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>
58 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
66 #include "osd_internal.h"
68 /* llo_* api support */
69 #include <md_object.h>
70 #include <lustre_quota.h>
73 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
74 "ldiskfs with parallel directory operations");
76 static const char dot[] = ".";
77 static const char dotdot[] = "..";
78 static const char remote_obj_dir[] = "REM_OBJ_DIR";
80 static const struct lu_object_operations osd_lu_obj_ops;
81 static const struct dt_object_operations osd_obj_ops;
82 static const struct dt_object_operations osd_obj_ea_ops;
83 static const struct dt_object_operations osd_obj_otable_it_ops;
84 static const struct dt_index_operations osd_index_iam_ops;
85 static const struct dt_index_operations osd_index_ea_ops;
87 #ifdef OSD_TRACK_DECLARES
88 int osd_trans_declare_op2rb[] = {
89 [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET,
90 [OSD_OT_PUNCH] = OSD_OT_MAX,
91 [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET,
92 [OSD_OT_CREATE] = OSD_OT_DESTROY,
93 [OSD_OT_DESTROY] = OSD_OT_CREATE,
94 [OSD_OT_REF_ADD] = OSD_OT_REF_DEL,
95 [OSD_OT_REF_DEL] = OSD_OT_REF_ADD,
96 [OSD_OT_WRITE] = OSD_OT_WRITE,
97 [OSD_OT_INSERT] = OSD_OT_DELETE,
98 [OSD_OT_DELETE] = OSD_OT_INSERT,
99 [OSD_OT_QUOTA] = OSD_OT_MAX,
103 static int osd_has_index(const struct osd_object *obj)
105 return obj->oo_dt.do_index_ops != NULL;
108 static int osd_object_invariant(const struct lu_object *l)
110 return osd_invariant(osd_obj(l));
114 * Concurrency: doesn't matter
116 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
118 return osd_oti_get(env)->oti_r_locks > 0;
122 * Concurrency: doesn't matter
124 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
126 struct osd_thread_info *oti = osd_oti_get(env);
127 return oti->oti_w_locks > 0 && o->oo_owner == env;
131 * Concurrency: doesn't access mutable data
133 static int osd_root_get(const struct lu_env *env,
134 struct dt_device *dev, struct lu_fid *f)
136 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
141 * OSD object methods.
145 * Concurrency: no concurrent access is possible that early in object
148 static struct lu_object *osd_object_alloc(const struct lu_env *env,
149 const struct lu_object_header *hdr,
152 struct osd_object *mo;
158 l = &mo->oo_dt.do_lu;
159 dt_object_init(&mo->oo_dt, NULL, d);
160 mo->oo_dt.do_ops = &osd_obj_ea_ops;
161 l->lo_ops = &osd_lu_obj_ops;
162 init_rwsem(&mo->oo_sem);
163 init_rwsem(&mo->oo_ext_idx_sem);
164 spin_lock_init(&mo->oo_guard);
171 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
172 const char *name, void *buf, int len)
174 dentry->d_inode = inode;
175 return inode->i_op->getxattr(dentry, name, buf, len);
178 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
179 struct dentry *dentry, struct lustre_mdt_attrs *lma)
183 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma,
186 /* try with old lma size */
187 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
188 info->oti_mdt_attrs_old,
191 memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
194 /* Check LMA compatibility */
195 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
196 CWARN("%.16s: unsupported incompat LMA feature(s) "
198 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
199 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
203 lustre_lma_swab(lma);
206 } else if (rc == 0) {
214 * retrieve object from backend ext fs.
216 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
217 struct osd_inode_id *id)
219 struct inode *inode = NULL;
221 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
223 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
224 id->oii_ino, PTR_ERR(inode));
225 } else if (id->oii_gen != OSD_OII_NOGEN &&
226 inode->i_generation != id->oii_gen) {
227 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
229 id->oii_ino, id->oii_gen, inode->i_generation);
231 inode = ERR_PTR(-ESTALE);
232 } else if (inode->i_nlink == 0) {
233 /* due to parallel readdir and unlink,
234 * we can have dead inode here. */
235 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
236 make_bad_inode(inode);
238 inode = ERR_PTR(-ESTALE);
239 } else if (is_bad_inode(inode)) {
240 CWARN("%.16s: bad inode: ino = %u\n",
241 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
243 inode = ERR_PTR(-ENOENT);
245 if (id->oii_gen == OSD_OII_NOGEN)
246 osd_id_gen(id, inode->i_ino, inode->i_generation);
248 /* Do not update file c/mtime in ldiskfs.
249 * NB: we don't have any lock to protect this because we don't
250 * have reference on osd_object now, but contention with
251 * another lookup + attr_set can't happen in the tiny window
252 * between if (...) and set S_NOCMTIME. */
253 if (!(inode->i_flags & S_NOCMTIME))
254 inode->i_flags |= S_NOCMTIME;
259 static struct inode *
260 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
261 struct osd_inode_id *id, struct lu_fid *fid)
263 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
267 inode = osd_iget(info, dev, id);
271 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
273 *fid = lma->lma_self_fid;
274 } else if (rc == -ENODATA) {
275 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
276 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
278 lu_igif_build(fid, inode->i_ino, inode->i_generation);
286 static struct inode *
287 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
288 struct osd_inode_id *id, const struct lu_fid *fid)
290 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
294 inode = osd_iget(info, dev, id);
298 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
307 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
308 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
309 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
311 return ERR_PTR(-EREMCHG);
317 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
318 const struct lu_fid *fid,
319 const struct lu_object_conf *conf)
321 struct osd_thread_info *info;
322 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
323 struct osd_device *dev;
324 struct osd_idmap_cache *oic;
325 struct osd_inode_id *id;
327 struct osd_scrub *scrub;
328 struct scrub_file *sf;
333 LINVRNT(osd_invariant(obj));
334 LASSERT(obj->oo_inode == NULL);
335 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
338 scrub = &dev->od_scrub;
339 sf = &scrub->os_file;
340 info = osd_oti_get(env);
342 oic = &info->oti_cache;
344 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
347 /* Search order: 1. per-thread cache. */
348 if (lu_fid_eq(fid, &oic->oic_fid)) {
354 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
355 /* Search order: 2. OI scrub pending list. */
356 result = osd_oii_lookup(dev, fid, id);
361 if (sf->sf_flags & SF_INCONSISTENT)
365 * Objects are created as locking anchors or place holders for objects
366 * yet to be created. No need to osd_oi_lookup() at here because FID
367 * shouldn't never be re-used, if it's really a duplicate FID from
368 * unexpected reason, we should be able to detect it later by calling
369 * do_create->osd_oi_insert()
371 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
372 GOTO(out, result = 0);
374 /* Search order: 3. OI files. */
375 result = osd_oi_lookup(info, dev, fid, id, true);
376 if (result == -ENOENT) {
377 if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
378 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
380 GOTO(out, result = 0);
390 inode = osd_iget(info, dev, id);
392 inode = osd_iget_verify(info, dev, id, fid);
394 result = PTR_ERR(inode);
395 if (result == -ENOENT || result == -ESTALE) {
396 fid_zero(&oic->oic_fid);
398 } else if (result == -EREMCHG) {
401 if (thread_is_running(&scrub->os_thread)) {
402 result = -EINPROGRESS;
403 } else if (!dev->od_noscrub) {
404 result = osd_scrub_start(dev);
405 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
406 "for "DFID", rc = %d [1]\n",
407 LDISKFS_SB(osd_sb(dev))->s_es->\
408 s_volume_name,PFID(fid), result);
409 if (result == 0 || result == -EALREADY)
410 result = -EINPROGRESS;
419 obj->oo_inode = inode;
420 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
422 obj->oo_compat_dot_created = 1;
423 obj->oo_compat_dotdot_created = 1;
425 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
426 GOTO(out, result = 0);
428 LASSERT(obj->oo_hl_head == NULL);
429 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
430 if (obj->oo_hl_head == NULL) {
431 obj->oo_inode = NULL;
433 GOTO(out, result = -ENOMEM);
435 GOTO(out, result = 0);
438 LINVRNT(osd_invariant(obj));
443 * Concurrency: shouldn't matter.
445 static void osd_object_init0(struct osd_object *obj)
447 LASSERT(obj->oo_inode != NULL);
448 obj->oo_dt.do_body_ops = &osd_body_ops;
449 obj->oo_dt.do_lu.lo_header->loh_attr |=
450 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
454 * Concurrency: no concurrent access is possible that early in object
457 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
458 const struct lu_object_conf *conf)
460 struct osd_object *obj = osd_obj(l);
463 LINVRNT(osd_invariant(obj));
465 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
466 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
467 l->lo_header->loh_attr |= LOHA_EXISTS;
471 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
472 obj->oo_dt.do_body_ops = &osd_body_ops_new;
473 if (result == 0 && obj->oo_inode != NULL)
474 osd_object_init0(obj);
476 LINVRNT(osd_invariant(obj));
481 * Concurrency: no concurrent access is possible that late in object
484 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
486 struct osd_object *obj = osd_obj(l);
488 LINVRNT(osd_invariant(obj));
490 dt_object_fini(&obj->oo_dt);
491 if (obj->oo_hl_head != NULL)
492 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
497 * Concurrency: no concurrent access is possible that late in object
500 static void osd_index_fini(struct osd_object *o)
502 struct iam_container *bag;
504 if (o->oo_dir != NULL) {
505 bag = &o->oo_dir->od_container;
506 if (o->oo_inode != NULL) {
507 if (bag->ic_object == o->oo_inode)
508 iam_container_fini(bag);
510 OBD_FREE_PTR(o->oo_dir);
516 * Concurrency: no concurrent access is possible that late in object
517 * life-cycle (for all existing callers, that is. New callers have to provide
518 * their own locking.)
520 static int osd_inode_unlinked(const struct inode *inode)
522 return inode->i_nlink == 0;
526 OSD_TXN_OI_DELETE_CREDITS = 20,
527 OSD_TXN_INODE_DELETE_CREDITS = 20
534 #if OSD_THANDLE_STATS
536 * Set time when the handle is allocated
538 static void osd_th_alloced(struct osd_thandle *oth)
540 oth->oth_alloced = cfs_time_current();
544 * Set time when the handle started
546 static void osd_th_started(struct osd_thandle *oth)
548 oth->oth_started = cfs_time_current();
552 * Helper function to convert time interval to microseconds packed in
555 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
559 cfs_duration_usec(cfs_time_sub(end, start), &val);
560 return val.tv_sec * 1000000 + val.tv_usec;
564 * Check whether the we deal with this handle for too long.
566 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
567 cfs_time_t alloced, cfs_time_t started,
570 cfs_time_t now = cfs_time_current();
572 LASSERT(dev != NULL);
574 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
575 interval_to_usec(alloced, started));
576 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
577 interval_to_usec(started, closed));
578 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
579 interval_to_usec(closed, now));
581 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
582 CWARN("transaction handle %p was open for too long: "
584 "alloced "CFS_TIME_T" ,"
585 "started "CFS_TIME_T" ,"
586 "closed "CFS_TIME_T"\n",
587 oth, now, alloced, started, closed);
588 libcfs_debug_dumpstack(NULL);
592 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
594 cfs_time_t __closed = cfs_time_current(); \
595 cfs_time_t __alloced = oth->oth_alloced; \
596 cfs_time_t __started = oth->oth_started; \
599 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
602 #else /* OSD_THANDLE_STATS */
604 #define osd_th_alloced(h) do {} while(0)
605 #define osd_th_started(h) do {} while(0)
606 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
608 #endif /* OSD_THANDLE_STATS */
611 * Concurrency: doesn't access mutable data.
613 static int osd_param_is_not_sane(const struct osd_device *dev,
614 const struct thandle *th)
616 struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
618 return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
622 * Concurrency: shouldn't matter.
624 static void osd_trans_commit_cb(struct super_block *sb,
625 struct ldiskfs_journal_cb_entry *jcb, int error)
627 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
628 struct thandle *th = &oh->ot_super;
629 struct lu_device *lud = &th->th_dev->dd_lu_dev;
630 struct dt_txn_commit_cb *dcb, *tmp;
632 LASSERT(oh->ot_handle == NULL);
635 CERROR("transaction @0x%p commit error: %d\n", th, error);
637 dt_txn_hook_commit(th);
639 /* call per-transaction callbacks if any */
640 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
641 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
642 "commit callback entry: magic=%x name='%s'\n",
643 dcb->dcb_magic, dcb->dcb_name);
644 cfs_list_del_init(&dcb->dcb_linkage);
645 dcb->dcb_func(NULL, th, dcb, error);
648 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
652 lu_context_exit(&th->th_ctx);
653 lu_context_fini(&th->th_ctx);
657 static struct thandle *osd_trans_create(const struct lu_env *env,
660 struct osd_thread_info *oti = osd_oti_get(env);
661 struct osd_iobuf *iobuf = &oti->oti_iobuf;
662 struct osd_thandle *oh;
666 /* on pending IO in this thread should left from prev. request */
667 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
669 th = ERR_PTR(-ENOMEM);
670 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
672 oh->ot_quota_trans = &oti->oti_quota_trans;
673 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
677 th->th_tags = LCT_TX_HANDLE;
679 oti->oti_dev = osd_dt_dev(d);
680 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
683 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
684 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
685 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
686 oti->oti_rollback = false;
692 * Concurrency: shouldn't matter.
694 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
697 struct osd_thread_info *oti = osd_oti_get(env);
698 struct osd_device *dev = osd_dt_dev(d);
700 struct osd_thandle *oh;
705 LASSERT(current->journal_info == NULL);
707 oh = container_of0(th, struct osd_thandle, ot_super);
709 LASSERT(oh->ot_handle == NULL);
711 rc = dt_txn_hook_start(env, d, th);
715 if (unlikely(osd_param_is_not_sane(dev, th))) {
716 static unsigned long last_printed;
717 static int last_credits;
719 CWARN("%.16s: too many transaction credits (%d > %d)\n",
720 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
722 osd_journal(dev)->j_max_transaction_buffers);
723 #ifdef OSD_TRACK_DECLARES
724 CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
725 oti->oti_declare_ops[OSD_OT_CREATE],
726 oti->oti_declare_ops_cred[OSD_OT_CREATE],
727 oti->oti_declare_ops[OSD_OT_DELETE],
728 oti->oti_declare_ops_cred[OSD_OT_DELETE],
729 oti->oti_declare_ops[OSD_OT_DESTROY],
730 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
731 CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n",
732 oti->oti_declare_ops[OSD_OT_ATTR_SET],
733 oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
734 oti->oti_declare_ops[OSD_OT_XATTR_SET],
735 oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
736 CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n",
737 oti->oti_declare_ops[OSD_OT_WRITE],
738 oti->oti_declare_ops_cred[OSD_OT_WRITE],
739 oti->oti_declare_ops[OSD_OT_PUNCH],
740 oti->oti_declare_ops_cred[OSD_OT_PUNCH],
741 oti->oti_declare_ops[OSD_OT_QUOTA],
742 oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
743 CWARN(" insert: %u/%u, delete: %u/%u\n",
744 oti->oti_declare_ops[OSD_OT_INSERT],
745 oti->oti_declare_ops_cred[OSD_OT_INSERT],
746 oti->oti_declare_ops[OSD_OT_DESTROY],
747 oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
748 CWARN(" ref_add: %u/%u, ref_del: %u/%u\n",
749 oti->oti_declare_ops[OSD_OT_REF_ADD],
750 oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
751 oti->oti_declare_ops[OSD_OT_REF_DEL],
752 oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
754 if (last_credits != oh->ot_credits &&
755 time_after(jiffies, last_printed + 60 * HZ)) {
756 libcfs_debug_dumpstack(NULL);
757 last_credits = oh->ot_credits;
758 last_printed = jiffies;
761 /* XXX Limit the credits to 'max_transaction_buffers', and
762 * let the underlying filesystem to catch the error if
763 * we really need so many credits.
765 * This should be removed when we can calculate the
766 * credits precisely. */
767 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
771 * XXX temporary stuff. Some abstraction layer should
774 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
778 LASSERT(oti->oti_txns == 0);
779 lu_context_init(&th->th_ctx, th->th_tags);
780 lu_context_enter(&th->th_ctx);
782 lu_device_get(&d->dd_lu_dev);
783 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
795 * Concurrency: shouldn't matter.
797 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
800 struct osd_thandle *oh;
801 struct osd_thread_info *oti = osd_oti_get(env);
802 struct osd_iobuf *iobuf = &oti->oti_iobuf;
803 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
806 oh = container_of0(th, struct osd_thandle, ot_super);
809 /* inform the quota slave device that the transaction is
811 qsd_op_end(env, qsd, oh->ot_quota_trans);
812 oh->ot_quota_trans = NULL;
814 if (oh->ot_handle != NULL) {
815 handle_t *hdl = oh->ot_handle;
818 * add commit callback
819 * notice we don't do this in osd_trans_start()
820 * as underlying transaction can change during truncate
822 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
825 LASSERT(oti->oti_txns == 1);
827 rc = dt_txn_hook_stop(env, th);
829 CERROR("Failure in transaction hook: %d\n", rc);
831 /* hook functions might modify th_sync */
832 hdl->h_sync = th->th_sync;
834 oh->ot_handle = NULL;
835 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
836 rc = ldiskfs_journal_stop(hdl));
838 CERROR("Failure to stop transaction: %d\n", rc);
843 /* as we want IO to journal and data IO be concurrent, we don't block
844 * awaiting data IO completion in osd_do_bio(), instead we wait here
845 * once transaction is submitted to the journal. all reqular requests
846 * don't do direct IO (except read/write), thus this wait_event becomes
849 * IMPORTANT: we have to wait till any IO submited by the thread is
850 * completed otherwise iobuf may be corrupted by different request
852 cfs_wait_event(iobuf->dr_wait,
853 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
855 rc = iobuf->dr_error;
860 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
862 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
865 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
866 LASSERT(&dcb->dcb_func != NULL);
867 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
873 * Called just before object is freed. Releases all resources except for
874 * object itself (that is released by osd_object_free()).
876 * Concurrency: no concurrent access is possible that late in object
879 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
881 struct osd_object *obj = osd_obj(l);
882 struct inode *inode = obj->oo_inode;
884 LINVRNT(osd_invariant(obj));
887 * If object is unlinked remove fid->ino mapping from object index.
892 struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave;
893 qid_t uid = inode->i_uid;
894 qid_t gid = inode->i_gid;
897 obj->oo_inode = NULL;
900 struct osd_thread_info *info = osd_oti_get(env);
901 struct lquota_id_info *qi = &info->oti_qi;
903 /* Release granted quota to master if necessary */
904 qi->lqi_id.qid_uid = uid;
905 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
907 qi->lqi_id.qid_uid = gid;
908 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
914 * Concurrency: ->loo_object_release() is called under site spin-lock.
916 static void osd_object_release(const struct lu_env *env,
922 * Concurrency: shouldn't matter.
924 static int osd_object_print(const struct lu_env *env, void *cookie,
925 lu_printer_t p, const struct lu_object *l)
927 struct osd_object *o = osd_obj(l);
930 if (o->oo_dir != NULL)
931 d = o->oo_dir->od_container.ic_descr;
934 return (*p)(env, cookie,
935 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
937 o->oo_inode ? o->oo_inode->i_ino : 0UL,
938 o->oo_inode ? o->oo_inode->i_generation : 0,
939 d ? d->id_ops->id_name : "plain");
943 * Concurrency: shouldn't matter.
945 int osd_statfs(const struct lu_env *env, struct dt_device *d,
946 struct obd_statfs *sfs)
948 struct osd_device *osd = osd_dt_dev(d);
949 struct super_block *sb = osd_sb(osd);
950 struct kstatfs *ksfs;
953 if (unlikely(osd->od_mnt == NULL))
956 /* osd_lproc.c call this without env, allocate ksfs for that case */
957 if (unlikely(env == NULL)) {
962 ksfs = &osd_oti_get(env)->oti_ksfs;
965 spin_lock(&osd->od_osfs_lock);
967 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
968 result = sb->s_op->statfs(sb->s_root, ksfs);
969 if (likely(result == 0)) { /* N.B. statfs can't really fail */
970 osd->od_osfs_age = cfs_time_current_64();
971 statfs_pack(&osd->od_statfs, ksfs);
972 if (sb->s_flags & MS_RDONLY)
973 sfs->os_state = OS_STATE_READONLY;
977 if (likely(result == 0))
978 *sfs = osd->od_statfs;
979 spin_unlock(&osd->od_osfs_lock);
981 if (unlikely(env == NULL))
988 * Estimate space needed for file creations. We assume the largest filename
989 * which is 2^64 - 1, hence a filename of 20 chars.
990 * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
992 #ifdef __LDISKFS_DIR_REC_LEN
993 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
995 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
999 * Concurrency: doesn't access mutable data.
1001 static void osd_conf_get(const struct lu_env *env,
1002 const struct dt_device *dev,
1003 struct dt_device_param *param)
1005 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1008 * XXX should be taken from not-yet-existing fs abstraction layer.
1010 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1011 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1012 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1013 param->ddp_block_shift = sb->s_blocksize_bits;
1014 param->ddp_mount_type = LDD_MT_LDISKFS;
1015 param->ddp_maxbytes = sb->s_maxbytes;
1016 /* Overhead estimate should be fairly accurate, so we really take a tiny
1017 * error margin which also avoids fragmenting the filesystem too much */
1018 param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1019 /* inode are statically allocated, so per-inode space consumption
1020 * is the space consumed by the directory entry */
1021 param->ddp_inodespace = PER_OBJ_USAGE;
1022 /* per-fragment overhead to be used by the client code */
1023 param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb);
1024 param->ddp_mntopts = 0;
1025 if (test_opt(sb, XATTR_USER))
1026 param->ddp_mntopts |= MNTOPT_USERXATTR;
1027 if (test_opt(sb, POSIX_ACL))
1028 param->ddp_mntopts |= MNTOPT_ACL;
1030 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1031 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1032 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1035 param->ddp_max_ea_size = sb->s_blocksize;
1040 * Concurrency: shouldn't matter.
1042 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1044 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1045 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1049 * Start commit for OSD device.
1051 * An implementation of dt_commit_async method for OSD device.
1052 * Asychronously starts underlayng fs sync and thereby a transaction
1055 * \param env environment
1056 * \param d dt device
1058 * \see dt_device_operations
1060 static int osd_commit_async(const struct lu_env *env,
1061 struct dt_device *d)
1063 struct super_block *s = osd_sb(osd_dt_dev(d));
1066 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1067 RETURN(s->s_op->sync_fs(s, 0));
1071 * Concurrency: shouldn't matter.
1074 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1076 struct super_block *sb = osd_sb(osd_dt_dev(d));
1077 struct block_device *dev = sb->s_bdev;
1078 #ifdef HAVE_DEV_SET_RDONLY
1079 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1082 int rc = -EOPNOTSUPP;
1086 #ifdef HAVE_DEV_SET_RDONLY
1087 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1089 if (jdev && (jdev != dev)) {
1090 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1092 dev_set_rdonly(jdev);
1094 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1095 dev_set_rdonly(dev);
1097 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1098 osd_dt_dev(d)->od_svname, (long)dev, rc);
1104 * Concurrency: serialization provided by callers.
1106 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1107 int mode, unsigned long timeout, __u32 alg,
1108 struct lustre_capa_key *keys)
1110 struct osd_device *dev = osd_dt_dev(d);
1113 dev->od_fl_capa = mode;
1114 dev->od_capa_timeout = timeout;
1115 dev->od_capa_alg = alg;
1116 dev->od_capa_keys = keys;
1121 * Note: we do not count into QUOTA here.
1122 * If we mount with --data_journal we may need more.
1124 const int osd_dto_credits_noquota[DTO_NR] = {
1127 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1128 * SINGLEDATA_TRANS_BLOCKS(8)
1129 * XXX Note: maybe iam need more, since iam have more level than
1132 [DTO_INDEX_INSERT] = 16,
1133 [DTO_INDEX_DELETE] = 16,
1137 [DTO_INDEX_UPDATE] = 16,
1139 * Create a object. The same as create object in EXT3.
1140 * DATA_TRANS_BLOCKS(14) +
1141 * INDEX_EXTRA_BLOCKS(8) +
1142 * 3(inode bits, groups, GDT)
1144 [DTO_OBJECT_CREATE] = 25,
1146 * XXX: real credits to be fixed
1148 [DTO_OBJECT_DELETE] = 25,
1150 * Attr set credits (inode)
1152 [DTO_ATTR_SET_BASE] = 1,
1154 * Xattr set. The same as xattr of EXT3.
1155 * DATA_TRANS_BLOCKS(14)
1156 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1157 * are also counted in. Do not know why?
1159 [DTO_XATTR_SET] = 14,
1162 * credits for inode change during write.
1164 [DTO_WRITE_BASE] = 3,
1166 * credits for single block write.
1168 [DTO_WRITE_BLOCK] = 14,
1170 * Attr set credits for chown.
1171 * This is extra credits for setattr, and it is null without quota
1173 [DTO_ATTR_SET_CHOWN]= 0
1176 static const struct dt_device_operations osd_dt_ops = {
1177 .dt_root_get = osd_root_get,
1178 .dt_statfs = osd_statfs,
1179 .dt_trans_create = osd_trans_create,
1180 .dt_trans_start = osd_trans_start,
1181 .dt_trans_stop = osd_trans_stop,
1182 .dt_trans_cb_add = osd_trans_cb_add,
1183 .dt_conf_get = osd_conf_get,
1184 .dt_sync = osd_sync,
1186 .dt_commit_async = osd_commit_async,
1187 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1190 static void osd_object_read_lock(const struct lu_env *env,
1191 struct dt_object *dt, unsigned role)
1193 struct osd_object *obj = osd_dt_obj(dt);
1194 struct osd_thread_info *oti = osd_oti_get(env);
1196 LINVRNT(osd_invariant(obj));
1198 LASSERT(obj->oo_owner != env);
1199 down_read_nested(&obj->oo_sem, role);
1201 LASSERT(obj->oo_owner == NULL);
1205 static void osd_object_write_lock(const struct lu_env *env,
1206 struct dt_object *dt, unsigned role)
1208 struct osd_object *obj = osd_dt_obj(dt);
1209 struct osd_thread_info *oti = osd_oti_get(env);
1211 LINVRNT(osd_invariant(obj));
1213 LASSERT(obj->oo_owner != env);
1214 down_write_nested(&obj->oo_sem, role);
1216 LASSERT(obj->oo_owner == NULL);
1217 obj->oo_owner = env;
1221 static void osd_object_read_unlock(const struct lu_env *env,
1222 struct dt_object *dt)
1224 struct osd_object *obj = osd_dt_obj(dt);
1225 struct osd_thread_info *oti = osd_oti_get(env);
1227 LINVRNT(osd_invariant(obj));
1229 LASSERT(oti->oti_r_locks > 0);
1231 up_read(&obj->oo_sem);
1234 static void osd_object_write_unlock(const struct lu_env *env,
1235 struct dt_object *dt)
1237 struct osd_object *obj = osd_dt_obj(dt);
1238 struct osd_thread_info *oti = osd_oti_get(env);
1240 LINVRNT(osd_invariant(obj));
1242 LASSERT(obj->oo_owner == env);
1243 LASSERT(oti->oti_w_locks > 0);
1245 obj->oo_owner = NULL;
1246 up_write(&obj->oo_sem);
1249 static int osd_object_write_locked(const struct lu_env *env,
1250 struct dt_object *dt)
1252 struct osd_object *obj = osd_dt_obj(dt);
1254 LINVRNT(osd_invariant(obj));
1256 return obj->oo_owner == env;
1259 static int capa_is_sane(const struct lu_env *env,
1260 struct osd_device *dev,
1261 struct lustre_capa *capa,
1262 struct lustre_capa_key *keys)
1264 struct osd_thread_info *oti = osd_oti_get(env);
1265 struct lustre_capa *tcapa = &oti->oti_capa;
1266 struct obd_capa *oc;
1270 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1272 if (capa_is_expired(oc)) {
1273 DEBUG_CAPA(D_ERROR, capa, "expired");
1280 if (capa_is_expired_sec(capa)) {
1281 DEBUG_CAPA(D_ERROR, capa, "expired");
1285 spin_lock(&capa_lock);
1286 for (i = 0; i < 2; i++) {
1287 if (keys[i].lk_keyid == capa->lc_keyid) {
1288 oti->oti_capa_key = keys[i];
1292 spin_unlock(&capa_lock);
1295 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1299 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1303 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1304 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1308 oc = capa_add(dev->od_capa_hash, capa);
1314 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1315 struct lustre_capa *capa, __u64 opc)
1317 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1318 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1319 struct md_capainfo *ci;
1322 if (!dev->od_fl_capa)
1325 if (capa == BYPASS_CAPA)
1328 ci = md_capainfo(env);
1332 if (ci->mc_auth == LC_ID_NONE)
1336 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1340 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1341 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1346 if (!capa_opc_supported(capa, opc)) {
1347 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1351 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1352 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1359 static struct timespec *osd_inode_time(const struct lu_env *env,
1360 struct inode *inode, __u64 seconds)
1362 struct osd_thread_info *oti = osd_oti_get(env);
1363 struct timespec *t = &oti->oti_time;
1365 t->tv_sec = seconds;
1367 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1372 static void osd_inode_getattr(const struct lu_env *env,
1373 struct inode *inode, struct lu_attr *attr)
1375 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1376 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1377 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1380 attr->la_atime = LTIME_S(inode->i_atime);
1381 attr->la_mtime = LTIME_S(inode->i_mtime);
1382 attr->la_ctime = LTIME_S(inode->i_ctime);
1383 attr->la_mode = inode->i_mode;
1384 attr->la_size = i_size_read(inode);
1385 attr->la_blocks = inode->i_blocks;
1386 attr->la_uid = inode->i_uid;
1387 attr->la_gid = inode->i_gid;
1388 attr->la_flags = LDISKFS_I(inode)->i_flags;
1389 attr->la_nlink = inode->i_nlink;
1390 attr->la_rdev = inode->i_rdev;
1391 attr->la_blksize = 1 << inode->i_blkbits;
1392 attr->la_blkbits = inode->i_blkbits;
1395 static int osd_attr_get(const struct lu_env *env,
1396 struct dt_object *dt,
1397 struct lu_attr *attr,
1398 struct lustre_capa *capa)
1400 struct osd_object *obj = osd_dt_obj(dt);
1402 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1403 LINVRNT(osd_invariant(obj));
1405 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1408 spin_lock(&obj->oo_guard);
1409 osd_inode_getattr(env, obj->oo_inode, attr);
1410 spin_unlock(&obj->oo_guard);
1414 static int osd_declare_attr_set(const struct lu_env *env,
1415 struct dt_object *dt,
1416 const struct lu_attr *attr,
1417 struct thandle *handle)
1419 struct osd_thandle *oh;
1420 struct osd_object *obj;
1421 struct osd_thread_info *info = osd_oti_get(env);
1422 struct lquota_id_info *qi = &info->oti_qi;
1428 LASSERT(dt != NULL);
1429 LASSERT(handle != NULL);
1431 obj = osd_dt_obj(dt);
1432 LASSERT(osd_invariant(obj));
1434 oh = container_of0(handle, struct osd_thandle, ot_super);
1435 LASSERT(oh->ot_handle == NULL);
1437 osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1438 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1440 if (attr == NULL || obj->oo_inode == NULL)
1443 bspace = obj->oo_inode->i_blocks;
1444 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1445 bspace = toqb(bspace);
1447 /* Changing ownership is always preformed by super user, it should not
1450 * We still need to call the osd_declare_qid() to calculate the journal
1451 * credits for updating quota accounting files and to trigger quota
1452 * space adjustment once the operation is completed.*/
1453 if ((attr->la_valid & LA_UID) != 0 &&
1454 attr->la_uid != obj->oo_inode->i_uid) {
1455 qi->lqi_type = USRQUOTA;
1457 /* inode accounting */
1458 qi->lqi_is_blk = false;
1460 /* one more inode for the new owner ... */
1461 qi->lqi_id.qid_uid = attr->la_uid;
1463 allocated = (attr->la_uid == 0) ? true : false;
1464 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1465 if (rc == -EDQUOT || rc == -EINPROGRESS)
1470 /* and one less inode for the current uid */
1471 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1473 rc = osd_declare_qid(env, oh, qi, true, NULL);
1474 if (rc == -EDQUOT || rc == -EINPROGRESS)
1479 /* block accounting */
1480 qi->lqi_is_blk = true;
1482 /* more blocks for the new owner ... */
1483 qi->lqi_id.qid_uid = attr->la_uid;
1484 qi->lqi_space = bspace;
1485 allocated = (attr->la_uid == 0) ? true : false;
1486 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1487 if (rc == -EDQUOT || rc == -EINPROGRESS)
1492 /* and finally less blocks for the current owner */
1493 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1494 qi->lqi_space = -bspace;
1495 rc = osd_declare_qid(env, oh, qi, true, NULL);
1496 if (rc == -EDQUOT || rc == -EINPROGRESS)
1502 if (attr->la_valid & LA_GID &&
1503 attr->la_gid != obj->oo_inode->i_gid) {
1504 qi->lqi_type = GRPQUOTA;
1506 /* inode accounting */
1507 qi->lqi_is_blk = false;
1509 /* one more inode for the new group owner ... */
1510 qi->lqi_id.qid_gid = attr->la_gid;
1512 allocated = (attr->la_gid == 0) ? true : false;
1513 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1514 if (rc == -EDQUOT || rc == -EINPROGRESS)
1519 /* and one less inode for the current gid */
1520 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1522 rc = osd_declare_qid(env, oh, qi, true, NULL);
1523 if (rc == -EDQUOT || rc == -EINPROGRESS)
1528 /* block accounting */
1529 qi->lqi_is_blk = true;
1531 /* more blocks for the new owner ... */
1532 qi->lqi_id.qid_gid = attr->la_gid;
1533 qi->lqi_space = bspace;
1534 allocated = (attr->la_gid == 0) ? true : false;
1535 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1536 if (rc == -EDQUOT || rc == -EINPROGRESS)
1541 /* and finally less blocks for the current owner */
1542 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1543 qi->lqi_space = -bspace;
1544 rc = osd_declare_qid(env, oh, qi, true, NULL);
1545 if (rc == -EDQUOT || rc == -EINPROGRESS)
1554 static int osd_inode_setattr(const struct lu_env *env,
1555 struct inode *inode, const struct lu_attr *attr)
1559 bits = attr->la_valid;
1561 if (bits & LA_ATIME)
1562 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1563 if (bits & LA_CTIME)
1564 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1565 if (bits & LA_MTIME)
1566 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1567 if (bits & LA_SIZE) {
1568 LDISKFS_I(inode)->i_disksize = attr->la_size;
1569 i_size_write(inode, attr->la_size);
1573 /* OSD should not change "i_blocks" which is used by quota.
1574 * "i_blocks" should be changed by ldiskfs only. */
1575 if (bits & LA_BLOCKS)
1576 inode->i_blocks = attr->la_blocks;
1579 inode->i_mode = (inode->i_mode & S_IFMT) |
1580 (attr->la_mode & ~S_IFMT);
1582 inode->i_uid = attr->la_uid;
1584 inode->i_gid = attr->la_gid;
1585 if (bits & LA_NLINK)
1586 set_nlink(inode, attr->la_nlink);
1588 inode->i_rdev = attr->la_rdev;
1590 if (bits & LA_FLAGS) {
1591 /* always keep S_NOCMTIME */
1592 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1598 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1600 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1601 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1606 if (attr->la_valid & LA_UID)
1607 iattr.ia_valid |= ATTR_UID;
1608 if (attr->la_valid & LA_GID)
1609 iattr.ia_valid |= ATTR_GID;
1610 iattr.ia_uid = attr->la_uid;
1611 iattr.ia_gid = attr->la_gid;
1613 rc = ll_vfs_dq_transfer(inode, &iattr);
1615 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1616 "enforcement enabled on the ldiskfs filesystem?",
1617 inode->i_sb->s_id, rc);
1624 static int osd_attr_set(const struct lu_env *env,
1625 struct dt_object *dt,
1626 const struct lu_attr *attr,
1627 struct thandle *handle,
1628 struct lustre_capa *capa)
1630 struct osd_object *obj = osd_dt_obj(dt);
1631 struct inode *inode;
1634 LASSERT(handle != NULL);
1635 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1636 LASSERT(osd_invariant(obj));
1638 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1641 osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1643 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1644 struct osd_thread_info *oti = osd_oti_get(env);
1645 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
1646 struct lu_fid *fid1 = &oti->oti_fid;
1647 struct osd_inode_id *id = &oti->oti_id;
1648 struct iam_path_descr *ipd;
1649 struct iam_container *bag;
1650 struct osd_thandle *oh;
1653 fid_cpu_to_be(fid1, fid0);
1654 memset(id, 1, sizeof(*id));
1655 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1656 fid0)->oi_dir.od_container;
1657 ipd = osd_idx_ipd_get(env, bag);
1658 if (unlikely(ipd == NULL))
1661 oh = container_of0(handle, struct osd_thandle, ot_super);
1662 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1663 (const struct iam_rec *)id, ipd);
1664 osd_ipd_put(env, bag, ipd);
1665 return(rc > 0 ? 0 : rc);
1668 inode = obj->oo_inode;
1669 ll_vfs_dq_init(inode);
1671 rc = osd_quota_transfer(inode, attr);
1675 spin_lock(&obj->oo_guard);
1676 rc = osd_inode_setattr(env, inode, attr);
1677 spin_unlock(&obj->oo_guard);
1680 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1684 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1685 struct osd_object *obj,
1686 const char *name, const int namelen)
1688 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1691 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1693 struct dt_allocation_hint *hint,
1697 struct osd_device *osd = osd_obj2dev(obj);
1698 struct osd_thandle *oth;
1699 struct dt_object *parent = NULL;
1700 struct inode *inode;
1702 LINVRNT(osd_invariant(obj));
1703 LASSERT(obj->oo_inode == NULL);
1704 LASSERT(obj->oo_hl_head == NULL);
1706 if (S_ISDIR(mode) && ldiskfs_pdo) {
1707 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1708 if (obj->oo_hl_head == NULL)
1712 oth = container_of(th, struct osd_thandle, ot_super);
1713 LASSERT(oth->ot_handle->h_transaction != NULL);
1715 if (hint && hint->dah_parent)
1716 parent = hint->dah_parent;
1718 inode = ldiskfs_create_inode(oth->ot_handle,
1719 parent ? osd_dt_obj(parent)->oo_inode :
1720 osd_sb(osd)->s_root->d_inode,
1722 if (!IS_ERR(inode)) {
1723 /* Do not update file c/mtime in ldiskfs.
1724 * NB: don't need any lock because no contention at this
1726 inode->i_flags |= S_NOCMTIME;
1728 /* For new created object, it must be consistent,
1729 * and it is unnecessary to scrub against it. */
1730 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1731 obj->oo_inode = inode;
1734 if (obj->oo_hl_head != NULL) {
1735 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1736 obj->oo_hl_head = NULL;
1738 result = PTR_ERR(inode);
1740 LINVRNT(osd_invariant(obj));
1748 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1749 struct lu_attr *attr,
1750 struct dt_allocation_hint *hint,
1751 struct dt_object_format *dof,
1755 struct osd_thandle *oth;
1756 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1758 LASSERT(S_ISDIR(attr->la_mode));
1760 oth = container_of(th, struct osd_thandle, ot_super);
1761 LASSERT(oth->ot_handle->h_transaction != NULL);
1762 result = osd_mkfile(info, obj, mode, hint, th);
1767 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1768 struct lu_attr *attr,
1769 struct dt_allocation_hint *hint,
1770 struct dt_object_format *dof,
1774 struct osd_thandle *oth;
1775 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1777 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1779 LASSERT(S_ISREG(attr->la_mode));
1781 oth = container_of(th, struct osd_thandle, ot_super);
1782 LASSERT(oth->ot_handle->h_transaction != NULL);
1784 result = osd_mkfile(info, obj, mode, hint, th);
1786 LASSERT(obj->oo_inode != NULL);
1787 if (feat->dif_flags & DT_IND_VARKEY)
1788 result = iam_lvar_create(obj->oo_inode,
1789 feat->dif_keysize_max,
1791 feat->dif_recsize_max,
1794 result = iam_lfix_create(obj->oo_inode,
1795 feat->dif_keysize_max,
1797 feat->dif_recsize_max,
1804 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1805 struct lu_attr *attr,
1806 struct dt_allocation_hint *hint,
1807 struct dt_object_format *dof,
1810 LASSERT(S_ISREG(attr->la_mode));
1811 return osd_mkfile(info, obj, (attr->la_mode &
1812 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1815 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1816 struct lu_attr *attr,
1817 struct dt_allocation_hint *hint,
1818 struct dt_object_format *dof,
1821 LASSERT(S_ISLNK(attr->la_mode));
1822 return osd_mkfile(info, obj, (attr->la_mode &
1823 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1826 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1827 struct lu_attr *attr,
1828 struct dt_allocation_hint *hint,
1829 struct dt_object_format *dof,
1832 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1835 LINVRNT(osd_invariant(obj));
1836 LASSERT(obj->oo_inode == NULL);
1837 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1838 S_ISFIFO(mode) || S_ISSOCK(mode));
1840 result = osd_mkfile(info, obj, mode, hint, th);
1842 LASSERT(obj->oo_inode != NULL);
1844 * This inode should be marked dirty for i_rdev. Currently
1845 * that is done in the osd_attr_init().
1847 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1850 LINVRNT(osd_invariant(obj));
1854 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1856 struct dt_allocation_hint *hint,
1857 struct dt_object_format *dof,
1860 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1862 osd_obj_type_f result;
1878 result = osd_mk_index;
1889 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1890 struct dt_object *parent, struct dt_object *child,
1891 cfs_umode_t child_mode)
1895 memset(ah, 0, sizeof(*ah));
1896 ah->dah_parent = parent;
1897 ah->dah_mode = child_mode;
1900 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1901 struct lu_attr *attr, struct dt_object_format *dof)
1903 struct inode *inode = obj->oo_inode;
1904 __u64 valid = attr->la_valid;
1907 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1909 if (dof->dof_type != DFT_NODE)
1910 attr->la_valid &= ~LA_RDEV;
1911 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1912 attr->la_valid &= ~LA_ATIME;
1913 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1914 attr->la_valid &= ~LA_CTIME;
1915 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1916 attr->la_valid &= ~LA_MTIME;
1918 result = osd_quota_transfer(inode, attr);
1922 if (attr->la_valid != 0) {
1923 result = osd_inode_setattr(info->oti_env, inode, attr);
1925 * The osd_inode_setattr() should always succeed here. The
1926 * only error that could be returned is EDQUOT when we are
1927 * trying to change the UID or GID of the inode. However, this
1928 * should not happen since quota enforcement is no longer
1929 * enabled on ldiskfs (lquota takes care of it).
1931 LASSERTF(result == 0, "%d", result);
1932 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1935 attr->la_valid = valid;
1939 * Helper function for osd_object_create()
1941 * \retval 0, on success
1943 static int __osd_object_create(struct osd_thread_info *info,
1944 struct osd_object *obj, struct lu_attr *attr,
1945 struct dt_allocation_hint *hint,
1946 struct dt_object_format *dof,
1952 /* we drop umask so that permissions we pass are not affected */
1953 umask = current->fs->umask;
1954 current->fs->umask = 0;
1956 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1959 osd_attr_init(info, obj, attr, dof);
1960 osd_object_init0(obj);
1962 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1963 unlock_new_inode(obj->oo_inode);
1966 /* restore previous umask value */
1967 current->fs->umask = umask;
1973 * Helper function for osd_object_create()
1975 * \retval 0, on success
1977 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1978 const struct lu_fid *fid, struct thandle *th)
1980 struct osd_thread_info *info = osd_oti_get(env);
1981 struct osd_inode_id *id = &info->oti_id;
1982 struct osd_device *osd = osd_obj2dev(obj);
1984 LASSERT(obj->oo_inode != NULL);
1986 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1987 return osd_oi_insert(info, osd, fid, id, th);
1990 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1991 const struct lu_fid *fid, struct lu_seq_range *range)
1993 struct seq_server_site *ss = osd_seq_site(osd);
1996 if (fid_is_idif(fid)) {
1997 range->lsr_flags = LU_SEQ_RANGE_OST;
1998 range->lsr_index = fid_idif_ost_idx(fid);
2002 if (!fid_seq_in_fldb(fid_seq(fid))) {
2003 range->lsr_flags = LU_SEQ_RANGE_MDT;
2005 /* FIXME: If ss is NULL, it suppose not get lsr_index
2007 range->lsr_index = ss->ss_node_id;
2011 LASSERT(ss != NULL);
2012 range->lsr_flags = -1;
2013 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2015 CERROR("%s can not find "DFID": rc = %d\n",
2016 osd_name(osd), PFID(fid), rc);
2022 * Concurrency: no external locking is necessary.
2024 static int osd_declare_object_create(const struct lu_env *env,
2025 struct dt_object *dt,
2026 struct lu_attr *attr,
2027 struct dt_allocation_hint *hint,
2028 struct dt_object_format *dof,
2029 struct thandle *handle)
2031 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
2032 struct osd_thandle *oh;
2036 LASSERT(handle != NULL);
2038 oh = container_of0(handle, struct osd_thandle, ot_super);
2039 LASSERT(oh->ot_handle == NULL);
2041 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2042 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2043 if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2044 lu_object_fid(&dt->do_lu)))
2045 /* Reuse idle OI block may cause additional one OI block
2047 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2048 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2050 /* If this is directory, then we expect . and .. to be inserted as
2051 * well. The one directory block always needs to be created for the
2052 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2053 * block), there is no danger of needing a tree for the first block.
2055 if (attr && S_ISDIR(attr->la_mode)) {
2056 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2057 osd_dto_credits_noquota[DTO_WRITE_BASE]);
2058 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2064 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2065 false, false, NULL, false);
2069 /* It does fld look up inside declare, and the result will be
2070 * added to fld cache, so the following fld lookup inside insert
2071 * does not need send RPC anymore, so avoid send rpc with holding
2073 if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2074 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2075 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2076 lu_object_fid(&dt->do_lu), range);
2082 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2083 struct lu_attr *attr,
2084 struct dt_allocation_hint *hint,
2085 struct dt_object_format *dof,
2088 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2089 struct osd_object *obj = osd_dt_obj(dt);
2090 struct osd_thread_info *info = osd_oti_get(env);
2095 LINVRNT(osd_invariant(obj));
2096 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2097 LASSERT(osd_write_locked(env, obj));
2098 LASSERT(th != NULL);
2100 if (unlikely(fid_is_acct(fid)))
2101 /* Quota files can't be created from the kernel any more,
2102 * 'tune2fs -O quota' will take care of creating them */
2105 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2106 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2108 result = __osd_object_create(info, obj, attr, hint, dof, th);
2110 result = __osd_oi_insert(env, obj, fid, th);
2112 LASSERT(ergo(result == 0,
2113 dt_object_exists(dt) && !dt_object_remote(dt)));
2115 LASSERT(osd_invariant(obj));
2120 * Called to destroy on-disk representation of the object
2122 * Concurrency: must be locked
2124 static int osd_declare_object_destroy(const struct lu_env *env,
2125 struct dt_object *dt,
2128 struct osd_object *obj = osd_dt_obj(dt);
2129 struct inode *inode = obj->oo_inode;
2130 struct osd_thandle *oh;
2134 oh = container_of0(th, struct osd_thandle, ot_super);
2135 LASSERT(oh->ot_handle == NULL);
2138 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2139 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2140 /* Recycle idle OI leaf may cause additional three OI blocks
2142 osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2143 osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2145 /* one less inode */
2146 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2147 false, true, NULL, false);
2150 /* data to be truncated */
2151 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2152 true, true, NULL, false);
2156 static int osd_object_destroy(const struct lu_env *env,
2157 struct dt_object *dt,
2160 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2161 struct osd_object *obj = osd_dt_obj(dt);
2162 struct inode *inode = obj->oo_inode;
2163 struct osd_device *osd = osd_obj2dev(obj);
2164 struct osd_thandle *oh;
2168 oh = container_of0(th, struct osd_thandle, ot_super);
2169 LASSERT(oh->ot_handle);
2171 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2173 if (unlikely(fid_is_acct(fid)))
2176 /* Parallel control for OI scrub. For most of cases, there is no
2177 * lock contention. So it will not affect unlink performance. */
2178 mutex_lock(&inode->i_mutex);
2179 if (S_ISDIR(inode->i_mode)) {
2180 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2181 /* it will check/delete the inode from remote parent,
2182 * how to optimize it? unlink performance impaction XXX */
2183 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2184 if (result != 0 && result != -ENOENT) {
2185 CERROR("%s: delete inode "DFID": rc = %d\n",
2186 osd_name(osd), PFID(fid), result);
2188 spin_lock(&obj->oo_guard);
2190 spin_unlock(&obj->oo_guard);
2191 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2194 osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2196 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2197 mutex_unlock(&inode->i_mutex);
2199 /* XXX: add to ext3 orphan list */
2200 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2202 /* not needed in the cache anymore */
2203 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2209 * Put the fid into lustre_mdt_attrs, and then place the structure
2210 * inode's ea. This fid should not be altered during the life time
2213 * \retval +ve, on success
2214 * \retval -ve, on error
2216 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2218 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2219 const struct lu_fid *fid, __u64 flags)
2221 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2224 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2227 lustre_lma_init(lma, fid, flags);
2228 lustre_lma_swab(lma);
2230 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2232 /* Someone may created the EA by race. */
2233 if (unlikely(rc == -EEXIST))
2239 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2240 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2241 * To have compatilibility with 1.8 ldiskfs driver we need to have
2242 * magic number at start of fid data.
2243 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2246 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2247 const struct dt_rec *fid)
2249 if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2250 OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2251 param->edp_magic = 0;
2255 param->edp_magic = LDISKFS_LUFID_MAGIC;
2256 param->edp_len = sizeof(struct lu_fid) + 1;
2257 fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2261 * Try to read the fid from inode ea into dt_rec.
2263 * \param fid object fid.
2265 * \retval 0 on success
2267 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2268 __u32 ino, struct lu_fid *fid,
2269 struct osd_inode_id *id)
2271 struct osd_thread_info *info = osd_oti_get(env);
2272 struct inode *inode;
2275 osd_id_gen(id, ino, OSD_OII_NOGEN);
2276 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2278 RETURN(PTR_ERR(inode));
2284 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2286 struct inode *parent_dir,
2287 const struct dt_rec *dot_fid,
2288 const struct dt_rec *dot_dot_fid,
2289 struct osd_thandle *oth)
2291 struct ldiskfs_dentry_param *dot_ldp;
2292 struct ldiskfs_dentry_param *dot_dot_ldp;
2294 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2295 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2297 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2298 dot_ldp->edp_magic = 0;
2299 return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2300 dir, dot_ldp, dot_dot_ldp);
2304 * Create an local agent inode for remote entry
2306 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2307 struct osd_device *osd,
2308 struct osd_object *pobj,
2309 const struct lu_fid *fid,
2312 struct osd_thread_info *info = osd_oti_get(env);
2313 struct inode *local;
2314 struct osd_thandle *oh;
2319 oh = container_of(th, struct osd_thandle, ot_super);
2320 LASSERT(oh->ot_handle->h_transaction != NULL);
2322 /* FIXME: Insert index api needs to know the mode of
2323 * the remote object. Just use S_IFDIR for now */
2324 local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2325 if (IS_ERR(local)) {
2326 CERROR("%s: create local error %d\n", osd_name(osd),
2327 (int)PTR_ERR(local));
2331 /* Set special LMA flag for local agent inode */
2332 rc = osd_ea_fid_set(info, local, fid, LMAI_AGENT);
2334 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2335 osd_name(osd), PFID(fid), rc);
2336 RETURN(ERR_PTR(rc));
2339 rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2340 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2341 (const struct dt_rec *)fid, oh);
2343 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2344 osd_name(osd), PFID(fid), rc);
2345 RETURN(ERR_PTR(rc));
2352 * Delete local agent inode for remote entry
2354 static int osd_delete_local_agent_inode(const struct lu_env *env,
2355 struct osd_device *osd,
2356 const struct lu_fid *fid,
2357 __u32 ino, struct osd_thandle *oh)
2359 struct osd_thread_info *oti = osd_oti_get(env);
2360 struct osd_inode_id *id = &oti->oti_id;
2361 struct inode *inode;
2364 id->oii_ino = le32_to_cpu(ino);
2365 id->oii_gen = OSD_OII_NOGEN;
2366 inode = osd_iget(oti, osd, id);
2367 if (IS_ERR(inode)) {
2368 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2369 PFID(fid), id->oii_ino, id->oii_gen);
2370 RETURN(PTR_ERR(inode));
2374 mark_inode_dirty(inode);
2375 CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2376 osd_name(osd), PFID(fid), inode->i_ino);
2382 * OSD layer object create function for interoperability mode (b11826).
2383 * This is mostly similar to osd_object_create(). Only difference being, fid is
2384 * inserted into inode ea here.
2386 * \retval 0, on success
2387 * \retval -ve, on error
2389 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2390 struct lu_attr *attr,
2391 struct dt_allocation_hint *hint,
2392 struct dt_object_format *dof,
2395 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2396 struct osd_object *obj = osd_dt_obj(dt);
2397 struct osd_thread_info *info = osd_oti_get(env);
2402 LASSERT(osd_invariant(obj));
2403 LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2404 LASSERT(osd_write_locked(env, obj));
2405 LASSERT(th != NULL);
2407 if (unlikely(fid_is_acct(fid)))
2408 /* Quota files can't be created from the kernel any more,
2409 * 'tune2fs -O quota' will take care of creating them */
2412 osd_trans_exec_op(env, th, OSD_OT_CREATE);
2413 osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2415 result = __osd_object_create(info, obj, attr, hint, dof, th);
2417 result = osd_ea_fid_set(info, obj->oo_inode, fid, 0);
2420 result = __osd_oi_insert(env, obj, fid, th);
2422 LASSERT(ergo(result == 0,
2423 dt_object_exists(dt) && !dt_object_remote(dt)));
2424 LINVRNT(osd_invariant(obj));
2428 static int osd_declare_object_ref_add(const struct lu_env *env,
2429 struct dt_object *dt,
2430 struct thandle *handle)
2432 struct osd_thandle *oh;
2434 /* it's possible that object doesn't exist yet */
2435 LASSERT(handle != NULL);
2437 oh = container_of0(handle, struct osd_thandle, ot_super);
2438 LASSERT(oh->ot_handle == NULL);
2440 osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2441 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2447 * Concurrency: @dt is write locked.
2449 static int osd_object_ref_add(const struct lu_env *env,
2450 struct dt_object *dt, struct thandle *th)
2452 struct osd_object *obj = osd_dt_obj(dt);
2453 struct inode *inode = obj->oo_inode;
2455 LINVRNT(osd_invariant(obj));
2456 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2457 LASSERT(osd_write_locked(env, obj));
2458 LASSERT(th != NULL);
2460 osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2463 * DIR_NLINK feature is set for compatibility reasons if:
2464 * 1) nlinks > LDISKFS_LINK_MAX, or
2465 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2467 * It is easier to always set this flag (rather than check and set),
2468 * since it has less overhead, and the superblock will be dirtied
2469 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2470 * do not actually care whether this flag is set or not.
2472 spin_lock(&obj->oo_guard);
2473 /* inc_nlink from 0 may cause WARN_ON */
2474 if(inode->i_nlink == 0)
2475 set_nlink(inode, 1);
2478 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2479 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2480 inode->i_nlink == 2)
2481 set_nlink(inode, 1);
2483 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2484 spin_unlock(&obj->oo_guard);
2485 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2486 LINVRNT(osd_invariant(obj));
2491 static int osd_declare_object_ref_del(const struct lu_env *env,
2492 struct dt_object *dt,
2493 struct thandle *handle)
2495 struct osd_thandle *oh;
2497 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2498 LASSERT(handle != NULL);
2500 oh = container_of0(handle, struct osd_thandle, ot_super);
2501 LASSERT(oh->ot_handle == NULL);
2503 osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2504 osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2510 * Concurrency: @dt is write locked.
2512 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2515 struct osd_object *obj = osd_dt_obj(dt);
2516 struct inode *inode = obj->oo_inode;
2518 LINVRNT(osd_invariant(obj));
2519 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2520 LASSERT(osd_write_locked(env, obj));
2521 LASSERT(th != NULL);
2523 osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2525 spin_lock(&obj->oo_guard);
2526 LASSERT(inode->i_nlink > 0);
2528 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2529 * then the nlink count is 1. Don't let it be set to 0 or the directory
2530 * inode will be deleted incorrectly. */
2531 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2532 set_nlink(inode, 1);
2533 spin_unlock(&obj->oo_guard);
2534 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2535 LINVRNT(osd_invariant(obj));
2541 * Get the 64-bit version for an inode.
2543 static int osd_object_version_get(const struct lu_env *env,
2544 struct dt_object *dt, dt_obj_version_t *ver)
2546 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2548 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2549 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2550 *ver = LDISKFS_I(inode)->i_fs_version;
2555 * Concurrency: @dt is read locked.
2557 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2558 struct lu_buf *buf, const char *name,
2559 struct lustre_capa *capa)
2561 struct osd_object *obj = osd_dt_obj(dt);
2562 struct inode *inode = obj->oo_inode;
2563 struct osd_thread_info *info = osd_oti_get(env);
2564 struct dentry *dentry = &info->oti_obj_dentry;
2566 /* version get is not real XATTR but uses xattr API */
2567 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2568 /* for version we are just using xattr API but change inode
2570 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2571 osd_object_version_get(env, dt, buf->lb_buf);
2572 return sizeof(dt_obj_version_t);
2575 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2576 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2578 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2581 return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2585 static int osd_declare_xattr_set(const struct lu_env *env,
2586 struct dt_object *dt,
2587 const struct lu_buf *buf, const char *name,
2588 int fl, struct thandle *handle)
2590 struct osd_thandle *oh;
2592 LASSERT(handle != NULL);
2594 oh = container_of0(handle, struct osd_thandle, ot_super);
2595 LASSERT(oh->ot_handle == NULL);
2597 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2598 strcmp(name, XATTR_NAME_VERSION) == 0 ?
2599 osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2600 osd_dto_credits_noquota[DTO_XATTR_SET]);
2606 * Set the 64-bit version for object
2608 static void osd_object_version_set(const struct lu_env *env,
2609 struct dt_object *dt,
2610 dt_obj_version_t *new_version)
2612 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2614 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2615 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2617 LDISKFS_I(inode)->i_fs_version = *new_version;
2618 /** Version is set after all inode operations are finished,
2619 * so we should mark it dirty here */
2620 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2624 * Concurrency: @dt is write locked.
2626 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2627 const struct lu_buf *buf, const char *name, int fl,
2628 struct thandle *handle, struct lustre_capa *capa)
2630 struct osd_object *obj = osd_dt_obj(dt);
2631 struct inode *inode = obj->oo_inode;
2632 struct osd_thread_info *info = osd_oti_get(env);
2636 LASSERT(handle != NULL);
2638 /* version set is not real XATTR */
2639 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2640 /* for version we are just using xattr API but change inode
2642 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2643 osd_object_version_set(env, dt, buf->lb_buf);
2644 return sizeof(dt_obj_version_t);
2647 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2650 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2651 if (fl & LU_XATTR_REPLACE)
2652 fs_flags |= XATTR_REPLACE;
2654 if (fl & LU_XATTR_CREATE)
2655 fs_flags |= XATTR_CREATE;
2657 return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2662 * Concurrency: @dt is read locked.
2664 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2665 struct lu_buf *buf, struct lustre_capa *capa)
2667 struct osd_object *obj = osd_dt_obj(dt);
2668 struct inode *inode = obj->oo_inode;
2669 struct osd_thread_info *info = osd_oti_get(env);
2670 struct dentry *dentry = &info->oti_obj_dentry;
2672 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2673 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2674 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2676 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2679 dentry->d_inode = inode;
2680 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2683 static int osd_declare_xattr_del(const struct lu_env *env,
2684 struct dt_object *dt, const char *name,
2685 struct thandle *handle)
2687 struct osd_thandle *oh;
2689 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2690 LASSERT(handle != NULL);
2692 oh = container_of0(handle, struct osd_thandle, ot_super);
2693 LASSERT(oh->ot_handle == NULL);
2695 osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2696 osd_dto_credits_noquota[DTO_XATTR_SET]);
2702 * Concurrency: @dt is write locked.
2704 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2705 const char *name, struct thandle *handle,
2706 struct lustre_capa *capa)
2708 struct osd_object *obj = osd_dt_obj(dt);
2709 struct inode *inode = obj->oo_inode;
2710 struct osd_thread_info *info = osd_oti_get(env);
2711 struct dentry *dentry = &info->oti_obj_dentry;
2714 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2715 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2716 LASSERT(handle != NULL);
2718 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2721 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2723 ll_vfs_dq_init(inode);
2724 dentry->d_inode = inode;
2725 rc = inode->i_op->removexattr(dentry, name);
2729 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2730 struct dt_object *dt,
2731 struct lustre_capa *old,
2734 struct osd_thread_info *info = osd_oti_get(env);
2735 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2736 struct osd_object *obj = osd_dt_obj(dt);
2737 struct osd_device *dev = osd_obj2dev(obj);
2738 struct lustre_capa_key *key = &info->oti_capa_key;
2739 struct lustre_capa *capa = &info->oti_capa;
2740 struct obd_capa *oc;
2741 struct md_capainfo *ci;
2745 if (!dev->od_fl_capa)
2746 RETURN(ERR_PTR(-ENOENT));
2748 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2749 LINVRNT(osd_invariant(obj));
2751 /* renewal sanity check */
2752 if (old && osd_object_auth(env, dt, old, opc))
2753 RETURN(ERR_PTR(-EACCES));
2755 ci = md_capainfo(env);
2757 RETURN(ERR_PTR(-ENOENT));
2759 switch (ci->mc_auth) {
2763 capa->lc_uid = obj->oo_inode->i_uid;
2764 capa->lc_gid = obj->oo_inode->i_gid;
2765 capa->lc_flags = LC_ID_PLAIN;
2767 case LC_ID_CONVERT: {
2770 s[0] = obj->oo_inode->i_uid;
2771 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2772 s[2] = obj->oo_inode->i_gid;
2773 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2774 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2776 RETURN(ERR_PTR(rc));
2778 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2779 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2780 capa->lc_flags = LC_ID_CONVERT;
2784 RETURN(ERR_PTR(-EINVAL));
2787 capa->lc_fid = *fid;
2789 capa->lc_flags |= dev->od_capa_alg << 24;
2790 capa->lc_timeout = dev->od_capa_timeout;
2791 capa->lc_expiry = 0;
2793 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2795 LASSERT(!capa_is_expired(oc));
2799 spin_lock(&capa_lock);
2800 *key = dev->od_capa_keys[1];
2801 spin_unlock(&capa_lock);
2803 capa->lc_keyid = key->lk_keyid;
2804 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2806 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2808 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2809 RETURN(ERR_PTR(rc));
2812 oc = capa_add(dev->od_capa_hash, capa);
2816 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2818 struct osd_object *obj = osd_dt_obj(dt);
2819 struct inode *inode = obj->oo_inode;
2820 struct osd_thread_info *info = osd_oti_get(env);
2821 struct dentry *dentry = &info->oti_obj_dentry;
2822 struct file *file = &info->oti_file;
2827 dentry->d_inode = inode;
2828 file->f_dentry = dentry;
2829 file->f_mapping = inode->i_mapping;
2830 file->f_op = inode->i_fop;
2831 #ifndef HAVE_FILE_FSYNC_4ARGS
2832 mutex_lock(&inode->i_mutex);
2834 rc = do_fsync(file, 0);
2835 #ifndef HAVE_FILE_FSYNC_4ARGS
2836 mutex_unlock(&inode->i_mutex);
2841 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2844 struct osd_object *obj = osd_dt_obj(dt);
2847 *data = (void *)obj->oo_inode;
2855 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2856 const struct dt_index_features *feat)
2858 struct iam_descr *descr;
2860 if (osd_object_is_root(o))
2861 return feat == &dt_directory_features;
2863 LASSERT(o->oo_dir != NULL);
2865 descr = o->oo_dir->od_container.ic_descr;
2866 if (feat == &dt_directory_features) {
2867 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2873 feat->dif_keysize_min <= descr->id_key_size &&
2874 descr->id_key_size <= feat->dif_keysize_max &&
2875 feat->dif_recsize_min <= descr->id_rec_size &&
2876 descr->id_rec_size <= feat->dif_recsize_max &&
2877 !(feat->dif_flags & (DT_IND_VARKEY |
2878 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2879 ergo(feat->dif_flags & DT_IND_UPDATE,
2880 1 /* XXX check that object (and file system) is
2885 static int osd_iam_container_init(const struct lu_env *env,
2886 struct osd_object *obj,
2887 struct osd_directory *dir)
2889 struct iam_container *bag = &dir->od_container;
2892 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2896 result = iam_container_setup(bag);
2898 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2900 iam_container_fini(bag);
2907 * Concurrency: no external locking is necessary.
2909 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2910 const struct dt_index_features *feat)
2914 struct osd_object *obj = osd_dt_obj(dt);
2916 LINVRNT(osd_invariant(obj));
2918 if (osd_object_is_root(obj)) {
2919 dt->do_index_ops = &osd_index_ea_ops;
2921 } else if (feat == &dt_directory_features) {
2922 dt->do_index_ops = &osd_index_ea_ops;
2923 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2928 } else if (unlikely(feat == &dt_otable_features)) {
2929 dt->do_index_ops = &osd_otable_ops;
2931 } else if (unlikely(feat == &dt_acct_features)) {
2932 dt->do_index_ops = &osd_acct_index_ops;
2935 } else if (!osd_has_index(obj)) {
2936 struct osd_directory *dir;
2941 spin_lock(&obj->oo_guard);
2942 if (obj->oo_dir == NULL)
2946 * Concurrent thread allocated container data.
2949 spin_unlock(&obj->oo_guard);
2951 * Now, that we have container data, serialize its
2954 down_write(&obj->oo_ext_idx_sem);
2956 * recheck under lock.
2958 if (!osd_has_index(obj))
2959 result = osd_iam_container_init(env, obj, dir);
2962 up_write(&obj->oo_ext_idx_sem);
2970 if (result == 0 && skip_iam == 0) {
2971 if (!osd_iam_index_probe(env, obj, feat))
2974 LINVRNT(osd_invariant(obj));
2976 if (result == 0 && is_quota_glb_feat(feat) &&
2977 fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
2978 result = osd_quota_migration(env, dt, feat);
2983 static int osd_otable_it_attr_get(const struct lu_env *env,
2984 struct dt_object *dt,
2985 struct lu_attr *attr,
2986 struct lustre_capa *capa)
2992 static const struct dt_object_operations osd_obj_ops = {
2993 .do_read_lock = osd_object_read_lock,
2994 .do_write_lock = osd_object_write_lock,
2995 .do_read_unlock = osd_object_read_unlock,
2996 .do_write_unlock = osd_object_write_unlock,
2997 .do_write_locked = osd_object_write_locked,
2998 .do_attr_get = osd_attr_get,
2999 .do_declare_attr_set = osd_declare_attr_set,
3000 .do_attr_set = osd_attr_set,
3001 .do_ah_init = osd_ah_init,
3002 .do_declare_create = osd_declare_object_create,
3003 .do_create = osd_object_create,
3004 .do_declare_destroy = osd_declare_object_destroy,
3005 .do_destroy = osd_object_destroy,
3006 .do_index_try = osd_index_try,
3007 .do_declare_ref_add = osd_declare_object_ref_add,
3008 .do_ref_add = osd_object_ref_add,
3009 .do_declare_ref_del = osd_declare_object_ref_del,
3010 .do_ref_del = osd_object_ref_del,
3011 .do_xattr_get = osd_xattr_get,
3012 .do_declare_xattr_set = osd_declare_xattr_set,
3013 .do_xattr_set = osd_xattr_set,
3014 .do_declare_xattr_del = osd_declare_xattr_del,
3015 .do_xattr_del = osd_xattr_del,
3016 .do_xattr_list = osd_xattr_list,
3017 .do_capa_get = osd_capa_get,
3018 .do_object_sync = osd_object_sync,
3019 .do_data_get = osd_data_get,
3023 * dt_object_operations for interoperability mode
3024 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3026 static const struct dt_object_operations osd_obj_ea_ops = {
3027 .do_read_lock = osd_object_read_lock,
3028 .do_write_lock = osd_object_write_lock,
3029 .do_read_unlock = osd_object_read_unlock,
3030 .do_write_unlock = osd_object_write_unlock,
3031 .do_write_locked = osd_object_write_locked,
3032 .do_attr_get = osd_attr_get,
3033 .do_declare_attr_set = osd_declare_attr_set,
3034 .do_attr_set = osd_attr_set,
3035 .do_ah_init = osd_ah_init,
3036 .do_declare_create = osd_declare_object_create,
3037 .do_create = osd_object_ea_create,
3038 .do_declare_destroy = osd_declare_object_destroy,
3039 .do_destroy = osd_object_destroy,
3040 .do_index_try = osd_index_try,
3041 .do_declare_ref_add = osd_declare_object_ref_add,
3042 .do_ref_add = osd_object_ref_add,
3043 .do_declare_ref_del = osd_declare_object_ref_del,
3044 .do_ref_del = osd_object_ref_del,
3045 .do_xattr_get = osd_xattr_get,
3046 .do_declare_xattr_set = osd_declare_xattr_set,
3047 .do_xattr_set = osd_xattr_set,
3048 .do_declare_xattr_del = osd_declare_xattr_del,
3049 .do_xattr_del = osd_xattr_del,
3050 .do_xattr_list = osd_xattr_list,
3051 .do_capa_get = osd_capa_get,
3052 .do_object_sync = osd_object_sync,
3053 .do_data_get = osd_data_get,
3056 static const struct dt_object_operations osd_obj_otable_it_ops = {
3057 .do_attr_get = osd_otable_it_attr_get,
3058 .do_index_try = osd_index_try,
3061 static int osd_index_declare_iam_delete(const struct lu_env *env,
3062 struct dt_object *dt,
3063 const struct dt_key *key,
3064 struct thandle *handle)
3066 struct osd_thandle *oh;
3068 oh = container_of0(handle, struct osd_thandle, ot_super);
3069 LASSERT(oh->ot_handle == NULL);
3071 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3072 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3078 * delete a (key, value) pair from index \a dt specified by \a key
3080 * \param dt osd index object
3081 * \param key key for index
3082 * \param rec record reference
3083 * \param handle transaction handler
3086 * \retval -ve failure
3089 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3090 const struct dt_key *key,
3091 struct thandle *handle,
3092 struct lustre_capa *capa)
3094 struct osd_thread_info *oti = osd_oti_get(env);
3095 struct osd_object *obj = osd_dt_obj(dt);
3096 struct osd_thandle *oh;
3097 struct iam_path_descr *ipd;
3098 struct iam_container *bag = &obj->oo_dir->od_container;
3103 LINVRNT(osd_invariant(obj));
3104 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3105 LASSERT(bag->ic_object == obj->oo_inode);
3106 LASSERT(handle != NULL);
3108 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3111 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3113 ipd = osd_idx_ipd_get(env, bag);
3114 if (unlikely(ipd == NULL))
3117 oh = container_of0(handle, struct osd_thandle, ot_super);
3118 LASSERT(oh->ot_handle != NULL);
3119 LASSERT(oh->ot_handle->h_transaction != NULL);
3121 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3122 /* swab quota uid/gid provided by caller */
3123 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3124 key = (const struct dt_key *)&oti->oti_quota_id;
3127 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3128 osd_ipd_put(env, bag, ipd);
3129 LINVRNT(osd_invariant(obj));
3133 static int osd_index_declare_ea_delete(const struct lu_env *env,
3134 struct dt_object *dt,
3135 const struct dt_key *key,
3136 struct thandle *handle)
3138 struct osd_thandle *oh;
3139 struct inode *inode;
3143 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3144 LASSERT(handle != NULL);
3146 oh = container_of0(handle, struct osd_thandle, ot_super);
3147 LASSERT(oh->ot_handle == NULL);
3149 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3150 osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3152 inode = osd_dt_obj(dt)->oo_inode;
3155 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3156 true, true, NULL, false);
3160 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3163 struct osd_fid_pack *rec;
3166 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3167 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3168 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3173 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3176 struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
3177 struct seq_server_site *ss = osd_seq_site(osd);
3181 /* Those FID seqs, which are not in FLDB, must be local seq */
3182 if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
3185 rc = osd_fld_lookup(env, osd, fid, range);
3187 CERROR("%s: Can not lookup fld for "DFID"\n",
3188 osd_name(osd), PFID(fid));
3192 RETURN(ss->ss_node_id != range->lsr_index);
3196 * Index delete function for interoperability mode (b11826).
3197 * It will remove the directory entry added by osd_index_ea_insert().
3198 * This entry is needed to maintain name->fid mapping.
3200 * \param key, key i.e. file entry to be deleted
3202 * \retval 0, on success
3203 * \retval -ve, on error
3205 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3206 const struct dt_key *key,
3207 struct thandle *handle,
3208 struct lustre_capa *capa)
3210 struct osd_object *obj = osd_dt_obj(dt);
3211 struct inode *dir = obj->oo_inode;
3212 struct dentry *dentry;
3213 struct osd_thandle *oh;
3214 struct ldiskfs_dir_entry_2 *de = NULL;
3215 struct buffer_head *bh;
3216 struct htree_lock *hlock = NULL;
3217 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
3218 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3222 LINVRNT(osd_invariant(obj));
3223 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3224 LASSERT(handle != NULL);
3226 osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3228 oh = container_of(handle, struct osd_thandle, ot_super);
3229 LASSERT(oh->ot_handle != NULL);
3230 LASSERT(oh->ot_handle->h_transaction != NULL);
3232 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3235 ll_vfs_dq_init(dir);
3236 dentry = osd_child_dentry_get(env, obj,
3237 (char *)key, strlen((char *)key));
3239 if (obj->oo_hl_head != NULL) {
3240 hlock = osd_oti_get(env)->oti_hlock;
3241 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3242 dir, LDISKFS_HLOCK_DEL);
3244 down_write(&obj->oo_ext_idx_sem);
3247 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3251 /* Tried to delete local agent inode for the entries,
3252 * If it tries to delete .., which only happens during
3253 * rename or move the dir to /ORPHAN, even .. might
3254 * point to the remote parent, but it do not have
3255 * local agent inode, so we do not need check .. at all,
3256 * Note: we need delete entry first, then inode to keep
3258 if (strcmp((char *)key, dotdot) != 0) {
3259 LASSERT(de != NULL);
3260 rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3262 unlikely(osd_remote_fid(env, osd, fid)))
3263 /* Need to delete agent inode */
3264 ino = le32_to_cpu(de->inode);
3266 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3268 if (rc == 0 && unlikely(ino != 0)) {
3269 rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3272 CERROR("%s: del local inode "DFID": rc = %d\n",
3273 osd_name(osd), PFID(fid), rc);
3279 ldiskfs_htree_unlock(hlock);
3281 up_write(&obj->oo_ext_idx_sem);
3286 /* For inode on the remote MDT, .. will point to
3287 * /Agent directory, Check whether it needs to delete
3288 * from agent directory */
3289 if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3290 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3292 if (rc != 0 && rc != -ENOENT) {
3293 CERROR("%s: delete agent inode "DFID": rc = %d\n",
3294 osd_name(osd), PFID(fid), rc);
3304 LASSERT(osd_invariant(obj));
3309 * Lookup index for \a key and copy record to \a rec.
3311 * \param dt osd index object
3312 * \param key key for index
3313 * \param rec record reference
3315 * \retval +ve success : exact mach
3316 * \retval 0 return record with key not greater than \a key
3317 * \retval -ve failure
3319 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3320 struct dt_rec *rec, const struct dt_key *key,
3321 struct lustre_capa *capa)
3323 struct osd_object *obj = osd_dt_obj(dt);
3324 struct iam_path_descr *ipd;
3325 struct iam_container *bag = &obj->oo_dir->od_container;
3326 struct osd_thread_info *oti = osd_oti_get(env);
3327 struct iam_iterator *it = &oti->oti_idx_it;
3328 struct iam_rec *iam_rec;
3333 LASSERT(osd_invariant(obj));
3334 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3335 LASSERT(bag->ic_object == obj->oo_inode);
3337 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3340 ipd = osd_idx_ipd_get(env, bag);
3344 /* got ipd now we can start iterator. */
3345 iam_it_init(it, bag, 0, ipd);
3347 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3348 /* swab quota uid/gid provided by caller */
3349 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3350 key = (const struct dt_key *)&oti->oti_quota_id;
3353 rc = iam_it_get(it, (struct iam_key *)key);
3355 if (S_ISDIR(obj->oo_inode->i_mode))
3356 iam_rec = (struct iam_rec *)oti->oti_ldp;
3358 iam_rec = (struct iam_rec *) rec;
3360 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3362 if (S_ISDIR(obj->oo_inode->i_mode))
3363 osd_fid_unpack((struct lu_fid *) rec,
3364 (struct osd_fid_pack *)iam_rec);
3365 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3366 osd_quota_unpack(obj, rec);
3371 osd_ipd_put(env, bag, ipd);
3373 LINVRNT(osd_invariant(obj));
3378 static int osd_index_declare_iam_insert(const struct lu_env *env,
3379 struct dt_object *dt,
3380 const struct dt_rec *rec,
3381 const struct dt_key *key,
3382 struct thandle *handle)
3384 struct osd_thandle *oh;
3386 LASSERT(handle != NULL);
3388 oh = container_of0(handle, struct osd_thandle, ot_super);
3389 LASSERT(oh->ot_handle == NULL);
3391 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3392 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3398 * Inserts (key, value) pair in \a dt index object.
3400 * \param dt osd index object
3401 * \param key key for index
3402 * \param rec record reference
3403 * \param th transaction handler
3406 * \retval -ve failure
3408 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3409 const struct dt_rec *rec,
3410 const struct dt_key *key, struct thandle *th,
3411 struct lustre_capa *capa, int ignore_quota)
3413 struct osd_object *obj = osd_dt_obj(dt);
3414 struct iam_path_descr *ipd;
3415 struct osd_thandle *oh;
3416 struct iam_container *bag = &obj->oo_dir->od_container;
3417 struct osd_thread_info *oti = osd_oti_get(env);
3418 struct iam_rec *iam_rec;
3423 LINVRNT(osd_invariant(obj));
3424 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3425 LASSERT(bag->ic_object == obj->oo_inode);
3426 LASSERT(th != NULL);
3428 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3431 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3433 ipd = osd_idx_ipd_get(env, bag);
3434 if (unlikely(ipd == NULL))
3437 oh = container_of0(th, struct osd_thandle, ot_super);
3438 LASSERT(oh->ot_handle != NULL);
3439 LASSERT(oh->ot_handle->h_transaction != NULL);
3440 if (S_ISDIR(obj->oo_inode->i_mode)) {
3441 iam_rec = (struct iam_rec *)oti->oti_ldp;
3442 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3443 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3444 /* pack quota uid/gid */
3445 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3446 key = (const struct dt_key *)&oti->oti_quota_id;
3447 /* pack quota record */
3448 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3449 iam_rec = (struct iam_rec *)rec;
3451 iam_rec = (struct iam_rec *)rec;
3454 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3456 osd_ipd_put(env, bag, ipd);
3457 LINVRNT(osd_invariant(obj));
3462 * Calls ldiskfs_add_entry() to add directory entry
3463 * into the directory. This is required for
3464 * interoperability mode (b11826)
3466 * \retval 0, on success
3467 * \retval -ve, on error
3469 static int __osd_ea_add_rec(struct osd_thread_info *info,
3470 struct osd_object *pobj, struct inode *cinode,
3471 const char *name, const struct dt_rec *fid,
3472 struct htree_lock *hlock, struct thandle *th)
3474 struct ldiskfs_dentry_param *ldp;
3475 struct dentry *child;
3476 struct osd_thandle *oth;
3479 oth = container_of(th, struct osd_thandle, ot_super);
3480 LASSERT(oth->ot_handle != NULL);
3481 LASSERT(oth->ot_handle->h_transaction != NULL);
3482 LASSERT(pobj->oo_inode);
3484 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3485 if (unlikely(pobj->oo_inode ==
3486 osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3489 osd_get_ldiskfs_dirent_param(ldp, fid);
3490 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3491 child->d_fsdata = (void *)ldp;
3492 ll_vfs_dq_init(pobj->oo_inode);
3493 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3499 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3500 * into the directory.Also sets flags into osd object to
3501 * indicate dot and dotdot are created. This is required for
3502 * interoperability mode (b11826)
3504 * \param dir directory for dot and dotdot fixup.
3505 * \param obj child object for linking
3507 * \retval 0, on success
3508 * \retval -ve, on error
3510 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3511 struct osd_object *dir,
3512 struct inode *parent_dir, const char *name,
3513 const struct dt_rec *dot_fid,
3514 const struct dt_rec *dot_dot_fid,
3517 struct inode *inode = dir->oo_inode;
3518 struct osd_thandle *oth;
3521 oth = container_of(th, struct osd_thandle, ot_super);
3522 LASSERT(oth->ot_handle->h_transaction != NULL);
3523 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3525 if (strcmp(name, dot) == 0) {
3526 if (dir->oo_compat_dot_created) {
3529 LASSERT(inode == parent_dir);
3530 dir->oo_compat_dot_created = 1;
3533 } else if (strcmp(name, dotdot) == 0) {
3534 if (!dir->oo_compat_dot_created)
3536 /* in case of rename, dotdot is already created */
3537 if (dir->oo_compat_dotdot_created) {
3538 return __osd_ea_add_rec(info, dir, parent_dir, name,
3539 dot_dot_fid, NULL, th);
3542 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3543 parent_dir, dot_fid,
3546 dir->oo_compat_dotdot_created = 1;
3554 * It will call the appropriate osd_add* function and return the
3555 * value, return by respective functions.
3557 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3558 struct inode *cinode, const char *name,
3559 const struct dt_rec *fid, struct thandle *th)
3561 struct osd_thread_info *info = osd_oti_get(env);
3562 struct htree_lock *hlock;
3565 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3567 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3569 if (hlock != NULL) {
3570 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3573 down_write(&pobj->oo_ext_idx_sem);
3575 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3576 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3579 if (hlock != NULL) {
3580 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3581 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3583 down_write(&pobj->oo_ext_idx_sem);
3586 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3587 struct lu_fid *tfid = &info->oti_fid;
3589 *tfid = *(const struct lu_fid *)fid;
3591 rc = __osd_ea_add_rec(info, pobj, cinode, name,
3592 (const struct dt_rec *)tfid,
3595 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3600 ldiskfs_htree_unlock(hlock);
3602 up_write(&pobj->oo_ext_idx_sem);
3608 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3609 struct osd_idmap_cache *oic)
3611 struct osd_scrub *scrub = &dev->od_scrub;
3612 struct lu_fid *fid = &oic->oic_fid;
3613 struct osd_inode_id *id = &oti->oti_id;
3618 if (!fid_is_norm(fid) && !fid_is_igif(fid))
3622 rc = osd_oi_lookup(oti, dev, fid, id, true);
3623 if (rc != 0 && rc != -ENOENT)
3626 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3629 if (thread_is_running(&scrub->os_thread)) {
3630 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3631 /* There is race condition between osd_oi_lookup and OI scrub.
3632 * The OI scrub finished just after osd_oi_lookup() failure.
3633 * Under such case, it is unnecessary to trigger OI scrub again,
3634 * but try to call osd_oi_lookup() again. */
3635 if (unlikely(rc == -EAGAIN))
3641 if (!dev->od_noscrub && ++once == 1) {
3642 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3644 rc = osd_scrub_start(dev);
3645 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3647 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3656 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3657 struct osd_device *dev,
3658 struct osd_idmap_cache *oic,
3659 struct lu_fid *fid, __u32 ino)
3661 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
3662 struct inode *inode;
3665 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3666 inode = osd_iget(oti, dev, &oic->oic_lid);
3667 if (IS_ERR(inode)) {
3668 fid_zero(&oic->oic_fid);
3669 return PTR_ERR(inode);
3672 rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3675 fid_zero(&oic->oic_fid);
3677 *fid = oic->oic_fid = lma->lma_self_fid;
3682 * Calls ->lookup() to find dentry. From dentry get inode and
3683 * read inode's ea to get fid. This is required for interoperability
3686 * \retval 0, on success
3687 * \retval -ve, on error
3689 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3690 struct dt_rec *rec, const struct dt_key *key)
3692 struct inode *dir = obj->oo_inode;
3693 struct dentry *dentry;
3694 struct ldiskfs_dir_entry_2 *de;
3695 struct buffer_head *bh;
3696 struct lu_fid *fid = (struct lu_fid *) rec;
3697 struct htree_lock *hlock = NULL;
3702 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3704 dentry = osd_child_dentry_get(env, obj,
3705 (char *)key, strlen((char *)key));
3707 if (obj->oo_hl_head != NULL) {
3708 hlock = osd_oti_get(env)->oti_hlock;
3709 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3710 dir, LDISKFS_HLOCK_LOOKUP);
3712 down_read(&obj->oo_ext_idx_sem);
3715 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3717 struct osd_thread_info *oti = osd_oti_get(env);
3718 struct osd_inode_id *id = &oti->oti_id;
3719 struct osd_idmap_cache *oic = &oti->oti_cache;
3720 struct osd_device *dev = osd_obj2dev(obj);
3721 struct osd_scrub *scrub = &dev->od_scrub;
3722 struct scrub_file *sf = &scrub->os_file;
3724 ino = le32_to_cpu(de->inode);
3725 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
3727 rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
3731 rc = osd_get_fid_from_dentry(de, rec);
3733 /* done with de, release bh */
3736 rc = osd_ea_fid_get(env, obj, ino, fid, id);
3738 osd_id_gen(id, ino, OSD_OII_NOGEN);
3739 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
3740 fid_zero(&oic->oic_fid);
3745 oic->oic_fid = *fid;
3746 if ((scrub->os_pos_current <= ino) &&
3747 ((sf->sf_flags & SF_INCONSISTENT) ||
3748 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
3749 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3751 osd_consistency_check(oti, dev, oic);
3760 ldiskfs_htree_unlock(hlock);
3762 up_read(&obj->oo_ext_idx_sem);
3767 * Find the osd object for given fid.
3769 * \param fid need to find the osd object having this fid
3771 * \retval osd_object on success
3772 * \retval -ve on error
3774 struct osd_object *osd_object_find(const struct lu_env *env,
3775 struct dt_object *dt,
3776 const struct lu_fid *fid)
3778 struct lu_device *ludev = dt->do_lu.lo_dev;
3779 struct osd_object *child = NULL;
3780 struct lu_object *luch;
3781 struct lu_object *lo;
3784 * at this point topdev might not exist yet
3785 * (i.e. MGS is preparing profiles). so we can
3786 * not rely on topdev and instead lookup with
3787 * our device passed as topdev. this can't work
3788 * if the object isn't cached yet (as osd doesn't
3789 * allocate lu_header). IOW, the object must be
3790 * in the cache, otherwise lu_object_alloc() crashes
3793 luch = lu_object_find_at(env, ludev, fid, NULL);
3794 if (!IS_ERR(luch)) {
3795 if (lu_object_exists(luch)) {
3796 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3798 child = osd_obj(lo);
3800 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3801 "lu_object can't be located"
3802 DFID"\n", PFID(fid));
3804 if (child == NULL) {
3805 lu_object_put(env, luch);
3806 CERROR("Unable to get osd_object\n");
3807 child = ERR_PTR(-ENOENT);
3810 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3811 "lu_object does not exists "DFID"\n",
3813 lu_object_put(env, luch);
3814 child = ERR_PTR(-ENOENT);
3817 child = (void *)luch;
3823 * Put the osd object once done with it.
3825 * \param obj osd object that needs to be put
3827 static inline void osd_object_put(const struct lu_env *env,
3828 struct osd_object *obj)
3830 lu_object_put(env, &obj->oo_dt.do_lu);
3833 static int osd_index_declare_ea_insert(const struct lu_env *env,
3834 struct dt_object *dt,
3835 const struct dt_rec *rec,
3836 const struct dt_key *key,
3837 struct thandle *handle)
3839 struct osd_thandle *oh;
3840 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3841 struct lu_fid *fid = (struct lu_fid *)rec;
3845 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3846 LASSERT(handle != NULL);
3848 oh = container_of0(handle, struct osd_thandle, ot_super);
3849 LASSERT(oh->ot_handle == NULL);
3851 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3852 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3854 if (osd_dt_obj(dt)->oo_inode == NULL) {
3855 const char *name = (const char *)key;
3856 /* Object is not being created yet. Only happens when
3857 * 1. declare directory create
3858 * 2. declare insert .
3859 * 3. declare insert ..
3861 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
3863 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3865 /* We ignore block quota on meta pool (MDTs), so needn't
3866 * calculate how many blocks will be consumed by this index
3868 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
3869 oh, true, true, NULL, false);
3875 rc = osd_remote_fid(env, osd, fid);
3881 osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3882 osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
3883 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3884 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3885 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3886 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3892 * Index add function for interoperability mode (b11826).
3893 * It will add the directory entry.This entry is needed to
3894 * maintain name->fid mapping.
3896 * \param key it is key i.e. file entry to be inserted
3897 * \param rec it is value of given key i.e. fid
3899 * \retval 0, on success
3900 * \retval -ve, on error
3902 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3903 const struct dt_rec *rec,
3904 const struct dt_key *key, struct thandle *th,
3905 struct lustre_capa *capa, int ignore_quota)
3907 struct osd_object *obj = osd_dt_obj(dt);
3908 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3909 struct lu_fid *fid = (struct lu_fid *) rec;
3910 const char *name = (const char *)key;
3911 struct osd_thread_info *oti = osd_oti_get(env);
3912 struct osd_inode_id *id = &oti->oti_id;
3913 struct inode *child_inode = NULL;
3914 struct osd_object *child = NULL;
3918 LASSERT(osd_invariant(obj));
3919 LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3920 LASSERT(th != NULL);
3922 osd_trans_exec_op(env, th, OSD_OT_INSERT);
3924 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3927 LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
3929 rc = osd_remote_fid(env, osd, fid);
3931 CERROR("%s: Can not find object "DFID" rc %d\n",
3932 osd_name(osd), PFID(fid), rc);
3937 /* Insert remote entry */
3938 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
3939 struct osd_mdobj_map *omm = osd->od_mdt_map;
3940 struct osd_thandle *oh;
3942 /* If parent on remote MDT, we need put this object
3944 oh = container_of(th, typeof(*oh), ot_super);
3945 rc = osd_add_to_remote_parent(env, osd, obj, oh);
3947 CERROR("%s: add "DFID" error: rc = %d\n",
3949 PFID(lu_object_fid(&dt->do_lu)), rc);
3953 child_inode = igrab(omm->omm_remote_parent->d_inode);
3955 child_inode = osd_create_local_agent_inode(env, osd,
3958 if (IS_ERR(child_inode))
3959 RETURN(PTR_ERR(child_inode));
3962 /* Insert local entry */
3963 child = osd_object_find(env, dt, fid);
3964 if (IS_ERR(child)) {
3965 CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
3966 osd_name(osd), PFID(fid),
3967 id->oii_ino, id->oii_gen,
3968 (int)PTR_ERR(child_inode));
3969 RETURN(PTR_ERR(child_inode));
3971 child_inode = igrab(child->oo_inode);
3974 rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
3978 osd_object_put(env, child);
3979 LASSERT(osd_invariant(obj));
3984 * Initialize osd Iterator for given osd index object.
3986 * \param dt osd index object
3989 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3990 struct dt_object *dt,
3992 struct lustre_capa *capa)
3994 struct osd_it_iam *it;
3995 struct osd_thread_info *oti = osd_oti_get(env);
3996 struct osd_object *obj = osd_dt_obj(dt);
3997 struct lu_object *lo = &dt->do_lu;
3998 struct iam_path_descr *ipd;
3999 struct iam_container *bag = &obj->oo_dir->od_container;
4001 LASSERT(lu_object_exists(lo));
4003 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4004 return ERR_PTR(-EACCES);
4007 ipd = osd_it_ipd_get(env, bag);
4008 if (likely(ipd != NULL)) {
4012 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4013 return (struct dt_it *)it;
4015 return ERR_PTR(-ENOMEM);
4019 * free given Iterator.
4022 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4024 struct osd_it_iam *it = (struct osd_it_iam *)di;
4025 struct osd_object *obj = it->oi_obj;
4027 iam_it_fini(&it->oi_it);
4028 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4029 lu_object_put(env, &obj->oo_dt.do_lu);
4033 * Move Iterator to record specified by \a key
4035 * \param di osd iterator
4036 * \param key key for index
4038 * \retval +ve di points to record with least key not larger than key
4039 * \retval 0 di points to exact matched key
4040 * \retval -ve failure
4043 static int osd_it_iam_get(const struct lu_env *env,
4044 struct dt_it *di, const struct dt_key *key)
4046 struct osd_thread_info *oti = osd_oti_get(env);
4047 struct osd_it_iam *it = (struct osd_it_iam *)di;
4049 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4050 /* swab quota uid/gid */
4051 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4052 key = (struct dt_key *)&oti->oti_quota_id;
4055 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4061 * \param di osd iterator
4063 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4065 struct osd_it_iam *it = (struct osd_it_iam *)di;
4067 iam_it_put(&it->oi_it);
4071 * Move iterator by one record
4073 * \param di osd iterator
4075 * \retval +1 end of container reached
4077 * \retval -ve failure
4080 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4082 struct osd_it_iam *it = (struct osd_it_iam *)di;
4084 return iam_it_next(&it->oi_it);
4088 * Return pointer to the key under iterator.
4091 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4092 const struct dt_it *di)
4094 struct osd_thread_info *oti = osd_oti_get(env);
4095 struct osd_it_iam *it = (struct osd_it_iam *)di;
4096 struct osd_object *obj = it->oi_obj;
4099 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4101 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4102 /* swab quota uid/gid */
4103 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4104 key = (struct dt_key *)&oti->oti_quota_id;
4111 * Return size of key under iterator (in bytes)
4114 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4116 struct osd_it_iam *it = (struct osd_it_iam *)di;
4118 return iam_it_key_size(&it->oi_it);
4122 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4124 /* check if file type is required */
4125 if (ent->lde_attrs & LUDA_TYPE) {
4126 struct luda_type *lt;
4127 int align = sizeof(*lt) - 1;
4129 len = (len + align) & ~align;
4130 lt = (struct luda_type *)(ent->lde_name + len);
4131 lt->lt_type = cpu_to_le16(DTTOIF(type));
4134 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4138 * build lu direct from backend fs dirent.
4142 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4143 char *name, __u16 namelen, __u16 type, __u32 attr)
4145 ent->lde_attrs = attr | LUDA_FID;
4146 fid_cpu_to_le(&ent->lde_fid, fid);
4148 ent->lde_hash = cpu_to_le64(offset);
4149 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4151 strncpy(ent->lde_name, name, namelen);
4152 ent->lde_name[namelen] = '\0';
4153 ent->lde_namelen = cpu_to_le16(namelen);
4155 /* append lustre attributes */
4156 osd_it_append_attrs(ent, namelen, type);
4160 * Return pointer to the record under iterator.
4162 static int osd_it_iam_rec(const struct lu_env *env,
4163 const struct dt_it *di,
4164 struct dt_rec *dtrec, __u32 attr)
4166 struct osd_it_iam *it = (struct osd_it_iam *)di;
4167 struct osd_thread_info *info = osd_oti_get(env);
4170 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4171 const struct osd_fid_pack *rec;
4172 struct lu_fid *fid = &info->oti_fid;
4173 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4179 name = (char *)iam_it_key_get(&it->oi_it);
4181 RETURN(PTR_ERR(name));
4183 namelen = iam_it_key_size(&it->oi_it);
4185 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4187 RETURN(PTR_ERR(rec));
4189 rc = osd_fid_unpack(fid, rec);
4193 hash = iam_it_store(&it->oi_it);
4195 /* IAM does not store object type in IAM index (dir) */
4196 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4198 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4199 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4200 (struct iam_rec *)dtrec);
4201 osd_quota_unpack(it->oi_obj, dtrec);
4203 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4204 (struct iam_rec *)dtrec);
4211 * Returns cookie for current Iterator position.
4213 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4215 struct osd_it_iam *it = (struct osd_it_iam *)di;
4217 return iam_it_store(&it->oi_it);
4221 * Restore iterator from cookie.
4223 * \param di osd iterator
4224 * \param hash Iterator location cookie
4226 * \retval +ve di points to record with least key not larger than key.
4227 * \retval 0 di points to exact matched key
4228 * \retval -ve failure
4231 static int osd_it_iam_load(const struct lu_env *env,
4232 const struct dt_it *di, __u64 hash)
4234 struct osd_it_iam *it = (struct osd_it_iam *)di;
4236 return iam_it_load(&it->oi_it, hash);
4239 static const struct dt_index_operations osd_index_iam_ops = {
4240 .dio_lookup = osd_index_iam_lookup,
4241 .dio_declare_insert = osd_index_declare_iam_insert,
4242 .dio_insert = osd_index_iam_insert,
4243 .dio_declare_delete = osd_index_declare_iam_delete,
4244 .dio_delete = osd_index_iam_delete,
4246 .init = osd_it_iam_init,
4247 .fini = osd_it_iam_fini,
4248 .get = osd_it_iam_get,
4249 .put = osd_it_iam_put,
4250 .next = osd_it_iam_next,
4251 .key = osd_it_iam_key,
4252 .key_size = osd_it_iam_key_size,
4253 .rec = osd_it_iam_rec,
4254 .store = osd_it_iam_store,
4255 .load = osd_it_iam_load
4261 * Creates or initializes iterator context.
4263 * \retval struct osd_it_ea, iterator structure on success
4266 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4267 struct dt_object *dt,
4269 struct lustre_capa *capa)
4271 struct osd_object *obj = osd_dt_obj(dt);
4272 struct osd_thread_info *info = osd_oti_get(env);
4273 struct osd_it_ea *it = &info->oti_it_ea;
4274 struct lu_object *lo = &dt->do_lu;
4275 struct dentry *obj_dentry = &info->oti_it_dentry;
4277 LASSERT(lu_object_exists(lo));
4279 obj_dentry->d_inode = obj->oo_inode;
4280 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4281 obj_dentry->d_name.hash = 0;
4283 it->oie_rd_dirent = 0;
4284 it->oie_it_dirent = 0;
4285 it->oie_dirent = NULL;
4286 it->oie_buf = info->oti_it_ea_buf;
4288 it->oie_file.f_pos = 0;
4289 it->oie_file.f_dentry = obj_dentry;
4290 if (attr & LUDA_64BITHASH)
4291 it->oie_file.f_mode |= FMODE_64BITHASH;
4293 it->oie_file.f_mode |= FMODE_32BITHASH;
4294 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
4295 it->oie_file.f_op = obj->oo_inode->i_fop;
4296 it->oie_file.private_data = NULL;
4298 RETURN((struct dt_it *) it);
4302 * Destroy or finishes iterator context.
4304 * \param di iterator structure to be destroyed
4306 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4308 struct osd_it_ea *it = (struct osd_it_ea *)di;
4309 struct osd_object *obj = it->oie_obj;
4310 struct inode *inode = obj->oo_inode;
4313 it->oie_file.f_op->release(inode, &it->oie_file);
4314 lu_object_put(env, &obj->oo_dt.do_lu);
4319 * It position the iterator at given key, so that next lookup continues from
4320 * that key Or it is similar to dio_it->load() but based on a key,
4321 * rather than file position.
4323 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4326 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4328 static int osd_it_ea_get(const struct lu_env *env,
4329 struct dt_it *di, const struct dt_key *key)
4331 struct osd_it_ea *it = (struct osd_it_ea *)di;
4334 LASSERT(((const char *)key)[0] == '\0');
4335 it->oie_file.f_pos = 0;
4336 it->oie_rd_dirent = 0;
4337 it->oie_it_dirent = 0;
4338 it->oie_dirent = NULL;
4346 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4351 * It is called internally by ->readdir(). It fills the
4352 * iterator's in-memory data structure with required
4353 * information i.e. name, namelen, rec_size etc.
4355 * \param buf in which information to be filled in.
4356 * \param name name of the file in given dir
4358 * \retval 0 on success
4359 * \retval 1 on buffer full
4361 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4362 loff_t offset, __u64 ino,
4365 struct osd_it_ea *it = (struct osd_it_ea *)buf;
4366 struct osd_object *obj = it->oie_obj;
4367 struct osd_it_ea_dirent *ent = it->oie_dirent;
4368 struct lu_fid *fid = &ent->oied_fid;
4369 struct osd_fid_pack *rec;
4372 /* this should never happen */
4373 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4374 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4378 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4382 /* "." is just the object itself. */
4383 if (namelen == 1 && name[0] == '.') {
4384 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4385 } else if (d_type & LDISKFS_DIRENT_LUFID) {
4386 rec = (struct osd_fid_pack*) (name + namelen + 1);
4387 if (osd_fid_unpack(fid, rec) != 0)
4392 d_type &= ~LDISKFS_DIRENT_LUFID;
4394 /* NOT export local root. */
4395 if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4396 ino = obj->oo_inode->i_ino;
4397 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4400 ent->oied_ino = ino;
4401 ent->oied_off = offset;
4402 ent->oied_namelen = namelen;
4403 ent->oied_type = d_type;
4405 memcpy(ent->oied_name, name, namelen);
4407 it->oie_rd_dirent++;
4408 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4413 * Calls ->readdir() to load a directory entry at a time
4414 * and stored it in iterator's in-memory data structure.
4416 * \param di iterator's in memory structure
4418 * \retval 0 on success
4419 * \retval -ve on error
4421 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4422 const struct dt_it *di)
4424 struct osd_it_ea *it = (struct osd_it_ea *)di;
4425 struct osd_object *obj = it->oie_obj;
4426 struct inode *inode = obj->oo_inode;
4427 struct htree_lock *hlock = NULL;
4431 it->oie_dirent = it->oie_buf;
4432 it->oie_rd_dirent = 0;
4434 if (obj->oo_hl_head != NULL) {
4435 hlock = osd_oti_get(env)->oti_hlock;
4436 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4437 inode, LDISKFS_HLOCK_READDIR);
4439 down_read(&obj->oo_ext_idx_sem);
4442 result = inode->i_fop->readdir(&it->oie_file, it,
4443 (filldir_t) osd_ldiskfs_filldir);
4446 ldiskfs_htree_unlock(hlock);
4448 up_read(&obj->oo_ext_idx_sem);
4450 if (it->oie_rd_dirent == 0) {
4453 it->oie_dirent = it->oie_buf;
4454 it->oie_it_dirent = 1;
4461 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4462 * to load a directory entry at a time and stored it in
4463 * iterator's in-memory data structure.
4465 * \param di iterator's in memory structure
4467 * \retval +ve iterator reached to end
4468 * \retval 0 iterator not reached to end
4469 * \retval -ve on error
4471 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4473 struct osd_it_ea *it = (struct osd_it_ea *)di;
4478 if (it->oie_it_dirent < it->oie_rd_dirent) {
4480 (void *) it->oie_dirent +
4481 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4482 it->oie_dirent->oied_namelen);
4483 it->oie_it_dirent++;
4486 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4489 rc = osd_ldiskfs_it_fill(env, di);
4496 * Returns the key at current position from iterator's in memory structure.
4498 * \param di iterator's in memory structure
4500 * \retval key i.e. struct dt_key on success
4502 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4503 const struct dt_it *di)
4505 struct osd_it_ea *it = (struct osd_it_ea *)di;
4507 return (struct dt_key *)it->oie_dirent->oied_name;
4511 * Returns the key's size at current position from iterator's in memory structure.
4513 * \param di iterator's in memory structure
4515 * \retval key_size i.e. struct dt_key on success
4517 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4519 struct osd_it_ea *it = (struct osd_it_ea *)di;
4521 return it->oie_dirent->oied_namelen;
4525 osd_dirent_update(handle_t *jh, struct super_block *sb,
4526 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4527 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4529 struct osd_fid_pack *rec;
4533 LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4534 LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4536 rc = ldiskfs_journal_get_write_access(jh, bh);
4538 CERROR("%.16s: fail to write access for update dirent: "
4539 "name = %.*s, rc = %d\n",
4540 LDISKFS_SB(sb)->s_es->s_volume_name,
4541 ent->oied_namelen, ent->oied_name, rc);
4545 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4546 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4547 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4549 CERROR("%.16s: fail to dirty metadata for update dirent: "
4550 "name = %.*s, rc = %d\n",
4551 LDISKFS_SB(sb)->s_es->s_volume_name,
4552 ent->oied_namelen, ent->oied_name, rc);
4558 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4560 if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4561 __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4568 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
4570 LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
4571 "dot_dotdot = %d\n", dot_dotdot);
4573 if (LDISKFS_DIR_REC_LEN(de) >=
4574 __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
4581 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
4582 struct inode *dir, struct inode *inode,
4583 struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4584 struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
4585 struct htree_lock *hlock)
4587 struct dentry *dentry;
4588 struct osd_fid_pack *rec;
4589 struct ldiskfs_dentry_param *ldp;
4593 if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
4594 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
4597 /* There is enough space to hold the FID-in-dirent. */
4598 if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
4599 dir->i_sb->s_blocksize)) {
4600 rc = ldiskfs_journal_get_write_access(jh, bh);
4602 CERROR("%.16s: fail to write access for reinsert "
4603 "dirent: name = %.*s, rc = %d\n",
4604 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4605 ent->oied_namelen, ent->oied_name, rc);
4609 de->name[de->name_len] = 0;
4610 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4611 rec->fp_len = sizeof(struct lu_fid) + 1;
4612 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4613 de->file_type |= LDISKFS_DIRENT_LUFID;
4615 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4617 CERROR("%.16s: fail to dirty metadata for reinsert "
4618 "dirent: name = %.*s, rc = %d\n",
4619 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4620 ent->oied_namelen, ent->oied_name, rc);
4625 rc = ldiskfs_delete_entry(jh, dir, de, bh);
4627 CERROR("%.16s: fail to delete entry for reinsert dirent: "
4628 "name = %.*s, rc = %d\n",
4629 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4630 ent->oied_namelen, ent->oied_name, rc);
4634 dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
4636 ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
4637 osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
4638 dentry->d_fsdata = (void *)ldp;
4639 ll_vfs_dq_init(dir);
4640 rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
4641 /* It is too bad, we cannot reinsert the name entry back.
4642 * That means we lose it! */
4644 CERROR("%.16s: fail to insert entry for reinsert dirent: "
4645 "name = %.*s, rc = %d\n",
4646 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4647 ent->oied_namelen, ent->oied_name, rc);
4653 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
4654 struct osd_it_ea *it, struct lu_fid *fid,
4655 struct osd_inode_id *id, __u32 *attr)
4657 struct osd_thread_info *info = osd_oti_get(env);
4658 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
4659 struct osd_device *dev = osd_obj2dev(obj);
4660 struct super_block *sb = osd_sb(dev);
4661 const char *devname =
4662 LDISKFS_SB(sb)->s_es->s_volume_name;
4663 struct osd_it_ea_dirent *ent = it->oie_dirent;
4664 struct inode *dir = obj->oo_inode;
4665 struct htree_lock *hlock = NULL;
4666 struct buffer_head *bh = NULL;
4667 handle_t *jh = NULL;
4668 struct ldiskfs_dir_entry_2 *de;
4669 struct dentry *dentry;
4670 struct inode *inode;
4677 if (ent->oied_name[0] == '.') {
4678 if (ent->oied_namelen == 1)
4680 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
4684 dentry = osd_child_dentry_get(env, obj, ent->oied_name,
4687 /* We need to ensure that the name entry is still valid.
4688 * Because it may be removed or renamed by other already.
4690 * The unlink or rename operation will start journal before PDO lock,
4691 * so to avoid deadlock, here we need to start journal handle before
4692 * related PDO lock also. But because we do not know whether there
4693 * will be something to be repaired before PDO lock, we just start
4694 * journal without conditions.
4696 * We may need to remove the name entry firstly, then insert back.
4697 * One credit is for user quota file update.
4698 * One credit is for group quota file update.
4699 * Two credits are for dirty inode. */
4700 credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4701 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
4704 if (dev->od_dirent_journal) {
4705 jh = ldiskfs_journal_start_sb(sb, credits);
4708 CERROR("%.16s: fail to start trans for dirent "
4709 "check_repair: credits %d, name %.*s, rc %d\n",
4710 devname, credits, ent->oied_namelen,
4711 ent->oied_name, rc);
4715 if (obj->oo_hl_head != NULL) {
4716 hlock = osd_oti_get(env)->oti_hlock;
4717 /* "0" means exclusive lock for the whole directory.
4718 * We need to prevent others access such name entry
4719 * during the delete + insert. Neither HLOCK_ADD nor
4720 * HLOCK_DEL cannot guarantee the atomicity. */
4721 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
4723 down_write(&obj->oo_ext_idx_sem);
4726 if (obj->oo_hl_head != NULL) {
4727 hlock = osd_oti_get(env)->oti_hlock;
4728 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
4729 LDISKFS_HLOCK_LOOKUP);
4731 down_read(&obj->oo_ext_idx_sem);
4735 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
4736 /* For dot/dotdot entry, if there is not enough space to hold the
4737 * FID-in-dirent, just keep them there. It only happens when the
4738 * device upgraded from 1.8 or restored from MDT file-level backup.
4739 * For the whole directory, only dot/dotdot entry have no FID-in-dirent
4740 * and needs to get FID from LMA when readdir, it will not affect the
4741 * performance much. */
4742 if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
4743 (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
4744 *attr |= LUDA_IGNORE;
4745 GOTO(out_journal, rc = 0);
4748 osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
4749 inode = osd_iget(info, dev, id);
4750 if (IS_ERR(inode)) {
4751 rc = PTR_ERR(inode);
4752 if (rc == -ENOENT || rc == -ESTALE) {
4753 *attr |= LUDA_IGNORE;
4757 GOTO(out_journal, rc);
4760 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
4762 if (fid_is_sane(fid)) {
4763 /* FID-in-dirent is valid. */
4764 if (lu_fid_eq(fid, &lma->lma_self_fid))
4765 GOTO(out_inode, rc = 0);
4767 /* Do not repair under dryrun mode. */
4768 if (*attr & LUDA_VERIFY_DRYRUN) {
4769 *attr |= LUDA_REPAIR;
4770 GOTO(out_inode, rc = 0);
4773 if (!dev->od_dirent_journal) {
4777 ldiskfs_htree_unlock(hlock);
4779 up_read(&obj->oo_ext_idx_sem);
4780 dev->od_dirent_journal = 1;
4784 *fid = lma->lma_self_fid;
4786 /* Update the FID-in-dirent. */
4787 rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
4789 *attr |= LUDA_REPAIR;
4791 /* Do not repair under dryrun mode. */
4792 if (*attr & LUDA_VERIFY_DRYRUN) {
4793 *fid = lma->lma_self_fid;
4794 *attr |= LUDA_REPAIR;
4795 GOTO(out_inode, rc = 0);
4798 if (!dev->od_dirent_journal) {
4802 ldiskfs_htree_unlock(hlock);
4804 up_read(&obj->oo_ext_idx_sem);
4805 dev->od_dirent_journal = 1;
4809 *fid = lma->lma_self_fid;
4811 /* Append the FID-in-dirent. */
4812 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4813 fid, bh, de, hlock);
4815 *attr |= LUDA_REPAIR;
4817 } else if (rc == -ENODATA) {
4818 /* Do not repair under dryrun mode. */
4819 if (*attr & LUDA_VERIFY_DRYRUN) {
4820 if (fid_is_sane(fid)) {
4821 *attr |= LUDA_REPAIR;
4823 lu_igif_build(fid, inode->i_ino,
4824 inode->i_generation);
4825 *attr |= LUDA_UPGRADE;
4827 GOTO(out_inode, rc = 0);
4830 if (!dev->od_dirent_journal) {
4834 ldiskfs_htree_unlock(hlock);
4836 up_read(&obj->oo_ext_idx_sem);
4837 dev->od_dirent_journal = 1;
4842 if (unlikely(fid_is_sane(fid))) {
4843 /* FID-in-dirent exists, but FID-in-LMA is lost.
4844 * Trust the FID-in-dirent, and add FID-in-LMA. */
4845 rc = osd_ea_fid_set(info, inode, fid, 0);
4847 *attr |= LUDA_REPAIR;
4849 lu_igif_build(fid, inode->i_ino, inode->i_generation);
4850 /* It is probably IGIF object. Only aappend the
4851 * FID-in-dirent. OI scrub will process FID-in-LMA. */
4852 rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4853 fid, bh, de, hlock);
4855 *attr |= LUDA_UPGRADE;
4859 GOTO(out_inode, rc);
4866 if (hlock != NULL) {
4867 ldiskfs_htree_unlock(hlock);
4869 if (dev->od_dirent_journal)
4870 up_write(&obj->oo_ext_idx_sem);
4872 up_read(&obj->oo_ext_idx_sem);
4875 ldiskfs_journal_stop(jh);
4876 if (rc >= 0 && !dirty)
4877 dev->od_dirent_journal = 0;
4882 * Returns the value at current position from iterator's in memory structure.
4884 * \param di struct osd_it_ea, iterator's in memory structure
4885 * \param attr attr requested for dirent.
4886 * \param lde lustre dirent
4888 * \retval 0 no error and \param lde has correct lustre dirent.
4889 * \retval -ve on error
4891 static inline int osd_it_ea_rec(const struct lu_env *env,
4892 const struct dt_it *di,
4893 struct dt_rec *dtrec, __u32 attr)
4895 struct osd_it_ea *it = (struct osd_it_ea *)di;
4896 struct osd_object *obj = it->oie_obj;
4897 struct osd_device *dev = osd_obj2dev(obj);
4898 struct osd_scrub *scrub = &dev->od_scrub;
4899 struct scrub_file *sf = &scrub->os_file;
4900 struct osd_thread_info *oti = osd_oti_get(env);
4901 struct osd_inode_id *id = &oti->oti_id;
4902 struct osd_idmap_cache *oic = &oti->oti_cache;
4903 struct lu_fid *fid = &it->oie_dirent->oied_fid;
4904 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4905 __u32 ino = it->oie_dirent->oied_ino;
4909 if (attr & LUDA_VERIFY) {
4911 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
4912 attr |= LUDA_IGNORE;
4917 rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr);
4919 attr &= ~LU_DIRENT_ATTRS_MASK;
4920 if (!fid_is_sane(fid)) {
4921 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
4924 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4926 osd_id_gen(id, ino, OSD_OII_NOGEN);
4934 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4935 it->oie_dirent->oied_name,
4936 it->oie_dirent->oied_namelen,
4937 it->oie_dirent->oied_type, attr);
4939 if (osd_remote_fid(env, dev, fid))
4942 if (likely(!(attr & LUDA_IGNORE))) {
4944 oic->oic_fid = *fid;
4947 if (!(attr & LUDA_VERIFY) &&
4948 (scrub->os_pos_current <= ino) &&
4949 ((sf->sf_flags & SF_INCONSISTENT) ||
4950 (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4951 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4952 osd_consistency_check(oti, dev, oic);
4958 * Returns a cookie for current position of the iterator head, so that
4959 * user can use this cookie to load/start the iterator next time.
4961 * \param di iterator's in memory structure
4963 * \retval cookie for current position, on success
4965 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4967 struct osd_it_ea *it = (struct osd_it_ea *)di;
4969 return it->oie_dirent->oied_off;
4973 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4974 * to load a directory entry at a time and stored it i inn,
4975 * in iterator's in-memory data structure.
4977 * \param di struct osd_it_ea, iterator's in memory structure
4979 * \retval +ve on success
4980 * \retval -ve on error
4982 static int osd_it_ea_load(const struct lu_env *env,
4983 const struct dt_it *di, __u64 hash)
4985 struct osd_it_ea *it = (struct osd_it_ea *)di;
4989 it->oie_file.f_pos = hash;
4991 rc = osd_ldiskfs_it_fill(env, di);
4999 * Index lookup function for interoperability mode (b11826).
5001 * \param key, key i.e. file name to be searched
5003 * \retval +ve, on success
5004 * \retval -ve, on error
5006 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5007 struct dt_rec *rec, const struct dt_key *key,
5008 struct lustre_capa *capa)
5010 struct osd_object *obj = osd_dt_obj(dt);
5015 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5016 LINVRNT(osd_invariant(obj));
5018 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5021 rc = osd_ea_lookup_rec(env, obj, rec, key);
5028 * Index and Iterator operations for interoperability
5029 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5031 static const struct dt_index_operations osd_index_ea_ops = {
5032 .dio_lookup = osd_index_ea_lookup,
5033 .dio_declare_insert = osd_index_declare_ea_insert,
5034 .dio_insert = osd_index_ea_insert,
5035 .dio_declare_delete = osd_index_declare_ea_delete,
5036 .dio_delete = osd_index_ea_delete,
5038 .init = osd_it_ea_init,
5039 .fini = osd_it_ea_fini,
5040 .get = osd_it_ea_get,
5041 .put = osd_it_ea_put,
5042 .next = osd_it_ea_next,
5043 .key = osd_it_ea_key,
5044 .key_size = osd_it_ea_key_size,
5045 .rec = osd_it_ea_rec,
5046 .store = osd_it_ea_store,
5047 .load = osd_it_ea_load
5051 static void *osd_key_init(const struct lu_context *ctx,
5052 struct lu_context_key *key)
5054 struct osd_thread_info *info;
5056 OBD_ALLOC_PTR(info);
5058 return ERR_PTR(-ENOMEM);
5060 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5061 if (info->oti_it_ea_buf == NULL)
5064 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5066 info->oti_hlock = ldiskfs_htree_lock_alloc();
5067 if (info->oti_hlock == NULL)
5073 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5076 return ERR_PTR(-ENOMEM);
5079 static void osd_key_fini(const struct lu_context *ctx,
5080 struct lu_context_key *key, void* data)
5082 struct osd_thread_info *info = data;
5084 if (info->oti_hlock != NULL)
5085 ldiskfs_htree_lock_free(info->oti_hlock);
5086 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5087 lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5088 lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5092 static void osd_key_exit(const struct lu_context *ctx,
5093 struct lu_context_key *key, void *data)
5095 struct osd_thread_info *info = data;
5097 LASSERT(info->oti_r_locks == 0);
5098 LASSERT(info->oti_w_locks == 0);
5099 LASSERT(info->oti_txns == 0);
5102 /* type constructor/destructor: osd_type_init, osd_type_fini */
5103 LU_TYPE_INIT_FINI(osd, &osd_key);
5105 struct lu_context_key osd_key = {
5106 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5107 .lct_init = osd_key_init,
5108 .lct_fini = osd_key_fini,
5109 .lct_exit = osd_key_exit
5113 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5114 const char *name, struct lu_device *next)
5116 struct osd_device *osd = osd_dev(d);
5118 if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5119 >= sizeof(osd->od_svname))
5121 return osd_procfs_init(osd, name);
5124 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5128 osd_scrub_cleanup(env, o);
5131 fsfilt_put_ops(o->od_fsops);
5135 /* shutdown quota slave instance associated with the device */
5136 if (o->od_quota_slave != NULL) {
5137 qsd_fini(env, o->od_quota_slave);
5138 o->od_quota_slave = NULL;
5144 static int osd_mount(const struct lu_env *env,
5145 struct osd_device *o, struct lustre_cfg *cfg)
5147 const char *name = lustre_cfg_string(cfg, 0);
5148 const char *dev = lustre_cfg_string(cfg, 1);
5150 unsigned long page, s_flags, lmd_flags = 0;
5151 struct page *__page;
5152 struct file_system_type *type;
5153 char *options = NULL;
5158 if (o->od_mnt != NULL)
5161 if (strlen(dev) >= sizeof(o->od_mntdev))
5163 strcpy(o->od_mntdev, dev);
5165 o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
5166 if (o->od_fsops == NULL) {
5167 CERROR("Can't find fsfilt_ldiskfs\n");
5171 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
5173 GOTO(out, rc = -ENOMEM);
5175 str = lustre_cfg_string(cfg, 2);
5176 s_flags = simple_strtoul(str, NULL, 0);
5177 str = strstr(str, ":");
5179 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5180 opts = lustre_cfg_string(cfg, 3);
5181 page = (unsigned long)cfs_page_address(__page);
5182 options = (char *)page;
5185 strcat(options, "user_xattr,acl");
5187 strcat(options, opts);
5189 /* Glom up mount options */
5190 if (*options != '\0')
5191 strcat(options, ",");
5192 strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
5194 type = get_fs_type("ldiskfs");
5196 CERROR("%s: cannot find ldiskfs module\n", name);
5197 GOTO(out, rc = -ENODEV);
5200 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5201 cfs_module_put(type->owner);
5203 if (IS_ERR(o->od_mnt)) {
5204 rc = PTR_ERR(o->od_mnt);
5205 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5210 #ifdef HAVE_DEV_SET_RDONLY
5211 if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5212 CERROR("%s: underlying device %s is marked as read-only. "
5213 "Setup failed\n", name, dev);
5216 GOTO(out, rc = -EROFS);
5220 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5221 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5222 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5225 GOTO(out, rc = -EINVAL);
5228 ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
5229 LDISKFS_STATE_LUSTRE_NO_OI);
5230 if (lmd_flags & LMD_FLG_NOSCRUB)
5235 OBD_PAGE_FREE(__page);
5237 fsfilt_put_ops(o->od_fsops);
5242 static struct lu_device *osd_device_fini(const struct lu_env *env,
5243 struct lu_device *d)
5248 rc = osd_shutdown(env, osd_dev(d));
5250 osd_obj_map_fini(osd_dev(d));
5252 shrink_dcache_sb(osd_sb(osd_dev(d)));
5253 osd_sync(env, lu2dt_dev(d));
5255 rc = osd_procfs_fini(osd_dev(d));
5257 CERROR("proc fini error %d \n", rc);
5258 RETURN (ERR_PTR(rc));
5261 if (osd_dev(d)->od_mnt) {
5262 mntput(osd_dev(d)->od_mnt);
5263 osd_dev(d)->od_mnt = NULL;
5269 static int osd_device_init0(const struct lu_env *env,
5270 struct osd_device *o,
5271 struct lustre_cfg *cfg)
5273 struct lu_device *l = osd2lu_dev(o);
5274 struct osd_thread_info *info;
5278 /* if the module was re-loaded, env can loose its keys */
5279 rc = lu_env_refill((struct lu_env *) env);
5282 info = osd_oti_get(env);
5285 l->ld_ops = &osd_lu_ops;
5286 o->od_dt_dev.dd_ops = &osd_dt_ops;
5288 spin_lock_init(&o->od_osfs_lock);
5289 mutex_init(&o->od_otable_mutex);
5290 o->od_osfs_age = cfs_time_shift_64(-1000);
5292 o->od_capa_hash = init_capa_hash();
5293 if (o->od_capa_hash == NULL)
5294 GOTO(out, rc = -ENOMEM);
5296 o->od_read_cache = 1;
5297 o->od_writethrough_cache = 1;
5298 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5300 rc = osd_mount(env, o, cfg);
5304 CFS_INIT_LIST_HEAD(&o->od_ios_list);
5305 /* setup scrub, including OI files initialization */
5306 rc = osd_scrub_setup(env, o);
5310 cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5311 sizeof(o->od_svname));
5312 if (cplen >= sizeof(o->od_svname)) {
5317 rc = osd_obj_map_init(env, o);
5319 GOTO(out_scrub, rc);
5321 rc = lu_site_init(&o->od_site, l);
5323 GOTO(out_compat, rc);
5324 o->od_site.ls_bottom_dev = l;
5326 rc = lu_site_init_finish(&o->od_site);
5330 rc = osd_procfs_init(o, o->od_svname);
5332 CERROR("%s: can't initialize procfs: rc = %d\n",
5337 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5339 /* initialize quota slave instance */
5340 o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5342 if (IS_ERR(o->od_quota_slave)) {
5343 rc = PTR_ERR(o->od_quota_slave);
5344 o->od_quota_slave = NULL;
5345 GOTO(out_procfs, rc);
5352 lu_site_fini(&o->od_site);
5354 osd_obj_map_fini(o);
5356 osd_scrub_cleanup(env, o);
5358 osd_oi_fini(info, o);
5359 osd_shutdown(env, o);
5363 cleanup_capa_hash(o->od_capa_hash);
5368 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5369 struct lu_device_type *t,
5370 struct lustre_cfg *cfg)
5372 struct osd_device *o;
5377 return ERR_PTR(-ENOMEM);
5379 rc = dt_device_init(&o->od_dt_dev, t);
5381 /* Because the ctx might be revived in dt_device_init,
5382 * refill the env here */
5383 lu_env_refill((struct lu_env *)env);
5384 rc = osd_device_init0(env, o, cfg);
5386 dt_device_fini(&o->od_dt_dev);
5389 if (unlikely(rc != 0))
5392 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5395 static struct lu_device *osd_device_free(const struct lu_env *env,
5396 struct lu_device *d)
5398 struct osd_device *o = osd_dev(d);
5401 cleanup_capa_hash(o->od_capa_hash);
5402 /* XXX: make osd top device in order to release reference */
5403 d->ld_site->ls_top_dev = d;
5404 lu_site_purge(env, d->ld_site, -1);
5405 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5406 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5407 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5409 lu_site_fini(&o->od_site);
5410 dt_device_fini(&o->od_dt_dev);
5415 static int osd_process_config(const struct lu_env *env,
5416 struct lu_device *d, struct lustre_cfg *cfg)
5418 struct osd_device *o = osd_dev(d);
5422 switch(cfg->lcfg_command) {
5424 err = osd_mount(env, o, cfg);
5427 lu_dev_del_linkage(d->ld_site, d);
5428 err = osd_shutdown(env, o);
5437 static int osd_recovery_complete(const struct lu_env *env,
5438 struct lu_device *d)
5440 struct osd_device *osd = osd_dev(d);
5444 if (osd->od_quota_slave == NULL)
5447 /* start qsd instance on recovery completion, this notifies the quota
5448 * slave code that we are about to process new requests now */
5449 rc = qsd_start(env, osd->od_quota_slave);
5454 * we use exports to track all osd users
5456 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5457 struct obd_device *obd, struct obd_uuid *cluuid,
5458 struct obd_connect_data *data, void *localdata)
5460 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5461 struct lustre_handle conn;
5465 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5467 rc = class_connect(&conn, obd, cluuid);
5471 *exp = class_conn2export(&conn);
5473 spin_lock(&osd->od_osfs_lock);
5475 spin_unlock(&osd->od_osfs_lock);
5481 * once last export (we don't count self-export) disappeared
5482 * osd can be released
5484 static int osd_obd_disconnect(struct obd_export *exp)
5486 struct obd_device *obd = exp->exp_obd;
5487 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5488 int rc, release = 0;
5491 /* Only disconnect the underlying layers on the final disconnect. */
5492 spin_lock(&osd->od_osfs_lock);
5494 if (osd->od_connects == 0)
5496 spin_unlock(&osd->od_osfs_lock);
5498 rc = class_disconnect(exp); /* bz 9811 */
5500 if (rc == 0 && release)
5501 class_manual_cleanup(obd);
5505 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5506 struct lu_device *dev)
5508 struct osd_device *osd = osd_dev(dev);
5512 if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
5513 /* MDT/MDD still use old infrastructure to create
5515 result = llo_local_objects_setup(env, lu2md_dev(pdev),
5521 if (osd->od_quota_slave != NULL)
5522 /* set up quota slave objects */
5523 result = qsd_prepare(env, osd->od_quota_slave);
5528 static const struct lu_object_operations osd_lu_obj_ops = {
5529 .loo_object_init = osd_object_init,
5530 .loo_object_delete = osd_object_delete,
5531 .loo_object_release = osd_object_release,
5532 .loo_object_free = osd_object_free,
5533 .loo_object_print = osd_object_print,
5534 .loo_object_invariant = osd_object_invariant
5537 const struct lu_device_operations osd_lu_ops = {
5538 .ldo_object_alloc = osd_object_alloc,
5539 .ldo_process_config = osd_process_config,
5540 .ldo_recovery_complete = osd_recovery_complete,
5541 .ldo_prepare = osd_prepare,
5544 static const struct lu_device_type_operations osd_device_type_ops = {
5545 .ldto_init = osd_type_init,
5546 .ldto_fini = osd_type_fini,
5548 .ldto_start = osd_type_start,
5549 .ldto_stop = osd_type_stop,
5551 .ldto_device_alloc = osd_device_alloc,
5552 .ldto_device_free = osd_device_free,
5554 .ldto_device_init = osd_device_init,
5555 .ldto_device_fini = osd_device_fini
5558 struct lu_device_type osd_device_type = {
5559 .ldt_tags = LU_DEVICE_DT,
5560 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
5561 .ldt_ops = &osd_device_type_ops,
5562 .ldt_ctx_tags = LCT_LOCAL,
5566 * lprocfs legacy support.
5568 static struct obd_ops osd_obd_device_ops = {
5569 .o_owner = THIS_MODULE,
5570 .o_connect = osd_obd_connect,
5571 .o_disconnect = osd_obd_disconnect
5574 static int __init osd_mod_init(void)
5576 struct lprocfs_static_vars lvars;
5579 lprocfs_osd_init_vars(&lvars);
5580 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
5581 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5584 static void __exit osd_mod_exit(void)
5586 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
5589 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5590 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
5591 MODULE_LICENSE("GPL");
5593 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);