1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
50 /* LUSTRE_VERSION_CODE */
51 #include <lustre_ver.h>
52 /* prerequisite for linux/xattr.h */
53 #include <linux/types.h>
54 /* prerequisite for linux/xattr.h */
56 /* XATTR_{REPLACE,CREATE} */
57 #include <linux/xattr.h>
59 * XXX temporary stuff: direct access to ldiskfs/jdb. Interface between osd
60 * and file system is not yet specified.
62 /* handle_t, journal_start(), journal_stop() */
63 #include <linux/jbd.h>
65 #include <linux/ldiskfs_fs.h>
66 #include <linux/ldiskfs_jbd.h>
71 * struct OBD_{ALLOC,FREE}*()
74 #include <obd_support.h>
75 /* struct ptlrpc_thread */
76 #include <lustre_net.h>
79 #include <lustre_fid.h>
80 #include <linux/lustre_iam.h>
82 #include "osd_internal.h"
85 /* llo_* api support */
86 #include <md_object.h>
88 static const char dot[] = ".";
89 static const char dotdot[] = "..";
90 static const char remote_obj_dir[] = "REM_OBJ_DIR";
92 struct osd_directory {
93 struct iam_container od_container;
94 struct iam_descr od_descr;
98 struct dt_object oo_dt;
100 * Inode for file system object represented by this osd_object. This
101 * inode is pinned for the whole duration of lu_object life.
103 * Not modified concurrently (either setup early during object
104 * creation, or assigned by osd_object_create() under write lock).
106 struct inode *oo_inode;
108 * to protect index ops.
110 struct rw_semaphore oo_ext_idx_sem;
111 struct rw_semaphore oo_sem;
112 struct osd_directory *oo_dir;
113 /** protects inode attributes. */
116 * Following two members are used to indicate the presence of dot and
117 * dotdot in the given directory. This is required for interop mode
120 int oo_compat_dot_created;
121 int oo_compat_dotdot_created;
123 const struct lu_env *oo_owner;
124 #ifdef CONFIG_LOCKDEP
125 struct lockdep_map oo_dep_map;
129 static const struct lu_object_operations osd_lu_obj_ops;
130 static const struct lu_device_operations osd_lu_ops;
131 static struct lu_context_key osd_key;
132 static const struct dt_object_operations osd_obj_ops;
133 static const struct dt_object_operations osd_obj_ea_ops;
134 static const struct dt_body_operations osd_body_ops;
135 static const struct dt_index_operations osd_index_iam_ops;
136 static const struct dt_index_operations osd_index_ea_ops;
139 struct thandle ot_super;
141 struct journal_callback ot_jcb;
142 /* Link to the device, for debugging. */
143 struct lu_ref_link *ot_dev_link;
150 static int lu_device_is_osd(const struct lu_device *d)
152 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
155 static struct osd_device *osd_dt_dev(const struct dt_device *d)
157 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
158 return container_of0(d, struct osd_device, od_dt_dev);
161 static struct osd_device *osd_dev(const struct lu_device *d)
163 LASSERT(lu_device_is_osd(d));
164 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
167 static struct osd_device *osd_obj2dev(const struct osd_object *o)
169 return osd_dev(o->oo_dt.do_lu.lo_dev);
172 static struct super_block *osd_sb(const struct osd_device *dev)
174 return dev->od_mount->lmi_mnt->mnt_sb;
177 static int osd_object_is_root(const struct osd_object *obj)
179 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
182 static struct osd_object *osd_obj(const struct lu_object *o)
184 LASSERT(lu_device_is_osd(o->lo_dev));
185 return container_of0(o, struct osd_object, oo_dt.do_lu);
188 static struct osd_object *osd_dt_obj(const struct dt_object *d)
190 return osd_obj(&d->do_lu);
193 static struct lu_device *osd2lu_dev(struct osd_device *osd)
195 return &osd->od_dt_dev.dd_lu_dev;
198 static journal_t *osd_journal(const struct osd_device *dev)
200 return LDISKFS_SB(osd_sb(dev))->s_journal;
203 static int osd_has_index(const struct osd_object *obj)
205 return obj->oo_dt.do_index_ops != NULL;
208 static int osd_object_invariant(const struct lu_object *l)
210 return osd_invariant(osd_obj(l));
213 #ifdef HAVE_QUOTA_SUPPORT
215 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
217 struct md_ucred *uc = md_ucred(env);
221 save->oc_uid = current->fsuid;
222 save->oc_gid = current->fsgid;
223 save->oc_cap = current->cap_effective;
224 current->fsuid = uc->mu_fsuid;
225 current->fsgid = uc->mu_fsgid;
226 current->cap_effective = uc->mu_cap;
230 osd_pop_ctxt(struct osd_ctxt *save)
232 current->fsuid = save->oc_uid;
233 current->fsgid = save->oc_gid;
234 current->cap_effective = save->oc_cap;
238 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
240 return lu_context_key_get(&env->le_ctx, &osd_key);
244 * Concurrency: doesn't matter
246 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
248 return osd_oti_get(env)->oti_r_locks > 0;
252 * Concurrency: doesn't matter
254 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
256 struct osd_thread_info *oti = osd_oti_get(env);
257 return oti->oti_w_locks > 0 && o->oo_owner == env;
261 * Concurrency: doesn't access mutable data
263 static int osd_root_get(const struct lu_env *env,
264 struct dt_device *dev, struct lu_fid *f)
268 inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
269 lu_igif_build(f, inode->i_ino, inode->i_generation);
274 * OSD object methods.
278 * Concurrency: no concurrent access is possible that early in object
281 static struct lu_object *osd_object_alloc(const struct lu_env *env,
282 const struct lu_object_header *hdr,
285 struct osd_object *mo;
291 l = &mo->oo_dt.do_lu;
292 dt_object_init(&mo->oo_dt, NULL, d);
293 if (osd_dev(d)->od_iop_mode)
294 mo->oo_dt.do_ops = &osd_obj_ea_ops;
296 mo->oo_dt.do_ops = &osd_obj_ops;
298 l->lo_ops = &osd_lu_obj_ops;
299 init_rwsem(&mo->oo_sem);
300 init_rwsem(&mo->oo_ext_idx_sem);
301 spin_lock_init(&mo->oo_guard);
308 * retrieve object from backend ext fs.
310 static struct inode *osd_iget(struct osd_thread_info *info,
311 struct osd_device *dev,
312 const struct osd_inode_id *id)
316 inode = iget(osd_sb(dev), id->oii_ino);
318 CERROR("no inode\n");
319 inode = ERR_PTR(-EACCES);
320 } else if (id->oii_gen != OSD_OII_NOGEN &&
321 inode->i_generation != id->oii_gen) {
323 inode = ERR_PTR(-ESTALE);
324 } else if (inode->i_nlink == 0) {
325 /* due to parallel readdir and unlink,
326 * we can have dead inode here. */
327 CWARN("stale inode\n");
328 make_bad_inode(inode);
330 inode = ERR_PTR(-ESTALE);
331 } else if (is_bad_inode(inode)) {
332 CERROR("bad inode %lx\n",inode->i_ino);
334 inode = ERR_PTR(-ENOENT);
339 static int osd_fid_lookup(const struct lu_env *env,
340 struct osd_object *obj, const struct lu_fid *fid)
342 struct osd_thread_info *info;
343 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
344 struct osd_device *dev;
345 struct osd_inode_id *id;
350 LINVRNT(osd_invariant(obj));
351 LASSERT(obj->oo_inode == NULL);
352 LASSERT(fid_is_sane(fid));
354 * This assertion checks that osd layer sees only local
355 * fids. Unfortunately it is somewhat expensive (does a
356 * cache-lookup). Disabling it for production/acceptance-testing.
358 LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
362 info = osd_oti_get(env);
367 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
370 result = osd_oi_lookup(info, oi, fid, id);
372 inode = osd_iget(info, dev, id);
373 if (!IS_ERR(inode)) {
374 obj->oo_inode = inode;
375 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
376 if (dev->od_iop_mode) {
377 obj->oo_compat_dot_created = 1;
378 obj->oo_compat_dotdot_created = 1;
383 * If fid wasn't found in oi, inode-less object is
384 * created, for which lu_object_exists() returns
385 * false. This is used in a (frequent) case when
386 * objects are created as locking anchors or
387 * place holders for objects yet to be created.
389 result = PTR_ERR(inode);
390 } else if (result == -ENOENT)
392 LINVRNT(osd_invariant(obj));
398 * Concurrency: shouldn't matter.
400 static void osd_object_init0(struct osd_object *obj)
402 LASSERT(obj->oo_inode != NULL);
403 obj->oo_dt.do_body_ops = &osd_body_ops;
404 obj->oo_dt.do_lu.lo_header->loh_attr |=
405 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
409 * Concurrency: no concurrent access is possible that early in object
412 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
413 const struct lu_object_conf *unused)
415 struct osd_object *obj = osd_obj(l);
418 LINVRNT(osd_invariant(obj));
420 result = osd_fid_lookup(env, obj, lu_object_fid(l));
422 if (obj->oo_inode != NULL)
423 osd_object_init0(obj);
425 LINVRNT(osd_invariant(obj));
430 * Concurrency: no concurrent access is possible that late in object
433 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
435 struct osd_object *obj = osd_obj(l);
437 LINVRNT(osd_invariant(obj));
439 dt_object_fini(&obj->oo_dt);
446 static struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
447 const struct iam_container *bag)
449 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
450 osd_oti_get(env)->oti_it_ipd);
453 static struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
454 const struct iam_container *bag)
456 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
457 osd_oti_get(env)->oti_idx_ipd);
460 static void osd_ipd_put(const struct lu_env *env,
461 const struct iam_container *bag,
462 struct iam_path_descr *ipd)
464 bag->ic_descr->id_ops->id_ipd_free(ipd);
468 * Concurrency: no concurrent access is possible that late in object
471 static void osd_index_fini(struct osd_object *o)
473 struct iam_container *bag;
475 if (o->oo_dir != NULL) {
476 bag = &o->oo_dir->od_container;
477 if (o->oo_inode != NULL) {
478 if (bag->ic_object == o->oo_inode)
479 iam_container_fini(bag);
481 OBD_FREE_PTR(o->oo_dir);
487 * Concurrency: no concurrent access is possible that late in object
488 * life-cycle (for all existing callers, that is. New callers have to provide
489 * their own locking.)
491 static int osd_inode_unlinked(const struct inode *inode)
493 return inode->i_nlink == 0;
497 OSD_TXN_OI_DELETE_CREDITS = 20,
498 OSD_TXN_INODE_DELETE_CREDITS = 20
506 * Concurrency: doesn't access mutable data.
508 static int osd_param_is_sane(const struct osd_device *dev,
509 const struct txn_param *param)
511 return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
515 * Concurrency: shouldn't matter.
517 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
519 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
520 struct thandle *th = &oh->ot_super;
521 struct dt_device *dev = th->th_dev;
522 struct lu_device *lud = &dev->dd_lu_dev;
524 LASSERT(dev != NULL);
525 LASSERT(oh->ot_handle == NULL);
528 CERROR("transaction @0x%p commit error: %d\n", th, error);
530 struct lu_env *env = &osd_dt_dev(dev)->od_env_for_commit;
532 * This od_env_for_commit is only for commit usage. see
535 lu_context_enter(&env->le_ctx);
536 dt_txn_hook_commit(env, th);
537 lu_context_exit(&env->le_ctx);
540 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
544 lu_context_exit(&th->th_ctx);
545 lu_context_fini(&th->th_ctx);
550 * Concurrency: shouldn't matter.
552 static struct thandle *osd_trans_start(const struct lu_env *env,
556 struct osd_device *dev = osd_dt_dev(d);
558 struct osd_thandle *oh;
564 hook_res = dt_txn_hook_start(env, d, p);
566 RETURN(ERR_PTR(hook_res));
568 if (osd_param_is_sane(dev, p)) {
569 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
571 struct osd_thread_info *oti = osd_oti_get(env);
574 * XXX temporary stuff. Some abstraction layer should
578 jh = journal_start(osd_journal(dev), p->tp_credits);
584 jh->h_sync = p->tp_sync;
585 lu_device_get(&d->dd_lu_dev);
586 oh->ot_dev_link = lu_ref_add
587 (&d->dd_lu_dev.ld_reference,
589 /* add commit callback */
590 lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
591 lu_context_enter(&th->th_ctx);
592 journal_callback_set(jh, osd_trans_commit_cb,
593 (struct journal_callback *)&oh->ot_jcb);
594 LASSERT(oti->oti_txns == 0);
595 LASSERT(oti->oti_r_locks == 0);
596 LASSERT(oti->oti_w_locks == 0);
603 th = ERR_PTR(-ENOMEM);
605 CERROR("Invalid transaction parameters\n");
606 th = ERR_PTR(-EINVAL);
613 * Concurrency: shouldn't matter.
615 static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
618 struct osd_thandle *oh;
619 struct osd_thread_info *oti = osd_oti_get(env);
623 oh = container_of0(th, struct osd_thandle, ot_super);
624 if (oh->ot_handle != NULL) {
625 handle_t *hdl = oh->ot_handle;
627 LASSERT(oti->oti_txns == 1);
629 LASSERT(oti->oti_r_locks == 0);
630 LASSERT(oti->oti_w_locks == 0);
631 result = dt_txn_hook_stop(env, th);
633 CERROR("Failure in transaction hook: %d\n", result);
634 oh->ot_handle = NULL;
635 result = journal_stop(hdl);
637 CERROR("Failure to stop transaction: %d\n", result);
643 * Concurrency: no concurrent access is possible that late in object
646 static int osd_inode_remove(const struct lu_env *env, struct osd_object *obj)
648 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
649 struct osd_device *osd = osd_obj2dev(obj);
650 struct osd_thread_info *oti = osd_oti_get(env);
651 struct txn_param *prm = &oti->oti_txn;
652 struct lu_env *env_del_obj = &oti->oti_obj_delete_tx_env;
656 lu_env_init(env_del_obj, LCT_DT_THREAD);
657 txn_param_init(prm, OSD_TXN_OI_DELETE_CREDITS +
658 OSD_TXN_INODE_DELETE_CREDITS);
659 th = osd_trans_start(env_del_obj, &osd->od_dt_dev, prm);
661 result = osd_oi_delete(osd_oti_get(env_del_obj),
662 &osd->od_oi, fid, th);
663 osd_trans_stop(env_del_obj, th);
665 result = PTR_ERR(th);
667 lu_env_fini(env_del_obj);
672 * Called just before object is freed. Releases all resources except for
673 * object itself (that is released by osd_object_free()).
675 * Concurrency: no concurrent access is possible that late in object
678 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
680 struct osd_object *obj = osd_obj(l);
681 struct inode *inode = obj->oo_inode;
683 LINVRNT(osd_invariant(obj));
686 * If object is unlinked remove fid->ino mapping from object index.
693 if (osd_inode_unlinked(inode)) {
694 result = osd_inode_remove(env, obj);
696 LU_OBJECT_DEBUG(D_ERROR, env, l,
697 "Failed to cleanup: %d\n",
702 obj->oo_inode = NULL;
707 * Concurrency: ->loo_object_release() is called under site spin-lock.
709 static void osd_object_release(const struct lu_env *env,
712 struct osd_object *o = osd_obj(l);
714 LASSERT(!lu_object_is_dying(l->lo_header));
715 if (o->oo_inode != NULL && osd_inode_unlinked(o->oo_inode))
716 set_bit(LU_OBJECT_HEARD_BANSHEE, &l->lo_header->loh_flags);
720 * Concurrency: shouldn't matter.
722 static int osd_object_print(const struct lu_env *env, void *cookie,
723 lu_printer_t p, const struct lu_object *l)
725 struct osd_object *o = osd_obj(l);
728 if (o->oo_dir != NULL)
729 d = o->oo_dir->od_container.ic_descr;
732 return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
734 o->oo_inode ? o->oo_inode->i_ino : 0UL,
735 o->oo_inode ? o->oo_inode->i_generation : 0,
736 d ? d->id_ops->id_name : "plain");
740 * Concurrency: shouldn't matter.
742 int osd_statfs(const struct lu_env *env, struct dt_device *d,
745 struct osd_device *osd = osd_dt_dev(d);
746 struct super_block *sb = osd_sb(osd);
749 spin_lock(&osd->od_osfs_lock);
751 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
752 result = ll_do_statfs(sb, &osd->od_kstatfs);
753 if (likely(result == 0)) /* N.B. statfs can't really fail */
754 osd->od_osfs_age = cfs_time_current_64();
757 if (likely(result == 0))
758 *sfs = osd->od_kstatfs;
759 spin_unlock(&osd->od_osfs_lock);
765 * Concurrency: doesn't access mutable data.
767 static void osd_conf_get(const struct lu_env *env,
768 const struct dt_device *dev,
769 struct dt_device_param *param)
772 * XXX should be taken from not-yet-existing fs abstraction layer.
774 param->ddp_max_name_len = LDISKFS_NAME_LEN;
775 param->ddp_max_nlink = LDISKFS_LINK_MAX;
776 param->ddp_block_shift = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
780 * Helper function to get and fill the buffer with input values.
782 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
786 buf = &osd_oti_get(env)->oti_buf;
793 * Concurrency: shouldn't matter.
795 static int osd_sync(const struct lu_env *env, struct dt_device *d)
797 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
798 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
802 * Start commit for OSD device.
804 * An implementation of dt_commit_async method for OSD device.
805 * Asychronously starts underlayng fs sync and thereby a transaction
808 * \param env environment
811 * \see dt_device_operations
813 static int osd_commit_async(const struct lu_env *env,
816 struct super_block *s = osd_sb(osd_dt_dev(d));
819 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
820 RETURN(s->s_op->sync_fs(s, 0));
824 * Concurrency: shouldn't matter.
826 lvfs_sbdev_type fsfilt_ldiskfs_journal_sbdev(struct super_block *);
828 static void osd_ro(const struct lu_env *env, struct dt_device *d)
832 CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
834 __lvfs_set_rdonly(lvfs_sbdev(osd_sb(osd_dt_dev(d))),
835 fsfilt_ldiskfs_journal_sbdev(osd_sb(osd_dt_dev(d))));
841 * Concurrency: serialization provided by callers.
843 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
844 int mode, unsigned long timeout, __u32 alg,
845 struct lustre_capa_key *keys)
847 struct osd_device *dev = osd_dt_dev(d);
850 dev->od_fl_capa = mode;
851 dev->od_capa_timeout = timeout;
852 dev->od_capa_alg = alg;
853 dev->od_capa_keys = keys;
858 * Concurrency: serialization provided by callers.
860 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
861 struct dt_quota_ctxt *ctxt, void *data)
863 struct obd_device *obd = (void *)ctxt;
864 struct vfsmount *mnt = (struct vfsmount *)data;
867 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
868 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
869 obd->obd_lvfs_ctxt.pwdmnt = mnt;
870 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
871 obd->obd_lvfs_ctxt.fs = get_ds();
877 * Note: we do not count into QUOTA here.
878 * If we mount with --data_journal we may need more.
880 static const int osd_dto_credits_noquota[DTO_NR] = {
883 * INDEX_EXTRA_TRANS_BLOCKS(8) +
884 * SINGLEDATA_TRANS_BLOCKS(8)
885 * XXX Note: maybe iam need more, since iam have more level than
888 [DTO_INDEX_INSERT] = 16,
889 [DTO_INDEX_DELETE] = 16,
893 [DTO_IDNEX_UPDATE] = 16,
895 * Create a object. The same as create object in EXT3.
896 * DATA_TRANS_BLOCKS(14) +
897 * INDEX_EXTRA_BLOCKS(8) +
898 * 3(inode bits, groups, GDT)
900 [DTO_OBJECT_CREATE] = 25,
904 [DTO_OBJECT_DELETE] = 25,
907 * 3(inode bits, group, GDT)
909 [DTO_ATTR_SET_BASE] = 3,
911 * Xattr set. The same as xattr of EXT3.
912 * DATA_TRANS_BLOCKS(14)
913 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS are
914 * also counted in. Do not know why?
916 [DTO_XATTR_SET] = 14,
919 * creadits for inode change during write.
921 [DTO_WRITE_BASE] = 3,
923 * credits for single block write.
925 [DTO_WRITE_BLOCK] = 14,
927 * Attr set credits for chown.
928 * 3 (inode bit, group, GDT)
930 [DTO_ATTR_SET_CHOWN]= 3
934 * Note: we count into QUOTA here.
935 * If we mount with --data_journal we may need more.
937 static const int osd_dto_credits_quota[DTO_NR] = {
939 * INDEX_EXTRA_TRANS_BLOCKS(8) +
940 * SINGLEDATA_TRANS_BLOCKS(8) +
941 * 2 * QUOTA_TRANS_BLOCKS(2)
943 [DTO_INDEX_INSERT] = 20,
945 * INDEX_EXTRA_TRANS_BLOCKS(8) +
946 * SINGLEDATA_TRANS_BLOCKS(8) +
947 * 2 * QUOTA_TRANS_BLOCKS(2)
949 [DTO_INDEX_DELETE] = 20,
953 [DTO_IDNEX_UPDATE] = 16,
955 * Create a object. Same as create object in EXT3 filesystem.
956 * DATA_TRANS_BLOCKS(16) +
957 * INDEX_EXTRA_BLOCKS(8) +
958 * 3(inode bits, groups, GDT) +
959 * 2 * QUOTA_INIT_BLOCKS(25)
961 [DTO_OBJECT_CREATE] = 77,
964 * DATA_TRANS_BLOCKS(16) +
965 * INDEX_EXTRA_BLOCKS(8) +
966 * 3(inode bits, groups, GDT) +
969 [DTO_OBJECT_DELETE] = 27,
972 * 3 (inode bit, group, GDT) +
974 [DTO_ATTR_SET_BASE] = 3,
976 * Xattr set. The same as xattr of EXT3.
977 * DATA_TRANS_BLOCKS(16)
978 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS are
979 * also counted in. Do not know why?
981 [DTO_XATTR_SET] = 16,
984 * creadits for inode change during write.
986 [DTO_WRITE_BASE] = 3,
988 * credits for single block write.
990 [DTO_WRITE_BLOCK] = 16,
992 * Attr set credits for chown.
993 * 3 (inode bit, group, GDT) +
994 * 2 * QUOTA_INIT_BLOCKS(25) +
995 * 2 * QUOTA_DEL_BLOCKS(9)
997 [DTO_ATTR_SET_CHOWN]= 71
1000 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
1003 LASSERT(ARRAY_SIZE(osd_dto_credits_noquota) ==
1004 ARRAY_SIZE(osd_dto_credits_quota));
1005 LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits_noquota));
1006 #ifdef HAVE_QUOTA_SUPPORT
1007 if (test_opt(osd_sb(osd_dt_dev(d)), QUOTA))
1008 return osd_dto_credits_quota[op];
1011 return osd_dto_credits_noquota[op];
1014 static const struct dt_device_operations osd_dt_ops = {
1015 .dt_root_get = osd_root_get,
1016 .dt_statfs = osd_statfs,
1017 .dt_trans_start = osd_trans_start,
1018 .dt_trans_stop = osd_trans_stop,
1019 .dt_conf_get = osd_conf_get,
1020 .dt_sync = osd_sync,
1022 .dt_commit_async = osd_commit_async,
1023 .dt_credit_get = osd_credit_get,
1024 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1025 .dt_init_quota_ctxt= osd_init_quota_ctxt,
1028 static void osd_object_read_lock(const struct lu_env *env,
1029 struct dt_object *dt, unsigned role)
1031 struct osd_object *obj = osd_dt_obj(dt);
1032 struct osd_thread_info *oti = osd_oti_get(env);
1034 LINVRNT(osd_invariant(obj));
1036 LASSERT(obj->oo_owner != env);
1037 down_read_nested(&obj->oo_sem, role);
1039 LASSERT(obj->oo_owner == NULL);
1043 static void osd_object_write_lock(const struct lu_env *env,
1044 struct dt_object *dt, unsigned role)
1046 struct osd_object *obj = osd_dt_obj(dt);
1047 struct osd_thread_info *oti = osd_oti_get(env);
1049 LINVRNT(osd_invariant(obj));
1051 LASSERT(obj->oo_owner != env);
1052 down_write_nested(&obj->oo_sem, role);
1054 LASSERT(obj->oo_owner == NULL);
1055 obj->oo_owner = env;
1059 static void osd_object_read_unlock(const struct lu_env *env,
1060 struct dt_object *dt)
1062 struct osd_object *obj = osd_dt_obj(dt);
1063 struct osd_thread_info *oti = osd_oti_get(env);
1065 LINVRNT(osd_invariant(obj));
1067 LASSERT(oti->oti_r_locks > 0);
1069 up_read(&obj->oo_sem);
1072 static void osd_object_write_unlock(const struct lu_env *env,
1073 struct dt_object *dt)
1075 struct osd_object *obj = osd_dt_obj(dt);
1076 struct osd_thread_info *oti = osd_oti_get(env);
1078 LINVRNT(osd_invariant(obj));
1080 LASSERT(obj->oo_owner == env);
1081 LASSERT(oti->oti_w_locks > 0);
1083 obj->oo_owner = NULL;
1084 up_write(&obj->oo_sem);
1087 static int osd_object_write_locked(const struct lu_env *env,
1088 struct dt_object *dt)
1090 struct osd_object *obj = osd_dt_obj(dt);
1092 LINVRNT(osd_invariant(obj));
1094 return obj->oo_owner == env;
1097 static int capa_is_sane(const struct lu_env *env,
1098 struct osd_device *dev,
1099 struct lustre_capa *capa,
1100 struct lustre_capa_key *keys)
1102 struct osd_thread_info *oti = osd_oti_get(env);
1103 struct lustre_capa *tcapa = &oti->oti_capa;
1104 struct obd_capa *oc;
1108 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1110 if (capa_is_expired(oc)) {
1111 DEBUG_CAPA(D_ERROR, capa, "expired");
1118 if (capa_is_expired_sec(capa)) {
1119 DEBUG_CAPA(D_ERROR, capa, "expired");
1123 spin_lock(&capa_lock);
1124 for (i = 0; i < 2; i++) {
1125 if (keys[i].lk_keyid == capa->lc_keyid) {
1126 oti->oti_capa_key = keys[i];
1130 spin_unlock(&capa_lock);
1133 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1137 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1141 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1142 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1146 oc = capa_add(dev->od_capa_hash, capa);
1152 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1153 struct lustre_capa *capa, __u64 opc)
1155 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1156 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1157 struct md_capainfo *ci;
1160 if (!dev->od_fl_capa)
1163 if (capa == BYPASS_CAPA)
1166 ci = md_capainfo(env);
1170 if (ci->mc_auth == LC_ID_NONE)
1174 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1178 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1179 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1184 if (!capa_opc_supported(capa, opc)) {
1185 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1189 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1190 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1197 static struct timespec *osd_inode_time(const struct lu_env *env,
1198 struct inode *inode, __u64 seconds)
1200 struct osd_thread_info *oti = osd_oti_get(env);
1201 struct timespec *t = &oti->oti_time;
1203 t->tv_sec = seconds;
1205 *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1210 static void osd_inode_getattr(const struct lu_env *env,
1211 struct inode *inode, struct lu_attr *attr)
1213 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1214 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1215 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1217 attr->la_atime = LTIME_S(inode->i_atime);
1218 attr->la_mtime = LTIME_S(inode->i_mtime);
1219 attr->la_ctime = LTIME_S(inode->i_ctime);
1220 attr->la_mode = inode->i_mode;
1221 attr->la_size = i_size_read(inode);
1222 attr->la_blocks = inode->i_blocks;
1223 attr->la_uid = inode->i_uid;
1224 attr->la_gid = inode->i_gid;
1225 attr->la_flags = LDISKFS_I(inode)->i_flags;
1226 attr->la_nlink = inode->i_nlink;
1227 attr->la_rdev = inode->i_rdev;
1228 attr->la_blksize = ll_inode_blksize(inode);
1229 attr->la_blkbits = inode->i_blkbits;
1232 static int osd_attr_get(const struct lu_env *env,
1233 struct dt_object *dt,
1234 struct lu_attr *attr,
1235 struct lustre_capa *capa)
1237 struct osd_object *obj = osd_dt_obj(dt);
1239 LASSERT(dt_object_exists(dt));
1240 LINVRNT(osd_invariant(obj));
1242 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1245 spin_lock(&obj->oo_guard);
1246 osd_inode_getattr(env, obj->oo_inode, attr);
1247 spin_unlock(&obj->oo_guard);
1251 static int osd_inode_setattr(const struct lu_env *env,
1252 struct inode *inode, const struct lu_attr *attr)
1256 bits = attr->la_valid;
1258 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1260 #ifdef HAVE_QUOTA_SUPPORT
1261 if ((bits & LA_UID && attr->la_uid != inode->i_uid) ||
1262 (bits & LA_GID && attr->la_gid != inode->i_gid)) {
1263 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1269 iattr.ia_valid |= ATTR_UID;
1271 iattr.ia_valid |= ATTR_GID;
1272 iattr.ia_uid = attr->la_uid;
1273 iattr.ia_gid = attr->la_gid;
1274 osd_push_ctxt(env, save);
1275 rc = DQUOT_TRANSFER(inode, &iattr) ? -EDQUOT : 0;
1282 if (bits & LA_ATIME)
1283 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1284 if (bits & LA_CTIME)
1285 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1286 if (bits & LA_MTIME)
1287 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1288 if (bits & LA_SIZE) {
1289 LDISKFS_I(inode)->i_disksize = attr->la_size;
1290 i_size_write(inode, attr->la_size);
1293 /* OSD should not change "i_blocks" which is used by quota.
1294 * "i_blocks" should be changed by ldiskfs only.
1295 * Enable this assignment for SOM purpose now, until it is
1296 * stored in SOM EA. */
1297 if (bits & LA_BLOCKS)
1298 inode->i_blocks = attr->la_blocks;
1301 inode->i_mode = (inode->i_mode & S_IFMT) |
1302 (attr->la_mode & ~S_IFMT);
1304 inode->i_uid = attr->la_uid;
1306 inode->i_gid = attr->la_gid;
1307 if (bits & LA_NLINK)
1308 inode->i_nlink = attr->la_nlink;
1310 inode->i_rdev = attr->la_rdev;
1312 if (bits & LA_FLAGS) {
1313 struct ldiskfs_inode_info *li = LDISKFS_I(inode);
1315 li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
1316 (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
1321 static int osd_attr_set(const struct lu_env *env,
1322 struct dt_object *dt,
1323 const struct lu_attr *attr,
1324 struct thandle *handle,
1325 struct lustre_capa *capa)
1327 struct osd_object *obj = osd_dt_obj(dt);
1330 LASSERT(handle != NULL);
1331 LASSERT(dt_object_exists(dt));
1332 LASSERT(osd_invariant(obj));
1334 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1337 spin_lock(&obj->oo_guard);
1338 rc = osd_inode_setattr(env, obj->oo_inode, attr);
1339 spin_unlock(&obj->oo_guard);
1342 mark_inode_dirty(obj->oo_inode);
1349 * XXX temporary solution.
1351 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1352 struct lu_attr *attr, struct thandle *th)
1357 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1358 struct lu_attr *attr, struct thandle *th)
1360 osd_object_init0(obj);
1364 extern struct inode *ldiskfs_create_inode(handle_t *handle,
1365 struct inode * dir, int mode);
1366 extern int ldiskfs_add_entry(handle_t *handle, struct dentry *dentry,
1367 struct inode *inode);
1368 extern int ldiskfs_delete_entry(handle_t *handle,
1370 struct ldiskfs_dir_entry_2 * de_del,
1371 struct buffer_head * bh);
1372 extern struct buffer_head * ldiskfs_find_entry(struct dentry *dentry,
1373 struct ldiskfs_dir_entry_2
1375 extern int ldiskfs_add_dot_dotdot(handle_t *handle, struct inode *dir,
1376 struct inode *inode);
1378 extern int ldiskfs_xattr_set_handle(handle_t *handle, struct inode *inode,
1379 int name_index, const char *name,
1380 const void *value, size_t value_len,
1383 static struct dentry * osd_child_dentry_get(const struct lu_env *env,
1384 struct osd_object *obj,
1388 struct osd_thread_info *info = osd_oti_get(env);
1389 struct dentry *child_dentry = &info->oti_child_dentry;
1390 struct dentry *obj_dentry = &info->oti_obj_dentry;
1392 obj_dentry->d_inode = obj->oo_inode;
1393 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
1394 obj_dentry->d_name.hash = 0;
1396 child_dentry->d_name.hash = 0;
1397 child_dentry->d_parent = obj_dentry;
1398 child_dentry->d_name.name = name;
1399 child_dentry->d_name.len = namelen;
1400 return child_dentry;
1404 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1406 struct dt_allocation_hint *hint,
1410 struct osd_device *osd = osd_obj2dev(obj);
1411 struct osd_thandle *oth;
1412 struct dt_object *parent;
1413 struct inode *inode;
1414 #ifdef HAVE_QUOTA_SUPPORT
1415 struct osd_ctxt *save = &info->oti_ctxt;
1418 LINVRNT(osd_invariant(obj));
1419 LASSERT(obj->oo_inode == NULL);
1421 oth = container_of(th, struct osd_thandle, ot_super);
1422 LASSERT(oth->ot_handle->h_transaction != NULL);
1424 if (hint && hint->dah_parent)
1425 parent = hint->dah_parent;
1427 parent = osd->od_obj_area;
1429 LASSERT(parent != NULL);
1430 LASSERT(osd_dt_obj(parent)->oo_inode->i_op != NULL);
1432 #ifdef HAVE_QUOTA_SUPPORT
1433 osd_push_ctxt(info->oti_env, save);
1435 inode = ldiskfs_create_inode(oth->ot_handle,
1436 osd_dt_obj(parent)->oo_inode, mode);
1437 #ifdef HAVE_QUOTA_SUPPORT
1440 if (!IS_ERR(inode)) {
1441 obj->oo_inode = inode;
1444 result = PTR_ERR(inode);
1445 LINVRNT(osd_invariant(obj));
1450 extern int iam_lvar_create(struct inode *obj, int keysize, int ptrsize,
1451 int recsize, handle_t *handle);
1453 extern int iam_lfix_create(struct inode *obj, int keysize, int ptrsize,
1454 int recsize, handle_t *handle);
1460 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1461 struct lu_attr *attr,
1462 struct dt_allocation_hint *hint,
1463 struct dt_object_format *dof,
1467 struct osd_thandle *oth;
1468 struct osd_device *osd = osd_obj2dev(obj);
1469 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1471 LASSERT(S_ISDIR(attr->la_mode));
1473 oth = container_of(th, struct osd_thandle, ot_super);
1474 LASSERT(oth->ot_handle->h_transaction != NULL);
1475 result = osd_mkfile(info, obj, mode, hint, th);
1476 if (result == 0 && osd->od_iop_mode == 0) {
1477 LASSERT(obj->oo_inode != NULL);
1479 * XXX uh-oh... call low-level iam function directly.
1482 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1483 sizeof (struct lu_fid_pack),
1489 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1490 struct lu_attr *attr,
1491 struct dt_allocation_hint *hint,
1492 struct dt_object_format *dof,
1496 struct osd_thandle *oth;
1497 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1499 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1501 LASSERT(S_ISREG(attr->la_mode));
1503 oth = container_of(th, struct osd_thandle, ot_super);
1504 LASSERT(oth->ot_handle->h_transaction != NULL);
1506 result = osd_mkfile(info, obj, mode, hint, th);
1508 LASSERT(obj->oo_inode != NULL);
1509 if (feat->dif_flags & DT_IND_VARKEY)
1510 result = iam_lvar_create(obj->oo_inode,
1511 feat->dif_keysize_max,
1513 feat->dif_recsize_max,
1516 result = iam_lfix_create(obj->oo_inode,
1517 feat->dif_keysize_max,
1519 feat->dif_recsize_max,
1526 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1527 struct lu_attr *attr,
1528 struct dt_allocation_hint *hint,
1529 struct dt_object_format *dof,
1532 LASSERT(S_ISREG(attr->la_mode));
1533 return osd_mkfile(info, obj, (attr->la_mode &
1534 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1537 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1538 struct lu_attr *attr,
1539 struct dt_allocation_hint *hint,
1540 struct dt_object_format *dof,
1543 LASSERT(S_ISLNK(attr->la_mode));
1544 return osd_mkfile(info, obj, (attr->la_mode &
1545 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1548 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1549 struct lu_attr *attr,
1550 struct dt_allocation_hint *hint,
1551 struct dt_object_format *dof,
1554 umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1557 LINVRNT(osd_invariant(obj));
1558 LASSERT(obj->oo_inode == NULL);
1559 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1560 S_ISFIFO(mode) || S_ISSOCK(mode));
1562 result = osd_mkfile(info, obj, mode, hint, th);
1564 LASSERT(obj->oo_inode != NULL);
1565 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1567 LINVRNT(osd_invariant(obj));
1571 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1573 struct dt_allocation_hint *hint,
1574 struct dt_object_format *dof,
1577 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1579 osd_obj_type_f result;
1595 result = osd_mk_index;
1606 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1607 struct dt_object *parent, umode_t child_mode)
1611 memset(ah, 0, sizeof(*ah));
1612 ah->dah_parent = parent;
1613 ah->dah_mode = child_mode;
1617 * Helper function for osd_object_create()
1619 * \retval 0, on success
1621 static int __osd_object_create(struct osd_thread_info *info,
1622 struct osd_object *obj, struct lu_attr *attr,
1623 struct dt_allocation_hint *hint,
1624 struct dt_object_format *dof,
1630 result = osd_create_pre(info, obj, attr, th);
1632 result = osd_create_type_f(dof->dof_type)(info, obj,
1633 attr, hint, dof, th);
1635 result = osd_create_post(info, obj, attr, th);
1641 * Helper function for osd_object_create()
1643 * \retval 0, on success
1645 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1646 const struct lu_fid *fid, struct thandle *th)
1648 struct osd_thread_info *info = osd_oti_get(env);
1649 struct osd_inode_id *id = &info->oti_id;
1650 struct osd_device *osd = osd_obj2dev(obj);
1651 struct md_ucred *uc = md_ucred(env);
1653 LASSERT(obj->oo_inode != NULL);
1654 LASSERT(uc != NULL);
1656 id->oii_ino = obj->oo_inode->i_ino;
1657 id->oii_gen = obj->oo_inode->i_generation;
1659 return osd_oi_insert(info, &osd->od_oi, fid, id, th,
1660 uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
1663 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1664 struct lu_attr *attr,
1665 struct dt_allocation_hint *hint,
1666 struct dt_object_format *dof,
1669 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1670 struct osd_object *obj = osd_dt_obj(dt);
1671 struct osd_thread_info *info = osd_oti_get(env);
1676 LINVRNT(osd_invariant(obj));
1677 LASSERT(!dt_object_exists(dt));
1678 LASSERT(osd_write_locked(env, obj));
1679 LASSERT(th != NULL);
1681 result = __osd_object_create(info, obj, attr, hint, dof, th);
1683 result = __osd_oi_insert(env, obj, fid, th);
1685 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1686 LASSERT(osd_invariant(obj));
1691 * Helper function for osd_xattr_set()
1693 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1694 const struct lu_buf *buf, const char *name, int fl)
1696 struct osd_object *obj = osd_dt_obj(dt);
1697 struct inode *inode = obj->oo_inode;
1698 struct osd_thread_info *info = osd_oti_get(env);
1699 struct dentry *dentry = &info->oti_child_dentry;
1700 struct timespec *t = &info->oti_time;
1704 LASSERT(dt_object_exists(dt));
1705 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1706 LASSERT(osd_write_locked(env, obj));
1708 if (fl & LU_XATTR_REPLACE)
1709 fs_flags |= XATTR_REPLACE;
1711 if (fl & LU_XATTR_CREATE)
1712 fs_flags |= XATTR_CREATE;
1714 dentry->d_inode = inode;
1715 *t = inode->i_ctime;
1716 rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
1717 buf->lb_len, fs_flags);
1718 /* ctime should not be updated with server-side time. */
1719 spin_lock(&obj->oo_guard);
1720 inode->i_ctime = *t;
1721 spin_unlock(&obj->oo_guard);
1722 mark_inode_dirty(inode);
1727 * Put the fid into lustre_mdt_attrs, and then place the structure
1728 * inode's ea. This fid should not be altered during the life time
1731 * \retval +ve, on success
1732 * \retval -ve, on error
1734 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
1736 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
1737 const struct lu_fid *fid)
1739 struct osd_thread_info *info = osd_oti_get(env);
1740 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1742 fid_cpu_to_be(&mdt_attrs->lma_self_fid, fid);
1744 return __osd_xattr_set(env, dt,
1745 osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
1746 XATTR_NAME_LMA, LU_XATTR_CREATE);
1751 * Helper function to form igif
1753 static inline void osd_igif_get(const struct lu_env *env, struct inode *inode,
1756 lu_igif_build(fid, inode->i_ino, inode->i_generation);
1760 * Helper function to pack the fid
1762 static inline void osd_fid_pack(const struct lu_env *env, const struct lu_fid *fid,
1763 struct dt_rec *pack)
1765 fid_pack((struct lu_fid_pack *)pack, fid, &osd_oti_get(env)->oti_fid);
1769 * Try to read the fid from inode ea into dt_rec, if return value
1770 * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
1772 * \param fid, object fid.
1774 * \retval 0, on success
1776 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
1777 __u32 ino, struct lu_fid *fid)
1779 struct osd_thread_info *info = osd_oti_get(env);
1780 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1781 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1782 struct dentry *dentry = &info->oti_child_dentry;
1783 struct osd_inode_id *id = &info->oti_id;
1784 struct osd_device *dev;
1785 struct inode *inode;
1789 dev = osd_dev(ldev);
1792 id->oii_gen = OSD_OII_NOGEN;
1794 inode = osd_iget(info, dev, id);
1795 if (IS_ERR(inode)) {
1796 rc = PTR_ERR(inode);
1799 dentry->d_inode = inode;
1801 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1802 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
1805 /* Check LMA compatibility */
1807 (mdt_attrs->lma_incompat & ~cpu_to_be32(LMA_INCOMPAT_SUPP))) {
1808 CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
1809 inode->i_ino, be32_to_cpu(mdt_attrs->lma_incompat) &
1810 ~LMA_INCOMPAT_SUPP);
1815 fid_be_to_cpu(fid, &mdt_attrs->lma_self_fid);
1817 } else if (rc == -ENODATA) {
1818 osd_igif_get(env, inode, fid);
1828 * OSD layer object create function for interoperability mode (b11826).
1829 * This is mostly similar to osd_object_create(). Only difference being, fid is
1830 * inserted into inode ea here.
1832 * \retval 0, on success
1833 * \retval -ve, on error
1835 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
1836 struct lu_attr *attr,
1837 struct dt_allocation_hint *hint,
1838 struct dt_object_format *dof,
1841 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1842 struct osd_object *obj = osd_dt_obj(dt);
1843 struct osd_thread_info *info = osd_oti_get(env);
1849 LASSERT(osd_invariant(obj));
1850 LASSERT(!dt_object_exists(dt));
1851 LASSERT(osd_write_locked(env, obj));
1852 LASSERT(th != NULL);
1854 result = __osd_object_create(info, obj, attr, hint, dof, th);
1856 if (hint && hint->dah_parent)
1857 is_root = osd_object_is_root(osd_dt_obj(hint->dah_parent));
1859 /* objects under osd root shld have igif fid, so dont add fid EA */
1860 if (result == 0 && is_root == 0)
1861 result = osd_ea_fid_set(env, dt, fid);
1864 result = __osd_oi_insert(env, obj, fid, th);
1866 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1867 LINVRNT(osd_invariant(obj));
1872 * Concurrency: @dt is write locked.
1874 static void osd_object_ref_add(const struct lu_env *env,
1875 struct dt_object *dt,
1878 struct osd_object *obj = osd_dt_obj(dt);
1879 struct inode *inode = obj->oo_inode;
1881 LINVRNT(osd_invariant(obj));
1882 LASSERT(dt_object_exists(dt));
1883 LASSERT(osd_write_locked(env, obj));
1884 LASSERT(th != NULL);
1886 spin_lock(&obj->oo_guard);
1887 LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
1889 spin_unlock(&obj->oo_guard);
1890 mark_inode_dirty(inode);
1891 LINVRNT(osd_invariant(obj));
1895 * Concurrency: @dt is write locked.
1897 static void osd_object_ref_del(const struct lu_env *env,
1898 struct dt_object *dt,
1901 struct osd_object *obj = osd_dt_obj(dt);
1902 struct inode *inode = obj->oo_inode;
1904 LINVRNT(osd_invariant(obj));
1905 LASSERT(dt_object_exists(dt));
1906 LASSERT(osd_write_locked(env, obj));
1907 LASSERT(th != NULL);
1909 spin_lock(&obj->oo_guard);
1910 LASSERT(inode->i_nlink > 0);
1912 spin_unlock(&obj->oo_guard);
1913 mark_inode_dirty(inode);
1914 LINVRNT(osd_invariant(obj));
1918 * Concurrency: @dt is read locked.
1920 static int osd_xattr_get(const struct lu_env *env,
1921 struct dt_object *dt,
1924 struct lustre_capa *capa)
1926 struct osd_object *obj = osd_dt_obj(dt);
1927 struct inode *inode = obj->oo_inode;
1928 struct osd_thread_info *info = osd_oti_get(env);
1929 struct dentry *dentry = &info->oti_obj_dentry;
1931 LASSERT(dt_object_exists(dt));
1932 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1933 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1935 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1938 dentry->d_inode = inode;
1939 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
1943 * Concurrency: @dt is write locked.
1945 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1946 const struct lu_buf *buf, const char *name, int fl,
1947 struct thandle *handle, struct lustre_capa *capa)
1949 LASSERT(handle != NULL);
1951 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1954 return __osd_xattr_set(env, dt, buf, name, fl);
1958 * Concurrency: @dt is read locked.
1960 static int osd_xattr_list(const struct lu_env *env,
1961 struct dt_object *dt,
1963 struct lustre_capa *capa)
1965 struct osd_object *obj = osd_dt_obj(dt);
1966 struct inode *inode = obj->oo_inode;
1967 struct osd_thread_info *info = osd_oti_get(env);
1968 struct dentry *dentry = &info->oti_obj_dentry;
1970 LASSERT(dt_object_exists(dt));
1971 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
1972 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1974 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1977 dentry->d_inode = inode;
1978 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
1982 * Concurrency: @dt is write locked.
1984 static int osd_xattr_del(const struct lu_env *env,
1985 struct dt_object *dt,
1987 struct thandle *handle,
1988 struct lustre_capa *capa)
1990 struct osd_object *obj = osd_dt_obj(dt);
1991 struct inode *inode = obj->oo_inode;
1992 struct osd_thread_info *info = osd_oti_get(env);
1993 struct dentry *dentry = &info->oti_obj_dentry;
1994 struct timespec *t = &info->oti_time;
1997 LASSERT(dt_object_exists(dt));
1998 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
1999 LASSERT(osd_write_locked(env, obj));
2000 LASSERT(handle != NULL);
2002 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2005 dentry->d_inode = inode;
2006 *t = inode->i_ctime;
2007 rc = inode->i_op->removexattr(dentry, name);
2008 /* ctime should not be updated with server-side time. */
2009 spin_lock(&obj->oo_guard);
2010 inode->i_ctime = *t;
2011 spin_unlock(&obj->oo_guard);
2012 mark_inode_dirty(inode);
2016 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2017 struct dt_object *dt,
2018 struct lustre_capa *old,
2021 struct osd_thread_info *info = osd_oti_get(env);
2022 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2023 struct osd_object *obj = osd_dt_obj(dt);
2024 struct osd_device *dev = osd_obj2dev(obj);
2025 struct lustre_capa_key *key = &info->oti_capa_key;
2026 struct lustre_capa *capa = &info->oti_capa;
2027 struct obd_capa *oc;
2028 struct md_capainfo *ci;
2032 if (!dev->od_fl_capa)
2033 RETURN(ERR_PTR(-ENOENT));
2035 LASSERT(dt_object_exists(dt));
2036 LINVRNT(osd_invariant(obj));
2038 /* renewal sanity check */
2039 if (old && osd_object_auth(env, dt, old, opc))
2040 RETURN(ERR_PTR(-EACCES));
2042 ci = md_capainfo(env);
2044 RETURN(ERR_PTR(-ENOENT));
2046 switch (ci->mc_auth) {
2050 capa->lc_uid = obj->oo_inode->i_uid;
2051 capa->lc_gid = obj->oo_inode->i_gid;
2052 capa->lc_flags = LC_ID_PLAIN;
2054 case LC_ID_CONVERT: {
2057 s[0] = obj->oo_inode->i_uid;
2058 get_random_bytes(&(s[1]), sizeof(__u32));
2059 s[2] = obj->oo_inode->i_gid;
2060 get_random_bytes(&(s[3]), sizeof(__u32));
2061 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2063 RETURN(ERR_PTR(rc));
2065 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2066 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2067 capa->lc_flags = LC_ID_CONVERT;
2071 RETURN(ERR_PTR(-EINVAL));
2074 capa->lc_fid = *fid;
2076 capa->lc_flags |= dev->od_capa_alg << 24;
2077 capa->lc_timeout = dev->od_capa_timeout;
2078 capa->lc_expiry = 0;
2080 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2082 LASSERT(!capa_is_expired(oc));
2086 spin_lock(&capa_lock);
2087 *key = dev->od_capa_keys[1];
2088 spin_unlock(&capa_lock);
2090 capa->lc_keyid = key->lk_keyid;
2091 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2093 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2095 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2096 RETURN(ERR_PTR(rc));
2099 oc = capa_add(dev->od_capa_hash, capa);
2103 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2106 struct osd_object *obj = osd_dt_obj(dt);
2107 struct inode *inode = obj->oo_inode;
2108 struct osd_thread_info *info = osd_oti_get(env);
2109 struct dentry *dentry = &info->oti_obj_dentry;
2110 struct file *file = &info->oti_file;
2113 dentry->d_inode = inode;
2114 file->f_dentry = dentry;
2115 file->f_mapping = inode->i_mapping;
2116 file->f_op = inode->i_fop;
2117 LOCK_INODE_MUTEX(inode);
2118 rc = file->f_op->fsync(file, dentry, 0);
2119 UNLOCK_INODE_MUTEX(inode);
2124 * Get the 64-bit version for an inode.
2126 static dt_obj_version_t osd_object_version_get(const struct lu_env *env,
2127 struct dt_object *dt)
2129 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2131 CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
2132 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2133 return LDISKFS_I(inode)->i_fs_version;
2137 * Set the 64-bit version and return the old version.
2139 static void osd_object_version_set(const struct lu_env *env, struct dt_object *dt,
2140 dt_obj_version_t new_version)
2142 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2144 CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2145 new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2146 LDISKFS_I(inode)->i_fs_version = new_version;
2147 /** Version is set after all inode operations are finished,
2148 * so we should mark it dirty here */
2149 inode->i_sb->s_op->dirty_inode(inode);
2152 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2155 struct osd_object *obj = osd_dt_obj(dt);
2158 *data = (void *)obj->oo_inode;
2166 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2167 const struct dt_index_features *feat)
2169 struct iam_descr *descr;
2170 struct dt_object *dt = &o->oo_dt;
2172 if (osd_object_is_root(o))
2173 return feat == &dt_directory_features;
2175 LASSERT(o->oo_dir != NULL);
2177 descr = o->oo_dir->od_container.ic_descr;
2178 if (feat == &dt_directory_features) {
2179 if (descr->id_rec_size == sizeof(struct lu_fid_pack))
2182 if (descr == &iam_htree_compat_param) {
2183 /* if it is a HTREE dir then there is good chance that,
2184 * we dealing with ext3 directory here with no FIDs. */
2186 if (descr->id_rec_size ==
2187 sizeof ((struct ldiskfs_dir_entry_2 *)NULL)->inode) {
2189 dt->do_index_ops = &osd_index_ea_ops;
2196 feat->dif_keysize_min <= descr->id_key_size &&
2197 descr->id_key_size <= feat->dif_keysize_max &&
2198 feat->dif_recsize_min <= descr->id_rec_size &&
2199 descr->id_rec_size <= feat->dif_recsize_max &&
2200 !(feat->dif_flags & (DT_IND_VARKEY |
2201 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2202 ergo(feat->dif_flags & DT_IND_UPDATE,
2203 1 /* XXX check that object (and file system) is
2208 static int osd_iam_container_init(const struct lu_env *env,
2209 struct osd_object *obj,
2210 struct osd_directory *dir)
2213 struct iam_container *bag;
2215 bag = &dir->od_container;
2216 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2218 result = iam_container_setup(bag);
2220 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2222 iam_container_fini(bag);
2229 * Concurrency: no external locking is necessary.
2231 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2232 const struct dt_index_features *feat)
2236 struct osd_object *obj = osd_dt_obj(dt);
2237 struct osd_device *osd = osd_obj2dev(obj);
2239 LINVRNT(osd_invariant(obj));
2240 LASSERT(dt_object_exists(dt));
2242 if (osd_object_is_root(obj)) {
2243 dt->do_index_ops = &osd_index_ea_ops;
2245 } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2246 dt->do_index_ops = &osd_index_ea_ops;
2247 if (S_ISDIR(obj->oo_inode->i_mode))
2252 } else if (!osd_has_index(obj)) {
2253 struct osd_directory *dir;
2258 spin_lock(&obj->oo_guard);
2259 if (obj->oo_dir == NULL)
2263 * Concurrent thread allocated container data.
2266 spin_unlock(&obj->oo_guard);
2268 * Now, that we have container data, serialize its
2271 down_write(&obj->oo_ext_idx_sem);
2273 * recheck under lock.
2275 if (!osd_has_index(obj))
2276 result = osd_iam_container_init(env, obj, dir);
2279 up_write(&obj->oo_ext_idx_sem);
2285 if (result == 0 && ea_dir == 0) {
2286 if (!osd_iam_index_probe(env, obj, feat))
2289 LINVRNT(osd_invariant(obj));
2294 static const struct dt_object_operations osd_obj_ops = {
2295 .do_read_lock = osd_object_read_lock,
2296 .do_write_lock = osd_object_write_lock,
2297 .do_read_unlock = osd_object_read_unlock,
2298 .do_write_unlock = osd_object_write_unlock,
2299 .do_write_locked = osd_object_write_locked,
2300 .do_attr_get = osd_attr_get,
2301 .do_attr_set = osd_attr_set,
2302 .do_ah_init = osd_ah_init,
2303 .do_create = osd_object_create,
2304 .do_index_try = osd_index_try,
2305 .do_ref_add = osd_object_ref_add,
2306 .do_ref_del = osd_object_ref_del,
2307 .do_xattr_get = osd_xattr_get,
2308 .do_xattr_set = osd_xattr_set,
2309 .do_xattr_del = osd_xattr_del,
2310 .do_xattr_list = osd_xattr_list,
2311 .do_capa_get = osd_capa_get,
2312 .do_object_sync = osd_object_sync,
2313 .do_version_get = osd_object_version_get,
2314 .do_version_set = osd_object_version_set,
2315 .do_data_get = osd_data_get,
2319 * dt_object_operations for interoperability mode
2320 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2322 static const struct dt_object_operations osd_obj_ea_ops = {
2323 .do_read_lock = osd_object_read_lock,
2324 .do_write_lock = osd_object_write_lock,
2325 .do_read_unlock = osd_object_read_unlock,
2326 .do_write_unlock = osd_object_write_unlock,
2327 .do_write_locked = osd_object_write_locked,
2328 .do_attr_get = osd_attr_get,
2329 .do_attr_set = osd_attr_set,
2330 .do_ah_init = osd_ah_init,
2331 .do_create = osd_object_ea_create,
2332 .do_index_try = osd_index_try,
2333 .do_ref_add = osd_object_ref_add,
2334 .do_ref_del = osd_object_ref_del,
2335 .do_xattr_get = osd_xattr_get,
2336 .do_xattr_set = osd_xattr_set,
2337 .do_xattr_del = osd_xattr_del,
2338 .do_xattr_list = osd_xattr_list,
2339 .do_capa_get = osd_capa_get,
2340 .do_object_sync = osd_object_sync,
2341 .do_version_get = osd_object_version_get,
2342 .do_version_set = osd_object_version_set,
2343 .do_data_get = osd_data_get,
2351 * XXX: Another layering violation for now.
2353 * We don't want to use ->f_op->read methods, because generic file write
2355 * - serializes on ->i_sem, and
2357 * - does a lot of extra work like balance_dirty_pages(),
2359 * which doesn't work for globally shared files like /last-received.
2361 int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
2362 int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
2363 loff_t *offs, handle_t *handle);
2365 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
2366 struct lu_buf *buf, loff_t *pos,
2367 struct lustre_capa *capa)
2369 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2371 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2374 return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
2377 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
2378 const struct lu_buf *buf, loff_t *pos,
2379 struct thandle *handle, struct lustre_capa *capa,
2382 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2383 struct osd_thandle *oh;
2385 #ifdef HAVE_QUOTA_SUPPORT
2386 cfs_cap_t save = current->cap_effective;
2389 LASSERT(handle != NULL);
2391 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
2394 oh = container_of(handle, struct osd_thandle, ot_super);
2395 LASSERT(oh->ot_handle->h_transaction != NULL);
2396 #ifdef HAVE_QUOTA_SUPPORT
2398 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2400 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2402 result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
2403 pos, oh->ot_handle);
2404 #ifdef HAVE_QUOTA_SUPPORT
2405 current->cap_effective = save;
2408 result = buf->lb_len;
2412 static const struct dt_body_operations osd_body_ops = {
2413 .dbo_read = osd_read,
2414 .dbo_write = osd_write
2419 * delete a (key, value) pair from index \a dt specified by \a key
2421 * \param dt_object osd index object
2422 * \param key key for index
2423 * \param rec record reference
2424 * \param handle transaction handler
2427 * \retval -ve failure
2430 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2431 const struct dt_key *key, struct thandle *handle,
2432 struct lustre_capa *capa)
2434 struct osd_object *obj = osd_dt_obj(dt);
2435 struct osd_thandle *oh;
2436 struct iam_path_descr *ipd;
2437 struct iam_container *bag = &obj->oo_dir->od_container;
2442 LINVRNT(osd_invariant(obj));
2443 LASSERT(dt_object_exists(dt));
2444 LASSERT(bag->ic_object == obj->oo_inode);
2445 LASSERT(handle != NULL);
2447 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2450 ipd = osd_idx_ipd_get(env, bag);
2451 if (unlikely(ipd == NULL))
2454 oh = container_of0(handle, struct osd_thandle, ot_super);
2455 LASSERT(oh->ot_handle != NULL);
2456 LASSERT(oh->ot_handle->h_transaction != NULL);
2458 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2459 osd_ipd_put(env, bag, ipd);
2460 LINVRNT(osd_invariant(obj));
2465 * Index delete function for interoperability mode (b11826).
2466 * It will remove the directory entry added by osd_index_ea_insert().
2467 * This entry is needed to maintain name->fid mapping.
2469 * \param key, key i.e. file entry to be deleted
2471 * \retval 0, on success
2472 * \retval -ve, on error
2474 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2475 const struct dt_key *key, struct thandle *handle,
2476 struct lustre_capa *capa)
2478 struct osd_object *obj = osd_dt_obj(dt);
2479 struct inode *dir = obj->oo_inode;
2480 struct dentry *dentry;
2481 struct osd_thandle *oh;
2482 struct ldiskfs_dir_entry_2 *de;
2483 struct buffer_head *bh;
2489 LINVRNT(osd_invariant(obj));
2490 LASSERT(dt_object_exists(dt));
2491 LASSERT(handle != NULL);
2493 oh = container_of(handle, struct osd_thandle, ot_super);
2494 LASSERT(oh->ot_handle != NULL);
2495 LASSERT(oh->ot_handle->h_transaction != NULL);
2497 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2500 dentry = osd_child_dentry_get(env, obj,
2501 (char *)key, strlen((char *)key));
2503 down_write(&obj->oo_ext_idx_sem);
2504 bh = ldiskfs_find_entry(dentry, &de);
2506 struct osd_thread_info *oti = osd_oti_get(env);
2507 struct timespec *ctime = &oti->oti_time;
2508 struct timespec *mtime = &oti->oti_time2;
2510 *ctime = dir->i_ctime;
2511 *mtime = dir->i_mtime;
2512 rc = ldiskfs_delete_entry(oh->ot_handle,
2514 /* xtime should not be updated with server-side time. */
2515 spin_lock(&obj->oo_guard);
2516 dir->i_ctime = *ctime;
2517 dir->i_mtime = *mtime;
2518 spin_unlock(&obj->oo_guard);
2519 mark_inode_dirty(dir);
2524 up_write(&obj->oo_ext_idx_sem);
2525 LASSERT(osd_invariant(obj));
2530 * Lookup index for \a key and copy record to \a rec.
2532 * \param dt_object osd index object
2533 * \param key key for index
2534 * \param rec record reference
2536 * \retval +ve success : exact mach
2537 * \retval 0 return record with key not greater than \a key
2538 * \retval -ve failure
2540 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
2541 struct dt_rec *rec, const struct dt_key *key,
2542 struct lustre_capa *capa)
2544 struct osd_object *obj = osd_dt_obj(dt);
2545 struct iam_path_descr *ipd;
2546 struct iam_container *bag = &obj->oo_dir->od_container;
2547 struct osd_thread_info *oti = osd_oti_get(env);
2548 struct iam_iterator *it = &oti->oti_idx_it;
2552 LASSERT(osd_invariant(obj));
2553 LASSERT(dt_object_exists(dt));
2554 LASSERT(bag->ic_object == obj->oo_inode);
2556 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2559 ipd = osd_idx_ipd_get(env, bag);
2563 /* got ipd now we can start iterator. */
2564 iam_it_init(it, bag, 0, ipd);
2566 rc = iam_it_get(it, (struct iam_key *)key);
2568 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
2572 osd_ipd_put(env, bag, ipd);
2574 LINVRNT(osd_invariant(obj));
2580 * Inserts (key, value) pair in \a dt index object.
2582 * \param dt osd index object
2583 * \param key key for index
2584 * \param rec record reference
2585 * \param th transaction handler
2588 * \retval -ve failure
2590 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
2591 const struct dt_rec *rec, const struct dt_key *key,
2592 struct thandle *th, struct lustre_capa *capa,
2595 struct osd_object *obj = osd_dt_obj(dt);
2596 struct iam_path_descr *ipd;
2597 struct osd_thandle *oh;
2598 struct iam_container *bag = &obj->oo_dir->od_container;
2599 #ifdef HAVE_QUOTA_SUPPORT
2600 cfs_cap_t save = current->cap_effective;
2606 LINVRNT(osd_invariant(obj));
2607 LASSERT(dt_object_exists(dt));
2608 LASSERT(bag->ic_object == obj->oo_inode);
2609 LASSERT(th != NULL);
2611 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2614 ipd = osd_idx_ipd_get(env, bag);
2615 if (unlikely(ipd == NULL))
2618 oh = container_of0(th, struct osd_thandle, ot_super);
2619 LASSERT(oh->ot_handle != NULL);
2620 LASSERT(oh->ot_handle->h_transaction != NULL);
2621 #ifdef HAVE_QUOTA_SUPPORT
2623 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2625 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2627 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
2628 (struct iam_rec *)rec, ipd);
2629 #ifdef HAVE_QUOTA_SUPPORT
2630 current->cap_effective = save;
2632 osd_ipd_put(env, bag, ipd);
2633 LINVRNT(osd_invariant(obj));
2638 * Calls ldiskfs_add_entry() to add directory entry
2639 * into the directory. This is required for
2640 * interoperability mode (b11826)
2642 * \retval 0, on success
2643 * \retval -ve, on error
2645 static int __osd_ea_add_rec(struct osd_thread_info *info,
2646 struct osd_object *pobj,
2647 struct osd_object *cobj,
2651 struct dentry *child;
2652 struct osd_thandle *oth;
2653 struct inode *cinode = cobj->oo_inode;
2656 oth = container_of(th, struct osd_thandle, ot_super);
2657 LASSERT(oth->ot_handle != NULL);
2658 LASSERT(oth->ot_handle->h_transaction != NULL);
2660 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
2661 rc = ldiskfs_add_entry(oth->ot_handle, child, cinode);
2667 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
2668 * into the directory.Also sets flags into osd object to
2669 * indicate dot and dotdot are created. This is required for
2670 * interoperability mode (b11826)
2672 * \param dir directory for dot and dotdot fixup.
2673 * \param obj child object for linking
2675 * \retval 0, on success
2676 * \retval -ve, on error
2678 static int osd_add_dot_dotdot(struct osd_thread_info *info,
2679 struct osd_object *dir,
2680 struct osd_object *obj, const char *name,
2683 struct inode *parent_dir = obj->oo_inode;
2684 struct inode *inode = dir->oo_inode;
2685 struct osd_thandle *oth;
2688 oth = container_of(th, struct osd_thandle, ot_super);
2689 LASSERT(oth->ot_handle->h_transaction != NULL);
2690 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
2692 if (strcmp(name, dot) == 0) {
2693 if (dir->oo_compat_dot_created) {
2696 LASSERT(obj == dir);
2697 dir->oo_compat_dot_created = 1;
2700 } else if(strcmp(name, dotdot) == 0) {
2701 if (!dir->oo_compat_dot_created)
2703 if (dir->oo_compat_dotdot_created)
2704 return __osd_ea_add_rec(info, dir, obj, name, th);
2706 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, inode);
2708 dir->oo_compat_dotdot_created = 1;
2716 * It will call the appropriate osd_add* function and return the
2717 * value, return by respective functions.
2719 static int osd_ea_add_rec(const struct lu_env *env,
2720 struct osd_object *pobj,
2721 struct osd_object *cobj,
2725 struct osd_thread_info *info = osd_oti_get(env);
2728 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
2730 rc = osd_add_dot_dotdot(info, pobj, cobj, name, th);
2732 rc = __osd_ea_add_rec(info, pobj, cobj, name, th);
2738 * Calls ->lookup() to find dentry. From dentry get inode and
2739 * read inode's ea to get fid. This is required for interoperability
2742 * \retval 0, on success
2743 * \retval -ve, on error
2745 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
2746 struct dt_rec *rec, const struct dt_key *key)
2748 struct inode *dir = obj->oo_inode;
2749 struct osd_thread_info *info = osd_oti_get(env);
2750 struct dentry *dentry;
2751 struct ldiskfs_dir_entry_2 *de;
2752 struct buffer_head *bh;
2753 struct lu_fid *fid = &info->oti_fid;
2757 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2759 dentry = osd_child_dentry_get(env, obj,
2760 (char *)key, strlen((char *)key));
2762 down_read(&obj->oo_ext_idx_sem);
2763 bh = ldiskfs_find_entry(dentry, &de);
2765 ino = le32_to_cpu(de->inode);
2767 rc = osd_ea_fid_get(env, obj, ino, fid);
2769 osd_fid_pack(env, fid, rec);
2773 up_read(&obj->oo_ext_idx_sem);
2778 * Find the osd object for given fid.
2780 * \param fid, need to find the osd object having this fid
2782 * \retval osd_object, on success
2783 * \retval -ve, on error
2785 struct osd_object *osd_object_find(const struct lu_env *env,
2786 struct dt_object *dt,
2787 const struct lu_fid *fid)
2789 struct lu_device *ludev = dt->do_lu.lo_dev;
2790 struct osd_object *child = NULL;
2791 struct lu_object *luch;
2792 struct lu_object *lo;
2794 luch = lu_object_find(env, ludev, fid, NULL);
2795 if (!IS_ERR(luch)) {
2796 if (lu_object_exists(luch)) {
2797 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
2799 child = osd_obj(lo);
2801 LU_OBJECT_DEBUG(D_ERROR, env, luch,
2802 "lu_object can't be located"
2803 ""DFID"\n", PFID(fid));
2805 if (child == NULL) {
2806 lu_object_put(env, luch);
2807 CERROR("Unable to get osd_object\n");
2808 child = ERR_PTR(-ENOENT);
2811 LU_OBJECT_DEBUG(D_ERROR, env, luch,
2812 "lu_object does not exists "DFID"\n",
2814 child = ERR_PTR(-ENOENT);
2817 child = (void *)luch;
2823 * Put the osd object once done with it.
2825 * \param obj, osd object that needs to be put
2827 static inline void osd_object_put(const struct lu_env *env,
2828 struct osd_object *obj)
2830 lu_object_put(env, &obj->oo_dt.do_lu);
2834 * Index add function for interoperability mode (b11826).
2835 * It will add the directory entry.This entry is needed to
2836 * maintain name->fid mapping.
2838 * \param key, it is key i.e. file entry to be inserted
2839 * \param rec, it is value of given key i.e. fid
2841 * \retval 0, on success
2842 * \retval -ve, on error
2844 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
2845 const struct dt_rec *rec,
2846 const struct dt_key *key, struct thandle *th,
2847 struct lustre_capa *capa, int ignore_quota)
2849 struct osd_object *obj = osd_dt_obj(dt);
2850 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
2851 const struct lu_fid_pack *pack = (const struct lu_fid_pack *)rec;
2852 const char *name = (const char *)key;
2853 struct osd_object *child;
2854 #ifdef HAVE_QUOTA_SUPPORT
2855 cfs_cap_t save = current->cap_effective;
2861 LASSERT(osd_invariant(obj));
2862 LASSERT(dt_object_exists(dt));
2863 LASSERT(th != NULL);
2865 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2868 rc = fid_unpack(pack, fid);
2871 child = osd_object_find(env, dt, fid);
2872 if (!IS_ERR(child)) {
2873 struct inode *inode = obj->oo_inode;
2874 struct osd_thread_info *oti = osd_oti_get(env);
2875 struct timespec *ctime = &oti->oti_time;
2876 struct timespec *mtime = &oti->oti_time2;
2878 *ctime = inode->i_ctime;
2879 *mtime = inode->i_mtime;
2880 #ifdef HAVE_QUOTA_SUPPORT
2882 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2884 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2886 down_write(&obj->oo_ext_idx_sem);
2887 rc = osd_ea_add_rec(env, obj, child, name, th);
2888 up_write(&obj->oo_ext_idx_sem);
2889 #ifdef HAVE_QUOTA_SUPPORT
2890 current->cap_effective = save;
2892 osd_object_put(env, child);
2893 /* xtime should not be updated with server-side time. */
2894 spin_lock(&obj->oo_guard);
2895 inode->i_ctime = *ctime;
2896 inode->i_mtime = *mtime;
2897 spin_unlock(&obj->oo_guard);
2898 mark_inode_dirty(inode);
2900 rc = PTR_ERR(child);
2903 LASSERT(osd_invariant(obj));
2908 * Initialize osd Iterator for given osd index object.
2910 * \param dt osd index object
2913 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
2914 struct dt_object *dt,
2915 struct lustre_capa *capa)
2917 struct osd_it_iam *it;
2918 struct osd_thread_info *oti = osd_oti_get(env);
2919 struct osd_object *obj = osd_dt_obj(dt);
2920 struct lu_object *lo = &dt->do_lu;
2921 struct iam_path_descr *ipd;
2922 struct iam_container *bag = &obj->oo_dir->od_container;
2924 LASSERT(lu_object_exists(lo));
2926 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2927 return ERR_PTR(-EACCES);
2930 ipd = osd_it_ipd_get(env, bag);
2931 if (likely(ipd != NULL)) {
2935 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
2936 return (struct dt_it *)it;
2938 return ERR_PTR(-ENOMEM);
2942 * free given Iterator.
2945 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
2947 struct osd_it_iam *it = (struct osd_it_iam *)di;
2948 struct osd_object *obj = it->oi_obj;
2950 iam_it_fini(&it->oi_it);
2951 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
2952 lu_object_put(env, &obj->oo_dt.do_lu);
2956 * Move Iterator to record specified by \a key
2958 * \param di osd iterator
2959 * \param key key for index
2961 * \retval +ve di points to record with least key not larger than key
2962 * \retval 0 di points to exact matched key
2963 * \retval -ve failure
2966 static int osd_it_iam_get(const struct lu_env *env,
2967 struct dt_it *di, const struct dt_key *key)
2969 struct osd_it_iam *it = (struct osd_it_iam *)di;
2971 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
2977 * \param di osd iterator
2980 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
2982 struct osd_it_iam *it = (struct osd_it_iam *)di;
2984 iam_it_put(&it->oi_it);
2988 * Move iterator by one record
2990 * \param di osd iterator
2992 * \retval +1 end of container reached
2994 * \retval -ve failure
2997 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
2999 struct osd_it_iam *it = (struct osd_it_iam *)di;
3001 return iam_it_next(&it->oi_it);
3005 * Return pointer to the key under iterator.
3008 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3009 const struct dt_it *di)
3011 struct osd_it_iam *it = (struct osd_it_iam *)di;
3013 return (struct dt_key *)iam_it_key_get(&it->oi_it);
3017 * Return size of key under iterator (in bytes)
3020 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3022 struct osd_it_iam *it = (struct osd_it_iam *)di;
3024 return iam_it_key_size(&it->oi_it);
3027 static inline void osd_it_append_attrs(struct lu_dirent*ent,
3032 struct luda_type *lt;
3033 const unsigned align = sizeof(struct luda_type) - 1;
3035 /* check if file type is required */
3036 if (attr & LUDA_TYPE) {
3037 len = (len + align) & ~align;
3039 lt = (void *) ent->lde_name + len;
3040 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3041 ent->lde_attrs |= LUDA_TYPE;
3044 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3048 * build lu direct from backend fs dirent.
3051 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3059 fid_cpu_to_le(&ent->lde_fid, fid);
3060 ent->lde_attrs = LUDA_FID;
3062 ent->lde_hash = cpu_to_le64(offset);
3063 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3065 strncpy(ent->lde_name, name, namelen);
3066 ent->lde_namelen = cpu_to_le16(namelen);
3068 /* append lustre attributes */
3069 osd_it_append_attrs(ent, attr, namelen, type);
3073 * Return pointer to the record under iterator.
3075 static int osd_it_iam_rec(const struct lu_env *env,
3076 const struct dt_it *di,
3077 struct lu_dirent *lde,
3080 struct osd_it_iam *it = (struct osd_it_iam *)di;
3081 struct osd_thread_info *info = osd_oti_get(env);
3082 struct lu_fid *fid = &info->oti_fid;
3083 const struct lu_fid_pack *rec;
3089 name = (char *)iam_it_key_get(&it->oi_it);
3091 RETURN(PTR_ERR(name));
3093 namelen = iam_it_key_size(&it->oi_it);
3095 rec = (const struct lu_fid_pack *) iam_it_rec_get(&it->oi_it);
3097 RETURN(PTR_ERR(rec));
3099 rc = fid_unpack(rec, fid);
3103 hash = iam_it_store(&it->oi_it);
3105 /* IAM does not store object type in IAM index (dir) */
3106 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3113 * Returns cookie for current Iterator position.
3115 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3117 struct osd_it_iam *it = (struct osd_it_iam *)di;
3119 return iam_it_store(&it->oi_it);
3123 * Restore iterator from cookie.
3125 * \param di osd iterator
3126 * \param hash Iterator location cookie
3128 * \retval +ve di points to record with least key not larger than key.
3129 * \retval 0 di points to exact matched key
3130 * \retval -ve failure
3133 static int osd_it_iam_load(const struct lu_env *env,
3134 const struct dt_it *di, __u64 hash)
3136 struct osd_it_iam *it = (struct osd_it_iam *)di;
3138 return iam_it_load(&it->oi_it, hash);
3141 static const struct dt_index_operations osd_index_iam_ops = {
3142 .dio_lookup = osd_index_iam_lookup,
3143 .dio_insert = osd_index_iam_insert,
3144 .dio_delete = osd_index_iam_delete,
3146 .init = osd_it_iam_init,
3147 .fini = osd_it_iam_fini,
3148 .get = osd_it_iam_get,
3149 .put = osd_it_iam_put,
3150 .next = osd_it_iam_next,
3151 .key = osd_it_iam_key,
3152 .key_size = osd_it_iam_key_size,
3153 .rec = osd_it_iam_rec,
3154 .store = osd_it_iam_store,
3155 .load = osd_it_iam_load
3160 * Creates or initializes iterator context.
3162 * \retval struct osd_it_ea, iterator structure on success
3165 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3166 struct dt_object *dt,
3167 struct lustre_capa *capa)
3169 struct osd_object *obj = osd_dt_obj(dt);
3170 struct osd_thread_info *info = osd_oti_get(env);
3171 struct osd_it_ea *it = &info->oti_it_ea;
3172 struct lu_object *lo = &dt->do_lu;
3173 struct dentry *obj_dentry = &info->oti_it_dentry;
3175 LASSERT(lu_object_exists(lo));
3177 obj_dentry->d_inode = obj->oo_inode;
3178 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3179 obj_dentry->d_name.hash = 0;
3181 it->oie_rd_dirent = 0;
3182 it->oie_it_dirent = 0;
3183 it->oie_dirent = NULL;
3184 it->oie_buf = info->oti_it_ea_buf;
3186 it->oie_file.f_pos = 0;
3187 it->oie_file.f_dentry = obj_dentry;
3188 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
3189 it->oie_file.f_op = obj->oo_inode->i_fop;
3190 it->oie_file.private_data = NULL;
3192 RETURN((struct dt_it*) it);
3196 * Destroy or finishes iterator context.
3198 * \param di, struct osd_it_ea, iterator structure to be destroyed
3200 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3202 struct osd_it_ea *it = (struct osd_it_ea *)di;
3203 struct osd_object *obj = it->oie_obj;
3204 struct inode *inode = obj->oo_inode;
3207 it->oie_file.f_op->release(inode, &it->oie_file);
3208 lu_object_put(env, &obj->oo_dt.do_lu);
3213 * It position the iterator at given key, so that next lookup continues from
3214 * that key Or it is similar to dio_it->load() but based on a key,
3215 * rather than file position.
3217 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3220 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3222 static int osd_it_ea_get(const struct lu_env *env,
3223 struct dt_it *di, const struct dt_key *key)
3225 struct osd_it_ea *it = (struct osd_it_ea *)di;
3228 LASSERT(((const char *)key)[0] == '\0');
3229 it->oie_file.f_pos = 0;
3230 it->oie_rd_dirent = 0;
3231 it->oie_it_dirent = 0;
3232 it->oie_dirent = NULL;
3240 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3245 * It is called internally by ->readdir(). It fills the
3246 * iterator's in-memory data structure with required
3247 * information i.e. name, namelen, rec_size etc.
3249 * \param buf, in which information to be filled in.
3250 * \param name, name of the file in given dir
3252 * \retval 0, on success
3253 * \retval 1, on buffer full
3255 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3256 loff_t offset, __u64 ino,
3259 struct osd_it_ea *it = (struct osd_it_ea *)buf;
3260 struct osd_it_ea_dirent *ent = it->oie_dirent;
3263 /* this should never happen */
3264 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3265 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3269 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3273 ent->oied_ino = ino;
3274 ent->oied_off = offset;
3275 ent->oied_namelen = namelen;
3276 ent->oied_type = d_type;
3278 memcpy(ent->oied_name, name, namelen);
3280 it->oie_rd_dirent++;
3281 it->oie_dirent = (void *) ent + size_round(sizeof(*ent) + namelen);
3286 * Calls ->readdir() to load a directory entry at a time
3287 * and stored it in iterator's in-memory data structure.
3289 * \param di, struct osd_it_ea, iterator's in memory structure
3291 * \retval 0, on success
3292 * \retval -ve, on error
3294 static int osd_ldiskfs_it_fill(const struct dt_it *di)
3296 struct osd_it_ea *it = (struct osd_it_ea *)di;
3297 struct osd_object *obj = it->oie_obj;
3298 struct inode *inode = obj->oo_inode;
3302 it->oie_dirent = it->oie_buf;
3303 it->oie_rd_dirent = 0;
3305 down_read(&obj->oo_ext_idx_sem);
3306 result = inode->i_fop->readdir(&it->oie_file, it,
3307 (filldir_t) osd_ldiskfs_filldir);
3309 up_read(&obj->oo_ext_idx_sem);
3311 if (it->oie_rd_dirent == 0) {
3314 it->oie_dirent = it->oie_buf;
3315 it->oie_it_dirent = 1;
3322 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3323 * to load a directory entry at a time and stored it in
3324 * iterator's in-memory data structure.
3326 * \param di, struct osd_it_ea, iterator's in memory structure
3328 * \retval +ve, iterator reached to end
3329 * \retval 0, iterator not reached to end
3330 * \retval -ve, on error
3332 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
3334 struct osd_it_ea *it = (struct osd_it_ea *)di;
3339 if (it->oie_it_dirent < it->oie_rd_dirent) {
3340 it->oie_dirent = (void *) it->oie_dirent +
3341 size_round(sizeof(struct osd_it_ea_dirent) +
3342 it->oie_dirent->oied_namelen);
3343 it->oie_it_dirent++;
3346 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
3349 rc = osd_ldiskfs_it_fill(di);
3356 * Returns the key at current position from iterator's in memory structure.
3358 * \param di, struct osd_it_ea, iterator's in memory structure
3360 * \retval key i.e. struct dt_key on success
3362 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
3363 const struct dt_it *di)
3365 struct osd_it_ea *it = (struct osd_it_ea *)di;
3367 RETURN((struct dt_key *)it->oie_dirent->oied_name);
3371 * Returns the key's size at current position from iterator's in memory structure.
3373 * \param di, struct osd_it_ea, iterator's in memory structure
3375 * \retval key_size i.e. struct dt_key on success
3377 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
3379 struct osd_it_ea *it = (struct osd_it_ea *)di;
3381 RETURN(it->oie_dirent->oied_namelen);
3386 * Returns the value (i.e. fid/igif) at current position from iterator's
3387 * in memory structure.
3389 * \param di, struct osd_it_ea, iterator's in memory structure
3390 * \param attr, attr requested for dirent.
3391 * \param lde, lustre dirent
3393 * \retval 0, no error and \param lde has correct lustre dirent.
3394 * \retval -ve, on error
3396 static inline int osd_it_ea_rec(const struct lu_env *env,
3397 const struct dt_it *di,
3398 struct lu_dirent *lde,
3401 struct osd_it_ea *it = (struct osd_it_ea *)di;
3402 struct osd_object *obj = it->oie_obj;
3403 struct osd_thread_info *info = osd_oti_get(env);
3404 struct lu_fid *fid = &info->oti_fid;
3409 rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
3412 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
3413 it->oie_dirent->oied_name,
3414 it->oie_dirent->oied_namelen,
3415 it->oie_dirent->oied_type,
3421 * Returns a cookie for current position of the iterator head, so that
3422 * user can use this cookie to load/start the iterator next time.
3424 * \param di, struct osd_it_ea, iterator's in memory structure
3426 * \retval cookie for current position, on success
3428 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
3430 struct osd_it_ea *it = (struct osd_it_ea *)di;
3432 RETURN(it->oie_dirent->oied_off);
3436 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3437 * to load a directory entry at a time and stored it i inn,
3438 * in iterator's in-memory data structure.
3440 * \param di, struct osd_it_ea, iterator's in memory structure
3442 * \retval +ve, on success
3443 * \retval -ve, on error
3445 static int osd_it_ea_load(const struct lu_env *env,
3446 const struct dt_it *di, __u64 hash)
3448 struct osd_it_ea *it = (struct osd_it_ea *)di;
3452 it->oie_file.f_pos = hash;
3454 rc = osd_ldiskfs_it_fill(di);
3462 * Index lookup function for interoperability mode (b11826).
3464 * \param key, key i.e. file name to be searched
3466 * \retval +ve, on success
3467 * \retval -ve, on error
3469 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
3470 struct dt_rec *rec, const struct dt_key *key,
3471 struct lustre_capa *capa)
3473 struct osd_object *obj = osd_dt_obj(dt);
3478 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
3479 LINVRNT(osd_invariant(obj));
3481 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3484 rc = osd_ea_lookup_rec(env, obj, rec, key);
3492 * Index and Iterator operations for interoperability
3493 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3495 static const struct dt_index_operations osd_index_ea_ops = {
3496 .dio_lookup = osd_index_ea_lookup,
3497 .dio_insert = osd_index_ea_insert,
3498 .dio_delete = osd_index_ea_delete,
3500 .init = osd_it_ea_init,
3501 .fini = osd_it_ea_fini,
3502 .get = osd_it_ea_get,
3503 .put = osd_it_ea_put,
3504 .next = osd_it_ea_next,
3505 .key = osd_it_ea_key,
3506 .key_size = osd_it_ea_key_size,
3507 .rec = osd_it_ea_rec,
3508 .store = osd_it_ea_store,
3509 .load = osd_it_ea_load
3513 static void *osd_key_init(const struct lu_context *ctx,
3514 struct lu_context_key *key)
3516 struct osd_thread_info *info;
3518 OBD_ALLOC_PTR(info);
3520 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3521 if (info->oti_it_ea_buf != NULL) {
3522 info->oti_env = container_of(ctx, struct lu_env,
3526 info = ERR_PTR(-ENOMEM);
3529 info = ERR_PTR(-ENOMEM);
3534 static void osd_key_fini(const struct lu_context *ctx,
3535 struct lu_context_key *key, void* data)
3537 struct osd_thread_info *info = data;
3539 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3543 static void osd_key_exit(const struct lu_context *ctx,
3544 struct lu_context_key *key, void *data)
3546 struct osd_thread_info *info = data;
3548 LASSERT(info->oti_r_locks == 0);
3549 LASSERT(info->oti_w_locks == 0);
3550 LASSERT(info->oti_txns == 0);
3553 /* type constructor/destructor: osd_type_init, osd_type_fini */
3554 LU_TYPE_INIT_FINI(osd, &osd_key);
3556 static struct lu_context_key osd_key = {
3557 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
3558 .lct_init = osd_key_init,
3559 .lct_fini = osd_key_fini,
3560 .lct_exit = osd_key_exit
3564 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
3565 const char *name, struct lu_device *next)
3568 struct lu_context *ctx;
3570 /* context for commit hooks */
3571 ctx = &osd_dev(d)->od_env_for_commit.le_ctx;
3572 rc = lu_context_init(ctx, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
3574 rc = osd_procfs_init(osd_dev(d), name);
3575 ctx->lc_cookie = 0x3;
3580 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
3582 struct osd_thread_info *info = osd_oti_get(env);
3584 if (o->od_obj_area != NULL) {
3585 lu_object_put(env, &o->od_obj_area->do_lu);
3586 o->od_obj_area = NULL;
3588 osd_oi_fini(info, &o->od_oi);
3593 static int osd_mount(const struct lu_env *env,
3594 struct osd_device *o, struct lustre_cfg *cfg)
3596 struct lustre_mount_info *lmi;
3597 const char *dev = lustre_cfg_string(cfg, 0);
3598 struct lustre_disk_data *ldd;
3599 struct lustre_sb_info *lsi;
3603 if (o->od_mount != NULL) {
3604 CERROR("Already mounted (%s)\n", dev);
3609 lmi = server_get_mount(dev);
3611 CERROR("Cannot get mount info for %s!\n", dev);
3615 LASSERT(lmi != NULL);
3616 /* save lustre_mount_info in dt_device */
3619 lsi = s2lsi(lmi->lmi_sb);
3622 if (ldd->ldd_flags & LDD_F_IAM_DIR) {
3624 LCONSOLE_WARN("OSD: IAM mode enabled\n");
3628 o->od_obj_area = NULL;
3632 static struct lu_device *osd_device_fini(const struct lu_env *env,
3633 struct lu_device *d)
3638 shrink_dcache_sb(osd_sb(osd_dev(d)));
3639 osd_sync(env, lu2dt_dev(d));
3641 rc = osd_procfs_fini(osd_dev(d));
3643 CERROR("proc fini error %d \n", rc);
3644 RETURN (ERR_PTR(rc));
3647 if (osd_dev(d)->od_mount)
3648 server_put_mount(osd_dev(d)->od_mount->lmi_name,
3649 osd_dev(d)->od_mount->lmi_mnt);
3650 osd_dev(d)->od_mount = NULL;
3652 lu_context_fini(&osd_dev(d)->od_env_for_commit.le_ctx);
3656 static struct lu_device *osd_device_alloc(const struct lu_env *env,
3657 struct lu_device_type *t,
3658 struct lustre_cfg *cfg)
3660 struct lu_device *l;
3661 struct osd_device *o;
3667 result = dt_device_init(&o->od_dt_dev, t);
3670 l->ld_ops = &osd_lu_ops;
3671 o->od_dt_dev.dd_ops = &osd_dt_ops;
3672 spin_lock_init(&o->od_osfs_lock);
3673 o->od_osfs_age = cfs_time_shift_64(-1000);
3674 o->od_capa_hash = init_capa_hash();
3675 if (o->od_capa_hash == NULL) {
3676 dt_device_fini(&o->od_dt_dev);
3677 l = ERR_PTR(-ENOMEM);
3680 l = ERR_PTR(result);
3685 l = ERR_PTR(-ENOMEM);
3689 static struct lu_device *osd_device_free(const struct lu_env *env,
3690 struct lu_device *d)
3692 struct osd_device *o = osd_dev(d);
3695 cleanup_capa_hash(o->od_capa_hash);
3696 dt_device_fini(&o->od_dt_dev);
3701 static int osd_process_config(const struct lu_env *env,
3702 struct lu_device *d, struct lustre_cfg *cfg)
3704 struct osd_device *o = osd_dev(d);
3708 switch(cfg->lcfg_command) {
3710 err = osd_mount(env, o, cfg);
3713 err = osd_shutdown(env, o);
3722 extern void ldiskfs_orphan_cleanup (struct super_block * sb,
3723 struct ldiskfs_super_block * es);
3725 static int osd_recovery_complete(const struct lu_env *env,
3726 struct lu_device *d)
3728 struct osd_device *o = osd_dev(d);
3730 /* TODO: orphans handling */
3731 ldiskfs_orphan_cleanup(osd_sb(o), LDISKFS_SB(osd_sb(o))->s_es);
3735 static int osd_prepare(const struct lu_env *env,
3736 struct lu_device *pdev,
3737 struct lu_device *dev)
3739 struct osd_device *osd = osd_dev(dev);
3740 struct lustre_sb_info *lsi;
3741 struct lustre_disk_data *ldd;
3742 struct lustre_mount_info *lmi;
3743 struct osd_thread_info *oti = osd_oti_get(env);
3744 struct dt_object *d;
3748 /* 1. initialize oi before any file create or file open */
3749 result = osd_oi_init(oti, &osd->od_oi,
3750 &osd->od_dt_dev, lu2md_dev(pdev));
3754 lmi = osd->od_mount;
3755 lsi = s2lsi(lmi->lmi_sb);
3758 /* 2. setup local objects */
3759 result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
3763 /* 3. open remote object dir */
3764 d = dt_store_open(env, lu2dt_dev(dev), "",
3765 remote_obj_dir, &oti->oti_fid);
3767 osd->od_obj_area = d;
3770 result = PTR_ERR(d);
3771 osd->od_obj_area = NULL;
3778 static const struct lu_object_operations osd_lu_obj_ops = {
3779 .loo_object_init = osd_object_init,
3780 .loo_object_delete = osd_object_delete,
3781 .loo_object_release = osd_object_release,
3782 .loo_object_free = osd_object_free,
3783 .loo_object_print = osd_object_print,
3784 .loo_object_invariant = osd_object_invariant
3787 static const struct lu_device_operations osd_lu_ops = {
3788 .ldo_object_alloc = osd_object_alloc,
3789 .ldo_process_config = osd_process_config,
3790 .ldo_recovery_complete = osd_recovery_complete,
3791 .ldo_prepare = osd_prepare,
3794 static const struct lu_device_type_operations osd_device_type_ops = {
3795 .ldto_init = osd_type_init,
3796 .ldto_fini = osd_type_fini,
3798 .ldto_start = osd_type_start,
3799 .ldto_stop = osd_type_stop,
3801 .ldto_device_alloc = osd_device_alloc,
3802 .ldto_device_free = osd_device_free,
3804 .ldto_device_init = osd_device_init,
3805 .ldto_device_fini = osd_device_fini
3808 static struct lu_device_type osd_device_type = {
3809 .ldt_tags = LU_DEVICE_DT,
3810 .ldt_name = LUSTRE_OSD_NAME,
3811 .ldt_ops = &osd_device_type_ops,
3812 .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3816 * lprocfs legacy support.
3818 static struct obd_ops osd_obd_device_ops = {
3819 .o_owner = THIS_MODULE
3822 static struct lu_local_obj_desc llod_osd_rem_obj_dir = {
3823 .llod_name = remote_obj_dir,
3824 .llod_oid = OSD_REM_OBJ_DIR_OID,
3826 .llod_feat = &dt_directory_features,
3829 static int __init osd_mod_init(void)
3831 struct lprocfs_static_vars lvars;
3834 llo_local_obj_register(&llod_osd_rem_obj_dir);
3835 lprocfs_osd_init_vars(&lvars);
3836 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
3837 LUSTRE_OSD_NAME, &osd_device_type);
3840 static void __exit osd_mod_exit(void)
3842 llo_local_obj_unregister(&llod_osd_rem_obj_dir);
3843 class_unregister_type(LUSTRE_OSD_NAME);
3846 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3847 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
3848 MODULE_LICENSE("GPL");
3850 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);