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 <md_object.h>
51 #include <lustre_fid.h>
52 #include <lustre_acl.h>
53 #include <lustre_net.h>
56 #include "echo_internal.h"
58 /** \defgroup echo_client Echo Client
63 struct cl_device ed_cl;
64 struct echo_client_obd *ed_ec;
66 struct cl_site ed_site_myself;
67 struct cl_site *ed_site;
68 struct lu_device *ed_next;
71 struct lu_client_seq *ed_cl_seq;
75 struct cl_object eo_cl;
76 struct cl_object_header eo_hdr;
78 struct echo_device *eo_dev;
79 cfs_list_t eo_obj_chain;
80 struct lov_stripe_md *eo_lsm;
81 cfs_atomic_t eo_npages;
85 struct echo_object_conf {
86 struct cl_object_conf eoc_cl;
87 struct lov_stripe_md **eoc_md;
91 struct cl_page_slice ep_cl;
93 struct page *ep_vmpage;
97 struct cl_lock_slice el_cl;
99 struct echo_object *el_object;
101 cfs_atomic_t el_refcount;
104 static int echo_client_setup(const struct lu_env *env,
105 struct obd_device *obddev,
106 struct lustre_cfg *lcfg);
107 static int echo_client_cleanup(struct obd_device *obddev);
110 /** \defgroup echo_helpers Helper functions
113 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
115 return container_of0(dev, struct echo_device, ed_cl);
118 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
123 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
125 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
128 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
133 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
135 return container_of(o, struct echo_object, eo_cl);
138 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
140 return container_of(s, struct echo_page, ep_cl);
143 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
145 return container_of(s, struct echo_lock, el_cl);
148 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
150 return ecl->el_cl.cls_lock;
153 static struct lu_context_key echo_thread_key;
154 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
156 struct echo_thread_info *info;
157 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
158 LASSERT(info != NULL);
163 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
165 return container_of(c, struct echo_object_conf, eoc_cl);
168 /** @} echo_helpers */
170 static struct echo_object *cl_echo_object_find(struct echo_device *d,
171 struct lov_stripe_md **lsm);
172 static int cl_echo_object_put(struct echo_object *eco);
173 static int cl_echo_enqueue (struct echo_object *eco, obd_off start,
174 obd_off end, int mode, __u64 *cookie);
175 static int cl_echo_cancel (struct echo_device *d, __u64 cookie);
176 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
177 struct page **pages, int npages, int async);
179 struct echo_thread_info {
180 struct echo_object_conf eti_conf;
181 struct lustre_md eti_md;
183 struct cl_2queue eti_queue;
185 struct cl_lock_descr eti_descr;
186 struct lu_fid eti_fid;
187 struct lu_fid eti_fid2;
188 #ifdef HAVE_SERVER_SUPPORT
189 struct md_op_spec eti_spec;
190 struct lov_mds_md_v3 eti_lmm;
191 struct lov_user_md_v3 eti_lum;
192 struct md_attr eti_ma;
193 struct lu_name eti_lname;
194 /* per-thread values, can be re-used */
198 struct lu_buf eti_buf;
199 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
203 /* No session used right now */
204 struct echo_session_info {
208 static struct kmem_cache *echo_lock_kmem;
209 static struct kmem_cache *echo_object_kmem;
210 static struct kmem_cache *echo_thread_kmem;
211 static struct kmem_cache *echo_session_kmem;
212 /* static struct kmem_cache *echo_req_kmem; */
214 static struct lu_kmem_descr echo_caches[] = {
216 .ckd_cache = &echo_lock_kmem,
217 .ckd_name = "echo_lock_kmem",
218 .ckd_size = sizeof (struct echo_lock)
221 .ckd_cache = &echo_object_kmem,
222 .ckd_name = "echo_object_kmem",
223 .ckd_size = sizeof (struct echo_object)
226 .ckd_cache = &echo_thread_kmem,
227 .ckd_name = "echo_thread_kmem",
228 .ckd_size = sizeof (struct echo_thread_info)
231 .ckd_cache = &echo_session_kmem,
232 .ckd_name = "echo_session_kmem",
233 .ckd_size = sizeof (struct echo_session_info)
240 /** \defgroup echo_page Page operations
242 * Echo page operations.
246 static struct page *echo_page_vmpage(const struct lu_env *env,
247 const struct cl_page_slice *slice)
249 return cl2echo_page(slice)->ep_vmpage;
252 static int echo_page_own(const struct lu_env *env,
253 const struct cl_page_slice *slice,
254 struct cl_io *io, int nonblock)
256 struct echo_page *ep = cl2echo_page(slice);
259 mutex_lock(&ep->ep_lock);
260 else if (!mutex_trylock(&ep->ep_lock))
265 static void echo_page_disown(const struct lu_env *env,
266 const struct cl_page_slice *slice,
269 struct echo_page *ep = cl2echo_page(slice);
271 LASSERT(mutex_is_locked(&ep->ep_lock));
272 mutex_unlock(&ep->ep_lock);
275 static void echo_page_discard(const struct lu_env *env,
276 const struct cl_page_slice *slice,
277 struct cl_io *unused)
279 cl_page_delete(env, slice->cpl_page);
282 static int echo_page_is_vmlocked(const struct lu_env *env,
283 const struct cl_page_slice *slice)
285 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
290 static void echo_page_completion(const struct lu_env *env,
291 const struct cl_page_slice *slice,
294 LASSERT(slice->cpl_page->cp_sync_io != NULL);
297 static void echo_page_fini(const struct lu_env *env,
298 struct cl_page_slice *slice)
300 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
303 cfs_atomic_dec(&eco->eo_npages);
304 page_cache_release(cl2echo_page(slice)->ep_vmpage);
308 static int echo_page_prep(const struct lu_env *env,
309 const struct cl_page_slice *slice,
310 struct cl_io *unused)
315 static int echo_page_print(const struct lu_env *env,
316 const struct cl_page_slice *slice,
317 void *cookie, lu_printer_t printer)
319 struct echo_page *ep = cl2echo_page(slice);
321 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
322 ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
326 static const struct cl_page_operations echo_page_ops = {
327 .cpo_own = echo_page_own,
328 .cpo_disown = echo_page_disown,
329 .cpo_discard = echo_page_discard,
330 .cpo_vmpage = echo_page_vmpage,
331 .cpo_fini = echo_page_fini,
332 .cpo_print = echo_page_print,
333 .cpo_is_vmlocked = echo_page_is_vmlocked,
336 .cpo_prep = echo_page_prep,
337 .cpo_completion = echo_page_completion,
340 .cpo_prep = echo_page_prep,
341 .cpo_completion = echo_page_completion,
347 /** \defgroup echo_lock Locking
349 * echo lock operations
353 static void echo_lock_fini(const struct lu_env *env,
354 struct cl_lock_slice *slice)
356 struct echo_lock *ecl = cl2echo_lock(slice);
358 LASSERT(cfs_list_empty(&ecl->el_chain));
359 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
362 static void echo_lock_delete(const struct lu_env *env,
363 const struct cl_lock_slice *slice)
365 struct echo_lock *ecl = cl2echo_lock(slice);
367 LASSERT(cfs_list_empty(&ecl->el_chain));
370 static int echo_lock_fits_into(const struct lu_env *env,
371 const struct cl_lock_slice *slice,
372 const struct cl_lock_descr *need,
373 const struct cl_io *unused)
378 static struct cl_lock_operations echo_lock_ops = {
379 .clo_fini = echo_lock_fini,
380 .clo_delete = echo_lock_delete,
381 .clo_fits_into = echo_lock_fits_into
386 /** \defgroup echo_cl_ops cl_object operations
388 * operations for cl_object
392 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
393 struct cl_page *page, struct page *vmpage)
395 struct echo_page *ep = cl_object_page_slice(obj, page);
396 struct echo_object *eco = cl2echo_obj(obj);
399 ep->ep_vmpage = vmpage;
400 page_cache_get(vmpage);
401 mutex_init(&ep->ep_lock);
402 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
403 cfs_atomic_inc(&eco->eo_npages);
407 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
413 static int echo_lock_init(const struct lu_env *env,
414 struct cl_object *obj, struct cl_lock *lock,
415 const struct cl_io *unused)
417 struct echo_lock *el;
420 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, __GFP_IO);
422 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
423 el->el_object = cl2echo_obj(obj);
424 CFS_INIT_LIST_HEAD(&el->el_chain);
425 cfs_atomic_set(&el->el_refcount, 0);
427 RETURN(el == NULL ? -ENOMEM : 0);
430 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
431 const struct cl_object_conf *conf)
436 static const struct cl_object_operations echo_cl_obj_ops = {
437 .coo_page_init = echo_page_init,
438 .coo_lock_init = echo_lock_init,
439 .coo_io_init = echo_io_init,
440 .coo_conf_set = echo_conf_set
442 /** @} echo_cl_ops */
444 /** \defgroup echo_lu_ops lu_object operations
446 * operations for echo lu object.
450 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
451 const struct lu_object_conf *conf)
453 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
454 struct echo_client_obd *ec = ed->ed_ec;
455 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
459 struct lu_object *below;
460 struct lu_device *under;
463 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
467 lu_object_add(obj, below);
470 if (!ed->ed_next_ismd) {
471 const struct cl_object_conf *cconf = lu2cl_conf(conf);
472 struct echo_object_conf *econf = cl2echo_conf(cconf);
474 LASSERT(econf->eoc_md);
475 eco->eo_lsm = *econf->eoc_md;
476 /* clear the lsm pointer so that it won't get freed. */
477 *econf->eoc_md = NULL;
483 cfs_atomic_set(&eco->eo_npages, 0);
484 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
486 spin_lock(&ec->ec_lock);
487 cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
488 spin_unlock(&ec->ec_lock);
493 /* taken from osc_unpackmd() */
494 static int echo_alloc_memmd(struct echo_device *ed,
495 struct lov_stripe_md **lsmp)
501 /* If export is lov/osc then use their obd method */
502 if (ed->ed_next != NULL)
503 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
504 /* OFD has no unpackmd method, do everything here */
505 lsm_size = lov_stripe_md_size(1);
507 LASSERT(*lsmp == NULL);
508 OBD_ALLOC(*lsmp, lsm_size);
512 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
513 if ((*lsmp)->lsm_oinfo[0] == NULL) {
514 OBD_FREE(*lsmp, lsm_size);
518 loi_init((*lsmp)->lsm_oinfo[0]);
519 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
520 ostid_set_seq_echo(&(*lsmp)->lsm_oi);
525 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
531 /* If export is lov/osc then use their obd method */
532 if (ed->ed_next != NULL)
533 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
534 /* OFD has no unpackmd method, do everything here */
535 lsm_size = lov_stripe_md_size(1);
537 LASSERT(*lsmp != NULL);
538 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
539 OBD_FREE(*lsmp, lsm_size);
544 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
546 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
547 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
550 LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
552 spin_lock(&ec->ec_lock);
553 cfs_list_del_init(&eco->eo_obj_chain);
554 spin_unlock(&ec->ec_lock);
557 lu_object_header_fini(obj->lo_header);
560 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
561 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
565 static int echo_object_print(const struct lu_env *env, void *cookie,
566 lu_printer_t p, const struct lu_object *o)
568 struct echo_object *obj = cl2echo_obj(lu2cl(o));
570 return (*p)(env, cookie, "echoclient-object@%p", obj);
573 static const struct lu_object_operations echo_lu_obj_ops = {
574 .loo_object_init = echo_object_init,
575 .loo_object_delete = NULL,
576 .loo_object_release = NULL,
577 .loo_object_free = echo_object_free,
578 .loo_object_print = echo_object_print,
579 .loo_object_invariant = NULL
581 /** @} echo_lu_ops */
583 /** \defgroup echo_lu_dev_ops lu_device operations
585 * Operations for echo lu device.
589 static struct lu_object *echo_object_alloc(const struct lu_env *env,
590 const struct lu_object_header *hdr,
591 struct lu_device *dev)
593 struct echo_object *eco;
594 struct lu_object *obj = NULL;
597 /* we're the top dev. */
598 LASSERT(hdr == NULL);
599 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, __GFP_IO);
601 struct cl_object_header *hdr = &eco->eo_hdr;
603 obj = &echo_obj2cl(eco)->co_lu;
604 cl_object_header_init(hdr);
605 lu_object_init(obj, &hdr->coh_lu, dev);
606 lu_object_add_top(&hdr->coh_lu, obj);
608 eco->eo_cl.co_ops = &echo_cl_obj_ops;
609 obj->lo_ops = &echo_lu_obj_ops;
614 static struct lu_device_operations echo_device_lu_ops = {
615 .ldo_object_alloc = echo_object_alloc,
618 /** @} echo_lu_dev_ops */
620 static struct cl_device_operations echo_device_cl_ops = {
623 /** \defgroup echo_init Setup and teardown
625 * Init and fini functions for echo client.
629 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
631 struct cl_site *site = &ed->ed_site_myself;
634 /* initialize site */
635 rc = cl_site_init(site, &ed->ed_cl);
637 CERROR("Cannot initilize site for echo client(%d)\n", rc);
641 rc = lu_site_init_finish(&site->cs_lu);
649 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
652 if (!ed->ed_next_ismd)
653 cl_site_fini(ed->ed_site);
658 static void *echo_thread_key_init(const struct lu_context *ctx,
659 struct lu_context_key *key)
661 struct echo_thread_info *info;
663 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, __GFP_IO);
665 info = ERR_PTR(-ENOMEM);
669 static void echo_thread_key_fini(const struct lu_context *ctx,
670 struct lu_context_key *key, void *data)
672 struct echo_thread_info *info = data;
673 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
676 static void echo_thread_key_exit(const struct lu_context *ctx,
677 struct lu_context_key *key, void *data)
681 static struct lu_context_key echo_thread_key = {
682 .lct_tags = LCT_CL_THREAD,
683 .lct_init = echo_thread_key_init,
684 .lct_fini = echo_thread_key_fini,
685 .lct_exit = echo_thread_key_exit
688 static void *echo_session_key_init(const struct lu_context *ctx,
689 struct lu_context_key *key)
691 struct echo_session_info *session;
693 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, __GFP_IO);
695 session = ERR_PTR(-ENOMEM);
699 static void echo_session_key_fini(const struct lu_context *ctx,
700 struct lu_context_key *key, void *data)
702 struct echo_session_info *session = data;
703 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
706 static void echo_session_key_exit(const struct lu_context *ctx,
707 struct lu_context_key *key, void *data)
711 static struct lu_context_key echo_session_key = {
712 .lct_tags = LCT_SESSION,
713 .lct_init = echo_session_key_init,
714 .lct_fini = echo_session_key_fini,
715 .lct_exit = echo_session_key_exit
718 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
720 #ifdef HAVE_SERVER_SUPPORT
721 # define ECHO_SEQ_WIDTH 0xffffffff
722 static int echo_fid_init(struct echo_device *ed, char *obd_name,
723 struct seq_server_site *ss)
729 OBD_ALLOC_PTR(ed->ed_cl_seq);
730 if (ed->ed_cl_seq == NULL)
733 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
735 GOTO(out_free_seq, rc = -ENOMEM);
737 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
739 /* Init client side sequence-manager */
740 rc = seq_client_init(ed->ed_cl_seq, NULL,
742 prefix, ss->ss_server_seq);
743 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
744 OBD_FREE(prefix, MAX_OBD_NAME + 5);
746 GOTO(out_free_seq, rc);
751 OBD_FREE_PTR(ed->ed_cl_seq);
752 ed->ed_cl_seq = NULL;
756 static int echo_fid_fini(struct obd_device *obddev)
758 struct echo_device *ed = obd2echo_dev(obddev);
761 if (ed->ed_cl_seq != NULL) {
762 seq_client_fini(ed->ed_cl_seq);
763 OBD_FREE_PTR(ed->ed_cl_seq);
764 ed->ed_cl_seq = NULL;
769 #endif /* HAVE_SERVER_SUPPORT */
771 static struct lu_device *echo_device_alloc(const struct lu_env *env,
772 struct lu_device_type *t,
773 struct lustre_cfg *cfg)
775 struct lu_device *next;
776 struct echo_device *ed;
777 struct cl_device *cd;
778 struct obd_device *obd = NULL; /* to keep compiler happy */
779 struct obd_device *tgt;
780 const char *tgt_type_name;
787 GOTO(out, rc = -ENOMEM);
791 rc = cl_device_init(cd, t);
795 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
796 cd->cd_ops = &echo_device_cl_ops;
799 obd = class_name2obd(lustre_cfg_string(cfg, 0));
800 LASSERT(obd != NULL);
801 LASSERT(env != NULL);
803 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
805 CERROR("Can not find tgt device %s\n",
806 lustre_cfg_string(cfg, 1));
807 GOTO(out, rc = -ENODEV);
810 next = tgt->obd_lu_dev;
811 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
812 ed->ed_next_ismd = 1;
814 ed->ed_next_ismd = 0;
815 rc = echo_site_init(env, ed);
821 rc = echo_client_setup(env, obd, cfg);
825 ed->ed_ec = &obd->u.echo_client;
828 if (ed->ed_next_ismd) {
829 #ifdef HAVE_SERVER_SUPPORT
830 /* Suppose to connect to some Metadata layer */
832 struct lu_device *ld;
836 CERROR("%s is not lu device type!\n",
837 lustre_cfg_string(cfg, 1));
838 GOTO(out, rc = -EINVAL);
841 tgt_type_name = lustre_cfg_string(cfg, 2);
842 if (!tgt_type_name) {
843 CERROR("%s no type name for echo %s setup\n",
844 lustre_cfg_string(cfg, 1),
845 tgt->obd_type->typ_name);
846 GOTO(out, rc = -EINVAL);
851 spin_lock(&ls->ls_ld_lock);
852 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
853 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
858 spin_unlock(&ls->ls_ld_lock);
861 CERROR("%s is not lu device type!\n",
862 lustre_cfg_string(cfg, 1));
863 GOTO(out, rc = -EINVAL);
867 /* For MD echo client, it will use the site in MDS stack */
868 ed->ed_site_myself.cs_lu = *ls;
869 ed->ed_site = &ed->ed_site_myself;
870 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
871 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
873 CERROR("echo fid init error %d\n", rc);
876 #else /* !HAVE_SERVER_SUPPORT */
877 CERROR("Local operations are NOT supported on client side. "
878 "Only remote operations are supported. Metadata client "
879 "must be run on server side.\n");
880 GOTO(out, rc = -EOPNOTSUPP);
883 /* if echo client is to be stacked upon ost device, the next is
884 * NULL since ost is not a clio device so far */
885 if (next != NULL && !lu_device_is_cl(next))
888 tgt_type_name = tgt->obd_type->typ_name;
890 LASSERT(next != NULL);
891 if (next->ld_site != NULL)
892 GOTO(out, rc = -EBUSY);
894 next->ld_site = &ed->ed_site->cs_lu;
895 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
896 next->ld_type->ldt_name,
901 /* Tricky case, I have to determine the obd type since
902 * CLIO uses the different parameters to initialize
903 * objects for lov & osc. */
904 if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
905 ed->ed_next_islov = 1;
907 LASSERT(strcmp(tgt_type_name,
908 LUSTRE_OSC_NAME) == 0);
910 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
914 RETURN(&cd->cd_lu_dev);
919 rc2 = echo_client_cleanup(obd);
921 CERROR("Cleanup obd device %s error(%d)\n",
926 echo_site_fini(env, ed);
928 cl_device_fini(&ed->ed_cl);
938 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
939 const char *name, struct lu_device *next)
945 static struct lu_device *echo_device_fini(const struct lu_env *env,
948 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
949 struct lu_device *next = ed->ed_next;
951 while (next && !ed->ed_next_ismd)
952 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
956 static void echo_lock_release(const struct lu_env *env,
957 struct echo_lock *ecl,
960 struct cl_lock *clk = echo_lock2cl(ecl);
964 cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
966 cl_lock_mutex_get(env, clk);
967 cl_lock_cancel(env, clk);
968 cl_lock_delete(env, clk);
969 cl_lock_mutex_put(env, clk);
971 cl_lock_put(env, clk);
974 static struct lu_device *echo_device_free(const struct lu_env *env,
977 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
978 struct echo_client_obd *ec = ed->ed_ec;
979 struct echo_object *eco;
980 struct lu_device *next = ed->ed_next;
982 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
985 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
987 /* check if there are objects still alive.
988 * It shouldn't have any object because lu_site_purge would cleanup
989 * all of cached objects. Anyway, probably the echo device is being
990 * parallelly accessed.
992 spin_lock(&ec->ec_lock);
993 cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
995 spin_unlock(&ec->ec_lock);
998 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1001 "Waiting for the reference of echo object to be dropped\n");
1003 /* Wait for the last reference to be dropped. */
1004 spin_lock(&ec->ec_lock);
1005 while (!cfs_list_empty(&ec->ec_objects)) {
1006 spin_unlock(&ec->ec_lock);
1007 CERROR("echo_client still has objects at cleanup time, "
1008 "wait for 1 second\n");
1009 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
1010 cfs_time_seconds(1));
1011 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1012 spin_lock(&ec->ec_lock);
1014 spin_unlock(&ec->ec_lock);
1016 LASSERT(cfs_list_empty(&ec->ec_locks));
1018 CDEBUG(D_INFO, "No object exists, exiting...\n");
1020 echo_client_cleanup(d->ld_obd);
1021 #ifdef HAVE_SERVER_SUPPORT
1022 echo_fid_fini(d->ld_obd);
1024 while (next && !ed->ed_next_ismd)
1025 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1027 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1028 echo_site_fini(env, ed);
1029 cl_device_fini(&ed->ed_cl);
1035 static const struct lu_device_type_operations echo_device_type_ops = {
1036 .ldto_init = echo_type_init,
1037 .ldto_fini = echo_type_fini,
1039 .ldto_start = echo_type_start,
1040 .ldto_stop = echo_type_stop,
1042 .ldto_device_alloc = echo_device_alloc,
1043 .ldto_device_free = echo_device_free,
1044 .ldto_device_init = echo_device_init,
1045 .ldto_device_fini = echo_device_fini
1048 static struct lu_device_type echo_device_type = {
1049 .ldt_tags = LU_DEVICE_CL,
1050 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1051 .ldt_ops = &echo_device_type_ops,
1052 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1056 /** \defgroup echo_exports Exported operations
1058 * exporting functions to echo client
1063 /* Interfaces to echo client obd device */
1064 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1065 struct lov_stripe_md **lsmp)
1068 struct echo_thread_info *info;
1069 struct echo_object_conf *conf;
1070 struct lov_stripe_md *lsm;
1071 struct echo_object *eco;
1072 struct cl_object *obj;
1081 LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi));
1082 LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n",
1083 POSTID(&lsm->lsm_oi));
1085 /* Never return an object if the obd is to be freed. */
1086 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1087 RETURN(ERR_PTR(-ENODEV));
1089 env = cl_env_get(&refcheck);
1091 RETURN((void *)env);
1093 info = echo_env_info(env);
1094 conf = &info->eti_conf;
1096 if (!d->ed_next_islov) {
1097 struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1098 LASSERT(oinfo != NULL);
1099 oinfo->loi_oi = lsm->lsm_oi;
1100 conf->eoc_cl.u.coc_oinfo = oinfo;
1102 struct lustre_md *md;
1104 memset(md, 0, sizeof *md);
1106 conf->eoc_cl.u.coc_md = md;
1109 conf->eoc_md = lsmp;
1111 fid = &info->eti_fid;
1112 rc = ostid_to_fid(fid, &lsm->lsm_oi, 0);
1114 GOTO(out, eco = ERR_PTR(rc));
1116 /* In the function below, .hs_keycmp resolves to
1117 * lu_obj_hop_keycmp() */
1118 /* coverity[overrun-buffer-val] */
1119 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1121 GOTO(out, eco = (void*)obj);
1123 eco = cl2echo_obj(obj);
1124 if (eco->eo_deleted) {
1125 cl_object_put(env, obj);
1126 eco = ERR_PTR(-EAGAIN);
1130 cl_env_put(env, &refcheck);
1134 static int cl_echo_object_put(struct echo_object *eco)
1137 struct cl_object *obj = echo_obj2cl(eco);
1141 env = cl_env_get(&refcheck);
1143 RETURN(PTR_ERR(env));
1145 /* an external function to kill an object? */
1146 if (eco->eo_deleted) {
1147 struct lu_object_header *loh = obj->co_lu.lo_header;
1148 LASSERT(&eco->eo_hdr == luh2coh(loh));
1149 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1152 cl_object_put(env, obj);
1153 cl_env_put(env, &refcheck);
1157 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1158 obd_off start, obd_off end, int mode,
1159 __u64 *cookie , __u32 enqflags)
1162 struct cl_lock *lck;
1163 struct cl_object *obj;
1164 struct cl_lock_descr *descr;
1165 struct echo_thread_info *info;
1169 info = echo_env_info(env);
1171 descr = &info->eti_descr;
1172 obj = echo_obj2cl(eco);
1174 descr->cld_obj = obj;
1175 descr->cld_start = cl_index(obj, start);
1176 descr->cld_end = cl_index(obj, end);
1177 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1178 descr->cld_enq_flags = enqflags;
1181 lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1183 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1184 struct echo_lock *el;
1186 rc = cl_wait(env, lck);
1188 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1189 spin_lock(&ec->ec_lock);
1190 if (cfs_list_empty(&el->el_chain)) {
1191 cfs_list_add(&el->el_chain, &ec->ec_locks);
1192 el->el_cookie = ++ec->ec_unique;
1194 cfs_atomic_inc(&el->el_refcount);
1195 *cookie = el->el_cookie;
1196 spin_unlock(&ec->ec_lock);
1198 cl_lock_release(env, lck, "ec enqueue", current);
1204 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1205 int mode, __u64 *cookie)
1207 struct echo_thread_info *info;
1214 env = cl_env_get(&refcheck);
1216 RETURN(PTR_ERR(env));
1218 info = echo_env_info(env);
1221 io->ci_ignore_layout = 1;
1222 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1225 LASSERT(result == 0);
1227 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1228 cl_io_fini(env, io);
1232 cl_env_put(env, &refcheck);
1236 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1239 struct echo_client_obd *ec = ed->ed_ec;
1240 struct echo_lock *ecl = NULL;
1242 int found = 0, still_used = 0;
1245 LASSERT(ec != NULL);
1246 spin_lock(&ec->ec_lock);
1247 cfs_list_for_each (el, &ec->ec_locks) {
1248 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1249 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1250 found = (ecl->el_cookie == cookie);
1252 if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1253 cfs_list_del_init(&ecl->el_chain);
1259 spin_unlock(&ec->ec_lock);
1264 echo_lock_release(env, ecl, still_used);
1268 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1275 env = cl_env_get(&refcheck);
1277 RETURN(PTR_ERR(env));
1279 rc = cl_echo_cancel0(env, ed, cookie);
1281 cl_env_put(env, &refcheck);
1285 static void echo_commit_callback(const struct lu_env *env, struct cl_io *io,
1286 struct cl_page *page)
1288 struct echo_thread_info *info;
1289 struct cl_2queue *queue;
1291 info = echo_env_info(env);
1292 LASSERT(io == &info->eti_io);
1294 queue = &info->eti_queue;
1295 cl_page_list_add(&queue->c2_qout, page);
1298 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1299 struct page **pages, int npages, int async)
1302 struct echo_thread_info *info;
1303 struct cl_object *obj = echo_obj2cl(eco);
1304 struct echo_device *ed = eco->eo_dev;
1305 struct cl_2queue *queue;
1307 struct cl_page *clp;
1308 struct lustre_handle lh = { 0 };
1309 int page_size = cl_page_size(obj);
1315 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1316 LASSERT(ed->ed_next != NULL);
1317 env = cl_env_get(&refcheck);
1319 RETURN(PTR_ERR(env));
1321 info = echo_env_info(env);
1323 queue = &info->eti_queue;
1325 cl_2queue_init(queue);
1327 io->ci_ignore_layout = 1;
1328 rc = cl_io_init(env, io, CIT_MISC, obj);
1334 rc = cl_echo_enqueue0(env, eco, offset,
1335 offset + npages * PAGE_CACHE_SIZE - 1,
1336 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1339 GOTO(error_lock, rc);
1341 for (i = 0; i < npages; i++) {
1343 clp = cl_page_find(env, obj, cl_index(obj, offset),
1344 pages[i], CPT_TRANSIENT);
1349 LASSERT(clp->cp_type == CPT_TRANSIENT);
1351 rc = cl_page_own(env, io, clp);
1353 LASSERT(clp->cp_state == CPS_FREEING);
1354 cl_page_put(env, clp);
1358 cl_2queue_add(queue, clp);
1360 /* drop the reference count for cl_page_find, so that the page
1361 * will be freed in cl_2queue_fini. */
1362 cl_page_put(env, clp);
1363 cl_page_clip(env, clp, 0, page_size);
1365 offset += page_size;
1369 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1371 async = async && (typ == CRT_WRITE);
1373 rc = cl_io_commit_async(env, io, &queue->c2_qin,
1375 echo_commit_callback);
1377 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1378 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1379 async ? "async" : "sync", rc);
1382 cl_echo_cancel0(env, ed, lh.cookie);
1385 cl_2queue_discard(env, io, queue);
1386 cl_2queue_disown(env, io, queue);
1387 cl_2queue_fini(env, queue);
1388 cl_io_fini(env, io);
1390 cl_env_put(env, &refcheck);
1393 /** @} echo_exports */
1396 static obd_id last_object_id;
1399 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1401 struct lov_stripe_md *ulsm = _ulsm;
1404 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1408 if (copy_to_user (ulsm, lsm, sizeof(*ulsm)))
1411 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1412 if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1413 sizeof(lsm->lsm_oinfo[0])))
1420 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1421 void *ulsm, int ulsm_nob)
1423 struct echo_client_obd *ec = ed->ed_ec;
1426 if (ulsm_nob < sizeof (*lsm))
1429 if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
1432 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1433 lsm->lsm_magic != LOV_MAGIC ||
1434 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1435 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1439 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1440 if (copy_from_user(lsm->lsm_oinfo[i],
1441 ((struct lov_stripe_md *)ulsm)-> \
1443 sizeof(lsm->lsm_oinfo[0])))
1449 #ifdef HAVE_SERVER_SUPPORT
1450 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1453 sprintf(name, LPU64, id);
1454 lname->ln_name = name;
1455 lname->ln_namelen = strlen(name);
1458 /* similar to mdt_attr_get_complex */
1459 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1462 struct echo_thread_info *info = echo_env_info(env);
1467 LASSERT(ma->ma_lmm_size > 0);
1469 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1473 /* big_lmm may need to be grown */
1474 if (info->eti_big_lmmsize < rc) {
1475 int size = size_roundup_power2(rc);
1477 if (info->eti_big_lmmsize > 0) {
1478 /* free old buffer */
1479 LASSERT(info->eti_big_lmm);
1480 OBD_FREE_LARGE(info->eti_big_lmm,
1481 info->eti_big_lmmsize);
1482 info->eti_big_lmm = NULL;
1483 info->eti_big_lmmsize = 0;
1486 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1487 if (info->eti_big_lmm == NULL)
1489 info->eti_big_lmmsize = size;
1491 LASSERT(info->eti_big_lmmsize >= rc);
1493 info->eti_buf.lb_buf = info->eti_big_lmm;
1494 info->eti_buf.lb_len = info->eti_big_lmmsize;
1495 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1499 ma->ma_valid |= MA_LOV;
1500 ma->ma_lmm = info->eti_big_lmm;
1501 ma->ma_lmm_size = rc;
1506 static int echo_attr_get_complex(const struct lu_env *env,
1507 struct md_object *next,
1510 struct echo_thread_info *info = echo_env_info(env);
1511 struct lu_buf *buf = &info->eti_buf;
1512 umode_t mode = lu_object_attr(&next->mo_lu);
1513 int need = ma->ma_need;
1520 if (need & MA_INODE) {
1521 ma->ma_need = MA_INODE;
1522 rc = mo_attr_get(env, next, ma);
1525 ma->ma_valid |= MA_INODE;
1528 if (need & MA_LOV) {
1529 if (S_ISREG(mode) || S_ISDIR(mode)) {
1530 LASSERT(ma->ma_lmm_size > 0);
1531 buf->lb_buf = ma->ma_lmm;
1532 buf->lb_len = ma->ma_lmm_size;
1533 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1535 ma->ma_lmm_size = rc2;
1536 ma->ma_valid |= MA_LOV;
1537 } else if (rc2 == -ENODATA) {
1539 ma->ma_lmm_size = 0;
1540 } else if (rc2 == -ERANGE) {
1541 rc2 = echo_big_lmm_get(env, next, ma);
1543 GOTO(out, rc = rc2);
1545 GOTO(out, rc = rc2);
1550 #ifdef CONFIG_FS_POSIX_ACL
1551 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1552 buf->lb_buf = ma->ma_acl;
1553 buf->lb_len = ma->ma_acl_size;
1554 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1556 ma->ma_acl_size = rc2;
1557 ma->ma_valid |= MA_ACL_DEF;
1558 } else if (rc2 == -ENODATA) {
1560 ma->ma_acl_size = 0;
1562 GOTO(out, rc = rc2);
1568 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1569 rc, ma->ma_valid, ma->ma_lmm);
1574 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1575 struct md_object *parent, struct lu_fid *fid,
1576 struct lu_name *lname, struct md_op_spec *spec,
1579 struct lu_object *ec_child, *child;
1580 struct lu_device *ld = ed->ed_next;
1581 struct echo_thread_info *info = echo_env_info(env);
1582 struct lu_fid *fid2 = &info->eti_fid2;
1583 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1588 rc = mdo_lookup(env, parent, lname, fid2, spec);
1591 else if (rc != -ENOENT)
1594 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1596 if (IS_ERR(ec_child)) {
1597 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1599 RETURN(PTR_ERR(ec_child));
1602 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1603 if (child == NULL) {
1604 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1605 GOTO(out_put, rc = -EINVAL);
1608 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1609 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1612 * Do not perform lookup sanity check. We know that name does not exist.
1614 spec->sp_cr_lookup = 0;
1615 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1617 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1620 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1621 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1624 lu_object_put(env, ec_child);
1628 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1631 struct echo_thread_info *info = echo_env_info(env);
1633 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1634 ma->ma_lmm = (void *)&info->eti_lmm;
1635 ma->ma_lmm_size = sizeof(info->eti_lmm);
1637 LASSERT(info->eti_big_lmmsize);
1638 ma->ma_lmm = info->eti_big_lmm;
1639 ma->ma_lmm_size = info->eti_big_lmmsize;
1645 static int echo_create_md_object(const struct lu_env *env,
1646 struct echo_device *ed,
1647 struct lu_object *ec_parent,
1649 char *name, int namelen,
1650 __u64 id, __u32 mode, int count,
1651 int stripe_count, int stripe_offset)
1653 struct lu_object *parent;
1654 struct echo_thread_info *info = echo_env_info(env);
1655 struct lu_name *lname = &info->eti_lname;
1656 struct md_op_spec *spec = &info->eti_spec;
1657 struct md_attr *ma = &info->eti_ma;
1658 struct lu_device *ld = ed->ed_next;
1664 if (ec_parent == NULL)
1666 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1670 memset(ma, 0, sizeof(*ma));
1671 memset(spec, 0, sizeof(*spec));
1672 if (stripe_count != 0) {
1673 spec->sp_cr_flags |= FMODE_WRITE;
1674 echo_set_lmm_size(env, ld, ma);
1675 if (stripe_count != -1) {
1676 struct lov_user_md_v3 *lum = &info->eti_lum;
1678 lum->lmm_magic = LOV_USER_MAGIC_V3;
1679 lum->lmm_stripe_count = stripe_count;
1680 lum->lmm_stripe_offset = stripe_offset;
1681 lum->lmm_pattern = 0;
1682 spec->u.sp_ea.eadata = lum;
1683 spec->u.sp_ea.eadatalen = sizeof(*lum);
1684 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1688 ma->ma_attr.la_mode = mode;
1689 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1690 ma->ma_attr.la_ctime = cfs_time_current_64();
1693 lname->ln_name = name;
1694 lname->ln_namelen = namelen;
1695 /* If name is specified, only create one object by name */
1696 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1701 /* Create multiple object sequenced by id */
1702 for (i = 0; i < count; i++) {
1703 char *tmp_name = info->eti_name;
1705 echo_md_build_name(lname, tmp_name, id);
1707 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1710 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1721 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1722 struct echo_device *ed,
1723 struct md_object *parent,
1724 struct lu_name *lname)
1726 struct echo_thread_info *info = echo_env_info(env);
1727 struct lu_fid *fid = &info->eti_fid;
1728 struct lu_object *child;
1732 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1734 rc = mdo_lookup(env, parent, lname, fid, NULL);
1736 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1737 RETURN(ERR_PTR(rc));
1740 /* In the function below, .hs_keycmp resolves to
1741 * lu_obj_hop_keycmp() */
1742 /* coverity[overrun-buffer-val] */
1743 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1748 static int echo_setattr_object(const struct lu_env *env,
1749 struct echo_device *ed,
1750 struct lu_object *ec_parent,
1751 __u64 id, int count)
1753 struct lu_object *parent;
1754 struct echo_thread_info *info = echo_env_info(env);
1755 struct lu_name *lname = &info->eti_lname;
1756 char *name = info->eti_name;
1757 struct lu_device *ld = ed->ed_next;
1758 struct lu_buf *buf = &info->eti_buf;
1764 if (ec_parent == NULL)
1766 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1770 for (i = 0; i < count; i++) {
1771 struct lu_object *ec_child, *child;
1773 echo_md_build_name(lname, name, id);
1775 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1776 if (IS_ERR(ec_child)) {
1777 CERROR("Can't find child %s: rc = %ld\n",
1778 lname->ln_name, PTR_ERR(ec_child));
1779 RETURN(PTR_ERR(ec_child));
1782 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1783 if (child == NULL) {
1784 CERROR("Can not locate the child %s\n", lname->ln_name);
1785 lu_object_put(env, ec_child);
1790 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1791 PFID(lu_object_fid(child)));
1793 buf->lb_buf = info->eti_xattr_buf;
1794 buf->lb_len = sizeof(info->eti_xattr_buf);
1796 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1797 rc = mo_xattr_set(env, lu2md(child), buf, name,
1800 CERROR("Can not setattr child "DFID": rc = %d\n",
1801 PFID(lu_object_fid(child)), rc);
1802 lu_object_put(env, ec_child);
1805 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1806 PFID(lu_object_fid(child)));
1808 lu_object_put(env, ec_child);
1813 static int echo_getattr_object(const struct lu_env *env,
1814 struct echo_device *ed,
1815 struct lu_object *ec_parent,
1816 __u64 id, int count)
1818 struct lu_object *parent;
1819 struct echo_thread_info *info = echo_env_info(env);
1820 struct lu_name *lname = &info->eti_lname;
1821 char *name = info->eti_name;
1822 struct md_attr *ma = &info->eti_ma;
1823 struct lu_device *ld = ed->ed_next;
1829 if (ec_parent == NULL)
1831 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1835 memset(ma, 0, sizeof(*ma));
1836 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1837 ma->ma_acl = info->eti_xattr_buf;
1838 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1840 for (i = 0; i < count; i++) {
1841 struct lu_object *ec_child, *child;
1844 echo_md_build_name(lname, name, id);
1845 echo_set_lmm_size(env, ld, ma);
1847 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1848 if (IS_ERR(ec_child)) {
1849 CERROR("Can't find child %s: rc = %ld\n",
1850 lname->ln_name, PTR_ERR(ec_child));
1851 RETURN(PTR_ERR(ec_child));
1854 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1855 if (child == NULL) {
1856 CERROR("Can not locate the child %s\n", lname->ln_name);
1857 lu_object_put(env, ec_child);
1861 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1862 PFID(lu_object_fid(child)));
1863 rc = echo_attr_get_complex(env, lu2md(child), ma);
1865 CERROR("Can not getattr child "DFID": rc = %d\n",
1866 PFID(lu_object_fid(child)), rc);
1867 lu_object_put(env, ec_child);
1870 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1871 PFID(lu_object_fid(child)));
1873 lu_object_put(env, ec_child);
1879 static int echo_lookup_object(const struct lu_env *env,
1880 struct echo_device *ed,
1881 struct lu_object *ec_parent,
1882 __u64 id, int count)
1884 struct lu_object *parent;
1885 struct echo_thread_info *info = echo_env_info(env);
1886 struct lu_name *lname = &info->eti_lname;
1887 char *name = info->eti_name;
1888 struct lu_fid *fid = &info->eti_fid;
1889 struct lu_device *ld = ed->ed_next;
1893 if (ec_parent == NULL)
1895 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1899 /*prepare the requests*/
1900 for (i = 0; i < count; i++) {
1901 echo_md_build_name(lname, name, id);
1903 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1904 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1906 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1908 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1911 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1912 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1919 static int echo_md_destroy_internal(const struct lu_env *env,
1920 struct echo_device *ed,
1921 struct md_object *parent,
1922 struct lu_name *lname,
1925 struct lu_device *ld = ed->ed_next;
1926 struct lu_object *ec_child;
1927 struct lu_object *child;
1932 ec_child = echo_md_lookup(env, ed, parent, lname);
1933 if (IS_ERR(ec_child)) {
1934 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1936 RETURN(PTR_ERR(ec_child));
1939 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1940 if (child == NULL) {
1941 CERROR("Can not locate the child %s\n", lname->ln_name);
1942 GOTO(out_put, rc = -EINVAL);
1945 if (lu_object_remote(child)) {
1946 CERROR("Can not destroy remote object %s: rc = %d\n",
1947 lname->ln_name, -EPERM);
1948 GOTO(out_put, rc = -EPERM);
1950 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1951 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1953 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1955 CERROR("Can not unlink child %s: rc = %d\n",
1956 lname->ln_name, rc);
1959 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1960 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1962 lu_object_put(env, ec_child);
1966 static int echo_destroy_object(const struct lu_env *env,
1967 struct echo_device *ed,
1968 struct lu_object *ec_parent,
1969 char *name, int namelen,
1970 __u64 id, __u32 mode,
1973 struct echo_thread_info *info = echo_env_info(env);
1974 struct lu_name *lname = &info->eti_lname;
1975 struct md_attr *ma = &info->eti_ma;
1976 struct lu_device *ld = ed->ed_next;
1977 struct lu_object *parent;
1982 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1986 memset(ma, 0, sizeof(*ma));
1987 ma->ma_attr.la_mode = mode;
1988 ma->ma_attr.la_valid = LA_CTIME;
1989 ma->ma_attr.la_ctime = cfs_time_current_64();
1990 ma->ma_need = MA_INODE;
1994 lname->ln_name = name;
1995 lname->ln_namelen = namelen;
1996 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2001 /*prepare the requests*/
2002 for (i = 0; i < count; i++) {
2003 char *tmp_name = info->eti_name;
2006 echo_md_build_name(lname, tmp_name, id);
2008 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2011 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2020 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2021 struct echo_device *ed, char *path,
2024 struct lu_device *ld = ed->ed_next;
2025 struct md_device *md = lu2md_dev(ld);
2026 struct echo_thread_info *info = echo_env_info(env);
2027 struct lu_fid *fid = &info->eti_fid;
2028 struct lu_name *lname = &info->eti_lname;
2029 struct lu_object *parent = NULL;
2030 struct lu_object *child = NULL;
2034 /*Only support MDD layer right now*/
2035 rc = md->md_ops->mdo_root_get(env, md, fid);
2037 CERROR("get root error: rc = %d\n", rc);
2038 RETURN(ERR_PTR(rc));
2041 /* In the function below, .hs_keycmp resolves to
2042 * lu_obj_hop_keycmp() */
2043 /* coverity[overrun-buffer-val] */
2044 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2045 if (IS_ERR(parent)) {
2046 CERROR("Can not find the parent "DFID": rc = %ld\n",
2047 PFID(fid), PTR_ERR(parent));
2052 struct lu_object *ld_parent;
2055 e = strsep(&path, "/");
2060 if (!path || path[0] == '\0')
2066 lname->ln_namelen = strlen(e);
2068 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2069 if (ld_parent == NULL) {
2070 lu_object_put(env, parent);
2075 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2076 lu_object_put(env, parent);
2077 if (IS_ERR(child)) {
2078 rc = (int)PTR_ERR(child);
2079 CERROR("lookup %s under parent "DFID": rc = %d\n",
2080 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2087 RETURN(ERR_PTR(rc));
2092 static void echo_ucred_init(struct lu_env *env)
2094 struct lu_ucred *ucred = lu_ucred(env);
2096 ucred->uc_valid = UCRED_INVALID;
2098 ucred->uc_suppgids[0] = -1;
2099 ucred->uc_suppgids[1] = -1;
2101 ucred->uc_uid = ucred->uc_o_uid = current_uid();
2102 ucred->uc_gid = ucred->uc_o_gid = current_gid();
2103 ucred->uc_fsuid = ucred->uc_o_fsuid = current_fsuid();
2104 ucred->uc_fsgid = ucred->uc_o_fsgid = current_fsgid();
2105 ucred->uc_cap = cfs_curproc_cap_pack();
2107 /* remove fs privilege for non-root user. */
2108 if (ucred->uc_fsuid)
2109 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2110 ucred->uc_valid = UCRED_NEW;
2113 static void echo_ucred_fini(struct lu_env *env)
2115 struct lu_ucred *ucred = lu_ucred(env);
2116 ucred->uc_valid = UCRED_INIT;
2119 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2120 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2121 static int echo_md_handler(struct echo_device *ed, int command,
2122 char *path, int path_len, __u64 id, int count,
2123 struct obd_ioctl_data *data)
2125 struct echo_thread_info *info;
2126 struct lu_device *ld = ed->ed_next;
2129 struct lu_object *parent;
2131 int namelen = data->ioc_plen2;
2136 CERROR("MD echo client is not being initialized properly\n");
2140 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2141 CERROR("Only support MDD layer right now!\n");
2145 env = cl_env_get(&refcheck);
2147 RETURN(PTR_ERR(env));
2149 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2153 /* init big_lmm buffer */
2154 info = echo_env_info(env);
2155 LASSERT(info->eti_big_lmm == NULL);
2156 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2157 if (info->eti_big_lmm == NULL)
2158 GOTO(out_env, rc = -ENOMEM);
2159 info->eti_big_lmmsize = MIN_MD_SIZE;
2161 parent = echo_resolve_path(env, ed, path, path_len);
2162 if (IS_ERR(parent)) {
2163 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2165 GOTO(out_free, rc = PTR_ERR(parent));
2169 OBD_ALLOC(name, namelen + 1);
2171 GOTO(out_put, rc = -ENOMEM);
2172 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2173 GOTO(out_name, rc = -EFAULT);
2176 echo_ucred_init(env);
2179 case ECHO_MD_CREATE:
2180 case ECHO_MD_MKDIR: {
2181 struct echo_thread_info *info = echo_env_info(env);
2182 __u32 mode = data->ioc_obdo2.o_mode;
2183 struct lu_fid *fid = &info->eti_fid;
2184 int stripe_count = (int)data->ioc_obdo2.o_misc;
2185 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2187 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2191 /* In the function below, .hs_keycmp resolves to
2192 * lu_obj_hop_keycmp() */
2193 /* coverity[overrun-buffer-val] */
2194 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2195 id, mode, count, stripe_count,
2199 case ECHO_MD_DESTROY:
2200 case ECHO_MD_RMDIR: {
2201 __u32 mode = data->ioc_obdo2.o_mode;
2203 rc = echo_destroy_object(env, ed, parent, name, namelen,
2207 case ECHO_MD_LOOKUP:
2208 rc = echo_lookup_object(env, ed, parent, id, count);
2210 case ECHO_MD_GETATTR:
2211 rc = echo_getattr_object(env, ed, parent, id, count);
2213 case ECHO_MD_SETATTR:
2214 rc = echo_setattr_object(env, ed, parent, id, count);
2217 CERROR("unknown command %d\n", command);
2221 echo_ucred_fini(env);
2225 OBD_FREE(name, namelen + 1);
2227 lu_object_put(env, parent);
2229 LASSERT(info->eti_big_lmm);
2230 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2231 info->eti_big_lmm = NULL;
2232 info->eti_big_lmmsize = 0;
2234 cl_env_put(env, &refcheck);
2237 #endif /* HAVE_SERVER_SUPPORT */
2239 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2240 int on_target, struct obdo *oa, void *ulsm,
2241 int ulsm_nob, struct obd_trans_info *oti)
2243 struct echo_object *eco;
2244 struct echo_client_obd *ec = ed->ed_ec;
2245 struct lov_stripe_md *lsm = NULL;
2250 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2251 (on_target || /* set_stripe */
2252 ec->ec_nstripes != 0)) { /* LOV */
2253 CERROR ("No valid oid\n");
2257 rc = echo_alloc_memmd(ed, &lsm);
2259 CERROR("Cannot allocate md: rc = %d\n", rc);
2266 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2270 if (lsm->lsm_stripe_count == 0)
2271 lsm->lsm_stripe_count = ec->ec_nstripes;
2273 if (lsm->lsm_stripe_size == 0)
2274 lsm->lsm_stripe_size = PAGE_CACHE_SIZE;
2278 /* setup stripes: indices + default ids if required */
2279 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2280 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2281 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2283 lsm->lsm_oinfo[i]->loi_ost_idx =
2284 (idx + i) % ec->ec_nstripes;
2288 /* setup object ID here for !on_target and LOV hint */
2289 if (oa->o_valid & OBD_MD_FLID) {
2290 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2291 lsm->lsm_oi = oa->o_oi;
2294 if (ostid_id(&lsm->lsm_oi) == 0)
2295 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2299 /* Only echo objects are allowed to be created */
2300 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2301 (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2302 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2304 CERROR("Cannot create objects: rc = %d\n", rc);
2310 /* See what object ID we were given */
2311 oa->o_oi = lsm->lsm_oi;
2312 oa->o_valid |= OBD_MD_FLID;
2314 eco = cl_echo_object_find(ed, &lsm);
2316 GOTO(failed, rc = PTR_ERR(eco));
2317 cl_echo_object_put(eco);
2319 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2324 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2326 echo_free_memmd(ed, &lsm);
2328 CERROR("create object failed with: rc = %d\n", rc);
2332 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2335 struct lov_stripe_md *lsm = NULL;
2336 struct echo_object *eco;
2340 if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2341 /* disallow use of object id 0 */
2342 CERROR ("No valid oid\n");
2346 rc = echo_alloc_memmd(ed, &lsm);
2350 lsm->lsm_oi = oa->o_oi;
2351 if (!(oa->o_valid & OBD_MD_FLGROUP))
2352 ostid_set_seq_echo(&lsm->lsm_oi);
2355 eco = cl_echo_object_find(ed, &lsm);
2361 echo_free_memmd(ed, &lsm);
2365 static void echo_put_object(struct echo_object *eco)
2367 if (cl_echo_object_put(eco))
2368 CERROR("echo client: drop an object failed");
2372 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2374 unsigned long stripe_count;
2375 unsigned long stripe_size;
2376 unsigned long width;
2377 unsigned long woffset;
2381 if (lsm->lsm_stripe_count <= 1)
2385 stripe_size = lsm->lsm_stripe_size;
2386 stripe_count = lsm->lsm_stripe_count;
2388 /* width = # bytes in all stripes */
2389 width = stripe_size * stripe_count;
2391 /* woffset = offset within a width; offset = whole number of widths */
2392 woffset = do_div (offset, width);
2394 stripe_index = woffset / stripe_size;
2396 *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2397 *offp = offset * stripe_size + woffset % stripe_size;
2401 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2402 struct page *page, int rw, obd_id id,
2403 obd_off offset, obd_off count)
2410 /* no partial pages on the client */
2411 LASSERT(count == PAGE_CACHE_SIZE);
2415 for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2416 if (rw == OBD_BRW_WRITE) {
2417 stripe_off = offset + delta;
2419 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2421 stripe_off = 0xdeadbeef00c0ffeeULL;
2422 stripe_id = 0xdeadbeef00c0ffeeULL;
2424 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2425 stripe_off, stripe_id);
2431 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2432 struct page *page, obd_id id,
2433 obd_off offset, obd_off count)
2442 /* no partial pages on the client */
2443 LASSERT(count == PAGE_CACHE_SIZE);
2447 for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2448 stripe_off = offset + delta;
2450 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2452 rc2 = block_debug_check("test_brw",
2453 addr + delta, OBD_ECHO_BLOCK_SIZE,
2454 stripe_off, stripe_id);
2456 CERROR ("Error in echo object "LPX64"\n", id);
2465 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2466 struct echo_object *eco, obd_off offset,
2467 obd_size count, int async,
2468 struct obd_trans_info *oti)
2470 struct lov_stripe_md *lsm = eco->eo_lsm;
2472 struct brw_page *pga;
2473 struct brw_page *pgp;
2474 struct page **pages;
2483 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2484 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2485 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2487 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2489 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2490 LASSERT(lsm != NULL);
2491 LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2494 (count & (~CFS_PAGE_MASK)) != 0)
2497 /* XXX think again with misaligned I/O */
2498 npages = count >> PAGE_CACHE_SHIFT;
2500 if (rw == OBD_BRW_WRITE)
2501 brw_flags = OBD_BRW_ASYNC;
2503 OBD_ALLOC(pga, npages * sizeof(*pga));
2507 OBD_ALLOC(pages, npages * sizeof(*pages));
2508 if (pages == NULL) {
2509 OBD_FREE(pga, npages * sizeof(*pga));
2513 for (i = 0, pgp = pga, off = offset;
2515 i++, pgp++, off += PAGE_CACHE_SIZE) {
2517 LASSERT (pgp->pg == NULL); /* for cleanup */
2520 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2521 if (pgp->pg == NULL)
2525 pgp->count = PAGE_CACHE_SIZE;
2527 pgp->flag = brw_flags;
2530 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2531 ostid_id(&oa->o_oi), off,
2535 /* brw mode can only be used at client */
2536 LASSERT(ed->ed_next != NULL);
2537 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2540 if (rc != 0 || rw != OBD_BRW_READ)
2543 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2544 if (pgp->pg == NULL)
2549 vrc = echo_client_page_debug_check(lsm, pgp->pg,
2550 ostid_id(&oa->o_oi),
2551 pgp->off, pgp->count);
2552 if (vrc != 0 && rc == 0)
2555 OBD_PAGE_FREE(pgp->pg);
2557 OBD_FREE(pga, npages * sizeof(*pga));
2558 OBD_FREE(pages, npages * sizeof(*pages));
2562 static int echo_client_prep_commit(const struct lu_env *env,
2563 struct obd_export *exp, int rw,
2564 struct obdo *oa, struct echo_object *eco,
2565 obd_off offset, obd_size count,
2566 obd_size batch, struct obd_trans_info *oti,
2569 struct lov_stripe_md *lsm = eco->eo_lsm;
2570 struct obd_ioobj ioo;
2571 struct niobuf_local *lnb;
2572 struct niobuf_remote *rnb;
2574 obd_size npages, tot_pages;
2575 int i, ret = 0, brw_flags = 0;
2579 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2580 (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2583 npages = batch >> PAGE_CACHE_SHIFT;
2584 tot_pages = count >> PAGE_CACHE_SHIFT;
2586 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2587 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2589 if (lnb == NULL || rnb == NULL)
2590 GOTO(out, ret = -ENOMEM);
2592 if (rw == OBD_BRW_WRITE && async)
2593 brw_flags |= OBD_BRW_ASYNC;
2595 obdo_to_ioobj(oa, &ioo);
2599 for(; tot_pages; tot_pages -= npages) {
2602 if (tot_pages < npages)
2605 for (i = 0; i < npages; i++, off += PAGE_CACHE_SIZE) {
2606 rnb[i].offset = off;
2607 rnb[i].len = PAGE_CACHE_SIZE;
2608 rnb[i].flags = brw_flags;
2611 ioo.ioo_bufcnt = npages;
2612 oti->oti_transno = 0;
2615 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2619 LASSERT(lpages == npages);
2621 for (i = 0; i < lpages; i++) {
2622 struct page *page = lnb[i].page;
2624 /* read past eof? */
2625 if (page == NULL && lnb[i].rc == 0)
2629 lnb[i].flags |= OBD_BRW_ASYNC;
2631 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2632 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2633 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2636 if (rw == OBD_BRW_WRITE)
2637 echo_client_page_debug_setup(lsm, page, rw,
2638 ostid_id(&oa->o_oi),
2642 echo_client_page_debug_check(lsm, page,
2643 ostid_id(&oa->o_oi),
2648 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2649 rnb, npages, lnb, oti, ret);
2653 /* Reset oti otherwise it would confuse ldiskfs. */
2654 memset(oti, 0, sizeof(*oti));
2656 /* Reuse env context. */
2657 lu_context_exit((struct lu_context *)&env->le_ctx);
2658 lu_context_enter((struct lu_context *)&env->le_ctx);
2663 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2665 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2669 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2670 struct obd_export *exp,
2671 struct obd_ioctl_data *data,
2672 struct obd_trans_info *dummy_oti)
2674 struct obd_device *obd = class_exp2obd(exp);
2675 struct echo_device *ed = obd2echo_dev(obd);
2676 struct echo_client_obd *ec = ed->ed_ec;
2677 struct obdo *oa = &data->ioc_obdo1;
2678 struct echo_object *eco;
2684 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2686 rc = echo_get_object(&eco, ed, oa);
2690 oa->o_valid &= ~OBD_MD_FLHANDLE;
2692 /* OFD/obdfilter works only via prep/commit */
2693 test_mode = (long)data->ioc_pbuf1;
2697 if (ed->ed_next == NULL && test_mode != 3) {
2699 data->ioc_plen1 = data->ioc_count;
2702 /* Truncate batch size to maximum */
2703 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2704 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2706 switch (test_mode) {
2710 rc = echo_client_kbrw(ed, rw, oa,
2711 eco, data->ioc_offset,
2712 data->ioc_count, async, dummy_oti);
2715 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2716 eco, data->ioc_offset,
2717 data->ioc_count, data->ioc_plen1,
2723 echo_put_object(eco);
2728 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2729 int mode, obd_off offset, obd_size nob)
2731 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2732 struct lustre_handle *ulh = &oa->o_handle;
2733 struct echo_object *eco;
2738 if (ed->ed_next == NULL)
2739 RETURN(-EOPNOTSUPP);
2741 if (!(mode == LCK_PR || mode == LCK_PW))
2744 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2745 (nob & (~CFS_PAGE_MASK)) != 0)
2748 rc = echo_get_object (&eco, ed, oa);
2752 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2753 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2755 oa->o_valid |= OBD_MD_FLHANDLE;
2756 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2758 echo_put_object(eco);
2763 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2765 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2766 __u64 cookie = oa->o_handle.cookie;
2768 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2771 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2772 return cl_echo_cancel(ed, cookie);
2776 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2777 void *karg, void *uarg)
2779 #ifdef HAVE_SERVER_SUPPORT
2780 struct tgt_session_info *tsi;
2782 struct obd_device *obd = exp->exp_obd;
2783 struct echo_device *ed = obd2echo_dev(obd);
2784 struct echo_client_obd *ec = ed->ed_ec;
2785 struct echo_object *eco;
2786 struct obd_ioctl_data *data = karg;
2787 struct obd_trans_info dummy_oti;
2789 struct oti_req_ack_lock *ack_lock;
2792 int rw = OBD_BRW_READ;
2795 #ifdef HAVE_SERVER_SUPPORT
2796 struct lu_context echo_session;
2800 memset(&dummy_oti, 0, sizeof(dummy_oti));
2802 oa = &data->ioc_obdo1;
2803 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2804 oa->o_valid |= OBD_MD_FLGROUP;
2805 ostid_set_seq_echo(&oa->o_oi);
2808 /* This FID is unpacked just for validation at this point */
2809 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2817 rc = lu_env_init(env, LCT_DT_THREAD);
2819 GOTO(out_alloc, rc = -ENOMEM);
2821 #ifdef HAVE_SERVER_SUPPORT
2822 env->le_ses = &echo_session;
2823 rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2824 if (unlikely(rc < 0))
2826 lu_context_enter(env->le_ses);
2828 tsi = tgt_ses_info(env);
2829 tsi->tsi_exp = ec->ec_exp;
2832 case OBD_IOC_CREATE: /* may create echo object */
2833 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2834 GOTO (out, rc = -EPERM);
2836 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2837 data->ioc_plen1, &dummy_oti);
2840 #ifdef HAVE_SERVER_SUPPORT
2841 case OBD_IOC_ECHO_MD: {
2848 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2849 GOTO(out, rc = -EPERM);
2851 count = data->ioc_count;
2852 cmd = data->ioc_command;
2854 id = data->ioc_obdo2.o_oi.oi.oi_id;
2855 dirlen = data->ioc_plen1;
2856 OBD_ALLOC(dir, dirlen + 1);
2858 GOTO(out, rc = -ENOMEM);
2860 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2861 OBD_FREE(dir, data->ioc_plen1 + 1);
2862 GOTO(out, rc = -EFAULT);
2865 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2866 OBD_FREE(dir, dirlen + 1);
2869 case OBD_IOC_ECHO_ALLOC_SEQ: {
2870 struct lu_env *cl_env;
2875 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2876 GOTO(out, rc = -EPERM);
2878 cl_env = cl_env_get(&refcheck);
2880 GOTO(out, rc = PTR_ERR(cl_env));
2882 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2885 cl_env_put(cl_env, &refcheck);
2889 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2890 cl_env_put(cl_env, &refcheck);
2892 CERROR("%s: Can not alloc seq: rc = %d\n",
2897 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2900 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2901 if (copy_to_user(data->ioc_pbuf2, &max_count,
2906 #endif /* HAVE_SERVER_SUPPORT */
2907 case OBD_IOC_DESTROY:
2908 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2909 GOTO (out, rc = -EPERM);
2911 rc = echo_get_object(&eco, ed, oa);
2913 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2914 &dummy_oti, NULL, NULL);
2916 eco->eo_deleted = 1;
2917 echo_put_object(eco);
2921 case OBD_IOC_GETATTR:
2922 rc = echo_get_object(&eco, ed, oa);
2924 struct obd_info oinfo = { { { 0 } } };
2925 oinfo.oi_md = eco->eo_lsm;
2927 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2928 echo_put_object(eco);
2932 case OBD_IOC_SETATTR:
2933 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2934 GOTO (out, rc = -EPERM);
2936 rc = echo_get_object(&eco, ed, oa);
2938 struct obd_info oinfo = { { { 0 } } };
2940 oinfo.oi_md = eco->eo_lsm;
2942 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2943 echo_put_object(eco);
2947 case OBD_IOC_BRW_WRITE:
2948 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2949 GOTO (out, rc = -EPERM);
2953 case OBD_IOC_BRW_READ:
2954 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2957 case ECHO_IOC_GET_STRIPE:
2958 rc = echo_get_object(&eco, ed, oa);
2960 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2962 echo_put_object(eco);
2966 case ECHO_IOC_SET_STRIPE:
2967 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2968 GOTO (out, rc = -EPERM);
2970 if (data->ioc_pbuf1 == NULL) { /* unset */
2971 rc = echo_get_object(&eco, ed, oa);
2973 eco->eo_deleted = 1;
2974 echo_put_object(eco);
2977 rc = echo_create_object(env, ed, 0, oa,
2979 data->ioc_plen1, &dummy_oti);
2983 case ECHO_IOC_ENQUEUE:
2984 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2985 GOTO (out, rc = -EPERM);
2987 rc = echo_client_enqueue(exp, oa,
2988 data->ioc_conn1, /* lock mode */
2990 data->ioc_count);/*extent*/
2993 case ECHO_IOC_CANCEL:
2994 rc = echo_client_cancel(exp, oa);
2998 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2999 GOTO (out, rc = -ENOTTY);
3004 #ifdef HAVE_SERVER_SUPPORT
3005 lu_context_exit(env->le_ses);
3006 lu_context_fini(env->le_ses);
3013 /* XXX this should be in a helper also called by target_send_reply */
3014 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
3016 if (!ack_lock->mode)
3018 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
3024 static int echo_client_setup(const struct lu_env *env,
3025 struct obd_device *obddev, struct lustre_cfg *lcfg)
3027 struct echo_client_obd *ec = &obddev->u.echo_client;
3028 struct obd_device *tgt;
3029 struct obd_uuid echo_uuid = { "ECHO_UUID" };
3030 struct obd_connect_data *ocd = NULL;
3034 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3035 CERROR("requires a TARGET OBD name\n");
3039 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3040 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3041 CERROR("device not attached or not set up (%s)\n",
3042 lustre_cfg_string(lcfg, 1));
3046 spin_lock_init(&ec->ec_lock);
3047 CFS_INIT_LIST_HEAD (&ec->ec_objects);
3048 CFS_INIT_LIST_HEAD (&ec->ec_locks);
3050 ec->ec_nstripes = 0;
3052 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3053 #ifdef HAVE_SERVER_SUPPORT
3054 lu_context_tags_update(ECHO_MD_CTX_TAG);
3055 lu_session_tags_update(ECHO_MD_SES_TAG);
3057 CERROR("Local operations are NOT supported on client side. "
3058 "Only remote operations are supported. Metadata client "
3059 "must be run on server side.\n");
3064 OBD_ALLOC(ocd, sizeof(*ocd));
3066 CERROR("Can't alloc ocd connecting to %s\n",
3067 lustre_cfg_string(lcfg, 1));
3071 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3072 OBD_CONNECT_BRW_SIZE |
3073 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3074 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
3076 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
3077 ocd->ocd_version = LUSTRE_VERSION_CODE;
3078 ocd->ocd_group = FID_SEQ_ECHO;
3080 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3082 /* Turn off pinger because it connects to tgt obd directly. */
3083 spin_lock(&tgt->obd_dev_lock);
3084 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3085 spin_unlock(&tgt->obd_dev_lock);
3088 OBD_FREE(ocd, sizeof(*ocd));
3091 CERROR("fail to connect to device %s\n",
3092 lustre_cfg_string(lcfg, 1));
3099 static int echo_client_cleanup(struct obd_device *obddev)
3101 struct echo_device *ed = obd2echo_dev(obddev);
3102 struct echo_client_obd *ec = &obddev->u.echo_client;
3106 /*Do nothing for Metadata echo client*/
3110 if (ed->ed_next_ismd) {
3111 #ifdef HAVE_SERVER_SUPPORT
3112 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3113 lu_session_tags_clear(ECHO_MD_SES_TAG);
3115 CERROR("This is client-side only module, does not support "
3116 "metadata echo client.\n");
3121 if (!cfs_list_empty(&obddev->obd_exports)) {
3122 CERROR("still has clients!\n");
3126 LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3127 rc = obd_disconnect(ec->ec_exp);
3129 CERROR("fail to disconnect device: %d\n", rc);
3134 static int echo_client_connect(const struct lu_env *env,
3135 struct obd_export **exp,
3136 struct obd_device *src, struct obd_uuid *cluuid,
3137 struct obd_connect_data *data, void *localdata)
3140 struct lustre_handle conn = { 0 };
3143 rc = class_connect(&conn, src, cluuid);
3145 *exp = class_conn2export(&conn);
3151 static int echo_client_disconnect(struct obd_export *exp)
3157 GOTO(out, rc = -EINVAL);
3159 rc = class_disconnect(exp);
3165 static struct obd_ops echo_client_obd_ops = {
3166 .o_owner = THIS_MODULE,
3167 .o_iocontrol = echo_client_iocontrol,
3168 .o_connect = echo_client_connect,
3169 .o_disconnect = echo_client_disconnect
3172 int echo_client_init(void)
3174 struct lprocfs_static_vars lvars = { 0 };
3177 lprocfs_echo_init_vars(&lvars);
3179 rc = lu_kmem_init(echo_caches);
3181 rc = class_register_type(&echo_client_obd_ops, NULL, NULL,
3182 #ifndef HAVE_ONLY_PROCFS_SEQ
3185 LUSTRE_ECHO_CLIENT_NAME,
3188 lu_kmem_fini(echo_caches);
3193 void echo_client_exit(void)
3195 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3196 lu_kmem_fini(echo_caches);
3200 static int __init obdecho_init(void)
3202 struct lprocfs_static_vars lvars;
3206 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3208 LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3210 lprocfs_echo_init_vars(&lvars);
3212 # ifdef HAVE_SERVER_SUPPORT
3213 rc = echo_persistent_pages_init();
3217 rc = class_register_type(&echo_obd_ops, NULL, NULL,
3218 #ifndef HAVE_ONLY_PROCFS_SEQ
3221 LUSTRE_ECHO_NAME, NULL);
3226 rc = echo_client_init();
3228 # ifdef HAVE_SERVER_SUPPORT
3232 class_unregister_type(LUSTRE_ECHO_NAME);
3234 echo_persistent_pages_fini();
3240 static void /*__exit*/ obdecho_exit(void)
3244 # ifdef HAVE_SERVER_SUPPORT
3245 class_unregister_type(LUSTRE_ECHO_NAME);
3246 echo_persistent_pages_fini();
3250 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3251 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3252 MODULE_LICENSE("GPL");
3254 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3255 #endif /* __KERNEL__ */
3257 /** @} echo_client */