;
5926 struct lu_fid *fid = &ent->oied_fid;
5927 struct osd_fid_pack *rec;
5930 /* this should never happen */
5931 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5932 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5936 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5940 /* "." is just the object itself. */
5941 if (namelen == 1 && name[0] == '.') {
5942 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5943 } else if (d_type & LDISKFS_DIRENT_LUFID) {
5944 rec = (struct osd_fid_pack*) (name + namelen + 1);
5945 if (osd_fid_unpack(fid, rec) != 0)
5950 d_type &= ~LDISKFS_DIRENT_LUFID;
5952 /* NOT export local root. */
5953 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5954 ino = obj->oo_inode->i_ino;
5955 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5958 ent->oied_ino = ino;
5959 ent->oied_off = offset;
5960 ent->oied_namelen = namelen;
5961 ent->oied_type = d_type;
5963 memcpy(ent->oied_name, name, namelen);
5965 it->oie_rd_dirent++;
5966 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5971 * Calls ->readdir() to load a directory entry at a time
5972 * and stored it in iterator's in-memory data structure.
5974 * \param di iterator's in memory structure
5976 * \retval 0 on success
5977 * \retval -ve on error
5978 * \retval +1 reach the end of entry
5980 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5981 const struct dt_it *di)
5983 struct osd_it_ea *it = (struct osd_it_ea *)di;
5984 struct osd_object *obj = it->oie_obj;
5985 struct inode *inode = obj->oo_inode;
5986 struct htree_lock *hlock = NULL;
5987 struct file *filp = &it->oie_file;
5989 struct osd_filldir_cbs buf = {
5990 #ifdef HAVE_DIR_CONTEXT
5991 .ctx.actor = osd_ldiskfs_filldir,
5997 it->oie_dirent = it->oie_buf;
5998 it->oie_rd_dirent = 0;
6000 if (obj->oo_hl_head != NULL) {
6001 hlock = osd_oti_get(env)->oti_hlock;
6002 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6003 inode, LDISKFS_HLOCK_READDIR);
6005 down_read(&obj->oo_ext_idx_sem);
6008 #ifdef HAVE_DIR_CONTEXT
6009 buf.ctx.pos = filp->f_pos;
6010 rc = inode->i_fop->iterate(filp, &buf.ctx);
6011 filp->f_pos = buf.ctx.pos;
6013 rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6017 ldiskfs_htree_unlock(hlock);
6019 up_read(&obj->oo_ext_idx_sem);
6021 if (it->oie_rd_dirent == 0) {
6022 /*If it does not get any dirent, it means it has been reached
6023 *to the end of the dir */
6024 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6028 it->oie_dirent = it->oie_buf;
6029 it->oie_it_dirent = 1;
6036 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6037 * to load a directory entry at a time and stored it in
6038 * iterator's in-memory data structure.
6040 * \param di iterator's in memory structure
6042 * \retval +ve iterator reached to end
6043 * \retval 0 iterator not reached to end
6044 * \retval -ve on error
6046 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6048 struct osd_it_ea *it = (struct osd_it_ea *)di;
6053 if (it->oie_it_dirent < it->oie_rd_dirent) {
6055 (void *) it->oie_dirent +
6056 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6057 it->oie_dirent->oied_namelen);
6058 it->oie_it_dirent++;
6061 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6064 rc = osd_ldiskfs_it_fill(env, di);
6071 * Returns the key at current position from iterator's in memory structure.
6073 * \param di iterator's in memory structure
6075 * \retval key i.e. struct dt_key on success
6077 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6078 const struct dt_it *di)
6080 struct osd_it_ea *it = (struct osd_it_ea *)di;
6082 return (struct dt_key *)it->oie_dirent->oied_name;
6086 * Returns the key's size at current position from iterator's in memory structure.
6088 * \param di iterator's in memory structure
6090 * \retval key_size i.e. struct dt_key on success
6092 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6094 struct osd_it_ea *it = (struct osd_it_ea *)di;
6096 return it->oie_dirent->oied_namelen;
6099 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6101 if (LDISKFS_DIR_REC_LEN(de) >=
6102 __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6109 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6110 unsigned blocksize, bool dotdot)
6113 return osd_dotdot_has_space(de);
6115 if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6116 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6123 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6124 handle_t *jh, struct dentry *dentry,
6125 const struct lu_fid *fid, struct buffer_head *bh,
6126 struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6129 struct inode *dir = dentry->d_parent->d_inode;
6130 struct inode *inode = dentry->d_inode;
6131 struct osd_fid_pack *rec;
6132 struct ldiskfs_dentry_param *ldp;
6133 int namelen = dentry->d_name.len;
6135 struct osd_thread_info *info = osd_oti_get(env);
6138 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6139 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6142 /* There is enough space to hold the FID-in-dirent. */
6143 if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6144 rc = ldiskfs_journal_get_write_access(jh, bh);
6148 de->name[namelen] = 0;
6149 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6150 rec->fp_len = sizeof(struct lu_fid) + 1;
6151 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6152 de->file_type |= LDISKFS_DIRENT_LUFID;
6153 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6160 rc = ldiskfs_delete_entry(jh, dir, de, bh);
6164 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6165 osd_get_ldiskfs_dirent_param(ldp, fid);
6166 dentry->d_fsdata = (void *)ldp;
6167 ll_vfs_dq_init(dir);
6168 rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6169 /* It is too bad, we cannot reinsert the name entry back.
6170 * That means we lose it! */
6172 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6173 "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6174 osd_ino2name(inode),
6175 dir->i_ino, dir->i_generation, namelen,
6176 dentry->d_name.name, PFID(fid), rc);
6182 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6183 struct osd_it_ea *it, struct lu_fid *fid,
6184 struct osd_inode_id *id, __u32 *attr)
6186 struct osd_thread_info *info = osd_oti_get(env);
6187 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6188 struct osd_device *dev = osd_obj2dev(obj);
6189 struct super_block *sb = osd_sb(dev);
6190 const char *devname = osd_name(dev);
6191 struct osd_it_ea_dirent *ent = it->oie_dirent;
6192 struct inode *dir = obj->oo_inode;
6193 struct htree_lock *hlock = NULL;
6194 struct buffer_head *bh = NULL;
6195 handle_t *jh = NULL;
6196 struct ldiskfs_dir_entry_2 *de;
6197 struct dentry *dentry;
6198 struct inode *inode;
6199 const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6202 bool dotdot = false;
6206 if (ent->oied_name[0] == '.') {
6207 if (ent->oied_namelen == 1)
6210 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6214 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6215 inode = osd_iget(info, dev, id);
6216 if (IS_ERR(inode)) {
6217 rc = PTR_ERR(inode);
6218 if (rc == -ENOENT || rc == -ESTALE) {
6219 /* Maybe dangling name entry, or
6220 * corrupted directory entry. */
6221 *attr |= LUDA_UNKNOWN;
6224 CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6225 "check_repair, dir = %lu/%u, name = %.*s, "
6226 "ino = %llu, rc = %d\n",
6227 devname, dir->i_ino, dir->i_generation,
6228 ent->oied_namelen, ent->oied_name,
6235 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6237 rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6238 if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6243 /* We need to ensure that the name entry is still valid.
6244 * Because it may be removed or renamed by other already.
6246 * The unlink or rename operation will start journal before PDO lock,
6247 * so to avoid deadlock, here we need to start journal handle before
6248 * related PDO lock also. But because we do not know whether there
6249 * will be something to be repaired before PDO lock, we just start
6250 * journal without conditions.
6252 * We may need to remove the name entry firstly, then insert back.
6253 * One credit is for user quota file update.
6254 * One credit is for group quota file update.
6255 * Two credits are for dirty inode. */
6256 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6257 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6259 if (dev->od_dirent_journal != 0) {
6262 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6265 CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6266 "check_repair, dir = %lu/%u, credits = %d, "
6267 "name = %.*s, ino = %llu: rc = %d\n",
6268 devname, dir->i_ino, dir->i_generation, credits,
6269 ent->oied_namelen, ent->oied_name,
6272 GOTO(out_inode, rc);
6275 if (obj->oo_hl_head != NULL) {
6276 hlock = osd_oti_get(env)->oti_hlock;
6277 /* "0" means exclusive lock for the whole directory.
6278 * We need to prevent others access such name entry
6279 * during the delete + insert. Neither HLOCK_ADD nor
6280 * HLOCK_DEL cannot guarantee the atomicity. */
6281 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6283 down_write(&obj->oo_ext_idx_sem);
6286 if (obj->oo_hl_head != NULL) {
6287 hlock = osd_oti_get(env)->oti_hlock;
6288 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6289 LDISKFS_HLOCK_LOOKUP);
6291 down_read(&obj->oo_ext_idx_sem);
6295 bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6296 if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6297 *attr |= LUDA_IGNORE;
6302 /* For dotdot entry, if there is not enough space to hold the
6303 * FID-in-dirent, just keep them there. It only happens when the
6304 * device upgraded from 1.8 or restored from MDT file-level backup.
6305 * For the whole directory, only dotdot entry have no FID-in-dirent
6306 * and needs to get FID from LMA when readdir, it will not affect the
6307 * performance much. */
6308 if (dotdot && !osd_dotdot_has_space(de)) {
6309 *attr |= LUDA_UNKNOWN;
6315 if (lu_fid_eq(fid, &lma->lma_self_fid))
6318 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6319 struct lu_fid *tfid = &lma->lma_self_fid;
6321 if (likely(dotdot &&
6322 fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6323 fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6324 /* It must be REMOTE_PARENT_DIR and as the
6325 * 'dotdot' entry of remote directory */
6326 *attr |= LUDA_IGNORE;
6328 CDEBUG(D_LFSCK, "%s: expect remote agent "
6329 "parent directory, but got %.*s under "
6330 "dir = %lu/%u with the FID "DFID"\n",
6331 devname, ent->oied_namelen,
6332 ent->oied_name, dir->i_ino,
6333 dir->i_generation, PFID(tfid));
6335 *attr |= LUDA_UNKNOWN;
6342 if (!fid_is_zero(fid)) {
6343 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6345 if (rc == -ENOENT ||
6347 !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) {
6348 /* linkEA does not recognize the dirent entry,
6349 * it may because the dirent entry corruption
6350 * and points to other's inode. */
6351 CDEBUG(D_LFSCK, "%s: the target inode does not "
6352 "recognize the dirent, dir = %lu/%u, "
6353 " name = %.*s, ino = %llu, "
6354 DFID": rc = %d\n", devname, dir->i_ino,
6355 dir->i_generation, ent->oied_namelen,
6356 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6357 *attr |= LUDA_UNKNOWN;
6362 if (rc && rc != -ENODATA) {
6363 CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6364 "dir = %lu/%u, name = %.*s, ino = %llu, "
6365 DFID": rc = %d\n", devname, dir->i_ino,
6366 dir->i_generation, ent->oied_namelen,
6367 ent->oied_name, ent->oied_ino, PFID(fid), rc);
6368 *attr |= LUDA_UNKNOWN;
6375 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6376 * valid, trusted, in spite of fid_is_sane(fid) or not. */
6377 if (*attr & LUDA_VERIFY_DRYRUN) {
6378 *fid = lma->lma_self_fid;
6379 *attr |= LUDA_REPAIR;
6386 dev->od_dirent_journal = 1;
6387 if (hlock != NULL) {
6388 ldiskfs_htree_unlock(hlock);
6391 up_read(&obj->oo_ext_idx_sem);
6397 *fid = lma->lma_self_fid;
6399 /* Update or append the FID-in-dirent. */
6400 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6401 bh, de, hlock, dotdot);
6403 *attr |= LUDA_REPAIR;
6405 CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6406 "the dirent, dir = %lu/%u, name = %.*s, "
6407 "ino = %llu, "DFID": rc = %d\n",
6408 devname, dir->i_ino, dir->i_generation,
6409 ent->oied_namelen, ent->oied_name,
6410 ent->oied_ino, PFID(fid), rc);
6412 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6413 if (*attr & LUDA_VERIFY_DRYRUN) {
6414 if (fid_is_sane(fid)) {
6415 *attr |= LUDA_REPAIR;
6416 } else if (dev->od_index == 0) {
6417 lu_igif_build(fid, inode->i_ino,
6418 inode->i_generation);
6419 *attr |= LUDA_UPGRADE;
6427 dev->od_dirent_journal = 1;
6428 if (hlock != NULL) {
6429 ldiskfs_htree_unlock(hlock);
6432 up_read(&obj->oo_ext_idx_sem);
6439 if (unlikely(fid_is_sane(fid))) {
6440 /* FID-in-dirent exists, but FID-in-LMA is lost.
6441 * Trust the FID-in-dirent, and add FID-in-LMA. */
6442 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6444 *attr |= LUDA_REPAIR;
6446 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6447 "update dirent, dir = %lu/%u, "
6448 "name = %.*s, ino = %llu, "
6450 devname, dir->i_ino, dir->i_generation,
6451 ent->oied_namelen, ent->oied_name,
6452 ent->oied_ino, PFID(fid), rc);
6453 } else if (dev->od_index == 0) {
6454 lu_igif_build(fid, inode->i_ino, inode->i_generation);
6455 /* It is probably IGIF object. Only aappend the
6456 * FID-in-dirent. OI scrub will process FID-in-LMA. */
6457 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6458 bh, de, hlock, dotdot);
6460 *attr |= LUDA_UPGRADE;
6462 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6463 "after the dirent, dir = %lu/%u, "
6464 "name = %.*s, ino = %llu, "
6466 devname, dir->i_ino, dir->i_generation,
6467 ent->oied_namelen, ent->oied_name,
6468 ent->oied_ino, PFID(fid), rc);
6477 if (hlock != NULL) {
6478 ldiskfs_htree_unlock(hlock);
6480 if (dev->od_dirent_journal != 0)
6481 up_write(&obj->oo_ext_idx_sem);
6483 up_read(&obj->oo_ext_idx_sem);
6487 ldiskfs_journal_stop(jh);
6491 if (rc >= 0 && !dirty)
6492 dev->od_dirent_journal = 0;
6498 * Returns the value at current position from iterator's in memory structure.
6500 * \param di struct osd_it_ea, iterator's in memory structure
6501 * \param attr attr requested for dirent.
6502 * \param lde lustre dirent
6504 * \retval 0 no error and \param lde has correct lustre dirent.
6505 * \retval -ve on error
6507 static inline int osd_it_ea_rec(const struct lu_env *env,
6508 const struct dt_it *di,
6509 struct dt_rec *dtrec, __u32 attr)
6511 struct osd_it_ea *it = (struct osd_it_ea *)di;
6512 struct osd_object *obj = it->oie_obj;
6513 struct osd_device *dev = osd_obj2dev(obj);
6514 struct osd_thread_info *oti = osd_oti_get(env);
6515 struct osd_inode_id *id = &oti->oti_id;
6516 struct lu_fid *fid = &it->oie_dirent->oied_fid;
6517 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6518 __u32 ino = it->oie_dirent->oied_ino;
6522 LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6524 if (attr & LUDA_VERIFY) {
6525 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6526 attr |= LUDA_IGNORE;
6527 /* If the parent is on remote MDT, and there
6528 * is no FID-in-dirent, then we have to get
6529 * the parent FID from the linkEA. */
6530 if (!fid_is_sane(fid) &&
6531 it->oie_dirent->oied_namelen == 2 &&
6532 it->oie_dirent->oied_name[0] == '.' &&
6533 it->oie_dirent->oied_name[1] == '.')
6534 osd_get_pfid_from_linkea(env, obj, fid);
6536 rc = osd_dirent_check_repair(env, obj, it, fid, id,
6540 if (!fid_is_sane(fid))
6541 attr |= LUDA_UNKNOWN;
6543 attr &= ~LU_DIRENT_ATTRS_MASK;
6544 if (!fid_is_sane(fid)) {
6545 bool is_dotdot = false;
6546 if (it->oie_dirent->oied_namelen == 2 &&
6547 it->oie_dirent->oied_name[0] == '.' &&
6548 it->oie_dirent->oied_name[1] == '.')
6550 /* If the parent is on remote MDT, and there
6551 * is no FID-in-dirent, then we have to get
6552 * the parent FID from the linkEA. */
6553 if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6554 rc = osd_get_pfid_from_linkea(env, obj, fid);
6556 if (is_dotdot == false &&
6557 OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6560 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6563 osd_id_gen(id, ino, OSD_OII_NOGEN);
6567 /* Pack the entry anyway, at least the offset is right. */
6568 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6569 it->oie_dirent->oied_name,
6570 it->oie_dirent->oied_namelen,
6571 it->oie_dirent->oied_type, attr);
6576 if (osd_remote_fid(env, dev, fid))
6579 if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6580 osd_add_oi_cache(oti, dev, id, fid);
6582 RETURN(rc > 0 ? 0 : rc);
6586 * Returns the record size size at current position.
6588 * This function will return record(lu_dirent) size in bytes.
6590 * \param[in] env execution environment
6591 * \param[in] di iterator's in memory structure
6592 * \param[in] attr attribute of the entry, only requires LUDA_TYPE to
6593 * calculate the lu_dirent size.
6595 * \retval record size(in bytes & in memory) of the current lu_dirent
6598 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6601 struct osd_it_ea *it = (struct osd_it_ea *)di;
6603 return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6607 * Returns a cookie for current position of the iterator head, so that
6608 * user can use this cookie to load/start the iterator next time.
6610 * \param di iterator's in memory structure
6612 * \retval cookie for current position, on success
6614 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6616 struct osd_it_ea *it = (struct osd_it_ea *)di;
6618 return it->oie_dirent->oied_off;
6622 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6623 * to load a directory entry at a time and stored it i inn,
6624 * in iterator's in-memory data structure.
6626 * \param di struct osd_it_ea, iterator's in memory structure
6628 * \retval +ve on success
6629 * \retval -ve on error
6631 static int osd_it_ea_load(const struct lu_env *env,
6632 const struct dt_it *di, __u64 hash)
6634 struct osd_it_ea *it = (struct osd_it_ea *)di;
6638 it->oie_file.f_pos = hash;
6640 rc = osd_ldiskfs_it_fill(env, di);
6651 * Index lookup function for interoperability mode (b11826).
6653 * \param key, key i.e. file name to be searched
6655 * \retval +ve, on success
6656 * \retval -ve, on error
6658 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6659 struct dt_rec *rec, const struct dt_key *key)
6661 struct osd_object *obj = osd_dt_obj(dt);
6666 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6667 LINVRNT(osd_invariant(obj));
6669 rc = osd_ea_lookup_rec(env, obj, rec, key);
6676 * Index and Iterator operations for interoperability
6677 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6679 static const struct dt_index_operations osd_index_ea_ops = {
6680 .dio_lookup = osd_index_ea_lookup,
6681 .dio_declare_insert = osd_index_declare_ea_insert,
6682 .dio_insert = osd_index_ea_insert,
6683 .dio_declare_delete = osd_index_declare_ea_delete,
6684 .dio_delete = osd_index_ea_delete,
6686 .init = osd_it_ea_init,
6687 .fini = osd_it_ea_fini,
6688 .get = osd_it_ea_get,
6689 .put = osd_it_ea_put,
6690 .next = osd_it_ea_next,
6691 .key = osd_it_ea_key,
6692 .key_size = osd_it_ea_key_size,
6693 .rec = osd_it_ea_rec,
6694 .rec_size = osd_it_ea_rec_size,
6695 .store = osd_it_ea_store,
6696 .load = osd_it_ea_load
6700 static void *osd_key_init(const struct lu_context *ctx,
6701 struct lu_context_key *key)
6703 struct osd_thread_info *info;
6705 OBD_ALLOC_PTR(info);
6707 return ERR_PTR(-ENOMEM);
6709 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6710 if (info->oti_it_ea_buf == NULL)
6713 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6715 info->oti_hlock = ldiskfs_htree_lock_alloc();
6716 if (info->oti_hlock == NULL)
6722 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6725 return ERR_PTR(-ENOMEM);
6728 static void osd_key_fini(const struct lu_context *ctx,
6729 struct lu_context_key *key, void* data)
6731 struct osd_thread_info *info = data;
6732 struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6733 struct osd_idmap_cache *idc = info->oti_ins_cache;
6735 if (info->oti_inode != NULL)
6737 if (info->oti_hlock != NULL)
6738 ldiskfs_htree_lock_free(info->oti_hlock);
6739 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6740 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6741 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6742 lu_buf_free(&info->oti_big_buf);
6744 LASSERT(info->oti_ins_cache_size > 0);
6745 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6746 info->oti_ins_cache = NULL;
6747 info->oti_ins_cache_size = 0;
6752 static void osd_key_exit(const struct lu_context *ctx,
6753 struct lu_context_key *key, void *data)
6755 struct osd_thread_info *info = data;
6757 LASSERT(info->oti_r_locks == 0);
6758 LASSERT(info->oti_w_locks == 0);
6759 LASSERT(info->oti_txns == 0);
6762 /* type constructor/destructor: osd_type_init, osd_type_fini */
6763 LU_TYPE_INIT_FINI(osd, &osd_key);
6765 struct lu_context_key osd_key = {
6766 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6767 .lct_init = osd_key_init,
6768 .lct_fini = osd_key_fini,
6769 .lct_exit = osd_key_exit
6773 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6774 const char *name, struct lu_device *next)
6776 struct osd_device *osd = osd_dev(d);
6778 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6779 >= sizeof(osd->od_svname))
6781 return osd_procfs_init(osd, name);
6784 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6786 struct seq_server_site *ss = osd_seq_site(osd);
6790 if (osd->od_is_ost || osd->od_cl_seq != NULL)
6793 if (unlikely(ss == NULL))
6796 OBD_ALLOC_PTR(osd->od_cl_seq);
6797 if (osd->od_cl_seq == NULL)
6800 rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6801 osd->od_svname, ss->ss_server_seq);
6803 OBD_FREE_PTR(osd->od_cl_seq);
6804 osd->od_cl_seq = NULL;
6808 if (ss->ss_node_id == 0) {
6809 /* If the OSD on the sequence controller(MDT0), then allocate
6810 * sequence here, otherwise allocate sequence after connected
6811 * to MDT0 (see mdt_register_lwp_callback()). */
6812 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6813 &osd->od_cl_seq->lcs_space, env);
6819 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6821 if (osd->od_cl_seq == NULL)
6824 seq_client_fini(osd->od_cl_seq);
6825 OBD_FREE_PTR(osd->od_cl_seq);
6826 osd->od_cl_seq = NULL;
6829 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6833 /* shutdown quota slave instance associated with the device */
6834 if (o->od_quota_slave != NULL) {
6835 qsd_fini(env, o->od_quota_slave);
6836 o->od_quota_slave = NULL;
6839 osd_fid_fini(env, o);
6844 static void osd_umount(const struct lu_env *env, struct osd_device *o)
6848 if (o->od_mnt != NULL) {
6849 shrink_dcache_sb(osd_sb(o));
6850 osd_sync(env, &o->od_dt_dev);
6859 static int osd_mount(const struct lu_env *env,
6860 struct osd_device *o, struct lustre_cfg *cfg)
6862 const char *name = lustre_cfg_string(cfg, 0);
6863 const char *dev = lustre_cfg_string(cfg, 1);
6865 unsigned long page, s_flags, lmd_flags = 0;
6866 struct page *__page;
6867 struct file_system_type *type;
6868 char *options = NULL;
6870 struct osd_thread_info *info = osd_oti_get(env);
6871 struct lu_fid *fid = &info->oti_fid;
6872 struct inode *inode;
6873 int rc = 0, force_over_512tb = 0;
6876 if (o->od_mnt != NULL)
6879 if (strlen(dev) >= sizeof(o->od_mntdev))
6881 strcpy(o->od_mntdev, dev);
6883 str = lustre_cfg_string(cfg, 2);
6884 s_flags = simple_strtoul(str, NULL, 0);
6885 str = strstr(str, ":");
6887 lmd_flags = simple_strtoul(str + 1, NULL, 0);
6888 opts = lustre_cfg_string(cfg, 3);
6890 if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
6891 CERROR("%s: device %s extents feature is not guaranteed to "
6892 "work on big-endian systems. Use \"bigendian_extents\" "
6893 "mount option to override.\n", name, dev);
6897 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
6898 if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
6899 CWARN("force_over_128tb option is deprecated. "
6900 "Filesystems less than 512TB can be created without any "
6901 "force options. Use force_over_512tb option for "
6902 "filesystems greater than 512TB.\n");
6905 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
6906 if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
6907 CWARN("force_over_256tb option is deprecated. "
6908 "Filesystems less than 512TB can be created without any "
6909 "force options. Use force_over_512tb option for "
6910 "filesystems greater than 512TB.\n");
6914 if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
6915 force_over_512tb = 1;
6917 __page = alloc_page(GFP_KERNEL);
6919 GOTO(out, rc = -ENOMEM);
6920 page = (unsigned long)page_address(__page);
6921 options = (char *)page;
6924 /* strip out the options for back compatiblity */
6925 static char *sout[] = {
6932 /* strip out option we processed in osd */
6933 "bigendian_extents",
6939 strcat(options, opts);
6940 for (rc = 0, str = options; sout[rc]; ) {
6941 char *op = strstr(str, sout[rc]);
6947 if (op == options || *(op - 1) == ',') {
6948 str = op + strlen(sout[rc]);
6949 if (*str == ',' || *str == '\0') {
6950 *str == ',' ? str++ : str;
6951 memmove(op, str, strlen(str) + 1);
6954 for (str = op; *str != ',' && *str != '\0'; str++)
6958 strncat(options, "user_xattr,acl", 14);
6961 /* Glom up mount options */
6962 if (*options != '\0')
6963 strcat(options, ",");
6964 strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
6966 type = get_fs_type("ldiskfs");
6968 CERROR("%s: cannot find ldiskfs module\n", name);
6969 GOTO(out, rc = -ENODEV);
6972 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6973 module_put(type->owner);
6975 if (IS_ERR(o->od_mnt)) {
6976 rc = PTR_ERR(o->od_mnt);
6978 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6982 if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
6983 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
6984 force_over_512tb == 0) {
6985 CERROR("%s: device %s LDISKFS does not support filesystems "
6986 "greater than 512TB and can cause data corruption. "
6987 "Use \"force_over_512tb\" mount option to override.\n",
6989 GOTO(out_mnt, rc = -EINVAL);
6992 if (lmd_flags & LMD_FLG_DEV_RDONLY) {
6993 if (priv_dev_set_rdonly) {
6994 priv_dev_set_rdonly(osd_sb(o)->s_bdev);
6995 o->od_dt_dev.dd_rdonly = 1;
6996 LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
6999 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7002 GOTO(out_mnt, rc = -EOPNOTSUPP);
7004 } else if (priv_dev_check_rdonly &&
7005 priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7006 CERROR("%s: underlying device %s is marked as "
7007 "read-only. Setup failed\n", name, dev);
7009 GOTO(out_mnt, rc = -EROFS);
7012 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7013 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7014 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7015 GOTO(out_mnt, rc = -EINVAL);
7018 #ifdef LDISKFS_MOUNT_DIRDATA
7019 if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7020 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7021 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7022 else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7023 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7024 "enabling the dirdata feature. If you do not want to "
7025 "downgrade to Lustre-1.x again, you can enable it via "
7026 "'tune2fs -O dirdata device'\n", name, dev);
7028 inode = osd_sb(o)->s_root->d_inode;
7029 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7030 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7032 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7036 if (lmd_flags & LMD_FLG_NOSCRUB)
7047 __free_page(__page);
7052 static struct lu_device *osd_device_fini(const struct lu_env *env,
7053 struct lu_device *d)
7055 struct osd_device *o = osd_dev(d);
7058 osd_shutdown(env, o);
7060 osd_scrub_cleanup(env, o);
7061 osd_obj_map_fini(o);
7067 static int osd_device_init0(const struct lu_env *env,
7068 struct osd_device *o,
7069 struct lustre_cfg *cfg)
7071 struct lu_device *l = osd2lu_dev(o);
7072 struct osd_thread_info *info;
7076 /* if the module was re-loaded, env can loose its keys */
7077 rc = lu_env_refill((struct lu_env *) env);
7080 info = osd_oti_get(env);
7083 l->ld_ops = &osd_lu_ops;
7084 o->od_dt_dev.dd_ops = &osd_dt_ops;
7086 spin_lock_init(&o->od_osfs_lock);
7087 mutex_init(&o->od_otable_mutex);
7088 INIT_LIST_HEAD(&o->od_orphan_list);
7090 o->od_read_cache = 1;
7091 o->od_writethrough_cache = 1;
7092 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7094 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7095 sizeof(o->od_svname));
7096 if (cplen >= sizeof(o->od_svname)) {
7101 o->od_index = -1; /* -1 means index is invalid */
7102 rc = server_name2index(o->od_svname, &o->od_index, NULL);
7103 if (rc == LDD_F_SV_TYPE_OST)
7106 o->od_full_scrub_ratio = OFSR_DEFAULT;
7107 o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7108 rc = osd_mount(env, o, cfg);
7112 rc = osd_obj_map_init(env, o);
7116 rc = lu_site_init(&o->od_site, l);
7118 GOTO(out_compat, rc);
7119 o->od_site.ls_bottom_dev = l;
7121 rc = lu_site_init_finish(&o->od_site);
7125 INIT_LIST_HEAD(&o->od_ios_list);
7126 /* setup scrub, including OI files initialization */
7127 rc = osd_scrub_setup(env, o);
7131 rc = osd_procfs_init(o, o->od_svname);
7133 CERROR("%s: can't initialize procfs: rc = %d\n",
7135 GOTO(out_scrub, rc);
7138 LASSERT(l->ld_site->ls_linkage.next != NULL);
7139 LASSERT(l->ld_site->ls_linkage.prev != NULL);
7141 /* initialize quota slave instance */
7142 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7144 if (IS_ERR(o->od_quota_slave)) {
7145 rc = PTR_ERR(o->od_quota_slave);
7146 o->od_quota_slave = NULL;
7147 GOTO(out_procfs, rc);
7155 osd_scrub_cleanup(env, o);
7157 lu_site_fini(&o->od_site);
7159 osd_obj_map_fini(o);
7166 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7167 struct lu_device_type *t,
7168 struct lustre_cfg *cfg)
7170 struct osd_device *o;
7175 return ERR_PTR(-ENOMEM);
7177 rc = dt_device_init(&o->od_dt_dev, t);
7179 /* Because the ctx might be revived in dt_device_init,
7180 * refill the env here */
7181 lu_env_refill((struct lu_env *)env);
7182 rc = osd_device_init0(env, o, cfg);
7184 dt_device_fini(&o->od_dt_dev);
7187 if (unlikely(rc != 0))
7190 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7193 static struct lu_device *osd_device_free(const struct lu_env *env,
7194 struct lu_device *d)
7196 struct osd_device *o = osd_dev(d);
7199 /* XXX: make osd top device in order to release reference */
7200 d->ld_site->ls_top_dev = d;
7201 lu_site_purge(env, d->ld_site, -1);
7202 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7203 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7204 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7206 lu_site_fini(&o->od_site);
7207 dt_device_fini(&o->od_dt_dev);
7212 static int osd_process_config(const struct lu_env *env,
7213 struct lu_device *d, struct lustre_cfg *cfg)
7215 struct osd_device *o = osd_dev(d);
7219 switch (cfg->lcfg_command) {
7221 rc = osd_mount(env, o, cfg);
7224 lu_dev_del_linkage(d->ld_site, d);
7225 rc = osd_shutdown(env, o);
7228 LASSERT(&o->od_dt_dev);
7229 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7230 cfg, &o->od_dt_dev);
7231 if (rc > 0 || rc == -ENOSYS) {
7232 rc = class_process_proc_param(PARAM_OST,
7233 lprocfs_osd_obd_vars,
7234 cfg, &o->od_dt_dev);
7246 static int osd_recovery_complete(const struct lu_env *env,
7247 struct lu_device *d)
7249 struct osd_device *osd = osd_dev(d);
7253 if (osd->od_quota_slave == NULL)
7256 /* start qsd instance on recovery completion, this notifies the quota
7257 * slave code that we are about to process new requests now */
7258 rc = qsd_start(env, osd->od_quota_slave);
7263 * we use exports to track all osd users
7265 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7266 struct obd_device *obd, struct obd_uuid *cluuid,
7267 struct obd_connect_data *data, void *localdata)
7269 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7270 struct lustre_handle conn;
7274 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7276 rc = class_connect(&conn, obd, cluuid);
7280 *exp = class_conn2export(&conn);
7282 spin_lock(&osd->od_osfs_lock);
7284 spin_unlock(&osd->od_osfs_lock);
7290 * once last export (we don't count self-export) disappeared
7291 * osd can be released
7293 static int osd_obd_disconnect(struct obd_export *exp)
7295 struct obd_device *obd = exp->exp_obd;
7296 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7297 int rc, release = 0;
7300 /* Only disconnect the underlying layers on the final disconnect. */
7301 spin_lock(&osd->od_osfs_lock);
7303 if (osd->od_connects == 0)
7305 spin_unlock(&osd->od_osfs_lock);
7307 rc = class_disconnect(exp); /* bz 9811 */
7309 if (rc == 0 && release)
7310 class_manual_cleanup(obd);
7314 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7315 struct lu_device *dev)
7317 struct osd_device *osd = osd_dev(dev);
7318 struct lr_server_data *lsd =
7319 &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7323 if (osd->od_quota_slave != NULL) {
7324 /* set up quota slave objects */
7325 result = qsd_prepare(env, osd->od_quota_slave);
7330 if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7331 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7332 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7333 osd->od_index_in_idif = 1;
7335 osd->od_index_in_idif = 0;
7336 result = osd_register_proc_index_in_idif(osd);
7341 osd->od_index_in_idif = 1;
7345 result = osd_fid_init(env, osd);
7350 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7351 struct lu_fid *fid, struct md_op_data *op_data)
7353 struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7355 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7358 static const struct lu_object_operations osd_lu_obj_ops = {
7359 .loo_object_init = osd_object_init,
7360 .loo_object_delete = osd_object_delete,
7361 .loo_object_release = osd_object_release,
7362 .loo_object_free = osd_object_free,
7363 .loo_object_print = osd_object_print,
7364 .loo_object_invariant = osd_object_invariant
7367 const struct lu_device_operations osd_lu_ops = {
7368 .ldo_object_alloc = osd_object_alloc,
7369 .ldo_process_config = osd_process_config,
7370 .ldo_recovery_complete = osd_recovery_complete,
7371 .ldo_prepare = osd_prepare,
7374 static const struct lu_device_type_operations osd_device_type_ops = {
7375 .ldto_init = osd_type_init,
7376 .ldto_fini = osd_type_fini,
7378 .ldto_start = osd_type_start,
7379 .ldto_stop = osd_type_stop,
7381 .ldto_device_alloc = osd_device_alloc,
7382 .ldto_device_free = osd_device_free,
7384 .ldto_device_init = osd_device_init,
7385 .ldto_device_fini = osd_device_fini
7388 static struct lu_device_type osd_device_type = {
7389 .ldt_tags = LU_DEVICE_DT,
7390 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
7391 .ldt_ops = &osd_device_type_ops,
7392 .ldt_ctx_tags = LCT_LOCAL,
7395 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7397 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7398 struct super_block *sb = osd_sb(osd);
7400 return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7404 * lprocfs legacy support.
7406 static struct obd_ops osd_obd_device_ops = {
7407 .o_owner = THIS_MODULE,
7408 .o_connect = osd_obd_connect,
7409 .o_disconnect = osd_obd_disconnect,
7410 .o_fid_alloc = osd_fid_alloc,
7411 .o_health_check = osd_health_check,
7414 static int __init osd_init(void)
7418 CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7419 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7420 /* please, try to keep osd_thread_info smaller than a page */
7421 CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7426 rc = lu_kmem_init(ldiskfs_caches);
7430 #ifdef CONFIG_KALLSYMS
7431 priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7432 priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7433 /* Clear readonly is unused at this time */
7434 /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7437 rc = class_register_type(&osd_obd_device_ops, NULL, true,
7438 lprocfs_osd_module_vars,
7439 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7441 lu_kmem_fini(ldiskfs_caches);
7445 static void __exit osd_exit(void)
7447 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7448 lu_kmem_fini(ldiskfs_caches);
7451 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7452 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7453 MODULE_VERSION(LUSTRE_VERSION_STRING);
7454 MODULE_LICENSE("GPL");
7456 module_init(osd_init);
7457 module_exit(osd_exit);