4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <linux/module.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
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 "osd_internal.h"
70 /* llo_* api support */
71 #include <md_object.h>
72 #include <lustre_quota.h>
75 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
76 "ldiskfs with parallel directory operations");
78 static const char dot[] = ".";
79 static const char dotdot[] = "..";
80 static const char remote_obj_dir[] = "REM_OBJ_DIR";
82 static const struct lu_object_operations osd_lu_obj_ops;
83 static const struct dt_object_operations osd_obj_ops;
84 static const struct dt_object_operations osd_obj_ea_ops;
85 static const struct dt_object_operations osd_obj_otable_it_ops;
86 static const struct dt_index_operations osd_index_iam_ops;
87 static const struct dt_index_operations osd_index_ea_ops;
89 #ifdef OSD_TRACK_DECLARES
90 int osd_trans_declare_op2rb[] = {
91 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
92 [OSD_OT_PUNCH] = OSD_OT_MAX,
93 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
94 [OSD_OT_CREATE] = OSD_OT_DESTROY,
95 [OSD_OT_DESTROY] = OSD_OT_CREATE,
96 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
97 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
98 [OSD_OT_WRITE] = OSD_OT_WRITE,
99 [OSD_OT_INSERT] = OSD_OT_DELETE,
100 [OSD_OT_DELETE] = OSD_OT_INSERT,
101 [OSD_OT_QUOTA] = OSD_OT_MAX,
105 static int osd_has_index(const struct osd_object *obj)
107 return obj->oo_dt.do_index_ops != NULL;
110 static int osd_object_invariant(const struct lu_object *l)
112 return osd_invariant(osd_obj(l));
116 * Concurrency: doesn't matter
118 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
120 return osd_oti_get(env)->oti_r_locks > 0;
124 * Concurrency: doesn't matter
126 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
128 struct osd_thread_info *oti = osd_oti_get(env);
129 return oti->oti_w_locks > 0 && o->oo_owner == env;
133 * Concurrency: doesn't access mutable data
135 static int osd_root_get(const struct lu_env *env,
136 struct dt_device *dev, struct lu_fid *f)
138 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
143 * OSD object methods.
147 * Concurrency: no concurrent access is possible that early in object
150 static struct lu_object *osd_object_alloc(const struct lu_env *env,
151 const struct lu_object_header *hdr,
154 struct osd_object *mo;
160 l = &mo->oo_dt.do_lu;
161 dt_object_init(&mo->oo_dt, NULL, d);
162 mo->oo_dt.do_ops = &osd_obj_ea_ops;
163 l->lo_ops = &osd_lu_obj_ops;
164 init_rwsem(&mo->oo_sem);
165 init_rwsem(&mo->oo_ext_idx_sem);
166 spin_lock_init(&mo->oo_guard);
173 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
174 const char *name, void *buf, int len)
176 dentry->d_inode = inode;
177 return inode->i_op->getxattr(dentry, name, buf, len);
180 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
181 struct dentry *dentry, struct lustre_mdt_attrs *lma)
185 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma,
188 /* try with old lma size */
189 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
190 info->oti_mdt_attrs_old,
193 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
196 /* Check LMA compatibility */
197 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
198 CWARN("%.16s: unsupported incompat LMA feature(s) "
200 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
201 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
205 lustre_lma_swab(lma);
208 } else if (rc == 0) {
216 * retrieve object from backend ext fs.
218 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
219 struct osd_inode_id *id)
221 struct inode *inode = NULL;
223 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
225 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
226 id->oii_ino, PTR_ERR(inode));
227 } else if (id->oii_gen != OSD_OII_NOGEN &&
228 inode->i_generation != id->oii_gen) {
229 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
231 id->oii_ino, id->oii_gen, inode->i_generation);
233 inode = ERR_PTR(-ESTALE);
234 } else if (inode->i_nlink == 0) {
235 /* due to parallel readdir and unlink,
236 * we can have dead inode here. */
237 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
238 make_bad_inode(inode);
240 inode = ERR_PTR(-ESTALE);
241 } else if (is_bad_inode(inode)) {
242 CWARN("%.16s: bad inode: ino = %u\n",
243 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
245 inode = ERR_PTR(-ENOENT);
247 if (id->oii_gen == OSD_OII_NOGEN)
248 osd_id_gen(id, inode->i_ino, inode->i_generation);
250 /* Do not update file c/mtime in ldiskfs.
251 * NB: we don't have any lock to protect this because we don't
252 * have reference on osd_object now, but contention with
253 * another lookup + attr_set can't happen in the tiny window
254 * between if (...) and set S_NOCMTIME. */
255 if (!(inode->i_flags & S_NOCMTIME))
256 inode->i_flags |= S_NOCMTIME;
261 static struct inode *
262 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
263 struct osd_inode_id *id, struct lu_fid *fid)
265 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
269 inode = osd_iget(info, dev, id);
273 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
275 *fid = lma->lma_self_fid;
276 } else if (rc == -ENODATA) {
277 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
278 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
280 lu_igif_build(fid, inode->i_ino, inode->i_generation);
288 static struct inode *
289 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
290 struct osd_inode_id *id, const struct lu_fid *fid)
292 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
296 inode = osd_iget(info, dev, id);
300 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
309 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
310 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
311 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
313 return ERR_PTR(-EREMCHG);
319 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
320 const struct lu_fid *fid,
321 const struct lu_object_conf *conf)
323 struct osd_thread_info *info;
324 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
325 struct osd_device *dev;
326 struct osd_idmap_cache *oic;
327 struct osd_inode_id *id;
329 struct osd_scrub *scrub;
330 struct scrub_file *sf;
335 LINVRNT(osd_invariant(obj));
336 LASSERT(obj->oo_inode == NULL);
337 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
340 scrub = &dev->od_scrub;
341 sf = &scrub->os_file;
342 info = osd_oti_get(env);
344 oic = &info->oti_cache;
346 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
349 /* Search order: 1. per-thread cache. */
350 if (lu_fid_eq(fid, &oic->oic_fid)) {
356 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
357 /* Search order: 2. OI scrub pending list. */
358 result = osd_oii_lookup(dev, fid, id);
363 if (sf->sf_flags & SF_INCONSISTENT)
367 * Objects are created as locking anchors or place holders for objects
368 * yet to be created. No need to osd_oi_lookup() at here because FID
369 * shouldn't never be re-used, if it's really a duplicate FID from
370 * unexpected reason, we should be able to detect it later by calling
371 * do_create->osd_oi_insert()
373 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
374 GOTO(out, result = 0);
376 /* Search order: 3. OI files. */
377 result = osd_oi_lookup(info, dev, fid, id, true);
378 if (result == -ENOENT) {
379 if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
380 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
382 GOTO(out, result = 0);
392 inode = osd_iget(info, dev, id);
394 inode = osd_iget_verify(info, dev, id, fid);
396 result = PTR_ERR(inode);
397 if (result == -ENOENT || result == -ESTALE) {
398 fid_zero(&oic->oic_fid);
400 } else if (result == -EREMCHG) {
403 if (thread_is_running(&scrub->os_thread)) {
404 result = -EINPROGRESS;
405 } else if (!dev->od_noscrub) {
406 result = osd_scrub_start(dev);
407 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
408 "for "DFID", rc = %d [1]\n",
409 LDISKFS_SB(osd_sb(dev))->s_es->\
410 s_volume_name,PFID(fid), result);
411 if (result == 0 || result == -EALREADY)
412 result = -EINPROGRESS;
421 obj->oo_inode = inode;
422 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
424 obj->oo_compat_dot_created = 1;
425 obj->oo_compat_dotdot_created = 1;
427 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
428 GOTO(out, result = 0);
430 LASSERT(obj->oo_hl_head == NULL);
431 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
432 if (obj->oo_hl_head == NULL) {
433 obj->oo_inode = NULL;
435 GOTO(out, result = -ENOMEM);
437 GOTO(out, result = 0);
440 LINVRNT(osd_invariant(obj));
445 * Concurrency: shouldn't matter.
447 static void osd_object_init0(struct osd_object *obj)
449 LASSERT(obj->oo_inode != NULL);
450 obj->oo_dt.do_body_ops = &osd_body_ops;
451 obj->oo_dt.do_lu.lo_header->loh_attr |=
452 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
456 * Concurrency: no concurrent access is possible that early in object
459 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
460 const struct lu_object_conf *conf)
462 struct osd_object *obj = osd_obj(l);
465 LINVRNT(osd_invariant(obj));
467 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
468 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
469 l->lo_header->loh_attr |= LOHA_EXISTS;
473 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
474 obj->oo_dt.do_body_ops = &osd_body_ops_new;
475 if (result == 0 && obj->oo_inode != NULL)
476 osd_object_init0(obj);
478 LINVRNT(osd_invariant(obj));
483 * Concurrency: no concurrent access is possible that late in object
486 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
488 struct osd_object *obj = osd_obj(l);
490 LINVRNT(osd_invariant(obj));
492 dt_object_fini(&obj->oo_dt);
493 if (obj->oo_hl_head != NULL)
494 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
499 * Concurrency: no concurrent access is possible that late in object
502 static void osd_index_fini(struct osd_object *o)
504 struct iam_container *bag;
506 if (o->oo_dir != NULL) {
507 bag = &o->oo_dir->od_container;
508 if (o->oo_inode != NULL) {
509 if (bag->ic_object == o->oo_inode)
510 iam_container_fini(bag);
512 OBD_FREE_PTR(o->oo_dir);
518 * Concurrency: no concurrent access is possible that late in object
519 * life-cycle (for all existing callers, that is. New callers have to provide
520 * their own locking.)
522 static int osd_inode_unlinked(const struct inode *inode)
524 return inode->i_nlink == 0;
528 OSD_TXN_OI_DELETE_CREDITS = 20,
529 OSD_TXN_INODE_DELETE_CREDITS = 20
536 #if OSD_THANDLE_STATS
538 * Set time when the handle is allocated
540 static void osd_th_alloced(struct osd_thandle *oth)
542 oth->oth_alloced = cfs_time_current();
546 * Set time when the handle started
548 static void osd_th_started(struct osd_thandle *oth)
550 oth->oth_started = cfs_time_current();
554 * Helper function to convert time interval to microseconds packed in
557 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
561 cfs_duration_usec(cfs_time_sub(end, start), &val);
562 return val.tv_sec * 1000000 + val.tv_usec;
566 * Check whether the we deal with this handle for too long.
568 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
569 cfs_time_t alloced, cfs_time_t started,
572 cfs_time_t now = cfs_time_current();
574 LASSERT(dev != NULL);
576 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
577 interval_to_usec(alloced, started));
578 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
579 interval_to_usec(started, closed));
580 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
581 interval_to_usec(closed, now));
583 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
584 CWARN("transaction handle %p was open for too long: "
586 "alloced "CFS_TIME_T" ,"
587 "started "CFS_TIME_T" ,"
588 "closed "CFS_TIME_T"\n",
589 oth, now, alloced, started, closed);
590 libcfs_debug_dumpstack(NULL);
594 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
596 cfs_time_t __closed = cfs_time_current(); \
597 cfs_time_t __alloced = oth->oth_alloced; \
598 cfs_time_t __started = oth->oth_started; \
601 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
604 #else /* OSD_THANDLE_STATS */
606 #define osd_th_alloced(h) do {} while(0)
607 #define osd_th_started(h) do {} while(0)
608 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
610 #endif /* OSD_THANDLE_STATS */
613 * Concurrency: doesn't access mutable data.
615 static int osd_param_is_not_sane(const struct osd_device *dev,
616 const struct thandle *th)
618 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
620 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
624 * Concurrency: shouldn't matter.
626 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
627 static void osd_trans_commit_cb(struct super_block *sb,
628 struct journal_callback *jcb, int error)
630 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
633 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
634 struct thandle *th = &oh->ot_super;
635 struct lu_device *lud = &th->th_dev->dd_lu_dev;
636 struct dt_txn_commit_cb *dcb, *tmp;
638 LASSERT(oh->ot_handle == NULL);
641 CERROR("transaction @0x%p commit error: %d\n", th, error);
643 dt_txn_hook_commit(th);
645 /* call per-transaction callbacks if any */
646 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
647 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
648 "commit callback entry: magic=%x name='%s'\n",
649 dcb->dcb_magic, dcb->dcb_name);
650 cfs_list_del_init(&dcb->dcb_linkage);
651 dcb->dcb_func(NULL, th, dcb, error);
654 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
658 lu_context_exit(&th->th_ctx);
659 lu_context_fini(&th->th_ctx);
663 static struct thandle *osd_trans_create(const struct lu_env *env,
666 struct osd_thread_info *oti = osd_oti_get(env);
667 struct osd_iobuf *iobuf = &oti->oti_iobuf;
668 struct osd_thandle *oh;
672 /* on pending IO in this thread should left from prev. request */
673 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
675 th = ERR_PTR(-ENOMEM);
676 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
678 oh->ot_quota_trans = &oti->oti_quota_trans;
679 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
683 th->th_tags = LCT_TX_HANDLE;
685 oti->oti_dev = osd_dt_dev(d);
686 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
689 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
690 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
691 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
692 oti->oti_rollback = false;
698 * Concurrency: shouldn't matter.
700 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
703 struct osd_thread_info *oti = osd_oti_get(env);
704 struct osd_device *dev = osd_dt_dev(d);
706 struct osd_thandle *oh;
711 LASSERT(current->journal_info == NULL);
713 oh = container_of0(th, struct osd_thandle, ot_super);
715 LASSERT(oh->ot_handle == NULL);
717 rc = dt_txn_hook_start(env, d, th);
721 if (unlikely(osd_param_is_not_sane(dev, th))) {
722 static unsigned long last_printed;
723 static int last_credits;
725 CWARN("%.16s: too many transaction credits (%d > %d)\n",
726 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
728 osd_journal(dev)->j_max_transaction_buffers);
729 #ifdef OSD_TRACK_DECLARES
730 CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
731 oti->oti_declare_ops[OSD_OT_CREATE],
732 oti->oti_declare_ops_cred[OSD_OT_CREATE],
733 oti->oti_declare_ops[OSD_OT_DELETE],
734 oti->oti_declare_ops_cred[OSD_OT_DELETE],
735 oti->oti_declare_ops[OSD_OT_DESTROY],
736 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
737 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
738 oti->oti_declare_ops[OSD_OT_ATTR_SET],
739 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
740 oti->oti_declare_ops[OSD_OT_XATTR_SET],
741 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
742 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
743 oti->oti_declare_ops[OSD_OT_WRITE],
744 oti->oti_declare_ops_cred[OSD_OT_WRITE],
745 oti->oti_declare_ops[OSD_OT_PUNCH],
746 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
747 oti->oti_declare_ops[OSD_OT_QUOTA],
748 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
749 CWARN(" insert: %u/%u, delete: %u/%u\n",
750 oti->oti_declare_ops[OSD_OT_INSERT],
751 oti->oti_declare_ops_cred[OSD_OT_INSERT],
752 oti->oti_declare_ops[OSD_OT_DESTROY],
753 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
754 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
755 oti->oti_declare_ops[OSD_OT_REF_ADD],
756 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
757 oti->oti_declare_ops[OSD_OT_REF_DEL],
758 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
760 if (last_credits != oh->ot_credits &&
761 time_after(jiffies, last_printed + 60 * HZ)) {
762 libcfs_debug_dumpstack(NULL);
763 last_credits = oh->ot_credits;
764 last_printed = jiffies;
767 /* XXX Limit the credits to 'max_transaction_buffers', and
768 * let the underlying filesystem to catch the error if
769 * we really need so many credits.
771 * This should be removed when we can calculate the
772 * credits precisely. */
773 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
777 * XXX temporary stuff. Some abstraction layer should
780 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
784 LASSERT(oti->oti_txns == 0);
785 lu_context_init(&th->th_ctx, th->th_tags);
786 lu_context_enter(&th->th_ctx);
788 lu_device_get(&d->dd_lu_dev);
789 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
801 * Concurrency: shouldn't matter.
803 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
806 struct osd_thandle *oh;
807 struct osd_thread_info *oti = osd_oti_get(env);
808 struct osd_iobuf *iobuf = &oti->oti_iobuf;
809 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
812 oh = container_of0(th, struct osd_thandle, ot_super);
815 /* inform the quota slave device that the transaction is
817 qsd_op_end(env, qsd, oh->ot_quota_trans);
818 oh->ot_quota_trans = NULL;
820 if (oh->ot_handle != NULL) {
821 handle_t *hdl = oh->ot_handle;
824 * add commit callback
825 * notice we don't do this in osd_trans_start()
826 * as underlying transaction can change during truncate
828 osd_journal_callback_set(hdl, osd_trans_commit_cb,
831 LASSERT(oti->oti_txns == 1);
833 rc = dt_txn_hook_stop(env, th);
835 CERROR("Failure in transaction hook: %d\n", rc);
837 /* hook functions might modify th_sync */
838 hdl->h_sync = th->th_sync;
840 oh->ot_handle = NULL;
841 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
842 rc = ldiskfs_journal_stop(hdl));
844 CERROR("Failure to stop transaction: %d\n", rc);
849 /* as we want IO to journal and data IO be concurrent, we don't block
850 * awaiting data IO completion in osd_do_bio(), instead we wait here
851 * once transaction is submitted to the journal. all reqular requests
852 * don't do direct IO (except read/write), thus this wait_event becomes
855 * IMPORTANT: we have to wait till any IO submited by the thread is
856 * completed otherwise iobuf may be corrupted by different request
858 cfs_wait_event(iobuf->dr_wait,
859 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
861 rc = iobuf->dr_error;
866 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
868 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
871 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
872 LASSERT(&dcb->dcb_func != NULL);
873 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
879 * Called just before object is freed. Releases all resources except for
880 * object itself (that is released by osd_object_free()).
882 * Concurrency: no concurrent access is possible that late in object
885 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
887 struct osd_object *obj = osd_obj(l);
888 struct inode *inode = obj->oo_inode;
890 LINVRNT(osd_invariant(obj));
893 * If object is unlinked remove fid->ino mapping from object index.
898 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
899 qid_t uid = inode->i_uid;
900 qid_t gid = inode->i_gid;
903 obj->oo_inode = NULL;
906 struct osd_thread_info *info = osd_oti_get(env);
907 struct lquota_id_info *qi = &info->oti_qi;
909 /* Release granted quota to master if necessary */
910 qi->lqi_id.qid_uid = uid;
911 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
913 qi->lqi_id.qid_uid = gid;
914 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
920 * Concurrency: ->loo_object_release() is called under site spin-lock.
922 static void osd_object_release(const struct lu_env *env,
928 * Concurrency: shouldn't matter.
930 static int osd_object_print(const struct lu_env *env, void *cookie,
931 lu_printer_t p, const struct lu_object *l)
933 struct osd_object *o = osd_obj(l);
936 if (o->oo_dir != NULL)
937 d = o->oo_dir->od_container.ic_descr;
940 return (*p)(env, cookie,
941 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
943 o->oo_inode ? o->oo_inode->i_ino : 0UL,
944 o->oo_inode ? o->oo_inode->i_generation : 0,
945 d ? d->id_ops->id_name : "plain");
949 * Concurrency: shouldn't matter.
951 int osd_statfs(const struct lu_env *env, struct dt_device *d,
952 struct obd_statfs *sfs)
954 struct osd_device *osd = osd_dt_dev(d);
955 struct super_block *sb = osd_sb(osd);
956 struct kstatfs *ksfs;
959 if (unlikely(osd->od_mnt == NULL))
962 /* osd_lproc.c call this without env, allocate ksfs for that case */
963 if (unlikely(env == NULL)) {
968 ksfs = &osd_oti_get(env)->oti_ksfs;
971 spin_lock(&osd->od_osfs_lock);
973 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
974 result = sb->s_op->statfs(sb->s_root, ksfs);
975 if (likely(result == 0)) { /* N.B. statfs can't really fail */
976 osd->od_osfs_age = cfs_time_current_64();
977 statfs_pack(&osd->od_statfs, ksfs);
978 if (sb->s_flags & MS_RDONLY)
979 sfs->os_state = OS_STATE_READONLY;
983 if (likely(result == 0))
984 *sfs = osd->od_statfs;
985 spin_unlock(&osd->od_osfs_lock);
987 if (unlikely(env == NULL))
994 * Estimate space needed for file creations. We assume the largest filename
995 * which is 2^64 - 1, hence a filename of 20 chars.
996 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
998 #ifdef __LDISKFS_DIR_REC_LEN
999 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1001 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1005 * Concurrency: doesn't access mutable data.
1007 static void osd_conf_get(const struct lu_env *env,
1008 const struct dt_device *dev,
1009 struct dt_device_param *param)
1011 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1014 * XXX should be taken from not-yet-existing fs abstraction layer.
1016 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1017 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1018 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1019 param->ddp_block_shift = sb->s_blocksize_bits;
1020 param->ddp_mount_type = LDD_MT_LDISKFS;
1021 param->ddp_maxbytes = sb->s_maxbytes;
1022 /* Overhead estimate should be fairly accurate, so we really take a tiny
1023 * error margin which also avoids fragmenting the filesystem too much */
1024 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1025 /* inode are statically allocated, so per-inode space consumption
1026 * is the space consumed by the directory entry */
1027 param->ddp_inodespace = PER_OBJ_USAGE;
1028 /* per-fragment overhead to be used by the client code */
1029 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1030 param->ddp_mntopts = 0;
1031 if (test_opt(sb, XATTR_USER))
1032 param->ddp_mntopts |= MNTOPT_USERXATTR;
1033 if (test_opt(sb, POSIX_ACL))
1034 param->ddp_mntopts |= MNTOPT_ACL;
1036 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1037 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1038 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1041 param->ddp_max_ea_size = sb->s_blocksize;
1046 * Concurrency: shouldn't matter.
1048 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1050 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1051 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1055 * Start commit for OSD device.
1057 * An implementation of dt_commit_async method for OSD device.
1058 * Asychronously starts underlayng fs sync and thereby a transaction
1061 * \param env environment
1062 * \param d dt device
1064 * \see dt_device_operations
1066 static int osd_commit_async(const struct lu_env *env,
1067 struct dt_device *d)
1069 struct super_block *s = osd_sb(osd_dt_dev(d));
1072 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1073 RETURN(s->s_op->sync_fs(s, 0));
1077 * Concurrency: shouldn't matter.
1080 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1082 struct super_block *sb = osd_sb(osd_dt_dev(d));
1086 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1088 rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1093 * Concurrency: serialization provided by callers.
1095 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1096 int mode, unsigned long timeout, __u32 alg,
1097 struct lustre_capa_key *keys)
1099 struct osd_device *dev = osd_dt_dev(d);
1102 dev->od_fl_capa = mode;
1103 dev->od_capa_timeout = timeout;
1104 dev->od_capa_alg = alg;
1105 dev->od_capa_keys = keys;
1110 * Note: we do not count into QUOTA here.
1111 * If we mount with --data_journal we may need more.
1113 const int osd_dto_credits_noquota[DTO_NR] = {
1116 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1117 * SINGLEDATA_TRANS_BLOCKS(8)
1118 * XXX Note: maybe iam need more, since iam have more level than
1121 [DTO_INDEX_INSERT] = 16,
1122 [DTO_INDEX_DELETE] = 16,
1126 [DTO_INDEX_UPDATE] = 16,
1128 * Create a object. The same as create object in EXT3.
1129 * DATA_TRANS_BLOCKS(14) +
1130 * INDEX_EXTRA_BLOCKS(8) +
1131 * 3(inode bits, groups, GDT)
1133 [DTO_OBJECT_CREATE] = 25,
1135 * XXX: real credits to be fixed
1137 [DTO_OBJECT_DELETE] = 25,
1139 * Attr set credits (inode)
1141 [DTO_ATTR_SET_BASE] = 1,
1143 * Xattr set. The same as xattr of EXT3.
1144 * DATA_TRANS_BLOCKS(14)
1145 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1146 * are also counted in. Do not know why?
1148 [DTO_XATTR_SET] = 14,
1151 * credits for inode change during write.
1153 [DTO_WRITE_BASE] = 3,
1155 * credits for single block write.
1157 [DTO_WRITE_BLOCK] = 14,
1159 * Attr set credits for chown.
1160 * This is extra credits for setattr, and it is null without quota
1162 [DTO_ATTR_SET_CHOWN]= 0
1165 static const struct dt_device_operations osd_dt_ops = {
1166 .dt_root_get = osd_root_get,
1167 .dt_statfs = osd_statfs,
1168 .dt_trans_create = osd_trans_create,
1169 .dt_trans_start = osd_trans_start,
1170 .dt_trans_stop = osd_trans_stop,
1171 .dt_trans_cb_add = osd_trans_cb_add,
1172 .dt_conf_get = osd_conf_get,
1173 .dt_sync = osd_sync,
1175 .dt_commit_async = osd_commit_async,
1176 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1179 static void osd_object_read_lock(const struct lu_env *env,
1180 struct dt_object *dt, unsigned role)
1182 struct osd_object *obj = osd_dt_obj(dt);
1183 struct osd_thread_info *oti = osd_oti_get(env);
1185 LINVRNT(osd_invariant(obj));
1187 LASSERT(obj->oo_owner != env);
1188 down_read_nested(&obj->oo_sem, role);
1190 LASSERT(obj->oo_owner == NULL);
1194 static void osd_object_write_lock(const struct lu_env *env,
1195 struct dt_object *dt, unsigned role)
1197 struct osd_object *obj = osd_dt_obj(dt);
1198 struct osd_thread_info *oti = osd_oti_get(env);
1200 LINVRNT(osd_invariant(obj));
1202 LASSERT(obj->oo_owner != env);
1203 down_write_nested(&obj->oo_sem, role);
1205 LASSERT(obj->oo_owner == NULL);
1206 obj->oo_owner = env;
1210 static void osd_object_read_unlock(const struct lu_env *env,
1211 struct dt_object *dt)
1213 struct osd_object *obj = osd_dt_obj(dt);
1214 struct osd_thread_info *oti = osd_oti_get(env);
1216 LINVRNT(osd_invariant(obj));
1218 LASSERT(oti->oti_r_locks > 0);
1220 up_read(&obj->oo_sem);
1223 static void osd_object_write_unlock(const struct lu_env *env,
1224 struct dt_object *dt)
1226 struct osd_object *obj = osd_dt_obj(dt);
1227 struct osd_thread_info *oti = osd_oti_get(env);
1229 LINVRNT(osd_invariant(obj));
1231 LASSERT(obj->oo_owner == env);
1232 LASSERT(oti->oti_w_locks > 0);
1234 obj->oo_owner = NULL;
1235 up_write(&obj->oo_sem);
1238 static int osd_object_write_locked(const struct lu_env *env,
1239 struct dt_object *dt)
1241 struct osd_object *obj = osd_dt_obj(dt);
1243 LINVRNT(osd_invariant(obj));
1245 return obj->oo_owner == env;
1248 static int capa_is_sane(const struct lu_env *env,
1249 struct osd_device *dev,
1250 struct lustre_capa *capa,
1251 struct lustre_capa_key *keys)
1253 struct osd_thread_info *oti = osd_oti_get(env);
1254 struct lustre_capa *tcapa = &oti->oti_capa;
1255 struct obd_capa *oc;
1259 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1261 if (capa_is_expired(oc)) {
1262 DEBUG_CAPA(D_ERROR, capa, "expired");
1269 if (capa_is_expired_sec(capa)) {
1270 DEBUG_CAPA(D_ERROR, capa, "expired");
1274 spin_lock(&capa_lock);
1275 for (i = 0; i < 2; i++) {
1276 if (keys[i].lk_keyid == capa->lc_keyid) {
1277 oti->oti_capa_key = keys[i];
1281 spin_unlock(&capa_lock);
1284 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1288 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1292 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1293 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1297 oc = capa_add(dev->od_capa_hash, capa);
1303 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1304 struct lustre_capa *capa, __u64 opc)
1306 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1307 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1308 struct md_capainfo *ci;
1311 if (!dev->od_fl_capa)
1314 if (capa == BYPASS_CAPA)
1317 ci = md_capainfo(env);
1321 if (ci->mc_auth == LC_ID_NONE)
1325 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1329 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1330 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1335 if (!capa_opc_supported(capa, opc)) {
1336 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1340 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1341 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1348 static struct timespec *osd_inode_time(const struct lu_env *env,
1349 struct inode *inode, __u64 seconds)
1351 struct osd_thread_info *oti = osd_oti_get(env);
1352 struct timespec *t = &oti->oti_time;
1354 t->tv_sec = seconds;
1356 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1361 static void osd_inode_getattr(const struct lu_env *env,
1362 struct inode *inode, struct lu_attr *attr)
1364 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1365 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1366 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1368 attr->la_atime = LTIME_S(inode->i_atime);
1369 attr->la_mtime = LTIME_S(inode->i_mtime);
1370 attr->la_ctime = LTIME_S(inode->i_ctime);
1371 attr->la_mode = inode->i_mode;
1372 attr->la_size = i_size_read(inode);
1373 attr->la_blocks = inode->i_blocks;
1374 attr->la_uid = inode->i_uid;
1375 attr->la_gid = inode->i_gid;
1376 attr->la_flags = LDISKFS_I(inode)->i_flags;
1377 attr->la_nlink = inode->i_nlink;
1378 attr->la_rdev = inode->i_rdev;
1379 attr->la_blksize = 1 << inode->i_blkbits;
1380 attr->la_blkbits = inode->i_blkbits;
1383 static int osd_attr_get(const struct lu_env *env,
1384 struct dt_object *dt,
1385 struct lu_attr *attr,
1386 struct lustre_capa *capa)
1388 struct osd_object *obj = osd_dt_obj(dt);
1390 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1391 LINVRNT(osd_invariant(obj));
1393 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1396 spin_lock(&obj->oo_guard);
1397 osd_inode_getattr(env, obj->oo_inode, attr);
1398 spin_unlock(&obj->oo_guard);
1402 static int osd_declare_attr_set(const struct lu_env *env,
1403 struct dt_object *dt,
1404 const struct lu_attr *attr,
1405 struct thandle *handle)
1407 struct osd_thandle *oh;
1408 struct osd_object *obj;
1409 struct osd_thread_info *info = osd_oti_get(env);
1410 struct lquota_id_info *qi = &info->oti_qi;
1416 LASSERT(dt != NULL);
1417 LASSERT(handle != NULL);
1419 obj = osd_dt_obj(dt);
1420 LASSERT(osd_invariant(obj));
1422 oh = container_of0(handle, struct osd_thandle, ot_super);
1423 LASSERT(oh->ot_handle == NULL);
1425 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1426 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1428 if (attr == NULL || obj->oo_inode == NULL)
1431 bspace = obj->oo_inode->i_blocks;
1432 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1433 bspace = toqb(bspace);
1435 /* Changing ownership is always preformed by super user, it should not
1438 * We still need to call the osd_declare_qid() to calculate the journal
1439 * credits for updating quota accounting files and to trigger quota
1440 * space adjustment once the operation is completed.*/
1441 if ((attr->la_valid & LA_UID) != 0 &&
1442 attr->la_uid != obj->oo_inode->i_uid) {
1443 qi->lqi_type = USRQUOTA;
1445 /* inode accounting */
1446 qi->lqi_is_blk = false;
1448 /* one more inode for the new owner ... */
1449 qi->lqi_id.qid_uid = attr->la_uid;
1451 allocated = (attr->la_uid == 0) ? true : false;
1452 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1453 if (rc == -EDQUOT || rc == -EINPROGRESS)
1458 /* and one less inode for the current uid */
1459 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1461 rc = osd_declare_qid(env, oh, qi, true, NULL);
1462 if (rc == -EDQUOT || rc == -EINPROGRESS)
1467 /* block accounting */
1468 qi->lqi_is_blk = true;
1470 /* more blocks for the new owner ... */
1471 qi->lqi_id.qid_uid = attr->la_uid;
1472 qi->lqi_space = bspace;
1473 allocated = (attr->la_uid == 0) ? true : false;
1474 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1475 if (rc == -EDQUOT || rc == -EINPROGRESS)
1480 /* and finally less blocks for the current owner */
1481 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1482 qi->lqi_space = -bspace;
1483 rc = osd_declare_qid(env, oh, qi, true, NULL);
1484 if (rc == -EDQUOT || rc == -EINPROGRESS)
1490 if (attr->la_valid & LA_GID &&
1491 attr->la_gid != obj->oo_inode->i_gid) {
1492 qi->lqi_type = GRPQUOTA;
1494 /* inode accounting */
1495 qi->lqi_is_blk = false;
1497 /* one more inode for the new group owner ... */
1498 qi->lqi_id.qid_gid = attr->la_gid;
1500 allocated = (attr->la_gid == 0) ? true : false;
1501 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1502 if (rc == -EDQUOT || rc == -EINPROGRESS)
1507 /* and one less inode for the current gid */
1508 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1510 rc = osd_declare_qid(env, oh, qi, true, NULL);
1511 if (rc == -EDQUOT || rc == -EINPROGRESS)
1516 /* block accounting */
1517 qi->lqi_is_blk = true;
1519 /* more blocks for the new owner ... */
1520 qi->lqi_id.qid_gid = attr->la_gid;
1521 qi->lqi_space = bspace;
1522 allocated = (attr->la_gid == 0) ? true : false;
1523 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1524 if (rc == -EDQUOT || rc == -EINPROGRESS)
1529 /* and finally less blocks for the current owner */
1530 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1531 qi->lqi_space = -bspace;
1532 rc = osd_declare_qid(env, oh, qi, true, NULL);
1533 if (rc == -EDQUOT || rc == -EINPROGRESS)
1542 static int osd_inode_setattr(const struct lu_env *env,
1543 struct inode *inode, const struct lu_attr *attr)
1547 bits = attr->la_valid;
1549 if (bits & LA_ATIME)
1550 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1551 if (bits & LA_CTIME)
1552 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1553 if (bits & LA_MTIME)
1554 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1555 if (bits & LA_SIZE) {
1556 LDISKFS_I(inode)->i_disksize = attr->la_size;
1557 i_size_write(inode, attr->la_size);
1561 /* OSD should not change "i_blocks" which is used by quota.
1562 * "i_blocks" should be changed by ldiskfs only. */
1563 if (bits & LA_BLOCKS)
1564 inode->i_blocks = attr->la_blocks;
1567 inode->i_mode = (inode->i_mode & S_IFMT) |
1568 (attr->la_mode & ~S_IFMT);
1570 inode->i_uid = attr->la_uid;
1572 inode->i_gid = attr->la_gid;
1573 if (bits & LA_NLINK)
1574 set_nlink(inode, attr->la_nlink);
1576 inode->i_rdev = attr->la_rdev;
1578 if (bits & LA_FLAGS) {
1579 /* always keep S_NOCMTIME */
1580 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1586 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1588 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1589 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1594 if (attr->la_valid & LA_UID)
1595 iattr.ia_valid |= ATTR_UID;
1596 if (attr->la_valid & LA_GID)
1597 iattr.ia_valid |= ATTR_GID;
1598 iattr.ia_uid = attr->la_uid;
1599 iattr.ia_gid = attr->la_gid;
1601 rc = ll_vfs_dq_transfer(inode, &iattr);
1603 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1604 "enforcement enabled on the ldiskfs filesystem?",
1605 inode->i_sb->s_id, rc);
1612 static int osd_attr_set(const struct lu_env *env,
1613 struct dt_object *dt,
1614 const struct lu_attr *attr,
1615 struct thandle *handle,
1616 struct lustre_capa *capa)
1618 struct osd_object *obj = osd_dt_obj(dt);
1619 struct inode *inode;
1622 LASSERT(handle != NULL);
1623 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1624 LASSERT(osd_invariant(obj));
1626 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1629 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1631 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1632 struct osd_thread_info *oti = osd_oti_get(env);
1633 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1634 struct lu_fid *fid1 = &oti->oti_fid;
1635 struct osd_inode_id *id = &oti->oti_id;
1636 struct iam_path_descr *ipd;
1637 struct iam_container *bag;
1638 struct osd_thandle *oh;
1641 fid_cpu_to_be(fid1, fid0);
1642 memset(id, 1, sizeof(*id));
1643 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1644 fid0)->oi_dir.od_container;
1645 ipd = osd_idx_ipd_get(env, bag);
1646 if (unlikely(ipd == NULL))
1649 oh = container_of0(handle, struct osd_thandle, ot_super);
1650 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1651 (const struct iam_rec *)id, ipd);
1652 osd_ipd_put(env, bag, ipd);
1653 return(rc > 0 ? 0 : rc);
1656 inode = obj->oo_inode;
1657 ll_vfs_dq_init(inode);
1659 rc = osd_quota_transfer(inode, attr);
1663 spin_lock(&obj->oo_guard);
1664 rc = osd_inode_setattr(env, inode, attr);
1665 spin_unlock(&obj->oo_guard);
1668 inode->i_sb->s_op->dirty_inode(inode);
1672 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1673 struct osd_object *obj,
1674 const char *name, const int namelen)
1676 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1679 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1681 struct dt_allocation_hint *hint,
1685 struct osd_device *osd = osd_obj2dev(obj);
1686 struct osd_thandle *oth;
1687 struct dt_object *parent = NULL;
1688 struct inode *inode;
1690 LINVRNT(osd_invariant(obj));
1691 LASSERT(obj->oo_inode == NULL);
1692 LASSERT(obj->oo_hl_head == NULL);
1694 if (S_ISDIR(mode) && ldiskfs_pdo) {
1695 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1696 if (obj->oo_hl_head == NULL)
1700 oth = container_of(th, struct osd_thandle, ot_super);
1701 LASSERT(oth->ot_handle->h_transaction != NULL);
1703 if (hint && hint->dah_parent)
1704 parent = hint->dah_parent;
1706 inode = ldiskfs_create_inode(oth->ot_handle,
1707 parent ? osd_dt_obj(parent)->oo_inode :
1708 osd_sb(osd)->s_root->d_inode,
1710 if (!IS_ERR(inode)) {
1711 /* Do not update file c/mtime in ldiskfs.
1712 * NB: don't need any lock because no contention at this
1714 inode->i_flags |= S_NOCMTIME;
1716 /* For new created object, it must be consistent,
1717 * and it is unnecessary to scrub against it. */
1718 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1719 obj->oo_inode = inode;
1722 if (obj->oo_hl_head != NULL) {
1723 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1724 obj->oo_hl_head = NULL;
1726 result = PTR_ERR(inode);
1728 LINVRNT(osd_invariant(obj));
1736 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1737 struct lu_attr *attr,
1738 struct dt_allocation_hint *hint,
1739 struct dt_object_format *dof,
1743 struct osd_thandle *oth;
1744 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1746 LASSERT(S_ISDIR(attr->la_mode));
1748 oth = container_of(th, struct osd_thandle, ot_super);
1749 LASSERT(oth->ot_handle->h_transaction != NULL);
1750 result = osd_mkfile(info, obj, mode, hint, th);
1755 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1756 struct lu_attr *attr,
1757 struct dt_allocation_hint *hint,
1758 struct dt_object_format *dof,
1762 struct osd_thandle *oth;
1763 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1765 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1767 LASSERT(S_ISREG(attr->la_mode));
1769 oth = container_of(th, struct osd_thandle, ot_super);
1770 LASSERT(oth->ot_handle->h_transaction != NULL);
1772 result = osd_mkfile(info, obj, mode, hint, th);
1774 LASSERT(obj->oo_inode != NULL);
1775 if (feat->dif_flags & DT_IND_VARKEY)
1776 result = iam_lvar_create(obj->oo_inode,
1777 feat->dif_keysize_max,
1779 feat->dif_recsize_max,
1782 result = iam_lfix_create(obj->oo_inode,
1783 feat->dif_keysize_max,
1785 feat->dif_recsize_max,
1792 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1793 struct lu_attr *attr,
1794 struct dt_allocation_hint *hint,
1795 struct dt_object_format *dof,
1798 LASSERT(S_ISREG(attr->la_mode));
1799 return osd_mkfile(info, obj, (attr->la_mode &
1800 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1803 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1804 struct lu_attr *attr,
1805 struct dt_allocation_hint *hint,
1806 struct dt_object_format *dof,
1809 LASSERT(S_ISLNK(attr->la_mode));
1810 return osd_mkfile(info, obj, (attr->la_mode &
1811 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1814 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1815 struct lu_attr *attr,
1816 struct dt_allocation_hint *hint,
1817 struct dt_object_format *dof,
1820 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1823 LINVRNT(osd_invariant(obj));
1824 LASSERT(obj->oo_inode == NULL);
1825 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1826 S_ISFIFO(mode) || S_ISSOCK(mode));
1828 result = osd_mkfile(info, obj, mode, hint, th);
1830 LASSERT(obj->oo_inode != NULL);
1832 * This inode should be marked dirty for i_rdev. Currently
1833 * that is done in the osd_attr_init().
1835 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1838 LINVRNT(osd_invariant(obj));
1842 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1844 struct dt_allocation_hint *hint,
1845 struct dt_object_format *dof,
1848 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1850 osd_obj_type_f result;
1866 result = osd_mk_index;
1877 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1878 struct dt_object *parent, struct dt_object *child,
1879 cfs_umode_t child_mode)
1883 memset(ah, 0, sizeof(*ah));
1884 ah->dah_parent = parent;
1885 ah->dah_mode = child_mode;
1888 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1889 struct lu_attr *attr, struct dt_object_format *dof)
1891 struct inode *inode = obj->oo_inode;
1892 __u64 valid = attr->la_valid;
1895 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1897 if (dof->dof_type != DFT_NODE)
1898 attr->la_valid &= ~LA_RDEV;
1899 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1900 attr->la_valid &= ~LA_ATIME;
1901 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1902 attr->la_valid &= ~LA_CTIME;
1903 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1904 attr->la_valid &= ~LA_MTIME;
1906 result = osd_quota_transfer(inode, attr);
1910 if (attr->la_valid != 0) {
1911 result = osd_inode_setattr(info->oti_env, inode, attr);
1913 * The osd_inode_setattr() should always succeed here. The
1914 * only error that could be returned is EDQUOT when we are
1915 * trying to change the UID or GID of the inode. However, this
1916 * should not happen since quota enforcement is no longer
1917 * enabled on ldiskfs (lquota takes care of it).
1919 LASSERTF(result == 0, "%d", result);
1920 inode->i_sb->s_op->dirty_inode(inode);
1923 attr->la_valid = valid;
1927 * Helper function for osd_object_create()
1929 * \retval 0, on success
1931 static int __osd_object_create(struct osd_thread_info *info,
1932 struct osd_object *obj, struct lu_attr *attr,
1933 struct dt_allocation_hint *hint,
1934 struct dt_object_format *dof,
1939 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1942 osd_attr_init(info, obj, attr, dof);
1943 osd_object_init0(obj);
1945 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1946 unlock_new_inode(obj->oo_inode);
1953 * Helper function for osd_object_create()
1955 * \retval 0, on success
1957 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1958 const struct lu_fid *fid, struct thandle *th)
1960 struct osd_thread_info *info = osd_oti_get(env);
1961 struct osd_inode_id *id = &info->oti_id;
1962 struct osd_device *osd = osd_obj2dev(obj);
1964 LASSERT(obj->oo_inode != NULL);
1966 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1967 return osd_oi_insert(info, osd, fid, id, th);
1970 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1971 const struct lu_fid *fid, struct lu_seq_range *range)
1973 struct seq_server_site *ss = osd_seq_site(osd);
1976 if (fid_is_igif(fid)) {
1977 range->lsr_flags = LU_SEQ_RANGE_MDT;
1978 range->lsr_index = 0;
1982 if (fid_is_idif(fid)) {
1983 range->lsr_flags = LU_SEQ_RANGE_OST;
1984 range->lsr_index = fid_idif_ost_idx(fid);
1988 if (!fid_is_norm(fid)) {
1989 range->lsr_flags = LU_SEQ_RANGE_MDT;
1991 /* FIXME: If ss is NULL, it suppose not get lsr_index
1993 range->lsr_index = ss->ss_node_id;
1997 LASSERT(ss != NULL);
1998 range->lsr_flags = -1;
1999 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2001 CERROR("%s can not find "DFID": rc = %d\n",
2002 osd_name(osd), PFID(fid), rc);
2008 * Concurrency: no external locking is necessary.
2010 static int osd_declare_object_create(const struct lu_env *env,
2011 struct dt_object *dt,
2012 struct lu_attr *attr,
2013 struct dt_allocation_hint *hint,
2014 struct dt_object_format *dof,
2015 struct thandle *handle)
2017 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
2018 struct osd_thandle *oh;
2022 LASSERT(handle != NULL);
2024 oh = container_of0(handle, struct osd_thandle, ot_super);
2025 LASSERT(oh->ot_handle == NULL);
2027 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2028 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2029 if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2030 lu_object_fid(&dt->do_lu)))
2031 /* Reuse idle OI block may cause additional one OI block
2033 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2034 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2036 /* If this is directory, then we expect . and .. to be inserted as
2037 * well. The one directory block always needs to be created for the
2038 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2039 * block), there is no danger of needing a tree for the first block.
2041 if (attr && S_ISDIR(attr->la_mode)) {
2042 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2043 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2044 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2050 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2051 false, false, NULL, false);
2055 /* It does fld look up inside declare, and the result will be
2056 * added to fld cache, so the following fld lookup inside insert
2057 * does not need send RPC anymore, so avoid send rpc with holding
2059 if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2060 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2061 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2062 lu_object_fid(&dt->do_lu), range);
2068 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2069 struct lu_attr *attr,
2070 struct dt_allocation_hint *hint,
2071 struct dt_object_format *dof,
2074 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2075 struct osd_object *obj = osd_dt_obj(dt);
2076 struct osd_thread_info *info = osd_oti_get(env);
2081 LINVRNT(osd_invariant(obj));
2082 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2083 LASSERT(osd_write_locked(env, obj));
2084 LASSERT(th != NULL);
2086 if (unlikely(fid_is_acct(fid)))
2087 /* Quota files can't be created from the kernel any more,
2088 * 'tune2fs -O quota' will take care of creating them */
2091 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2092 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2094 result = __osd_object_create(info, obj, attr, hint, dof, th);
2096 result = __osd_oi_insert(env, obj, fid, th);
2098 LASSERT(ergo(result == 0,
2099 dt_object_exists(dt) && !dt_object_remote(dt)));
2101 LASSERT(osd_invariant(obj));
2106 * Called to destroy on-disk representation of the object
2108 * Concurrency: must be locked
2110 static int osd_declare_object_destroy(const struct lu_env *env,
2111 struct dt_object *dt,
2114 struct osd_object *obj = osd_dt_obj(dt);
2115 struct inode *inode = obj->oo_inode;
2116 struct osd_thandle *oh;
2120 oh = container_of0(th, struct osd_thandle, ot_super);
2121 LASSERT(oh->ot_handle == NULL);
2124 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2125 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2126 /* Recycle idle OI leaf may cause additional three OI blocks
2128 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2129 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2131 /* one less inode */
2132 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2133 false, true, NULL, false);
2136 /* data to be truncated */
2137 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2138 true, true, NULL, false);
2142 static int osd_object_destroy(const struct lu_env *env,
2143 struct dt_object *dt,
2146 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2147 struct osd_object *obj = osd_dt_obj(dt);
2148 struct inode *inode = obj->oo_inode;
2149 struct osd_device *osd = osd_obj2dev(obj);
2150 struct osd_thandle *oh;
2154 oh = container_of0(th, struct osd_thandle, ot_super);
2155 LASSERT(oh->ot_handle);
2157 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2159 if (unlikely(fid_is_acct(fid)))
2162 /* Parallel control for OI scrub. For most of cases, there is no
2163 * lock contention. So it will not affect unlink performance. */
2164 mutex_lock(&inode->i_mutex);
2165 if (S_ISDIR(inode->i_mode)) {
2166 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2167 /* it will check/delete the agent inode for every dir
2168 * destory, how to optimize it? unlink performance
2170 result = osd_delete_from_agent(env, osd, obj, oh);
2171 if (result != 0 && result != -ENOENT) {
2172 CERROR("%s: delete agent inode "DFID": rc = %d\n",
2173 osd_name(osd), PFID(fid), result);
2175 spin_lock(&obj->oo_guard);
2177 spin_unlock(&obj->oo_guard);
2178 inode->i_sb->s_op->dirty_inode(inode);
2181 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2183 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2184 mutex_unlock(&inode->i_mutex);
2186 /* XXX: add to ext3 orphan list */
2187 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2189 /* not needed in the cache anymore */
2190 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2195 static inline int __osd_xattr_set(struct osd_thread_info *info,
2196 struct inode *inode, const char *name,
2197 const void *buf, int buflen, int fl)
2199 struct dentry *dentry = &info->oti_child_dentry;
2201 ll_vfs_dq_init(inode);
2202 dentry->d_inode = inode;
2203 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
2207 * Put the fid into lustre_mdt_attrs, and then place the structure
2208 * inode's ea. This fid should not be altered during the life time
2211 * \retval +ve, on success
2212 * \retval -ve, on error
2214 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2216 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2217 const struct lu_fid *fid)
2219 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2222 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2225 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid))
2228 lustre_lma_init(lma, fid);
2229 lustre_lma_swab(lma);
2231 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2233 /* Someone may created the EA by race. */
2234 if (unlikely(rc == -EEXIST))
2240 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2241 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2242 * To have compatilibility with 1.8 ldiskfs driver we need to have
2243 * magic number at start of fid data.
2244 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2247 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2248 const struct dt_rec *fid)
2250 if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) {
2251 param->edp_magic = 0;
2255 param->edp_magic = LDISKFS_LUFID_MAGIC;
2256 param->edp_len = sizeof(struct lu_fid) + 1;
2257 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2261 * Try to read the fid from inode ea into dt_rec.
2263 * \param fid object fid.
2265 * \retval 0 on success
2267 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2268 __u32 ino, struct lu_fid *fid,
2269 struct osd_inode_id *id)
2271 struct osd_thread_info *info = osd_oti_get(env);
2272 struct inode *inode;
2275 osd_id_gen(id, ino, OSD_OII_NOGEN);
2276 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2278 RETURN(PTR_ERR(inode));
2284 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2286 struct inode *parent_dir,
2287 const struct dt_rec *dot_fid,
2288 const struct dt_rec *dot_dot_fid,
2289 struct osd_thandle *oth)
2291 struct ldiskfs_dentry_param *dot_ldp;
2292 struct ldiskfs_dentry_param *dot_dot_ldp;
2294 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2295 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2297 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2298 dot_ldp->edp_magic = 0;
2299 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2300 dir, dot_ldp, dot_dot_ldp);
2304 * Create an local inode for remote entry
2306 static struct inode *osd_create_remote_inode(const struct lu_env *env,
2307 struct osd_device *osd,
2308 struct osd_object *pobj,
2309 const struct lu_fid *fid,
2312 struct osd_thread_info *info = osd_oti_get(env);
2313 struct inode *local;
2314 struct osd_thandle *oh;
2319 oh = container_of(th, struct osd_thandle, ot_super);
2320 LASSERT(oh->ot_handle->h_transaction != NULL);
2322 /* FIXME: Insert index api needs to know the mode of
2323 * the remote object. Just use S_IFDIR for now */
2324 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2325 if (IS_ERR(local)) {
2326 CERROR("%s: create local error %d\n", osd_name(osd),
2327 (int)PTR_ERR(local));
2331 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2332 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2333 (const struct dt_rec *)fid, oh);
2335 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2336 osd_name(osd), PFID(fid), rc);
2337 RETURN(ERR_PTR(rc));
2344 * Delete local inode for remote entry
2346 static int osd_delete_remote_inode(const struct lu_env *env,
2347 struct osd_device *osd,
2348 const struct lu_fid *fid,
2349 __u32 ino, struct osd_thandle *oh)
2351 struct osd_thread_info *oti = osd_oti_get(env);
2352 struct osd_inode_id *id = &oti->oti_id;
2353 struct inode *inode;
2356 id->oii_ino = le32_to_cpu(ino);
2357 id->oii_gen = OSD_OII_NOGEN;
2358 inode = osd_iget(oti, osd, id);
2359 if (IS_ERR(inode)) {
2360 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2361 PFID(fid), id->oii_ino, id->oii_gen);
2362 RETURN(PTR_ERR(inode));
2366 mark_inode_dirty(inode);
2367 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2368 osd_name(osd), PFID(fid), inode->i_ino);
2374 * OSD layer object create function for interoperability mode (b11826).
2375 * This is mostly similar to osd_object_create(). Only difference being, fid is
2376 * inserted into inode ea here.
2378 * \retval 0, on success
2379 * \retval -ve, on error
2381 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2382 struct lu_attr *attr,
2383 struct dt_allocation_hint *hint,
2384 struct dt_object_format *dof,
2387 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2388 struct osd_object *obj = osd_dt_obj(dt);
2389 struct osd_thread_info *info = osd_oti_get(env);
2394 LASSERT(osd_invariant(obj));
2395 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2396 LASSERT(osd_write_locked(env, obj));
2397 LASSERT(th != NULL);
2399 if (unlikely(fid_is_acct(fid)))
2400 /* Quota files can't be created from the kernel any more,
2401 * 'tune2fs -O quota' will take care of creating them */
2404 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2405 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2407 result = __osd_object_create(info, obj, attr, hint, dof, th);
2408 if ((result == 0) &&
2409 (fid_is_last_id(fid) ||
2410 !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)))
2411 result = osd_ea_fid_set(info, obj->oo_inode, fid);
2414 result = __osd_oi_insert(env, obj, fid, th);
2416 LASSERT(ergo(result == 0,
2417 dt_object_exists(dt) && !dt_object_remote(dt)));
2418 LINVRNT(osd_invariant(obj));
2422 static int osd_declare_object_ref_add(const struct lu_env *env,
2423 struct dt_object *dt,
2424 struct thandle *handle)
2426 struct osd_thandle *oh;
2428 /* it's possible that object doesn't exist yet */
2429 LASSERT(handle != NULL);
2431 oh = container_of0(handle, struct osd_thandle, ot_super);
2432 LASSERT(oh->ot_handle == NULL);
2434 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2435 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2441 * Concurrency: @dt is write locked.
2443 static int osd_object_ref_add(const struct lu_env *env,
2444 struct dt_object *dt, struct thandle *th)
2446 struct osd_object *obj = osd_dt_obj(dt);
2447 struct inode *inode = obj->oo_inode;
2449 LINVRNT(osd_invariant(obj));
2450 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2451 LASSERT(osd_write_locked(env, obj));
2452 LASSERT(th != NULL);
2454 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2457 * DIR_NLINK feature is set for compatibility reasons if:
2458 * 1) nlinks > LDISKFS_LINK_MAX, or
2459 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2461 * It is easier to always set this flag (rather than check and set),
2462 * since it has less overhead, and the superblock will be dirtied
2463 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2464 * do not actually care whether this flag is set or not.
2466 spin_lock(&obj->oo_guard);
2467 /* inc_nlink from 0 may cause WARN_ON */
2468 if(inode->i_nlink == 0)
2469 set_nlink(inode, 1);
2472 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2473 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2474 inode->i_nlink == 2)
2475 set_nlink(inode, 1);
2477 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2478 spin_unlock(&obj->oo_guard);
2479 inode->i_sb->s_op->dirty_inode(inode);
2480 LINVRNT(osd_invariant(obj));
2485 static int osd_declare_object_ref_del(const struct lu_env *env,
2486 struct dt_object *dt,
2487 struct thandle *handle)
2489 struct osd_thandle *oh;
2491 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2492 LASSERT(handle != NULL);
2494 oh = container_of0(handle, struct osd_thandle, ot_super);
2495 LASSERT(oh->ot_handle == NULL);
2497 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2498 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2504 * Concurrency: @dt is write locked.
2506 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2509 struct osd_object *obj = osd_dt_obj(dt);
2510 struct inode *inode = obj->oo_inode;
2512 LINVRNT(osd_invariant(obj));
2513 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2514 LASSERT(osd_write_locked(env, obj));
2515 LASSERT(th != NULL);
2517 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2519 spin_lock(&obj->oo_guard);
2520 LASSERT(inode->i_nlink > 0);
2522 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2523 * then the nlink count is 1. Don't let it be set to 0 or the directory
2524 * inode will be deleted incorrectly. */
2525 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2526 set_nlink(inode, 1);
2527 spin_unlock(&obj->oo_guard);
2528 inode->i_sb->s_op->dirty_inode(inode);
2529 LINVRNT(osd_invariant(obj));
2535 * Get the 64-bit version for an inode.
2537 static int osd_object_version_get(const struct lu_env *env,
2538 struct dt_object *dt, dt_obj_version_t *ver)
2540 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2542 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2543 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2544 *ver = LDISKFS_I(inode)->i_fs_version;
2549 * Concurrency: @dt is read locked.
2551 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2552 struct lu_buf *buf, const char *name,
2553 struct lustre_capa *capa)
2555 struct osd_object *obj = osd_dt_obj(dt);
2556 struct inode *inode = obj->oo_inode;
2557 struct osd_thread_info *info = osd_oti_get(env);
2558 struct dentry *dentry = &info->oti_obj_dentry;
2560 /* version get is not real XATTR but uses xattr API */
2561 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2562 /* for version we are just using xattr API but change inode
2564 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2565 osd_object_version_get(env, dt, buf->lb_buf);
2566 return sizeof(dt_obj_version_t);
2569 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2570 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2572 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2575 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2579 static int osd_declare_xattr_set(const struct lu_env *env,
2580 struct dt_object *dt,
2581 const struct lu_buf *buf, const char *name,
2582 int fl, struct thandle *handle)
2584 struct osd_thandle *oh;
2586 LASSERT(handle != NULL);
2588 oh = container_of0(handle, struct osd_thandle, ot_super);
2589 LASSERT(oh->ot_handle == NULL);
2591 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2592 strcmp(name, XATTR_NAME_VERSION) == 0 ?
2593 osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2594 osd_dto_credits_noquota[DTO_XATTR_SET]);
2600 * Set the 64-bit version for object
2602 static void osd_object_version_set(const struct lu_env *env,
2603 struct dt_object *dt,
2604 dt_obj_version_t *new_version)
2606 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2608 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2609 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2611 LDISKFS_I(inode)->i_fs_version = *new_version;
2612 /** Version is set after all inode operations are finished,
2613 * so we should mark it dirty here */
2614 inode->i_sb->s_op->dirty_inode(inode);
2618 * Concurrency: @dt is write locked.
2620 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2621 const struct lu_buf *buf, const char *name, int fl,
2622 struct thandle *handle, struct lustre_capa *capa)
2624 struct osd_object *obj = osd_dt_obj(dt);
2625 struct inode *inode = obj->oo_inode;
2626 struct osd_thread_info *info = osd_oti_get(env);
2629 LASSERT(handle != NULL);
2631 /* version set is not real XATTR */
2632 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2633 /* for version we are just using xattr API but change inode
2635 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2636 osd_object_version_set(env, dt, buf->lb_buf);
2637 return sizeof(dt_obj_version_t);
2640 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2643 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2644 if (fl & LU_XATTR_REPLACE)
2645 fs_flags |= XATTR_REPLACE;
2647 if (fl & LU_XATTR_CREATE)
2648 fs_flags |= XATTR_CREATE;
2650 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2655 * Concurrency: @dt is read locked.
2657 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2658 struct lu_buf *buf, struct lustre_capa *capa)
2660 struct osd_object *obj = osd_dt_obj(dt);
2661 struct inode *inode = obj->oo_inode;
2662 struct osd_thread_info *info = osd_oti_get(env);
2663 struct dentry *dentry = &info->oti_obj_dentry;
2665 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2666 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2667 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2669 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2672 dentry->d_inode = inode;
2673 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2676 static int osd_declare_xattr_del(const struct lu_env *env,
2677 struct dt_object *dt, const char *name,
2678 struct thandle *handle)
2680 struct osd_thandle *oh;
2682 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2683 LASSERT(handle != NULL);
2685 oh = container_of0(handle, struct osd_thandle, ot_super);
2686 LASSERT(oh->ot_handle == NULL);
2688 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2689 osd_dto_credits_noquota[DTO_XATTR_SET]);
2695 * Concurrency: @dt is write locked.
2697 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2698 const char *name, struct thandle *handle,
2699 struct lustre_capa *capa)
2701 struct osd_object *obj = osd_dt_obj(dt);
2702 struct inode *inode = obj->oo_inode;
2703 struct osd_thread_info *info = osd_oti_get(env);
2704 struct dentry *dentry = &info->oti_obj_dentry;
2707 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2708 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2709 LASSERT(handle != NULL);
2711 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2714 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2716 ll_vfs_dq_init(inode);
2717 dentry->d_inode = inode;
2718 rc = inode->i_op->removexattr(dentry, name);
2722 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2723 struct dt_object *dt,
2724 struct lustre_capa *old,
2727 struct osd_thread_info *info = osd_oti_get(env);
2728 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2729 struct osd_object *obj = osd_dt_obj(dt);
2730 struct osd_device *dev = osd_obj2dev(obj);
2731 struct lustre_capa_key *key = &info->oti_capa_key;
2732 struct lustre_capa *capa = &info->oti_capa;
2733 struct obd_capa *oc;
2734 struct md_capainfo *ci;
2738 if (!dev->od_fl_capa)
2739 RETURN(ERR_PTR(-ENOENT));
2741 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2742 LINVRNT(osd_invariant(obj));
2744 /* renewal sanity check */
2745 if (old && osd_object_auth(env, dt, old, opc))
2746 RETURN(ERR_PTR(-EACCES));
2748 ci = md_capainfo(env);
2750 RETURN(ERR_PTR(-ENOENT));
2752 switch (ci->mc_auth) {
2756 capa->lc_uid = obj->oo_inode->i_uid;
2757 capa->lc_gid = obj->oo_inode->i_gid;
2758 capa->lc_flags = LC_ID_PLAIN;
2760 case LC_ID_CONVERT: {
2763 s[0] = obj->oo_inode->i_uid;
2764 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2765 s[2] = obj->oo_inode->i_gid;
2766 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2767 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2769 RETURN(ERR_PTR(rc));
2771 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2772 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2773 capa->lc_flags = LC_ID_CONVERT;
2777 RETURN(ERR_PTR(-EINVAL));
2780 capa->lc_fid = *fid;
2782 capa->lc_flags |= dev->od_capa_alg << 24;
2783 capa->lc_timeout = dev->od_capa_timeout;
2784 capa->lc_expiry = 0;
2786 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2788 LASSERT(!capa_is_expired(oc));
2792 spin_lock(&capa_lock);
2793 *key = dev->od_capa_keys[1];
2794 spin_unlock(&capa_lock);
2796 capa->lc_keyid = key->lk_keyid;
2797 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2799 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2801 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2802 RETURN(ERR_PTR(rc));
2805 oc = capa_add(dev->od_capa_hash, capa);
2809 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2811 struct osd_object *obj = osd_dt_obj(dt);
2812 struct inode *inode = obj->oo_inode;
2813 struct osd_thread_info *info = osd_oti_get(env);
2814 struct dentry *dentry = &info->oti_obj_dentry;
2815 struct file *file = &info->oti_file;
2820 dentry->d_inode = inode;
2821 file->f_dentry = dentry;
2822 file->f_mapping = inode->i_mapping;
2823 file->f_op = inode->i_fop;
2824 mutex_lock(&inode->i_mutex);
2825 rc = file->f_op->fsync(file, dentry, 0);
2826 mutex_unlock(&inode->i_mutex);
2830 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2833 struct osd_object *obj = osd_dt_obj(dt);
2836 *data = (void *)obj->oo_inode;
2844 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2845 const struct dt_index_features *feat)
2847 struct iam_descr *descr;
2849 if (osd_object_is_root(o))
2850 return feat == &dt_directory_features;
2852 LASSERT(o->oo_dir != NULL);
2854 descr = o->oo_dir->od_container.ic_descr;
2855 if (feat == &dt_directory_features) {
2856 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2862 feat->dif_keysize_min <= descr->id_key_size &&
2863 descr->id_key_size <= feat->dif_keysize_max &&
2864 feat->dif_recsize_min <= descr->id_rec_size &&
2865 descr->id_rec_size <= feat->dif_recsize_max &&
2866 !(feat->dif_flags & (DT_IND_VARKEY |
2867 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2868 ergo(feat->dif_flags & DT_IND_UPDATE,
2869 1 /* XXX check that object (and file system) is
2874 static int osd_iam_container_init(const struct lu_env *env,
2875 struct osd_object *obj,
2876 struct osd_directory *dir)
2878 struct iam_container *bag = &dir->od_container;
2881 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2885 result = iam_container_setup(bag);
2887 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2889 iam_container_fini(bag);
2896 * Concurrency: no external locking is necessary.
2898 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2899 const struct dt_index_features *feat)
2903 struct osd_object *obj = osd_dt_obj(dt);
2905 LINVRNT(osd_invariant(obj));
2907 if (osd_object_is_root(obj)) {
2908 dt->do_index_ops = &osd_index_ea_ops;
2910 } else if (feat == &dt_directory_features) {
2911 dt->do_index_ops = &osd_index_ea_ops;
2912 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2917 } else if (unlikely(feat == &dt_otable_features)) {
2918 dt->do_index_ops = &osd_otable_ops;
2920 } else if (unlikely(feat == &dt_acct_features)) {
2921 dt->do_index_ops = &osd_acct_index_ops;
2924 } else if (!osd_has_index(obj)) {
2925 struct osd_directory *dir;
2930 spin_lock(&obj->oo_guard);
2931 if (obj->oo_dir == NULL)
2935 * Concurrent thread allocated container data.
2938 spin_unlock(&obj->oo_guard);
2940 * Now, that we have container data, serialize its
2943 down_write(&obj->oo_ext_idx_sem);
2945 * recheck under lock.
2947 if (!osd_has_index(obj))
2948 result = osd_iam_container_init(env, obj, dir);
2951 up_write(&obj->oo_ext_idx_sem);
2959 if (result == 0 && skip_iam == 0) {
2960 if (!osd_iam_index_probe(env, obj, feat))
2963 LINVRNT(osd_invariant(obj));
2965 if (is_quota_glb_feat(feat))
2966 result = osd_quota_migration(env, dt, feat);
2971 static int osd_otable_it_attr_get(const struct lu_env *env,
2972 struct dt_object *dt,
2973 struct lu_attr *attr,
2974 struct lustre_capa *capa)
2980 static const struct dt_object_operations osd_obj_ops = {
2981 .do_read_lock = osd_object_read_lock,
2982 .do_write_lock = osd_object_write_lock,
2983 .do_read_unlock = osd_object_read_unlock,
2984 .do_write_unlock = osd_object_write_unlock,
2985 .do_write_locked = osd_object_write_locked,
2986 .do_attr_get = osd_attr_get,
2987 .do_declare_attr_set = osd_declare_attr_set,
2988 .do_attr_set = osd_attr_set,
2989 .do_ah_init = osd_ah_init,
2990 .do_declare_create = osd_declare_object_create,
2991 .do_create = osd_object_create,
2992 .do_declare_destroy = osd_declare_object_destroy,
2993 .do_destroy = osd_object_destroy,
2994 .do_index_try = osd_index_try,
2995 .do_declare_ref_add = osd_declare_object_ref_add,
2996 .do_ref_add = osd_object_ref_add,
2997 .do_declare_ref_del = osd_declare_object_ref_del,
2998 .do_ref_del = osd_object_ref_del,
2999 .do_xattr_get = osd_xattr_get,
3000 .do_declare_xattr_set = osd_declare_xattr_set,
3001 .do_xattr_set = osd_xattr_set,
3002 .do_declare_xattr_del = osd_declare_xattr_del,
3003 .do_xattr_del = osd_xattr_del,
3004 .do_xattr_list = osd_xattr_list,
3005 .do_capa_get = osd_capa_get,
3006 .do_object_sync = osd_object_sync,
3007 .do_data_get = osd_data_get,
3011 * dt_object_operations for interoperability mode
3012 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3014 static const struct dt_object_operations osd_obj_ea_ops = {
3015 .do_read_lock = osd_object_read_lock,
3016 .do_write_lock = osd_object_write_lock,
3017 .do_read_unlock = osd_object_read_unlock,
3018 .do_write_unlock = osd_object_write_unlock,
3019 .do_write_locked = osd_object_write_locked,
3020 .do_attr_get = osd_attr_get,
3021 .do_declare_attr_set = osd_declare_attr_set,
3022 .do_attr_set = osd_attr_set,
3023 .do_ah_init = osd_ah_init,
3024 .do_declare_create = osd_declare_object_create,
3025 .do_create = osd_object_ea_create,
3026 .do_declare_destroy = osd_declare_object_destroy,
3027 .do_destroy = osd_object_destroy,
3028 .do_index_try = osd_index_try,
3029 .do_declare_ref_add = osd_declare_object_ref_add,
3030 .do_ref_add = osd_object_ref_add,
3031 .do_declare_ref_del = osd_declare_object_ref_del,
3032 .do_ref_del = osd_object_ref_del,
3033 .do_xattr_get = osd_xattr_get,
3034 .do_declare_xattr_set = osd_declare_xattr_set,
3035 .do_xattr_set = osd_xattr_set,
3036 .do_declare_xattr_del = osd_declare_xattr_del,
3037 .do_xattr_del = osd_xattr_del,
3038 .do_xattr_list = osd_xattr_list,
3039 .do_capa_get = osd_capa_get,
3040 .do_object_sync = osd_object_sync,
3041 .do_data_get = osd_data_get,
3044 static const struct dt_object_operations osd_obj_otable_it_ops = {
3045 .do_attr_get = osd_otable_it_attr_get,
3046 .do_index_try = osd_index_try,
3049 static int osd_index_declare_iam_delete(const struct lu_env *env,
3050 struct dt_object *dt,
3051 const struct dt_key *key,
3052 struct thandle *handle)
3054 struct osd_thandle *oh;
3056 oh = container_of0(handle, struct osd_thandle, ot_super);
3057 LASSERT(oh->ot_handle == NULL);
3059 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3060 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3066 * delete a (key, value) pair from index \a dt specified by \a key
3068 * \param dt osd index object
3069 * \param key key for index
3070 * \param rec record reference
3071 * \param handle transaction handler
3074 * \retval -ve failure
3077 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3078 const struct dt_key *key,
3079 struct thandle *handle,
3080 struct lustre_capa *capa)
3082 struct osd_thread_info *oti = osd_oti_get(env);
3083 struct osd_object *obj = osd_dt_obj(dt);
3084 struct osd_thandle *oh;
3085 struct iam_path_descr *ipd;
3086 struct iam_container *bag = &obj->oo_dir->od_container;
3091 LINVRNT(osd_invariant(obj));
3092 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3093 LASSERT(bag->ic_object == obj->oo_inode);
3094 LASSERT(handle != NULL);
3096 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3099 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3101 ipd = osd_idx_ipd_get(env, bag);
3102 if (unlikely(ipd == NULL))
3105 oh = container_of0(handle, struct osd_thandle, ot_super);
3106 LASSERT(oh->ot_handle != NULL);
3107 LASSERT(oh->ot_handle->h_transaction != NULL);
3109 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3110 /* swab quota uid/gid provided by caller */
3111 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3112 key = (const struct dt_key *)&oti->oti_quota_id;
3115 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3116 osd_ipd_put(env, bag, ipd);
3117 LINVRNT(osd_invariant(obj));
3121 static int osd_index_declare_ea_delete(const struct lu_env *env,
3122 struct dt_object *dt,
3123 const struct dt_key *key,
3124 struct thandle *handle)
3126 struct osd_thandle *oh;
3127 struct inode *inode;
3131 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3132 LASSERT(handle != NULL);
3134 oh = container_of0(handle, struct osd_thandle, ot_super);
3135 LASSERT(oh->ot_handle == NULL);
3137 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3138 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3140 inode = osd_dt_obj(dt)->oo_inode;
3143 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3144 true, true, NULL, false);
3148 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3151 struct osd_fid_pack *rec;
3154 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3155 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3156 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3161 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3164 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
3165 struct seq_server_site *ss = osd_seq_site(osd);
3169 if ((!fid_is_norm(fid) && !fid_is_igif(fid)) || ss == NULL)
3172 rc = osd_fld_lookup(env, osd, fid, range);
3174 CERROR("%s: Can not lookup fld for "DFID"\n",
3175 osd_name(osd), PFID(fid));
3179 RETURN(ss->ss_node_id != range->lsr_index);
3183 * Index delete function for interoperability mode (b11826).
3184 * It will remove the directory entry added by osd_index_ea_insert().
3185 * This entry is needed to maintain name->fid mapping.
3187 * \param key, key i.e. file entry to be deleted
3189 * \retval 0, on success
3190 * \retval -ve, on error
3192 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3193 const struct dt_key *key,
3194 struct thandle *handle,
3195 struct lustre_capa *capa)
3197 struct osd_object *obj = osd_dt_obj(dt);
3198 struct inode *dir = obj->oo_inode;
3199 struct dentry *dentry;
3200 struct osd_thandle *oh;
3201 struct ldiskfs_dir_entry_2 *de = NULL;
3202 struct buffer_head *bh;
3203 struct htree_lock *hlock = NULL;
3204 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3205 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3209 LINVRNT(osd_invariant(obj));
3210 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3211 LASSERT(handle != NULL);
3213 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3215 oh = container_of(handle, struct osd_thandle, ot_super);
3216 LASSERT(oh->ot_handle != NULL);
3217 LASSERT(oh->ot_handle->h_transaction != NULL);
3219 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3222 ll_vfs_dq_init(dir);
3223 dentry = osd_child_dentry_get(env, obj,
3224 (char *)key, strlen((char *)key));
3226 if (obj->oo_hl_head != NULL) {
3227 hlock = osd_oti_get(env)->oti_hlock;
3228 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3229 dir, LDISKFS_HLOCK_DEL);
3231 down_write(&obj->oo_ext_idx_sem);
3234 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3236 rc = ldiskfs_delete_entry(oh->ot_handle,
3243 ldiskfs_htree_unlock(hlock);
3245 up_write(&obj->oo_ext_idx_sem);
3250 /* For inode on the remote MDT, .. will point to
3251 * /Agent directory. So do not try to lookup/delete
3252 * remote inode for .. */
3253 if (strcmp((char *)key, dotdot) == 0)
3256 LASSERT(de != NULL);
3257 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3258 if (rc == 0 && osd_remote_fid(env, osd, fid)) {
3259 __u32 ino = le32_to_cpu(de->inode);
3261 rc = osd_delete_remote_inode(env, osd, fid, ino, oh);
3263 CERROR("%s: del local inode "DFID": rc = %d\n",
3264 osd_name(osd), PFID(fid), rc);
3271 LASSERT(osd_invariant(obj));
3276 * Lookup index for \a key and copy record to \a rec.
3278 * \param dt osd index object
3279 * \param key key for index
3280 * \param rec record reference
3282 * \retval +ve success : exact mach
3283 * \retval 0 return record with key not greater than \a key
3284 * \retval -ve failure
3286 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3287 struct dt_rec *rec, const struct dt_key *key,
3288 struct lustre_capa *capa)
3290 struct osd_object *obj = osd_dt_obj(dt);
3291 struct iam_path_descr *ipd;
3292 struct iam_container *bag = &obj->oo_dir->od_container;
3293 struct osd_thread_info *oti = osd_oti_get(env);
3294 struct iam_iterator *it = &oti->oti_idx_it;
3295 struct iam_rec *iam_rec;
3300 LASSERT(osd_invariant(obj));
3301 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3302 LASSERT(bag->ic_object == obj->oo_inode);
3304 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3307 ipd = osd_idx_ipd_get(env, bag);
3311 /* got ipd now we can start iterator. */
3312 iam_it_init(it, bag, 0, ipd);
3314 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3315 /* swab quota uid/gid provided by caller */
3316 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3317 key = (const struct dt_key *)&oti->oti_quota_id;
3320 rc = iam_it_get(it, (struct iam_key *)key);
3322 if (S_ISDIR(obj->oo_inode->i_mode))
3323 iam_rec = (struct iam_rec *)oti->oti_ldp;
3325 iam_rec = (struct iam_rec *) rec;
3327 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3329 if (S_ISDIR(obj->oo_inode->i_mode))
3330 osd_fid_unpack((struct lu_fid *) rec,
3331 (struct osd_fid_pack *)iam_rec);
3332 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3333 osd_quota_unpack(obj, rec);
3338 osd_ipd_put(env, bag, ipd);
3340 LINVRNT(osd_invariant(obj));
3345 static int osd_index_declare_iam_insert(const struct lu_env *env,
3346 struct dt_object *dt,
3347 const struct dt_rec *rec,
3348 const struct dt_key *key,
3349 struct thandle *handle)
3351 struct osd_thandle *oh;
3353 LASSERT(handle != NULL);
3355 oh = container_of0(handle, struct osd_thandle, ot_super);
3356 LASSERT(oh->ot_handle == NULL);
3358 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3359 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3365 * Inserts (key, value) pair in \a dt index object.
3367 * \param dt osd index object
3368 * \param key key for index
3369 * \param rec record reference
3370 * \param th transaction handler
3373 * \retval -ve failure
3375 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3376 const struct dt_rec *rec,
3377 const struct dt_key *key, struct thandle *th,
3378 struct lustre_capa *capa, int ignore_quota)
3380 struct osd_object *obj = osd_dt_obj(dt);
3381 struct iam_path_descr *ipd;
3382 struct osd_thandle *oh;
3383 struct iam_container *bag = &obj->oo_dir->od_container;
3384 struct osd_thread_info *oti = osd_oti_get(env);
3385 struct iam_rec *iam_rec;
3390 LINVRNT(osd_invariant(obj));
3391 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3392 LASSERT(bag->ic_object == obj->oo_inode);
3393 LASSERT(th != NULL);
3395 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3398 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3400 ipd = osd_idx_ipd_get(env, bag);
3401 if (unlikely(ipd == NULL))
3404 oh = container_of0(th, struct osd_thandle, ot_super);
3405 LASSERT(oh->ot_handle != NULL);
3406 LASSERT(oh->ot_handle->h_transaction != NULL);
3407 if (S_ISDIR(obj->oo_inode->i_mode)) {
3408 iam_rec = (struct iam_rec *)oti->oti_ldp;
3409 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3410 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3411 /* pack quota uid/gid */
3412 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3413 key = (const struct dt_key *)&oti->oti_quota_id;
3414 /* pack quota record */
3415 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3416 iam_rec = (struct iam_rec *)rec;
3418 iam_rec = (struct iam_rec *)rec;
3421 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3423 osd_ipd_put(env, bag, ipd);
3424 LINVRNT(osd_invariant(obj));
3429 * Calls ldiskfs_add_entry() to add directory entry
3430 * into the directory. This is required for
3431 * interoperability mode (b11826)
3433 * \retval 0, on success
3434 * \retval -ve, on error
3436 static int __osd_ea_add_rec(struct osd_thread_info *info,
3437 struct osd_object *pobj, struct inode *cinode,
3438 const char *name, const struct dt_rec *fid,
3439 struct htree_lock *hlock, struct thandle *th)
3441 struct ldiskfs_dentry_param *ldp;
3442 struct dentry *child;
3443 struct osd_thandle *oth;
3446 oth = container_of(th, struct osd_thandle, ot_super);
3447 LASSERT(oth->ot_handle != NULL);
3448 LASSERT(oth->ot_handle->h_transaction != NULL);
3449 LASSERT(pobj->oo_inode);
3451 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3452 if (unlikely(pobj->oo_inode ==
3453 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3456 osd_get_ldiskfs_dirent_param(ldp, fid);
3457 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3458 child->d_fsdata = (void *)ldp;
3459 ll_vfs_dq_init(pobj->oo_inode);
3460 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3466 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3467 * into the directory.Also sets flags into osd object to
3468 * indicate dot and dotdot are created. This is required for
3469 * interoperability mode (b11826)
3471 * \param dir directory for dot and dotdot fixup.
3472 * \param obj child object for linking
3474 * \retval 0, on success
3475 * \retval -ve, on error
3477 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3478 struct osd_object *dir,
3479 struct inode *parent_dir, const char *name,
3480 const struct dt_rec *dot_fid,
3481 const struct dt_rec *dot_dot_fid,
3484 struct inode *inode = dir->oo_inode;
3485 struct osd_thandle *oth;
3488 oth = container_of(th, struct osd_thandle, ot_super);
3489 LASSERT(oth->ot_handle->h_transaction != NULL);
3490 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3492 if (strcmp(name, dot) == 0) {
3493 if (dir->oo_compat_dot_created) {
3496 LASSERT(inode == parent_dir);
3497 dir->oo_compat_dot_created = 1;
3500 } else if (strcmp(name, dotdot) == 0) {
3501 if (!dir->oo_compat_dot_created)
3503 /* in case of rename, dotdot is already created */
3504 if (dir->oo_compat_dotdot_created) {
3505 return __osd_ea_add_rec(info, dir, parent_dir, name,
3506 dot_dot_fid, NULL, th);
3509 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3510 parent_dir, dot_fid,
3513 dir->oo_compat_dotdot_created = 1;
3521 * It will call the appropriate osd_add* function and return the
3522 * value, return by respective functions.
3524 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3525 struct inode *cinode, const char *name,
3526 const struct dt_rec *fid, struct thandle *th)
3528 struct osd_thread_info *info = osd_oti_get(env);
3529 struct htree_lock *hlock;
3532 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3534 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3536 if (hlock != NULL) {
3537 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3540 down_write(&pobj->oo_ext_idx_sem);
3542 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3543 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3546 if (hlock != NULL) {
3547 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3548 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3550 down_write(&pobj->oo_ext_idx_sem);
3553 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3554 struct lu_fid *tfid = &info->oti_fid;
3556 *tfid = *(const struct lu_fid *)fid;
3558 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3559 (const struct dt_rec *)tfid,
3562 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3567 ldiskfs_htree_unlock(hlock);
3569 up_write(&pobj->oo_ext_idx_sem);
3575 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3576 struct osd_idmap_cache *oic)
3578 struct osd_scrub *scrub = &dev->od_scrub;
3579 struct lu_fid *fid = &oic->oic_fid;
3580 struct osd_inode_id *id = &oti->oti_id;
3585 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3589 rc = osd_oi_lookup(oti, dev, fid, id, true);
3590 if (rc != 0 && rc != -ENOENT)
3593 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3596 if (thread_is_running(&scrub->os_thread)) {
3597 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3598 /* There is race condition between osd_oi_lookup and OI scrub.
3599 * The OI scrub finished just after osd_oi_lookup() failure.
3600 * Under such case, it is unnecessary to trigger OI scrub again,
3601 * but try to call osd_oi_lookup() again. */
3602 if (unlikely(rc == -EAGAIN))
3608 if (!dev->od_noscrub && ++once == 1) {
3609 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3611 rc = osd_scrub_start(dev);
3612 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3614 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3623 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3624 struct osd_device *dev,
3625 struct osd_idmap_cache *oic,
3626 struct lu_fid *fid, __u32 ino)
3628 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
3629 struct inode *inode;
3632 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3633 inode = osd_iget(oti, dev, &oic->oic_lid);
3634 if (IS_ERR(inode)) {
3635 fid_zero(&oic->oic_fid);
3636 return PTR_ERR(inode);
3639 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3642 fid_zero(&oic->oic_fid);
3644 *fid = oic->oic_fid = lma->lma_self_fid;
3649 * Calls ->lookup() to find dentry. From dentry get inode and
3650 * read inode's ea to get fid. This is required for interoperability
3653 * \retval 0, on success
3654 * \retval -ve, on error
3656 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3657 struct dt_rec *rec, const struct dt_key *key)
3659 struct inode *dir = obj->oo_inode;
3660 struct dentry *dentry;
3661 struct ldiskfs_dir_entry_2 *de;
3662 struct buffer_head *bh;
3663 struct lu_fid *fid = (struct lu_fid *) rec;
3664 struct htree_lock *hlock = NULL;
3669 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3671 dentry = osd_child_dentry_get(env, obj,
3672 (char *)key, strlen((char *)key));
3674 if (obj->oo_hl_head != NULL) {
3675 hlock = osd_oti_get(env)->oti_hlock;
3676 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3677 dir, LDISKFS_HLOCK_LOOKUP);
3679 down_read(&obj->oo_ext_idx_sem);
3682 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3684 struct osd_thread_info *oti = osd_oti_get(env);
3685 struct osd_inode_id *id = &oti->oti_id;
3686 struct osd_idmap_cache *oic = &oti->oti_cache;
3687 struct osd_device *dev = osd_obj2dev(obj);
3688 struct osd_scrub *scrub = &dev->od_scrub;
3689 struct scrub_file *sf = &scrub->os_file;
3691 ino = le32_to_cpu(de->inode);
3692 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
3694 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
3698 rc = osd_get_fid_from_dentry(de, rec);
3700 /* done with de, release bh */
3703 rc = osd_ea_fid_get(env, obj, ino, fid, id);
3705 osd_id_gen(id, ino, OSD_OII_NOGEN);
3706 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
3707 fid_zero(&oic->oic_fid);
3712 oic->oic_fid = *fid;
3713 if ((scrub->os_pos_current <= ino) &&
3714 ((sf->sf_flags & SF_INCONSISTENT) ||
3715 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
3716 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3718 osd_consistency_check(oti, dev, oic);
3727 ldiskfs_htree_unlock(hlock);
3729 up_read(&obj->oo_ext_idx_sem);
3734 * Find the osd object for given fid.
3736 * \param fid need to find the osd object having this fid
3738 * \retval osd_object on success
3739 * \retval -ve on error
3741 struct osd_object *osd_object_find(const struct lu_env *env,
3742 struct dt_object *dt,
3743 const struct lu_fid *fid)
3745 struct lu_device *ludev = dt->do_lu.lo_dev;
3746 struct osd_object *child = NULL;
3747 struct lu_object *luch;
3748 struct lu_object *lo;
3751 * at this point topdev might not exist yet
3752 * (i.e. MGS is preparing profiles). so we can
3753 * not rely on topdev and instead lookup with
3754 * our device passed as topdev. this can't work
3755 * if the object isn't cached yet (as osd doesn't
3756 * allocate lu_header). IOW, the object must be
3757 * in the cache, otherwise lu_object_alloc() crashes
3760 luch = lu_object_find_at(env, ludev, fid, NULL);
3761 if (!IS_ERR(luch)) {
3762 if (lu_object_exists(luch)) {
3763 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3765 child = osd_obj(lo);
3767 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3768 "lu_object can't be located"
3769 DFID"\n", PFID(fid));
3771 if (child == NULL) {
3772 lu_object_put(env, luch);
3773 CERROR("Unable to get osd_object\n");
3774 child = ERR_PTR(-ENOENT);
3777 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3778 "lu_object does not exists "DFID"\n",
3780 lu_object_put(env, luch);
3781 child = ERR_PTR(-ENOENT);
3784 child = (void *)luch;
3790 * Put the osd object once done with it.
3792 * \param obj osd object that needs to be put
3794 static inline void osd_object_put(const struct lu_env *env,
3795 struct osd_object *obj)
3797 lu_object_put(env, &obj->oo_dt.do_lu);
3800 static int osd_index_declare_ea_insert(const struct lu_env *env,
3801 struct dt_object *dt,
3802 const struct dt_rec *rec,
3803 const struct dt_key *key,
3804 struct thandle *handle)
3806 struct osd_thandle *oh;
3807 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3808 struct lu_fid *fid = (struct lu_fid *)rec;
3812 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3813 LASSERT(handle != NULL);
3815 oh = container_of0(handle, struct osd_thandle, ot_super);
3816 LASSERT(oh->ot_handle == NULL);
3818 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3819 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3821 if (osd_dt_obj(dt)->oo_inode == NULL) {
3822 const char *name = (const char *)key;
3823 /* Object is not being created yet. Only happens when
3824 * 1. declare directory create
3825 * 2. declare insert .
3826 * 3. declare insert ..
3828 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
3830 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3832 /* We ignore block quota on meta pool (MDTs), so needn't
3833 * calculate how many blocks will be consumed by this index
3835 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
3836 oh, true, true, NULL, false);
3842 rc = osd_remote_fid(env, osd, fid);
3848 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3849 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
3850 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3851 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3852 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3853 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3859 * Index add function for interoperability mode (b11826).
3860 * It will add the directory entry.This entry is needed to
3861 * maintain name->fid mapping.
3863 * \param key it is key i.e. file entry to be inserted
3864 * \param rec it is value of given key i.e. fid
3866 * \retval 0, on success
3867 * \retval -ve, on error
3869 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3870 const struct dt_rec *rec,
3871 const struct dt_key *key, struct thandle *th,
3872 struct lustre_capa *capa, int ignore_quota)
3874 struct osd_object *obj = osd_dt_obj(dt);
3875 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3876 struct lu_fid *fid = (struct lu_fid *) rec;
3877 const char *name = (const char *)key;
3878 struct osd_thread_info *oti = osd_oti_get(env);
3879 struct osd_inode_id *id = &oti->oti_id;
3880 struct inode *child_inode = NULL;
3881 struct osd_object *child = NULL;
3885 LASSERT(osd_invariant(obj));
3886 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3887 LASSERT(th != NULL);
3889 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3891 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3894 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
3896 rc = osd_remote_fid(env, osd, fid);
3898 CERROR("%s: Can not find object "DFID" rc %d\n",
3899 osd_name(osd), PFID(fid), rc);
3904 /* Insert remote entry */
3905 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
3906 struct osd_mdobj_map *omm = osd->od_mdt_map;
3907 struct osd_thandle *oh;
3909 /* If parent on remote MDT, we need put this object
3911 oh = container_of(th, typeof(*oh), ot_super);
3912 rc = osd_add_to_agent(env, osd, obj, oh);
3914 CERROR("%s: add agent "DFID" error: rc = %d\n",
3916 PFID(lu_object_fid(&dt->do_lu)), rc);
3920 child_inode = igrab(omm->omm_agent_dentry->d_inode);
3922 child_inode = osd_create_remote_inode(env, osd, obj,
3924 if (IS_ERR(child_inode))
3925 RETURN(PTR_ERR(child_inode));
3928 /* Insert local entry */
3929 child = osd_object_find(env, dt, fid);
3930 if (IS_ERR(child)) {
3931 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
3932 osd_name(osd), PFID(fid),
3933 id->oii_ino, id->oii_gen,
3934 (int)PTR_ERR(child_inode));
3935 RETURN(PTR_ERR(child_inode));
3937 child_inode = igrab(child->oo_inode);
3940 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
3944 osd_object_put(env, child);
3945 LASSERT(osd_invariant(obj));
3950 * Initialize osd Iterator for given osd index object.
3952 * \param dt osd index object
3955 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3956 struct dt_object *dt,
3958 struct lustre_capa *capa)
3960 struct osd_it_iam *it;
3961 struct osd_thread_info *oti = osd_oti_get(env);
3962 struct osd_object *obj = osd_dt_obj(dt);
3963 struct lu_object *lo = &dt->do_lu;
3964 struct iam_path_descr *ipd;
3965 struct iam_container *bag = &obj->oo_dir->od_container;
3967 LASSERT(lu_object_exists(lo));
3969 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3970 return ERR_PTR(-EACCES);
3973 ipd = osd_it_ipd_get(env, bag);
3974 if (likely(ipd != NULL)) {
3978 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3979 return (struct dt_it *)it;
3981 return ERR_PTR(-ENOMEM);
3985 * free given Iterator.
3988 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3990 struct osd_it_iam *it = (struct osd_it_iam *)di;
3991 struct osd_object *obj = it->oi_obj;
3993 iam_it_fini(&it->oi_it);
3994 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3995 lu_object_put(env, &obj->oo_dt.do_lu);
3999 * Move Iterator to record specified by \a key
4001 * \param di osd iterator
4002 * \param key key for index
4004 * \retval +ve di points to record with least key not larger than key
4005 * \retval 0 di points to exact matched key
4006 * \retval -ve failure
4009 static int osd_it_iam_get(const struct lu_env *env,
4010 struct dt_it *di, const struct dt_key *key)
4012 struct osd_thread_info *oti = osd_oti_get(env);
4013 struct osd_it_iam *it = (struct osd_it_iam *)di;
4015 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4016 /* swab quota uid/gid */
4017 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4018 key = (struct dt_key *)&oti->oti_quota_id;
4021 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4027 * \param di osd iterator
4029 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4031 struct osd_it_iam *it = (struct osd_it_iam *)di;
4033 iam_it_put(&it->oi_it);
4037 * Move iterator by one record
4039 * \param di osd iterator
4041 * \retval +1 end of container reached
4043 * \retval -ve failure
4046 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4048 struct osd_it_iam *it = (struct osd_it_iam *)di;
4050 return iam_it_next(&it->oi_it);
4054 * Return pointer to the key under iterator.
4057 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4058 const struct dt_it *di)
4060 struct osd_thread_info *oti = osd_oti_get(env);
4061 struct osd_it_iam *it = (struct osd_it_iam *)di;
4062 struct osd_object *obj = it->oi_obj;
4065 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4067 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4068 /* swab quota uid/gid */
4069 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4070 key = (struct dt_key *)&oti->oti_quota_id;
4077 * Return size of key under iterator (in bytes)
4080 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4082 struct osd_it_iam *it = (struct osd_it_iam *)di;
4084 return iam_it_key_size(&it->oi_it);
4088 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4090 /* check if file type is required */
4091 if (ent->lde_attrs & LUDA_TYPE) {
4092 int align = sizeof(struct luda_type) - 1;
4093 struct luda_type *lt;
4095 len = (len + align) & ~align;
4096 lt = (struct luda_type *)(ent->lde_name + len);
4097 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
4100 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4104 * build lu direct from backend fs dirent.
4108 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4109 char *name, __u16 namelen, __u16 type, __u32 attr)
4111 ent->lde_attrs = attr | LUDA_FID;
4112 fid_cpu_to_le(&ent->lde_fid, fid);
4114 ent->lde_hash = cpu_to_le64(offset);
4115 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4117 strncpy(ent->lde_name, name, namelen);
4118 ent->lde_namelen = cpu_to_le16(namelen);
4120 /* append lustre attributes */
4121 osd_it_append_attrs(ent, namelen, type);
4125 * Return pointer to the record under iterator.
4127 static int osd_it_iam_rec(const struct lu_env *env,
4128 const struct dt_it *di,
4129 struct dt_rec *dtrec, __u32 attr)
4131 struct osd_it_iam *it = (struct osd_it_iam *)di;
4132 struct osd_thread_info *info = osd_oti_get(env);
4135 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4136 const struct osd_fid_pack *rec;
4137 struct lu_fid *fid = &info->oti_fid;
4138 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4144 name = (char *)iam_it_key_get(&it->oi_it);
4146 RETURN(PTR_ERR(name));
4148 namelen = iam_it_key_size(&it->oi_it);
4150 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4152 RETURN(PTR_ERR(rec));
4154 rc = osd_fid_unpack(fid, rec);
4158 hash = iam_it_store(&it->oi_it);
4160 /* IAM does not store object type in IAM index (dir) */
4161 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4163 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4164 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4165 (struct iam_rec *)dtrec);
4166 osd_quota_unpack(it->oi_obj, dtrec);
4168 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4169 (struct iam_rec *)dtrec);
4176 * Returns cookie for current Iterator position.
4178 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4180 struct osd_it_iam *it = (struct osd_it_iam *)di;
4182 return iam_it_store(&it->oi_it);
4186 * Restore iterator from cookie.
4188 * \param di osd iterator
4189 * \param hash Iterator location cookie
4191 * \retval +ve di points to record with least key not larger than key.
4192 * \retval 0 di points to exact matched key
4193 * \retval -ve failure
4196 static int osd_it_iam_load(const struct lu_env *env,
4197 const struct dt_it *di, __u64 hash)
4199 struct osd_it_iam *it = (struct osd_it_iam *)di;
4201 return iam_it_load(&it->oi_it, hash);
4204 static const struct dt_index_operations osd_index_iam_ops = {
4205 .dio_lookup = osd_index_iam_lookup,
4206 .dio_declare_insert = osd_index_declare_iam_insert,
4207 .dio_insert = osd_index_iam_insert,
4208 .dio_declare_delete = osd_index_declare_iam_delete,
4209 .dio_delete = osd_index_iam_delete,
4211 .init = osd_it_iam_init,
4212 .fini = osd_it_iam_fini,
4213 .get = osd_it_iam_get,
4214 .put = osd_it_iam_put,
4215 .next = osd_it_iam_next,
4216 .key = osd_it_iam_key,
4217 .key_size = osd_it_iam_key_size,
4218 .rec = osd_it_iam_rec,
4219 .store = osd_it_iam_store,
4220 .load = osd_it_iam_load
4226 * Creates or initializes iterator context.
4228 * \retval struct osd_it_ea, iterator structure on success
4231 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4232 struct dt_object *dt,
4234 struct lustre_capa *capa)
4236 struct osd_object *obj = osd_dt_obj(dt);
4237 struct osd_thread_info *info = osd_oti_get(env);
4238 struct osd_it_ea *it = &info->oti_it_ea;
4239 struct lu_object *lo = &dt->do_lu;
4240 struct dentry *obj_dentry = &info->oti_it_dentry;
4242 LASSERT(lu_object_exists(lo));
4244 obj_dentry->d_inode = obj->oo_inode;
4245 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4246 obj_dentry->d_name.hash = 0;
4248 it->oie_rd_dirent = 0;
4249 it->oie_it_dirent = 0;
4250 it->oie_dirent = NULL;
4251 it->oie_buf = info->oti_it_ea_buf;
4253 it->oie_file.f_pos = 0;
4254 it->oie_file.f_dentry = obj_dentry;
4255 if (attr & LUDA_64BITHASH)
4256 it->oie_file.f_mode |= FMODE_64BITHASH;
4258 it->oie_file.f_mode |= FMODE_32BITHASH;
4259 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
4260 it->oie_file.f_op = obj->oo_inode->i_fop;
4261 it->oie_file.private_data = NULL;
4263 RETURN((struct dt_it *) it);
4267 * Destroy or finishes iterator context.
4269 * \param di iterator structure to be destroyed
4271 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4273 struct osd_it_ea *it = (struct osd_it_ea *)di;
4274 struct osd_object *obj = it->oie_obj;
4275 struct inode *inode = obj->oo_inode;
4278 it->oie_file.f_op->release(inode, &it->oie_file);
4279 lu_object_put(env, &obj->oo_dt.do_lu);
4284 * It position the iterator at given key, so that next lookup continues from
4285 * that key Or it is similar to dio_it->load() but based on a key,
4286 * rather than file position.
4288 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4291 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4293 static int osd_it_ea_get(const struct lu_env *env,
4294 struct dt_it *di, const struct dt_key *key)
4296 struct osd_it_ea *it = (struct osd_it_ea *)di;
4299 LASSERT(((const char *)key)[0] == '\0');
4300 it->oie_file.f_pos = 0;
4301 it->oie_rd_dirent = 0;
4302 it->oie_it_dirent = 0;
4303 it->oie_dirent = NULL;
4311 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4316 * It is called internally by ->readdir(). It fills the
4317 * iterator's in-memory data structure with required
4318 * information i.e. name, namelen, rec_size etc.
4320 * \param buf in which information to be filled in.
4321 * \param name name of the file in given dir
4323 * \retval 0 on success
4324 * \retval 1 on buffer full
4326 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4327 loff_t offset, __u64 ino,
4330 struct osd_it_ea *it = (struct osd_it_ea *)buf;
4331 struct osd_object *obj = it->oie_obj;
4332 struct osd_it_ea_dirent *ent = it->oie_dirent;
4333 struct lu_fid *fid = &ent->oied_fid;
4334 struct osd_fid_pack *rec;
4337 /* this should never happen */
4338 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4339 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4343 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4347 /* "." is just the object itself. */
4348 if (namelen == 1 && name[0] == '.') {
4349 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4350 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4351 rec = (struct osd_fid_pack*) (name + namelen + 1);
4352 if (osd_fid_unpack(fid, rec) != 0)
4357 d_type &= ~LDISKFS_DIRENT_LUFID;
4359 /* NOT export local root. */
4360 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4361 ino = obj->oo_inode->i_ino;
4362 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4365 ent->oied_ino = ino;
4366 ent->oied_off = offset;
4367 ent->oied_namelen = namelen;
4368 ent->oied_type = d_type;
4370 memcpy(ent->oied_name, name, namelen);
4372 it->oie_rd_dirent++;
4373 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4378 * Calls ->readdir() to load a directory entry at a time
4379 * and stored it in iterator's in-memory data structure.
4381 * \param di iterator's in memory structure
4383 * \retval 0 on success
4384 * \retval -ve on error
4386 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4387 const struct dt_it *di)
4389 struct osd_it_ea *it = (struct osd_it_ea *)di;
4390 struct osd_object *obj = it->oie_obj;
4391 struct inode *inode = obj->oo_inode;
4392 struct htree_lock *hlock = NULL;
4396 it->oie_dirent = it->oie_buf;
4397 it->oie_rd_dirent = 0;
4399 if (obj->oo_hl_head != NULL) {
4400 hlock = osd_oti_get(env)->oti_hlock;
4401 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4402 inode, LDISKFS_HLOCK_READDIR);
4404 down_read(&obj->oo_ext_idx_sem);
4407 result = inode->i_fop->readdir(&it->oie_file, it,
4408 (filldir_t) osd_ldiskfs_filldir);
4411 ldiskfs_htree_unlock(hlock);
4413 up_read(&obj->oo_ext_idx_sem);
4415 if (it->oie_rd_dirent == 0) {
4418 it->oie_dirent = it->oie_buf;
4419 it->oie_it_dirent = 1;
4426 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4427 * to load a directory entry at a time and stored it in
4428 * iterator's in-memory data structure.
4430 * \param di iterator's in memory structure
4432 * \retval +ve iterator reached to end
4433 * \retval 0 iterator not reached to end
4434 * \retval -ve on error
4436 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4438 struct osd_it_ea *it = (struct osd_it_ea *)di;
4443 if (it->oie_it_dirent < it->oie_rd_dirent) {
4445 (void *) it->oie_dirent +
4446 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4447 it->oie_dirent->oied_namelen);
4448 it->oie_it_dirent++;
4451 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4454 rc = osd_ldiskfs_it_fill(env, di);
4461 * Returns the key at current position from iterator's in memory structure.
4463 * \param di iterator's in memory structure
4465 * \retval key i.e. struct dt_key on success
4467 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4468 const struct dt_it *di)
4470 struct osd_it_ea *it = (struct osd_it_ea *)di;
4472 return (struct dt_key *)it->oie_dirent->oied_name;
4476 * Returns the key's size at current position from iterator's in memory structure.
4478 * \param di iterator's in memory structure
4480 * \retval key_size i.e. struct dt_key on success
4482 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4484 struct osd_it_ea *it = (struct osd_it_ea *)di;
4486 return it->oie_dirent->oied_namelen;
4490 osd_dirent_update(handle_t *jh, struct super_block *sb,
4491 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4492 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4494 struct osd_fid_pack *rec;
4498 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4499 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4501 rc = ldiskfs_journal_get_write_access(jh, bh);
4503 CERROR("%.16s: fail to write access for update dirent: "
4504 "name = %.*s, rc = %d\n",
4505 LDISKFS_SB(sb)->s_es->s_volume_name,
4506 ent->oied_namelen, ent->oied_name, rc);
4510 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4511 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4512 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4514 CERROR("%.16s: fail to dirty metadata for update dirent: "
4515 "name = %.*s, rc = %d\n",
4516 LDISKFS_SB(sb)->s_es->s_volume_name,
4517 ent->oied_namelen, ent->oied_name, rc);
4523 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4525 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4526 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4533 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
4534 struct inode *dir, struct inode *inode,
4535 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4536 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
4537 struct htree_lock *hlock)
4539 struct dentry *dentry;
4540 struct osd_fid_pack *rec;
4541 struct ldiskfs_dentry_param *ldp;
4545 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
4546 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
4549 /* There is enough space to hold the FID-in-dirent. */
4550 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
4551 dir->i_sb->s_blocksize)) {
4552 rc = ldiskfs_journal_get_write_access(jh, bh);
4554 CERROR("%.16s: fail to write access for reinsert "
4555 "dirent: name = %.*s, rc = %d\n",
4556 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4557 ent->oied_namelen, ent->oied_name, rc);
4561 de->name[de->name_len] = 0;
4562 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4563 rec->fp_len = sizeof(struct lu_fid) + 1;
4564 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4565 de->file_type |= LDISKFS_DIRENT_LUFID;
4567 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4569 CERROR("%.16s: fail to dirty metadata for reinsert "
4570 "dirent: name = %.*s, rc = %d\n",
4571 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4572 ent->oied_namelen, ent->oied_name, rc);
4577 rc = ldiskfs_delete_entry(jh, dir, de, bh);
4579 CERROR("%.16s: fail to delete entry for reinsert dirent: "
4580 "name = %.*s, rc = %d\n",
4581 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4582 ent->oied_namelen, ent->oied_name, rc);
4586 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
4588 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
4589 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
4590 dentry->d_fsdata = (void *)ldp;
4591 ll_vfs_dq_init(dir);
4592 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
4593 /* It is too bad, we cannot reinsert the name entry back.
4594 * That means we lose it! */
4596 CERROR("%.16s: fail to insert entry for reinsert dirent: "
4597 "name = %.*s, rc = %d\n",
4598 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4599 ent->oied_namelen, ent->oied_name, rc);
4605 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
4606 struct osd_it_ea *it, struct lu_fid *fid,
4607 struct osd_inode_id *id, __u32 *attr)
4609 struct osd_thread_info *info = osd_oti_get(env);
4610 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
4611 struct osd_device *dev = osd_obj2dev(obj);
4612 struct super_block *sb = osd_sb(dev);
4613 const char *devname =
4614 LDISKFS_SB(sb)->s_es->s_volume_name;
4615 struct osd_it_ea_dirent *ent = it->oie_dirent;
4616 struct inode *dir = obj->oo_inode;
4617 struct htree_lock *hlock = NULL;
4618 struct buffer_head *bh = NULL;
4619 handle_t *jh = NULL;
4620 struct ldiskfs_dir_entry_2 *de;
4621 struct dentry *dentry;
4622 struct inode *inode;
4626 bool is_dotdot = false;
4629 if (ent->oied_name[0] == '.') {
4630 /* Skip dot entry, even if it has stale FID-in-dirent, because
4631 * we do not use such FID-in-dirent anymore, it is harmless. */
4632 if (ent->oied_namelen == 1)
4635 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
4639 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
4642 /* We need to ensure that the name entry is still valid.
4643 * Because it may be removed or renamed by other already.
4645 * The unlink or rename operation will start journal before PDO lock,
4646 * so to avoid deadlock, here we need to start journal handle before
4647 * related PDO lock also. But because we do not know whether there
4648 * will be something to be repaired before PDO lock, we just start
4649 * journal without conditions.
4651 * We may need to remove the name entry firstly, then insert back.
4652 * One credit is for user quota file update.
4653 * One credit is for group quota file update.
4654 * Two credits are for dirty inode. */
4655 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4656 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
4659 if (dev->od_dirent_journal) {
4660 jh = ldiskfs_journal_start_sb(sb, credits);
4663 CERROR("%.16s: fail to start trans for dirent "
4664 "check_repair: credits %d, name %.*s, rc %d\n",
4665 devname, credits, ent->oied_namelen,
4666 ent->oied_name, rc);
4671 if (obj->oo_hl_head != NULL) {
4672 hlock = osd_oti_get(env)->oti_hlock;
4673 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
4676 down_write(&obj->oo_ext_idx_sem);
4679 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
4680 /* For dotdot entry, if there is not enough space to hold FID-in-dirent,
4681 * just keep it there. It only happens when the device upgraded from 1.8
4682 * or restored from MDT file-level backup. For the whole directory, only
4683 * dotdot entry has no FID-in-dirent and needs to get FID from LMA when
4684 * readdir, it will not affect the performance much. */
4685 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
4686 (is_dotdot && !osd_dirent_has_space(de->rec_len,
4688 sb->s_blocksize))) {
4689 *attr |= LUDA_IGNORE;
4690 GOTO(out_journal, rc = 0);
4693 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
4694 inode = osd_iget(info, dev, id);
4695 if (IS_ERR(inode)) {
4696 rc = PTR_ERR(inode);
4697 if (rc == -ENOENT || rc == -ESTALE) {
4698 *attr |= LUDA_IGNORE;
4702 GOTO(out_journal, rc);
4705 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
4707 if (fid_is_sane(fid)) {
4708 /* FID-in-dirent is valid. */
4709 if (lu_fid_eq(fid, &lma->lma_self_fid))
4710 GOTO(out_inode, rc = 0);
4712 /* Do not repair under dryrun mode. */
4713 if (*attr & LUDA_VERIFY_DRYRUN) {
4714 *attr |= LUDA_REPAIR;
4715 GOTO(out_inode, rc = 0);
4718 if (!dev->od_dirent_journal) {
4722 ldiskfs_htree_unlock(hlock);
4724 up_write(&obj->oo_ext_idx_sem);
4725 dev->od_dirent_journal = 1;
4729 *fid = lma->lma_self_fid;
4731 /* Update the FID-in-dirent. */
4732 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
4734 *attr |= LUDA_REPAIR;
4736 /* Do not repair under dryrun mode. */
4737 if (*attr & LUDA_VERIFY_DRYRUN) {
4738 *attr |= LUDA_REPAIR;
4739 GOTO(out_inode, rc = 0);
4742 if (!dev->od_dirent_journal) {
4746 ldiskfs_htree_unlock(hlock);
4748 up_write(&obj->oo_ext_idx_sem);
4749 dev->od_dirent_journal = 1;
4753 *fid = lma->lma_self_fid;
4755 /* Append the FID-in-dirent. */
4756 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4757 fid, bh, de, hlock);
4759 *attr |= LUDA_REPAIR;
4761 } else if (rc == -ENODATA) {
4762 /* Do not repair under dryrun mode. */
4763 if (*attr & LUDA_VERIFY_DRYRUN) {
4764 if (fid_is_sane(fid))
4765 *attr |= LUDA_REPAIR;
4767 *attr |= LUDA_UPGRADE;
4768 GOTO(out_inode, rc = 0);
4771 if (!dev->od_dirent_journal) {
4775 ldiskfs_htree_unlock(hlock);
4777 up_write(&obj->oo_ext_idx_sem);
4778 dev->od_dirent_journal = 1;
4783 if (unlikely(fid_is_sane(fid))) {
4784 /* FID-in-dirent exists, but FID-in-LMA is lost.
4785 * Trust the FID-in-dirent, and add FID-in-LMA. */
4786 rc = osd_ea_fid_set(info, inode, fid);
4788 *attr |= LUDA_REPAIR;
4790 lu_igif_build(fid, inode->i_ino, inode->i_generation);
4791 /* It is probably IGIF object. Only aappend the
4792 * FID-in-dirent. OI scrub will process FID-in-LMA. */
4793 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4794 fid, bh, de, hlock);
4796 *attr |= LUDA_UPGRADE;
4800 GOTO(out_inode, rc);
4808 ldiskfs_htree_unlock(hlock);
4810 up_write(&obj->oo_ext_idx_sem);
4812 ldiskfs_journal_stop(jh);
4813 if (rc >= 0 && !dirty)
4814 dev->od_dirent_journal = 0;
4819 * Returns the value at current position from iterator's in memory structure.
4821 * \param di struct osd_it_ea, iterator's in memory structure
4822 * \param attr attr requested for dirent.
4823 * \param lde lustre dirent
4825 * \retval 0 no error and \param lde has correct lustre dirent.
4826 * \retval -ve on error
4828 static inline int osd_it_ea_rec(const struct lu_env *env,
4829 const struct dt_it *di,
4830 struct dt_rec *dtrec, __u32 attr)
4832 struct osd_it_ea *it = (struct osd_it_ea *)di;
4833 struct osd_object *obj = it->oie_obj;
4834 struct osd_device *dev = osd_obj2dev(obj);
4835 struct osd_scrub *scrub = &dev->od_scrub;
4836 struct scrub_file *sf = &scrub->os_file;
4837 struct osd_thread_info *oti = osd_oti_get(env);
4838 struct osd_inode_id *id = &oti->oti_id;
4839 struct osd_idmap_cache *oic = &oti->oti_cache;
4840 struct lu_fid *fid = &it->oie_dirent->oied_fid;
4841 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4842 __u32 ino = it->oie_dirent->oied_ino;
4846 if (attr & LUDA_VERIFY) {
4848 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
4849 attr |= LUDA_IGNORE;
4854 rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr);
4856 attr &= ~LU_DIRENT_ATTRS_MASK;
4857 if (!fid_is_sane(fid)) {
4858 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
4861 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4863 osd_id_gen(id, ino, OSD_OII_NOGEN);
4871 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4872 it->oie_dirent->oied_name,
4873 it->oie_dirent->oied_namelen,
4874 it->oie_dirent->oied_type, attr);
4876 if (osd_remote_fid(env, dev, fid))
4879 if (likely(!(attr & LUDA_IGNORE))) {
4881 oic->oic_fid = *fid;
4884 if (!(attr & LUDA_VERIFY) &&
4885 (scrub->os_pos_current <= ino) &&
4886 ((sf->sf_flags & SF_INCONSISTENT) ||
4887 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4888 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4889 osd_consistency_check(oti, dev, oic);
4895 * Returns a cookie for current position of the iterator head, so that
4896 * user can use this cookie to load/start the iterator next time.
4898 * \param di iterator's in memory structure
4900 * \retval cookie for current position, on success
4902 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4904 struct osd_it_ea *it = (struct osd_it_ea *)di;
4906 return it->oie_dirent->oied_off;
4910 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4911 * to load a directory entry at a time and stored it i inn,
4912 * in iterator's in-memory data structure.
4914 * \param di struct osd_it_ea, iterator's in memory structure
4916 * \retval +ve on success
4917 * \retval -ve on error
4919 static int osd_it_ea_load(const struct lu_env *env,
4920 const struct dt_it *di, __u64 hash)
4922 struct osd_it_ea *it = (struct osd_it_ea *)di;
4926 it->oie_file.f_pos = hash;
4928 rc = osd_ldiskfs_it_fill(env, di);
4936 * Index lookup function for interoperability mode (b11826).
4938 * \param key, key i.e. file name to be searched
4940 * \retval +ve, on success
4941 * \retval -ve, on error
4943 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4944 struct dt_rec *rec, const struct dt_key *key,
4945 struct lustre_capa *capa)
4947 struct osd_object *obj = osd_dt_obj(dt);
4952 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4953 LINVRNT(osd_invariant(obj));
4955 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4958 rc = osd_ea_lookup_rec(env, obj, rec, key);
4965 * Index and Iterator operations for interoperability
4966 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4968 static const struct dt_index_operations osd_index_ea_ops = {
4969 .dio_lookup = osd_index_ea_lookup,
4970 .dio_declare_insert = osd_index_declare_ea_insert,
4971 .dio_insert = osd_index_ea_insert,
4972 .dio_declare_delete = osd_index_declare_ea_delete,
4973 .dio_delete = osd_index_ea_delete,
4975 .init = osd_it_ea_init,
4976 .fini = osd_it_ea_fini,
4977 .get = osd_it_ea_get,
4978 .put = osd_it_ea_put,
4979 .next = osd_it_ea_next,
4980 .key = osd_it_ea_key,
4981 .key_size = osd_it_ea_key_size,
4982 .rec = osd_it_ea_rec,
4983 .store = osd_it_ea_store,
4984 .load = osd_it_ea_load
4988 static void *osd_key_init(const struct lu_context *ctx,
4989 struct lu_context_key *key)
4991 struct osd_thread_info *info;
4993 OBD_ALLOC_PTR(info);
4995 return ERR_PTR(-ENOMEM);
4997 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4998 if (info->oti_it_ea_buf == NULL)
5001 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5003 info->oti_hlock = ldiskfs_htree_lock_alloc();
5004 if (info->oti_hlock == NULL)
5010 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5013 return ERR_PTR(-ENOMEM);
5016 static void osd_key_fini(const struct lu_context *ctx,
5017 struct lu_context_key *key, void* data)
5019 struct osd_thread_info *info = data;
5021 if (info->oti_hlock != NULL)
5022 ldiskfs_htree_lock_free(info->oti_hlock);
5023 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5027 static void osd_key_exit(const struct lu_context *ctx,
5028 struct lu_context_key *key, void *data)
5030 struct osd_thread_info *info = data;
5032 LASSERT(info->oti_r_locks == 0);
5033 LASSERT(info->oti_w_locks == 0);
5034 LASSERT(info->oti_txns == 0);
5037 /* type constructor/destructor: osd_type_init, osd_type_fini */
5038 LU_TYPE_INIT_FINI(osd, &osd_key);
5040 struct lu_context_key osd_key = {
5041 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5042 .lct_init = osd_key_init,
5043 .lct_fini = osd_key_fini,
5044 .lct_exit = osd_key_exit
5048 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5049 const char *name, struct lu_device *next)
5051 struct osd_device *osd = osd_dev(d);
5053 strncpy(osd->od_svname, name, MAX_OBD_NAME);
5054 return osd_procfs_init(osd, name);
5057 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5061 osd_scrub_cleanup(env, o);
5064 fsfilt_put_ops(o->od_fsops);
5068 /* shutdown quota slave instance associated with the device */
5069 if (o->od_quota_slave != NULL) {
5070 qsd_fini(env, o->od_quota_slave);
5071 o->od_quota_slave = NULL;
5077 static int osd_mount(const struct lu_env *env,
5078 struct osd_device *o, struct lustre_cfg *cfg)
5080 const char *name = lustre_cfg_string(cfg, 0);
5081 const char *dev = lustre_cfg_string(cfg, 1);
5083 unsigned long page, s_flags, lmd_flags = 0;
5084 struct page *__page;
5085 struct file_system_type *type;
5086 char *options = NULL;
5091 if (o->od_mnt != NULL)
5094 if (strlen(dev) >= sizeof(o->od_mntdev))
5096 strcpy(o->od_mntdev, dev);
5098 o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
5099 if (o->od_fsops == NULL) {
5100 CERROR("Can't find fsfilt_ldiskfs\n");
5104 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
5106 GOTO(out, rc = -ENOMEM);
5108 str = lustre_cfg_string(cfg, 2);
5109 s_flags = simple_strtoul(str, NULL, 0);
5110 str = strstr(str, ":");
5112 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5113 opts = lustre_cfg_string(cfg, 3);
5114 page = (unsigned long)cfs_page_address(__page);
5115 options = (char *)page;
5118 strcat(options, "user_xattr,acl");
5120 strcat(options, opts);
5122 /* Glom up mount options */
5123 if (*options != '\0')
5124 strcat(options, ",");
5125 strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
5127 type = get_fs_type("ldiskfs");
5129 CERROR("%s: cannot find ldiskfs module\n", name);
5130 GOTO(out, rc = -ENODEV);
5133 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5134 cfs_module_put(type->owner);
5136 if (IS_ERR(o->od_mnt)) {
5137 rc = PTR_ERR(o->od_mnt);
5138 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5143 if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5144 CERROR("%s: underlying device %s is marked as read-only. "
5145 "Setup failed\n", name, dev);
5148 GOTO(out, rc = -EROFS);
5151 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5152 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5153 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5156 GOTO(out, rc = -EINVAL);
5159 ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
5160 LDISKFS_STATE_LUSTRE_NO_OI);
5161 if (lmd_flags & LMD_FLG_NOSCRUB)
5166 OBD_PAGE_FREE(__page);
5168 fsfilt_put_ops(o->od_fsops);
5173 static struct lu_device *osd_device_fini(const struct lu_env *env,
5174 struct lu_device *d)
5179 rc = osd_shutdown(env, osd_dev(d));
5181 osd_obj_map_fini(osd_dev(d));
5183 shrink_dcache_sb(osd_sb(osd_dev(d)));
5184 osd_sync(env, lu2dt_dev(d));
5186 rc = osd_procfs_fini(osd_dev(d));
5188 CERROR("proc fini error %d \n", rc);
5189 RETURN (ERR_PTR(rc));
5192 if (osd_dev(d)->od_mnt) {
5193 mntput(osd_dev(d)->od_mnt);
5194 osd_dev(d)->od_mnt = NULL;
5200 static int osd_device_init0(const struct lu_env *env,
5201 struct osd_device *o,
5202 struct lustre_cfg *cfg)
5204 struct lu_device *l = osd2lu_dev(o);
5205 struct osd_thread_info *info;
5208 /* if the module was re-loaded, env can loose its keys */
5209 rc = lu_env_refill((struct lu_env *) env);
5212 info = osd_oti_get(env);
5215 l->ld_ops = &osd_lu_ops;
5216 o->od_dt_dev.dd_ops = &osd_dt_ops;
5218 spin_lock_init(&o->od_osfs_lock);
5219 mutex_init(&o->od_otable_mutex);
5220 o->od_osfs_age = cfs_time_shift_64(-1000);
5222 o->od_capa_hash = init_capa_hash();
5223 if (o->od_capa_hash == NULL)
5224 GOTO(out, rc = -ENOMEM);
5226 o->od_read_cache = 1;
5227 o->od_writethrough_cache = 1;
5228 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5230 rc = osd_mount(env, o, cfg);
5234 CFS_INIT_LIST_HEAD(&o->od_ios_list);
5235 /* setup scrub, including OI files initialization */
5236 rc = osd_scrub_setup(env, o);
5240 strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
5241 sizeof(o->od_svname) - 1);
5243 rc = osd_obj_map_init(o);
5245 GOTO(out_scrub, rc);
5247 rc = lu_site_init(&o->od_site, l);
5249 GOTO(out_compat, rc);
5250 o->od_site.ls_bottom_dev = l;
5252 rc = lu_site_init_finish(&o->od_site);
5256 rc = osd_procfs_init(o, o->od_svname);
5258 CERROR("%s: can't initialize procfs: rc = %d\n",
5263 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5265 /* initialize quota slave instance */
5266 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5268 if (IS_ERR(o->od_quota_slave)) {
5269 rc = PTR_ERR(o->od_quota_slave);
5270 o->od_quota_slave = NULL;
5271 GOTO(out_procfs, rc);
5278 lu_site_fini(&o->od_site);
5280 osd_obj_map_fini(o);
5282 osd_scrub_cleanup(env, o);
5284 osd_oi_fini(info, o);
5285 osd_shutdown(env, o);
5289 cleanup_capa_hash(o->od_capa_hash);
5294 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5295 struct lu_device_type *t,
5296 struct lustre_cfg *cfg)
5298 struct osd_device *o;
5303 return ERR_PTR(-ENOMEM);
5305 rc = dt_device_init(&o->od_dt_dev, t);
5307 /* Because the ctx might be revived in dt_device_init,
5308 * refill the env here */
5309 lu_env_refill((struct lu_env *)env);
5310 rc = osd_device_init0(env, o, cfg);
5312 dt_device_fini(&o->od_dt_dev);
5315 if (unlikely(rc != 0))
5318 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5321 static struct lu_device *osd_device_free(const struct lu_env *env,
5322 struct lu_device *d)
5324 struct osd_device *o = osd_dev(d);
5327 cleanup_capa_hash(o->od_capa_hash);
5328 /* XXX: make osd top device in order to release reference */
5329 d->ld_site->ls_top_dev = d;
5330 lu_site_purge(env, d->ld_site, -1);
5331 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5332 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5333 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5335 lu_site_fini(&o->od_site);
5336 dt_device_fini(&o->od_dt_dev);
5341 static int osd_process_config(const struct lu_env *env,
5342 struct lu_device *d, struct lustre_cfg *cfg)
5344 struct osd_device *o = osd_dev(d);
5348 switch(cfg->lcfg_command) {
5350 err = osd_mount(env, o, cfg);
5353 lu_dev_del_linkage(d->ld_site, d);
5354 err = osd_shutdown(env, o);
5363 static int osd_recovery_complete(const struct lu_env *env,
5364 struct lu_device *d)
5366 struct osd_device *osd = osd_dev(d);
5370 if (osd->od_quota_slave == NULL)
5373 /* start qsd instance on recovery completion, this notifies the quota
5374 * slave code that we are about to process new requests now */
5375 rc = qsd_start(env, osd->od_quota_slave);
5380 * we use exports to track all osd users
5382 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5383 struct obd_device *obd, struct obd_uuid *cluuid,
5384 struct obd_connect_data *data, void *localdata)
5386 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5387 struct lustre_handle conn;
5391 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5393 rc = class_connect(&conn, obd, cluuid);
5397 *exp = class_conn2export(&conn);
5399 spin_lock(&osd->od_osfs_lock);
5401 spin_unlock(&osd->od_osfs_lock);
5407 * once last export (we don't count self-export) disappeared
5408 * osd can be released
5410 static int osd_obd_disconnect(struct obd_export *exp)
5412 struct obd_device *obd = exp->exp_obd;
5413 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5414 int rc, release = 0;
5417 /* Only disconnect the underlying layers on the final disconnect. */
5418 spin_lock(&osd->od_osfs_lock);
5420 if (osd->od_connects == 0)
5422 spin_unlock(&osd->od_osfs_lock);
5424 rc = class_disconnect(exp); /* bz 9811 */
5426 if (rc == 0 && release)
5427 class_manual_cleanup(obd);
5431 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5432 struct lu_device *dev)
5434 struct osd_device *osd = osd_dev(dev);
5438 if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
5439 /* MDT/MDD still use old infrastructure to create
5441 result = llo_local_objects_setup(env, lu2md_dev(pdev),
5447 if (osd->od_quota_slave != NULL)
5448 /* set up quota slave objects */
5449 result = qsd_prepare(env, osd->od_quota_slave);
5454 static const struct lu_object_operations osd_lu_obj_ops = {
5455 .loo_object_init = osd_object_init,
5456 .loo_object_delete = osd_object_delete,
5457 .loo_object_release = osd_object_release,
5458 .loo_object_free = osd_object_free,
5459 .loo_object_print = osd_object_print,
5460 .loo_object_invariant = osd_object_invariant
5463 const struct lu_device_operations osd_lu_ops = {
5464 .ldo_object_alloc = osd_object_alloc,
5465 .ldo_process_config = osd_process_config,
5466 .ldo_recovery_complete = osd_recovery_complete,
5467 .ldo_prepare = osd_prepare,
5470 static const struct lu_device_type_operations osd_device_type_ops = {
5471 .ldto_init = osd_type_init,
5472 .ldto_fini = osd_type_fini,
5474 .ldto_start = osd_type_start,
5475 .ldto_stop = osd_type_stop,
5477 .ldto_device_alloc = osd_device_alloc,
5478 .ldto_device_free = osd_device_free,
5480 .ldto_device_init = osd_device_init,
5481 .ldto_device_fini = osd_device_fini
5484 struct lu_device_type osd_device_type = {
5485 .ldt_tags = LU_DEVICE_DT,
5486 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
5487 .ldt_ops = &osd_device_type_ops,
5488 .ldt_ctx_tags = LCT_LOCAL,
5492 * lprocfs legacy support.
5494 static struct obd_ops osd_obd_device_ops = {
5495 .o_owner = THIS_MODULE,
5496 .o_connect = osd_obd_connect,
5497 .o_disconnect = osd_obd_disconnect
5500 static int __init osd_mod_init(void)
5502 struct lprocfs_static_vars lvars;
5505 lprocfs_osd_init_vars(&lvars);
5506 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
5507 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5510 static void __exit osd_mod_exit(void)
5512 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
5515 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5516 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
5517 MODULE_LICENSE("GPL");
5519 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);