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, 2013, 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 <libcfs/libcfs.h>
41 #include <liblustre.h>
45 #include <obd_support.h>
46 #include <obd_class.h>
47 #include <lustre_debug.h>
48 #include <lprocfs_status.h>
49 #include <cl_object.h>
50 #include <lustre_fid.h>
51 #include <lustre_acl.h>
52 #include <lustre_net.h>
55 #include "echo_internal.h"
57 /** \defgroup echo_client Echo Client
62 struct cl_device ed_cl;
63 struct echo_client_obd *ed_ec;
65 struct cl_site ed_site_myself;
66 struct cl_site *ed_site;
67 struct lu_device *ed_next;
70 struct lu_client_seq *ed_cl_seq;
74 struct cl_object eo_cl;
75 struct cl_object_header eo_hdr;
77 struct echo_device *eo_dev;
78 cfs_list_t eo_obj_chain;
79 struct lov_stripe_md *eo_lsm;
80 cfs_atomic_t eo_npages;
84 struct echo_object_conf {
85 struct cl_object_conf eoc_cl;
86 struct lov_stripe_md **eoc_md;
90 struct cl_page_slice ep_cl;
92 cfs_page_t *ep_vmpage;
96 struct cl_lock_slice el_cl;
98 struct echo_object *el_object;
100 cfs_atomic_t el_refcount;
104 struct cl_io_slice ei_cl;
109 struct cl_req_slice er_cl;
113 static int echo_client_setup(const struct lu_env *env,
114 struct obd_device *obddev,
115 struct lustre_cfg *lcfg);
116 static int echo_client_cleanup(struct obd_device *obddev);
119 /** \defgroup echo_helpers Helper functions
122 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
124 return container_of0(dev, struct echo_device, ed_cl);
127 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
132 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
134 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
137 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
142 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
144 return container_of(o, struct echo_object, eo_cl);
147 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
149 return container_of(s, struct echo_page, ep_cl);
152 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
154 return container_of(s, struct echo_lock, el_cl);
157 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
159 return ecl->el_cl.cls_lock;
162 static struct lu_context_key echo_thread_key;
163 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
165 struct echo_thread_info *info;
166 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
167 LASSERT(info != NULL);
172 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
174 return container_of(c, struct echo_object_conf, eoc_cl);
177 /** @} echo_helpers */
179 static struct echo_object *cl_echo_object_find(struct echo_device *d,
180 struct lov_stripe_md **lsm);
181 static int cl_echo_object_put(struct echo_object *eco);
182 static int cl_echo_enqueue (struct echo_object *eco, obd_off start,
183 obd_off end, int mode, __u64 *cookie);
184 static int cl_echo_cancel (struct echo_device *d, __u64 cookie);
185 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
186 cfs_page_t **pages, int npages, int async);
188 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
190 struct echo_thread_info {
191 struct echo_object_conf eti_conf;
192 struct lustre_md eti_md;
194 struct cl_2queue eti_queue;
196 struct cl_lock_descr eti_descr;
197 struct lu_fid eti_fid;
198 struct lu_fid eti_fid2;
199 struct md_op_spec eti_spec;
200 struct lov_mds_md_v3 eti_lmm;
201 struct lov_user_md_v3 eti_lum;
202 struct md_attr eti_ma;
203 struct lu_name eti_lname;
204 /* per-thread values, can be re-used */
208 struct lu_buf eti_buf;
209 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
212 /* No session used right now */
213 struct echo_session_info {
217 static cfs_mem_cache_t *echo_lock_kmem;
218 static cfs_mem_cache_t *echo_object_kmem;
219 static cfs_mem_cache_t *echo_thread_kmem;
220 static cfs_mem_cache_t *echo_session_kmem;
221 //static cfs_mem_cache_t *echo_req_kmem;
223 static struct lu_kmem_descr echo_caches[] = {
225 .ckd_cache = &echo_lock_kmem,
226 .ckd_name = "echo_lock_kmem",
227 .ckd_size = sizeof (struct echo_lock)
230 .ckd_cache = &echo_object_kmem,
231 .ckd_name = "echo_object_kmem",
232 .ckd_size = sizeof (struct echo_object)
235 .ckd_cache = &echo_thread_kmem,
236 .ckd_name = "echo_thread_kmem",
237 .ckd_size = sizeof (struct echo_thread_info)
240 .ckd_cache = &echo_session_kmem,
241 .ckd_name = "echo_session_kmem",
242 .ckd_size = sizeof (struct echo_session_info)
246 .ckd_cache = &echo_req_kmem,
247 .ckd_name = "echo_req_kmem",
248 .ckd_size = sizeof (struct echo_req)
256 /** \defgroup echo_page Page operations
258 * Echo page operations.
262 static cfs_page_t *echo_page_vmpage(const struct lu_env *env,
263 const struct cl_page_slice *slice)
265 return cl2echo_page(slice)->ep_vmpage;
268 static int echo_page_own(const struct lu_env *env,
269 const struct cl_page_slice *slice,
270 struct cl_io *io, int nonblock)
272 struct echo_page *ep = cl2echo_page(slice);
275 mutex_lock(&ep->ep_lock);
276 else if (!mutex_trylock(&ep->ep_lock))
281 static void echo_page_disown(const struct lu_env *env,
282 const struct cl_page_slice *slice,
285 struct echo_page *ep = cl2echo_page(slice);
287 LASSERT(mutex_is_locked(&ep->ep_lock));
288 mutex_unlock(&ep->ep_lock);
291 static void echo_page_discard(const struct lu_env *env,
292 const struct cl_page_slice *slice,
293 struct cl_io *unused)
295 cl_page_delete(env, slice->cpl_page);
298 static int echo_page_is_vmlocked(const struct lu_env *env,
299 const struct cl_page_slice *slice)
301 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
306 static void echo_page_completion(const struct lu_env *env,
307 const struct cl_page_slice *slice,
310 LASSERT(slice->cpl_page->cp_sync_io != NULL);
313 static void echo_page_fini(const struct lu_env *env,
314 struct cl_page_slice *slice)
316 struct echo_page *ep = cl2echo_page(slice);
317 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
318 cfs_page_t *vmpage = ep->ep_vmpage;
321 cfs_atomic_dec(&eco->eo_npages);
322 page_cache_release(vmpage);
326 static int echo_page_prep(const struct lu_env *env,
327 const struct cl_page_slice *slice,
328 struct cl_io *unused)
333 static int echo_page_print(const struct lu_env *env,
334 const struct cl_page_slice *slice,
335 void *cookie, lu_printer_t printer)
337 struct echo_page *ep = cl2echo_page(slice);
339 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
340 ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
344 static const struct cl_page_operations echo_page_ops = {
345 .cpo_own = echo_page_own,
346 .cpo_disown = echo_page_disown,
347 .cpo_discard = echo_page_discard,
348 .cpo_vmpage = echo_page_vmpage,
349 .cpo_fini = echo_page_fini,
350 .cpo_print = echo_page_print,
351 .cpo_is_vmlocked = echo_page_is_vmlocked,
354 .cpo_prep = echo_page_prep,
355 .cpo_completion = echo_page_completion,
358 .cpo_prep = echo_page_prep,
359 .cpo_completion = echo_page_completion,
365 /** \defgroup echo_lock Locking
367 * echo lock operations
371 static void echo_lock_fini(const struct lu_env *env,
372 struct cl_lock_slice *slice)
374 struct echo_lock *ecl = cl2echo_lock(slice);
376 LASSERT(cfs_list_empty(&ecl->el_chain));
377 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
380 static void echo_lock_delete(const struct lu_env *env,
381 const struct cl_lock_slice *slice)
383 struct echo_lock *ecl = cl2echo_lock(slice);
385 LASSERT(cfs_list_empty(&ecl->el_chain));
388 static int echo_lock_fits_into(const struct lu_env *env,
389 const struct cl_lock_slice *slice,
390 const struct cl_lock_descr *need,
391 const struct cl_io *unused)
396 static struct cl_lock_operations echo_lock_ops = {
397 .clo_fini = echo_lock_fini,
398 .clo_delete = echo_lock_delete,
399 .clo_fits_into = echo_lock_fits_into
404 /** \defgroup echo_cl_ops cl_object operations
406 * operations for cl_object
410 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
411 struct cl_page *page, cfs_page_t *vmpage)
413 struct echo_page *ep = cl_object_page_slice(obj, page);
414 struct echo_object *eco = cl2echo_obj(obj);
417 ep->ep_vmpage = vmpage;
418 page_cache_get(vmpage);
419 mutex_init(&ep->ep_lock);
420 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
421 cfs_atomic_inc(&eco->eo_npages);
425 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
431 static int echo_lock_init(const struct lu_env *env,
432 struct cl_object *obj, struct cl_lock *lock,
433 const struct cl_io *unused)
435 struct echo_lock *el;
438 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, CFS_ALLOC_IO);
440 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
441 el->el_object = cl2echo_obj(obj);
442 CFS_INIT_LIST_HEAD(&el->el_chain);
443 cfs_atomic_set(&el->el_refcount, 0);
445 RETURN(el == NULL ? -ENOMEM : 0);
448 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
449 const struct cl_object_conf *conf)
454 static const struct cl_object_operations echo_cl_obj_ops = {
455 .coo_page_init = echo_page_init,
456 .coo_lock_init = echo_lock_init,
457 .coo_io_init = echo_io_init,
458 .coo_conf_set = echo_conf_set
460 /** @} echo_cl_ops */
462 /** \defgroup echo_lu_ops lu_object operations
464 * operations for echo lu object.
468 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
469 const struct lu_object_conf *conf)
471 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
472 struct echo_client_obd *ec = ed->ed_ec;
473 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
477 struct lu_object *below;
478 struct lu_device *under;
481 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
485 lu_object_add(obj, below);
488 if (!ed->ed_next_ismd) {
489 const struct cl_object_conf *cconf = lu2cl_conf(conf);
490 struct echo_object_conf *econf = cl2echo_conf(cconf);
492 LASSERT(econf->eoc_md);
493 eco->eo_lsm = *econf->eoc_md;
494 /* clear the lsm pointer so that it won't get freed. */
495 *econf->eoc_md = NULL;
501 cfs_atomic_set(&eco->eo_npages, 0);
502 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
504 spin_lock(&ec->ec_lock);
505 cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
506 spin_unlock(&ec->ec_lock);
511 /* taken from osc_unpackmd() */
512 static int echo_alloc_memmd(struct echo_device *ed,
513 struct lov_stripe_md **lsmp)
519 /* If export is lov/osc then use their obd method */
520 if (ed->ed_next != NULL)
521 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
522 /* OFD has no unpackmd method, do everything here */
523 lsm_size = lov_stripe_md_size(1);
525 LASSERT(*lsmp == NULL);
526 OBD_ALLOC(*lsmp, lsm_size);
530 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
531 if ((*lsmp)->lsm_oinfo[0] == NULL) {
532 OBD_FREE(*lsmp, lsm_size);
536 loi_init((*lsmp)->lsm_oinfo[0]);
537 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
538 ostid_set_seq_echo(&(*lsmp)->lsm_oi);
543 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
549 /* If export is lov/osc then use their obd method */
550 if (ed->ed_next != NULL)
551 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
552 /* OFD has no unpackmd method, do everything here */
553 lsm_size = lov_stripe_md_size(1);
555 LASSERT(*lsmp != NULL);
556 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
557 OBD_FREE(*lsmp, lsm_size);
562 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
564 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
565 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
568 LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
570 spin_lock(&ec->ec_lock);
571 cfs_list_del_init(&eco->eo_obj_chain);
572 spin_unlock(&ec->ec_lock);
575 lu_object_header_fini(obj->lo_header);
578 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
579 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
583 static int echo_object_print(const struct lu_env *env, void *cookie,
584 lu_printer_t p, const struct lu_object *o)
586 struct echo_object *obj = cl2echo_obj(lu2cl(o));
588 return (*p)(env, cookie, "echoclient-object@%p", obj);
591 static const struct lu_object_operations echo_lu_obj_ops = {
592 .loo_object_init = echo_object_init,
593 .loo_object_delete = NULL,
594 .loo_object_release = NULL,
595 .loo_object_free = echo_object_free,
596 .loo_object_print = echo_object_print,
597 .loo_object_invariant = NULL
599 /** @} echo_lu_ops */
601 /** \defgroup echo_lu_dev_ops lu_device operations
603 * Operations for echo lu device.
607 static struct lu_object *echo_object_alloc(const struct lu_env *env,
608 const struct lu_object_header *hdr,
609 struct lu_device *dev)
611 struct echo_object *eco;
612 struct lu_object *obj = NULL;
615 /* we're the top dev. */
616 LASSERT(hdr == NULL);
617 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, CFS_ALLOC_IO);
619 struct cl_object_header *hdr = &eco->eo_hdr;
621 obj = &echo_obj2cl(eco)->co_lu;
622 cl_object_header_init(hdr);
623 lu_object_init(obj, &hdr->coh_lu, dev);
624 lu_object_add_top(&hdr->coh_lu, obj);
626 eco->eo_cl.co_ops = &echo_cl_obj_ops;
627 obj->lo_ops = &echo_lu_obj_ops;
632 static struct lu_device_operations echo_device_lu_ops = {
633 .ldo_object_alloc = echo_object_alloc,
636 /** @} echo_lu_dev_ops */
638 static struct cl_device_operations echo_device_cl_ops = {
641 /** \defgroup echo_init Setup and teardown
643 * Init and fini functions for echo client.
647 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
649 struct cl_site *site = &ed->ed_site_myself;
652 /* initialize site */
653 rc = cl_site_init(site, &ed->ed_cl);
655 CERROR("Cannot initilize site for echo client(%d)\n", rc);
659 rc = lu_site_init_finish(&site->cs_lu);
667 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
670 if (!ed->ed_next_ismd)
671 cl_site_fini(ed->ed_site);
676 static void *echo_thread_key_init(const struct lu_context *ctx,
677 struct lu_context_key *key)
679 struct echo_thread_info *info;
681 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, CFS_ALLOC_IO);
683 info = ERR_PTR(-ENOMEM);
687 static void echo_thread_key_fini(const struct lu_context *ctx,
688 struct lu_context_key *key, void *data)
690 struct echo_thread_info *info = data;
691 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
694 static void echo_thread_key_exit(const struct lu_context *ctx,
695 struct lu_context_key *key, void *data)
699 static struct lu_context_key echo_thread_key = {
700 .lct_tags = LCT_CL_THREAD,
701 .lct_init = echo_thread_key_init,
702 .lct_fini = echo_thread_key_fini,
703 .lct_exit = echo_thread_key_exit
706 static void *echo_session_key_init(const struct lu_context *ctx,
707 struct lu_context_key *key)
709 struct echo_session_info *session;
711 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, CFS_ALLOC_IO);
713 session = ERR_PTR(-ENOMEM);
717 static void echo_session_key_fini(const struct lu_context *ctx,
718 struct lu_context_key *key, void *data)
720 struct echo_session_info *session = data;
721 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
724 static void echo_session_key_exit(const struct lu_context *ctx,
725 struct lu_context_key *key, void *data)
729 static struct lu_context_key echo_session_key = {
730 .lct_tags = LCT_SESSION,
731 .lct_init = echo_session_key_init,
732 .lct_fini = echo_session_key_fini,
733 .lct_exit = echo_session_key_exit
736 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
738 #define ECHO_SEQ_WIDTH 0xffffffff
739 static int echo_fid_init(struct echo_device *ed, char *obd_name,
740 struct seq_server_site *ss)
746 OBD_ALLOC_PTR(ed->ed_cl_seq);
747 if (ed->ed_cl_seq == NULL)
750 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
752 GOTO(out_free_seq, rc = -ENOMEM);
754 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
756 /* Init client side sequence-manager */
757 rc = seq_client_init(ed->ed_cl_seq, NULL,
759 prefix, ss->ss_server_seq);
760 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
761 OBD_FREE(prefix, MAX_OBD_NAME + 5);
763 GOTO(out_free_seq, rc);
768 OBD_FREE_PTR(ed->ed_cl_seq);
769 ed->ed_cl_seq = NULL;
773 static int echo_fid_fini(struct obd_device *obddev)
775 struct echo_device *ed = obd2echo_dev(obddev);
778 if (ed->ed_cl_seq != NULL) {
779 seq_client_fini(ed->ed_cl_seq);
780 OBD_FREE_PTR(ed->ed_cl_seq);
781 ed->ed_cl_seq = NULL;
787 static struct lu_device *echo_device_alloc(const struct lu_env *env,
788 struct lu_device_type *t,
789 struct lustre_cfg *cfg)
791 struct lu_device *next;
792 struct echo_device *ed;
793 struct cl_device *cd;
794 struct obd_device *obd = NULL; /* to keep compiler happy */
795 struct obd_device *tgt;
796 const char *tgt_type_name;
803 GOTO(out, rc = -ENOMEM);
807 rc = cl_device_init(cd, t);
811 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
812 cd->cd_ops = &echo_device_cl_ops;
815 obd = class_name2obd(lustre_cfg_string(cfg, 0));
816 LASSERT(obd != NULL);
817 LASSERT(env != NULL);
819 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
821 CERROR("Can not find tgt device %s\n",
822 lustre_cfg_string(cfg, 1));
823 GOTO(out, rc = -ENODEV);
826 next = tgt->obd_lu_dev;
827 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
828 ed->ed_next_ismd = 1;
830 ed->ed_next_ismd = 0;
831 rc = echo_site_init(env, ed);
837 rc = echo_client_setup(env, obd, cfg);
841 ed->ed_ec = &obd->u.echo_client;
844 if (ed->ed_next_ismd) {
845 /* Suppose to connect to some Metadata layer */
847 struct lu_device *ld;
851 CERROR("%s is not lu device type!\n",
852 lustre_cfg_string(cfg, 1));
853 GOTO(out, rc = -EINVAL);
856 tgt_type_name = lustre_cfg_string(cfg, 2);
857 if (!tgt_type_name) {
858 CERROR("%s no type name for echo %s setup\n",
859 lustre_cfg_string(cfg, 1),
860 tgt->obd_type->typ_name);
861 GOTO(out, rc = -EINVAL);
866 spin_lock(&ls->ls_ld_lock);
867 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
868 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
873 spin_unlock(&ls->ls_ld_lock);
876 CERROR("%s is not lu device type!\n",
877 lustre_cfg_string(cfg, 1));
878 GOTO(out, rc = -EINVAL);
882 /* For MD echo client, it will use the site in MDS stack */
883 ed->ed_site_myself.cs_lu = *ls;
884 ed->ed_site = &ed->ed_site_myself;
885 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
886 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
888 CERROR("echo fid init error %d\n", rc);
892 /* if echo client is to be stacked upon ost device, the next is
893 * NULL since ost is not a clio device so far */
894 if (next != NULL && !lu_device_is_cl(next))
897 tgt_type_name = tgt->obd_type->typ_name;
899 LASSERT(next != NULL);
900 if (next->ld_site != NULL)
901 GOTO(out, rc = -EBUSY);
903 next->ld_site = &ed->ed_site->cs_lu;
904 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
905 next->ld_type->ldt_name,
910 /* Tricky case, I have to determine the obd type since
911 * CLIO uses the different parameters to initialize
912 * objects for lov & osc. */
913 if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
914 ed->ed_next_islov = 1;
916 LASSERT(strcmp(tgt_type_name,
917 LUSTRE_OSC_NAME) == 0);
919 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
923 RETURN(&cd->cd_lu_dev);
928 rc2 = echo_client_cleanup(obd);
930 CERROR("Cleanup obd device %s error(%d)\n",
935 echo_site_fini(env, ed);
937 cl_device_fini(&ed->ed_cl);
947 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
948 const char *name, struct lu_device *next)
954 static struct lu_device *echo_device_fini(const struct lu_env *env,
957 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
958 struct lu_device *next = ed->ed_next;
960 while (next && !ed->ed_next_ismd)
961 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
965 static void echo_lock_release(const struct lu_env *env,
966 struct echo_lock *ecl,
969 struct cl_lock *clk = echo_lock2cl(ecl);
973 cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
975 cl_lock_mutex_get(env, clk);
976 cl_lock_cancel(env, clk);
977 cl_lock_delete(env, clk);
978 cl_lock_mutex_put(env, clk);
980 cl_lock_put(env, clk);
983 static struct lu_device *echo_device_free(const struct lu_env *env,
986 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
987 struct echo_client_obd *ec = ed->ed_ec;
988 struct echo_object *eco;
989 struct lu_device *next = ed->ed_next;
991 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
994 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
996 /* check if there are objects still alive.
997 * It shouldn't have any object because lu_site_purge would cleanup
998 * all of cached objects. Anyway, probably the echo device is being
999 * parallelly accessed.
1001 spin_lock(&ec->ec_lock);
1002 cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1003 eco->eo_deleted = 1;
1004 spin_unlock(&ec->ec_lock);
1007 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1010 "Waiting for the reference of echo object to be dropped\n");
1012 /* Wait for the last reference to be dropped. */
1013 spin_lock(&ec->ec_lock);
1014 while (!cfs_list_empty(&ec->ec_objects)) {
1015 spin_unlock(&ec->ec_lock);
1016 CERROR("echo_client still has objects at cleanup time, "
1017 "wait for 1 second\n");
1018 cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT,
1019 cfs_time_seconds(1));
1020 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1021 spin_lock(&ec->ec_lock);
1023 spin_unlock(&ec->ec_lock);
1025 LASSERT(cfs_list_empty(&ec->ec_locks));
1027 CDEBUG(D_INFO, "No object exists, exiting...\n");
1029 echo_client_cleanup(d->ld_obd);
1030 echo_fid_fini(d->ld_obd);
1031 while (next && !ed->ed_next_ismd)
1032 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1034 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1035 echo_site_fini(env, ed);
1036 cl_device_fini(&ed->ed_cl);
1042 static const struct lu_device_type_operations echo_device_type_ops = {
1043 .ldto_init = echo_type_init,
1044 .ldto_fini = echo_type_fini,
1046 .ldto_start = echo_type_start,
1047 .ldto_stop = echo_type_stop,
1049 .ldto_device_alloc = echo_device_alloc,
1050 .ldto_device_free = echo_device_free,
1051 .ldto_device_init = echo_device_init,
1052 .ldto_device_fini = echo_device_fini
1055 static struct lu_device_type echo_device_type = {
1056 .ldt_tags = LU_DEVICE_CL,
1057 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1058 .ldt_ops = &echo_device_type_ops,
1059 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1063 /** \defgroup echo_exports Exported operations
1065 * exporting functions to echo client
1070 /* Interfaces to echo client obd device */
1071 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1072 struct lov_stripe_md **lsmp)
1075 struct echo_thread_info *info;
1076 struct echo_object_conf *conf;
1077 struct lov_stripe_md *lsm;
1078 struct echo_object *eco;
1079 struct cl_object *obj;
1088 LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi));
1089 LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n",
1090 POSTID(&lsm->lsm_oi));
1092 /* Never return an object if the obd is to be freed. */
1093 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1094 RETURN(ERR_PTR(-ENODEV));
1096 env = cl_env_get(&refcheck);
1098 RETURN((void *)env);
1100 info = echo_env_info(env);
1101 conf = &info->eti_conf;
1103 if (!d->ed_next_islov) {
1104 struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1105 LASSERT(oinfo != NULL);
1106 oinfo->loi_oi = lsm->lsm_oi;
1107 conf->eoc_cl.u.coc_oinfo = oinfo;
1109 struct lustre_md *md;
1111 memset(md, 0, sizeof *md);
1113 conf->eoc_cl.u.coc_md = md;
1116 conf->eoc_md = lsmp;
1118 fid = &info->eti_fid;
1119 rc = ostid_to_fid(fid, &lsm->lsm_oi, 0);
1121 GOTO(out, eco = ERR_PTR(rc));
1123 /* In the function below, .hs_keycmp resolves to
1124 * lu_obj_hop_keycmp() */
1125 /* coverity[overrun-buffer-val] */
1126 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1128 GOTO(out, eco = (void*)obj);
1130 eco = cl2echo_obj(obj);
1131 if (eco->eo_deleted) {
1132 cl_object_put(env, obj);
1133 eco = ERR_PTR(-EAGAIN);
1137 cl_env_put(env, &refcheck);
1141 static int cl_echo_object_put(struct echo_object *eco)
1144 struct cl_object *obj = echo_obj2cl(eco);
1148 env = cl_env_get(&refcheck);
1150 RETURN(PTR_ERR(env));
1152 /* an external function to kill an object? */
1153 if (eco->eo_deleted) {
1154 struct lu_object_header *loh = obj->co_lu.lo_header;
1155 LASSERT(&eco->eo_hdr == luh2coh(loh));
1156 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1159 cl_object_put(env, obj);
1160 cl_env_put(env, &refcheck);
1164 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1165 obd_off start, obd_off end, int mode,
1166 __u64 *cookie , __u32 enqflags)
1169 struct cl_lock *lck;
1170 struct cl_object *obj;
1171 struct cl_lock_descr *descr;
1172 struct echo_thread_info *info;
1176 info = echo_env_info(env);
1178 descr = &info->eti_descr;
1179 obj = echo_obj2cl(eco);
1181 descr->cld_obj = obj;
1182 descr->cld_start = cl_index(obj, start);
1183 descr->cld_end = cl_index(obj, end);
1184 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1185 descr->cld_enq_flags = enqflags;
1188 lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1190 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1191 struct echo_lock *el;
1193 rc = cl_wait(env, lck);
1195 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1196 spin_lock(&ec->ec_lock);
1197 if (cfs_list_empty(&el->el_chain)) {
1198 cfs_list_add(&el->el_chain, &ec->ec_locks);
1199 el->el_cookie = ++ec->ec_unique;
1201 cfs_atomic_inc(&el->el_refcount);
1202 *cookie = el->el_cookie;
1203 spin_unlock(&ec->ec_lock);
1205 cl_lock_release(env, lck, "ec enqueue", cfs_current());
1211 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1212 int mode, __u64 *cookie)
1214 struct echo_thread_info *info;
1221 env = cl_env_get(&refcheck);
1223 RETURN(PTR_ERR(env));
1225 info = echo_env_info(env);
1228 io->ci_ignore_layout = 1;
1229 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1232 LASSERT(result == 0);
1234 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1235 cl_io_fini(env, io);
1239 cl_env_put(env, &refcheck);
1243 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1246 struct echo_client_obd *ec = ed->ed_ec;
1247 struct echo_lock *ecl = NULL;
1249 int found = 0, still_used = 0;
1252 LASSERT(ec != NULL);
1253 spin_lock(&ec->ec_lock);
1254 cfs_list_for_each (el, &ec->ec_locks) {
1255 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1256 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1257 found = (ecl->el_cookie == cookie);
1259 if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1260 cfs_list_del_init(&ecl->el_chain);
1266 spin_unlock(&ec->ec_lock);
1271 echo_lock_release(env, ecl, still_used);
1275 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1282 env = cl_env_get(&refcheck);
1284 RETURN(PTR_ERR(env));
1286 rc = cl_echo_cancel0(env, ed, cookie);
1288 cl_env_put(env, &refcheck);
1292 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1293 enum cl_req_type unused, struct cl_2queue *queue)
1295 struct cl_page *clp;
1296 struct cl_page *temp;
1300 cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1302 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1305 result = result ?: rc;
1310 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1311 cfs_page_t **pages, int npages, int async)
1314 struct echo_thread_info *info;
1315 struct cl_object *obj = echo_obj2cl(eco);
1316 struct echo_device *ed = eco->eo_dev;
1317 struct cl_2queue *queue;
1319 struct cl_page *clp;
1320 struct lustre_handle lh = { 0 };
1321 int page_size = cl_page_size(obj);
1327 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1328 LASSERT(ed->ed_next != NULL);
1329 env = cl_env_get(&refcheck);
1331 RETURN(PTR_ERR(env));
1333 info = echo_env_info(env);
1335 queue = &info->eti_queue;
1337 cl_2queue_init(queue);
1339 io->ci_ignore_layout = 1;
1340 rc = cl_io_init(env, io, CIT_MISC, obj);
1346 rc = cl_echo_enqueue0(env, eco, offset,
1347 offset + npages * CFS_PAGE_SIZE - 1,
1348 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1351 GOTO(error_lock, rc);
1353 for (i = 0; i < npages; i++) {
1355 clp = cl_page_find(env, obj, cl_index(obj, offset),
1356 pages[i], CPT_TRANSIENT);
1361 LASSERT(clp->cp_type == CPT_TRANSIENT);
1363 rc = cl_page_own(env, io, clp);
1365 LASSERT(clp->cp_state == CPS_FREEING);
1366 cl_page_put(env, clp);
1370 cl_2queue_add(queue, clp);
1372 /* drop the reference count for cl_page_find, so that the page
1373 * will be freed in cl_2queue_fini. */
1374 cl_page_put(env, clp);
1375 cl_page_clip(env, clp, 0, page_size);
1377 offset += page_size;
1381 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1383 async = async && (typ == CRT_WRITE);
1385 rc = cl_echo_async_brw(env, io, typ, queue);
1387 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1388 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1389 async ? "async" : "sync", rc);
1392 cl_echo_cancel0(env, ed, lh.cookie);
1395 cl_2queue_discard(env, io, queue);
1396 cl_2queue_disown(env, io, queue);
1397 cl_2queue_fini(env, queue);
1398 cl_io_fini(env, io);
1400 cl_env_put(env, &refcheck);
1403 /** @} echo_exports */
1406 static obd_id last_object_id;
1409 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1411 struct lov_stripe_md *ulsm = _ulsm;
1414 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1418 if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm)))
1421 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1422 if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1423 sizeof(lsm->lsm_oinfo[0])))
1430 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1431 void *ulsm, int ulsm_nob)
1433 struct echo_client_obd *ec = ed->ed_ec;
1436 if (ulsm_nob < sizeof (*lsm))
1439 if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm)))
1442 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1443 lsm->lsm_magic != LOV_MAGIC ||
1444 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1445 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1449 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1450 if (cfs_copy_from_user(lsm->lsm_oinfo[i],
1451 ((struct lov_stripe_md *)ulsm)-> \
1453 sizeof(lsm->lsm_oinfo[0])))
1459 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1462 sprintf(name, LPU64, id);
1463 lname->ln_name = name;
1464 lname->ln_namelen = strlen(name);
1467 /* similar to mdt_attr_get_complex */
1468 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1471 struct echo_thread_info *info = echo_env_info(env);
1476 LASSERT(ma->ma_lmm_size > 0);
1478 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1482 /* big_lmm may need to be grown */
1483 if (info->eti_big_lmmsize < rc) {
1484 int size = size_roundup_power2(rc);
1486 if (info->eti_big_lmmsize > 0) {
1487 /* free old buffer */
1488 LASSERT(info->eti_big_lmm);
1489 OBD_FREE_LARGE(info->eti_big_lmm,
1490 info->eti_big_lmmsize);
1491 info->eti_big_lmm = NULL;
1492 info->eti_big_lmmsize = 0;
1495 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1496 if (info->eti_big_lmm == NULL)
1498 info->eti_big_lmmsize = size;
1500 LASSERT(info->eti_big_lmmsize >= rc);
1502 info->eti_buf.lb_buf = info->eti_big_lmm;
1503 info->eti_buf.lb_len = info->eti_big_lmmsize;
1504 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1508 ma->ma_valid |= MA_LOV;
1509 ma->ma_lmm = info->eti_big_lmm;
1510 ma->ma_lmm_size = rc;
1515 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1518 struct echo_thread_info *info = echo_env_info(env);
1519 struct lu_buf *buf = &info->eti_buf;
1520 cfs_umode_t mode = lu_object_attr(&next->mo_lu);
1521 int need = ma->ma_need;
1528 if (need & MA_INODE) {
1529 ma->ma_need = MA_INODE;
1530 rc = mo_attr_get(env, next, ma);
1533 ma->ma_valid |= MA_INODE;
1536 if (need & MA_LOV) {
1537 if (S_ISREG(mode) || S_ISDIR(mode)) {
1538 LASSERT(ma->ma_lmm_size > 0);
1539 buf->lb_buf = ma->ma_lmm;
1540 buf->lb_len = ma->ma_lmm_size;
1541 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1543 ma->ma_lmm_size = rc2;
1544 ma->ma_valid |= MA_LOV;
1545 } else if (rc2 == -ENODATA) {
1547 ma->ma_lmm_size = 0;
1548 } else if (rc2 == -ERANGE) {
1549 rc2 = echo_big_lmm_get(env, next, ma);
1551 GOTO(out, rc = rc2);
1553 GOTO(out, rc = rc2);
1558 #ifdef CONFIG_FS_POSIX_ACL
1559 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1560 buf->lb_buf = ma->ma_acl;
1561 buf->lb_len = ma->ma_acl_size;
1562 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1564 ma->ma_acl_size = rc2;
1565 ma->ma_valid |= MA_ACL_DEF;
1566 } else if (rc2 == -ENODATA) {
1568 ma->ma_acl_size = 0;
1570 GOTO(out, rc = rc2);
1576 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1577 rc, ma->ma_valid, ma->ma_lmm);
1582 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1583 struct md_object *parent, struct lu_fid *fid,
1584 struct lu_name *lname, struct md_op_spec *spec,
1587 struct lu_object *ec_child, *child;
1588 struct lu_device *ld = ed->ed_next;
1589 struct echo_thread_info *info = echo_env_info(env);
1590 struct lu_fid *fid2 = &info->eti_fid2;
1591 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1596 rc = mdo_lookup(env, parent, lname, fid2, spec);
1599 else if (rc != -ENOENT)
1602 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1604 if (IS_ERR(ec_child)) {
1605 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1607 RETURN(PTR_ERR(ec_child));
1610 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1611 if (child == NULL) {
1612 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1613 GOTO(out_put, rc = -EINVAL);
1616 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1617 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1620 * Do not perform lookup sanity check. We know that name does not exist.
1622 spec->sp_cr_lookup = 0;
1623 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1625 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1628 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1629 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1632 lu_object_put(env, ec_child);
1636 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1639 struct echo_thread_info *info = echo_env_info(env);
1641 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1642 ma->ma_lmm = (void *)&info->eti_lmm;
1643 ma->ma_lmm_size = sizeof(info->eti_lmm);
1645 LASSERT(info->eti_big_lmmsize);
1646 ma->ma_lmm = info->eti_big_lmm;
1647 ma->ma_lmm_size = info->eti_big_lmmsize;
1653 static int echo_create_md_object(const struct lu_env *env,
1654 struct echo_device *ed,
1655 struct lu_object *ec_parent,
1657 char *name, int namelen,
1658 __u64 id, __u32 mode, int count,
1659 int stripe_count, int stripe_offset)
1661 struct lu_object *parent;
1662 struct echo_thread_info *info = echo_env_info(env);
1663 struct lu_name *lname = &info->eti_lname;
1664 struct md_op_spec *spec = &info->eti_spec;
1665 struct md_attr *ma = &info->eti_ma;
1666 struct lu_device *ld = ed->ed_next;
1672 if (ec_parent == NULL)
1674 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1678 memset(ma, 0, sizeof(*ma));
1679 memset(spec, 0, sizeof(*spec));
1680 if (stripe_count != 0) {
1681 spec->sp_cr_flags |= FMODE_WRITE;
1682 echo_set_lmm_size(env, ld, ma);
1683 if (stripe_count != -1) {
1684 struct lov_user_md_v3 *lum = &info->eti_lum;
1686 lum->lmm_magic = LOV_USER_MAGIC_V3;
1687 lum->lmm_stripe_count = stripe_count;
1688 lum->lmm_stripe_offset = stripe_offset;
1689 lum->lmm_pattern = 0;
1690 spec->u.sp_ea.eadata = lum;
1691 spec->u.sp_ea.eadatalen = sizeof(*lum);
1692 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1696 ma->ma_attr.la_mode = mode;
1697 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1698 ma->ma_attr.la_ctime = cfs_time_current_64();
1701 lname->ln_name = name;
1702 lname->ln_namelen = namelen;
1703 /* If name is specified, only create one object by name */
1704 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1709 /* Create multiple object sequenced by id */
1710 for (i = 0; i < count; i++) {
1711 char *tmp_name = info->eti_name;
1713 echo_md_build_name(lname, tmp_name, id);
1715 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1718 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1729 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1730 struct echo_device *ed,
1731 struct md_object *parent,
1732 struct lu_name *lname)
1734 struct echo_thread_info *info = echo_env_info(env);
1735 struct lu_fid *fid = &info->eti_fid;
1736 struct lu_object *child;
1740 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1742 rc = mdo_lookup(env, parent, lname, fid, NULL);
1744 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1745 RETURN(ERR_PTR(rc));
1748 /* In the function below, .hs_keycmp resolves to
1749 * lu_obj_hop_keycmp() */
1750 /* coverity[overrun-buffer-val] */
1751 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1756 static int echo_setattr_object(const struct lu_env *env,
1757 struct echo_device *ed,
1758 struct lu_object *ec_parent,
1759 __u64 id, int count)
1761 struct lu_object *parent;
1762 struct echo_thread_info *info = echo_env_info(env);
1763 struct lu_name *lname = &info->eti_lname;
1764 char *name = info->eti_name;
1765 struct lu_device *ld = ed->ed_next;
1766 struct lu_buf *buf = &info->eti_buf;
1772 if (ec_parent == NULL)
1774 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1778 for (i = 0; i < count; i++) {
1779 struct lu_object *ec_child, *child;
1781 echo_md_build_name(lname, name, id);
1783 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1784 if (IS_ERR(ec_child)) {
1785 CERROR("Can't find child %s: rc = %ld\n",
1786 lname->ln_name, PTR_ERR(ec_child));
1787 RETURN(PTR_ERR(ec_child));
1790 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1791 if (child == NULL) {
1792 CERROR("Can not locate the child %s\n", lname->ln_name);
1793 lu_object_put(env, ec_child);
1798 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1799 PFID(lu_object_fid(child)));
1801 buf->lb_buf = info->eti_xattr_buf;
1802 buf->lb_len = sizeof(info->eti_xattr_buf);
1804 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1805 rc = mo_xattr_set(env, lu2md(child), buf, name,
1808 CERROR("Can not setattr child "DFID": rc = %d\n",
1809 PFID(lu_object_fid(child)), rc);
1810 lu_object_put(env, ec_child);
1813 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1814 PFID(lu_object_fid(child)));
1816 lu_object_put(env, ec_child);
1821 static int echo_getattr_object(const struct lu_env *env,
1822 struct echo_device *ed,
1823 struct lu_object *ec_parent,
1824 __u64 id, int count)
1826 struct lu_object *parent;
1827 struct echo_thread_info *info = echo_env_info(env);
1828 struct lu_name *lname = &info->eti_lname;
1829 char *name = info->eti_name;
1830 struct md_attr *ma = &info->eti_ma;
1831 struct lu_device *ld = ed->ed_next;
1837 if (ec_parent == NULL)
1839 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1843 memset(ma, 0, sizeof(*ma));
1844 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1845 ma->ma_acl = info->eti_xattr_buf;
1846 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1848 for (i = 0; i < count; i++) {
1849 struct lu_object *ec_child, *child;
1852 echo_md_build_name(lname, name, id);
1853 echo_set_lmm_size(env, ld, ma);
1855 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1856 if (IS_ERR(ec_child)) {
1857 CERROR("Can't find child %s: rc = %ld\n",
1858 lname->ln_name, PTR_ERR(ec_child));
1859 RETURN(PTR_ERR(ec_child));
1862 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1863 if (child == NULL) {
1864 CERROR("Can not locate the child %s\n", lname->ln_name);
1865 lu_object_put(env, ec_child);
1869 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1870 PFID(lu_object_fid(child)));
1871 rc = echo_attr_get_complex(env, lu2md(child), ma);
1873 CERROR("Can not getattr child "DFID": rc = %d\n",
1874 PFID(lu_object_fid(child)), rc);
1875 lu_object_put(env, ec_child);
1878 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1879 PFID(lu_object_fid(child)));
1881 lu_object_put(env, ec_child);
1887 static int echo_lookup_object(const struct lu_env *env,
1888 struct echo_device *ed,
1889 struct lu_object *ec_parent,
1890 __u64 id, int count)
1892 struct lu_object *parent;
1893 struct echo_thread_info *info = echo_env_info(env);
1894 struct lu_name *lname = &info->eti_lname;
1895 char *name = info->eti_name;
1896 struct lu_fid *fid = &info->eti_fid;
1897 struct lu_device *ld = ed->ed_next;
1901 if (ec_parent == NULL)
1903 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1907 /*prepare the requests*/
1908 for (i = 0; i < count; i++) {
1909 echo_md_build_name(lname, name, id);
1911 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1912 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1914 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1916 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1919 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1920 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1927 static int echo_md_destroy_internal(const struct lu_env *env,
1928 struct echo_device *ed,
1929 struct md_object *parent,
1930 struct lu_name *lname,
1933 struct lu_device *ld = ed->ed_next;
1934 struct lu_object *ec_child;
1935 struct lu_object *child;
1940 ec_child = echo_md_lookup(env, ed, parent, lname);
1941 if (IS_ERR(ec_child)) {
1942 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1944 RETURN(PTR_ERR(ec_child));
1947 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1948 if (child == NULL) {
1949 CERROR("Can not locate the child %s\n", lname->ln_name);
1950 GOTO(out_put, rc = -EINVAL);
1953 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1954 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1956 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1958 CERROR("Can not unlink child %s: rc = %d\n",
1959 lname->ln_name, rc);
1962 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1963 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1965 lu_object_put(env, ec_child);
1969 static int echo_destroy_object(const struct lu_env *env,
1970 struct echo_device *ed,
1971 struct lu_object *ec_parent,
1972 char *name, int namelen,
1973 __u64 id, __u32 mode,
1976 struct echo_thread_info *info = echo_env_info(env);
1977 struct lu_name *lname = &info->eti_lname;
1978 struct md_attr *ma = &info->eti_ma;
1979 struct lu_device *ld = ed->ed_next;
1980 struct lu_object *parent;
1985 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1989 memset(ma, 0, sizeof(*ma));
1990 ma->ma_attr.la_mode = mode;
1991 ma->ma_attr.la_valid = LA_CTIME;
1992 ma->ma_attr.la_ctime = cfs_time_current_64();
1993 ma->ma_need = MA_INODE;
1997 lname->ln_name = name;
1998 lname->ln_namelen = namelen;
1999 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2004 /*prepare the requests*/
2005 for (i = 0; i < count; i++) {
2006 char *tmp_name = info->eti_name;
2009 echo_md_build_name(lname, tmp_name, id);
2011 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2014 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2023 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2024 struct echo_device *ed, char *path,
2027 struct lu_device *ld = ed->ed_next;
2028 struct md_device *md = lu2md_dev(ld);
2029 struct echo_thread_info *info = echo_env_info(env);
2030 struct lu_fid *fid = &info->eti_fid;
2031 struct lu_name *lname = &info->eti_lname;
2032 struct lu_object *parent = NULL;
2033 struct lu_object *child = NULL;
2037 /*Only support MDD layer right now*/
2038 rc = md->md_ops->mdo_root_get(env, md, fid);
2040 CERROR("get root error: rc = %d\n", rc);
2041 RETURN(ERR_PTR(rc));
2044 /* In the function below, .hs_keycmp resolves to
2045 * lu_obj_hop_keycmp() */
2046 /* coverity[overrun-buffer-val] */
2047 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2048 if (IS_ERR(parent)) {
2049 CERROR("Can not find the parent "DFID": rc = %ld\n",
2050 PFID(fid), PTR_ERR(parent));
2055 struct lu_object *ld_parent;
2058 e = strsep(&path, "/");
2063 if (!path || path[0] == '\0')
2069 lname->ln_namelen = strlen(e);
2071 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2072 if (ld_parent == NULL) {
2073 lu_object_put(env, parent);
2078 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2079 lu_object_put(env, parent);
2080 if (IS_ERR(child)) {
2081 rc = (int)PTR_ERR(child);
2082 CERROR("lookup %s under parent "DFID": rc = %d\n",
2083 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2090 RETURN(ERR_PTR(rc));
2095 static void echo_ucred_init(struct lu_env *env)
2097 struct lu_ucred *ucred = lu_ucred(env);
2099 ucred->uc_valid = UCRED_INVALID;
2101 ucred->uc_suppgids[0] = -1;
2102 ucred->uc_suppgids[1] = -1;
2104 ucred->uc_uid = ucred->uc_o_uid = cfs_curproc_uid();
2105 ucred->uc_gid = ucred->uc_o_gid = cfs_curproc_gid();
2106 ucred->uc_fsuid = ucred->uc_o_fsuid = cfs_curproc_fsuid();
2107 ucred->uc_fsgid = ucred->uc_o_fsgid = cfs_curproc_fsgid();
2108 ucred->uc_cap = cfs_curproc_cap_pack();
2110 /* remove fs privilege for non-root user. */
2111 if (ucred->uc_fsuid)
2112 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2113 ucred->uc_valid = UCRED_NEW;
2116 static void echo_ucred_fini(struct lu_env *env)
2118 struct lu_ucred *ucred = lu_ucred(env);
2119 ucred->uc_valid = UCRED_INIT;
2122 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2123 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2124 static int echo_md_handler(struct echo_device *ed, int command,
2125 char *path, int path_len, __u64 id, int count,
2126 struct obd_ioctl_data *data)
2128 struct echo_thread_info *info;
2129 struct lu_device *ld = ed->ed_next;
2132 struct lu_object *parent;
2134 int namelen = data->ioc_plen2;
2139 CERROR("MD echo client is not being initialized properly\n");
2143 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2144 CERROR("Only support MDD layer right now!\n");
2148 env = cl_env_get(&refcheck);
2150 RETURN(PTR_ERR(env));
2152 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2156 /* init big_lmm buffer */
2157 info = echo_env_info(env);
2158 LASSERT(info->eti_big_lmm == NULL);
2159 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2160 if (info->eti_big_lmm == NULL)
2161 GOTO(out_env, rc = -ENOMEM);
2162 info->eti_big_lmmsize = MIN_MD_SIZE;
2164 parent = echo_resolve_path(env, ed, path, path_len);
2165 if (IS_ERR(parent)) {
2166 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2168 GOTO(out_free, rc = PTR_ERR(parent));
2172 OBD_ALLOC(name, namelen + 1);
2174 GOTO(out_put, rc = -ENOMEM);
2175 if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen))
2176 GOTO(out_name, rc = -EFAULT);
2179 echo_ucred_init(env);
2182 case ECHO_MD_CREATE:
2183 case ECHO_MD_MKDIR: {
2184 struct echo_thread_info *info = echo_env_info(env);
2185 __u32 mode = data->ioc_obdo2.o_mode;
2186 struct lu_fid *fid = &info->eti_fid;
2187 int stripe_count = (int)data->ioc_obdo2.o_misc;
2188 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2190 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2194 /* In the function below, .hs_keycmp resolves to
2195 * lu_obj_hop_keycmp() */
2196 /* coverity[overrun-buffer-val] */
2197 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2198 id, mode, count, stripe_count,
2202 case ECHO_MD_DESTROY:
2203 case ECHO_MD_RMDIR: {
2204 __u32 mode = data->ioc_obdo2.o_mode;
2206 rc = echo_destroy_object(env, ed, parent, name, namelen,
2210 case ECHO_MD_LOOKUP:
2211 rc = echo_lookup_object(env, ed, parent, id, count);
2213 case ECHO_MD_GETATTR:
2214 rc = echo_getattr_object(env, ed, parent, id, count);
2216 case ECHO_MD_SETATTR:
2217 rc = echo_setattr_object(env, ed, parent, id, count);
2220 CERROR("unknown command %d\n", command);
2224 echo_ucred_fini(env);
2228 OBD_FREE(name, namelen + 1);
2230 lu_object_put(env, parent);
2232 LASSERT(info->eti_big_lmm);
2233 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2234 info->eti_big_lmm = NULL;
2235 info->eti_big_lmmsize = 0;
2237 cl_env_put(env, &refcheck);
2241 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2242 int on_target, struct obdo *oa, void *ulsm,
2243 int ulsm_nob, struct obd_trans_info *oti)
2245 struct echo_object *eco;
2246 struct echo_client_obd *ec = ed->ed_ec;
2247 struct lov_stripe_md *lsm = NULL;
2252 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2253 (on_target || /* set_stripe */
2254 ec->ec_nstripes != 0)) { /* LOV */
2255 CERROR ("No valid oid\n");
2259 rc = echo_alloc_memmd(ed, &lsm);
2261 CERROR("Cannot allocate md: rc = %d\n", rc);
2268 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2272 if (lsm->lsm_stripe_count == 0)
2273 lsm->lsm_stripe_count = ec->ec_nstripes;
2275 if (lsm->lsm_stripe_size == 0)
2276 lsm->lsm_stripe_size = CFS_PAGE_SIZE;
2280 /* setup stripes: indices + default ids if required */
2281 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2282 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2283 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2285 lsm->lsm_oinfo[i]->loi_ost_idx =
2286 (idx + i) % ec->ec_nstripes;
2290 /* setup object ID here for !on_target and LOV hint */
2291 if (oa->o_valid & OBD_MD_FLID) {
2292 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2293 lsm->lsm_oi = oa->o_oi;
2296 if (ostid_id(&lsm->lsm_oi) == 0)
2297 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2301 /* Only echo objects are allowed to be created */
2302 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2303 (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2304 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2306 CERROR("Cannot create objects: rc = %d\n", rc);
2312 /* See what object ID we were given */
2313 oa->o_oi = lsm->lsm_oi;
2314 oa->o_valid |= OBD_MD_FLID;
2316 eco = cl_echo_object_find(ed, &lsm);
2318 GOTO(failed, rc = PTR_ERR(eco));
2319 cl_echo_object_put(eco);
2321 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2326 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2328 echo_free_memmd(ed, &lsm);
2330 CERROR("create object failed with: rc = %d\n", rc);
2334 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2337 struct lov_stripe_md *lsm = NULL;
2338 struct echo_object *eco;
2342 if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2343 /* disallow use of object id 0 */
2344 CERROR ("No valid oid\n");
2348 rc = echo_alloc_memmd(ed, &lsm);
2352 lsm->lsm_oi = oa->o_oi;
2353 if (!(oa->o_valid & OBD_MD_FLGROUP))
2354 ostid_set_seq_echo(&lsm->lsm_oi);
2357 eco = cl_echo_object_find(ed, &lsm);
2363 echo_free_memmd(ed, &lsm);
2367 static void echo_put_object(struct echo_object *eco)
2369 if (cl_echo_object_put(eco))
2370 CERROR("echo client: drop an object failed");
2374 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2376 unsigned long stripe_count;
2377 unsigned long stripe_size;
2378 unsigned long width;
2379 unsigned long woffset;
2383 if (lsm->lsm_stripe_count <= 1)
2387 stripe_size = lsm->lsm_stripe_size;
2388 stripe_count = lsm->lsm_stripe_count;
2390 /* width = # bytes in all stripes */
2391 width = stripe_size * stripe_count;
2393 /* woffset = offset within a width; offset = whole number of widths */
2394 woffset = do_div (offset, width);
2396 stripe_index = woffset / stripe_size;
2398 *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2399 *offp = offset * stripe_size + woffset % stripe_size;
2403 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2404 cfs_page_t *page, int rw, obd_id id,
2405 obd_off offset, obd_off count)
2412 /* no partial pages on the client */
2413 LASSERT(count == CFS_PAGE_SIZE);
2415 addr = cfs_kmap(page);
2417 for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2418 if (rw == OBD_BRW_WRITE) {
2419 stripe_off = offset + delta;
2421 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2423 stripe_off = 0xdeadbeef00c0ffeeULL;
2424 stripe_id = 0xdeadbeef00c0ffeeULL;
2426 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2427 stripe_off, stripe_id);
2433 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2434 cfs_page_t *page, obd_id id,
2435 obd_off offset, obd_off count)
2444 /* no partial pages on the client */
2445 LASSERT(count == CFS_PAGE_SIZE);
2447 addr = cfs_kmap(page);
2449 for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2450 stripe_off = offset + delta;
2452 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2454 rc2 = block_debug_check("test_brw",
2455 addr + delta, OBD_ECHO_BLOCK_SIZE,
2456 stripe_off, stripe_id);
2458 CERROR ("Error in echo object "LPX64"\n", id);
2467 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2468 struct echo_object *eco, obd_off offset,
2469 obd_size count, int async,
2470 struct obd_trans_info *oti)
2472 struct lov_stripe_md *lsm = eco->eo_lsm;
2474 struct brw_page *pga;
2475 struct brw_page *pgp;
2485 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2486 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2487 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2489 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
2491 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2492 LASSERT(lsm != NULL);
2493 LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2496 (count & (~CFS_PAGE_MASK)) != 0)
2499 /* XXX think again with misaligned I/O */
2500 npages = count >> CFS_PAGE_SHIFT;
2502 if (rw == OBD_BRW_WRITE)
2503 brw_flags = OBD_BRW_ASYNC;
2505 OBD_ALLOC(pga, npages * sizeof(*pga));
2509 OBD_ALLOC(pages, npages * sizeof(*pages));
2510 if (pages == NULL) {
2511 OBD_FREE(pga, npages * sizeof(*pga));
2515 for (i = 0, pgp = pga, off = offset;
2517 i++, pgp++, off += CFS_PAGE_SIZE) {
2519 LASSERT (pgp->pg == NULL); /* for cleanup */
2522 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2523 if (pgp->pg == NULL)
2527 pgp->count = CFS_PAGE_SIZE;
2529 pgp->flag = brw_flags;
2532 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2533 ostid_id(&oa->o_oi), off,
2537 /* brw mode can only be used at client */
2538 LASSERT(ed->ed_next != NULL);
2539 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2542 if (rc != 0 || rw != OBD_BRW_READ)
2545 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2546 if (pgp->pg == NULL)
2551 vrc = echo_client_page_debug_check(lsm, pgp->pg,
2552 ostid_id(&oa->o_oi),
2553 pgp->off, pgp->count);
2554 if (vrc != 0 && rc == 0)
2557 OBD_PAGE_FREE(pgp->pg);
2559 OBD_FREE(pga, npages * sizeof(*pga));
2560 OBD_FREE(pages, npages * sizeof(*pages));
2564 static int echo_client_prep_commit(const struct lu_env *env,
2565 struct obd_export *exp, int rw,
2566 struct obdo *oa, struct echo_object *eco,
2567 obd_off offset, obd_size count,
2568 obd_size batch, struct obd_trans_info *oti,
2571 struct lov_stripe_md *lsm = eco->eo_lsm;
2572 struct obd_ioobj ioo;
2573 struct niobuf_local *lnb;
2574 struct niobuf_remote *rnb;
2576 obd_size npages, tot_pages;
2577 int i, ret = 0, brw_flags = 0;
2581 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2582 (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2585 npages = batch >> CFS_PAGE_SHIFT;
2586 tot_pages = count >> CFS_PAGE_SHIFT;
2588 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2589 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2591 if (lnb == NULL || rnb == NULL)
2592 GOTO(out, ret = -ENOMEM);
2594 if (rw == OBD_BRW_WRITE && async)
2595 brw_flags |= OBD_BRW_ASYNC;
2597 obdo_to_ioobj(oa, &ioo);
2601 for(; tot_pages; tot_pages -= npages) {
2604 if (tot_pages < npages)
2607 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
2608 rnb[i].offset = off;
2609 rnb[i].len = CFS_PAGE_SIZE;
2610 rnb[i].flags = brw_flags;
2613 ioo.ioo_bufcnt = npages;
2614 oti->oti_transno = 0;
2617 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2621 LASSERT(lpages == npages);
2623 for (i = 0; i < lpages; i++) {
2624 cfs_page_t *page = lnb[i].page;
2626 /* read past eof? */
2627 if (page == NULL && lnb[i].rc == 0)
2631 lnb[i].flags |= OBD_BRW_ASYNC;
2633 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2634 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2635 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2638 if (rw == OBD_BRW_WRITE)
2639 echo_client_page_debug_setup(lsm, page, rw,
2640 ostid_id(&oa->o_oi),
2644 echo_client_page_debug_check(lsm, page,
2645 ostid_id(&oa->o_oi),
2650 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2651 rnb, npages, lnb, oti, ret);
2655 /* Reset oti otherwise it would confuse ldiskfs. */
2656 memset(oti, 0, sizeof(*oti));
2661 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2663 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2667 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2668 struct obd_export *exp,
2669 struct obd_ioctl_data *data,
2670 struct obd_trans_info *dummy_oti)
2672 struct obd_device *obd = class_exp2obd(exp);
2673 struct echo_device *ed = obd2echo_dev(obd);
2674 struct echo_client_obd *ec = ed->ed_ec;
2675 struct obdo *oa = &data->ioc_obdo1;
2676 struct echo_object *eco;
2682 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2684 rc = echo_get_object(&eco, ed, oa);
2688 oa->o_valid &= ~OBD_MD_FLHANDLE;
2690 /* OFD/obdfilter works only via prep/commit */
2691 test_mode = (long)data->ioc_pbuf1;
2695 if (ed->ed_next == NULL && test_mode != 3) {
2697 data->ioc_plen1 = data->ioc_count;
2700 /* Truncate batch size to maximum */
2701 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2702 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2704 switch (test_mode) {
2708 rc = echo_client_kbrw(ed, rw, oa,
2709 eco, data->ioc_offset,
2710 data->ioc_count, async, dummy_oti);
2713 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2714 eco, data->ioc_offset,
2715 data->ioc_count, data->ioc_plen1,
2721 echo_put_object(eco);
2726 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2727 int mode, obd_off offset, obd_size nob)
2729 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2730 struct lustre_handle *ulh = &oa->o_handle;
2731 struct echo_object *eco;
2736 if (ed->ed_next == NULL)
2737 RETURN(-EOPNOTSUPP);
2739 if (!(mode == LCK_PR || mode == LCK_PW))
2742 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2743 (nob & (~CFS_PAGE_MASK)) != 0)
2746 rc = echo_get_object (&eco, ed, oa);
2750 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2751 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2753 oa->o_valid |= OBD_MD_FLHANDLE;
2754 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2756 echo_put_object(eco);
2761 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2763 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2764 __u64 cookie = oa->o_handle.cookie;
2766 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2769 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2770 return cl_echo_cancel(ed, cookie);
2774 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2775 void *karg, void *uarg)
2777 struct obd_device *obd = exp->exp_obd;
2778 struct echo_device *ed = obd2echo_dev(obd);
2779 struct echo_client_obd *ec = ed->ed_ec;
2780 struct echo_object *eco;
2781 struct obd_ioctl_data *data = karg;
2782 struct obd_trans_info dummy_oti;
2784 struct oti_req_ack_lock *ack_lock;
2787 int rw = OBD_BRW_READ;
2792 memset(&dummy_oti, 0, sizeof(dummy_oti));
2794 oa = &data->ioc_obdo1;
2795 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2796 oa->o_valid |= OBD_MD_FLGROUP;
2797 ostid_set_seq_echo(&oa->o_oi);
2800 /* This FID is unpacked just for validation at this point */
2801 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2809 rc = lu_env_init(env, LCT_DT_THREAD);
2811 GOTO(out, rc = -ENOMEM);
2814 case OBD_IOC_CREATE: /* may create echo object */
2815 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2816 GOTO (out, rc = -EPERM);
2818 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2819 data->ioc_plen1, &dummy_oti);
2822 case OBD_IOC_ECHO_MD: {
2829 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2830 GOTO(out, rc = -EPERM);
2832 count = data->ioc_count;
2833 cmd = data->ioc_command;
2835 id = ostid_id(&data->ioc_obdo2.o_oi);
2837 dirlen = data->ioc_plen1;
2838 OBD_ALLOC(dir, dirlen + 1);
2840 GOTO(out, rc = -ENOMEM);
2842 if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2843 OBD_FREE(dir, data->ioc_plen1 + 1);
2844 GOTO(out, rc = -EFAULT);
2847 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2848 OBD_FREE(dir, dirlen + 1);
2851 case OBD_IOC_ECHO_ALLOC_SEQ: {
2852 struct lu_env *cl_env;
2857 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2858 GOTO(out, rc = -EPERM);
2860 cl_env = cl_env_get(&refcheck);
2862 GOTO(out, rc = PTR_ERR(cl_env));
2864 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2867 cl_env_put(cl_env, &refcheck);
2871 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2872 cl_env_put(cl_env, &refcheck);
2874 CERROR("%s: Can not alloc seq: rc = %d\n",
2879 if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2882 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2883 if (cfs_copy_to_user(data->ioc_pbuf2, &max_count,
2888 case OBD_IOC_DESTROY:
2889 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2890 GOTO (out, rc = -EPERM);
2892 rc = echo_get_object(&eco, ed, oa);
2894 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2895 &dummy_oti, NULL, NULL);
2897 eco->eo_deleted = 1;
2898 echo_put_object(eco);
2902 case OBD_IOC_GETATTR:
2903 rc = echo_get_object(&eco, ed, oa);
2905 struct obd_info oinfo = { { { 0 } } };
2906 oinfo.oi_md = eco->eo_lsm;
2908 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2909 echo_put_object(eco);
2913 case OBD_IOC_SETATTR:
2914 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2915 GOTO (out, rc = -EPERM);
2917 rc = echo_get_object(&eco, ed, oa);
2919 struct obd_info oinfo = { { { 0 } } };
2921 oinfo.oi_md = eco->eo_lsm;
2923 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2924 echo_put_object(eco);
2928 case OBD_IOC_BRW_WRITE:
2929 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2930 GOTO (out, rc = -EPERM);
2934 case OBD_IOC_BRW_READ:
2935 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2938 case ECHO_IOC_GET_STRIPE:
2939 rc = echo_get_object(&eco, ed, oa);
2941 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2943 echo_put_object(eco);
2947 case ECHO_IOC_SET_STRIPE:
2948 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2949 GOTO (out, rc = -EPERM);
2951 if (data->ioc_pbuf1 == NULL) { /* unset */
2952 rc = echo_get_object(&eco, ed, oa);
2954 eco->eo_deleted = 1;
2955 echo_put_object(eco);
2958 rc = echo_create_object(env, ed, 0, oa,
2960 data->ioc_plen1, &dummy_oti);
2964 case ECHO_IOC_ENQUEUE:
2965 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2966 GOTO (out, rc = -EPERM);
2968 rc = echo_client_enqueue(exp, oa,
2969 data->ioc_conn1, /* lock mode */
2971 data->ioc_count);/*extent*/
2974 case ECHO_IOC_CANCEL:
2975 rc = echo_client_cancel(exp, oa);
2979 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2980 GOTO (out, rc = -ENOTTY);
2988 /* XXX this should be in a helper also called by target_send_reply */
2989 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2991 if (!ack_lock->mode)
2993 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
2999 static int echo_client_setup(const struct lu_env *env,
3000 struct obd_device *obddev, struct lustre_cfg *lcfg)
3002 struct echo_client_obd *ec = &obddev->u.echo_client;
3003 struct obd_device *tgt;
3004 struct obd_uuid echo_uuid = { "ECHO_UUID" };
3005 struct obd_connect_data *ocd = NULL;
3009 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3010 CERROR("requires a TARGET OBD name\n");
3014 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3015 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3016 CERROR("device not attached or not set up (%s)\n",
3017 lustre_cfg_string(lcfg, 1));
3021 spin_lock_init(&ec->ec_lock);
3022 CFS_INIT_LIST_HEAD (&ec->ec_objects);
3023 CFS_INIT_LIST_HEAD (&ec->ec_locks);
3025 ec->ec_nstripes = 0;
3027 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3028 lu_context_tags_update(ECHO_MD_CTX_TAG);
3029 lu_session_tags_update(ECHO_MD_SES_TAG);
3033 OBD_ALLOC(ocd, sizeof(*ocd));
3035 CERROR("Can't alloc ocd connecting to %s\n",
3036 lustre_cfg_string(lcfg, 1));
3040 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3041 OBD_CONNECT_BRW_SIZE |
3042 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3043 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
3045 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
3046 ocd->ocd_version = LUSTRE_VERSION_CODE;
3047 ocd->ocd_group = FID_SEQ_ECHO;
3049 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3051 /* Turn off pinger because it connects to tgt obd directly. */
3052 spin_lock(&tgt->obd_dev_lock);
3053 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3054 spin_unlock(&tgt->obd_dev_lock);
3057 OBD_FREE(ocd, sizeof(*ocd));
3060 CERROR("fail to connect to device %s\n",
3061 lustre_cfg_string(lcfg, 1));
3068 static int echo_client_cleanup(struct obd_device *obddev)
3070 struct echo_device *ed = obd2echo_dev(obddev);
3071 struct echo_client_obd *ec = &obddev->u.echo_client;
3075 /*Do nothing for Metadata echo client*/
3079 if (ed->ed_next_ismd) {
3080 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3081 lu_session_tags_clear(ECHO_MD_SES_TAG);
3085 if (!cfs_list_empty(&obddev->obd_exports)) {
3086 CERROR("still has clients!\n");
3090 LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3091 rc = obd_disconnect(ec->ec_exp);
3093 CERROR("fail to disconnect device: %d\n", rc);
3098 static int echo_client_connect(const struct lu_env *env,
3099 struct obd_export **exp,
3100 struct obd_device *src, struct obd_uuid *cluuid,
3101 struct obd_connect_data *data, void *localdata)
3104 struct lustre_handle conn = { 0 };
3107 rc = class_connect(&conn, src, cluuid);
3109 *exp = class_conn2export(&conn);
3115 static int echo_client_disconnect(struct obd_export *exp)
3118 struct obd_device *obd;
3119 struct echo_client_obd *ec;
3120 struct ec_lock *ecl;
3126 GOTO(out, rc = -EINVAL);
3130 ec = &obd->u.echo_client;
3132 /* no more contention on export's lock list */
3133 while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
3134 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
3135 struct ec_lock, ecl_exp_chain);
3136 cfs_list_del (&ecl->ecl_exp_chain);
3138 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3139 ecl->ecl_mode, &ecl->ecl_lock_handle);
3141 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3142 "(%d)\n", ecl->ecl_object->eco_id, rc);
3144 echo_put_object (ecl->ecl_object);
3145 OBD_FREE (ecl, sizeof (*ecl));
3149 rc = class_disconnect(exp);
3155 static struct obd_ops echo_client_obd_ops = {
3156 .o_owner = THIS_MODULE,
3159 .o_setup = echo_client_setup,
3160 .o_cleanup = echo_client_cleanup,
3163 .o_iocontrol = echo_client_iocontrol,
3164 .o_connect = echo_client_connect,
3165 .o_disconnect = echo_client_disconnect
3168 int echo_client_init(void)
3170 struct lprocfs_static_vars lvars = { 0 };
3173 lprocfs_echo_init_vars(&lvars);
3175 rc = lu_kmem_init(echo_caches);
3177 rc = class_register_type(&echo_client_obd_ops, NULL,
3179 LUSTRE_ECHO_CLIENT_NAME,
3182 lu_kmem_fini(echo_caches);
3187 void echo_client_exit(void)
3189 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3190 lu_kmem_fini(echo_caches);
3194 static int __init obdecho_init(void)
3196 struct lprocfs_static_vars lvars;
3200 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3202 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3204 lprocfs_echo_init_vars(&lvars);
3206 # ifdef HAVE_SERVER_SUPPORT
3207 rc = echo_persistent_pages_init();
3211 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3212 LUSTRE_ECHO_NAME, NULL);
3217 rc = echo_client_init();
3219 # ifdef HAVE_SERVER_SUPPORT
3223 class_unregister_type(LUSTRE_ECHO_NAME);
3225 echo_persistent_pages_fini();
3231 static void /*__exit*/ obdecho_exit(void)
3235 # ifdef HAVE_SERVER_SUPPORT
3236 class_unregister_type(LUSTRE_ECHO_NAME);
3237 echo_persistent_pages_fini();
3241 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3242 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3243 MODULE_LICENSE("GPL");
3245 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3246 #endif /* __KERNEL__ */
3248 /** @} echo_client */