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 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
174 struct dentry *dentry, struct lustre_mdt_attrs *lma)
178 dentry->d_inode = inode;
179 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
182 /* try with old lma size */
183 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
184 info->oti_mdt_attrs_old,
187 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
190 /* Check LMA compatibility */
191 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
192 CWARN("%.16s: unsupported incompat LMA feature(s) "
194 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
195 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
199 lustre_lma_swab(lma);
202 } else if (rc == 0) {
210 * retrieve object from backend ext fs.
212 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
213 struct osd_inode_id *id)
215 struct inode *inode = NULL;
217 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
219 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
220 id->oii_ino, PTR_ERR(inode));
221 } else if (id->oii_gen != OSD_OII_NOGEN &&
222 inode->i_generation != id->oii_gen) {
223 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
225 id->oii_ino, id->oii_gen, inode->i_generation);
227 inode = ERR_PTR(-ESTALE);
228 } else if (inode->i_nlink == 0) {
229 /* due to parallel readdir and unlink,
230 * we can have dead inode here. */
231 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
232 make_bad_inode(inode);
234 inode = ERR_PTR(-ESTALE);
235 } else if (is_bad_inode(inode)) {
236 CWARN("%.16s: bad inode: ino = %u\n",
237 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
239 inode = ERR_PTR(-ENOENT);
241 if (id->oii_gen == OSD_OII_NOGEN)
242 osd_id_gen(id, inode->i_ino, inode->i_generation);
244 /* Do not update file c/mtime in ldiskfs.
245 * NB: we don't have any lock to protect this because we don't
246 * have reference on osd_object now, but contention with
247 * another lookup + attr_set can't happen in the tiny window
248 * between if (...) and set S_NOCMTIME. */
249 if (!(inode->i_flags & S_NOCMTIME))
250 inode->i_flags |= S_NOCMTIME;
255 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
256 struct osd_inode_id *id, struct lu_fid *fid)
258 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
262 inode = osd_iget(info, dev, id);
266 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
268 *fid = lma->lma_self_fid;
269 } else if (rc == -ENODATA) {
270 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
271 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
273 lu_igif_build(fid, inode->i_ino, inode->i_generation);
281 static struct inode *
282 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
283 struct osd_inode_id *id, const struct lu_fid *fid)
285 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
289 inode = osd_iget(info, dev, id);
293 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
302 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
303 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
304 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
306 return ERR_PTR(-EREMCHG);
312 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
313 const struct lu_fid *fid,
314 const struct lu_object_conf *conf)
316 struct osd_thread_info *info;
317 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
318 struct osd_device *dev;
319 struct osd_idmap_cache *oic;
320 struct osd_inode_id *id;
322 struct osd_scrub *scrub;
323 struct scrub_file *sf;
328 LINVRNT(osd_invariant(obj));
329 LASSERT(obj->oo_inode == NULL);
330 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
333 scrub = &dev->od_scrub;
334 sf = &scrub->os_file;
335 info = osd_oti_get(env);
337 oic = &info->oti_cache;
339 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
342 /* Search order: 1. per-thread cache. */
343 if (lu_fid_eq(fid, &oic->oic_fid)) {
349 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
350 /* Search order: 2. OI scrub pending list. */
351 result = osd_oii_lookup(dev, fid, id);
356 if (sf->sf_flags & SF_INCONSISTENT)
360 * Objects are created as locking anchors or place holders for objects
361 * yet to be created. No need to osd_oi_lookup() at here because FID
362 * shouldn't never be re-used, if it's really a duplicate FID from
363 * unexpected reason, we should be able to detect it later by calling
364 * do_create->osd_oi_insert()
366 if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
367 GOTO(out, result = 0);
369 /* Search order: 3. OI files. */
370 result = osd_oi_lookup(info, dev, fid, id);
371 if (result == -ENOENT) {
372 if (!fid_is_norm(fid) ||
373 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
375 GOTO(out, result = 0);
385 inode = osd_iget(info, dev, id);
387 inode = osd_iget_verify(info, dev, id, fid);
389 result = PTR_ERR(inode);
390 if (result == -ENOENT || result == -ESTALE) {
391 fid_zero(&oic->oic_fid);
393 } else if (result == -EREMCHG) {
396 if (thread_is_running(&scrub->os_thread)) {
397 result = -EINPROGRESS;
398 } else if (!dev->od_noscrub) {
399 result = osd_scrub_start(dev);
400 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
401 "for "DFID", rc = %d [1]\n",
402 LDISKFS_SB(osd_sb(dev))->s_es->\
403 s_volume_name,PFID(fid), result);
404 if (result == 0 || result == -EALREADY)
405 result = -EINPROGRESS;
414 obj->oo_inode = inode;
415 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
417 obj->oo_compat_dot_created = 1;
418 obj->oo_compat_dotdot_created = 1;
420 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
421 GOTO(out, result = 0);
423 LASSERT(obj->oo_hl_head == NULL);
424 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
425 if (obj->oo_hl_head == NULL) {
426 obj->oo_inode = NULL;
428 GOTO(out, result = -ENOMEM);
430 GOTO(out, result = 0);
433 LINVRNT(osd_invariant(obj));
438 * Concurrency: shouldn't matter.
440 static void osd_object_init0(struct osd_object *obj)
442 LASSERT(obj->oo_inode != NULL);
443 obj->oo_dt.do_body_ops = &osd_body_ops;
444 obj->oo_dt.do_lu.lo_header->loh_attr |=
445 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
449 * Concurrency: no concurrent access is possible that early in object
452 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
453 const struct lu_object_conf *conf)
455 struct osd_object *obj = osd_obj(l);
458 LINVRNT(osd_invariant(obj));
460 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
461 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
462 l->lo_header->loh_attr |= LOHA_EXISTS;
466 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
467 obj->oo_dt.do_body_ops = &osd_body_ops_new;
468 if (result == 0 && obj->oo_inode != NULL)
469 osd_object_init0(obj);
471 LINVRNT(osd_invariant(obj));
476 * Concurrency: no concurrent access is possible that late in object
479 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
481 struct osd_object *obj = osd_obj(l);
483 LINVRNT(osd_invariant(obj));
485 dt_object_fini(&obj->oo_dt);
486 if (obj->oo_hl_head != NULL)
487 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
492 * Concurrency: no concurrent access is possible that late in object
495 static void osd_index_fini(struct osd_object *o)
497 struct iam_container *bag;
499 if (o->oo_dir != NULL) {
500 bag = &o->oo_dir->od_container;
501 if (o->oo_inode != NULL) {
502 if (bag->ic_object == o->oo_inode)
503 iam_container_fini(bag);
505 OBD_FREE_PTR(o->oo_dir);
511 * Concurrency: no concurrent access is possible that late in object
512 * life-cycle (for all existing callers, that is. New callers have to provide
513 * their own locking.)
515 static int osd_inode_unlinked(const struct inode *inode)
517 return inode->i_nlink == 0;
521 OSD_TXN_OI_DELETE_CREDITS = 20,
522 OSD_TXN_INODE_DELETE_CREDITS = 20
529 #if OSD_THANDLE_STATS
531 * Set time when the handle is allocated
533 static void osd_th_alloced(struct osd_thandle *oth)
535 oth->oth_alloced = cfs_time_current();
539 * Set time when the handle started
541 static void osd_th_started(struct osd_thandle *oth)
543 oth->oth_started = cfs_time_current();
547 * Helper function to convert time interval to microseconds packed in
550 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
554 cfs_duration_usec(cfs_time_sub(end, start), &val);
555 return val.tv_sec * 1000000 + val.tv_usec;
559 * Check whether the we deal with this handle for too long.
561 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
562 cfs_time_t alloced, cfs_time_t started,
565 cfs_time_t now = cfs_time_current();
567 LASSERT(dev != NULL);
569 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
570 interval_to_usec(alloced, started));
571 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
572 interval_to_usec(started, closed));
573 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
574 interval_to_usec(closed, now));
576 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
577 CWARN("transaction handle %p was open for too long: "
579 "alloced "CFS_TIME_T" ,"
580 "started "CFS_TIME_T" ,"
581 "closed "CFS_TIME_T"\n",
582 oth, now, alloced, started, closed);
583 libcfs_debug_dumpstack(NULL);
587 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
589 cfs_time_t __closed = cfs_time_current(); \
590 cfs_time_t __alloced = oth->oth_alloced; \
591 cfs_time_t __started = oth->oth_started; \
594 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
597 #else /* OSD_THANDLE_STATS */
599 #define osd_th_alloced(h) do {} while(0)
600 #define osd_th_started(h) do {} while(0)
601 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
603 #endif /* OSD_THANDLE_STATS */
606 * Concurrency: doesn't access mutable data.
608 static int osd_param_is_not_sane(const struct osd_device *dev,
609 const struct thandle *th)
611 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
613 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
617 * Concurrency: shouldn't matter.
619 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
620 static void osd_trans_commit_cb(struct super_block *sb,
621 struct journal_callback *jcb, int error)
623 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
626 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
627 struct thandle *th = &oh->ot_super;
628 struct lu_device *lud = &th->th_dev->dd_lu_dev;
629 struct dt_txn_commit_cb *dcb, *tmp;
631 LASSERT(oh->ot_handle == NULL);
634 CERROR("transaction @0x%p commit error: %d\n", th, error);
636 dt_txn_hook_commit(th);
638 /* call per-transaction callbacks if any */
639 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
640 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
641 "commit callback entry: magic=%x name='%s'\n",
642 dcb->dcb_magic, dcb->dcb_name);
643 cfs_list_del_init(&dcb->dcb_linkage);
644 dcb->dcb_func(NULL, th, dcb, error);
647 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
651 lu_context_exit(&th->th_ctx);
652 lu_context_fini(&th->th_ctx);
656 static struct thandle *osd_trans_create(const struct lu_env *env,
659 struct osd_thread_info *oti = osd_oti_get(env);
660 struct osd_iobuf *iobuf = &oti->oti_iobuf;
661 struct osd_thandle *oh;
665 /* on pending IO in this thread should left from prev. request */
666 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
668 th = ERR_PTR(-ENOMEM);
669 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
671 oh->ot_quota_trans = &oti->oti_quota_trans;
672 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
676 th->th_tags = LCT_TX_HANDLE;
678 oti->oti_dev = osd_dt_dev(d);
679 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
682 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
683 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
684 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
685 oti->oti_rollback = false;
691 * Concurrency: shouldn't matter.
693 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
696 struct osd_thread_info *oti = osd_oti_get(env);
697 struct osd_device *dev = osd_dt_dev(d);
699 struct osd_thandle *oh;
704 LASSERT(current->journal_info == NULL);
706 oh = container_of0(th, struct osd_thandle, ot_super);
708 LASSERT(oh->ot_handle == NULL);
710 rc = dt_txn_hook_start(env, d, th);
714 if (unlikely(osd_param_is_not_sane(dev, th))) {
715 static unsigned long last_printed;
716 static int last_credits;
718 CWARN("%.16s: too many transaction credits (%d > %d)\n",
719 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
721 osd_journal(dev)->j_max_transaction_buffers);
722 #ifdef OSD_TRACK_DECLARES
723 CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
724 oti->oti_declare_ops[OSD_OT_CREATE],
725 oti->oti_declare_ops_cred[OSD_OT_CREATE],
726 oti->oti_declare_ops[OSD_OT_DELETE],
727 oti->oti_declare_ops_cred[OSD_OT_DELETE],
728 oti->oti_declare_ops[OSD_OT_DESTROY],
729 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
730 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
731 oti->oti_declare_ops[OSD_OT_ATTR_SET],
732 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
733 oti->oti_declare_ops[OSD_OT_XATTR_SET],
734 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
735 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
736 oti->oti_declare_ops[OSD_OT_WRITE],
737 oti->oti_declare_ops_cred[OSD_OT_WRITE],
738 oti->oti_declare_ops[OSD_OT_PUNCH],
739 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
740 oti->oti_declare_ops[OSD_OT_QUOTA],
741 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
742 CWARN(" insert: %u/%u, delete: %u/%u\n",
743 oti->oti_declare_ops[OSD_OT_INSERT],
744 oti->oti_declare_ops_cred[OSD_OT_INSERT],
745 oti->oti_declare_ops[OSD_OT_DESTROY],
746 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
747 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
748 oti->oti_declare_ops[OSD_OT_REF_ADD],
749 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
750 oti->oti_declare_ops[OSD_OT_REF_DEL],
751 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
753 if (last_credits != oh->ot_credits &&
754 time_after(jiffies, last_printed + 60 * HZ)) {
755 libcfs_debug_dumpstack(NULL);
756 last_credits = oh->ot_credits;
757 last_printed = jiffies;
760 /* XXX Limit the credits to 'max_transaction_buffers', and
761 * let the underlying filesystem to catch the error if
762 * we really need so many credits.
764 * This should be removed when we can calculate the
765 * credits precisely. */
766 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
770 * XXX temporary stuff. Some abstraction layer should
773 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
777 LASSERT(oti->oti_txns == 0);
778 lu_context_init(&th->th_ctx, th->th_tags);
779 lu_context_enter(&th->th_ctx);
781 lu_device_get(&d->dd_lu_dev);
782 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
794 * Concurrency: shouldn't matter.
796 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
799 struct osd_thandle *oh;
800 struct osd_thread_info *oti = osd_oti_get(env);
801 struct osd_iobuf *iobuf = &oti->oti_iobuf;
802 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
805 oh = container_of0(th, struct osd_thandle, ot_super);
808 /* inform the quota slave device that the transaction is
810 qsd_op_end(env, qsd, oh->ot_quota_trans);
811 oh->ot_quota_trans = NULL;
813 if (oh->ot_handle != NULL) {
814 handle_t *hdl = oh->ot_handle;
817 * add commit callback
818 * notice we don't do this in osd_trans_start()
819 * as underlying transaction can change during truncate
821 osd_journal_callback_set(hdl, osd_trans_commit_cb,
824 LASSERT(oti->oti_txns == 1);
826 rc = dt_txn_hook_stop(env, th);
828 CERROR("Failure in transaction hook: %d\n", rc);
830 /* hook functions might modify th_sync */
831 hdl->h_sync = th->th_sync;
833 oh->ot_handle = NULL;
834 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
835 rc = ldiskfs_journal_stop(hdl));
837 CERROR("Failure to stop transaction: %d\n", rc);
842 /* as we want IO to journal and data IO be concurrent, we don't block
843 * awaiting data IO completion in osd_do_bio(), instead we wait here
844 * once transaction is submitted to the journal. all reqular requests
845 * don't do direct IO (except read/write), thus this wait_event becomes
848 * IMPORTANT: we have to wait till any IO submited by the thread is
849 * completed otherwise iobuf may be corrupted by different request
851 cfs_wait_event(iobuf->dr_wait,
852 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
854 rc = iobuf->dr_error;
859 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
861 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
864 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
865 LASSERT(&dcb->dcb_func != NULL);
866 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
872 * Called just before object is freed. Releases all resources except for
873 * object itself (that is released by osd_object_free()).
875 * Concurrency: no concurrent access is possible that late in object
878 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
880 struct osd_object *obj = osd_obj(l);
881 struct inode *inode = obj->oo_inode;
883 LINVRNT(osd_invariant(obj));
886 * If object is unlinked remove fid->ino mapping from object index.
891 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
892 qid_t uid = inode->i_uid;
893 qid_t gid = inode->i_gid;
896 obj->oo_inode = NULL;
899 struct osd_thread_info *info = osd_oti_get(env);
900 struct lquota_id_info *qi = &info->oti_qi;
902 /* Release granted quota to master if necessary */
903 qi->lqi_id.qid_uid = uid;
904 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
906 qi->lqi_id.qid_uid = gid;
907 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
913 * Concurrency: ->loo_object_release() is called under site spin-lock.
915 static void osd_object_release(const struct lu_env *env,
921 * Concurrency: shouldn't matter.
923 static int osd_object_print(const struct lu_env *env, void *cookie,
924 lu_printer_t p, const struct lu_object *l)
926 struct osd_object *o = osd_obj(l);
929 if (o->oo_dir != NULL)
930 d = o->oo_dir->od_container.ic_descr;
933 return (*p)(env, cookie,
934 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
936 o->oo_inode ? o->oo_inode->i_ino : 0UL,
937 o->oo_inode ? o->oo_inode->i_generation : 0,
938 d ? d->id_ops->id_name : "plain");
942 * Concurrency: shouldn't matter.
944 int osd_statfs(const struct lu_env *env, struct dt_device *d,
945 struct obd_statfs *sfs)
947 struct osd_device *osd = osd_dt_dev(d);
948 struct super_block *sb = osd_sb(osd);
949 struct kstatfs *ksfs;
952 if (unlikely(osd->od_mnt == NULL))
955 /* osd_lproc.c call this without env, allocate ksfs for that case */
956 if (unlikely(env == NULL)) {
961 ksfs = &osd_oti_get(env)->oti_ksfs;
964 spin_lock(&osd->od_osfs_lock);
966 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
967 result = sb->s_op->statfs(sb->s_root, ksfs);
968 if (likely(result == 0)) { /* N.B. statfs can't really fail */
969 osd->od_osfs_age = cfs_time_current_64();
970 statfs_pack(&osd->od_statfs, ksfs);
971 if (sb->s_flags & MS_RDONLY)
972 sfs->os_state = OS_STATE_READONLY;
976 if (likely(result == 0))
977 *sfs = osd->od_statfs;
978 spin_unlock(&osd->od_osfs_lock);
980 if (unlikely(env == NULL))
987 * Estimate space needed for file creations. We assume the largest filename
988 * which is 2^64 - 1, hence a filename of 20 chars.
989 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
991 #ifdef __LDISKFS_DIR_REC_LEN
992 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
994 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
998 * Concurrency: doesn't access mutable data.
1000 static void osd_conf_get(const struct lu_env *env,
1001 const struct dt_device *dev,
1002 struct dt_device_param *param)
1004 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1007 * XXX should be taken from not-yet-existing fs abstraction layer.
1009 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1010 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1011 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1012 param->ddp_block_shift = sb->s_blocksize_bits;
1013 param->ddp_mount_type = LDD_MT_LDISKFS;
1014 param->ddp_maxbytes = sb->s_maxbytes;
1015 /* Overhead estimate should be fairly accurate, so we really take a tiny
1016 * error margin which also avoids fragmenting the filesystem too much */
1017 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1018 /* inode are statically allocated, so per-inode space consumption
1019 * is the space consumed by the directory entry */
1020 param->ddp_inodespace = PER_OBJ_USAGE;
1021 /* per-fragment overhead to be used by the client code */
1022 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1023 param->ddp_mntopts = 0;
1024 if (test_opt(sb, XATTR_USER))
1025 param->ddp_mntopts |= MNTOPT_USERXATTR;
1026 if (test_opt(sb, POSIX_ACL))
1027 param->ddp_mntopts |= MNTOPT_ACL;
1029 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1030 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1031 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1034 param->ddp_max_ea_size = sb->s_blocksize;
1039 * Helper function to get and fill the buffer with input values.
1041 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1045 buf = &osd_oti_get(env)->oti_buf;
1052 * Concurrency: shouldn't matter.
1054 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1056 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1057 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1061 * Start commit for OSD device.
1063 * An implementation of dt_commit_async method for OSD device.
1064 * Asychronously starts underlayng fs sync and thereby a transaction
1067 * \param env environment
1068 * \param d dt device
1070 * \see dt_device_operations
1072 static int osd_commit_async(const struct lu_env *env,
1073 struct dt_device *d)
1075 struct super_block *s = osd_sb(osd_dt_dev(d));
1078 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1079 RETURN(s->s_op->sync_fs(s, 0));
1083 * Concurrency: shouldn't matter.
1086 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1088 struct super_block *sb = osd_sb(osd_dt_dev(d));
1092 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1094 rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1099 * Concurrency: serialization provided by callers.
1101 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1102 int mode, unsigned long timeout, __u32 alg,
1103 struct lustre_capa_key *keys)
1105 struct osd_device *dev = osd_dt_dev(d);
1108 dev->od_fl_capa = mode;
1109 dev->od_capa_timeout = timeout;
1110 dev->od_capa_alg = alg;
1111 dev->od_capa_keys = keys;
1116 * Note: we do not count into QUOTA here.
1117 * If we mount with --data_journal we may need more.
1119 const int osd_dto_credits_noquota[DTO_NR] = {
1122 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1123 * SINGLEDATA_TRANS_BLOCKS(8)
1124 * XXX Note: maybe iam need more, since iam have more level than
1127 [DTO_INDEX_INSERT] = 16,
1128 [DTO_INDEX_DELETE] = 16,
1132 [DTO_INDEX_UPDATE] = 16,
1134 * Create a object. The same as create object in EXT3.
1135 * DATA_TRANS_BLOCKS(14) +
1136 * INDEX_EXTRA_BLOCKS(8) +
1137 * 3(inode bits, groups, GDT)
1139 [DTO_OBJECT_CREATE] = 25,
1141 * XXX: real credits to be fixed
1143 [DTO_OBJECT_DELETE] = 25,
1145 * Attr set credits (inode)
1147 [DTO_ATTR_SET_BASE] = 1,
1149 * Xattr set. The same as xattr of EXT3.
1150 * DATA_TRANS_BLOCKS(14)
1151 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1152 * are also counted in. Do not know why?
1154 [DTO_XATTR_SET] = 14,
1157 * credits for inode change during write.
1159 [DTO_WRITE_BASE] = 3,
1161 * credits for single block write.
1163 [DTO_WRITE_BLOCK] = 14,
1165 * Attr set credits for chown.
1166 * This is extra credits for setattr, and it is null without quota
1168 [DTO_ATTR_SET_CHOWN]= 0
1171 static const struct dt_device_operations osd_dt_ops = {
1172 .dt_root_get = osd_root_get,
1173 .dt_statfs = osd_statfs,
1174 .dt_trans_create = osd_trans_create,
1175 .dt_trans_start = osd_trans_start,
1176 .dt_trans_stop = osd_trans_stop,
1177 .dt_trans_cb_add = osd_trans_cb_add,
1178 .dt_conf_get = osd_conf_get,
1179 .dt_sync = osd_sync,
1181 .dt_commit_async = osd_commit_async,
1182 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1185 static void osd_object_read_lock(const struct lu_env *env,
1186 struct dt_object *dt, unsigned role)
1188 struct osd_object *obj = osd_dt_obj(dt);
1189 struct osd_thread_info *oti = osd_oti_get(env);
1191 LINVRNT(osd_invariant(obj));
1193 LASSERT(obj->oo_owner != env);
1194 down_read_nested(&obj->oo_sem, role);
1196 LASSERT(obj->oo_owner == NULL);
1200 static void osd_object_write_lock(const struct lu_env *env,
1201 struct dt_object *dt, unsigned role)
1203 struct osd_object *obj = osd_dt_obj(dt);
1204 struct osd_thread_info *oti = osd_oti_get(env);
1206 LINVRNT(osd_invariant(obj));
1208 LASSERT(obj->oo_owner != env);
1209 down_write_nested(&obj->oo_sem, role);
1211 LASSERT(obj->oo_owner == NULL);
1212 obj->oo_owner = env;
1216 static void osd_object_read_unlock(const struct lu_env *env,
1217 struct dt_object *dt)
1219 struct osd_object *obj = osd_dt_obj(dt);
1220 struct osd_thread_info *oti = osd_oti_get(env);
1222 LINVRNT(osd_invariant(obj));
1224 LASSERT(oti->oti_r_locks > 0);
1226 up_read(&obj->oo_sem);
1229 static void osd_object_write_unlock(const struct lu_env *env,
1230 struct dt_object *dt)
1232 struct osd_object *obj = osd_dt_obj(dt);
1233 struct osd_thread_info *oti = osd_oti_get(env);
1235 LINVRNT(osd_invariant(obj));
1237 LASSERT(obj->oo_owner == env);
1238 LASSERT(oti->oti_w_locks > 0);
1240 obj->oo_owner = NULL;
1241 up_write(&obj->oo_sem);
1244 static int osd_object_write_locked(const struct lu_env *env,
1245 struct dt_object *dt)
1247 struct osd_object *obj = osd_dt_obj(dt);
1249 LINVRNT(osd_invariant(obj));
1251 return obj->oo_owner == env;
1254 static int capa_is_sane(const struct lu_env *env,
1255 struct osd_device *dev,
1256 struct lustre_capa *capa,
1257 struct lustre_capa_key *keys)
1259 struct osd_thread_info *oti = osd_oti_get(env);
1260 struct lustre_capa *tcapa = &oti->oti_capa;
1261 struct obd_capa *oc;
1265 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1267 if (capa_is_expired(oc)) {
1268 DEBUG_CAPA(D_ERROR, capa, "expired");
1275 if (capa_is_expired_sec(capa)) {
1276 DEBUG_CAPA(D_ERROR, capa, "expired");
1280 spin_lock(&capa_lock);
1281 for (i = 0; i < 2; i++) {
1282 if (keys[i].lk_keyid == capa->lc_keyid) {
1283 oti->oti_capa_key = keys[i];
1287 spin_unlock(&capa_lock);
1290 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1294 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1298 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1299 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1303 oc = capa_add(dev->od_capa_hash, capa);
1309 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1310 struct lustre_capa *capa, __u64 opc)
1312 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1313 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1314 struct md_capainfo *ci;
1317 if (!dev->od_fl_capa)
1320 if (capa == BYPASS_CAPA)
1323 ci = md_capainfo(env);
1327 if (ci->mc_auth == LC_ID_NONE)
1331 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1335 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1336 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1341 if (!capa_opc_supported(capa, opc)) {
1342 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1346 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1347 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1354 static struct timespec *osd_inode_time(const struct lu_env *env,
1355 struct inode *inode, __u64 seconds)
1357 struct osd_thread_info *oti = osd_oti_get(env);
1358 struct timespec *t = &oti->oti_time;
1360 t->tv_sec = seconds;
1362 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1367 static void osd_inode_getattr(const struct lu_env *env,
1368 struct inode *inode, struct lu_attr *attr)
1370 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1371 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1372 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1374 attr->la_atime = LTIME_S(inode->i_atime);
1375 attr->la_mtime = LTIME_S(inode->i_mtime);
1376 attr->la_ctime = LTIME_S(inode->i_ctime);
1377 attr->la_mode = inode->i_mode;
1378 attr->la_size = i_size_read(inode);
1379 attr->la_blocks = inode->i_blocks;
1380 attr->la_uid = inode->i_uid;
1381 attr->la_gid = inode->i_gid;
1382 attr->la_flags = LDISKFS_I(inode)->i_flags;
1383 attr->la_nlink = inode->i_nlink;
1384 attr->la_rdev = inode->i_rdev;
1385 attr->la_blksize = 1 << inode->i_blkbits;
1386 attr->la_blkbits = inode->i_blkbits;
1389 static int osd_attr_get(const struct lu_env *env,
1390 struct dt_object *dt,
1391 struct lu_attr *attr,
1392 struct lustre_capa *capa)
1394 struct osd_object *obj = osd_dt_obj(dt);
1396 LASSERT(dt_object_exists(dt));
1397 LINVRNT(osd_invariant(obj));
1399 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1402 spin_lock(&obj->oo_guard);
1403 osd_inode_getattr(env, obj->oo_inode, attr);
1404 spin_unlock(&obj->oo_guard);
1408 static int osd_declare_attr_set(const struct lu_env *env,
1409 struct dt_object *dt,
1410 const struct lu_attr *attr,
1411 struct thandle *handle)
1413 struct osd_thandle *oh;
1414 struct osd_object *obj;
1415 struct osd_thread_info *info = osd_oti_get(env);
1416 struct lquota_id_info *qi = &info->oti_qi;
1422 LASSERT(dt != NULL);
1423 LASSERT(handle != NULL);
1425 obj = osd_dt_obj(dt);
1426 LASSERT(osd_invariant(obj));
1428 oh = container_of0(handle, struct osd_thandle, ot_super);
1429 LASSERT(oh->ot_handle == NULL);
1431 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1432 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1434 if (attr == NULL || obj->oo_inode == NULL)
1437 bspace = obj->oo_inode->i_blocks;
1438 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1439 bspace = toqb(bspace);
1441 /* Changing ownership is always preformed by super user, it should not
1444 * We still need to call the osd_declare_qid() to calculate the journal
1445 * credits for updating quota accounting files and to trigger quota
1446 * space adjustment once the operation is completed.*/
1447 if ((attr->la_valid & LA_UID) != 0 &&
1448 attr->la_uid != obj->oo_inode->i_uid) {
1449 qi->lqi_type = USRQUOTA;
1451 /* inode accounting */
1452 qi->lqi_is_blk = false;
1454 /* one more inode for the new owner ... */
1455 qi->lqi_id.qid_uid = attr->la_uid;
1457 allocated = (attr->la_uid == 0) ? true : false;
1458 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1459 if (rc == -EDQUOT || rc == -EINPROGRESS)
1464 /* and one less inode for the current uid */
1465 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1467 rc = osd_declare_qid(env, oh, qi, true, NULL);
1468 if (rc == -EDQUOT || rc == -EINPROGRESS)
1473 /* block accounting */
1474 qi->lqi_is_blk = true;
1476 /* more blocks for the new owner ... */
1477 qi->lqi_id.qid_uid = attr->la_uid;
1478 qi->lqi_space = bspace;
1479 allocated = (attr->la_uid == 0) ? true : false;
1480 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1481 if (rc == -EDQUOT || rc == -EINPROGRESS)
1486 /* and finally less blocks for the current owner */
1487 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1488 qi->lqi_space = -bspace;
1489 rc = osd_declare_qid(env, oh, qi, true, NULL);
1490 if (rc == -EDQUOT || rc == -EINPROGRESS)
1496 if (attr->la_valid & LA_GID &&
1497 attr->la_gid != obj->oo_inode->i_gid) {
1498 qi->lqi_type = GRPQUOTA;
1500 /* inode accounting */
1501 qi->lqi_is_blk = false;
1503 /* one more inode for the new group owner ... */
1504 qi->lqi_id.qid_gid = attr->la_gid;
1506 allocated = (attr->la_gid == 0) ? true : false;
1507 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1508 if (rc == -EDQUOT || rc == -EINPROGRESS)
1513 /* and one less inode for the current gid */
1514 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1516 rc = osd_declare_qid(env, oh, qi, true, NULL);
1517 if (rc == -EDQUOT || rc == -EINPROGRESS)
1522 /* block accounting */
1523 qi->lqi_is_blk = true;
1525 /* more blocks for the new owner ... */
1526 qi->lqi_id.qid_gid = attr->la_gid;
1527 qi->lqi_space = bspace;
1528 allocated = (attr->la_gid == 0) ? true : false;
1529 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1530 if (rc == -EDQUOT || rc == -EINPROGRESS)
1535 /* and finally less blocks for the current owner */
1536 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1537 qi->lqi_space = -bspace;
1538 rc = osd_declare_qid(env, oh, qi, true, NULL);
1539 if (rc == -EDQUOT || rc == -EINPROGRESS)
1548 static int osd_inode_setattr(const struct lu_env *env,
1549 struct inode *inode, const struct lu_attr *attr)
1553 bits = attr->la_valid;
1555 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1557 if (bits & LA_ATIME)
1558 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1559 if (bits & LA_CTIME)
1560 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1561 if (bits & LA_MTIME)
1562 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1563 if (bits & LA_SIZE) {
1564 LDISKFS_I(inode)->i_disksize = attr->la_size;
1565 i_size_write(inode, attr->la_size);
1569 /* OSD should not change "i_blocks" which is used by quota.
1570 * "i_blocks" should be changed by ldiskfs only. */
1571 if (bits & LA_BLOCKS)
1572 inode->i_blocks = attr->la_blocks;
1575 inode->i_mode = (inode->i_mode & S_IFMT) |
1576 (attr->la_mode & ~S_IFMT);
1578 inode->i_uid = attr->la_uid;
1580 inode->i_gid = attr->la_gid;
1581 if (bits & LA_NLINK)
1582 set_nlink(inode, attr->la_nlink);
1584 inode->i_rdev = attr->la_rdev;
1586 if (bits & LA_FLAGS) {
1587 /* always keep S_NOCMTIME */
1588 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1594 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1596 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1597 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1602 if (attr->la_valid & LA_UID)
1603 iattr.ia_valid |= ATTR_UID;
1604 if (attr->la_valid & LA_GID)
1605 iattr.ia_valid |= ATTR_GID;
1606 iattr.ia_uid = attr->la_uid;
1607 iattr.ia_gid = attr->la_gid;
1609 rc = ll_vfs_dq_transfer(inode, &iattr);
1611 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1612 "enforcement enabled on the ldiskfs filesystem?",
1613 inode->i_sb->s_id, rc);
1620 static int osd_attr_set(const struct lu_env *env,
1621 struct dt_object *dt,
1622 const struct lu_attr *attr,
1623 struct thandle *handle,
1624 struct lustre_capa *capa)
1626 struct osd_object *obj = osd_dt_obj(dt);
1627 struct inode *inode;
1630 LASSERT(handle != NULL);
1631 LASSERT(dt_object_exists(dt));
1632 LASSERT(osd_invariant(obj));
1634 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1637 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1639 inode = obj->oo_inode;
1640 ll_vfs_dq_init(inode);
1642 rc = osd_quota_transfer(inode, attr);
1646 spin_lock(&obj->oo_guard);
1647 rc = osd_inode_setattr(env, inode, attr);
1648 spin_unlock(&obj->oo_guard);
1651 inode->i_sb->s_op->dirty_inode(inode);
1655 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1656 struct osd_object *obj,
1657 const char *name, const int namelen)
1659 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1662 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1664 struct dt_allocation_hint *hint,
1668 struct osd_device *osd = osd_obj2dev(obj);
1669 struct osd_thandle *oth;
1670 struct dt_object *parent = NULL;
1671 struct inode *inode;
1673 LINVRNT(osd_invariant(obj));
1674 LASSERT(obj->oo_inode == NULL);
1675 LASSERT(obj->oo_hl_head == NULL);
1677 if (S_ISDIR(mode) && ldiskfs_pdo) {
1678 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1679 if (obj->oo_hl_head == NULL)
1683 oth = container_of(th, struct osd_thandle, ot_super);
1684 LASSERT(oth->ot_handle->h_transaction != NULL);
1686 if (hint && hint->dah_parent)
1687 parent = hint->dah_parent;
1689 inode = ldiskfs_create_inode(oth->ot_handle,
1690 parent ? osd_dt_obj(parent)->oo_inode :
1691 osd_sb(osd)->s_root->d_inode,
1693 if (!IS_ERR(inode)) {
1694 /* Do not update file c/mtime in ldiskfs.
1695 * NB: don't need any lock because no contention at this
1697 inode->i_flags |= S_NOCMTIME;
1698 inode->i_state |= I_LUSTRE_NOSCRUB;
1699 obj->oo_inode = inode;
1702 if (obj->oo_hl_head != NULL) {
1703 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1704 obj->oo_hl_head = NULL;
1706 result = PTR_ERR(inode);
1708 LINVRNT(osd_invariant(obj));
1716 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1717 struct lu_attr *attr,
1718 struct dt_allocation_hint *hint,
1719 struct dt_object_format *dof,
1723 struct osd_thandle *oth;
1724 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1726 LASSERT(S_ISDIR(attr->la_mode));
1728 oth = container_of(th, struct osd_thandle, ot_super);
1729 LASSERT(oth->ot_handle->h_transaction != NULL);
1730 result = osd_mkfile(info, obj, mode, hint, th);
1735 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1736 struct lu_attr *attr,
1737 struct dt_allocation_hint *hint,
1738 struct dt_object_format *dof,
1742 struct osd_thandle *oth;
1743 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1745 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1747 LASSERT(S_ISREG(attr->la_mode));
1749 oth = container_of(th, struct osd_thandle, ot_super);
1750 LASSERT(oth->ot_handle->h_transaction != NULL);
1752 result = osd_mkfile(info, obj, mode, hint, th);
1754 LASSERT(obj->oo_inode != NULL);
1755 if (feat->dif_flags & DT_IND_VARKEY)
1756 result = iam_lvar_create(obj->oo_inode,
1757 feat->dif_keysize_max,
1759 feat->dif_recsize_max,
1762 result = iam_lfix_create(obj->oo_inode,
1763 feat->dif_keysize_max,
1765 feat->dif_recsize_max,
1772 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1773 struct lu_attr *attr,
1774 struct dt_allocation_hint *hint,
1775 struct dt_object_format *dof,
1778 LASSERT(S_ISREG(attr->la_mode));
1779 return osd_mkfile(info, obj, (attr->la_mode &
1780 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1783 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1784 struct lu_attr *attr,
1785 struct dt_allocation_hint *hint,
1786 struct dt_object_format *dof,
1789 LASSERT(S_ISLNK(attr->la_mode));
1790 return osd_mkfile(info, obj, (attr->la_mode &
1791 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1794 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1795 struct lu_attr *attr,
1796 struct dt_allocation_hint *hint,
1797 struct dt_object_format *dof,
1800 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1803 LINVRNT(osd_invariant(obj));
1804 LASSERT(obj->oo_inode == NULL);
1805 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1806 S_ISFIFO(mode) || S_ISSOCK(mode));
1808 result = osd_mkfile(info, obj, mode, hint, th);
1810 LASSERT(obj->oo_inode != NULL);
1812 * This inode should be marked dirty for i_rdev. Currently
1813 * that is done in the osd_attr_init().
1815 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1818 LINVRNT(osd_invariant(obj));
1822 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1824 struct dt_allocation_hint *hint,
1825 struct dt_object_format *dof,
1828 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1830 osd_obj_type_f result;
1846 result = osd_mk_index;
1857 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1858 struct dt_object *parent, struct dt_object *child,
1859 cfs_umode_t child_mode)
1863 memset(ah, 0, sizeof(*ah));
1864 ah->dah_parent = parent;
1865 ah->dah_mode = child_mode;
1868 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1869 struct lu_attr *attr, struct dt_object_format *dof)
1871 struct inode *inode = obj->oo_inode;
1872 __u64 valid = attr->la_valid;
1875 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1877 if (dof->dof_type != DFT_NODE)
1878 attr->la_valid &= ~LA_RDEV;
1879 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1880 attr->la_valid &= ~LA_ATIME;
1881 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1882 attr->la_valid &= ~LA_CTIME;
1883 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1884 attr->la_valid &= ~LA_MTIME;
1886 result = osd_quota_transfer(inode, attr);
1890 if (attr->la_valid != 0) {
1891 result = osd_inode_setattr(info->oti_env, inode, attr);
1893 * The osd_inode_setattr() should always succeed here. The
1894 * only error that could be returned is EDQUOT when we are
1895 * trying to change the UID or GID of the inode. However, this
1896 * should not happen since quota enforcement is no longer
1897 * enabled on ldiskfs (lquota takes care of it).
1899 LASSERTF(result == 0, "%d", result);
1900 inode->i_sb->s_op->dirty_inode(inode);
1903 attr->la_valid = valid;
1907 * Helper function for osd_object_create()
1909 * \retval 0, on success
1911 static int __osd_object_create(struct osd_thread_info *info,
1912 struct osd_object *obj, struct lu_attr *attr,
1913 struct dt_allocation_hint *hint,
1914 struct dt_object_format *dof,
1919 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1922 osd_attr_init(info, obj, attr, dof);
1923 osd_object_init0(obj);
1925 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1926 unlock_new_inode(obj->oo_inode);
1933 * Helper function for osd_object_create()
1935 * \retval 0, on success
1937 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1938 const struct lu_fid *fid, struct thandle *th)
1940 struct osd_thread_info *info = osd_oti_get(env);
1941 struct osd_inode_id *id = &info->oti_id;
1942 struct osd_device *osd = osd_obj2dev(obj);
1944 LASSERT(obj->oo_inode != NULL);
1946 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1947 return osd_oi_insert(info, osd, fid, id, th);
1950 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1951 const struct lu_fid *fid, struct lu_seq_range *range)
1953 struct seq_server_site *ss = osd_seq_site(osd);
1956 if (fid_is_igif(fid)) {
1957 range->lsr_flags = LU_SEQ_RANGE_MDT;
1958 range->lsr_index = 0;
1962 if (fid_is_idif(fid)) {
1963 range->lsr_flags = LU_SEQ_RANGE_OST;
1964 range->lsr_index = fid_idif_ost_idx(fid);
1968 if (!fid_is_norm(fid)) {
1969 range->lsr_flags = LU_SEQ_RANGE_MDT;
1971 /* FIXME: If ss is NULL, it suppose not get lsr_index
1973 range->lsr_index = ss->ss_node_id;
1977 LASSERT(ss != NULL);
1978 range->lsr_flags = -1;
1979 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1981 CERROR("%s can not find "DFID": rc = %d\n",
1982 osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc);
1988 static int osd_declare_object_create(const struct lu_env *env,
1989 struct dt_object *dt,
1990 struct lu_attr *attr,
1991 struct dt_allocation_hint *hint,
1992 struct dt_object_format *dof,
1993 struct thandle *handle)
1995 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1996 struct osd_thandle *oh;
2000 LASSERT(handle != NULL);
2002 oh = container_of0(handle, struct osd_thandle, ot_super);
2003 LASSERT(oh->ot_handle == NULL);
2005 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2006 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2007 /* XXX: So far, only normal fid needs be inserted into the oi,
2008 * things could be changed later. Revise following code then. */
2009 if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2010 !fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2011 lu_object_fid(&dt->do_lu))) {
2012 /* Reuse idle OI block may cause additional one OI block
2014 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2015 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2017 /* If this is directory, then we expect . and .. to be inserted as
2018 * well. The one directory block always needs to be created for the
2019 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2020 * block), there is no danger of needing a tree for the first block.
2022 if (attr && S_ISDIR(attr->la_mode)) {
2023 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2024 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2025 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2031 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2032 false, false, NULL, false);
2036 /* It does fld look up inside declare, and the result will be
2037 * added to fld cache, so the following fld lookup inside insert
2038 * does not need send RPC anymore, so avoid send rpc with holding
2040 if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2041 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2042 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2043 lu_object_fid(&dt->do_lu), range);
2049 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2050 struct lu_attr *attr,
2051 struct dt_allocation_hint *hint,
2052 struct dt_object_format *dof,
2055 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2056 struct osd_object *obj = osd_dt_obj(dt);
2057 struct osd_thread_info *info = osd_oti_get(env);
2062 LINVRNT(osd_invariant(obj));
2063 LASSERT(!dt_object_exists(dt));
2064 LASSERT(osd_write_locked(env, obj));
2065 LASSERT(th != NULL);
2067 if (unlikely(fid_is_acct(fid)))
2068 /* Quota files can't be created from the kernel any more,
2069 * 'tune2fs -O quota' will take care of creating them */
2072 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2073 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2075 result = __osd_object_create(info, obj, attr, hint, dof, th);
2077 result = __osd_oi_insert(env, obj, fid, th);
2079 LASSERT(ergo(result == 0, dt_object_exists(dt)));
2080 LASSERT(osd_invariant(obj));
2085 * Called to destroy on-disk representation of the object
2087 * Concurrency: must be locked
2089 static int osd_declare_object_destroy(const struct lu_env *env,
2090 struct dt_object *dt,
2093 struct osd_object *obj = osd_dt_obj(dt);
2094 struct inode *inode = obj->oo_inode;
2095 struct osd_thandle *oh;
2099 oh = container_of0(th, struct osd_thandle, ot_super);
2100 LASSERT(oh->ot_handle == NULL);
2103 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2104 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2105 /* XXX: So far, only normal fid needs to be inserted into the OI,
2106 * so only normal fid needs to be removed from the OI also.
2107 * Recycle idle OI leaf may cause additional three OI blocks
2109 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2110 fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2111 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0);
2113 /* one less inode */
2114 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2115 false, true, NULL, false);
2118 /* data to be truncated */
2119 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2120 true, true, NULL, false);
2124 static int osd_object_destroy(const struct lu_env *env,
2125 struct dt_object *dt,
2128 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2129 struct osd_object *obj = osd_dt_obj(dt);
2130 struct inode *inode = obj->oo_inode;
2131 struct osd_device *osd = osd_obj2dev(obj);
2132 struct osd_thandle *oh;
2136 oh = container_of0(th, struct osd_thandle, ot_super);
2137 LASSERT(oh->ot_handle);
2139 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2141 if (unlikely(fid_is_acct(fid)))
2144 /* Parallel control for OI scrub. For most of cases, there is no
2145 * lock contention. So it will not affect unlink performance. */
2146 mutex_lock(&inode->i_mutex);
2147 if (S_ISDIR(inode->i_mode)) {
2148 LASSERT(osd_inode_unlinked(inode) ||
2149 inode->i_nlink == 1);
2150 spin_lock(&obj->oo_guard);
2152 spin_unlock(&obj->oo_guard);
2153 inode->i_sb->s_op->dirty_inode(inode);
2156 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2158 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2159 mutex_unlock(&inode->i_mutex);
2161 /* XXX: add to ext3 orphan list */
2162 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2164 /* not needed in the cache anymore */
2165 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2171 * Helper function for osd_xattr_set()
2173 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2174 const struct lu_buf *buf, const char *name, int fl)
2176 struct osd_object *obj = osd_dt_obj(dt);
2177 struct inode *inode = obj->oo_inode;
2178 struct osd_thread_info *info = osd_oti_get(env);
2179 struct dentry *dentry = &info->oti_child_dentry;
2183 LASSERT(dt_object_exists(dt));
2184 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2186 if (fl & LU_XATTR_REPLACE)
2187 fs_flags |= XATTR_REPLACE;
2189 if (fl & LU_XATTR_CREATE)
2190 fs_flags |= XATTR_CREATE;
2192 ll_vfs_dq_init(inode);
2193 dentry->d_inode = inode;
2194 rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2195 buf->lb_len, fs_flags);
2200 * Put the fid into lustre_mdt_attrs, and then place the structure
2201 * inode's ea. This fid should not be altered during the life time
2204 * \retval +ve, on success
2205 * \retval -ve, on error
2207 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2209 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2210 const struct lu_fid *fid)
2212 struct osd_thread_info *info = osd_oti_get(env);
2213 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2215 lustre_lma_init(mdt_attrs, fid);
2216 lustre_lma_swab(mdt_attrs);
2217 return __osd_xattr_set(env, dt,
2218 osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2219 XATTR_NAME_LMA, LU_XATTR_CREATE);
2224 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2225 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2226 * To have compatilibility with 1.8 ldiskfs driver we need to have
2227 * magic number at start of fid data.
2228 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2231 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2232 const struct dt_rec *fid)
2234 /* XXX: replace the check with "!fid_is_client_mdt_visible()"
2235 * when FID in OI file introduced for local object. */
2236 if (!fid_is_norm((const struct lu_fid *)fid) &&
2237 !fid_is_igif((const struct lu_fid *)fid)) {
2238 param->edp_magic = 0;
2242 param->edp_magic = LDISKFS_LUFID_MAGIC;
2243 param->edp_len = sizeof(struct lu_fid) + 1;
2244 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2248 * Try to read the fid from inode ea into dt_rec, if return value
2249 * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2251 * \param fid object fid.
2253 * \retval 0 on success
2255 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2256 __u32 ino, struct lu_fid *fid,
2257 struct osd_inode_id *id)
2259 struct osd_thread_info *info = osd_oti_get(env);
2260 struct inode *inode;
2263 osd_id_gen(id, ino, OSD_OII_NOGEN);
2264 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2266 RETURN(PTR_ERR(inode));
2273 * OSD layer object create function for interoperability mode (b11826).
2274 * This is mostly similar to osd_object_create(). Only difference being, fid is
2275 * inserted into inode ea here.
2277 * \retval 0, on success
2278 * \retval -ve, on error
2280 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2281 struct lu_attr *attr,
2282 struct dt_allocation_hint *hint,
2283 struct dt_object_format *dof,
2286 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2287 struct osd_object *obj = osd_dt_obj(dt);
2288 struct osd_thread_info *info = osd_oti_get(env);
2293 LASSERT(osd_invariant(obj));
2294 LASSERT(!dt_object_exists(dt));
2295 LASSERT(osd_write_locked(env, obj));
2296 LASSERT(th != NULL);
2298 if (unlikely(fid_is_acct(fid)))
2299 /* Quota files can't be created from the kernel any more,
2300 * 'tune2fs -O quota' will take care of creating them */
2303 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2304 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2306 result = __osd_object_create(info, obj, attr, hint, dof, th);
2307 /* objects under osd root shld have igif fid, so dont add fid EA */
2308 /* For ost object, the fid will be stored during first write */
2309 if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL &&
2310 !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid))
2311 result = osd_ea_fid_set(env, dt, fid);
2314 result = __osd_oi_insert(env, obj, fid, th);
2316 LASSERT(ergo(result == 0, dt_object_exists(dt)));
2317 LINVRNT(osd_invariant(obj));
2321 static int osd_declare_object_ref_add(const struct lu_env *env,
2322 struct dt_object *dt,
2323 struct thandle *handle)
2325 struct osd_thandle *oh;
2327 /* it's possible that object doesn't exist yet */
2328 LASSERT(handle != NULL);
2330 oh = container_of0(handle, struct osd_thandle, ot_super);
2331 LASSERT(oh->ot_handle == NULL);
2333 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2334 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2340 * Concurrency: @dt is write locked.
2342 static int osd_object_ref_add(const struct lu_env *env,
2343 struct dt_object *dt, struct thandle *th)
2345 struct osd_object *obj = osd_dt_obj(dt);
2346 struct inode *inode = obj->oo_inode;
2348 LINVRNT(osd_invariant(obj));
2349 LASSERT(dt_object_exists(dt));
2350 LASSERT(osd_write_locked(env, obj));
2351 LASSERT(th != NULL);
2353 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2356 * DIR_NLINK feature is set for compatibility reasons if:
2357 * 1) nlinks > LDISKFS_LINK_MAX, or
2358 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2360 * It is easier to always set this flag (rather than check and set),
2361 * since it has less overhead, and the superblock will be dirtied
2362 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2363 * do not actually care whether this flag is set or not.
2365 spin_lock(&obj->oo_guard);
2366 /* inc_nlink from 0 may cause WARN_ON */
2367 if(inode->i_nlink == 0)
2368 set_nlink(inode, 1);
2371 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2372 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2373 inode->i_nlink == 2)
2374 set_nlink(inode, 1);
2376 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2377 spin_unlock(&obj->oo_guard);
2378 inode->i_sb->s_op->dirty_inode(inode);
2379 LINVRNT(osd_invariant(obj));
2384 static int osd_declare_object_ref_del(const struct lu_env *env,
2385 struct dt_object *dt,
2386 struct thandle *handle)
2388 struct osd_thandle *oh;
2390 LASSERT(dt_object_exists(dt));
2391 LASSERT(handle != NULL);
2393 oh = container_of0(handle, struct osd_thandle, ot_super);
2394 LASSERT(oh->ot_handle == NULL);
2396 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2397 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2403 * Concurrency: @dt is write locked.
2405 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2408 struct osd_object *obj = osd_dt_obj(dt);
2409 struct inode *inode = obj->oo_inode;
2411 LINVRNT(osd_invariant(obj));
2412 LASSERT(dt_object_exists(dt));
2413 LASSERT(osd_write_locked(env, obj));
2414 LASSERT(th != NULL);
2416 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2418 spin_lock(&obj->oo_guard);
2419 LASSERT(inode->i_nlink > 0);
2421 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2422 * then the nlink count is 1. Don't let it be set to 0 or the directory
2423 * inode will be deleted incorrectly. */
2424 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2425 set_nlink(inode, 1);
2426 spin_unlock(&obj->oo_guard);
2427 inode->i_sb->s_op->dirty_inode(inode);
2428 LINVRNT(osd_invariant(obj));
2434 * Get the 64-bit version for an inode.
2436 static int osd_object_version_get(const struct lu_env *env,
2437 struct dt_object *dt, dt_obj_version_t *ver)
2439 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2441 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2442 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2443 *ver = LDISKFS_I(inode)->i_fs_version;
2448 * Concurrency: @dt is read locked.
2450 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2451 struct lu_buf *buf, const char *name,
2452 struct lustre_capa *capa)
2454 struct osd_object *obj = osd_dt_obj(dt);
2455 struct inode *inode = obj->oo_inode;
2456 struct osd_thread_info *info = osd_oti_get(env);
2457 struct dentry *dentry = &info->oti_obj_dentry;
2459 /* version get is not real XATTR but uses xattr API */
2460 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2461 /* for version we are just using xattr API but change inode
2463 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2464 osd_object_version_get(env, dt, buf->lb_buf);
2465 return sizeof(dt_obj_version_t);
2468 LASSERT(dt_object_exists(dt));
2469 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2471 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2474 dentry->d_inode = inode;
2475 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2479 static int osd_declare_xattr_set(const struct lu_env *env,
2480 struct dt_object *dt,
2481 const struct lu_buf *buf, const char *name,
2482 int fl, struct thandle *handle)
2484 struct osd_thandle *oh;
2486 LASSERT(handle != NULL);
2488 oh = container_of0(handle, struct osd_thandle, ot_super);
2489 LASSERT(oh->ot_handle == NULL);
2491 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2492 strcmp(name, XATTR_NAME_VERSION) == 0 ?
2493 osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2494 osd_dto_credits_noquota[DTO_XATTR_SET]);
2500 * Set the 64-bit version for object
2502 static void osd_object_version_set(const struct lu_env *env,
2503 struct dt_object *dt,
2504 dt_obj_version_t *new_version)
2506 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2508 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2509 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2511 LDISKFS_I(inode)->i_fs_version = *new_version;
2512 /** Version is set after all inode operations are finished,
2513 * so we should mark it dirty here */
2514 inode->i_sb->s_op->dirty_inode(inode);
2518 * Concurrency: @dt is write locked.
2520 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2521 const struct lu_buf *buf, const char *name, int fl,
2522 struct thandle *handle, struct lustre_capa *capa)
2524 LASSERT(handle != NULL);
2526 /* version set is not real XATTR */
2527 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2528 /* for version we are just using xattr API but change inode
2530 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2531 osd_object_version_set(env, dt, buf->lb_buf);
2532 return sizeof(dt_obj_version_t);
2535 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2538 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2539 return __osd_xattr_set(env, dt, buf, name, fl);
2543 * Concurrency: @dt is read locked.
2545 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2546 struct lu_buf *buf, struct lustre_capa *capa)
2548 struct osd_object *obj = osd_dt_obj(dt);
2549 struct inode *inode = obj->oo_inode;
2550 struct osd_thread_info *info = osd_oti_get(env);
2551 struct dentry *dentry = &info->oti_obj_dentry;
2553 LASSERT(dt_object_exists(dt));
2554 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2555 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2557 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2560 dentry->d_inode = inode;
2561 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2564 static int osd_declare_xattr_del(const struct lu_env *env,
2565 struct dt_object *dt, const char *name,
2566 struct thandle *handle)
2568 struct osd_thandle *oh;
2570 LASSERT(dt_object_exists(dt));
2571 LASSERT(handle != NULL);
2573 oh = container_of0(handle, struct osd_thandle, ot_super);
2574 LASSERT(oh->ot_handle == NULL);
2576 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2577 osd_dto_credits_noquota[DTO_XATTR_SET]);
2583 * Concurrency: @dt is write locked.
2585 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2586 const char *name, struct thandle *handle,
2587 struct lustre_capa *capa)
2589 struct osd_object *obj = osd_dt_obj(dt);
2590 struct inode *inode = obj->oo_inode;
2591 struct osd_thread_info *info = osd_oti_get(env);
2592 struct dentry *dentry = &info->oti_obj_dentry;
2595 LASSERT(dt_object_exists(dt));
2596 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2597 LASSERT(osd_write_locked(env, obj));
2598 LASSERT(handle != NULL);
2600 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2603 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2605 ll_vfs_dq_init(inode);
2606 dentry->d_inode = inode;
2607 rc = inode->i_op->removexattr(dentry, name);
2611 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2612 struct dt_object *dt,
2613 struct lustre_capa *old,
2616 struct osd_thread_info *info = osd_oti_get(env);
2617 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2618 struct osd_object *obj = osd_dt_obj(dt);
2619 struct osd_device *dev = osd_obj2dev(obj);
2620 struct lustre_capa_key *key = &info->oti_capa_key;
2621 struct lustre_capa *capa = &info->oti_capa;
2622 struct obd_capa *oc;
2623 struct md_capainfo *ci;
2627 if (!dev->od_fl_capa)
2628 RETURN(ERR_PTR(-ENOENT));
2630 LASSERT(dt_object_exists(dt));
2631 LINVRNT(osd_invariant(obj));
2633 /* renewal sanity check */
2634 if (old && osd_object_auth(env, dt, old, opc))
2635 RETURN(ERR_PTR(-EACCES));
2637 ci = md_capainfo(env);
2639 RETURN(ERR_PTR(-ENOENT));
2641 switch (ci->mc_auth) {
2645 capa->lc_uid = obj->oo_inode->i_uid;
2646 capa->lc_gid = obj->oo_inode->i_gid;
2647 capa->lc_flags = LC_ID_PLAIN;
2649 case LC_ID_CONVERT: {
2652 s[0] = obj->oo_inode->i_uid;
2653 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2654 s[2] = obj->oo_inode->i_gid;
2655 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2656 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2658 RETURN(ERR_PTR(rc));
2660 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2661 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2662 capa->lc_flags = LC_ID_CONVERT;
2666 RETURN(ERR_PTR(-EINVAL));
2669 capa->lc_fid = *fid;
2671 capa->lc_flags |= dev->od_capa_alg << 24;
2672 capa->lc_timeout = dev->od_capa_timeout;
2673 capa->lc_expiry = 0;
2675 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2677 LASSERT(!capa_is_expired(oc));
2681 spin_lock(&capa_lock);
2682 *key = dev->od_capa_keys[1];
2683 spin_unlock(&capa_lock);
2685 capa->lc_keyid = key->lk_keyid;
2686 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2688 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2690 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2691 RETURN(ERR_PTR(rc));
2694 oc = capa_add(dev->od_capa_hash, capa);
2698 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2700 struct osd_object *obj = osd_dt_obj(dt);
2701 struct inode *inode = obj->oo_inode;
2702 struct osd_thread_info *info = osd_oti_get(env);
2703 struct dentry *dentry = &info->oti_obj_dentry;
2704 struct file *file = &info->oti_file;
2709 dentry->d_inode = inode;
2710 file->f_dentry = dentry;
2711 file->f_mapping = inode->i_mapping;
2712 file->f_op = inode->i_fop;
2713 mutex_lock(&inode->i_mutex);
2714 rc = file->f_op->fsync(file, dentry, 0);
2715 mutex_unlock(&inode->i_mutex);
2719 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2722 struct osd_object *obj = osd_dt_obj(dt);
2725 *data = (void *)obj->oo_inode;
2733 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2734 const struct dt_index_features *feat)
2736 struct iam_descr *descr;
2738 if (osd_object_is_root(o))
2739 return feat == &dt_directory_features;
2741 LASSERT(o->oo_dir != NULL);
2743 descr = o->oo_dir->od_container.ic_descr;
2744 if (feat == &dt_directory_features) {
2745 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2751 feat->dif_keysize_min <= descr->id_key_size &&
2752 descr->id_key_size <= feat->dif_keysize_max &&
2753 feat->dif_recsize_min <= descr->id_rec_size &&
2754 descr->id_rec_size <= feat->dif_recsize_max &&
2755 !(feat->dif_flags & (DT_IND_VARKEY |
2756 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2757 ergo(feat->dif_flags & DT_IND_UPDATE,
2758 1 /* XXX check that object (and file system) is
2763 static int osd_iam_container_init(const struct lu_env *env,
2764 struct osd_object *obj,
2765 struct osd_directory *dir)
2767 struct iam_container *bag = &dir->od_container;
2770 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2774 result = iam_container_setup(bag);
2776 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2778 iam_container_fini(bag);
2785 * Concurrency: no external locking is necessary.
2787 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2788 const struct dt_index_features *feat)
2792 struct osd_object *obj = osd_dt_obj(dt);
2794 LINVRNT(osd_invariant(obj));
2795 LASSERT(dt_object_exists(dt));
2797 if (osd_object_is_root(obj)) {
2798 dt->do_index_ops = &osd_index_ea_ops;
2800 } else if (feat == &dt_directory_features) {
2801 dt->do_index_ops = &osd_index_ea_ops;
2802 if (S_ISDIR(obj->oo_inode->i_mode))
2807 } else if (unlikely(feat == &dt_otable_features)) {
2808 dt->do_index_ops = &osd_otable_ops;
2810 } else if (unlikely(feat == &dt_acct_features)) {
2811 dt->do_index_ops = &osd_acct_index_ops;
2814 } else if (!osd_has_index(obj)) {
2815 struct osd_directory *dir;
2820 spin_lock(&obj->oo_guard);
2821 if (obj->oo_dir == NULL)
2825 * Concurrent thread allocated container data.
2828 spin_unlock(&obj->oo_guard);
2830 * Now, that we have container data, serialize its
2833 down_write(&obj->oo_ext_idx_sem);
2835 * recheck under lock.
2837 if (!osd_has_index(obj))
2838 result = osd_iam_container_init(env, obj, dir);
2841 up_write(&obj->oo_ext_idx_sem);
2849 if (result == 0 && skip_iam == 0) {
2850 if (!osd_iam_index_probe(env, obj, feat))
2853 LINVRNT(osd_invariant(obj));
2855 if (is_quota_glb_feat(feat))
2856 result = osd_quota_migration(env, dt, feat);
2861 static int osd_otable_it_attr_get(const struct lu_env *env,
2862 struct dt_object *dt,
2863 struct lu_attr *attr,
2864 struct lustre_capa *capa)
2870 static const struct dt_object_operations osd_obj_ops = {
2871 .do_read_lock = osd_object_read_lock,
2872 .do_write_lock = osd_object_write_lock,
2873 .do_read_unlock = osd_object_read_unlock,
2874 .do_write_unlock = osd_object_write_unlock,
2875 .do_write_locked = osd_object_write_locked,
2876 .do_attr_get = osd_attr_get,
2877 .do_declare_attr_set = osd_declare_attr_set,
2878 .do_attr_set = osd_attr_set,
2879 .do_ah_init = osd_ah_init,
2880 .do_declare_create = osd_declare_object_create,
2881 .do_create = osd_object_create,
2882 .do_declare_destroy = osd_declare_object_destroy,
2883 .do_destroy = osd_object_destroy,
2884 .do_index_try = osd_index_try,
2885 .do_declare_ref_add = osd_declare_object_ref_add,
2886 .do_ref_add = osd_object_ref_add,
2887 .do_declare_ref_del = osd_declare_object_ref_del,
2888 .do_ref_del = osd_object_ref_del,
2889 .do_xattr_get = osd_xattr_get,
2890 .do_declare_xattr_set = osd_declare_xattr_set,
2891 .do_xattr_set = osd_xattr_set,
2892 .do_declare_xattr_del = osd_declare_xattr_del,
2893 .do_xattr_del = osd_xattr_del,
2894 .do_xattr_list = osd_xattr_list,
2895 .do_capa_get = osd_capa_get,
2896 .do_object_sync = osd_object_sync,
2897 .do_data_get = osd_data_get,
2901 * dt_object_operations for interoperability mode
2902 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2904 static const struct dt_object_operations osd_obj_ea_ops = {
2905 .do_read_lock = osd_object_read_lock,
2906 .do_write_lock = osd_object_write_lock,
2907 .do_read_unlock = osd_object_read_unlock,
2908 .do_write_unlock = osd_object_write_unlock,
2909 .do_write_locked = osd_object_write_locked,
2910 .do_attr_get = osd_attr_get,
2911 .do_declare_attr_set = osd_declare_attr_set,
2912 .do_attr_set = osd_attr_set,
2913 .do_ah_init = osd_ah_init,
2914 .do_declare_create = osd_declare_object_create,
2915 .do_create = osd_object_ea_create,
2916 .do_declare_destroy = osd_declare_object_destroy,
2917 .do_destroy = osd_object_destroy,
2918 .do_index_try = osd_index_try,
2919 .do_declare_ref_add = osd_declare_object_ref_add,
2920 .do_ref_add = osd_object_ref_add,
2921 .do_declare_ref_del = osd_declare_object_ref_del,
2922 .do_ref_del = osd_object_ref_del,
2923 .do_xattr_get = osd_xattr_get,
2924 .do_declare_xattr_set = osd_declare_xattr_set,
2925 .do_xattr_set = osd_xattr_set,
2926 .do_declare_xattr_del = osd_declare_xattr_del,
2927 .do_xattr_del = osd_xattr_del,
2928 .do_xattr_list = osd_xattr_list,
2929 .do_capa_get = osd_capa_get,
2930 .do_object_sync = osd_object_sync,
2931 .do_data_get = osd_data_get,
2934 static const struct dt_object_operations osd_obj_otable_it_ops = {
2935 .do_attr_get = osd_otable_it_attr_get,
2936 .do_index_try = osd_index_try,
2939 static int osd_index_declare_iam_delete(const struct lu_env *env,
2940 struct dt_object *dt,
2941 const struct dt_key *key,
2942 struct thandle *handle)
2944 struct osd_thandle *oh;
2946 oh = container_of0(handle, struct osd_thandle, ot_super);
2947 LASSERT(oh->ot_handle == NULL);
2949 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2950 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2956 * delete a (key, value) pair from index \a dt specified by \a key
2958 * \param dt osd index object
2959 * \param key key for index
2960 * \param rec record reference
2961 * \param handle transaction handler
2964 * \retval -ve failure
2967 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2968 const struct dt_key *key,
2969 struct thandle *handle,
2970 struct lustre_capa *capa)
2972 struct osd_thread_info *oti = osd_oti_get(env);
2973 struct osd_object *obj = osd_dt_obj(dt);
2974 struct osd_thandle *oh;
2975 struct iam_path_descr *ipd;
2976 struct iam_container *bag = &obj->oo_dir->od_container;
2981 LINVRNT(osd_invariant(obj));
2982 LASSERT(dt_object_exists(dt));
2983 LASSERT(bag->ic_object == obj->oo_inode);
2984 LASSERT(handle != NULL);
2986 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2989 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
2991 ipd = osd_idx_ipd_get(env, bag);
2992 if (unlikely(ipd == NULL))
2995 oh = container_of0(handle, struct osd_thandle, ot_super);
2996 LASSERT(oh->ot_handle != NULL);
2997 LASSERT(oh->ot_handle->h_transaction != NULL);
2999 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3000 /* swab quota uid/gid provided by caller */
3001 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3002 key = (const struct dt_key *)&oti->oti_quota_id;
3005 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3006 osd_ipd_put(env, bag, ipd);
3007 LINVRNT(osd_invariant(obj));
3011 static int osd_index_declare_ea_delete(const struct lu_env *env,
3012 struct dt_object *dt,
3013 const struct dt_key *key,
3014 struct thandle *handle)
3016 struct osd_thandle *oh;
3017 struct inode *inode;
3021 LASSERT(dt_object_exists(dt));
3022 LASSERT(handle != NULL);
3024 oh = container_of0(handle, struct osd_thandle, ot_super);
3025 LASSERT(oh->ot_handle == NULL);
3027 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3028 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3030 inode = osd_dt_obj(dt)->oo_inode;
3033 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3034 true, true, NULL, false);
3038 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3041 struct osd_fid_pack *rec;
3044 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3045 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3046 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3052 * Index delete function for interoperability mode (b11826).
3053 * It will remove the directory entry added by osd_index_ea_insert().
3054 * This entry is needed to maintain name->fid mapping.
3056 * \param key, key i.e. file entry to be deleted
3058 * \retval 0, on success
3059 * \retval -ve, on error
3061 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3062 const struct dt_key *key,
3063 struct thandle *handle,
3064 struct lustre_capa *capa)
3066 struct osd_object *obj = osd_dt_obj(dt);
3067 struct inode *dir = obj->oo_inode;
3068 struct dentry *dentry;
3069 struct osd_thandle *oh;
3070 struct ldiskfs_dir_entry_2 *de;
3071 struct buffer_head *bh;
3072 struct htree_lock *hlock = NULL;
3077 LINVRNT(osd_invariant(obj));
3078 LASSERT(dt_object_exists(dt));
3079 LASSERT(handle != NULL);
3081 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3083 oh = container_of(handle, struct osd_thandle, ot_super);
3084 LASSERT(oh->ot_handle != NULL);
3085 LASSERT(oh->ot_handle->h_transaction != NULL);
3087 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3090 ll_vfs_dq_init(dir);
3091 dentry = osd_child_dentry_get(env, obj,
3092 (char *)key, strlen((char *)key));
3094 if (obj->oo_hl_head != NULL) {
3095 hlock = osd_oti_get(env)->oti_hlock;
3096 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3097 dir, LDISKFS_HLOCK_DEL);
3099 down_write(&obj->oo_ext_idx_sem);
3102 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3104 rc = ldiskfs_delete_entry(oh->ot_handle,
3111 ldiskfs_htree_unlock(hlock);
3113 up_write(&obj->oo_ext_idx_sem);
3115 LASSERT(osd_invariant(obj));
3120 * Lookup index for \a key and copy record to \a rec.
3122 * \param dt osd index object
3123 * \param key key for index
3124 * \param rec record reference
3126 * \retval +ve success : exact mach
3127 * \retval 0 return record with key not greater than \a key
3128 * \retval -ve failure
3130 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3131 struct dt_rec *rec, const struct dt_key *key,
3132 struct lustre_capa *capa)
3134 struct osd_object *obj = osd_dt_obj(dt);
3135 struct iam_path_descr *ipd;
3136 struct iam_container *bag = &obj->oo_dir->od_container;
3137 struct osd_thread_info *oti = osd_oti_get(env);
3138 struct iam_iterator *it = &oti->oti_idx_it;
3139 struct iam_rec *iam_rec;
3144 LASSERT(osd_invariant(obj));
3145 LASSERT(dt_object_exists(dt));
3146 LASSERT(bag->ic_object == obj->oo_inode);
3148 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3151 ipd = osd_idx_ipd_get(env, bag);
3155 /* got ipd now we can start iterator. */
3156 iam_it_init(it, bag, 0, ipd);
3158 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3159 /* swab quota uid/gid provided by caller */
3160 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3161 key = (const struct dt_key *)&oti->oti_quota_id;
3164 rc = iam_it_get(it, (struct iam_key *)key);
3166 if (S_ISDIR(obj->oo_inode->i_mode))
3167 iam_rec = (struct iam_rec *)oti->oti_ldp;
3169 iam_rec = (struct iam_rec *) rec;
3171 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3173 if (S_ISDIR(obj->oo_inode->i_mode))
3174 osd_fid_unpack((struct lu_fid *) rec,
3175 (struct osd_fid_pack *)iam_rec);
3176 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3177 osd_quota_unpack(obj, rec);
3182 osd_ipd_put(env, bag, ipd);
3184 LINVRNT(osd_invariant(obj));
3189 static int osd_index_declare_iam_insert(const struct lu_env *env,
3190 struct dt_object *dt,
3191 const struct dt_rec *rec,
3192 const struct dt_key *key,
3193 struct thandle *handle)
3195 struct osd_thandle *oh;
3197 LASSERT(dt_object_exists(dt));
3198 LASSERT(handle != NULL);
3200 oh = container_of0(handle, struct osd_thandle, ot_super);
3201 LASSERT(oh->ot_handle == NULL);
3203 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3204 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3210 * Inserts (key, value) pair in \a dt index object.
3212 * \param dt osd index object
3213 * \param key key for index
3214 * \param rec record reference
3215 * \param th transaction handler
3218 * \retval -ve failure
3220 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3221 const struct dt_rec *rec,
3222 const struct dt_key *key, struct thandle *th,
3223 struct lustre_capa *capa, int ignore_quota)
3225 struct osd_object *obj = osd_dt_obj(dt);
3226 struct iam_path_descr *ipd;
3227 struct osd_thandle *oh;
3228 struct iam_container *bag = &obj->oo_dir->od_container;
3229 struct osd_thread_info *oti = osd_oti_get(env);
3230 struct iam_rec *iam_rec;
3235 LINVRNT(osd_invariant(obj));
3236 LASSERT(dt_object_exists(dt));
3237 LASSERT(bag->ic_object == obj->oo_inode);
3238 LASSERT(th != NULL);
3240 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3243 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3245 ipd = osd_idx_ipd_get(env, bag);
3246 if (unlikely(ipd == NULL))
3249 oh = container_of0(th, struct osd_thandle, ot_super);
3250 LASSERT(oh->ot_handle != NULL);
3251 LASSERT(oh->ot_handle->h_transaction != NULL);
3252 if (S_ISDIR(obj->oo_inode->i_mode)) {
3253 iam_rec = (struct iam_rec *)oti->oti_ldp;
3254 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3255 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3256 /* pack quota uid/gid */
3257 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3258 key = (const struct dt_key *)&oti->oti_quota_id;
3259 /* pack quota record */
3260 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3261 iam_rec = (struct iam_rec *)rec;
3263 iam_rec = (struct iam_rec *)rec;
3266 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3268 osd_ipd_put(env, bag, ipd);
3269 LINVRNT(osd_invariant(obj));
3274 * Calls ldiskfs_add_entry() to add directory entry
3275 * into the directory. This is required for
3276 * interoperability mode (b11826)
3278 * \retval 0, on success
3279 * \retval -ve, on error
3281 static int __osd_ea_add_rec(struct osd_thread_info *info,
3282 struct osd_object *pobj, struct inode *cinode,
3283 const char *name, const struct dt_rec *fid,
3284 struct htree_lock *hlock, struct thandle *th)
3286 struct ldiskfs_dentry_param *ldp;
3287 struct dentry *child;
3288 struct osd_thandle *oth;
3291 oth = container_of(th, struct osd_thandle, ot_super);
3292 LASSERT(oth->ot_handle != NULL);
3293 LASSERT(oth->ot_handle->h_transaction != NULL);
3294 LASSERT(pobj->oo_inode);
3296 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3297 if (unlikely(pobj->oo_inode ==
3298 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3301 osd_get_ldiskfs_dirent_param(ldp, fid);
3302 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3303 child->d_fsdata = (void *)ldp;
3304 ll_vfs_dq_init(pobj->oo_inode);
3305 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3311 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3312 * into the directory.Also sets flags into osd object to
3313 * indicate dot and dotdot are created. This is required for
3314 * interoperability mode (b11826)
3316 * \param dir directory for dot and dotdot fixup.
3317 * \param obj child object for linking
3319 * \retval 0, on success
3320 * \retval -ve, on error
3322 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3323 struct osd_object *dir,
3324 struct inode *parent_dir, const char *name,
3325 const struct dt_rec *dot_fid,
3326 const struct dt_rec *dot_dot_fid,
3329 struct inode *inode = dir->oo_inode;
3330 struct ldiskfs_dentry_param *dot_ldp;
3331 struct ldiskfs_dentry_param *dot_dot_ldp;
3332 struct osd_thandle *oth;
3335 oth = container_of(th, struct osd_thandle, ot_super);
3336 LASSERT(oth->ot_handle->h_transaction != NULL);
3337 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3339 if (strcmp(name, dot) == 0) {
3340 if (dir->oo_compat_dot_created) {
3343 LASSERT(inode == parent_dir);
3344 dir->oo_compat_dot_created = 1;
3347 } else if(strcmp(name, dotdot) == 0) {
3348 if (!dir->oo_compat_dot_created)
3351 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3352 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3353 /* in case of rename, dotdot is already created */
3354 if (dir->oo_compat_dotdot_created)
3355 return __osd_ea_add_rec(info, dir, parent_dir, name,
3356 dot_dot_fid, NULL, th);
3358 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3359 dot_ldp->edp_magic = 0;
3360 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3361 inode, dot_ldp, dot_dot_ldp);
3363 dir->oo_compat_dotdot_created = 1;
3371 * It will call the appropriate osd_add* function and return the
3372 * value, return by respective functions.
3374 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3375 struct inode *cinode, const char *name,
3376 const struct dt_rec *fid, struct thandle *th)
3378 struct osd_thread_info *info = osd_oti_get(env);
3379 struct htree_lock *hlock;
3382 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3384 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3386 if (hlock != NULL) {
3387 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3390 down_write(&pobj->oo_ext_idx_sem);
3392 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3393 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3396 if (hlock != NULL) {
3397 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3398 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3400 down_write(&pobj->oo_ext_idx_sem);
3403 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3407 ldiskfs_htree_unlock(hlock);
3409 up_write(&pobj->oo_ext_idx_sem);
3415 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3416 struct osd_idmap_cache *oic)
3418 struct osd_scrub *scrub = &dev->od_scrub;
3419 struct lu_fid *fid = &oic->oic_fid;
3420 struct osd_inode_id *id = &oti->oti_id;
3425 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3429 rc = osd_oi_lookup(oti, dev, fid, id);
3430 if (rc != 0 && rc != -ENOENT)
3433 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3436 if (thread_is_running(&scrub->os_thread)) {
3437 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3438 /* There is race condition between osd_oi_lookup and OI scrub.
3439 * The OI scrub finished just after osd_oi_lookup() failure.
3440 * Under such case, it is unnecessary to trigger OI scrub again,
3441 * but try to call osd_oi_lookup() again. */
3442 if (unlikely(rc == -EAGAIN))
3448 if (!dev->od_noscrub && ++once == 1) {
3449 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3451 rc = osd_scrub_start(dev);
3452 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3454 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3464 * Calls ->lookup() to find dentry. From dentry get inode and
3465 * read inode's ea to get fid. This is required for interoperability
3468 * \retval 0, on success
3469 * \retval -ve, on error
3471 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3472 struct dt_rec *rec, const struct dt_key *key)
3474 struct inode *dir = obj->oo_inode;
3475 struct dentry *dentry;
3476 struct ldiskfs_dir_entry_2 *de;
3477 struct buffer_head *bh;
3478 struct lu_fid *fid = (struct lu_fid *) rec;
3479 struct htree_lock *hlock = NULL;
3484 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3486 dentry = osd_child_dentry_get(env, obj,
3487 (char *)key, strlen((char *)key));
3489 if (obj->oo_hl_head != NULL) {
3490 hlock = osd_oti_get(env)->oti_hlock;
3491 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3492 dir, LDISKFS_HLOCK_LOOKUP);
3494 down_read(&obj->oo_ext_idx_sem);
3497 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3499 struct osd_thread_info *oti = osd_oti_get(env);
3500 struct osd_idmap_cache *oic = &oti->oti_cache;
3501 struct osd_device *dev = osd_obj2dev(obj);
3502 struct osd_scrub *scrub = &dev->od_scrub;
3503 struct scrub_file *sf = &scrub->os_file;
3505 ino = le32_to_cpu(de->inode);
3506 rc = osd_get_fid_from_dentry(de, rec);
3508 /* done with de, release bh */
3511 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3513 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3515 fid_zero(&oic->oic_fid);
3519 oic->oic_fid = *fid;
3520 if ((scrub->os_pos_current <= ino) &&
3521 (sf->sf_flags & SF_INCONSISTENT ||
3522 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3524 osd_consistency_check(oti, dev, oic);
3533 ldiskfs_htree_unlock(hlock);
3535 up_read(&obj->oo_ext_idx_sem);
3540 * Find the osd object for given fid.
3542 * \param fid need to find the osd object having this fid
3544 * \retval osd_object on success
3545 * \retval -ve on error
3547 struct osd_object *osd_object_find(const struct lu_env *env,
3548 struct dt_object *dt,
3549 const struct lu_fid *fid)
3551 struct lu_device *ludev = dt->do_lu.lo_dev;
3552 struct osd_object *child = NULL;
3553 struct lu_object *luch;
3554 struct lu_object *lo;
3557 * at this point topdev might not exist yet
3558 * (i.e. MGS is preparing profiles). so we can
3559 * not rely on topdev and instead lookup with
3560 * our device passed as topdev. this can't work
3561 * if the object isn't cached yet (as osd doesn't
3562 * allocate lu_header). IOW, the object must be
3563 * in the cache, otherwise lu_object_alloc() crashes
3566 luch = lu_object_find_at(env, ludev, fid, NULL);
3567 if (!IS_ERR(luch)) {
3568 if (lu_object_exists(luch)) {
3569 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3571 child = osd_obj(lo);
3573 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3574 "lu_object can't be located"
3575 DFID"\n", PFID(fid));
3577 if (child == NULL) {
3578 lu_object_put(env, luch);
3579 CERROR("Unable to get osd_object\n");
3580 child = ERR_PTR(-ENOENT);
3583 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3584 "lu_object does not exists "DFID"\n",
3586 lu_object_put(env, luch);
3587 child = ERR_PTR(-ENOENT);
3590 child = (void *)luch;
3596 * Put the osd object once done with it.
3598 * \param obj osd object that needs to be put
3600 static inline void osd_object_put(const struct lu_env *env,
3601 struct osd_object *obj)
3603 lu_object_put(env, &obj->oo_dt.do_lu);
3606 static int osd_index_declare_ea_insert(const struct lu_env *env,
3607 struct dt_object *dt,
3608 const struct dt_rec *rec,
3609 const struct dt_key *key,
3610 struct thandle *handle)
3612 struct osd_thandle *oh;
3613 struct inode *inode;
3614 struct lu_fid *fid = (struct lu_fid *)rec;
3618 LASSERT(dt_object_exists(dt));
3619 LASSERT(handle != NULL);
3621 oh = container_of0(handle, struct osd_thandle, ot_super);
3622 LASSERT(oh->ot_handle == NULL);
3624 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3625 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3627 inode = osd_dt_obj(dt)->oo_inode;
3630 /* We ignore block quota on meta pool (MDTs), so needn't
3631 * calculate how many blocks will be consumed by this index
3633 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3634 true, true, NULL, false);
3638 /* It does fld look up inside declare, and the result will be
3639 * added to fld cache, so the following fld lookup inside insert
3640 * does not need send RPC anymore, so avoid send rpc with holding
3642 LASSERTF(fid_is_sane(fid), "fid is insane"DFID"\n", PFID(fid));
3643 osd_fld_lookup(env, osd_dt_dev(handle->th_dev), fid,
3644 &osd_oti_get(env)->oti_seq_range);
3650 * Index add function for interoperability mode (b11826).
3651 * It will add the directory entry.This entry is needed to
3652 * maintain name->fid mapping.
3654 * \param key it is key i.e. file entry to be inserted
3655 * \param rec it is value of given key i.e. fid
3657 * \retval 0, on success
3658 * \retval -ve, on error
3660 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3661 const struct dt_rec *rec,
3662 const struct dt_key *key, struct thandle *th,
3663 struct lustre_capa *capa, int ignore_quota)
3665 struct osd_object *obj = osd_dt_obj(dt);
3666 struct lu_fid *fid = (struct lu_fid *) rec;
3667 const char *name = (const char *)key;
3668 struct osd_object *child;
3673 LASSERT(osd_invariant(obj));
3674 LASSERT(dt_object_exists(dt));
3675 LASSERT(th != NULL);
3677 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3679 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3682 child = osd_object_find(env, dt, fid);
3683 if (!IS_ERR(child)) {
3684 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3685 osd_object_put(env, child);
3687 rc = PTR_ERR(child);
3690 LASSERT(osd_invariant(obj));
3695 * Initialize osd Iterator for given osd index object.
3697 * \param dt osd index object
3700 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3701 struct dt_object *dt,
3703 struct lustre_capa *capa)
3705 struct osd_it_iam *it;
3706 struct osd_thread_info *oti = osd_oti_get(env);
3707 struct osd_object *obj = osd_dt_obj(dt);
3708 struct lu_object *lo = &dt->do_lu;
3709 struct iam_path_descr *ipd;
3710 struct iam_container *bag = &obj->oo_dir->od_container;
3712 LASSERT(lu_object_exists(lo));
3714 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3715 return ERR_PTR(-EACCES);
3718 ipd = osd_it_ipd_get(env, bag);
3719 if (likely(ipd != NULL)) {
3723 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3724 return (struct dt_it *)it;
3726 return ERR_PTR(-ENOMEM);
3730 * free given Iterator.
3733 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3735 struct osd_it_iam *it = (struct osd_it_iam *)di;
3736 struct osd_object *obj = it->oi_obj;
3738 iam_it_fini(&it->oi_it);
3739 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3740 lu_object_put(env, &obj->oo_dt.do_lu);
3744 * Move Iterator to record specified by \a key
3746 * \param di osd iterator
3747 * \param key key for index
3749 * \retval +ve di points to record with least key not larger than key
3750 * \retval 0 di points to exact matched key
3751 * \retval -ve failure
3754 static int osd_it_iam_get(const struct lu_env *env,
3755 struct dt_it *di, const struct dt_key *key)
3757 struct osd_thread_info *oti = osd_oti_get(env);
3758 struct osd_it_iam *it = (struct osd_it_iam *)di;
3760 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3761 /* swab quota uid/gid */
3762 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3763 key = (struct dt_key *)&oti->oti_quota_id;
3766 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3772 * \param di osd iterator
3775 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3777 struct osd_it_iam *it = (struct osd_it_iam *)di;
3779 iam_it_put(&it->oi_it);
3783 * Move iterator by one record
3785 * \param di osd iterator
3787 * \retval +1 end of container reached
3789 * \retval -ve failure
3792 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3794 struct osd_it_iam *it = (struct osd_it_iam *)di;
3796 return iam_it_next(&it->oi_it);
3800 * Return pointer to the key under iterator.
3803 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3804 const struct dt_it *di)
3806 struct osd_thread_info *oti = osd_oti_get(env);
3807 struct osd_it_iam *it = (struct osd_it_iam *)di;
3808 struct osd_object *obj = it->oi_obj;
3811 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3813 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3814 /* swab quota uid/gid */
3815 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3816 key = (struct dt_key *)&oti->oti_quota_id;
3823 * Return size of key under iterator (in bytes)
3826 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3828 struct osd_it_iam *it = (struct osd_it_iam *)di;
3830 return iam_it_key_size(&it->oi_it);
3833 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3834 int len, __u16 type)
3836 struct luda_type *lt;
3837 const unsigned align = sizeof(struct luda_type) - 1;
3839 /* check if file type is required */
3840 if (attr & LUDA_TYPE) {
3841 len = (len + align) & ~align;
3843 lt = (void *) ent->lde_name + len;
3844 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3845 ent->lde_attrs |= LUDA_TYPE;
3848 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3852 * build lu direct from backend fs dirent.
3855 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3856 struct lu_fid *fid, __u64 offset,
3857 char *name, __u16 namelen,
3858 __u16 type, __u32 attr)
3860 fid_cpu_to_le(&ent->lde_fid, fid);
3861 ent->lde_attrs = LUDA_FID;
3863 ent->lde_hash = cpu_to_le64(offset);
3864 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3866 strncpy(ent->lde_name, name, namelen);
3867 ent->lde_namelen = cpu_to_le16(namelen);
3869 /* append lustre attributes */
3870 osd_it_append_attrs(ent, attr, namelen, type);
3874 * Return pointer to the record under iterator.
3876 static int osd_it_iam_rec(const struct lu_env *env,
3877 const struct dt_it *di,
3878 struct dt_rec *dtrec, __u32 attr)
3880 struct osd_it_iam *it = (struct osd_it_iam *)di;
3881 struct osd_thread_info *info = osd_oti_get(env);
3884 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3885 const struct osd_fid_pack *rec;
3886 struct lu_fid *fid = &info->oti_fid;
3887 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3893 name = (char *)iam_it_key_get(&it->oi_it);
3895 RETURN(PTR_ERR(name));
3897 namelen = iam_it_key_size(&it->oi_it);
3899 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3901 RETURN(PTR_ERR(rec));
3903 rc = osd_fid_unpack(fid, rec);
3907 hash = iam_it_store(&it->oi_it);
3909 /* IAM does not store object type in IAM index (dir) */
3910 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3912 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3913 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3914 (struct iam_rec *)dtrec);
3915 osd_quota_unpack(it->oi_obj, dtrec);
3917 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3918 (struct iam_rec *)dtrec);
3925 * Returns cookie for current Iterator position.
3927 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3929 struct osd_it_iam *it = (struct osd_it_iam *)di;
3931 return iam_it_store(&it->oi_it);
3935 * Restore iterator from cookie.
3937 * \param di osd iterator
3938 * \param hash Iterator location cookie
3940 * \retval +ve di points to record with least key not larger than key.
3941 * \retval 0 di points to exact matched key
3942 * \retval -ve failure
3945 static int osd_it_iam_load(const struct lu_env *env,
3946 const struct dt_it *di, __u64 hash)
3948 struct osd_it_iam *it = (struct osd_it_iam *)di;
3950 return iam_it_load(&it->oi_it, hash);
3953 static const struct dt_index_operations osd_index_iam_ops = {
3954 .dio_lookup = osd_index_iam_lookup,
3955 .dio_declare_insert = osd_index_declare_iam_insert,
3956 .dio_insert = osd_index_iam_insert,
3957 .dio_declare_delete = osd_index_declare_iam_delete,
3958 .dio_delete = osd_index_iam_delete,
3960 .init = osd_it_iam_init,
3961 .fini = osd_it_iam_fini,
3962 .get = osd_it_iam_get,
3963 .put = osd_it_iam_put,
3964 .next = osd_it_iam_next,
3965 .key = osd_it_iam_key,
3966 .key_size = osd_it_iam_key_size,
3967 .rec = osd_it_iam_rec,
3968 .store = osd_it_iam_store,
3969 .load = osd_it_iam_load
3975 * Creates or initializes iterator context.
3977 * \retval struct osd_it_ea, iterator structure on success
3980 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3981 struct dt_object *dt,
3983 struct lustre_capa *capa)
3985 struct osd_object *obj = osd_dt_obj(dt);
3986 struct osd_thread_info *info = osd_oti_get(env);
3987 struct osd_it_ea *it = &info->oti_it_ea;
3988 struct lu_object *lo = &dt->do_lu;
3989 struct dentry *obj_dentry = &info->oti_it_dentry;
3991 LASSERT(lu_object_exists(lo));
3993 obj_dentry->d_inode = obj->oo_inode;
3994 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3995 obj_dentry->d_name.hash = 0;
3997 it->oie_rd_dirent = 0;
3998 it->oie_it_dirent = 0;
3999 it->oie_dirent = NULL;
4000 it->oie_buf = info->oti_it_ea_buf;
4002 it->oie_file.f_pos = 0;
4003 it->oie_file.f_dentry = obj_dentry;
4004 if (attr & LUDA_64BITHASH)
4005 it->oie_file.f_mode |= FMODE_64BITHASH;
4007 it->oie_file.f_mode |= FMODE_32BITHASH;
4008 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
4009 it->oie_file.f_op = obj->oo_inode->i_fop;
4010 it->oie_file.private_data = NULL;
4012 RETURN((struct dt_it *) it);
4016 * Destroy or finishes iterator context.
4018 * \param di iterator structure to be destroyed
4020 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4022 struct osd_it_ea *it = (struct osd_it_ea *)di;
4023 struct osd_object *obj = it->oie_obj;
4024 struct inode *inode = obj->oo_inode;
4027 it->oie_file.f_op->release(inode, &it->oie_file);
4028 lu_object_put(env, &obj->oo_dt.do_lu);
4033 * It position the iterator at given key, so that next lookup continues from
4034 * that key Or it is similar to dio_it->load() but based on a key,
4035 * rather than file position.
4037 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4040 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4042 static int osd_it_ea_get(const struct lu_env *env,
4043 struct dt_it *di, const struct dt_key *key)
4045 struct osd_it_ea *it = (struct osd_it_ea *)di;
4048 LASSERT(((const char *)key)[0] == '\0');
4049 it->oie_file.f_pos = 0;
4050 it->oie_rd_dirent = 0;
4051 it->oie_it_dirent = 0;
4052 it->oie_dirent = NULL;
4060 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4065 * It is called internally by ->readdir(). It fills the
4066 * iterator's in-memory data structure with required
4067 * information i.e. name, namelen, rec_size etc.
4069 * \param buf in which information to be filled in.
4070 * \param name name of the file in given dir
4072 * \retval 0 on success
4073 * \retval 1 on buffer full
4075 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4076 loff_t offset, __u64 ino,
4079 struct osd_it_ea *it = (struct osd_it_ea *)buf;
4080 struct osd_object *obj = it->oie_obj;
4081 struct osd_it_ea_dirent *ent = it->oie_dirent;
4082 struct lu_fid *fid = &ent->oied_fid;
4083 struct osd_fid_pack *rec;
4086 /* this should never happen */
4087 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4088 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4092 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4096 /* "." is just the object itself. */
4097 if (namelen == 1 && name[0] == '.') {
4098 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4099 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4100 rec = (struct osd_fid_pack*) (name + namelen + 1);
4101 if (osd_fid_unpack(fid, rec) != 0)
4106 d_type &= ~LDISKFS_DIRENT_LUFID;
4108 /* NOT export local root. */
4109 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4110 ino = obj->oo_inode->i_ino;
4111 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4114 ent->oied_ino = ino;
4115 ent->oied_off = offset;
4116 ent->oied_namelen = namelen;
4117 ent->oied_type = d_type;
4119 memcpy(ent->oied_name, name, namelen);
4121 it->oie_rd_dirent++;
4122 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4127 * Calls ->readdir() to load a directory entry at a time
4128 * and stored it in iterator's in-memory data structure.
4130 * \param di iterator's in memory structure
4132 * \retval 0 on success
4133 * \retval -ve on error
4135 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4136 const struct dt_it *di)
4138 struct osd_it_ea *it = (struct osd_it_ea *)di;
4139 struct osd_object *obj = it->oie_obj;
4140 struct inode *inode = obj->oo_inode;
4141 struct htree_lock *hlock = NULL;
4145 it->oie_dirent = it->oie_buf;
4146 it->oie_rd_dirent = 0;
4148 if (obj->oo_hl_head != NULL) {
4149 hlock = osd_oti_get(env)->oti_hlock;
4150 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4151 inode, LDISKFS_HLOCK_READDIR);
4153 down_read(&obj->oo_ext_idx_sem);
4156 result = inode->i_fop->readdir(&it->oie_file, it,
4157 (filldir_t) osd_ldiskfs_filldir);
4160 ldiskfs_htree_unlock(hlock);
4162 up_read(&obj->oo_ext_idx_sem);
4164 if (it->oie_rd_dirent == 0) {
4167 it->oie_dirent = it->oie_buf;
4168 it->oie_it_dirent = 1;
4175 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4176 * to load a directory entry at a time and stored it in
4177 * iterator's in-memory data structure.
4179 * \param di iterator's in memory structure
4181 * \retval +ve iterator reached to end
4182 * \retval 0 iterator not reached to end
4183 * \retval -ve on error
4185 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4187 struct osd_it_ea *it = (struct osd_it_ea *)di;
4192 if (it->oie_it_dirent < it->oie_rd_dirent) {
4194 (void *) it->oie_dirent +
4195 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4196 it->oie_dirent->oied_namelen);
4197 it->oie_it_dirent++;
4200 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4203 rc = osd_ldiskfs_it_fill(env, di);
4210 * Returns the key at current position from iterator's in memory structure.
4212 * \param di iterator's in memory structure
4214 * \retval key i.e. struct dt_key on success
4216 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4217 const struct dt_it *di)
4219 struct osd_it_ea *it = (struct osd_it_ea *)di;
4221 return (struct dt_key *)it->oie_dirent->oied_name;
4225 * Returns the key's size at current position from iterator's in memory structure.
4227 * \param di iterator's in memory structure
4229 * \retval key_size i.e. struct dt_key on success
4231 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4233 struct osd_it_ea *it = (struct osd_it_ea *)di;
4235 return it->oie_dirent->oied_namelen;
4240 * Returns the value (i.e. fid/igif) at current position from iterator's
4241 * in memory structure.
4243 * \param di struct osd_it_ea, iterator's in memory structure
4244 * \param attr attr requested for dirent.
4245 * \param lde lustre dirent
4247 * \retval 0 no error and \param lde has correct lustre dirent.
4248 * \retval -ve on error
4250 static inline int osd_it_ea_rec(const struct lu_env *env,
4251 const struct dt_it *di,
4252 struct dt_rec *dtrec, __u32 attr)
4254 struct osd_it_ea *it = (struct osd_it_ea *)di;
4255 struct osd_object *obj = it->oie_obj;
4256 struct osd_device *dev = osd_obj2dev(obj);
4257 struct osd_scrub *scrub = &dev->od_scrub;
4258 struct scrub_file *sf = &scrub->os_file;
4259 struct osd_thread_info *oti = osd_oti_get(env);
4260 struct osd_idmap_cache *oic = &oti->oti_cache;
4261 struct lu_fid *fid = &it->oie_dirent->oied_fid;
4262 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4263 __u32 ino = it->oie_dirent->oied_ino;
4267 if (!fid_is_sane(fid)) {
4268 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4270 fid_zero(&oic->oic_fid);
4274 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4277 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4278 it->oie_dirent->oied_name,
4279 it->oie_dirent->oied_namelen,
4280 it->oie_dirent->oied_type, attr);
4281 oic->oic_fid = *fid;
4282 if ((scrub->os_pos_current <= ino) &&
4283 (sf->sf_flags & SF_INCONSISTENT ||
4284 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4285 osd_consistency_check(oti, dev, oic);
4291 * Returns a cookie for current position of the iterator head, so that
4292 * user can use this cookie to load/start the iterator next time.
4294 * \param di iterator's in memory structure
4296 * \retval cookie for current position, on success
4298 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4300 struct osd_it_ea *it = (struct osd_it_ea *)di;
4302 return it->oie_dirent->oied_off;
4306 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4307 * to load a directory entry at a time and stored it i inn,
4308 * in iterator's in-memory data structure.
4310 * \param di struct osd_it_ea, iterator's in memory structure
4312 * \retval +ve on success
4313 * \retval -ve on error
4315 static int osd_it_ea_load(const struct lu_env *env,
4316 const struct dt_it *di, __u64 hash)
4318 struct osd_it_ea *it = (struct osd_it_ea *)di;
4322 it->oie_file.f_pos = hash;
4324 rc = osd_ldiskfs_it_fill(env, di);
4332 * Index lookup function for interoperability mode (b11826).
4334 * \param key, key i.e. file name to be searched
4336 * \retval +ve, on success
4337 * \retval -ve, on error
4339 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4340 struct dt_rec *rec, const struct dt_key *key,
4341 struct lustre_capa *capa)
4343 struct osd_object *obj = osd_dt_obj(dt);
4348 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4349 LINVRNT(osd_invariant(obj));
4351 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4354 rc = osd_ea_lookup_rec(env, obj, rec, key);
4361 * Index and Iterator operations for interoperability
4362 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4364 static const struct dt_index_operations osd_index_ea_ops = {
4365 .dio_lookup = osd_index_ea_lookup,
4366 .dio_declare_insert = osd_index_declare_ea_insert,
4367 .dio_insert = osd_index_ea_insert,
4368 .dio_declare_delete = osd_index_declare_ea_delete,
4369 .dio_delete = osd_index_ea_delete,
4371 .init = osd_it_ea_init,
4372 .fini = osd_it_ea_fini,
4373 .get = osd_it_ea_get,
4374 .put = osd_it_ea_put,
4375 .next = osd_it_ea_next,
4376 .key = osd_it_ea_key,
4377 .key_size = osd_it_ea_key_size,
4378 .rec = osd_it_ea_rec,
4379 .store = osd_it_ea_store,
4380 .load = osd_it_ea_load
4384 static void *osd_key_init(const struct lu_context *ctx,
4385 struct lu_context_key *key)
4387 struct osd_thread_info *info;
4389 OBD_ALLOC_PTR(info);
4391 return ERR_PTR(-ENOMEM);
4393 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4394 if (info->oti_it_ea_buf == NULL)
4397 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4399 info->oti_hlock = ldiskfs_htree_lock_alloc();
4400 if (info->oti_hlock == NULL)
4406 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4409 return ERR_PTR(-ENOMEM);
4412 static void osd_key_fini(const struct lu_context *ctx,
4413 struct lu_context_key *key, void* data)
4415 struct osd_thread_info *info = data;
4417 if (info->oti_hlock != NULL)
4418 ldiskfs_htree_lock_free(info->oti_hlock);
4419 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4423 static void osd_key_exit(const struct lu_context *ctx,
4424 struct lu_context_key *key, void *data)
4426 struct osd_thread_info *info = data;
4428 LASSERT(info->oti_r_locks == 0);
4429 LASSERT(info->oti_w_locks == 0);
4430 LASSERT(info->oti_txns == 0);
4433 /* type constructor/destructor: osd_type_init, osd_type_fini */
4434 LU_TYPE_INIT_FINI(osd, &osd_key);
4436 struct lu_context_key osd_key = {
4437 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4438 .lct_init = osd_key_init,
4439 .lct_fini = osd_key_fini,
4440 .lct_exit = osd_key_exit
4444 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4445 const char *name, struct lu_device *next)
4447 struct osd_device *osd = osd_dev(d);
4449 strncpy(osd->od_svname, name, MAX_OBD_NAME);
4450 return osd_procfs_init(osd, name);
4453 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4457 osd_scrub_cleanup(env, o);
4460 fsfilt_put_ops(o->od_fsops);
4464 /* shutdown quota slave instance associated with the device */
4465 if (o->od_quota_slave != NULL) {
4466 qsd_fini(env, o->od_quota_slave);
4467 o->od_quota_slave = NULL;
4473 static int osd_mount(const struct lu_env *env,
4474 struct osd_device *o, struct lustre_cfg *cfg)
4476 const char *name = lustre_cfg_string(cfg, 0);
4477 const char *dev = lustre_cfg_string(cfg, 1);
4479 unsigned long page, s_flags, lmd_flags = 0;
4480 struct page *__page;
4481 struct file_system_type *type;
4482 char *options = NULL;
4487 if (o->od_mnt != NULL)
4490 if (strlen(dev) >= sizeof(o->od_mntdev))
4492 strcpy(o->od_mntdev, dev);
4494 o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4495 if (o->od_fsops == NULL) {
4496 CERROR("Can't find fsfilt_ldiskfs\n");
4500 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4502 GOTO(out, rc = -ENOMEM);
4504 str = lustre_cfg_string(cfg, 2);
4505 s_flags = simple_strtoul(str, NULL, 0);
4506 str = strstr(str, ":");
4508 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4509 opts = lustre_cfg_string(cfg, 3);
4510 page = (unsigned long)cfs_page_address(__page);
4511 options = (char *)page;
4514 strcat(options, "user_xattr,acl");
4516 strcat(options, opts);
4518 /* Glom up mount options */
4519 if (*options != '\0')
4520 strcat(options, ",");
4521 strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4523 type = get_fs_type("ldiskfs");
4525 CERROR("%s: cannot find ldiskfs module\n", name);
4526 GOTO(out, rc = -ENODEV);
4529 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4530 cfs_module_put(type->owner);
4532 if (IS_ERR(o->od_mnt)) {
4533 rc = PTR_ERR(o->od_mnt);
4534 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4539 if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4540 CERROR("%s: underlying device %s is marked as read-only. "
4541 "Setup failed\n", name, dev);
4544 GOTO(out, rc = -EROFS);
4547 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4548 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4549 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4552 GOTO(out, rc = -EINVAL);
4555 if (lmd_flags & LMD_FLG_NOSCRUB)
4560 OBD_PAGE_FREE(__page);
4562 fsfilt_put_ops(o->od_fsops);
4567 static struct lu_device *osd_device_fini(const struct lu_env *env,
4568 struct lu_device *d)
4573 rc = osd_shutdown(env, osd_dev(d));
4575 osd_obj_map_fini(osd_dev(d));
4577 shrink_dcache_sb(osd_sb(osd_dev(d)));
4578 osd_sync(env, lu2dt_dev(d));
4580 rc = osd_procfs_fini(osd_dev(d));
4582 CERROR("proc fini error %d \n", rc);
4583 RETURN (ERR_PTR(rc));
4586 if (osd_dev(d)->od_mnt) {
4587 mntput(osd_dev(d)->od_mnt);
4588 osd_dev(d)->od_mnt = NULL;
4594 static int osd_device_init0(const struct lu_env *env,
4595 struct osd_device *o,
4596 struct lustre_cfg *cfg)
4598 struct lu_device *l = osd2lu_dev(o);
4599 struct osd_thread_info *info;
4602 /* if the module was re-loaded, env can loose its keys */
4603 rc = lu_env_refill((struct lu_env *) env);
4606 info = osd_oti_get(env);
4609 l->ld_ops = &osd_lu_ops;
4610 o->od_dt_dev.dd_ops = &osd_dt_ops;
4612 spin_lock_init(&o->od_osfs_lock);
4613 mutex_init(&o->od_otable_mutex);
4614 o->od_osfs_age = cfs_time_shift_64(-1000);
4616 o->od_capa_hash = init_capa_hash();
4617 if (o->od_capa_hash == NULL)
4618 GOTO(out, rc = -ENOMEM);
4620 o->od_read_cache = 1;
4621 o->od_writethrough_cache = 1;
4622 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
4624 rc = osd_mount(env, o, cfg);
4628 /* setup scrub, including OI files initialization */
4629 rc = osd_scrub_setup(env, o);
4633 strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4634 sizeof(o->od_svname) - 1);
4636 rc = osd_obj_map_init(o);
4638 GOTO(out_scrub, rc);
4640 rc = lu_site_init(&o->od_site, l);
4642 GOTO(out_compat, rc);
4643 o->od_site.ls_bottom_dev = l;
4645 rc = lu_site_init_finish(&o->od_site);
4649 rc = osd_procfs_init(o, o->od_svname);
4651 CERROR("%s: can't initialize procfs: rc = %d\n",
4656 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4658 /* initialize quota slave instance */
4659 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
4661 if (IS_ERR(o->od_quota_slave)) {
4662 rc = PTR_ERR(o->od_quota_slave);
4663 o->od_quota_slave = NULL;
4664 GOTO(out_procfs, rc);
4671 lu_site_fini(&o->od_site);
4673 osd_obj_map_fini(o);
4675 osd_scrub_cleanup(env, o);
4677 osd_oi_fini(info, o);
4678 osd_shutdown(env, o);
4682 cleanup_capa_hash(o->od_capa_hash);
4687 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4688 struct lu_device_type *t,
4689 struct lustre_cfg *cfg)
4691 struct osd_device *o;
4696 return ERR_PTR(-ENOMEM);
4698 rc = dt_device_init(&o->od_dt_dev, t);
4700 /* Because the ctx might be revived in dt_device_init,
4701 * refill the env here */
4702 lu_env_refill((struct lu_env *)env);
4703 rc = osd_device_init0(env, o, cfg);
4705 dt_device_fini(&o->od_dt_dev);
4708 if (unlikely(rc != 0))
4711 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4714 static struct lu_device *osd_device_free(const struct lu_env *env,
4715 struct lu_device *d)
4717 struct osd_device *o = osd_dev(d);
4720 cleanup_capa_hash(o->od_capa_hash);
4721 /* XXX: make osd top device in order to release reference */
4722 d->ld_site->ls_top_dev = d;
4723 lu_site_purge(env, d->ld_site, -1);
4724 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4725 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4726 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4728 lu_site_fini(&o->od_site);
4729 dt_device_fini(&o->od_dt_dev);
4734 static int osd_process_config(const struct lu_env *env,
4735 struct lu_device *d, struct lustre_cfg *cfg)
4737 struct osd_device *o = osd_dev(d);
4741 switch(cfg->lcfg_command) {
4743 err = osd_mount(env, o, cfg);
4746 lu_dev_del_linkage(d->ld_site, d);
4747 err = osd_shutdown(env, o);
4756 static int osd_recovery_complete(const struct lu_env *env,
4757 struct lu_device *d)
4759 struct osd_device *osd = osd_dev(d);
4763 if (osd->od_quota_slave == NULL)
4766 /* start qsd instance on recovery completion, this notifies the quota
4767 * slave code that we are about to process new requests now */
4768 rc = qsd_start(env, osd->od_quota_slave);
4773 * we use exports to track all osd users
4775 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4776 struct obd_device *obd, struct obd_uuid *cluuid,
4777 struct obd_connect_data *data, void *localdata)
4779 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4780 struct lustre_handle conn;
4784 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4786 rc = class_connect(&conn, obd, cluuid);
4790 *exp = class_conn2export(&conn);
4792 spin_lock(&osd->od_osfs_lock);
4794 spin_unlock(&osd->od_osfs_lock);
4800 * once last export (we don't count self-export) disappeared
4801 * osd can be released
4803 static int osd_obd_disconnect(struct obd_export *exp)
4805 struct obd_device *obd = exp->exp_obd;
4806 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4807 int rc, release = 0;
4810 /* Only disconnect the underlying layers on the final disconnect. */
4811 spin_lock(&osd->od_osfs_lock);
4813 if (osd->od_connects == 0)
4815 spin_unlock(&osd->od_osfs_lock);
4817 rc = class_disconnect(exp); /* bz 9811 */
4819 if (rc == 0 && release)
4820 class_manual_cleanup(obd);
4824 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4825 struct lu_device *dev)
4827 struct osd_device *osd = osd_dev(dev);
4831 if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
4832 /* MDT/MDD still use old infrastructure to create
4834 result = llo_local_objects_setup(env, lu2md_dev(pdev),
4840 if (osd->od_quota_slave != NULL)
4841 /* set up quota slave objects */
4842 result = qsd_prepare(env, osd->od_quota_slave);
4847 static const struct lu_object_operations osd_lu_obj_ops = {
4848 .loo_object_init = osd_object_init,
4849 .loo_object_delete = osd_object_delete,
4850 .loo_object_release = osd_object_release,
4851 .loo_object_free = osd_object_free,
4852 .loo_object_print = osd_object_print,
4853 .loo_object_invariant = osd_object_invariant
4856 const struct lu_device_operations osd_lu_ops = {
4857 .ldo_object_alloc = osd_object_alloc,
4858 .ldo_process_config = osd_process_config,
4859 .ldo_recovery_complete = osd_recovery_complete,
4860 .ldo_prepare = osd_prepare,
4863 static const struct lu_device_type_operations osd_device_type_ops = {
4864 .ldto_init = osd_type_init,
4865 .ldto_fini = osd_type_fini,
4867 .ldto_start = osd_type_start,
4868 .ldto_stop = osd_type_stop,
4870 .ldto_device_alloc = osd_device_alloc,
4871 .ldto_device_free = osd_device_free,
4873 .ldto_device_init = osd_device_init,
4874 .ldto_device_fini = osd_device_fini
4877 struct lu_device_type osd_device_type = {
4878 .ldt_tags = LU_DEVICE_DT,
4879 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
4880 .ldt_ops = &osd_device_type_ops,
4881 .ldt_ctx_tags = LCT_LOCAL,
4885 * lprocfs legacy support.
4887 static struct obd_ops osd_obd_device_ops = {
4888 .o_owner = THIS_MODULE,
4889 .o_connect = osd_obd_connect,
4890 .o_disconnect = osd_obd_disconnect
4893 static int __init osd_mod_init(void)
4895 struct lprocfs_static_vars lvars;
4898 lprocfs_osd_init_vars(&lvars);
4899 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4900 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4903 static void __exit osd_mod_exit(void)
4905 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4908 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4909 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4910 MODULE_LICENSE("GPL");
4912 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);