4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_ECHO
39 #include <linux/user_namespace.h>
40 #ifdef HAVE_UIDGID_HEADER
41 # include <linux/uidgid.h>
43 #include <libcfs/libcfs.h>
46 #include <obd_support.h>
47 #include <obd_class.h>
48 #include <lustre_debug.h>
49 #include <lprocfs_status.h>
50 #include <cl_object.h>
51 #include <lustre_fid.h>
52 #include <lustre_acl.h>
53 #include <lustre_ioctl.h>
54 #include <lustre_net.h>
55 #ifdef HAVE_SERVER_SUPPORT
56 # include <md_object.h>
57 #endif /* HAVE_SERVER_SUPPORT */
59 #include "echo_internal.h"
61 /** \defgroup echo_client Echo Client
66 struct cl_device ed_cl;
67 struct echo_client_obd *ed_ec;
69 struct cl_site ed_site_myself;
70 struct cl_site *ed_site;
71 struct lu_device *ed_next;
73 struct lu_client_seq *ed_cl_seq;
74 #ifdef HAVE_SERVER_SUPPORT
75 struct local_oid_storage *ed_los;
76 struct lu_fid ed_root_fid;
77 #endif /* HAVE_SERVER_SUPPORT */
81 struct cl_object eo_cl;
82 struct cl_object_header eo_hdr;
83 struct echo_device *eo_dev;
84 struct list_head eo_obj_chain;
85 struct lov_oinfo *eo_oinfo;
90 struct echo_object_conf {
91 struct cl_object_conf eoc_cl;
92 struct lov_oinfo **eoc_oinfo;
96 struct cl_page_slice ep_cl;
101 struct cl_lock_slice el_cl;
102 struct list_head el_chain;
103 struct echo_object *el_object;
105 atomic_t el_refcount;
108 #ifdef HAVE_SERVER_SUPPORT
109 static const char echo_md_root_dir_name[] = "ROOT_ECHO";
112 * In order to use the values of members in struct mdd_device,
113 * we define an alias structure here.
115 struct echo_md_device {
116 struct md_device emd_md_dev;
117 struct obd_export *emd_child_exp;
118 struct dt_device *emd_child;
119 struct dt_device *emd_bottom;
120 struct lu_fid emd_root_fid;
121 struct lu_fid emd_local_root_fid;
123 #endif /* HAVE_SERVER_SUPPORT */
125 static int echo_client_setup(const struct lu_env *env,
126 struct obd_device *obddev,
127 struct lustre_cfg *lcfg);
128 static int echo_client_cleanup(struct obd_device *obddev);
131 /** \defgroup echo_helpers Helper functions
134 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
136 return container_of0(dev, struct echo_device, ed_cl);
139 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
144 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
146 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
149 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
154 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
156 return container_of(o, struct echo_object, eo_cl);
159 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
161 return container_of(s, struct echo_page, ep_cl);
164 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
166 return container_of(s, struct echo_lock, el_cl);
169 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
171 return ecl->el_cl.cls_lock;
174 static struct lu_context_key echo_thread_key;
175 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
177 struct echo_thread_info *info;
178 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
179 LASSERT(info != NULL);
184 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
186 return container_of(c, struct echo_object_conf, eoc_cl);
189 #ifdef HAVE_SERVER_SUPPORT
190 static inline struct echo_md_device *lu2emd_dev(struct lu_device *d)
192 return container_of0(d, struct echo_md_device, emd_md_dev.md_lu_dev);
195 static inline struct lu_device *emd2lu_dev(struct echo_md_device *d)
197 return &d->emd_md_dev.md_lu_dev;
200 static inline struct seq_server_site *echo_md_seq_site(struct echo_md_device *d)
202 return emd2lu_dev(d)->ld_site->ld_seq_site;
205 static inline struct obd_device *emd2obd_dev(struct echo_md_device *d)
207 return d->emd_md_dev.md_lu_dev.ld_obd;
209 #endif /* HAVE_SERVER_SUPPORT */
211 /** @} echo_helpers */
213 static int cl_echo_object_put(struct echo_object *eco);
214 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
215 struct page **pages, int npages, int async);
217 struct echo_thread_info {
218 struct echo_object_conf eti_conf;
219 struct lustre_md eti_md;
221 struct cl_2queue eti_queue;
223 struct cl_lock eti_lock;
224 struct lu_fid eti_fid;
225 struct lu_fid eti_fid2;
226 #ifdef HAVE_SERVER_SUPPORT
227 struct md_op_spec eti_spec;
228 struct lov_mds_md_v3 eti_lmm;
229 struct lov_user_md_v3 eti_lum;
230 struct md_attr eti_ma;
231 struct lu_name eti_lname;
232 /* per-thread values, can be re-used */
236 struct lu_buf eti_buf;
237 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
241 /* No session used right now */
242 struct echo_session_info {
246 static struct kmem_cache *echo_lock_kmem;
247 static struct kmem_cache *echo_object_kmem;
248 static struct kmem_cache *echo_thread_kmem;
249 static struct kmem_cache *echo_session_kmem;
250 /* static struct kmem_cache *echo_req_kmem; */
252 static struct lu_kmem_descr echo_caches[] = {
254 .ckd_cache = &echo_lock_kmem,
255 .ckd_name = "echo_lock_kmem",
256 .ckd_size = sizeof (struct echo_lock)
259 .ckd_cache = &echo_object_kmem,
260 .ckd_name = "echo_object_kmem",
261 .ckd_size = sizeof (struct echo_object)
264 .ckd_cache = &echo_thread_kmem,
265 .ckd_name = "echo_thread_kmem",
266 .ckd_size = sizeof (struct echo_thread_info)
269 .ckd_cache = &echo_session_kmem,
270 .ckd_name = "echo_session_kmem",
271 .ckd_size = sizeof (struct echo_session_info)
278 /** \defgroup echo_page Page operations
280 * Echo page operations.
284 static int echo_page_own(const struct lu_env *env,
285 const struct cl_page_slice *slice,
286 struct cl_io *io, int nonblock)
288 struct echo_page *ep = cl2echo_page(slice);
291 mutex_lock(&ep->ep_lock);
292 else if (!mutex_trylock(&ep->ep_lock))
297 static void echo_page_disown(const struct lu_env *env,
298 const struct cl_page_slice *slice,
301 struct echo_page *ep = cl2echo_page(slice);
303 LASSERT(mutex_is_locked(&ep->ep_lock));
304 mutex_unlock(&ep->ep_lock);
307 static void echo_page_discard(const struct lu_env *env,
308 const struct cl_page_slice *slice,
309 struct cl_io *unused)
311 cl_page_delete(env, slice->cpl_page);
314 static int echo_page_is_vmlocked(const struct lu_env *env,
315 const struct cl_page_slice *slice)
317 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
322 static void echo_page_completion(const struct lu_env *env,
323 const struct cl_page_slice *slice,
326 LASSERT(slice->cpl_page->cp_sync_io != NULL);
329 static void echo_page_fini(const struct lu_env *env,
330 struct cl_page_slice *slice)
332 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
335 atomic_dec(&eco->eo_npages);
336 page_cache_release(slice->cpl_page->cp_vmpage);
340 static int echo_page_prep(const struct lu_env *env,
341 const struct cl_page_slice *slice,
342 struct cl_io *unused)
347 static int echo_page_print(const struct lu_env *env,
348 const struct cl_page_slice *slice,
349 void *cookie, lu_printer_t printer)
351 struct echo_page *ep = cl2echo_page(slice);
353 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
354 ep, mutex_is_locked(&ep->ep_lock),
355 slice->cpl_page->cp_vmpage);
359 static const struct cl_page_operations echo_page_ops = {
360 .cpo_own = echo_page_own,
361 .cpo_disown = echo_page_disown,
362 .cpo_discard = echo_page_discard,
363 .cpo_fini = echo_page_fini,
364 .cpo_print = echo_page_print,
365 .cpo_is_vmlocked = echo_page_is_vmlocked,
368 .cpo_prep = echo_page_prep,
369 .cpo_completion = echo_page_completion,
372 .cpo_prep = echo_page_prep,
373 .cpo_completion = echo_page_completion,
379 /** \defgroup echo_lock Locking
381 * echo lock operations
385 static void echo_lock_fini(const struct lu_env *env,
386 struct cl_lock_slice *slice)
388 struct echo_lock *ecl = cl2echo_lock(slice);
390 LASSERT(list_empty(&ecl->el_chain));
391 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
394 static struct cl_lock_operations echo_lock_ops = {
395 .clo_fini = echo_lock_fini,
400 /** \defgroup echo_cl_ops cl_object operations
402 * operations for cl_object
406 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
407 struct cl_page *page, pgoff_t index)
409 struct echo_page *ep = cl_object_page_slice(obj, page);
410 struct echo_object *eco = cl2echo_obj(obj);
413 page_cache_get(page->cp_vmpage);
414 mutex_init(&ep->ep_lock);
415 cl_page_slice_add(page, &ep->ep_cl, obj, index, &echo_page_ops);
416 atomic_inc(&eco->eo_npages);
420 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
426 static int echo_lock_init(const struct lu_env *env,
427 struct cl_object *obj, struct cl_lock *lock,
428 const struct cl_io *unused)
430 struct echo_lock *el;
433 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, GFP_NOFS);
435 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
436 el->el_object = cl2echo_obj(obj);
437 INIT_LIST_HEAD(&el->el_chain);
438 atomic_set(&el->el_refcount, 0);
440 RETURN(el == NULL ? -ENOMEM : 0);
443 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
444 const struct cl_object_conf *conf)
449 static const struct cl_object_operations echo_cl_obj_ops = {
450 .coo_page_init = echo_page_init,
451 .coo_lock_init = echo_lock_init,
452 .coo_io_init = echo_io_init,
453 .coo_conf_set = echo_conf_set
455 /** @} echo_cl_ops */
457 /** \defgroup echo_lu_ops lu_object operations
459 * operations for echo lu object.
463 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
464 const struct lu_object_conf *conf)
466 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
467 struct echo_client_obd *ec = ed->ed_ec;
468 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
472 struct lu_object *below;
473 struct lu_device *under;
476 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
480 lu_object_add(obj, below);
483 if (!ed->ed_next_ismd) {
484 const struct cl_object_conf *cconf = lu2cl_conf(conf);
485 struct echo_object_conf *econf = cl2echo_conf(cconf);
487 LASSERT(econf->eoc_oinfo != NULL);
489 /* Transfer the oinfo pointer to eco that it won't be
491 eco->eo_oinfo = *econf->eoc_oinfo;
492 *econf->eoc_oinfo = NULL;
494 eco->eo_oinfo = NULL;
498 atomic_set(&eco->eo_npages, 0);
499 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
501 spin_lock(&ec->ec_lock);
502 list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
503 spin_unlock(&ec->ec_lock);
508 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
510 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
511 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
514 LASSERT(atomic_read(&eco->eo_npages) == 0);
516 spin_lock(&ec->ec_lock);
517 list_del_init(&eco->eo_obj_chain);
518 spin_unlock(&ec->ec_lock);
521 lu_object_header_fini(obj->lo_header);
523 if (eco->eo_oinfo != NULL)
524 OBD_FREE_PTR(eco->eo_oinfo);
526 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
530 static int echo_object_print(const struct lu_env *env, void *cookie,
531 lu_printer_t p, const struct lu_object *o)
533 struct echo_object *obj = cl2echo_obj(lu2cl(o));
535 return (*p)(env, cookie, "echoclient-object@%p", obj);
538 static const struct lu_object_operations echo_lu_obj_ops = {
539 .loo_object_init = echo_object_init,
540 .loo_object_delete = NULL,
541 .loo_object_release = NULL,
542 .loo_object_free = echo_object_free,
543 .loo_object_print = echo_object_print,
544 .loo_object_invariant = NULL
546 /** @} echo_lu_ops */
548 /** \defgroup echo_lu_dev_ops lu_device operations
550 * Operations for echo lu device.
554 static struct lu_object *echo_object_alloc(const struct lu_env *env,
555 const struct lu_object_header *hdr,
556 struct lu_device *dev)
558 struct echo_object *eco;
559 struct lu_object *obj = NULL;
562 /* we're the top dev. */
563 LASSERT(hdr == NULL);
564 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, GFP_NOFS);
566 struct cl_object_header *hdr = &eco->eo_hdr;
568 obj = &echo_obj2cl(eco)->co_lu;
569 cl_object_header_init(hdr);
570 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
572 lu_object_init(obj, &hdr->coh_lu, dev);
573 lu_object_add_top(&hdr->coh_lu, obj);
575 eco->eo_cl.co_ops = &echo_cl_obj_ops;
576 obj->lo_ops = &echo_lu_obj_ops;
581 static struct lu_device_operations echo_device_lu_ops = {
582 .ldo_object_alloc = echo_object_alloc,
585 /** @} echo_lu_dev_ops */
587 static struct cl_device_operations echo_device_cl_ops = {
590 /** \defgroup echo_init Setup and teardown
592 * Init and fini functions for echo client.
596 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
598 struct cl_site *site = &ed->ed_site_myself;
601 /* initialize site */
602 rc = cl_site_init(site, &ed->ed_cl);
604 CERROR("Cannot initialize site for echo client(%d)\n", rc);
608 rc = lu_site_init_finish(&site->cs_lu);
616 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
619 if (!ed->ed_next_ismd)
620 cl_site_fini(ed->ed_site);
625 static void *echo_thread_key_init(const struct lu_context *ctx,
626 struct lu_context_key *key)
628 struct echo_thread_info *info;
630 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, GFP_NOFS);
632 info = ERR_PTR(-ENOMEM);
636 static void echo_thread_key_fini(const struct lu_context *ctx,
637 struct lu_context_key *key, void *data)
639 struct echo_thread_info *info = data;
640 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
643 static void echo_thread_key_exit(const struct lu_context *ctx,
644 struct lu_context_key *key, void *data)
648 static struct lu_context_key echo_thread_key = {
649 .lct_tags = LCT_CL_THREAD,
650 .lct_init = echo_thread_key_init,
651 .lct_fini = echo_thread_key_fini,
652 .lct_exit = echo_thread_key_exit
655 static void *echo_session_key_init(const struct lu_context *ctx,
656 struct lu_context_key *key)
658 struct echo_session_info *session;
660 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, GFP_NOFS);
662 session = ERR_PTR(-ENOMEM);
666 static void echo_session_key_fini(const struct lu_context *ctx,
667 struct lu_context_key *key, void *data)
669 struct echo_session_info *session = data;
670 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
673 static void echo_session_key_exit(const struct lu_context *ctx,
674 struct lu_context_key *key, void *data)
678 static struct lu_context_key echo_session_key = {
679 .lct_tags = LCT_SESSION,
680 .lct_init = echo_session_key_init,
681 .lct_fini = echo_session_key_fini,
682 .lct_exit = echo_session_key_exit
685 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
687 #ifdef HAVE_SERVER_SUPPORT
688 # define ECHO_SEQ_WIDTH 0xffffffff
689 static int echo_fid_init(struct echo_device *ed, char *obd_name,
690 struct seq_server_site *ss)
696 OBD_ALLOC_PTR(ed->ed_cl_seq);
697 if (ed->ed_cl_seq == NULL)
700 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
702 GOTO(out_free_seq, rc = -ENOMEM);
704 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
706 /* Init client side sequence-manager */
707 rc = seq_client_init(ed->ed_cl_seq, NULL,
709 prefix, ss->ss_server_seq);
710 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
711 OBD_FREE(prefix, MAX_OBD_NAME + 5);
713 GOTO(out_free_seq, rc);
718 OBD_FREE_PTR(ed->ed_cl_seq);
719 ed->ed_cl_seq = NULL;
723 static int echo_fid_fini(struct obd_device *obddev)
725 struct echo_device *ed = obd2echo_dev(obddev);
728 if (ed->ed_cl_seq != NULL) {
729 seq_client_fini(ed->ed_cl_seq);
730 OBD_FREE_PTR(ed->ed_cl_seq);
731 ed->ed_cl_seq = NULL;
737 static void echo_ed_los_fini(const struct lu_env *env, struct echo_device *ed)
741 if (ed != NULL && ed->ed_next_ismd && ed->ed_los != NULL) {
742 local_oid_storage_fini(env, ed->ed_los);
748 echo_md_local_file_create(const struct lu_env *env, struct echo_md_device *emd,
749 struct local_oid_storage *los,
750 const struct lu_fid *pfid, const char *name,
751 __u32 mode, struct lu_fid *fid)
753 struct dt_object *parent = NULL;
754 struct dt_object *dto = NULL;
758 LASSERT(!fid_is_zero(pfid));
759 parent = dt_locate(env, emd->emd_bottom, pfid);
760 if (unlikely(IS_ERR(parent)))
761 RETURN(PTR_ERR(parent));
763 /* create local file with @fid */
764 dto = local_file_find_or_create_with_fid(env, emd->emd_bottom, fid,
767 GOTO(out_put, rc = PTR_ERR(dto));
769 *fid = *lu_object_fid(&dto->do_lu);
770 /* since stack is not fully set up the local_storage uses own stack
771 * and we should drop its object from cache */
772 lu_object_put_nocache(env, &dto->do_lu);
776 lu_object_put(env, &parent->do_lu);
781 echo_md_root_get(const struct lu_env *env, struct echo_md_device *emd,
782 struct echo_device *ed)
788 /* Setup local dirs */
789 fid.f_seq = FID_SEQ_LOCAL_NAME;
792 rc = local_oid_storage_init(env, emd->emd_bottom, &fid, &ed->ed_los);
796 lu_echo_root_fid(&fid);
797 if (echo_md_seq_site(emd)->ss_node_id == 0) {
798 rc = echo_md_local_file_create(env, emd, ed->ed_los,
799 &emd->emd_local_root_fid,
800 echo_md_root_dir_name, S_IFDIR |
801 S_IRUGO | S_IWUSR | S_IXUGO,
804 CERROR("%s: create md echo root fid failed: rc = %d\n",
805 emd2obd_dev(emd)->obd_name, rc);
809 ed->ed_root_fid = fid;
813 echo_ed_los_fini(env, ed);
817 #endif /* HAVE_SERVER_SUPPORT */
819 static struct lu_device *echo_device_alloc(const struct lu_env *env,
820 struct lu_device_type *t,
821 struct lustre_cfg *cfg)
823 struct lu_device *next;
824 struct echo_device *ed;
825 struct cl_device *cd;
826 struct obd_device *obd = NULL; /* to keep compiler happy */
827 struct obd_device *tgt;
828 const char *tgt_type_name;
835 GOTO(out, rc = -ENOMEM);
839 rc = cl_device_init(cd, t);
843 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
844 cd->cd_ops = &echo_device_cl_ops;
847 obd = class_name2obd(lustre_cfg_string(cfg, 0));
848 LASSERT(obd != NULL);
849 LASSERT(env != NULL);
851 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
853 CERROR("Can not find tgt device %s\n",
854 lustre_cfg_string(cfg, 1));
855 GOTO(out, rc = -ENODEV);
858 next = tgt->obd_lu_dev;
860 if (strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME) == 0) {
861 ed->ed_next_ismd = 1;
862 } else if (strcmp(tgt->obd_type->typ_name, LUSTRE_OST_NAME) == 0 ||
863 strcmp(tgt->obd_type->typ_name, LUSTRE_OSC_NAME) == 0) {
864 ed->ed_next_ismd = 0;
865 rc = echo_site_init(env, ed);
869 GOTO(out, rc = -EINVAL);
874 rc = echo_client_setup(env, obd, cfg);
878 ed->ed_ec = &obd->u.echo_client;
881 if (ed->ed_next_ismd) {
882 #ifdef HAVE_SERVER_SUPPORT
883 /* Suppose to connect to some Metadata layer */
884 struct lu_site *ls = NULL;
885 struct lu_device *ld = NULL;
886 struct md_device *md = NULL;
887 struct echo_md_device *emd = NULL;
891 CERROR("%s is not lu device type!\n",
892 lustre_cfg_string(cfg, 1));
893 GOTO(out, rc = -EINVAL);
896 tgt_type_name = lustre_cfg_string(cfg, 2);
897 if (!tgt_type_name) {
898 CERROR("%s no type name for echo %s setup\n",
899 lustre_cfg_string(cfg, 1),
900 tgt->obd_type->typ_name);
901 GOTO(out, rc = -EINVAL);
906 spin_lock(&ls->ls_ld_lock);
907 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
908 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
913 spin_unlock(&ls->ls_ld_lock);
916 CERROR("%s is not lu device type!\n",
917 lustre_cfg_string(cfg, 1));
918 GOTO(out, rc = -EINVAL);
922 /* For MD echo client, it will use the site in MDS stack */
923 ed->ed_site_myself.cs_lu = *ls;
924 ed->ed_site = &ed->ed_site_myself;
925 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
926 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
928 CERROR("echo fid init error %d\n", rc);
932 md = lu2md_dev(next);
933 emd = lu2emd_dev(&md->md_lu_dev);
934 rc = echo_md_root_get(env, emd, ed);
936 CERROR("%s: get root error: rc = %d\n",
937 emd2obd_dev(emd)->obd_name, rc);
940 #else /* !HAVE_SERVER_SUPPORT */
941 CERROR("Local operations are NOT supported on client side. "
942 "Only remote operations are supported. Metadata client "
943 "must be run on server side.\n");
944 GOTO(out, rc = -EOPNOTSUPP);
945 #endif /* HAVE_SERVER_SUPPORT */
947 /* if echo client is to be stacked upon ost device, the next is
948 * NULL since ost is not a clio device so far */
949 if (next != NULL && !lu_device_is_cl(next))
952 tgt_type_name = tgt->obd_type->typ_name;
954 LASSERT(next != NULL);
955 if (next->ld_site != NULL)
956 GOTO(out, rc = -EBUSY);
958 next->ld_site = &ed->ed_site->cs_lu;
959 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
960 next->ld_type->ldt_name,
965 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
969 RETURN(&cd->cd_lu_dev);
974 rc2 = echo_client_cleanup(obd);
976 CERROR("Cleanup obd device %s error(%d)\n",
981 echo_site_fini(env, ed);
983 cl_device_fini(&ed->ed_cl);
993 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
994 const char *name, struct lu_device *next)
1000 static struct lu_device *echo_device_fini(const struct lu_env *env,
1001 struct lu_device *d)
1003 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
1004 struct lu_device *next = ed->ed_next;
1006 while (next && !ed->ed_next_ismd)
1007 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
1011 static void echo_lock_release(const struct lu_env *env,
1012 struct echo_lock *ecl,
1015 struct cl_lock *clk = echo_lock2cl(ecl);
1017 cl_lock_release(env, clk);
1020 static struct lu_device *echo_device_free(const struct lu_env *env,
1021 struct lu_device *d)
1023 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
1024 struct echo_client_obd *ec = ed->ed_ec;
1025 struct echo_object *eco;
1026 struct lu_device *next = ed->ed_next;
1028 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1031 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1033 /* check if there are objects still alive.
1034 * It shouldn't have any object because lu_site_purge would cleanup
1035 * all of cached objects. Anyway, probably the echo device is being
1036 * parallelly accessed.
1038 spin_lock(&ec->ec_lock);
1039 list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1040 eco->eo_deleted = 1;
1041 spin_unlock(&ec->ec_lock);
1044 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1047 "Waiting for the reference of echo object to be dropped\n");
1049 /* Wait for the last reference to be dropped. */
1050 spin_lock(&ec->ec_lock);
1051 while (!list_empty(&ec->ec_objects)) {
1052 spin_unlock(&ec->ec_lock);
1053 CERROR("echo_client still has objects at cleanup time, "
1054 "wait for 1 second\n");
1055 set_current_state(TASK_UNINTERRUPTIBLE);
1056 schedule_timeout(cfs_time_seconds(1));
1057 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1058 spin_lock(&ec->ec_lock);
1060 spin_unlock(&ec->ec_lock);
1062 LASSERT(list_empty(&ec->ec_locks));
1064 CDEBUG(D_INFO, "No object exists, exiting...\n");
1066 echo_client_cleanup(d->ld_obd);
1067 #ifdef HAVE_SERVER_SUPPORT
1068 echo_fid_fini(d->ld_obd);
1069 echo_ed_los_fini(env, ed);
1071 while (next && !ed->ed_next_ismd)
1072 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1074 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1075 echo_site_fini(env, ed);
1076 cl_device_fini(&ed->ed_cl);
1082 static const struct lu_device_type_operations echo_device_type_ops = {
1083 .ldto_init = echo_type_init,
1084 .ldto_fini = echo_type_fini,
1086 .ldto_start = echo_type_start,
1087 .ldto_stop = echo_type_stop,
1089 .ldto_device_alloc = echo_device_alloc,
1090 .ldto_device_free = echo_device_free,
1091 .ldto_device_init = echo_device_init,
1092 .ldto_device_fini = echo_device_fini
1095 static struct lu_device_type echo_device_type = {
1096 .ldt_tags = LU_DEVICE_CL,
1097 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1098 .ldt_ops = &echo_device_type_ops,
1099 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1103 /** \defgroup echo_exports Exported operations
1105 * exporting functions to echo client
1110 /* Interfaces to echo client obd device */
1111 static struct echo_object *
1112 cl_echo_object_find(struct echo_device *d, const struct ost_id *oi)
1115 struct echo_thread_info *info;
1116 struct echo_object_conf *conf;
1117 struct echo_object *eco;
1118 struct cl_object *obj;
1119 struct lov_oinfo *oinfo = NULL;
1125 LASSERTF(ostid_id(oi) != 0, DOSTID"\n", POSTID(oi));
1126 LASSERTF(ostid_seq(oi) == FID_SEQ_ECHO, DOSTID"\n", POSTID(oi));
1128 /* Never return an object if the obd is to be freed. */
1129 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1130 RETURN(ERR_PTR(-ENODEV));
1132 env = cl_env_get(&refcheck);
1134 RETURN((void *)env);
1136 info = echo_env_info(env);
1137 conf = &info->eti_conf;
1139 OBD_ALLOC_PTR(oinfo);
1141 GOTO(out, eco = ERR_PTR(-ENOMEM));
1143 oinfo->loi_oi = *oi;
1144 conf->eoc_cl.u.coc_oinfo = oinfo;
1147 /* If echo_object_init() is successful then ownership of oinfo
1148 * is transferred to the object. */
1149 conf->eoc_oinfo = &oinfo;
1151 fid = &info->eti_fid;
1152 rc = ostid_to_fid(fid, oi, 0);
1154 GOTO(out, eco = ERR_PTR(rc));
1156 /* In the function below, .hs_keycmp resolves to
1157 * lu_obj_hop_keycmp() */
1158 /* coverity[overrun-buffer-val] */
1159 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1161 GOTO(out, eco = (void*)obj);
1163 eco = cl2echo_obj(obj);
1164 if (eco->eo_deleted) {
1165 cl_object_put(env, obj);
1166 eco = ERR_PTR(-EAGAIN);
1171 OBD_FREE_PTR(oinfo);
1173 cl_env_put(env, &refcheck);
1177 static int cl_echo_object_put(struct echo_object *eco)
1180 struct cl_object *obj = echo_obj2cl(eco);
1184 env = cl_env_get(&refcheck);
1186 RETURN(PTR_ERR(env));
1188 /* an external function to kill an object? */
1189 if (eco->eo_deleted) {
1190 struct lu_object_header *loh = obj->co_lu.lo_header;
1191 LASSERT(&eco->eo_hdr == luh2coh(loh));
1192 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1195 cl_object_put(env, obj);
1196 cl_env_put(env, &refcheck);
1200 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1201 u64 start, u64 end, int mode,
1202 __u64 *cookie , __u32 enqflags)
1205 struct cl_lock *lck;
1206 struct cl_object *obj;
1207 struct cl_lock_descr *descr;
1208 struct echo_thread_info *info;
1212 info = echo_env_info(env);
1214 lck = &info->eti_lock;
1215 obj = echo_obj2cl(eco);
1217 memset(lck, 0, sizeof(*lck));
1218 descr = &lck->cll_descr;
1219 descr->cld_obj = obj;
1220 descr->cld_start = cl_index(obj, start);
1221 descr->cld_end = cl_index(obj, end);
1222 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1223 descr->cld_enq_flags = enqflags;
1226 rc = cl_lock_request(env, io, lck);
1228 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1229 struct echo_lock *el;
1231 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1232 spin_lock(&ec->ec_lock);
1233 if (list_empty(&el->el_chain)) {
1234 list_add(&el->el_chain, &ec->ec_locks);
1235 el->el_cookie = ++ec->ec_unique;
1237 atomic_inc(&el->el_refcount);
1238 *cookie = el->el_cookie;
1239 spin_unlock(&ec->ec_lock);
1244 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1247 struct echo_client_obd *ec = ed->ed_ec;
1248 struct echo_lock *ecl = NULL;
1249 struct list_head *el;
1250 int found = 0, still_used = 0;
1253 LASSERT(ec != NULL);
1254 spin_lock(&ec->ec_lock);
1255 list_for_each(el, &ec->ec_locks) {
1256 ecl = list_entry(el, struct echo_lock, el_chain);
1257 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1258 found = (ecl->el_cookie == cookie);
1260 if (atomic_dec_and_test(&ecl->el_refcount))
1261 list_del_init(&ecl->el_chain);
1267 spin_unlock(&ec->ec_lock);
1272 echo_lock_release(env, ecl, still_used);
1276 static void echo_commit_callback(const struct lu_env *env, struct cl_io *io,
1277 struct cl_page *page)
1279 struct echo_thread_info *info;
1280 struct cl_2queue *queue;
1282 info = echo_env_info(env);
1283 LASSERT(io == &info->eti_io);
1285 queue = &info->eti_queue;
1286 cl_page_list_add(&queue->c2_qout, page);
1289 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
1290 struct page **pages, int npages, int async)
1293 struct echo_thread_info *info;
1294 struct cl_object *obj = echo_obj2cl(eco);
1295 struct echo_device *ed = eco->eo_dev;
1296 struct cl_2queue *queue;
1298 struct cl_page *clp;
1299 struct lustre_handle lh = { 0 };
1300 int page_size = cl_page_size(obj);
1306 LASSERT((offset & ~PAGE_MASK) == 0);
1307 LASSERT(ed->ed_next != NULL);
1308 env = cl_env_get(&refcheck);
1310 RETURN(PTR_ERR(env));
1312 info = echo_env_info(env);
1314 queue = &info->eti_queue;
1316 cl_2queue_init(queue);
1318 io->ci_ignore_layout = 1;
1319 rc = cl_io_init(env, io, CIT_MISC, obj);
1325 rc = cl_echo_enqueue0(env, eco, offset,
1326 offset + npages * PAGE_CACHE_SIZE - 1,
1327 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1330 GOTO(error_lock, rc);
1332 for (i = 0; i < npages; i++) {
1334 clp = cl_page_find(env, obj, cl_index(obj, offset),
1335 pages[i], CPT_TRANSIENT);
1340 LASSERT(clp->cp_type == CPT_TRANSIENT);
1342 rc = cl_page_own(env, io, clp);
1344 LASSERT(clp->cp_state == CPS_FREEING);
1345 cl_page_put(env, clp);
1349 cl_2queue_add(queue, clp);
1351 /* drop the reference count for cl_page_find, so that the page
1352 * will be freed in cl_2queue_fini. */
1353 cl_page_put(env, clp);
1354 cl_page_clip(env, clp, 0, page_size);
1356 offset += page_size;
1360 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1362 async = async && (typ == CRT_WRITE);
1364 rc = cl_io_commit_async(env, io, &queue->c2_qin,
1366 echo_commit_callback);
1368 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1369 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1370 async ? "async" : "sync", rc);
1373 cl_echo_cancel0(env, ed, lh.cookie);
1376 cl_2queue_discard(env, io, queue);
1377 cl_2queue_disown(env, io, queue);
1378 cl_2queue_fini(env, queue);
1379 cl_io_fini(env, io);
1381 cl_env_put(env, &refcheck);
1384 /** @} echo_exports */
1387 static u64 last_object_id;
1389 #ifdef HAVE_SERVER_SUPPORT
1390 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1393 sprintf(name, LPU64, id);
1394 lname->ln_name = name;
1395 lname->ln_namelen = strlen(name);
1398 /* similar to mdt_attr_get_complex */
1399 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1402 struct echo_thread_info *info = echo_env_info(env);
1407 LASSERT(ma->ma_lmm_size > 0);
1409 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1413 /* big_lmm may need to be grown */
1414 if (info->eti_big_lmmsize < rc) {
1415 int size = size_roundup_power2(rc);
1417 if (info->eti_big_lmmsize > 0) {
1418 /* free old buffer */
1419 LASSERT(info->eti_big_lmm);
1420 OBD_FREE_LARGE(info->eti_big_lmm,
1421 info->eti_big_lmmsize);
1422 info->eti_big_lmm = NULL;
1423 info->eti_big_lmmsize = 0;
1426 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1427 if (info->eti_big_lmm == NULL)
1429 info->eti_big_lmmsize = size;
1431 LASSERT(info->eti_big_lmmsize >= rc);
1433 info->eti_buf.lb_buf = info->eti_big_lmm;
1434 info->eti_buf.lb_len = info->eti_big_lmmsize;
1435 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1439 ma->ma_valid |= MA_LOV;
1440 ma->ma_lmm = info->eti_big_lmm;
1441 ma->ma_lmm_size = rc;
1446 static int echo_attr_get_complex(const struct lu_env *env,
1447 struct md_object *next,
1450 struct echo_thread_info *info = echo_env_info(env);
1451 struct lu_buf *buf = &info->eti_buf;
1452 umode_t mode = lu_object_attr(&next->mo_lu);
1453 int need = ma->ma_need;
1460 if (need & MA_INODE) {
1461 ma->ma_need = MA_INODE;
1462 rc = mo_attr_get(env, next, ma);
1465 ma->ma_valid |= MA_INODE;
1468 if (need & MA_LOV) {
1469 if (S_ISREG(mode) || S_ISDIR(mode)) {
1470 LASSERT(ma->ma_lmm_size > 0);
1471 buf->lb_buf = ma->ma_lmm;
1472 buf->lb_len = ma->ma_lmm_size;
1473 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1475 ma->ma_lmm_size = rc2;
1476 ma->ma_valid |= MA_LOV;
1477 } else if (rc2 == -ENODATA) {
1479 ma->ma_lmm_size = 0;
1480 } else if (rc2 == -ERANGE) {
1481 rc2 = echo_big_lmm_get(env, next, ma);
1483 GOTO(out, rc = rc2);
1485 GOTO(out, rc = rc2);
1490 #ifdef CONFIG_FS_POSIX_ACL
1491 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1492 buf->lb_buf = ma->ma_acl;
1493 buf->lb_len = ma->ma_acl_size;
1494 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1496 ma->ma_acl_size = rc2;
1497 ma->ma_valid |= MA_ACL_DEF;
1498 } else if (rc2 == -ENODATA) {
1500 ma->ma_acl_size = 0;
1502 GOTO(out, rc = rc2);
1508 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1509 rc, ma->ma_valid, ma->ma_lmm);
1514 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1515 struct md_object *parent, struct lu_fid *fid,
1516 struct lu_name *lname, struct md_op_spec *spec,
1519 struct lu_object *ec_child, *child;
1520 struct lu_device *ld = ed->ed_next;
1521 struct echo_thread_info *info = echo_env_info(env);
1522 struct lu_fid *fid2 = &info->eti_fid2;
1523 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1528 rc = mdo_lookup(env, parent, lname, fid2, spec);
1531 else if (rc != -ENOENT)
1534 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1536 if (IS_ERR(ec_child)) {
1537 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1539 RETURN(PTR_ERR(ec_child));
1542 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1543 if (child == NULL) {
1544 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1545 GOTO(out_put, rc = -EINVAL);
1548 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1549 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1552 * Do not perform lookup sanity check. We know that name does not exist.
1554 spec->sp_cr_lookup = 0;
1555 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1557 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1560 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1561 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1564 lu_object_put(env, ec_child);
1568 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1571 struct echo_thread_info *info = echo_env_info(env);
1573 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1574 ma->ma_lmm = (void *)&info->eti_lmm;
1575 ma->ma_lmm_size = sizeof(info->eti_lmm);
1577 LASSERT(info->eti_big_lmmsize);
1578 ma->ma_lmm = info->eti_big_lmm;
1579 ma->ma_lmm_size = info->eti_big_lmmsize;
1585 static int echo_create_md_object(const struct lu_env *env,
1586 struct echo_device *ed,
1587 struct lu_object *ec_parent,
1589 char *name, int namelen,
1590 __u64 id, __u32 mode, int count,
1591 int stripe_count, int stripe_offset)
1593 struct lu_object *parent;
1594 struct echo_thread_info *info = echo_env_info(env);
1595 struct lu_name *lname = &info->eti_lname;
1596 struct md_op_spec *spec = &info->eti_spec;
1597 struct md_attr *ma = &info->eti_ma;
1598 struct lu_device *ld = ed->ed_next;
1604 if (ec_parent == NULL)
1606 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1610 memset(ma, 0, sizeof(*ma));
1611 memset(spec, 0, sizeof(*spec));
1612 if (stripe_count != 0) {
1613 spec->sp_cr_flags |= FMODE_WRITE;
1614 echo_set_lmm_size(env, ld, ma);
1615 if (stripe_count != -1) {
1616 struct lov_user_md_v3 *lum = &info->eti_lum;
1618 lum->lmm_magic = LOV_USER_MAGIC_V3;
1619 lum->lmm_stripe_count = stripe_count;
1620 lum->lmm_stripe_offset = stripe_offset;
1621 lum->lmm_pattern = 0;
1622 spec->u.sp_ea.eadata = lum;
1623 spec->u.sp_ea.eadatalen = sizeof(*lum);
1624 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1628 ma->ma_attr.la_mode = mode;
1629 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1630 ma->ma_attr.la_ctime = cfs_time_current_64();
1633 lname->ln_name = name;
1634 lname->ln_namelen = namelen;
1635 /* If name is specified, only create one object by name */
1636 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1641 /* Create multiple object sequenced by id */
1642 for (i = 0; i < count; i++) {
1643 char *tmp_name = info->eti_name;
1645 echo_md_build_name(lname, tmp_name, id);
1647 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1650 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1661 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1662 struct echo_device *ed,
1663 struct md_object *parent,
1664 struct lu_name *lname)
1666 struct echo_thread_info *info = echo_env_info(env);
1667 struct lu_fid *fid = &info->eti_fid;
1668 struct lu_object *child;
1672 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1674 rc = mdo_lookup(env, parent, lname, fid, NULL);
1676 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1677 RETURN(ERR_PTR(rc));
1680 /* In the function below, .hs_keycmp resolves to
1681 * lu_obj_hop_keycmp() */
1682 /* coverity[overrun-buffer-val] */
1683 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1688 static int echo_setattr_object(const struct lu_env *env,
1689 struct echo_device *ed,
1690 struct lu_object *ec_parent,
1691 __u64 id, int count)
1693 struct lu_object *parent;
1694 struct echo_thread_info *info = echo_env_info(env);
1695 struct lu_name *lname = &info->eti_lname;
1696 char *name = info->eti_name;
1697 struct lu_device *ld = ed->ed_next;
1698 struct lu_buf *buf = &info->eti_buf;
1704 if (ec_parent == NULL)
1706 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1710 for (i = 0; i < count; i++) {
1711 struct lu_object *ec_child, *child;
1713 echo_md_build_name(lname, name, id);
1715 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1716 if (IS_ERR(ec_child)) {
1717 CERROR("Can't find child %s: rc = %ld\n",
1718 lname->ln_name, PTR_ERR(ec_child));
1719 RETURN(PTR_ERR(ec_child));
1722 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1723 if (child == NULL) {
1724 CERROR("Can not locate the child %s\n", lname->ln_name);
1725 lu_object_put(env, ec_child);
1730 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1731 PFID(lu_object_fid(child)));
1733 buf->lb_buf = info->eti_xattr_buf;
1734 buf->lb_len = sizeof(info->eti_xattr_buf);
1736 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1737 rc = mo_xattr_set(env, lu2md(child), buf, name,
1740 CERROR("Can not setattr child "DFID": rc = %d\n",
1741 PFID(lu_object_fid(child)), rc);
1742 lu_object_put(env, ec_child);
1745 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1746 PFID(lu_object_fid(child)));
1748 lu_object_put(env, ec_child);
1753 static int echo_getattr_object(const struct lu_env *env,
1754 struct echo_device *ed,
1755 struct lu_object *ec_parent,
1756 __u64 id, int count)
1758 struct lu_object *parent;
1759 struct echo_thread_info *info = echo_env_info(env);
1760 struct lu_name *lname = &info->eti_lname;
1761 char *name = info->eti_name;
1762 struct md_attr *ma = &info->eti_ma;
1763 struct lu_device *ld = ed->ed_next;
1769 if (ec_parent == NULL)
1771 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1775 memset(ma, 0, sizeof(*ma));
1776 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1777 ma->ma_acl = info->eti_xattr_buf;
1778 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1780 for (i = 0; i < count; i++) {
1781 struct lu_object *ec_child, *child;
1784 echo_md_build_name(lname, name, id);
1785 echo_set_lmm_size(env, ld, ma);
1787 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1788 if (IS_ERR(ec_child)) {
1789 CERROR("Can't find child %s: rc = %ld\n",
1790 lname->ln_name, PTR_ERR(ec_child));
1791 RETURN(PTR_ERR(ec_child));
1794 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1795 if (child == NULL) {
1796 CERROR("Can not locate the child %s\n", lname->ln_name);
1797 lu_object_put(env, ec_child);
1801 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1802 PFID(lu_object_fid(child)));
1803 rc = echo_attr_get_complex(env, lu2md(child), ma);
1805 CERROR("Can not getattr child "DFID": rc = %d\n",
1806 PFID(lu_object_fid(child)), rc);
1807 lu_object_put(env, ec_child);
1810 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1811 PFID(lu_object_fid(child)));
1813 lu_object_put(env, ec_child);
1819 static int echo_lookup_object(const struct lu_env *env,
1820 struct echo_device *ed,
1821 struct lu_object *ec_parent,
1822 __u64 id, int count)
1824 struct lu_object *parent;
1825 struct echo_thread_info *info = echo_env_info(env);
1826 struct lu_name *lname = &info->eti_lname;
1827 char *name = info->eti_name;
1828 struct lu_fid *fid = &info->eti_fid;
1829 struct lu_device *ld = ed->ed_next;
1833 if (ec_parent == NULL)
1835 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1839 /*prepare the requests*/
1840 for (i = 0; i < count; i++) {
1841 echo_md_build_name(lname, name, id);
1843 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1844 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1846 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1848 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1851 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1852 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1859 static int echo_md_destroy_internal(const struct lu_env *env,
1860 struct echo_device *ed,
1861 struct md_object *parent,
1862 struct lu_name *lname,
1865 struct lu_device *ld = ed->ed_next;
1866 struct lu_object *ec_child;
1867 struct lu_object *child;
1872 ec_child = echo_md_lookup(env, ed, parent, lname);
1873 if (IS_ERR(ec_child)) {
1874 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1876 RETURN(PTR_ERR(ec_child));
1879 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1880 if (child == NULL) {
1881 CERROR("Can not locate the child %s\n", lname->ln_name);
1882 GOTO(out_put, rc = -EINVAL);
1885 if (lu_object_remote(child)) {
1886 CERROR("Can not destroy remote object %s: rc = %d\n",
1887 lname->ln_name, -EPERM);
1888 GOTO(out_put, rc = -EPERM);
1890 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1891 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1893 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1895 CERROR("Can not unlink child %s: rc = %d\n",
1896 lname->ln_name, rc);
1899 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1900 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1902 lu_object_put(env, ec_child);
1906 static int echo_destroy_object(const struct lu_env *env,
1907 struct echo_device *ed,
1908 struct lu_object *ec_parent,
1909 char *name, int namelen,
1910 __u64 id, __u32 mode,
1913 struct echo_thread_info *info = echo_env_info(env);
1914 struct lu_name *lname = &info->eti_lname;
1915 struct md_attr *ma = &info->eti_ma;
1916 struct lu_device *ld = ed->ed_next;
1917 struct lu_object *parent;
1922 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1926 memset(ma, 0, sizeof(*ma));
1927 ma->ma_attr.la_mode = mode;
1928 ma->ma_attr.la_valid = LA_CTIME;
1929 ma->ma_attr.la_ctime = cfs_time_current_64();
1930 ma->ma_need = MA_INODE;
1934 lname->ln_name = name;
1935 lname->ln_namelen = namelen;
1936 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1941 /*prepare the requests*/
1942 for (i = 0; i < count; i++) {
1943 char *tmp_name = info->eti_name;
1946 echo_md_build_name(lname, tmp_name, id);
1948 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1951 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
1960 static struct lu_object *echo_resolve_path(const struct lu_env *env,
1961 struct echo_device *ed, char *path,
1964 struct lu_device *ld = ed->ed_next;
1965 struct echo_thread_info *info = echo_env_info(env);
1966 struct lu_fid *fid = &info->eti_fid;
1967 struct lu_name *lname = &info->eti_lname;
1968 struct lu_object *parent = NULL;
1969 struct lu_object *child = NULL;
1973 *fid = ed->ed_root_fid;
1975 /* In the function below, .hs_keycmp resolves to
1976 * lu_obj_hop_keycmp() */
1977 /* coverity[overrun-buffer-val] */
1978 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1979 if (IS_ERR(parent)) {
1980 CERROR("Can not find the parent "DFID": rc = %ld\n",
1981 PFID(fid), PTR_ERR(parent));
1986 struct lu_object *ld_parent;
1989 e = strsep(&path, "/");
1994 if (!path || path[0] == '\0')
2000 lname->ln_namelen = strlen(e);
2002 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2003 if (ld_parent == NULL) {
2004 lu_object_put(env, parent);
2009 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2010 lu_object_put(env, parent);
2011 if (IS_ERR(child)) {
2012 rc = (int)PTR_ERR(child);
2013 CERROR("lookup %s under parent "DFID": rc = %d\n",
2014 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2021 RETURN(ERR_PTR(rc));
2026 static void echo_ucred_init(struct lu_env *env)
2028 struct lu_ucred *ucred = lu_ucred(env);
2030 ucred->uc_valid = UCRED_INVALID;
2032 ucred->uc_suppgids[0] = -1;
2033 ucred->uc_suppgids[1] = -1;
2035 ucred->uc_uid = ucred->uc_o_uid =
2036 from_kuid(&init_user_ns, current_uid());
2037 ucred->uc_gid = ucred->uc_o_gid =
2038 from_kgid(&init_user_ns, current_gid());
2039 ucred->uc_fsuid = ucred->uc_o_fsuid =
2040 from_kuid(&init_user_ns, current_fsuid());
2041 ucred->uc_fsgid = ucred->uc_o_fsgid =
2042 from_kgid(&init_user_ns, current_fsgid());
2043 ucred->uc_cap = cfs_curproc_cap_pack();
2045 /* remove fs privilege for non-root user. */
2046 if (ucred->uc_fsuid)
2047 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2048 ucred->uc_valid = UCRED_NEW;
2051 static void echo_ucred_fini(struct lu_env *env)
2053 struct lu_ucred *ucred = lu_ucred(env);
2054 ucred->uc_valid = UCRED_INIT;
2057 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2058 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2059 static int echo_md_handler(struct echo_device *ed, int command,
2060 char *path, int path_len, __u64 id, int count,
2061 struct obd_ioctl_data *data)
2063 struct echo_thread_info *info;
2064 struct lu_device *ld = ed->ed_next;
2067 struct lu_object *parent;
2069 int namelen = data->ioc_plen2;
2074 CERROR("MD echo client is not being initialized properly\n");
2078 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2079 CERROR("Only support MDD layer right now!\n");
2083 env = cl_env_get(&refcheck);
2085 RETURN(PTR_ERR(env));
2087 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2091 /* init big_lmm buffer */
2092 info = echo_env_info(env);
2093 LASSERT(info->eti_big_lmm == NULL);
2094 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2095 if (info->eti_big_lmm == NULL)
2096 GOTO(out_env, rc = -ENOMEM);
2097 info->eti_big_lmmsize = MIN_MD_SIZE;
2099 parent = echo_resolve_path(env, ed, path, path_len);
2100 if (IS_ERR(parent)) {
2101 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2103 GOTO(out_free, rc = PTR_ERR(parent));
2107 OBD_ALLOC(name, namelen + 1);
2109 GOTO(out_put, rc = -ENOMEM);
2110 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2111 GOTO(out_name, rc = -EFAULT);
2114 echo_ucred_init(env);
2117 case ECHO_MD_CREATE:
2118 case ECHO_MD_MKDIR: {
2119 struct echo_thread_info *info = echo_env_info(env);
2120 __u32 mode = data->ioc_obdo2.o_mode;
2121 struct lu_fid *fid = &info->eti_fid;
2122 int stripe_count = (int)data->ioc_obdo2.o_misc;
2123 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2125 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2129 /* In the function below, .hs_keycmp resolves to
2130 * lu_obj_hop_keycmp() */
2131 /* coverity[overrun-buffer-val] */
2132 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2133 id, mode, count, stripe_count,
2137 case ECHO_MD_DESTROY:
2138 case ECHO_MD_RMDIR: {
2139 __u32 mode = data->ioc_obdo2.o_mode;
2141 rc = echo_destroy_object(env, ed, parent, name, namelen,
2145 case ECHO_MD_LOOKUP:
2146 rc = echo_lookup_object(env, ed, parent, id, count);
2148 case ECHO_MD_GETATTR:
2149 rc = echo_getattr_object(env, ed, parent, id, count);
2151 case ECHO_MD_SETATTR:
2152 rc = echo_setattr_object(env, ed, parent, id, count);
2155 CERROR("unknown command %d\n", command);
2159 echo_ucred_fini(env);
2163 OBD_FREE(name, namelen + 1);
2165 lu_object_put(env, parent);
2167 LASSERT(info->eti_big_lmm);
2168 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2169 info->eti_big_lmm = NULL;
2170 info->eti_big_lmmsize = 0;
2172 cl_env_put(env, &refcheck);
2175 #endif /* HAVE_SERVER_SUPPORT */
2177 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2180 struct echo_object *eco;
2181 struct echo_client_obd *ec = ed->ed_ec;
2186 if (!(oa->o_valid & OBD_MD_FLID) ||
2187 !(oa->o_valid & OBD_MD_FLGROUP) ||
2188 !fid_seq_is_echo(ostid_seq(&oa->o_oi))) {
2189 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2193 if (ostid_id(&oa->o_oi) == 0)
2194 ostid_set_id(&oa->o_oi, ++last_object_id);
2196 rc = obd_create(env, ec->ec_exp, oa);
2198 CERROR("Cannot create objects: rc = %d\n", rc);
2204 oa->o_valid |= OBD_MD_FLID;
2206 eco = cl_echo_object_find(ed, &oa->o_oi);
2208 GOTO(failed, rc = PTR_ERR(eco));
2209 cl_echo_object_put(eco);
2211 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2215 if (created && rc != 0)
2216 obd_destroy(env, ec->ec_exp, oa);
2219 CERROR("create object failed with: rc = %d\n", rc);
2224 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2227 struct echo_object *eco;
2231 if (!(oa->o_valid & OBD_MD_FLID) ||
2232 !(oa->o_valid & OBD_MD_FLGROUP) ||
2233 ostid_id(&oa->o_oi) == 0) {
2234 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2239 eco = cl_echo_object_find(ed, &oa->o_oi);
2248 static void echo_put_object(struct echo_object *eco)
2252 rc = cl_echo_object_put(eco);
2254 CERROR("%s: echo client drop an object failed: rc = %d\n",
2255 eco->eo_dev->ed_ec->ec_exp->exp_obd->obd_name, rc);
2258 static void echo_client_page_debug_setup(struct page *page, int rw, u64 id,
2259 u64 offset, u64 count)
2266 /* no partial pages on the client */
2267 LASSERT(count == PAGE_CACHE_SIZE);
2271 for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2272 if (rw == OBD_BRW_WRITE) {
2273 stripe_off = offset + delta;
2276 stripe_off = 0xdeadbeef00c0ffeeULL;
2277 stripe_id = 0xdeadbeef00c0ffeeULL;
2279 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2280 stripe_off, stripe_id);
2287 echo_client_page_debug_check(struct page *page, u64 id, u64 offset, u64 count)
2296 /* no partial pages on the client */
2297 LASSERT(count == PAGE_CACHE_SIZE);
2301 for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2302 stripe_off = offset + delta;
2305 rc2 = block_debug_check("test_brw",
2306 addr + delta, OBD_ECHO_BLOCK_SIZE,
2307 stripe_off, stripe_id);
2309 CERROR ("Error in echo object "LPX64"\n", id);
2318 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2319 struct echo_object *eco, u64 offset,
2320 u64 count, int async)
2323 struct brw_page *pga;
2324 struct brw_page *pgp;
2325 struct page **pages;
2334 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2335 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2336 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2338 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2340 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2342 if ((count & (~PAGE_MASK)) != 0)
2345 /* XXX think again with misaligned I/O */
2346 npages = count >> PAGE_CACHE_SHIFT;
2348 if (rw == OBD_BRW_WRITE)
2349 brw_flags = OBD_BRW_ASYNC;
2351 OBD_ALLOC(pga, npages * sizeof(*pga));
2355 OBD_ALLOC(pages, npages * sizeof(*pages));
2356 if (pages == NULL) {
2357 OBD_FREE(pga, npages * sizeof(*pga));
2361 for (i = 0, pgp = pga, off = offset;
2363 i++, pgp++, off += PAGE_CACHE_SIZE) {
2365 LASSERT (pgp->pg == NULL); /* for cleanup */
2368 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2369 if (pgp->pg == NULL)
2373 pgp->count = PAGE_CACHE_SIZE;
2375 pgp->flag = brw_flags;
2378 echo_client_page_debug_setup(pgp->pg, rw,
2379 ostid_id(&oa->o_oi), off,
2383 /* brw mode can only be used at client */
2384 LASSERT(ed->ed_next != NULL);
2385 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2388 if (rc != 0 || rw != OBD_BRW_READ)
2391 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2392 if (pgp->pg == NULL)
2397 vrc = echo_client_page_debug_check(pgp->pg,
2398 ostid_id(&oa->o_oi),
2399 pgp->off, pgp->count);
2400 if (vrc != 0 && rc == 0)
2403 OBD_PAGE_FREE(pgp->pg);
2405 OBD_FREE(pga, npages * sizeof(*pga));
2406 OBD_FREE(pages, npages * sizeof(*pages));
2410 static int echo_client_prep_commit(const struct lu_env *env,
2411 struct obd_export *exp, int rw,
2412 struct obdo *oa, struct echo_object *eco,
2413 u64 offset, u64 count,
2414 u64 batch, int async)
2416 struct obd_ioobj ioo;
2417 struct niobuf_local *lnb;
2418 struct niobuf_remote rnb;
2420 u64 npages, tot_pages;
2421 int i, ret = 0, brw_flags = 0;
2425 if (count <= 0 || (count & ~PAGE_CACHE_MASK) != 0)
2428 npages = batch >> PAGE_CACHE_SHIFT;
2429 tot_pages = count >> PAGE_CACHE_SHIFT;
2431 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2433 GOTO(out, ret = -ENOMEM);
2435 if (rw == OBD_BRW_WRITE && async)
2436 brw_flags |= OBD_BRW_ASYNC;
2438 obdo_to_ioobj(oa, &ioo);
2442 for (; tot_pages > 0; tot_pages -= npages) {
2445 if (tot_pages < npages)
2448 rnb.rnb_offset = off;
2449 rnb.rnb_len = npages * PAGE_CACHE_SIZE;
2450 rnb.rnb_flags = brw_flags;
2452 off += npages * PAGE_CACHE_SIZE;
2455 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, &rnb, &lpages, lnb);
2459 for (i = 0; i < lpages; i++) {
2460 struct page *page = lnb[i].lnb_page;
2462 /* read past eof? */
2463 if (page == NULL && lnb[i].lnb_rc == 0)
2467 lnb[i].lnb_flags |= OBD_BRW_ASYNC;
2469 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2470 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2471 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2474 if (rw == OBD_BRW_WRITE)
2475 echo_client_page_debug_setup(page, rw,
2476 ostid_id(&oa->o_oi),
2477 lnb[i].lnb_file_offset,
2480 echo_client_page_debug_check(page,
2481 ostid_id(&oa->o_oi),
2482 lnb[i].lnb_file_offset,
2486 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo, &rnb, npages, lnb,
2491 /* Reuse env context. */
2492 lu_context_exit((struct lu_context *)&env->le_ctx);
2493 lu_context_enter((struct lu_context *)&env->le_ctx);
2498 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2502 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2503 struct obd_export *exp,
2504 struct obd_ioctl_data *data)
2506 struct obd_device *obd = class_exp2obd(exp);
2507 struct echo_device *ed = obd2echo_dev(obd);
2508 struct echo_client_obd *ec = ed->ed_ec;
2509 struct obdo *oa = &data->ioc_obdo1;
2510 struct echo_object *eco;
2516 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2518 rc = echo_get_object(&eco, ed, oa);
2522 oa->o_valid &= ~OBD_MD_FLHANDLE;
2524 /* OFD/obdfilter works only via prep/commit */
2525 test_mode = (long)data->ioc_pbuf1;
2526 if (ed->ed_next == NULL && test_mode != 3) {
2528 data->ioc_plen1 = data->ioc_count;
2534 /* Truncate batch size to maximum */
2535 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2536 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2538 switch (test_mode) {
2542 rc = echo_client_kbrw(ed, rw, oa, eco, data->ioc_offset,
2543 data->ioc_count, async);
2546 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa, eco,
2547 data->ioc_offset, data->ioc_count,
2548 data->ioc_plen1, async);
2554 echo_put_object(eco);
2560 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2561 void *karg, void *uarg)
2563 #ifdef HAVE_SERVER_SUPPORT
2564 struct tgt_session_info *tsi;
2566 struct obd_device *obd = exp->exp_obd;
2567 struct echo_device *ed = obd2echo_dev(obd);
2568 struct echo_client_obd *ec = ed->ed_ec;
2569 struct echo_object *eco;
2570 struct obd_ioctl_data *data = karg;
2574 int rw = OBD_BRW_READ;
2576 #ifdef HAVE_SERVER_SUPPORT
2577 struct lu_context echo_session;
2581 oa = &data->ioc_obdo1;
2582 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2583 oa->o_valid |= OBD_MD_FLGROUP;
2584 ostid_set_seq_echo(&oa->o_oi);
2587 /* This FID is unpacked just for validation at this point */
2588 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2596 rc = lu_env_init(env, LCT_DT_THREAD);
2598 GOTO(out_alloc, rc = -ENOMEM);
2600 #ifdef HAVE_SERVER_SUPPORT
2601 env->le_ses = &echo_session;
2602 rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2603 if (unlikely(rc < 0))
2605 lu_context_enter(env->le_ses);
2607 tsi = tgt_ses_info(env);
2608 tsi->tsi_exp = ec->ec_exp;
2609 tsi->tsi_jobid = NULL;
2612 case OBD_IOC_CREATE: /* may create echo object */
2613 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2614 GOTO (out, rc = -EPERM);
2616 rc = echo_create_object(env, ed, oa);
2619 #ifdef HAVE_SERVER_SUPPORT
2620 case OBD_IOC_ECHO_MD: {
2627 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2628 GOTO(out, rc = -EPERM);
2630 count = data->ioc_count;
2631 cmd = data->ioc_command;
2633 id = data->ioc_obdo2.o_oi.oi.oi_id;
2634 dirlen = data->ioc_plen1;
2635 OBD_ALLOC(dir, dirlen + 1);
2637 GOTO(out, rc = -ENOMEM);
2639 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2640 OBD_FREE(dir, data->ioc_plen1 + 1);
2641 GOTO(out, rc = -EFAULT);
2644 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2645 OBD_FREE(dir, dirlen + 1);
2648 case OBD_IOC_ECHO_ALLOC_SEQ: {
2649 struct lu_env *cl_env;
2654 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2655 GOTO(out, rc = -EPERM);
2657 cl_env = cl_env_get(&refcheck);
2659 GOTO(out, rc = PTR_ERR(cl_env));
2661 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2664 cl_env_put(cl_env, &refcheck);
2668 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2669 cl_env_put(cl_env, &refcheck);
2671 CERROR("%s: Can not alloc seq: rc = %d\n",
2676 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2679 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2680 if (copy_to_user(data->ioc_pbuf2, &max_count,
2685 #endif /* HAVE_SERVER_SUPPORT */
2686 case OBD_IOC_DESTROY:
2687 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2688 GOTO (out, rc = -EPERM);
2690 rc = echo_get_object(&eco, ed, oa);
2692 rc = obd_destroy(env, ec->ec_exp, oa);
2694 eco->eo_deleted = 1;
2695 echo_put_object(eco);
2699 case OBD_IOC_GETATTR:
2700 rc = echo_get_object(&eco, ed, oa);
2702 rc = obd_getattr(env, ec->ec_exp, oa);
2703 echo_put_object(eco);
2707 case OBD_IOC_SETATTR:
2708 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2709 GOTO (out, rc = -EPERM);
2711 rc = echo_get_object(&eco, ed, oa);
2713 rc = obd_setattr(env, ec->ec_exp, oa);
2714 echo_put_object(eco);
2718 case OBD_IOC_BRW_WRITE:
2719 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2720 GOTO (out, rc = -EPERM);
2724 case OBD_IOC_BRW_READ:
2725 rc = echo_client_brw_ioctl(env, rw, exp, data);
2729 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2730 GOTO (out, rc = -ENOTTY);
2735 #ifdef HAVE_SERVER_SUPPORT
2736 lu_context_exit(env->le_ses);
2737 lu_context_fini(env->le_ses);
2747 static int echo_client_setup(const struct lu_env *env,
2748 struct obd_device *obddev, struct lustre_cfg *lcfg)
2750 struct echo_client_obd *ec = &obddev->u.echo_client;
2751 struct obd_device *tgt;
2752 struct obd_uuid echo_uuid = { "ECHO_UUID" };
2753 struct obd_connect_data *ocd = NULL;
2757 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2758 CERROR("requires a TARGET OBD name\n");
2762 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2763 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2764 CERROR("device not attached or not set up (%s)\n",
2765 lustre_cfg_string(lcfg, 1));
2769 spin_lock_init(&ec->ec_lock);
2770 INIT_LIST_HEAD(&ec->ec_objects);
2771 INIT_LIST_HEAD(&ec->ec_locks);
2774 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2775 #ifdef HAVE_SERVER_SUPPORT
2776 lu_context_tags_update(ECHO_MD_CTX_TAG);
2777 lu_session_tags_update(ECHO_MD_SES_TAG);
2779 CERROR("Local operations are NOT supported on client side. "
2780 "Only remote operations are supported. Metadata client "
2781 "must be run on server side.\n");
2786 OBD_ALLOC(ocd, sizeof(*ocd));
2788 CERROR("Can't alloc ocd connecting to %s\n",
2789 lustre_cfg_string(lcfg, 1));
2793 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2794 OBD_CONNECT_BRW_SIZE |
2795 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2796 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2798 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2799 ocd->ocd_version = LUSTRE_VERSION_CODE;
2800 ocd->ocd_group = FID_SEQ_ECHO;
2802 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2804 /* Turn off pinger because it connects to tgt obd directly. */
2805 spin_lock(&tgt->obd_dev_lock);
2806 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
2807 spin_unlock(&tgt->obd_dev_lock);
2810 OBD_FREE(ocd, sizeof(*ocd));
2813 CERROR("fail to connect to device %s\n",
2814 lustre_cfg_string(lcfg, 1));
2821 static int echo_client_cleanup(struct obd_device *obddev)
2823 struct echo_device *ed = obd2echo_dev(obddev);
2824 struct echo_client_obd *ec = &obddev->u.echo_client;
2828 /*Do nothing for Metadata echo client*/
2832 if (ed->ed_next_ismd) {
2833 #ifdef HAVE_SERVER_SUPPORT
2834 lu_context_tags_clear(ECHO_MD_CTX_TAG);
2835 lu_session_tags_clear(ECHO_MD_SES_TAG);
2837 CERROR("This is client-side only module, does not support "
2838 "metadata echo client.\n");
2843 if (!list_empty(&obddev->obd_exports)) {
2844 CERROR("still has clients!\n");
2848 LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
2849 rc = obd_disconnect(ec->ec_exp);
2851 CERROR("fail to disconnect device: %d\n", rc);
2856 static int echo_client_connect(const struct lu_env *env,
2857 struct obd_export **exp,
2858 struct obd_device *src, struct obd_uuid *cluuid,
2859 struct obd_connect_data *data, void *localdata)
2862 struct lustre_handle conn = { 0 };
2865 rc = class_connect(&conn, src, cluuid);
2867 *exp = class_conn2export(&conn);
2873 static int echo_client_disconnect(struct obd_export *exp)
2879 GOTO(out, rc = -EINVAL);
2881 rc = class_disconnect(exp);
2887 static struct obd_ops echo_client_obd_ops = {
2888 .o_owner = THIS_MODULE,
2889 .o_iocontrol = echo_client_iocontrol,
2890 .o_connect = echo_client_connect,
2891 .o_disconnect = echo_client_disconnect
2894 static int echo_client_init(void)
2898 rc = lu_kmem_init(echo_caches);
2900 rc = class_register_type(&echo_client_obd_ops, NULL, true, NULL,
2901 LUSTRE_ECHO_CLIENT_NAME,
2904 lu_kmem_fini(echo_caches);
2909 static void echo_client_exit(void)
2911 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
2912 lu_kmem_fini(echo_caches);
2915 static int __init obdecho_init(void)
2920 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
2922 LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
2924 # ifdef HAVE_SERVER_SUPPORT
2925 rc = echo_persistent_pages_init();
2929 rc = class_register_type(&echo_obd_ops, NULL, true, NULL,
2930 LUSTRE_ECHO_NAME, NULL);
2935 rc = echo_client_init();
2937 # ifdef HAVE_SERVER_SUPPORT
2941 class_unregister_type(LUSTRE_ECHO_NAME);
2943 echo_persistent_pages_fini();
2949 static void /*__exit*/ obdecho_exit(void)
2953 # ifdef HAVE_SERVER_SUPPORT
2954 class_unregister_type(LUSTRE_ECHO_NAME);
2955 echo_persistent_pages_fini();
2959 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2960 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
2961 MODULE_VERSION(LUSTRE_VERSION_STRING);
2962 MODULE_LICENSE("GPL");
2964 module_init(obdecho_init);
2965 module_exit(obdecho_exit);
2967 /** @} echo_client */