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>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
45 # define EXPORT_SYMTAB
47 #define DEBUG_SUBSYSTEM S_MDS
49 #include <linux/module.h>
51 /* LUSTRE_VERSION_CODE */
52 #include <lustre_ver.h>
53 /* prerequisite for linux/xattr.h */
54 #include <linux/types.h>
55 /* prerequisite for linux/xattr.h */
57 /* XATTR_{REPLACE,CREATE} */
58 #include <linux/xattr.h>
63 * struct OBD_{ALLOC,FREE}*()
66 #include <obd_support.h>
67 /* struct ptlrpc_thread */
68 #include <lustre_net.h>
71 #include <lustre_fid.h>
73 #include "osd_internal.h"
76 /* llo_* api support */
77 #include <md_object.h>
79 static const char dot[] = ".";
80 static const char dotdot[] = "..";
81 static const char remote_obj_dir[] = "REM_OBJ_DIR";
83 struct osd_directory {
84 struct iam_container od_container;
85 struct iam_descr od_descr;
89 struct dt_object oo_dt;
91 * Inode for file system object represented by this osd_object. This
92 * inode is pinned for the whole duration of lu_object life.
94 * Not modified concurrently (either setup early during object
95 * creation, or assigned by osd_object_create() under write lock).
97 struct inode *oo_inode;
99 * to protect index ops.
101 cfs_rw_semaphore_t oo_ext_idx_sem;
102 cfs_rw_semaphore_t oo_sem;
103 struct osd_directory *oo_dir;
104 /** protects inode attributes. */
105 cfs_spinlock_t oo_guard;
107 * Following two members are used to indicate the presence of dot and
108 * dotdot in the given directory. This is required for interop mode
111 int oo_compat_dot_created;
112 int oo_compat_dotdot_created;
114 const struct lu_env *oo_owner;
115 #ifdef CONFIG_LOCKDEP
116 struct lockdep_map oo_dep_map;
120 static const struct lu_object_operations osd_lu_obj_ops;
121 static const struct lu_device_operations osd_lu_ops;
122 static struct lu_context_key osd_key;
123 static const struct dt_object_operations osd_obj_ops;
124 static const struct dt_object_operations osd_obj_ea_ops;
125 static const struct dt_body_operations osd_body_ops;
126 static const struct dt_index_operations osd_index_iam_ops;
127 static const struct dt_index_operations osd_index_ea_ops;
130 struct thandle ot_super;
132 struct journal_callback ot_jcb;
133 /* Link to the device, for debugging. */
134 struct lu_ref_link *ot_dev_link;
141 static int lu_device_is_osd(const struct lu_device *d)
143 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
146 static struct osd_device *osd_dt_dev(const struct dt_device *d)
148 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
149 return container_of0(d, struct osd_device, od_dt_dev);
152 static struct osd_device *osd_dev(const struct lu_device *d)
154 LASSERT(lu_device_is_osd(d));
155 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
158 static struct osd_device *osd_obj2dev(const struct osd_object *o)
160 return osd_dev(o->oo_dt.do_lu.lo_dev);
163 static struct super_block *osd_sb(const struct osd_device *dev)
165 return dev->od_mount->lmi_mnt->mnt_sb;
168 static int osd_object_is_root(const struct osd_object *obj)
170 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
173 static struct osd_object *osd_obj(const struct lu_object *o)
175 LASSERT(lu_device_is_osd(o->lo_dev));
176 return container_of0(o, struct osd_object, oo_dt.do_lu);
179 static struct osd_object *osd_dt_obj(const struct dt_object *d)
181 return osd_obj(&d->do_lu);
184 static struct lu_device *osd2lu_dev(struct osd_device *osd)
186 return &osd->od_dt_dev.dd_lu_dev;
189 static journal_t *osd_journal(const struct osd_device *dev)
191 return LDISKFS_SB(osd_sb(dev))->s_journal;
194 static int osd_has_index(const struct osd_object *obj)
196 return obj->oo_dt.do_index_ops != NULL;
199 static int osd_object_invariant(const struct lu_object *l)
201 return osd_invariant(osd_obj(l));
204 #ifdef HAVE_QUOTA_SUPPORT
206 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
208 struct md_ucred *uc = md_ucred(env);
212 save->oc_uid = current->fsuid;
213 save->oc_gid = current->fsgid;
214 save->oc_cap = current->cap_effective;
215 current->fsuid = uc->mu_fsuid;
216 current->fsgid = uc->mu_fsgid;
217 current->cap_effective = uc->mu_cap;
221 osd_pop_ctxt(struct osd_ctxt *save)
223 current->fsuid = save->oc_uid;
224 current->fsgid = save->oc_gid;
225 current->cap_effective = save->oc_cap;
229 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
231 return lu_context_key_get(&env->le_ctx, &osd_key);
235 * Concurrency: doesn't matter
237 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
239 return osd_oti_get(env)->oti_r_locks > 0;
243 * Concurrency: doesn't matter
245 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
247 struct osd_thread_info *oti = osd_oti_get(env);
248 return oti->oti_w_locks > 0 && o->oo_owner == env;
252 * Concurrency: doesn't access mutable data
254 static int osd_root_get(const struct lu_env *env,
255 struct dt_device *dev, struct lu_fid *f)
259 inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
260 LU_IGIF_BUILD(f, inode->i_ino, inode->i_generation);
265 * OSD object methods.
269 * Concurrency: no concurrent access is possible that early in object
272 static struct lu_object *osd_object_alloc(const struct lu_env *env,
273 const struct lu_object_header *hdr,
276 struct osd_object *mo;
282 l = &mo->oo_dt.do_lu;
283 dt_object_init(&mo->oo_dt, NULL, d);
284 if (osd_dev(d)->od_iop_mode)
285 mo->oo_dt.do_ops = &osd_obj_ea_ops;
287 mo->oo_dt.do_ops = &osd_obj_ops;
289 l->lo_ops = &osd_lu_obj_ops;
290 cfs_init_rwsem(&mo->oo_sem);
291 cfs_init_rwsem(&mo->oo_ext_idx_sem);
292 cfs_spin_lock_init(&mo->oo_guard);
299 * retrieve object from backend ext fs.
301 static struct inode *osd_iget(struct osd_thread_info *info,
302 struct osd_device *dev,
303 const struct osd_inode_id *id)
305 struct inode *inode = NULL;
307 #ifdef HAVE_EXT4_LDISKFS
308 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
310 /* Newer kernels return an error instead of a NULL pointer */
313 inode = iget(osd_sb(dev), id->oii_ino);
316 CERROR("no inode\n");
317 inode = ERR_PTR(-EACCES);
318 } else if (id->oii_gen != OSD_OII_NOGEN &&
319 inode->i_generation != id->oii_gen) {
321 inode = ERR_PTR(-ESTALE);
322 } else if (inode->i_nlink == 0) {
323 /* due to parallel readdir and unlink,
324 * we can have dead inode here. */
325 CWARN("stale inode\n");
326 make_bad_inode(inode);
328 inode = ERR_PTR(-ESTALE);
329 } else if (is_bad_inode(inode)) {
330 CERROR("bad inode %lx\n",inode->i_ino);
332 inode = ERR_PTR(-ENOENT);
337 static int osd_fid_lookup(const struct lu_env *env,
338 struct osd_object *obj, const struct lu_fid *fid)
340 struct osd_thread_info *info;
341 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
342 struct osd_device *dev;
343 struct osd_inode_id *id;
348 LINVRNT(osd_invariant(obj));
349 LASSERT(obj->oo_inode == NULL);
350 LASSERT(fid_is_sane(fid) || osd_fid_is_root(fid));
352 * This assertion checks that osd layer sees only local
353 * fids. Unfortunately it is somewhat expensive (does a
354 * cache-lookup). Disabling it for production/acceptance-testing.
356 LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
360 info = osd_oti_get(env);
365 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
368 result = osd_oi_lookup(info, oi, fid, id);
370 inode = osd_iget(info, dev, id);
371 if (!IS_ERR(inode)) {
372 obj->oo_inode = inode;
373 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
374 if (dev->od_iop_mode) {
375 obj->oo_compat_dot_created = 1;
376 obj->oo_compat_dotdot_created = 1;
381 * If fid wasn't found in oi, inode-less object is
382 * created, for which lu_object_exists() returns
383 * false. This is used in a (frequent) case when
384 * objects are created as locking anchors or
385 * place holders for objects yet to be created.
387 result = PTR_ERR(inode);
388 } else if (result == -ENOENT)
390 LINVRNT(osd_invariant(obj));
396 * Concurrency: shouldn't matter.
398 static void osd_object_init0(struct osd_object *obj)
400 LASSERT(obj->oo_inode != NULL);
401 obj->oo_dt.do_body_ops = &osd_body_ops;
402 obj->oo_dt.do_lu.lo_header->loh_attr |=
403 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
407 * Concurrency: no concurrent access is possible that early in object
410 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
411 const struct lu_object_conf *unused)
413 struct osd_object *obj = osd_obj(l);
416 LINVRNT(osd_invariant(obj));
418 result = osd_fid_lookup(env, obj, lu_object_fid(l));
420 if (obj->oo_inode != NULL)
421 osd_object_init0(obj);
423 LINVRNT(osd_invariant(obj));
428 * Concurrency: no concurrent access is possible that late in object
431 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
433 struct osd_object *obj = osd_obj(l);
435 LINVRNT(osd_invariant(obj));
437 dt_object_fini(&obj->oo_dt);
444 static struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
445 const struct iam_container *bag)
447 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
448 osd_oti_get(env)->oti_it_ipd);
451 static struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
452 const struct iam_container *bag)
454 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
455 osd_oti_get(env)->oti_idx_ipd);
458 static void osd_ipd_put(const struct lu_env *env,
459 const struct iam_container *bag,
460 struct iam_path_descr *ipd)
462 bag->ic_descr->id_ops->id_ipd_free(ipd);
466 * Concurrency: no concurrent access is possible that late in object
469 static void osd_index_fini(struct osd_object *o)
471 struct iam_container *bag;
473 if (o->oo_dir != NULL) {
474 bag = &o->oo_dir->od_container;
475 if (o->oo_inode != NULL) {
476 if (bag->ic_object == o->oo_inode)
477 iam_container_fini(bag);
479 OBD_FREE_PTR(o->oo_dir);
485 * Concurrency: no concurrent access is possible that late in object
486 * life-cycle (for all existing callers, that is. New callers have to provide
487 * their own locking.)
489 static int osd_inode_unlinked(const struct inode *inode)
491 return inode->i_nlink == 0;
495 OSD_TXN_OI_DELETE_CREDITS = 20,
496 OSD_TXN_INODE_DELETE_CREDITS = 20
504 * Concurrency: doesn't access mutable data.
506 static int osd_param_is_sane(const struct osd_device *dev,
507 const struct txn_param *param)
509 return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
513 * Concurrency: shouldn't matter.
515 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
517 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
518 struct thandle *th = &oh->ot_super;
519 struct dt_device *dev = th->th_dev;
520 struct lu_device *lud = &dev->dd_lu_dev;
522 LASSERT(dev != NULL);
523 LASSERT(oh->ot_handle == NULL);
526 CERROR("transaction @0x%p commit error: %d\n", th, error);
528 struct lu_env *env = &osd_dt_dev(dev)->od_env_for_commit;
530 * This od_env_for_commit is only for commit usage. see
533 lu_context_enter(&env->le_ctx);
534 dt_txn_hook_commit(env, th);
535 lu_context_exit(&env->le_ctx);
538 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
542 lu_context_exit(&th->th_ctx);
543 lu_context_fini(&th->th_ctx);
548 * Concurrency: shouldn't matter.
550 static struct thandle *osd_trans_start(const struct lu_env *env,
554 struct osd_device *dev = osd_dt_dev(d);
556 struct osd_thandle *oh;
562 hook_res = dt_txn_hook_start(env, d, p);
564 RETURN(ERR_PTR(hook_res));
566 if (osd_param_is_sane(dev, p)) {
567 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
569 struct osd_thread_info *oti = osd_oti_get(env);
572 * XXX temporary stuff. Some abstraction layer should
576 jh = ldiskfs_journal_start_sb(osd_sb(dev), p->tp_credits);
582 jh->h_sync = p->tp_sync;
583 lu_device_get(&d->dd_lu_dev);
584 oh->ot_dev_link = lu_ref_add
585 (&d->dd_lu_dev.ld_reference,
587 /* add commit callback */
588 lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
589 lu_context_enter(&th->th_ctx);
590 osd_journal_callback_set(jh, osd_trans_commit_cb,
591 (struct journal_callback *)&oh->ot_jcb);
592 LASSERT(oti->oti_txns == 0);
593 LASSERT(oti->oti_r_locks == 0);
594 LASSERT(oti->oti_w_locks == 0);
601 th = ERR_PTR(-ENOMEM);
603 CERROR("Invalid transaction parameters\n");
604 th = ERR_PTR(-EINVAL);
611 * Concurrency: shouldn't matter.
613 static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
616 struct osd_thandle *oh;
617 struct osd_thread_info *oti = osd_oti_get(env);
621 oh = container_of0(th, struct osd_thandle, ot_super);
622 if (oh->ot_handle != NULL) {
623 handle_t *hdl = oh->ot_handle;
625 LASSERT(oti->oti_txns == 1);
627 LASSERT(oti->oti_r_locks == 0);
628 LASSERT(oti->oti_w_locks == 0);
629 result = dt_txn_hook_stop(env, th);
631 CERROR("Failure in transaction hook: %d\n", result);
632 oh->ot_handle = NULL;
633 result = ldiskfs_journal_stop(hdl);
635 CERROR("Failure to stop transaction: %d\n", result);
641 * Concurrency: no concurrent access is possible that late in object
644 static int osd_inode_remove(const struct lu_env *env, struct osd_object *obj)
646 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
647 struct osd_device *osd = osd_obj2dev(obj);
648 struct osd_thread_info *oti = osd_oti_get(env);
649 struct txn_param *prm = &oti->oti_txn;
650 struct lu_env *env_del_obj = &oti->oti_obj_delete_tx_env;
654 lu_env_init(env_del_obj, LCT_DT_THREAD);
655 txn_param_init(prm, OSD_TXN_OI_DELETE_CREDITS +
656 OSD_TXN_INODE_DELETE_CREDITS);
657 th = osd_trans_start(env_del_obj, &osd->od_dt_dev, prm);
659 result = osd_oi_delete(osd_oti_get(env_del_obj),
660 &osd->od_oi, fid, th);
661 osd_trans_stop(env_del_obj, th);
663 result = PTR_ERR(th);
665 lu_env_fini(env_del_obj);
670 * Called just before object is freed. Releases all resources except for
671 * object itself (that is released by osd_object_free()).
673 * Concurrency: no concurrent access is possible that late in object
676 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
678 struct osd_object *obj = osd_obj(l);
679 struct inode *inode = obj->oo_inode;
681 LINVRNT(osd_invariant(obj));
684 * If object is unlinked remove fid->ino mapping from object index.
691 if (osd_inode_unlinked(inode)) {
692 result = osd_inode_remove(env, obj);
694 LU_OBJECT_DEBUG(D_ERROR, env, l,
695 "Failed to cleanup: %d\n",
700 obj->oo_inode = NULL;
705 * Concurrency: ->loo_object_release() is called under site spin-lock.
707 static void osd_object_release(const struct lu_env *env,
710 struct osd_object *o = osd_obj(l);
712 LASSERT(!lu_object_is_dying(l->lo_header));
713 if (o->oo_inode != NULL && osd_inode_unlinked(o->oo_inode))
714 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE, &l->lo_header->loh_flags);
718 * Concurrency: shouldn't matter.
720 static int osd_object_print(const struct lu_env *env, void *cookie,
721 lu_printer_t p, const struct lu_object *l)
723 struct osd_object *o = osd_obj(l);
726 if (o->oo_dir != NULL)
727 d = o->oo_dir->od_container.ic_descr;
730 return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
732 o->oo_inode ? o->oo_inode->i_ino : 0UL,
733 o->oo_inode ? o->oo_inode->i_generation : 0,
734 d ? d->id_ops->id_name : "plain");
738 * Concurrency: shouldn't matter.
740 int osd_statfs(const struct lu_env *env, struct dt_device *d,
743 struct osd_device *osd = osd_dt_dev(d);
744 struct super_block *sb = osd_sb(osd);
747 cfs_spin_lock(&osd->od_osfs_lock);
749 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
750 result = ll_do_statfs(sb, &osd->od_kstatfs);
751 if (likely(result == 0)) /* N.B. statfs can't really fail */
752 osd->od_osfs_age = cfs_time_current_64();
755 if (likely(result == 0))
756 *sfs = osd->od_kstatfs;
757 cfs_spin_unlock(&osd->od_osfs_lock);
763 * Concurrency: doesn't access mutable data.
765 static void osd_conf_get(const struct lu_env *env,
766 const struct dt_device *dev,
767 struct dt_device_param *param)
770 * XXX should be taken from not-yet-existing fs abstraction layer.
772 param->ddp_max_name_len = LDISKFS_NAME_LEN;
773 param->ddp_max_nlink = LDISKFS_LINK_MAX;
774 param->ddp_block_shift = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
778 * Helper function to get and fill the buffer with input values.
780 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
784 buf = &osd_oti_get(env)->oti_buf;
791 * Concurrency: shouldn't matter.
793 static int osd_sync(const struct lu_env *env, struct dt_device *d)
795 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
796 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
800 * Start commit for OSD device.
802 * An implementation of dt_commit_async method for OSD device.
803 * Asychronously starts underlayng fs sync and thereby a transaction
806 * \param env environment
809 * \see dt_device_operations
811 static int osd_commit_async(const struct lu_env *env,
814 struct super_block *s = osd_sb(osd_dt_dev(d));
817 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
818 RETURN(s->s_op->sync_fs(s, 0));
822 * Concurrency: shouldn't matter.
824 lvfs_sbdev_type fsfilt_ldiskfs_journal_sbdev(struct super_block *);
826 static void osd_ro(const struct lu_env *env, struct dt_device *d)
830 CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
832 __lvfs_set_rdonly(lvfs_sbdev(osd_sb(osd_dt_dev(d))),
833 fsfilt_ldiskfs_journal_sbdev(osd_sb(osd_dt_dev(d))));
839 * Concurrency: serialization provided by callers.
841 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
842 int mode, unsigned long timeout, __u32 alg,
843 struct lustre_capa_key *keys)
845 struct osd_device *dev = osd_dt_dev(d);
848 dev->od_fl_capa = mode;
849 dev->od_capa_timeout = timeout;
850 dev->od_capa_alg = alg;
851 dev->od_capa_keys = keys;
856 * Concurrency: serialization provided by callers.
858 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
859 struct dt_quota_ctxt *ctxt, void *data)
861 struct obd_device *obd = (void *)ctxt;
862 struct vfsmount *mnt = (struct vfsmount *)data;
865 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
866 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
867 obd->obd_lvfs_ctxt.pwdmnt = mnt;
868 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
869 obd->obd_lvfs_ctxt.fs = get_ds();
875 * Note: we do not count into QUOTA here.
876 * If we mount with --data_journal we may need more.
878 static const int osd_dto_credits_noquota[DTO_NR] = {
881 * INDEX_EXTRA_TRANS_BLOCKS(8) +
882 * SINGLEDATA_TRANS_BLOCKS(8)
883 * XXX Note: maybe iam need more, since iam have more level than
886 [DTO_INDEX_INSERT] = 16,
887 [DTO_INDEX_DELETE] = 16,
891 [DTO_IDNEX_UPDATE] = 16,
893 * Create a object. The same as create object in EXT3.
894 * DATA_TRANS_BLOCKS(14) +
895 * INDEX_EXTRA_BLOCKS(8) +
896 * 3(inode bits, groups, GDT)
898 [DTO_OBJECT_CREATE] = 25,
902 [DTO_OBJECT_DELETE] = 25,
905 * 3(inode bits, group, GDT)
907 [DTO_ATTR_SET_BASE] = 3,
909 * Xattr set. The same as xattr of EXT3.
910 * DATA_TRANS_BLOCKS(14)
911 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
912 * are also counted in. Do not know why?
914 [DTO_XATTR_SET] = 14,
917 * creadits for inode change during write.
919 [DTO_WRITE_BASE] = 3,
921 * credits for single block write.
923 [DTO_WRITE_BLOCK] = 14,
925 * Attr set credits for chown.
926 * This is extra credits for setattr, and it is null without quota
928 [DTO_ATTR_SET_CHOWN]= 0
932 * Note: we count into QUOTA here.
933 * If we mount with --data_journal we may need more.
935 static const int osd_dto_credits_quota[DTO_NR] = {
937 * INDEX_EXTRA_TRANS_BLOCKS(8) +
938 * SINGLEDATA_TRANS_BLOCKS(8) +
939 * 2 * QUOTA_TRANS_BLOCKS(2)
941 [DTO_INDEX_INSERT] = 20,
943 * INDEX_EXTRA_TRANS_BLOCKS(8) +
944 * SINGLEDATA_TRANS_BLOCKS(8) +
945 * 2 * QUOTA_TRANS_BLOCKS(2)
947 [DTO_INDEX_DELETE] = 20,
951 [DTO_IDNEX_UPDATE] = 16,
953 * Create a object. Same as create object in EXT3 filesystem.
954 * DATA_TRANS_BLOCKS(16) +
955 * INDEX_EXTRA_BLOCKS(8) +
956 * 3(inode bits, groups, GDT) +
957 * 2 * QUOTA_INIT_BLOCKS(25)
959 [DTO_OBJECT_CREATE] = 77,
962 * DATA_TRANS_BLOCKS(16) +
963 * INDEX_EXTRA_BLOCKS(8) +
964 * 3(inode bits, groups, GDT) +
967 [DTO_OBJECT_DELETE] = 27,
970 * 3 (inode bit, group, GDT) +
972 [DTO_ATTR_SET_BASE] = 3,
974 * Xattr set. The same as xattr of EXT3.
975 * DATA_TRANS_BLOCKS(16)
976 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS are
977 * also counted in. Do not know why?
979 [DTO_XATTR_SET] = 16,
982 * creadits for inode change during write.
984 [DTO_WRITE_BASE] = 3,
986 * credits for single block write.
988 [DTO_WRITE_BLOCK] = 16,
990 * Attr set credits for chown.
991 * It is added to already set setattr credits
992 * 2 * QUOTA_INIT_BLOCKS(25) +
993 * 2 * QUOTA_DEL_BLOCKS(9)
995 [DTO_ATTR_SET_CHOWN]= 68,
998 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
1001 LASSERT(ARRAY_SIZE(osd_dto_credits_noquota) ==
1002 ARRAY_SIZE(osd_dto_credits_quota));
1003 LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits_noquota));
1004 #ifdef HAVE_QUOTA_SUPPORT
1005 if (test_opt(osd_sb(osd_dt_dev(d)), QUOTA))
1006 return osd_dto_credits_quota[op];
1009 return osd_dto_credits_noquota[op];
1012 static const struct dt_device_operations osd_dt_ops = {
1013 .dt_root_get = osd_root_get,
1014 .dt_statfs = osd_statfs,
1015 .dt_trans_start = osd_trans_start,
1016 .dt_trans_stop = osd_trans_stop,
1017 .dt_conf_get = osd_conf_get,
1018 .dt_sync = osd_sync,
1020 .dt_commit_async = osd_commit_async,
1021 .dt_credit_get = osd_credit_get,
1022 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1023 .dt_init_quota_ctxt= osd_init_quota_ctxt,
1026 static void osd_object_read_lock(const struct lu_env *env,
1027 struct dt_object *dt, unsigned role)
1029 struct osd_object *obj = osd_dt_obj(dt);
1030 struct osd_thread_info *oti = osd_oti_get(env);
1032 LINVRNT(osd_invariant(obj));
1034 LASSERT(obj->oo_owner != env);
1035 cfs_down_read_nested(&obj->oo_sem, role);
1037 LASSERT(obj->oo_owner == NULL);
1041 static void osd_object_write_lock(const struct lu_env *env,
1042 struct dt_object *dt, unsigned role)
1044 struct osd_object *obj = osd_dt_obj(dt);
1045 struct osd_thread_info *oti = osd_oti_get(env);
1047 LINVRNT(osd_invariant(obj));
1049 LASSERT(obj->oo_owner != env);
1050 cfs_down_write_nested(&obj->oo_sem, role);
1052 LASSERT(obj->oo_owner == NULL);
1053 obj->oo_owner = env;
1057 static void osd_object_read_unlock(const struct lu_env *env,
1058 struct dt_object *dt)
1060 struct osd_object *obj = osd_dt_obj(dt);
1061 struct osd_thread_info *oti = osd_oti_get(env);
1063 LINVRNT(osd_invariant(obj));
1065 LASSERT(oti->oti_r_locks > 0);
1067 cfs_up_read(&obj->oo_sem);
1070 static void osd_object_write_unlock(const struct lu_env *env,
1071 struct dt_object *dt)
1073 struct osd_object *obj = osd_dt_obj(dt);
1074 struct osd_thread_info *oti = osd_oti_get(env);
1076 LINVRNT(osd_invariant(obj));
1078 LASSERT(obj->oo_owner == env);
1079 LASSERT(oti->oti_w_locks > 0);
1081 obj->oo_owner = NULL;
1082 cfs_up_write(&obj->oo_sem);
1085 static int osd_object_write_locked(const struct lu_env *env,
1086 struct dt_object *dt)
1088 struct osd_object *obj = osd_dt_obj(dt);
1090 LINVRNT(osd_invariant(obj));
1092 return obj->oo_owner == env;
1095 static int capa_is_sane(const struct lu_env *env,
1096 struct osd_device *dev,
1097 struct lustre_capa *capa,
1098 struct lustre_capa_key *keys)
1100 struct osd_thread_info *oti = osd_oti_get(env);
1101 struct lustre_capa *tcapa = &oti->oti_capa;
1102 struct obd_capa *oc;
1106 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1108 if (capa_is_expired(oc)) {
1109 DEBUG_CAPA(D_ERROR, capa, "expired");
1116 if (capa_is_expired_sec(capa)) {
1117 DEBUG_CAPA(D_ERROR, capa, "expired");
1121 cfs_spin_lock(&capa_lock);
1122 for (i = 0; i < 2; i++) {
1123 if (keys[i].lk_keyid == capa->lc_keyid) {
1124 oti->oti_capa_key = keys[i];
1128 cfs_spin_unlock(&capa_lock);
1131 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1135 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1139 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1140 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1144 oc = capa_add(dev->od_capa_hash, capa);
1150 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1151 struct lustre_capa *capa, __u64 opc)
1153 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1154 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1155 struct md_capainfo *ci;
1158 if (!dev->od_fl_capa)
1161 if (capa == BYPASS_CAPA)
1164 ci = md_capainfo(env);
1168 if (ci->mc_auth == LC_ID_NONE)
1172 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1176 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1177 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1182 if (!capa_opc_supported(capa, opc)) {
1183 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1187 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1188 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1195 static struct timespec *osd_inode_time(const struct lu_env *env,
1196 struct inode *inode, __u64 seconds)
1198 struct osd_thread_info *oti = osd_oti_get(env);
1199 struct timespec *t = &oti->oti_time;
1201 t->tv_sec = seconds;
1203 *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1208 static void osd_inode_getattr(const struct lu_env *env,
1209 struct inode *inode, struct lu_attr *attr)
1211 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1212 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1213 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1215 attr->la_atime = LTIME_S(inode->i_atime);
1216 attr->la_mtime = LTIME_S(inode->i_mtime);
1217 attr->la_ctime = LTIME_S(inode->i_ctime);
1218 attr->la_mode = inode->i_mode;
1219 attr->la_size = i_size_read(inode);
1220 attr->la_blocks = inode->i_blocks;
1221 attr->la_uid = inode->i_uid;
1222 attr->la_gid = inode->i_gid;
1223 attr->la_flags = LDISKFS_I(inode)->i_flags;
1224 attr->la_nlink = inode->i_nlink;
1225 attr->la_rdev = inode->i_rdev;
1226 attr->la_blksize = ll_inode_blksize(inode);
1227 attr->la_blkbits = inode->i_blkbits;
1230 static int osd_attr_get(const struct lu_env *env,
1231 struct dt_object *dt,
1232 struct lu_attr *attr,
1233 struct lustre_capa *capa)
1235 struct osd_object *obj = osd_dt_obj(dt);
1237 LASSERT(dt_object_exists(dt));
1238 LINVRNT(osd_invariant(obj));
1240 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1243 cfs_spin_lock(&obj->oo_guard);
1244 osd_inode_getattr(env, obj->oo_inode, attr);
1245 cfs_spin_unlock(&obj->oo_guard);
1249 static int osd_inode_setattr(const struct lu_env *env,
1250 struct inode *inode, const struct lu_attr *attr)
1254 bits = attr->la_valid;
1256 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1258 #ifdef HAVE_QUOTA_SUPPORT
1259 if ((bits & LA_UID && attr->la_uid != inode->i_uid) ||
1260 (bits & LA_GID && attr->la_gid != inode->i_gid)) {
1261 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1267 iattr.ia_valid |= ATTR_UID;
1269 iattr.ia_valid |= ATTR_GID;
1270 iattr.ia_uid = attr->la_uid;
1271 iattr.ia_gid = attr->la_gid;
1272 osd_push_ctxt(env, save);
1273 rc = DQUOT_TRANSFER(inode, &iattr) ? -EDQUOT : 0;
1280 if (bits & LA_ATIME)
1281 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1282 if (bits & LA_CTIME)
1283 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1284 if (bits & LA_MTIME)
1285 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1286 if (bits & LA_SIZE) {
1287 LDISKFS_I(inode)->i_disksize = attr->la_size;
1288 i_size_write(inode, attr->la_size);
1292 /* OSD should not change "i_blocks" which is used by quota.
1293 * "i_blocks" should be changed by ldiskfs only. */
1294 if (bits & LA_BLOCKS)
1295 inode->i_blocks = attr->la_blocks;
1298 inode->i_mode = (inode->i_mode & S_IFMT) |
1299 (attr->la_mode & ~S_IFMT);
1301 inode->i_uid = attr->la_uid;
1303 inode->i_gid = attr->la_gid;
1304 if (bits & LA_NLINK)
1305 inode->i_nlink = attr->la_nlink;
1307 inode->i_rdev = attr->la_rdev;
1309 if (bits & LA_FLAGS)
1310 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags);
1314 static int osd_attr_set(const struct lu_env *env,
1315 struct dt_object *dt,
1316 const struct lu_attr *attr,
1317 struct thandle *handle,
1318 struct lustre_capa *capa)
1320 struct osd_object *obj = osd_dt_obj(dt);
1323 LASSERT(handle != NULL);
1324 LASSERT(dt_object_exists(dt));
1325 LASSERT(osd_invariant(obj));
1327 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1330 cfs_spin_lock(&obj->oo_guard);
1331 rc = osd_inode_setattr(env, obj->oo_inode, attr);
1332 cfs_spin_unlock(&obj->oo_guard);
1335 mark_inode_dirty(obj->oo_inode);
1342 * XXX temporary solution.
1344 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1345 struct lu_attr *attr, struct thandle *th)
1350 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1351 struct lu_attr *attr, struct thandle *th)
1353 osd_object_init0(obj);
1357 static struct dentry * osd_child_dentry_get(const struct lu_env *env,
1358 struct osd_object *obj,
1362 struct osd_thread_info *info = osd_oti_get(env);
1363 struct dentry *child_dentry = &info->oti_child_dentry;
1364 struct dentry *obj_dentry = &info->oti_obj_dentry;
1366 obj_dentry->d_inode = obj->oo_inode;
1367 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
1368 obj_dentry->d_name.hash = 0;
1370 child_dentry->d_name.hash = 0;
1371 child_dentry->d_parent = obj_dentry;
1372 child_dentry->d_name.name = name;
1373 child_dentry->d_name.len = namelen;
1374 return child_dentry;
1378 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1380 struct dt_allocation_hint *hint,
1384 struct osd_device *osd = osd_obj2dev(obj);
1385 struct osd_thandle *oth;
1386 struct dt_object *parent;
1387 struct inode *inode;
1388 #ifdef HAVE_QUOTA_SUPPORT
1389 struct osd_ctxt *save = &info->oti_ctxt;
1392 LINVRNT(osd_invariant(obj));
1393 LASSERT(obj->oo_inode == NULL);
1395 oth = container_of(th, struct osd_thandle, ot_super);
1396 LASSERT(oth->ot_handle->h_transaction != NULL);
1398 if (hint && hint->dah_parent)
1399 parent = hint->dah_parent;
1401 parent = osd->od_obj_area;
1403 LASSERT(parent != NULL);
1404 LASSERT(osd_dt_obj(parent)->oo_inode->i_op != NULL);
1406 #ifdef HAVE_QUOTA_SUPPORT
1407 osd_push_ctxt(info->oti_env, save);
1409 inode = ldiskfs_create_inode(oth->ot_handle,
1410 osd_dt_obj(parent)->oo_inode, mode);
1411 #ifdef HAVE_QUOTA_SUPPORT
1414 if (!IS_ERR(inode)) {
1415 obj->oo_inode = inode;
1418 result = PTR_ERR(inode);
1419 LINVRNT(osd_invariant(obj));
1427 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1428 struct lu_attr *attr,
1429 struct dt_allocation_hint *hint,
1430 struct dt_object_format *dof,
1434 struct osd_thandle *oth;
1435 struct osd_device *osd = osd_obj2dev(obj);
1436 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1438 LASSERT(S_ISDIR(attr->la_mode));
1440 oth = container_of(th, struct osd_thandle, ot_super);
1441 LASSERT(oth->ot_handle->h_transaction != NULL);
1442 result = osd_mkfile(info, obj, mode, hint, th);
1443 if (result == 0 && osd->od_iop_mode == 0) {
1444 LASSERT(obj->oo_inode != NULL);
1446 * XXX uh-oh... call low-level iam function directly.
1449 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1450 sizeof (struct osd_fid_pack),
1456 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1457 struct lu_attr *attr,
1458 struct dt_allocation_hint *hint,
1459 struct dt_object_format *dof,
1463 struct osd_thandle *oth;
1464 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1466 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1468 LASSERT(S_ISREG(attr->la_mode));
1470 oth = container_of(th, struct osd_thandle, ot_super);
1471 LASSERT(oth->ot_handle->h_transaction != NULL);
1473 result = osd_mkfile(info, obj, mode, hint, th);
1475 LASSERT(obj->oo_inode != NULL);
1476 if (feat->dif_flags & DT_IND_VARKEY)
1477 result = iam_lvar_create(obj->oo_inode,
1478 feat->dif_keysize_max,
1480 feat->dif_recsize_max,
1483 result = iam_lfix_create(obj->oo_inode,
1484 feat->dif_keysize_max,
1486 feat->dif_recsize_max,
1493 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1494 struct lu_attr *attr,
1495 struct dt_allocation_hint *hint,
1496 struct dt_object_format *dof,
1499 LASSERT(S_ISREG(attr->la_mode));
1500 return osd_mkfile(info, obj, (attr->la_mode &
1501 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1504 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1505 struct lu_attr *attr,
1506 struct dt_allocation_hint *hint,
1507 struct dt_object_format *dof,
1510 LASSERT(S_ISLNK(attr->la_mode));
1511 return osd_mkfile(info, obj, (attr->la_mode &
1512 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1515 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1516 struct lu_attr *attr,
1517 struct dt_allocation_hint *hint,
1518 struct dt_object_format *dof,
1521 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1524 LINVRNT(osd_invariant(obj));
1525 LASSERT(obj->oo_inode == NULL);
1526 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1527 S_ISFIFO(mode) || S_ISSOCK(mode));
1529 result = osd_mkfile(info, obj, mode, hint, th);
1531 LASSERT(obj->oo_inode != NULL);
1532 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1534 LINVRNT(osd_invariant(obj));
1538 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1540 struct dt_allocation_hint *hint,
1541 struct dt_object_format *dof,
1544 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1546 osd_obj_type_f result;
1562 result = osd_mk_index;
1573 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1574 struct dt_object *parent, cfs_umode_t child_mode)
1578 memset(ah, 0, sizeof(*ah));
1579 ah->dah_parent = parent;
1580 ah->dah_mode = child_mode;
1584 * Helper function for osd_object_create()
1586 * \retval 0, on success
1588 static int __osd_object_create(struct osd_thread_info *info,
1589 struct osd_object *obj, struct lu_attr *attr,
1590 struct dt_allocation_hint *hint,
1591 struct dt_object_format *dof,
1597 result = osd_create_pre(info, obj, attr, th);
1599 result = osd_create_type_f(dof->dof_type)(info, obj,
1600 attr, hint, dof, th);
1602 result = osd_create_post(info, obj, attr, th);
1608 * Helper function for osd_object_create()
1610 * \retval 0, on success
1612 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1613 const struct lu_fid *fid, struct thandle *th)
1615 struct osd_thread_info *info = osd_oti_get(env);
1616 struct osd_inode_id *id = &info->oti_id;
1617 struct osd_device *osd = osd_obj2dev(obj);
1618 struct md_ucred *uc = md_ucred(env);
1620 LASSERT(obj->oo_inode != NULL);
1621 LASSERT(uc != NULL);
1623 id->oii_ino = obj->oo_inode->i_ino;
1624 id->oii_gen = obj->oo_inode->i_generation;
1626 return osd_oi_insert(info, &osd->od_oi, fid, id, th,
1627 uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
1630 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1631 struct lu_attr *attr,
1632 struct dt_allocation_hint *hint,
1633 struct dt_object_format *dof,
1636 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1637 struct osd_object *obj = osd_dt_obj(dt);
1638 struct osd_thread_info *info = osd_oti_get(env);
1643 LINVRNT(osd_invariant(obj));
1644 LASSERT(!dt_object_exists(dt));
1645 LASSERT(osd_write_locked(env, obj));
1646 LASSERT(th != NULL);
1648 result = __osd_object_create(info, obj, attr, hint, dof, th);
1650 result = __osd_oi_insert(env, obj, fid, th);
1652 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1653 LASSERT(osd_invariant(obj));
1658 * Helper function for osd_xattr_set()
1660 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1661 const struct lu_buf *buf, const char *name, int fl)
1663 struct osd_object *obj = osd_dt_obj(dt);
1664 struct inode *inode = obj->oo_inode;
1665 struct osd_thread_info *info = osd_oti_get(env);
1666 struct dentry *dentry = &info->oti_child_dentry;
1667 struct timespec *t = &info->oti_time;
1671 LASSERT(dt_object_exists(dt));
1672 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1673 LASSERT(osd_write_locked(env, obj));
1675 if (fl & LU_XATTR_REPLACE)
1676 fs_flags |= XATTR_REPLACE;
1678 if (fl & LU_XATTR_CREATE)
1679 fs_flags |= XATTR_CREATE;
1681 dentry->d_inode = inode;
1682 *t = inode->i_ctime;
1683 rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
1684 buf->lb_len, fs_flags);
1685 /* ctime should not be updated with server-side time. */
1686 cfs_spin_lock(&obj->oo_guard);
1687 inode->i_ctime = *t;
1688 cfs_spin_unlock(&obj->oo_guard);
1689 mark_inode_dirty(inode);
1694 * Put the fid into lustre_mdt_attrs, and then place the structure
1695 * inode's ea. This fid should not be altered during the life time
1698 * \retval +ve, on success
1699 * \retval -ve, on error
1701 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
1703 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
1704 const struct lu_fid *fid)
1706 struct osd_thread_info *info = osd_oti_get(env);
1707 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1709 lustre_lma_init(mdt_attrs, fid);
1710 lustre_lma_swab(mdt_attrs);
1711 return __osd_xattr_set(env, dt,
1712 osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
1713 XATTR_NAME_LMA, LU_XATTR_CREATE);
1718 * Helper function to form igif
1720 static inline void osd_igif_get(const struct lu_env *env, struct inode *inode,
1723 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
1727 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1729 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1730 struct lu_fid *befider)
1732 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1733 memcpy(pack->fp_area, befider, sizeof(*befider));
1734 pack->fp_len = sizeof(*befider) + 1;
1738 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
1739 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
1740 * To have compatilibility with 1.8 ldiskfs driver we need to have
1741 * magic number at start of fid data.
1742 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
1745 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
1746 const struct dt_rec *fid)
1748 param->edp_magic = LDISKFS_LUFID_MAGIC;
1749 param->edp_len = sizeof(struct lu_fid) + 1;
1751 fid_cpu_to_be((struct lu_fid *)param->edp_data,
1752 (struct lu_fid *)fid);
1755 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1760 switch (pack->fp_len) {
1761 case sizeof *fid + 1:
1762 memcpy(fid, pack->fp_area, sizeof *fid);
1763 fid_be_to_cpu(fid, fid);
1766 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1773 * Try to read the fid from inode ea into dt_rec, if return value
1774 * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
1776 * \param fid object fid.
1778 * \retval 0 on success
1780 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
1781 __u32 ino, struct lu_fid *fid)
1783 struct osd_thread_info *info = osd_oti_get(env);
1784 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1785 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1786 struct dentry *dentry = &info->oti_child_dentry;
1787 struct osd_inode_id *id = &info->oti_id;
1788 struct osd_device *dev;
1789 struct inode *inode;
1793 dev = osd_dev(ldev);
1796 id->oii_gen = OSD_OII_NOGEN;
1798 inode = osd_iget(info, dev, id);
1799 if (IS_ERR(inode)) {
1800 rc = PTR_ERR(inode);
1803 dentry->d_inode = inode;
1805 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1806 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
1809 /* Check LMA compatibility */
1811 (mdt_attrs->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP))) {
1812 CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
1813 inode->i_ino, le32_to_cpu(mdt_attrs->lma_incompat) &
1814 ~LMA_INCOMPAT_SUPP);
1819 lustre_lma_swab(mdt_attrs);
1820 memcpy(fid, &mdt_attrs->lma_self_fid, sizeof(*fid));
1822 } else if (rc == -ENODATA) {
1823 osd_igif_get(env, inode, fid);
1832 * OSD layer object create function for interoperability mode (b11826).
1833 * This is mostly similar to osd_object_create(). Only difference being, fid is
1834 * inserted into inode ea here.
1836 * \retval 0, on success
1837 * \retval -ve, on error
1839 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
1840 struct lu_attr *attr,
1841 struct dt_allocation_hint *hint,
1842 struct dt_object_format *dof,
1845 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1846 struct osd_object *obj = osd_dt_obj(dt);
1847 struct osd_thread_info *info = osd_oti_get(env);
1852 LASSERT(osd_invariant(obj));
1853 LASSERT(!dt_object_exists(dt));
1854 LASSERT(osd_write_locked(env, obj));
1855 LASSERT(th != NULL);
1857 result = __osd_object_create(info, obj, attr, hint, dof, th);
1859 /* objects under osd root shld have igif fid, so dont add fid EA */
1860 if (result == 0 && fid_seq(fid) >= FID_SEQ_DISTRIBUTED_START)
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 cfs_spin_lock(&obj->oo_guard);
1887 LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
1889 cfs_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 cfs_spin_lock(&obj->oo_guard);
1910 LASSERT(inode->i_nlink > 0);
1912 cfs_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 cfs_spin_lock(&obj->oo_guard);
2010 inode->i_ctime = *t;
2011 cfs_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 ll_get_random_bytes(&(s[1]), sizeof(__u32));
2059 s[2] = obj->oo_inode->i_gid;
2060 ll_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 cfs_spin_lock(&capa_lock);
2087 *key = dev->od_capa_keys[1];
2088 cfs_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;
2171 if (osd_object_is_root(o))
2172 return feat == &dt_directory_features;
2174 LASSERT(o->oo_dir != NULL);
2176 descr = o->oo_dir->od_container.ic_descr;
2177 if (feat == &dt_directory_features) {
2178 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2184 feat->dif_keysize_min <= descr->id_key_size &&
2185 descr->id_key_size <= feat->dif_keysize_max &&
2186 feat->dif_recsize_min <= descr->id_rec_size &&
2187 descr->id_rec_size <= feat->dif_recsize_max &&
2188 !(feat->dif_flags & (DT_IND_VARKEY |
2189 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2190 ergo(feat->dif_flags & DT_IND_UPDATE,
2191 1 /* XXX check that object (and file system) is
2196 static int osd_iam_container_init(const struct lu_env *env,
2197 struct osd_object *obj,
2198 struct osd_directory *dir)
2201 struct iam_container *bag;
2203 bag = &dir->od_container;
2204 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2206 result = iam_container_setup(bag);
2208 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2210 iam_container_fini(bag);
2217 * Concurrency: no external locking is necessary.
2219 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2220 const struct dt_index_features *feat)
2224 struct osd_object *obj = osd_dt_obj(dt);
2225 struct osd_device *osd = osd_obj2dev(obj);
2227 LINVRNT(osd_invariant(obj));
2228 LASSERT(dt_object_exists(dt));
2230 if (osd_object_is_root(obj)) {
2231 dt->do_index_ops = &osd_index_ea_ops;
2233 } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2234 dt->do_index_ops = &osd_index_ea_ops;
2235 if (S_ISDIR(obj->oo_inode->i_mode))
2240 } else if (!osd_has_index(obj)) {
2241 struct osd_directory *dir;
2246 cfs_spin_lock(&obj->oo_guard);
2247 if (obj->oo_dir == NULL)
2251 * Concurrent thread allocated container data.
2254 cfs_spin_unlock(&obj->oo_guard);
2256 * Now, that we have container data, serialize its
2259 cfs_down_write(&obj->oo_ext_idx_sem);
2261 * recheck under lock.
2263 if (!osd_has_index(obj))
2264 result = osd_iam_container_init(env, obj, dir);
2267 cfs_up_write(&obj->oo_ext_idx_sem);
2273 if (result == 0 && ea_dir == 0) {
2274 if (!osd_iam_index_probe(env, obj, feat))
2277 LINVRNT(osd_invariant(obj));
2282 static const struct dt_object_operations osd_obj_ops = {
2283 .do_read_lock = osd_object_read_lock,
2284 .do_write_lock = osd_object_write_lock,
2285 .do_read_unlock = osd_object_read_unlock,
2286 .do_write_unlock = osd_object_write_unlock,
2287 .do_write_locked = osd_object_write_locked,
2288 .do_attr_get = osd_attr_get,
2289 .do_attr_set = osd_attr_set,
2290 .do_ah_init = osd_ah_init,
2291 .do_create = osd_object_create,
2292 .do_index_try = osd_index_try,
2293 .do_ref_add = osd_object_ref_add,
2294 .do_ref_del = osd_object_ref_del,
2295 .do_xattr_get = osd_xattr_get,
2296 .do_xattr_set = osd_xattr_set,
2297 .do_xattr_del = osd_xattr_del,
2298 .do_xattr_list = osd_xattr_list,
2299 .do_capa_get = osd_capa_get,
2300 .do_object_sync = osd_object_sync,
2301 .do_version_get = osd_object_version_get,
2302 .do_version_set = osd_object_version_set,
2303 .do_data_get = osd_data_get,
2307 * dt_object_operations for interoperability mode
2308 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2310 static const struct dt_object_operations osd_obj_ea_ops = {
2311 .do_read_lock = osd_object_read_lock,
2312 .do_write_lock = osd_object_write_lock,
2313 .do_read_unlock = osd_object_read_unlock,
2314 .do_write_unlock = osd_object_write_unlock,
2315 .do_write_locked = osd_object_write_locked,
2316 .do_attr_get = osd_attr_get,
2317 .do_attr_set = osd_attr_set,
2318 .do_ah_init = osd_ah_init,
2319 .do_create = osd_object_ea_create,
2320 .do_index_try = osd_index_try,
2321 .do_ref_add = osd_object_ref_add,
2322 .do_ref_del = osd_object_ref_del,
2323 .do_xattr_get = osd_xattr_get,
2324 .do_xattr_set = osd_xattr_set,
2325 .do_xattr_del = osd_xattr_del,
2326 .do_xattr_list = osd_xattr_list,
2327 .do_capa_get = osd_capa_get,
2328 .do_object_sync = osd_object_sync,
2329 .do_version_get = osd_object_version_get,
2330 .do_version_set = osd_object_version_set,
2331 .do_data_get = osd_data_get,
2339 * XXX: Another layering violation for now.
2341 * We don't want to use ->f_op->read methods, because generic file write
2343 * - serializes on ->i_sem, and
2345 * - does a lot of extra work like balance_dirty_pages(),
2347 * which doesn't work for globally shared files like /last-received.
2349 static int osd_ldiskfs_readlink(struct inode *inode, char *buffer, int buflen)
2351 struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2353 memcpy(buffer, (char*)ei->i_data, buflen);
2358 static int osd_ldiskfs_read(struct inode *inode, void *buf, int size,
2361 struct buffer_head *bh;
2362 unsigned long block;
2369 /* prevent reading after eof */
2370 spin_lock(&inode->i_lock);
2371 if (i_size_read(inode) < *offs + size) {
2372 size = i_size_read(inode) - *offs;
2373 spin_unlock(&inode->i_lock);
2375 CDEBUG(D_EXT2, "size %llu is too short to read @%llu\n",
2376 i_size_read(inode), *offs);
2378 } else if (size == 0) {
2382 spin_unlock(&inode->i_lock);
2385 blocksize = 1 << inode->i_blkbits;
2388 block = *offs >> inode->i_blkbits;
2389 boffs = *offs & (blocksize - 1);
2390 csize = min(blocksize - boffs, size);
2391 bh = ldiskfs_bread(NULL, inode, block, 0, &err);
2393 CERROR("can't read block: %d\n", err);
2397 memcpy(buf, bh->b_data + boffs, csize);
2407 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
2408 struct lu_buf *buf, loff_t *pos,
2409 struct lustre_capa *capa)
2411 struct osd_object *obj = osd_dt_obj(dt);
2412 struct inode *inode = obj->oo_inode;
2415 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
2418 /* Read small symlink from inode body as we need to maintain correct
2419 * on-disk symlinks for ldiskfs.
2421 if (S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
2422 (buf->lb_len <= sizeof (LDISKFS_I(inode)->i_data)))
2423 rc = osd_ldiskfs_readlink(inode, buf->lb_buf, buf->lb_len);
2425 rc = osd_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
2430 static int osd_ldiskfs_writelink(struct inode *inode, char *buffer, int buflen)
2433 memcpy((char*)&LDISKFS_I(inode)->i_data, (char *)buffer,
2435 LDISKFS_I(inode)->i_disksize = buflen;
2436 i_size_write(inode, buflen);
2437 inode->i_sb->s_op->dirty_inode(inode);
2442 static int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
2443 loff_t *offs, handle_t *handle)
2445 struct buffer_head *bh = NULL;
2446 loff_t offset = *offs;
2447 loff_t new_size = i_size_read(inode);
2448 unsigned long block;
2449 int blocksize = 1 << inode->i_blkbits;
2453 int dirty_inode = 0;
2455 while (bufsize > 0) {
2459 block = offset >> inode->i_blkbits;
2460 boffs = offset & (blocksize - 1);
2461 size = min(blocksize - boffs, bufsize);
2462 bh = ldiskfs_bread(handle, inode, block, 1, &err);
2464 CERROR("can't read/create block: %d\n", err);
2468 err = ldiskfs_journal_get_write_access(handle, bh);
2470 CERROR("journal_get_write_access() returned error %d\n",
2474 LASSERTF(boffs + size <= bh->b_size,
2475 "boffs %d size %d bh->b_size %lu",
2476 boffs, size, (unsigned long)bh->b_size);
2477 memcpy(bh->b_data + boffs, buf, size);
2478 err = ldiskfs_journal_dirty_metadata(handle, bh);
2482 if (offset + size > new_size)
2483 new_size = offset + size;
2491 /* correct in-core and on-disk sizes */
2492 if (new_size > i_size_read(inode)) {
2493 spin_lock(&inode->i_lock);
2494 if (new_size > i_size_read(inode))
2495 i_size_write(inode, new_size);
2496 if (i_size_read(inode) > LDISKFS_I(inode)->i_disksize) {
2497 LDISKFS_I(inode)->i_disksize = i_size_read(inode);
2500 spin_unlock(&inode->i_lock);
2502 inode->i_sb->s_op->dirty_inode(inode);
2510 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
2511 const struct lu_buf *buf, loff_t *pos,
2512 struct thandle *handle, struct lustre_capa *capa,
2515 struct osd_object *obj = osd_dt_obj(dt);
2516 struct inode *inode = obj->oo_inode;
2517 struct osd_thandle *oh;
2519 #ifdef HAVE_QUOTA_SUPPORT
2520 cfs_cap_t save = current->cap_effective;
2523 LASSERT(handle != NULL);
2525 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
2528 oh = container_of(handle, struct osd_thandle, ot_super);
2529 LASSERT(oh->ot_handle->h_transaction != NULL);
2530 #ifdef HAVE_QUOTA_SUPPORT
2532 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2534 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2536 /* Write small symlink to inode body as we need to maintain correct
2537 * on-disk symlinks for ldiskfs.
2539 if(S_ISLNK(obj->oo_dt.do_lu.lo_header->loh_attr) &&
2540 (buf->lb_len < sizeof (LDISKFS_I(inode)->i_data)))
2541 result = osd_ldiskfs_writelink(inode, buf->lb_buf, buf->lb_len);
2543 result = osd_ldiskfs_write_record(inode, buf->lb_buf,
2546 #ifdef HAVE_QUOTA_SUPPORT
2547 current->cap_effective = save;
2550 result = buf->lb_len;
2554 static const struct dt_body_operations osd_body_ops = {
2555 .dbo_read = osd_read,
2556 .dbo_write = osd_write
2561 * delete a (key, value) pair from index \a dt specified by \a key
2563 * \param dt osd index object
2564 * \param key key for index
2565 * \param rec record reference
2566 * \param handle transaction handler
2569 * \retval -ve failure
2572 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2573 const struct dt_key *key, struct thandle *handle,
2574 struct lustre_capa *capa)
2576 struct osd_object *obj = osd_dt_obj(dt);
2577 struct osd_thandle *oh;
2578 struct iam_path_descr *ipd;
2579 struct iam_container *bag = &obj->oo_dir->od_container;
2584 LINVRNT(osd_invariant(obj));
2585 LASSERT(dt_object_exists(dt));
2586 LASSERT(bag->ic_object == obj->oo_inode);
2587 LASSERT(handle != NULL);
2589 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2592 ipd = osd_idx_ipd_get(env, bag);
2593 if (unlikely(ipd == NULL))
2596 oh = container_of0(handle, struct osd_thandle, ot_super);
2597 LASSERT(oh->ot_handle != NULL);
2598 LASSERT(oh->ot_handle->h_transaction != NULL);
2600 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2601 osd_ipd_put(env, bag, ipd);
2602 LINVRNT(osd_invariant(obj));
2606 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2609 struct osd_fid_pack *rec;
2612 if (de->file_type & LDISKFS_DIRENT_LUFID) {
2613 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2614 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2620 * Index delete function for interoperability mode (b11826).
2621 * It will remove the directory entry added by osd_index_ea_insert().
2622 * This entry is needed to maintain name->fid mapping.
2624 * \param key, key i.e. file entry to be deleted
2626 * \retval 0, on success
2627 * \retval -ve, on error
2629 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2630 const struct dt_key *key, struct thandle *handle,
2631 struct lustre_capa *capa)
2633 struct osd_object *obj = osd_dt_obj(dt);
2634 struct inode *dir = obj->oo_inode;
2635 struct dentry *dentry;
2636 struct osd_thandle *oh;
2637 struct ldiskfs_dir_entry_2 *de;
2638 struct buffer_head *bh;
2644 LINVRNT(osd_invariant(obj));
2645 LASSERT(dt_object_exists(dt));
2646 LASSERT(handle != NULL);
2648 oh = container_of(handle, struct osd_thandle, ot_super);
2649 LASSERT(oh->ot_handle != NULL);
2650 LASSERT(oh->ot_handle->h_transaction != NULL);
2652 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2655 dentry = osd_child_dentry_get(env, obj,
2656 (char *)key, strlen((char *)key));
2658 cfs_down_write(&obj->oo_ext_idx_sem);
2659 bh = ll_ldiskfs_find_entry(dir, dentry, &de);
2661 struct osd_thread_info *oti = osd_oti_get(env);
2662 struct timespec *ctime = &oti->oti_time;
2663 struct timespec *mtime = &oti->oti_time2;
2665 *ctime = dir->i_ctime;
2666 *mtime = dir->i_mtime;
2667 rc = ldiskfs_delete_entry(oh->ot_handle,
2669 /* xtime should not be updated with server-side time. */
2670 cfs_spin_lock(&obj->oo_guard);
2671 dir->i_ctime = *ctime;
2672 dir->i_mtime = *mtime;
2673 cfs_spin_unlock(&obj->oo_guard);
2674 mark_inode_dirty(dir);
2679 cfs_up_write(&obj->oo_ext_idx_sem);
2680 LASSERT(osd_invariant(obj));
2685 * Lookup index for \a key and copy record to \a rec.
2687 * \param dt osd index object
2688 * \param key key for index
2689 * \param rec record reference
2691 * \retval +ve success : exact mach
2692 * \retval 0 return record with key not greater than \a key
2693 * \retval -ve failure
2695 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
2696 struct dt_rec *rec, const struct dt_key *key,
2697 struct lustre_capa *capa)
2699 struct osd_object *obj = osd_dt_obj(dt);
2700 struct iam_path_descr *ipd;
2701 struct iam_container *bag = &obj->oo_dir->od_container;
2702 struct osd_thread_info *oti = osd_oti_get(env);
2703 struct iam_iterator *it = &oti->oti_idx_it;
2704 struct iam_rec *iam_rec;
2708 LASSERT(osd_invariant(obj));
2709 LASSERT(dt_object_exists(dt));
2710 LASSERT(bag->ic_object == obj->oo_inode);
2712 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2715 ipd = osd_idx_ipd_get(env, bag);
2719 /* got ipd now we can start iterator. */
2720 iam_it_init(it, bag, 0, ipd);
2722 rc = iam_it_get(it, (struct iam_key *)key);
2724 if (S_ISDIR(obj->oo_inode->i_mode))
2725 iam_rec = (struct iam_rec *)oti->oti_ldp;
2727 iam_rec = (struct iam_rec *) rec;
2729 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
2730 if (S_ISDIR(obj->oo_inode->i_mode))
2731 osd_fid_unpack((struct lu_fid *) rec,
2732 (struct osd_fid_pack *)iam_rec);
2736 osd_ipd_put(env, bag, ipd);
2738 LINVRNT(osd_invariant(obj));
2744 * Inserts (key, value) pair in \a dt index object.
2746 * \param dt osd index object
2747 * \param key key for index
2748 * \param rec record reference
2749 * \param th transaction handler
2752 * \retval -ve failure
2754 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
2755 const struct dt_rec *rec, const struct dt_key *key,
2756 struct thandle *th, struct lustre_capa *capa,
2759 struct osd_object *obj = osd_dt_obj(dt);
2760 struct iam_path_descr *ipd;
2761 struct osd_thandle *oh;
2762 struct iam_container *bag = &obj->oo_dir->od_container;
2763 #ifdef HAVE_QUOTA_SUPPORT
2764 cfs_cap_t save = current->cap_effective;
2766 struct osd_thread_info *oti = osd_oti_get(env);
2767 struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_ldp;
2772 LINVRNT(osd_invariant(obj));
2773 LASSERT(dt_object_exists(dt));
2774 LASSERT(bag->ic_object == obj->oo_inode);
2775 LASSERT(th != NULL);
2777 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2780 ipd = osd_idx_ipd_get(env, bag);
2781 if (unlikely(ipd == NULL))
2784 oh = container_of0(th, struct osd_thandle, ot_super);
2785 LASSERT(oh->ot_handle != NULL);
2786 LASSERT(oh->ot_handle->h_transaction != NULL);
2787 #ifdef HAVE_QUOTA_SUPPORT
2789 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
2791 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
2793 if (S_ISDIR(obj->oo_inode->i_mode))
2794 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
2796 iam_rec = (struct iam_rec *) rec;
2797 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
2799 #ifdef HAVE_QUOTA_SUPPORT
2800 current->cap_effective = save;
2802 osd_ipd_put(env, bag, ipd);
2803 LINVRNT(osd_invariant(obj));
2808 * Calls ldiskfs_add_entry() to add directory entry
2809 * into the directory. This is required for
2810 * interoperability mode (b11826)
2812 * \retval 0, on success
2813 * \retval -ve, on error
2815 static int __osd_ea_add_rec(struct osd_thread_info *info,
2816 struct osd_object *pobj,
2817 struct inode *cinode,
2819 const struct dt_rec *fid,
2822 struct ldiskfs_dentry_param *ldp;
2823 struct dentry *child;
2824 struct osd_thandle *oth;
2827 oth = container_of(th, struct osd_thandle, ot_super);
2828 LASSERT(oth->ot_handle != NULL);
2829 LASSERT(oth->ot_handle->h_transaction != NULL);
2831 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
2833 if (fid_is_igif((struct lu_fid *)fid) ||
2834 fid_seq((struct lu_fid *)fid) >= FID_SEQ_DISTRIBUTED_START) {
2835 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2836 osd_get_ldiskfs_dirent_param(ldp, fid);
2837 child->d_fsdata = (void*) ldp;
2839 child->d_fsdata = NULL;
2840 rc = ldiskfs_add_entry(oth->ot_handle, child, cinode);
2846 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
2847 * into the directory.Also sets flags into osd object to
2848 * indicate dot and dotdot are created. This is required for
2849 * interoperability mode (b11826)
2851 * \param dir directory for dot and dotdot fixup.
2852 * \param obj child object for linking
2854 * \retval 0, on success
2855 * \retval -ve, on error
2857 static int osd_add_dot_dotdot(struct osd_thread_info *info,
2858 struct osd_object *dir,
2859 struct inode *parent_dir, const char *name,
2860 const struct dt_rec *dot_fid,
2861 const struct dt_rec *dot_dot_fid,
2864 struct inode *inode = dir->oo_inode;
2865 struct ldiskfs_dentry_param *dot_ldp;
2866 struct ldiskfs_dentry_param *dot_dot_ldp;
2867 struct osd_thandle *oth;
2870 oth = container_of(th, struct osd_thandle, ot_super);
2871 LASSERT(oth->ot_handle->h_transaction != NULL);
2872 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
2874 if (strcmp(name, dot) == 0) {
2875 if (dir->oo_compat_dot_created) {
2878 LASSERT(inode == parent_dir);
2879 dir->oo_compat_dot_created = 1;
2882 } else if(strcmp(name, dotdot) == 0) {
2883 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2884 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2886 if (!dir->oo_compat_dot_created)
2888 if (fid_seq((struct lu_fid *) dot_fid) >= FID_SEQ_DISTRIBUTED_START) {
2889 osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
2890 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2895 /* in case of rename, dotdot is already created */
2896 if (dir->oo_compat_dotdot_created) {
2897 return __osd_ea_add_rec(info, dir, parent_dir, name,
2901 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, inode,
2902 dot_ldp, dot_dot_ldp);
2904 dir->oo_compat_dotdot_created = 1;
2912 * It will call the appropriate osd_add* function and return the
2913 * value, return by respective functions.
2915 static int osd_ea_add_rec(const struct lu_env *env,
2916 struct osd_object *pobj,
2917 struct inode *cinode,
2919 const struct dt_rec *fid,
2922 struct osd_thread_info *info = osd_oti_get(env);
2925 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
2927 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
2928 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2931 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, th);
2937 * Calls ->lookup() to find dentry. From dentry get inode and
2938 * read inode's ea to get fid. This is required for interoperability
2941 * \retval 0, on success
2942 * \retval -ve, on error
2944 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
2945 struct dt_rec *rec, const struct dt_key *key)
2947 struct inode *dir = obj->oo_inode;
2948 struct dentry *dentry;
2949 struct ldiskfs_dir_entry_2 *de;
2950 struct buffer_head *bh;
2951 struct lu_fid *fid = (struct lu_fid *) rec;
2955 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2957 dentry = osd_child_dentry_get(env, obj,
2958 (char *)key, strlen((char *)key));
2960 cfs_down_read(&obj->oo_ext_idx_sem);
2961 bh = ll_ldiskfs_find_entry(dir, dentry, &de);
2963 ino = le32_to_cpu(de->inode);
2964 rc = osd_get_fid_from_dentry(de, rec);
2966 /* done with de, release bh */
2969 rc = osd_ea_fid_get(env, obj, ino, fid);
2973 cfs_up_read(&obj->oo_ext_idx_sem);
2978 * Find the osd object for given fid.
2980 * \param fid need to find the osd object having this fid
2982 * \retval osd_object on success
2983 * \retval -ve on error
2985 struct osd_object *osd_object_find(const struct lu_env *env,
2986 struct dt_object *dt,
2987 const struct lu_fid *fid)
2989 struct lu_device *ludev = dt->do_lu.lo_dev;
2990 struct osd_object *child = NULL;
2991 struct lu_object *luch;
2992 struct lu_object *lo;
2994 luch = lu_object_find(env, ludev, fid, NULL);
2995 if (!IS_ERR(luch)) {
2996 if (lu_object_exists(luch)) {
2997 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
2999 child = osd_obj(lo);
3001 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3002 "lu_object can't be located"
3003 ""DFID"\n", PFID(fid));
3005 if (child == NULL) {
3006 lu_object_put(env, luch);
3007 CERROR("Unable to get osd_object\n");
3008 child = ERR_PTR(-ENOENT);
3011 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3012 "lu_object does not exists "DFID"\n",
3014 child = ERR_PTR(-ENOENT);
3017 child = (void *)luch;
3023 * Put the osd object once done with it.
3025 * \param obj osd object that needs to be put
3027 static inline void osd_object_put(const struct lu_env *env,
3028 struct osd_object *obj)
3030 lu_object_put(env, &obj->oo_dt.do_lu);
3034 * Index add function for interoperability mode (b11826).
3035 * It will add the directory entry.This entry is needed to
3036 * maintain name->fid mapping.
3038 * \param key it is key i.e. file entry to be inserted
3039 * \param rec it is value of given key i.e. fid
3041 * \retval 0, on success
3042 * \retval -ve, on error
3044 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3045 const struct dt_rec *rec,
3046 const struct dt_key *key, struct thandle *th,
3047 struct lustre_capa *capa, int ignore_quota)
3049 struct osd_object *obj = osd_dt_obj(dt);
3050 struct lu_fid *fid = (struct lu_fid *) rec;
3051 const char *name = (const char *)key;
3052 struct osd_object *child;
3053 #ifdef HAVE_QUOTA_SUPPORT
3054 cfs_cap_t save = current->cap_effective;
3060 LASSERT(osd_invariant(obj));
3061 LASSERT(dt_object_exists(dt));
3062 LASSERT(th != NULL);
3064 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3067 child = osd_object_find(env, dt, fid);
3068 if (!IS_ERR(child)) {
3069 struct inode *inode = obj->oo_inode;
3070 struct osd_thread_info *oti = osd_oti_get(env);
3071 struct timespec *ctime = &oti->oti_time;
3072 struct timespec *mtime = &oti->oti_time2;
3074 *ctime = inode->i_ctime;
3075 *mtime = inode->i_mtime;
3076 #ifdef HAVE_QUOTA_SUPPORT
3078 current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
3080 current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
3082 cfs_down_write(&obj->oo_ext_idx_sem);
3083 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3084 cfs_up_write(&obj->oo_ext_idx_sem);
3085 #ifdef HAVE_QUOTA_SUPPORT
3086 current->cap_effective = save;
3088 osd_object_put(env, child);
3089 /* xtime should not be updated with server-side time. */
3090 cfs_spin_lock(&obj->oo_guard);
3091 inode->i_ctime = *ctime;
3092 inode->i_mtime = *mtime;
3093 cfs_spin_unlock(&obj->oo_guard);
3094 mark_inode_dirty(inode);
3096 rc = PTR_ERR(child);
3099 LASSERT(osd_invariant(obj));
3104 * Initialize osd Iterator for given osd index object.
3106 * \param dt osd index object
3109 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3110 struct dt_object *dt,
3111 struct lustre_capa *capa)
3113 struct osd_it_iam *it;
3114 struct osd_thread_info *oti = osd_oti_get(env);
3115 struct osd_object *obj = osd_dt_obj(dt);
3116 struct lu_object *lo = &dt->do_lu;
3117 struct iam_path_descr *ipd;
3118 struct iam_container *bag = &obj->oo_dir->od_container;
3120 LASSERT(lu_object_exists(lo));
3122 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3123 return ERR_PTR(-EACCES);
3126 ipd = osd_it_ipd_get(env, bag);
3127 if (likely(ipd != NULL)) {
3131 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3132 return (struct dt_it *)it;
3134 return ERR_PTR(-ENOMEM);
3138 * free given Iterator.
3141 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3143 struct osd_it_iam *it = (struct osd_it_iam *)di;
3144 struct osd_object *obj = it->oi_obj;
3146 iam_it_fini(&it->oi_it);
3147 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3148 lu_object_put(env, &obj->oo_dt.do_lu);
3152 * Move Iterator to record specified by \a key
3154 * \param di osd iterator
3155 * \param key key for index
3157 * \retval +ve di points to record with least key not larger than key
3158 * \retval 0 di points to exact matched key
3159 * \retval -ve failure
3162 static int osd_it_iam_get(const struct lu_env *env,
3163 struct dt_it *di, const struct dt_key *key)
3165 struct osd_it_iam *it = (struct osd_it_iam *)di;
3167 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3173 * \param di osd iterator
3176 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3178 struct osd_it_iam *it = (struct osd_it_iam *)di;
3180 iam_it_put(&it->oi_it);
3184 * Move iterator by one record
3186 * \param di osd iterator
3188 * \retval +1 end of container reached
3190 * \retval -ve failure
3193 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3195 struct osd_it_iam *it = (struct osd_it_iam *)di;
3197 return iam_it_next(&it->oi_it);
3201 * Return pointer to the key under iterator.
3204 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3205 const struct dt_it *di)
3207 struct osd_it_iam *it = (struct osd_it_iam *)di;
3209 return (struct dt_key *)iam_it_key_get(&it->oi_it);
3213 * Return size of key under iterator (in bytes)
3216 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3218 struct osd_it_iam *it = (struct osd_it_iam *)di;
3220 return iam_it_key_size(&it->oi_it);
3223 static inline void osd_it_append_attrs(struct lu_dirent*ent,
3228 struct luda_type *lt;
3229 const unsigned align = sizeof(struct luda_type) - 1;
3231 /* check if file type is required */
3232 if (attr & LUDA_TYPE) {
3233 len = (len + align) & ~align;
3235 lt = (void *) ent->lde_name + len;
3236 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3237 ent->lde_attrs |= LUDA_TYPE;
3240 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3244 * build lu direct from backend fs dirent.
3247 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3255 fid_cpu_to_le(&ent->lde_fid, fid);
3256 ent->lde_attrs = LUDA_FID;
3258 ent->lde_hash = cpu_to_le64(offset);
3259 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3261 strncpy(ent->lde_name, name, namelen);
3262 ent->lde_namelen = cpu_to_le16(namelen);
3264 /* append lustre attributes */
3265 osd_it_append_attrs(ent, attr, namelen, type);
3269 * Return pointer to the record under iterator.
3271 static int osd_it_iam_rec(const struct lu_env *env,
3272 const struct dt_it *di,
3273 struct lu_dirent *lde,
3276 struct osd_it_iam *it = (struct osd_it_iam *)di;
3277 struct osd_thread_info *info = osd_oti_get(env);
3278 struct lu_fid *fid = &info->oti_fid;
3279 const struct osd_fid_pack *rec;
3285 name = (char *)iam_it_key_get(&it->oi_it);
3287 RETURN(PTR_ERR(name));
3289 namelen = iam_it_key_size(&it->oi_it);
3291 rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
3293 RETURN(PTR_ERR(rec));
3295 rc = osd_fid_unpack(fid, rec);
3299 hash = iam_it_store(&it->oi_it);
3301 /* IAM does not store object type in IAM index (dir) */
3302 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3309 * Returns cookie for current Iterator position.
3311 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3313 struct osd_it_iam *it = (struct osd_it_iam *)di;
3315 return iam_it_store(&it->oi_it);
3319 * Restore iterator from cookie.
3321 * \param di osd iterator
3322 * \param hash Iterator location cookie
3324 * \retval +ve di points to record with least key not larger than key.
3325 * \retval 0 di points to exact matched key
3326 * \retval -ve failure
3329 static int osd_it_iam_load(const struct lu_env *env,
3330 const struct dt_it *di, __u64 hash)
3332 struct osd_it_iam *it = (struct osd_it_iam *)di;
3334 return iam_it_load(&it->oi_it, hash);
3337 static const struct dt_index_operations osd_index_iam_ops = {
3338 .dio_lookup = osd_index_iam_lookup,
3339 .dio_insert = osd_index_iam_insert,
3340 .dio_delete = osd_index_iam_delete,
3342 .init = osd_it_iam_init,
3343 .fini = osd_it_iam_fini,
3344 .get = osd_it_iam_get,
3345 .put = osd_it_iam_put,
3346 .next = osd_it_iam_next,
3347 .key = osd_it_iam_key,
3348 .key_size = osd_it_iam_key_size,
3349 .rec = osd_it_iam_rec,
3350 .store = osd_it_iam_store,
3351 .load = osd_it_iam_load
3356 * Creates or initializes iterator context.
3358 * \retval struct osd_it_ea, iterator structure on success
3361 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3362 struct dt_object *dt,
3363 struct lustre_capa *capa)
3365 struct osd_object *obj = osd_dt_obj(dt);
3366 struct osd_thread_info *info = osd_oti_get(env);
3367 struct osd_it_ea *it = &info->oti_it_ea;
3368 struct lu_object *lo = &dt->do_lu;
3369 struct dentry *obj_dentry = &info->oti_it_dentry;
3371 LASSERT(lu_object_exists(lo));
3373 obj_dentry->d_inode = obj->oo_inode;
3374 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3375 obj_dentry->d_name.hash = 0;
3377 it->oie_rd_dirent = 0;
3378 it->oie_it_dirent = 0;
3379 it->oie_dirent = NULL;
3380 it->oie_buf = info->oti_it_ea_buf;
3382 it->oie_file.f_pos = 0;
3383 it->oie_file.f_dentry = obj_dentry;
3384 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
3385 it->oie_file.f_op = obj->oo_inode->i_fop;
3386 it->oie_file.private_data = NULL;
3388 RETURN((struct dt_it *) it);
3392 * Destroy or finishes iterator context.
3394 * \param di iterator structure to be destroyed
3396 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3398 struct osd_it_ea *it = (struct osd_it_ea *)di;
3399 struct osd_object *obj = it->oie_obj;
3400 struct inode *inode = obj->oo_inode;
3403 it->oie_file.f_op->release(inode, &it->oie_file);
3404 lu_object_put(env, &obj->oo_dt.do_lu);
3409 * It position the iterator at given key, so that next lookup continues from
3410 * that key Or it is similar to dio_it->load() but based on a key,
3411 * rather than file position.
3413 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3416 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3418 static int osd_it_ea_get(const struct lu_env *env,
3419 struct dt_it *di, const struct dt_key *key)
3421 struct osd_it_ea *it = (struct osd_it_ea *)di;
3424 LASSERT(((const char *)key)[0] == '\0');
3425 it->oie_file.f_pos = 0;
3426 it->oie_rd_dirent = 0;
3427 it->oie_it_dirent = 0;
3428 it->oie_dirent = NULL;
3436 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3441 * It is called internally by ->readdir(). It fills the
3442 * iterator's in-memory data structure with required
3443 * information i.e. name, namelen, rec_size etc.
3445 * \param buf in which information to be filled in.
3446 * \param name name of the file in given dir
3448 * \retval 0 on success
3449 * \retval 1 on buffer full
3451 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3452 loff_t offset, __u64 ino,
3455 struct osd_it_ea *it = (struct osd_it_ea *)buf;
3456 struct osd_it_ea_dirent *ent = it->oie_dirent;
3457 struct lu_fid *fid = &ent->oied_fid;
3458 struct osd_fid_pack *rec;
3461 /* this should never happen */
3462 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3463 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3467 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3471 if (d_type & LDISKFS_DIRENT_LUFID) {
3472 rec = (struct osd_fid_pack*) (name + namelen + 1);
3474 if (osd_fid_unpack(fid, rec) != 0)
3477 d_type &= ~LDISKFS_DIRENT_LUFID;
3482 ent->oied_ino = ino;
3483 ent->oied_off = offset;
3484 ent->oied_namelen = namelen;
3485 ent->oied_type = d_type;
3487 memcpy(ent->oied_name, name, namelen);
3489 it->oie_rd_dirent++;
3490 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
3495 * Calls ->readdir() to load a directory entry at a time
3496 * and stored it in iterator's in-memory data structure.
3498 * \param di iterator's in memory structure
3500 * \retval 0 on success
3501 * \retval -ve on error
3503 static int osd_ldiskfs_it_fill(const struct dt_it *di)
3505 struct osd_it_ea *it = (struct osd_it_ea *)di;
3506 struct osd_object *obj = it->oie_obj;
3507 struct inode *inode = obj->oo_inode;
3511 it->oie_dirent = it->oie_buf;
3512 it->oie_rd_dirent = 0;
3514 cfs_down_read(&obj->oo_ext_idx_sem);
3515 result = inode->i_fop->readdir(&it->oie_file, it,
3516 (filldir_t) osd_ldiskfs_filldir);
3518 cfs_up_read(&obj->oo_ext_idx_sem);
3520 if (it->oie_rd_dirent == 0) {
3523 it->oie_dirent = it->oie_buf;
3524 it->oie_it_dirent = 1;
3531 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3532 * to load a directory entry at a time and stored it in
3533 * iterator's in-memory data structure.
3535 * \param di iterator's in memory structure
3537 * \retval +ve iterator reached to end
3538 * \retval 0 iterator not reached to end
3539 * \retval -ve on error
3541 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
3543 struct osd_it_ea *it = (struct osd_it_ea *)di;
3548 if (it->oie_it_dirent < it->oie_rd_dirent) {
3550 (void *) it->oie_dirent +
3551 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
3552 it->oie_dirent->oied_namelen);
3553 it->oie_it_dirent++;
3556 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
3559 rc = osd_ldiskfs_it_fill(di);
3566 * Returns the key at current position from iterator's in memory structure.
3568 * \param di iterator's in memory structure
3570 * \retval key i.e. struct dt_key on success
3572 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
3573 const struct dt_it *di)
3575 struct osd_it_ea *it = (struct osd_it_ea *)di;
3577 RETURN((struct dt_key *)it->oie_dirent->oied_name);
3581 * Returns the key's size at current position from iterator's in memory structure.
3583 * \param di iterator's in memory structure
3585 * \retval key_size i.e. struct dt_key on success
3587 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
3589 struct osd_it_ea *it = (struct osd_it_ea *)di;
3591 RETURN(it->oie_dirent->oied_namelen);
3596 * Returns the value (i.e. fid/igif) at current position from iterator's
3597 * in memory structure.
3599 * \param di struct osd_it_ea, iterator's in memory structure
3600 * \param attr attr requested for dirent.
3601 * \param lde lustre dirent
3603 * \retval 0 no error and \param lde has correct lustre dirent.
3604 * \retval -ve on error
3606 static inline int osd_it_ea_rec(const struct lu_env *env,
3607 const struct dt_it *di,
3608 struct lu_dirent *lde,
3611 struct osd_it_ea *it = (struct osd_it_ea *)di;
3612 struct osd_object *obj = it->oie_obj;
3613 struct lu_fid *fid = &it->oie_dirent->oied_fid;
3618 if (!fid_is_sane(fid))
3619 rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
3622 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
3623 it->oie_dirent->oied_name,
3624 it->oie_dirent->oied_namelen,
3625 it->oie_dirent->oied_type,
3631 * Returns a cookie for current position of the iterator head, so that
3632 * user can use this cookie to load/start the iterator next time.
3634 * \param di iterator's in memory structure
3636 * \retval cookie for current position, on success
3638 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
3640 struct osd_it_ea *it = (struct osd_it_ea *)di;
3642 RETURN(it->oie_dirent->oied_off);
3646 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3647 * to load a directory entry at a time and stored it i inn,
3648 * in iterator's in-memory data structure.
3650 * \param di struct osd_it_ea, iterator's in memory structure
3652 * \retval +ve on success
3653 * \retval -ve on error
3655 static int osd_it_ea_load(const struct lu_env *env,
3656 const struct dt_it *di, __u64 hash)
3658 struct osd_it_ea *it = (struct osd_it_ea *)di;
3662 it->oie_file.f_pos = hash;
3664 rc = osd_ldiskfs_it_fill(di);
3672 * Index lookup function for interoperability mode (b11826).
3674 * \param key, key i.e. file name to be searched
3676 * \retval +ve, on success
3677 * \retval -ve, on error
3679 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
3680 struct dt_rec *rec, const struct dt_key *key,
3681 struct lustre_capa *capa)
3683 struct osd_object *obj = osd_dt_obj(dt);
3688 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
3689 LINVRNT(osd_invariant(obj));
3691 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3694 rc = osd_ea_lookup_rec(env, obj, rec, key);
3702 * Index and Iterator operations for interoperability
3703 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3705 static const struct dt_index_operations osd_index_ea_ops = {
3706 .dio_lookup = osd_index_ea_lookup,
3707 .dio_insert = osd_index_ea_insert,
3708 .dio_delete = osd_index_ea_delete,
3710 .init = osd_it_ea_init,
3711 .fini = osd_it_ea_fini,
3712 .get = osd_it_ea_get,
3713 .put = osd_it_ea_put,
3714 .next = osd_it_ea_next,
3715 .key = osd_it_ea_key,
3716 .key_size = osd_it_ea_key_size,
3717 .rec = osd_it_ea_rec,
3718 .store = osd_it_ea_store,
3719 .load = osd_it_ea_load
3723 static void *osd_key_init(const struct lu_context *ctx,
3724 struct lu_context_key *key)
3726 struct osd_thread_info *info;
3728 OBD_ALLOC_PTR(info);
3730 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3731 if (info->oti_it_ea_buf != NULL) {
3732 info->oti_env = container_of(ctx, struct lu_env,
3736 info = ERR_PTR(-ENOMEM);
3739 info = ERR_PTR(-ENOMEM);
3744 static void osd_key_fini(const struct lu_context *ctx,
3745 struct lu_context_key *key, void* data)
3747 struct osd_thread_info *info = data;
3749 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3753 static void osd_key_exit(const struct lu_context *ctx,
3754 struct lu_context_key *key, void *data)
3756 struct osd_thread_info *info = data;
3758 LASSERT(info->oti_r_locks == 0);
3759 LASSERT(info->oti_w_locks == 0);
3760 LASSERT(info->oti_txns == 0);
3763 /* type constructor/destructor: osd_type_init, osd_type_fini */
3764 LU_TYPE_INIT_FINI(osd, &osd_key);
3766 static struct lu_context_key osd_key = {
3767 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
3768 .lct_init = osd_key_init,
3769 .lct_fini = osd_key_fini,
3770 .lct_exit = osd_key_exit
3774 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
3775 const char *name, struct lu_device *next)
3778 struct lu_context *ctx;
3780 /* context for commit hooks */
3781 ctx = &osd_dev(d)->od_env_for_commit.le_ctx;
3782 rc = lu_context_init(ctx, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
3784 rc = osd_procfs_init(osd_dev(d), name);
3785 ctx->lc_cookie = 0x3;
3790 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
3792 struct osd_thread_info *info = osd_oti_get(env);
3794 if (o->od_obj_area != NULL) {
3795 lu_object_put(env, &o->od_obj_area->do_lu);
3796 o->od_obj_area = NULL;
3798 osd_oi_fini(info, &o->od_oi);
3803 static int osd_mount(const struct lu_env *env,
3804 struct osd_device *o, struct lustre_cfg *cfg)
3806 struct lustre_mount_info *lmi;
3807 const char *dev = lustre_cfg_string(cfg, 0);
3808 struct lustre_disk_data *ldd;
3809 struct lustre_sb_info *lsi;
3812 if (o->od_mount != NULL) {
3813 CERROR("Already mounted (%s)\n", dev);
3818 lmi = server_get_mount(dev);
3820 CERROR("Cannot get mount info for %s!\n", dev);
3824 LASSERT(lmi != NULL);
3825 /* save lustre_mount_info in dt_device */
3828 lsi = s2lsi(lmi->lmi_sb);
3831 if (ldd->ldd_flags & LDD_F_IAM_DIR) {
3833 LCONSOLE_WARN("OSD: IAM mode enabled\n");
3837 o->od_obj_area = NULL;
3841 static struct lu_device *osd_device_fini(const struct lu_env *env,
3842 struct lu_device *d)
3847 shrink_dcache_sb(osd_sb(osd_dev(d)));
3848 osd_sync(env, lu2dt_dev(d));
3850 rc = osd_procfs_fini(osd_dev(d));
3852 CERROR("proc fini error %d \n", rc);
3853 RETURN (ERR_PTR(rc));
3856 if (osd_dev(d)->od_mount)
3857 server_put_mount(osd_dev(d)->od_mount->lmi_name,
3858 osd_dev(d)->od_mount->lmi_mnt);
3859 osd_dev(d)->od_mount = NULL;
3861 lu_context_fini(&osd_dev(d)->od_env_for_commit.le_ctx);
3865 static struct lu_device *osd_device_alloc(const struct lu_env *env,
3866 struct lu_device_type *t,
3867 struct lustre_cfg *cfg)
3869 struct lu_device *l;
3870 struct osd_device *o;
3876 result = dt_device_init(&o->od_dt_dev, t);
3879 l->ld_ops = &osd_lu_ops;
3880 o->od_dt_dev.dd_ops = &osd_dt_ops;
3881 cfs_spin_lock_init(&o->od_osfs_lock);
3882 o->od_osfs_age = cfs_time_shift_64(-1000);
3883 o->od_capa_hash = init_capa_hash();
3884 if (o->od_capa_hash == NULL) {
3885 dt_device_fini(&o->od_dt_dev);
3886 l = ERR_PTR(-ENOMEM);
3889 l = ERR_PTR(result);
3894 l = ERR_PTR(-ENOMEM);
3898 static struct lu_device *osd_device_free(const struct lu_env *env,
3899 struct lu_device *d)
3901 struct osd_device *o = osd_dev(d);
3904 cleanup_capa_hash(o->od_capa_hash);
3905 dt_device_fini(&o->od_dt_dev);
3910 static int osd_process_config(const struct lu_env *env,
3911 struct lu_device *d, struct lustre_cfg *cfg)
3913 struct osd_device *o = osd_dev(d);
3917 switch(cfg->lcfg_command) {
3919 err = osd_mount(env, o, cfg);
3922 err = osd_shutdown(env, o);
3931 static int osd_recovery_complete(const struct lu_env *env,
3932 struct lu_device *d)
3937 static int osd_prepare(const struct lu_env *env,
3938 struct lu_device *pdev,
3939 struct lu_device *dev)
3941 struct osd_device *osd = osd_dev(dev);
3942 struct lustre_sb_info *lsi;
3943 struct lustre_disk_data *ldd;
3944 struct lustre_mount_info *lmi;
3945 struct osd_thread_info *oti = osd_oti_get(env);
3946 struct dt_object *d;
3950 /* 1. initialize oi before any file create or file open */
3951 result = osd_oi_init(oti, &osd->od_oi,
3952 &osd->od_dt_dev, lu2md_dev(pdev));
3956 lmi = osd->od_mount;
3957 lsi = s2lsi(lmi->lmi_sb);
3960 /* 2. setup local objects */
3961 result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
3965 /* 3. open remote object dir */
3966 d = dt_store_open(env, lu2dt_dev(dev), "",
3967 remote_obj_dir, &oti->oti_fid);
3969 osd->od_obj_area = d;
3972 result = PTR_ERR(d);
3973 osd->od_obj_area = NULL;
3980 static const struct lu_object_operations osd_lu_obj_ops = {
3981 .loo_object_init = osd_object_init,
3982 .loo_object_delete = osd_object_delete,
3983 .loo_object_release = osd_object_release,
3984 .loo_object_free = osd_object_free,
3985 .loo_object_print = osd_object_print,
3986 .loo_object_invariant = osd_object_invariant
3989 static const struct lu_device_operations osd_lu_ops = {
3990 .ldo_object_alloc = osd_object_alloc,
3991 .ldo_process_config = osd_process_config,
3992 .ldo_recovery_complete = osd_recovery_complete,
3993 .ldo_prepare = osd_prepare,
3996 static const struct lu_device_type_operations osd_device_type_ops = {
3997 .ldto_init = osd_type_init,
3998 .ldto_fini = osd_type_fini,
4000 .ldto_start = osd_type_start,
4001 .ldto_stop = osd_type_stop,
4003 .ldto_device_alloc = osd_device_alloc,
4004 .ldto_device_free = osd_device_free,
4006 .ldto_device_init = osd_device_init,
4007 .ldto_device_fini = osd_device_fini
4010 static struct lu_device_type osd_device_type = {
4011 .ldt_tags = LU_DEVICE_DT,
4012 .ldt_name = LUSTRE_OSD_NAME,
4013 .ldt_ops = &osd_device_type_ops,
4014 .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4018 * lprocfs legacy support.
4020 static struct obd_ops osd_obd_device_ops = {
4021 .o_owner = THIS_MODULE
4024 static struct lu_local_obj_desc llod_osd_rem_obj_dir = {
4025 .llod_name = remote_obj_dir,
4026 .llod_oid = OSD_REM_OBJ_DIR_OID,
4028 .llod_feat = &dt_directory_features,
4031 static int __init osd_mod_init(void)
4033 struct lprocfs_static_vars lvars;
4036 llo_local_obj_register(&llod_osd_rem_obj_dir);
4037 lprocfs_osd_init_vars(&lvars);
4038 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4039 LUSTRE_OSD_NAME, &osd_device_type);
4042 static void __exit osd_mod_exit(void)
4044 llo_local_obj_unregister(&llod_osd_rem_obj_dir);
4045 class_unregister_type(LUSTRE_OSD_NAME);
4048 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4049 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4050 MODULE_LICENSE("GPL");
4052 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);