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, 2012, 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
38 #include <linux/libcfs/libcfs.h>
41 #include <obd_support.h>
42 #include <obd_class.h>
43 #include <lustre_debug.h>
44 #include <lprocfs_status.h>
45 #include <cl_object.h>
46 #include <md_object.h>
47 #include <lustre_fid.h>
48 #include <lustre_acl.h>
49 #include <lustre_net.h>
52 #include "echo_internal.h"
54 /** \defgroup echo_client Echo Client
59 struct cl_device ed_cl;
60 struct echo_client_obd *ed_ec;
62 struct cl_site ed_site_myself;
63 struct cl_site *ed_site;
64 struct lu_device *ed_next;
67 struct lu_client_seq *ed_cl_seq;
71 struct cl_object eo_cl;
72 struct cl_object_header eo_hdr;
74 struct echo_device *eo_dev;
75 struct list_head eo_obj_chain;
76 struct lov_stripe_md *eo_lsm;
81 struct echo_object_conf {
82 struct cl_object_conf eoc_cl;
83 struct lov_stripe_md **eoc_md;
87 struct cl_page_slice ep_cl;
89 struct page *ep_vmpage;
93 struct cl_lock_slice el_cl;
94 struct list_head el_chain;
95 struct echo_object *el_object;
101 struct cl_io_slice ei_cl;
106 struct cl_req_slice er_cl;
110 static int echo_client_setup(const struct lu_env *env,
111 struct obd_device *obddev,
112 struct lustre_cfg *lcfg);
113 static int echo_client_cleanup(struct obd_device *obddev);
116 /** \defgroup echo_helpers Helper functions
119 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
121 return container_of0(dev, struct echo_device, ed_cl);
124 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
129 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
131 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
134 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
139 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
141 return container_of(o, struct echo_object, eo_cl);
144 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
146 return container_of(s, struct echo_page, ep_cl);
149 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
151 return container_of(s, struct echo_lock, el_cl);
154 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
156 return ecl->el_cl.cls_lock;
159 static struct lu_context_key echo_thread_key;
160 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
162 struct echo_thread_info *info;
163 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
164 LASSERT(info != NULL);
169 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
171 return container_of(c, struct echo_object_conf, eoc_cl);
174 /** @} echo_helpers */
176 static struct echo_object *cl_echo_object_find(struct echo_device *d,
177 struct lov_stripe_md **lsm);
178 static int cl_echo_object_put(struct echo_object *eco);
179 static int cl_echo_enqueue (struct echo_object *eco, obd_off start,
180 obd_off end, int mode, __u64 *cookie);
181 static int cl_echo_cancel (struct echo_device *d, __u64 cookie);
182 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
183 struct page **pages, int npages, int async);
185 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
187 struct echo_thread_info {
188 struct echo_object_conf eti_conf;
189 struct lustre_md eti_md;
191 struct cl_2queue eti_queue;
193 struct cl_lock_descr eti_descr;
194 struct lu_fid eti_fid;
195 struct lu_fid eti_fid2;
196 struct md_op_spec eti_spec;
197 struct lov_mds_md_v3 eti_lmm;
198 struct lov_user_md_v3 eti_lum;
199 struct md_attr eti_ma;
200 struct lu_name eti_lname;
201 /* per-thread values, can be re-used */
205 struct lu_buf eti_buf;
206 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
209 /* No session used right now */
210 struct echo_session_info {
214 static struct kmem_cache *echo_lock_kmem;
215 static struct kmem_cache *echo_object_kmem;
216 static struct kmem_cache *echo_thread_kmem;
217 static struct kmem_cache *echo_session_kmem;
218 //static struct kmem_cache *echo_req_kmem;
220 static struct lu_kmem_descr echo_caches[] = {
222 .ckd_cache = &echo_lock_kmem,
223 .ckd_name = "echo_lock_kmem",
224 .ckd_size = sizeof (struct echo_lock)
227 .ckd_cache = &echo_object_kmem,
228 .ckd_name = "echo_object_kmem",
229 .ckd_size = sizeof (struct echo_object)
232 .ckd_cache = &echo_thread_kmem,
233 .ckd_name = "echo_thread_kmem",
234 .ckd_size = sizeof (struct echo_thread_info)
237 .ckd_cache = &echo_session_kmem,
238 .ckd_name = "echo_session_kmem",
239 .ckd_size = sizeof (struct echo_session_info)
243 .ckd_cache = &echo_req_kmem,
244 .ckd_name = "echo_req_kmem",
245 .ckd_size = sizeof (struct echo_req)
253 /** \defgroup echo_page Page operations
255 * Echo page operations.
259 static struct page *echo_page_vmpage(const struct lu_env *env,
260 const struct cl_page_slice *slice)
262 return cl2echo_page(slice)->ep_vmpage;
265 static int echo_page_own(const struct lu_env *env,
266 const struct cl_page_slice *slice,
267 struct cl_io *io, int nonblock)
269 struct echo_page *ep = cl2echo_page(slice);
272 mutex_lock(&ep->ep_lock);
273 else if (!mutex_trylock(&ep->ep_lock))
278 static void echo_page_disown(const struct lu_env *env,
279 const struct cl_page_slice *slice,
282 struct echo_page *ep = cl2echo_page(slice);
284 LASSERT(mutex_is_locked(&ep->ep_lock));
285 mutex_unlock(&ep->ep_lock);
288 static void echo_page_discard(const struct lu_env *env,
289 const struct cl_page_slice *slice,
290 struct cl_io *unused)
292 cl_page_delete(env, slice->cpl_page);
295 static int echo_page_is_vmlocked(const struct lu_env *env,
296 const struct cl_page_slice *slice)
298 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
303 static void echo_page_completion(const struct lu_env *env,
304 const struct cl_page_slice *slice,
307 LASSERT(slice->cpl_page->cp_sync_io != NULL);
310 static void echo_page_fini(const struct lu_env *env,
311 struct cl_page_slice *slice)
313 struct echo_page *ep = cl2echo_page(slice);
314 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
315 struct page *vmpage = ep->ep_vmpage;
317 atomic_dec(&eco->eo_npages);
318 page_cache_release(vmpage);
321 static int echo_page_prep(const struct lu_env *env,
322 const struct cl_page_slice *slice,
323 struct cl_io *unused)
328 static int echo_page_print(const struct lu_env *env,
329 const struct cl_page_slice *slice,
330 void *cookie, lu_printer_t printer)
332 struct echo_page *ep = cl2echo_page(slice);
334 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
335 ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
339 static const struct cl_page_operations echo_page_ops = {
340 .cpo_own = echo_page_own,
341 .cpo_disown = echo_page_disown,
342 .cpo_discard = echo_page_discard,
343 .cpo_vmpage = echo_page_vmpage,
344 .cpo_fini = echo_page_fini,
345 .cpo_print = echo_page_print,
346 .cpo_is_vmlocked = echo_page_is_vmlocked,
349 .cpo_prep = echo_page_prep,
350 .cpo_completion = echo_page_completion,
353 .cpo_prep = echo_page_prep,
354 .cpo_completion = echo_page_completion,
360 /** \defgroup echo_lock Locking
362 * echo lock operations
366 static void echo_lock_fini(const struct lu_env *env,
367 struct cl_lock_slice *slice)
369 struct echo_lock *ecl = cl2echo_lock(slice);
371 LASSERT(list_empty(&ecl->el_chain));
372 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
375 static void echo_lock_delete(const struct lu_env *env,
376 const struct cl_lock_slice *slice)
378 struct echo_lock *ecl = cl2echo_lock(slice);
380 LASSERT(list_empty(&ecl->el_chain));
383 static int echo_lock_fits_into(const struct lu_env *env,
384 const struct cl_lock_slice *slice,
385 const struct cl_lock_descr *need,
386 const struct cl_io *unused)
391 static struct cl_lock_operations echo_lock_ops = {
392 .clo_fini = echo_lock_fini,
393 .clo_delete = echo_lock_delete,
394 .clo_fits_into = echo_lock_fits_into
399 /** \defgroup echo_cl_ops cl_object operations
401 * operations for cl_object
405 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
406 struct cl_page *page, struct page *vmpage)
408 struct echo_page *ep = cl_object_page_slice(obj, page);
409 struct echo_object *eco = cl2echo_obj(obj);
411 ep->ep_vmpage = vmpage;
412 page_cache_get(vmpage);
413 mutex_init(&ep->ep_lock);
414 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
415 atomic_inc(&eco->eo_npages);
419 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
425 static int echo_lock_init(const struct lu_env *env,
426 struct cl_object *obj, struct cl_lock *lock,
427 const struct cl_io *unused)
429 struct echo_lock *el;
431 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, __GFP_IO);
433 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
434 el->el_object = cl2echo_obj(obj);
435 INIT_LIST_HEAD(&el->el_chain);
436 atomic_set(&el->el_refcount, 0);
438 return el == NULL ? -ENOMEM : 0;
441 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
442 const struct cl_object_conf *conf)
447 static const struct cl_object_operations echo_cl_obj_ops = {
448 .coo_page_init = echo_page_init,
449 .coo_lock_init = echo_lock_init,
450 .coo_io_init = echo_io_init,
451 .coo_conf_set = echo_conf_set
453 /** @} echo_cl_ops */
455 /** \defgroup echo_lu_ops lu_object operations
457 * operations for echo lu object.
461 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
462 const struct lu_object_conf *conf)
464 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
465 struct echo_client_obd *ec = ed->ed_ec;
466 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
469 struct lu_object *below;
470 struct lu_device *under;
473 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
477 lu_object_add(obj, below);
480 if (!ed->ed_next_ismd) {
481 const struct cl_object_conf *cconf = lu2cl_conf(conf);
482 struct echo_object_conf *econf = cl2echo_conf(cconf);
484 LASSERT(econf->eoc_md);
485 eco->eo_lsm = *econf->eoc_md;
486 /* clear the lsm pointer so that it won't get freed. */
487 *econf->eoc_md = NULL;
493 atomic_set(&eco->eo_npages, 0);
494 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
496 spin_lock(&ec->ec_lock);
497 list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
498 spin_unlock(&ec->ec_lock);
503 /* taken from osc_unpackmd() */
504 static int echo_alloc_memmd(struct echo_device *ed,
505 struct lov_stripe_md **lsmp)
509 /* If export is lov/osc then use their obd method */
510 if (ed->ed_next != NULL)
511 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
512 /* OFD has no unpackmd method, do everything here */
513 lsm_size = lov_stripe_md_size(1);
515 LASSERT(*lsmp == NULL);
516 OBD_ALLOC(*lsmp, lsm_size);
520 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
521 if ((*lsmp)->lsm_oinfo[0] == NULL) {
522 OBD_FREE(*lsmp, lsm_size);
526 loi_init((*lsmp)->lsm_oinfo[0]);
527 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
528 ostid_set_seq_echo(&(*lsmp)->lsm_oi);
533 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
537 /* If export is lov/osc then use their obd method */
538 if (ed->ed_next != NULL)
539 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
540 /* OFD has no unpackmd method, do everything here */
541 lsm_size = lov_stripe_md_size(1);
543 LASSERT(*lsmp != NULL);
544 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
545 OBD_FREE(*lsmp, lsm_size);
550 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
552 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
553 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
555 LASSERT(atomic_read(&eco->eo_npages) == 0);
557 spin_lock(&ec->ec_lock);
558 list_del_init(&eco->eo_obj_chain);
559 spin_unlock(&ec->ec_lock);
562 lu_object_header_fini(obj->lo_header);
565 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
566 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
569 static int echo_object_print(const struct lu_env *env, void *cookie,
570 lu_printer_t p, const struct lu_object *o)
572 struct echo_object *obj = cl2echo_obj(lu2cl(o));
574 return (*p)(env, cookie, "echoclient-object@%p", obj);
577 static const struct lu_object_operations echo_lu_obj_ops = {
578 .loo_object_init = echo_object_init,
579 .loo_object_delete = NULL,
580 .loo_object_release = NULL,
581 .loo_object_free = echo_object_free,
582 .loo_object_print = echo_object_print,
583 .loo_object_invariant = NULL
585 /** @} echo_lu_ops */
587 /** \defgroup echo_lu_dev_ops lu_device operations
589 * Operations for echo lu device.
593 static struct lu_object *echo_object_alloc(const struct lu_env *env,
594 const struct lu_object_header *hdr,
595 struct lu_device *dev)
597 struct echo_object *eco;
598 struct lu_object *obj = NULL;
600 /* we're the top dev. */
601 LASSERT(hdr == NULL);
602 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, __GFP_IO);
604 struct cl_object_header *hdr = &eco->eo_hdr;
606 obj = &echo_obj2cl(eco)->co_lu;
607 cl_object_header_init(hdr);
608 lu_object_init(obj, &hdr->coh_lu, dev);
609 lu_object_add_top(&hdr->coh_lu, obj);
611 eco->eo_cl.co_ops = &echo_cl_obj_ops;
612 obj->lo_ops = &echo_lu_obj_ops;
617 static struct lu_device_operations echo_device_lu_ops = {
618 .ldo_object_alloc = echo_object_alloc,
621 /** @} echo_lu_dev_ops */
623 static struct cl_device_operations echo_device_cl_ops = {
626 /** \defgroup echo_init Setup and teardown
628 * Init and fini functions for echo client.
632 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
634 struct cl_site *site = &ed->ed_site_myself;
637 /* initialize site */
638 rc = cl_site_init(site, &ed->ed_cl);
640 CERROR("Cannot initialize site for echo client(%d)\n", rc);
644 rc = lu_site_init_finish(&site->cs_lu);
652 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
655 if (!ed->ed_next_ismd)
656 cl_site_fini(ed->ed_site);
661 static void *echo_thread_key_init(const struct lu_context *ctx,
662 struct lu_context_key *key)
664 struct echo_thread_info *info;
666 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, __GFP_IO);
668 info = ERR_PTR(-ENOMEM);
672 static void echo_thread_key_fini(const struct lu_context *ctx,
673 struct lu_context_key *key, void *data)
675 struct echo_thread_info *info = data;
676 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
679 static void echo_thread_key_exit(const struct lu_context *ctx,
680 struct lu_context_key *key, void *data)
684 static struct lu_context_key echo_thread_key = {
685 .lct_tags = LCT_CL_THREAD,
686 .lct_init = echo_thread_key_init,
687 .lct_fini = echo_thread_key_fini,
688 .lct_exit = echo_thread_key_exit
691 static void *echo_session_key_init(const struct lu_context *ctx,
692 struct lu_context_key *key)
694 struct echo_session_info *session;
696 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, __GFP_IO);
698 session = ERR_PTR(-ENOMEM);
702 static void echo_session_key_fini(const struct lu_context *ctx,
703 struct lu_context_key *key, void *data)
705 struct echo_session_info *session = data;
706 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
709 static void echo_session_key_exit(const struct lu_context *ctx,
710 struct lu_context_key *key, void *data)
714 static struct lu_context_key echo_session_key = {
715 .lct_tags = LCT_SESSION,
716 .lct_init = echo_session_key_init,
717 .lct_fini = echo_session_key_fini,
718 .lct_exit = echo_session_key_exit
721 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
723 #define ECHO_SEQ_WIDTH 0xffffffff
724 static int echo_fid_init(struct echo_device *ed, char *obd_name,
725 struct seq_server_site *ss)
730 OBD_ALLOC_PTR(ed->ed_cl_seq);
731 if (ed->ed_cl_seq == NULL)
734 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
736 GOTO(out_free_seq, rc = -ENOMEM);
738 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
740 /* Init client side sequence-manager */
741 rc = seq_client_init(ed->ed_cl_seq, NULL,
743 prefix, ss->ss_server_seq);
744 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
745 OBD_FREE(prefix, MAX_OBD_NAME + 5);
747 GOTO(out_free_seq, rc);
752 OBD_FREE_PTR(ed->ed_cl_seq);
753 ed->ed_cl_seq = NULL;
757 static int echo_fid_fini(struct obd_device *obddev)
759 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;
770 static struct lu_device *echo_device_alloc(const struct lu_env *env,
771 struct lu_device_type *t,
772 struct lustre_cfg *cfg)
774 struct lu_device *next;
775 struct echo_device *ed;
776 struct cl_device *cd;
777 struct obd_device *obd = NULL; /* to keep compiler happy */
778 struct obd_device *tgt;
779 const char *tgt_type_name;
785 GOTO(out, rc = -ENOMEM);
789 rc = cl_device_init(cd, t);
793 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
794 cd->cd_ops = &echo_device_cl_ops;
797 obd = class_name2obd(lustre_cfg_string(cfg, 0));
798 LASSERT(obd != NULL);
799 LASSERT(env != NULL);
801 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
803 CERROR("Can not find tgt device %s\n",
804 lustre_cfg_string(cfg, 1));
805 GOTO(out, rc = -ENODEV);
808 next = tgt->obd_lu_dev;
809 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
810 ed->ed_next_ismd = 1;
812 ed->ed_next_ismd = 0;
813 rc = echo_site_init(env, ed);
819 rc = echo_client_setup(env, obd, cfg);
823 ed->ed_ec = &obd->u.echo_client;
826 if (ed->ed_next_ismd) {
827 /* Suppose to connect to some Metadata layer */
829 struct lu_device *ld;
833 CERROR("%s is not lu device type!\n",
834 lustre_cfg_string(cfg, 1));
835 GOTO(out, rc = -EINVAL);
838 tgt_type_name = lustre_cfg_string(cfg, 2);
839 if (!tgt_type_name) {
840 CERROR("%s no type name for echo %s setup\n",
841 lustre_cfg_string(cfg, 1),
842 tgt->obd_type->typ_name);
843 GOTO(out, rc = -EINVAL);
848 spin_lock(&ls->ls_ld_lock);
849 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
850 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
855 spin_unlock(&ls->ls_ld_lock);
858 CERROR("%s is not lu device type!\n",
859 lustre_cfg_string(cfg, 1));
860 GOTO(out, rc = -EINVAL);
864 /* For MD echo client, it will use the site in MDS stack */
865 ed->ed_site_myself.cs_lu = *ls;
866 ed->ed_site = &ed->ed_site_myself;
867 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
868 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
870 CERROR("echo fid init error %d\n", rc);
874 /* if echo client is to be stacked upon ost device, the next is
875 * NULL since ost is not a clio device so far */
876 if (next != NULL && !lu_device_is_cl(next))
879 tgt_type_name = tgt->obd_type->typ_name;
881 LASSERT(next != NULL);
882 if (next->ld_site != NULL)
883 GOTO(out, rc = -EBUSY);
885 next->ld_site = &ed->ed_site->cs_lu;
886 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
887 next->ld_type->ldt_name,
892 /* Tricky case, I have to determine the obd type since
893 * CLIO uses the different parameters to initialize
894 * objects for lov & osc. */
895 if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
896 ed->ed_next_islov = 1;
898 LASSERT(strcmp(tgt_type_name,
899 LUSTRE_OSC_NAME) == 0);
901 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
905 return &cd->cd_lu_dev;
910 rc2 = echo_client_cleanup(obd);
912 CERROR("Cleanup obd device %s error(%d)\n",
917 echo_site_fini(env, ed);
919 cl_device_fini(&ed->ed_cl);
929 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
930 const char *name, struct lu_device *next)
936 static struct lu_device *echo_device_fini(const struct lu_env *env,
939 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
940 struct lu_device *next = ed->ed_next;
942 while (next && !ed->ed_next_ismd)
943 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
947 static void echo_lock_release(const struct lu_env *env,
948 struct echo_lock *ecl,
951 struct cl_lock *clk = echo_lock2cl(ecl);
955 cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
957 cl_lock_mutex_get(env, clk);
958 cl_lock_cancel(env, clk);
959 cl_lock_delete(env, clk);
960 cl_lock_mutex_put(env, clk);
962 cl_lock_put(env, clk);
965 static struct lu_device *echo_device_free(const struct lu_env *env,
968 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
969 struct echo_client_obd *ec = ed->ed_ec;
970 struct echo_object *eco;
971 struct lu_device *next = ed->ed_next;
973 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
976 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
978 /* check if there are objects still alive.
979 * It shouldn't have any object because lu_site_purge would cleanup
980 * all of cached objects. Anyway, probably the echo device is being
981 * parallelly accessed.
983 spin_lock(&ec->ec_lock);
984 list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
986 spin_unlock(&ec->ec_lock);
989 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
992 "Waiting for the reference of echo object to be dropped\n");
994 /* Wait for the last reference to be dropped. */
995 spin_lock(&ec->ec_lock);
996 while (!list_empty(&ec->ec_objects)) {
997 spin_unlock(&ec->ec_lock);
998 CERROR("echo_client still has objects at cleanup time, "
999 "wait for 1 second\n");
1000 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
1001 cfs_time_seconds(1));
1002 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1003 spin_lock(&ec->ec_lock);
1005 spin_unlock(&ec->ec_lock);
1007 LASSERT(list_empty(&ec->ec_locks));
1009 CDEBUG(D_INFO, "No object exists, exiting...\n");
1011 echo_client_cleanup(d->ld_obd);
1012 echo_fid_fini(d->ld_obd);
1013 while (next && !ed->ed_next_ismd)
1014 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1016 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1017 echo_site_fini(env, ed);
1018 cl_device_fini(&ed->ed_cl);
1024 static const struct lu_device_type_operations echo_device_type_ops = {
1025 .ldto_init = echo_type_init,
1026 .ldto_fini = echo_type_fini,
1028 .ldto_start = echo_type_start,
1029 .ldto_stop = echo_type_stop,
1031 .ldto_device_alloc = echo_device_alloc,
1032 .ldto_device_free = echo_device_free,
1033 .ldto_device_init = echo_device_init,
1034 .ldto_device_fini = echo_device_fini
1037 static struct lu_device_type echo_device_type = {
1038 .ldt_tags = LU_DEVICE_CL,
1039 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1040 .ldt_ops = &echo_device_type_ops,
1041 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1045 /** \defgroup echo_exports Exported operations
1047 * exporting functions to echo client
1052 /* Interfaces to echo client obd device */
1053 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1054 struct lov_stripe_md **lsmp)
1057 struct echo_thread_info *info;
1058 struct echo_object_conf *conf;
1059 struct lov_stripe_md *lsm;
1060 struct echo_object *eco;
1061 struct cl_object *obj;
1069 LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi));
1070 LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n",
1071 POSTID(&lsm->lsm_oi));
1073 /* Never return an object if the obd is to be freed. */
1074 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1075 return ERR_PTR(-ENODEV);
1077 env = cl_env_get(&refcheck);
1081 info = echo_env_info(env);
1082 conf = &info->eti_conf;
1084 if (!d->ed_next_islov) {
1085 struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1086 LASSERT(oinfo != NULL);
1087 oinfo->loi_oi = lsm->lsm_oi;
1088 conf->eoc_cl.u.coc_oinfo = oinfo;
1090 struct lustre_md *md;
1092 memset(md, 0, sizeof(*md));
1094 conf->eoc_cl.u.coc_md = md;
1097 conf->eoc_md = lsmp;
1099 fid = &info->eti_fid;
1100 rc = ostid_to_fid(fid, &lsm->lsm_oi, 0);
1102 GOTO(out, eco = ERR_PTR(rc));
1104 /* In the function below, .hs_keycmp resolves to
1105 * lu_obj_hop_keycmp() */
1106 /* coverity[overrun-buffer-val] */
1107 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1109 GOTO(out, eco = (void*)obj);
1111 eco = cl2echo_obj(obj);
1112 if (eco->eo_deleted) {
1113 cl_object_put(env, obj);
1114 eco = ERR_PTR(-EAGAIN);
1118 cl_env_put(env, &refcheck);
1122 static int cl_echo_object_put(struct echo_object *eco)
1125 struct cl_object *obj = echo_obj2cl(eco);
1128 env = cl_env_get(&refcheck);
1130 return PTR_ERR(env);
1132 /* an external function to kill an object? */
1133 if (eco->eo_deleted) {
1134 struct lu_object_header *loh = obj->co_lu.lo_header;
1135 LASSERT(&eco->eo_hdr == luh2coh(loh));
1136 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1139 cl_object_put(env, obj);
1140 cl_env_put(env, &refcheck);
1144 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1145 obd_off start, obd_off end, int mode,
1146 __u64 *cookie , __u32 enqflags)
1149 struct cl_lock *lck;
1150 struct cl_object *obj;
1151 struct cl_lock_descr *descr;
1152 struct echo_thread_info *info;
1155 info = echo_env_info(env);
1157 descr = &info->eti_descr;
1158 obj = echo_obj2cl(eco);
1160 descr->cld_obj = obj;
1161 descr->cld_start = cl_index(obj, start);
1162 descr->cld_end = cl_index(obj, end);
1163 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1164 descr->cld_enq_flags = enqflags;
1167 lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1169 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1170 struct echo_lock *el;
1172 rc = cl_wait(env, lck);
1174 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1175 spin_lock(&ec->ec_lock);
1176 if (list_empty(&el->el_chain)) {
1177 list_add(&el->el_chain, &ec->ec_locks);
1178 el->el_cookie = ++ec->ec_unique;
1180 atomic_inc(&el->el_refcount);
1181 *cookie = el->el_cookie;
1182 spin_unlock(&ec->ec_lock);
1184 cl_lock_release(env, lck, "ec enqueue", current);
1190 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1191 int mode, __u64 *cookie)
1193 struct echo_thread_info *info;
1199 env = cl_env_get(&refcheck);
1201 return PTR_ERR(env);
1203 info = echo_env_info(env);
1206 io->ci_ignore_layout = 1;
1207 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1210 LASSERT(result == 0);
1212 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1213 cl_io_fini(env, io);
1216 cl_env_put(env, &refcheck);
1220 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1223 struct echo_client_obd *ec = ed->ed_ec;
1224 struct echo_lock *ecl = NULL;
1225 struct list_head *el;
1226 int found = 0, still_used = 0;
1228 LASSERT(ec != NULL);
1229 spin_lock(&ec->ec_lock);
1230 list_for_each (el, &ec->ec_locks) {
1231 ecl = list_entry (el, struct echo_lock, el_chain);
1232 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1233 found = (ecl->el_cookie == cookie);
1235 if (atomic_dec_and_test(&ecl->el_refcount))
1236 list_del_init(&ecl->el_chain);
1242 spin_unlock(&ec->ec_lock);
1247 echo_lock_release(env, ecl, still_used);
1251 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1257 env = cl_env_get(&refcheck);
1259 return PTR_ERR(env);
1261 rc = cl_echo_cancel0(env, ed, cookie);
1263 cl_env_put(env, &refcheck);
1267 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1268 enum cl_req_type unused, struct cl_2queue *queue)
1270 struct cl_page *clp;
1271 struct cl_page *temp;
1274 cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1276 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1279 result = result ?: rc;
1284 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1285 struct page **pages, int npages, int async)
1288 struct echo_thread_info *info;
1289 struct cl_object *obj = echo_obj2cl(eco);
1290 struct echo_device *ed = eco->eo_dev;
1291 struct cl_2queue *queue;
1293 struct cl_page *clp;
1294 struct lustre_handle lh = { 0 };
1295 int page_size = cl_page_size(obj);
1300 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1301 LASSERT(ed->ed_next != NULL);
1302 env = cl_env_get(&refcheck);
1304 return PTR_ERR(env);
1306 info = echo_env_info(env);
1308 queue = &info->eti_queue;
1310 cl_2queue_init(queue);
1312 io->ci_ignore_layout = 1;
1313 rc = cl_io_init(env, io, CIT_MISC, obj);
1319 rc = cl_echo_enqueue0(env, eco, offset,
1320 offset + npages * PAGE_CACHE_SIZE - 1,
1321 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1324 GOTO(error_lock, rc);
1326 for (i = 0; i < npages; i++) {
1328 clp = cl_page_find(env, obj, cl_index(obj, offset),
1329 pages[i], CPT_TRANSIENT);
1334 LASSERT(clp->cp_type == CPT_TRANSIENT);
1336 rc = cl_page_own(env, io, clp);
1338 LASSERT(clp->cp_state == CPS_FREEING);
1339 cl_page_put(env, clp);
1343 cl_2queue_add(queue, clp);
1345 /* drop the reference count for cl_page_find, so that the page
1346 * will be freed in cl_2queue_fini. */
1347 cl_page_put(env, clp);
1348 cl_page_clip(env, clp, 0, page_size);
1350 offset += page_size;
1354 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1356 async = async && (typ == CRT_WRITE);
1358 rc = cl_echo_async_brw(env, io, typ, queue);
1360 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1361 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1362 async ? "async" : "sync", rc);
1365 cl_echo_cancel0(env, ed, lh.cookie);
1367 cl_2queue_discard(env, io, queue);
1368 cl_2queue_disown(env, io, queue);
1369 cl_2queue_fini(env, queue);
1370 cl_io_fini(env, io);
1372 cl_env_put(env, &refcheck);
1375 /** @} echo_exports */
1378 static obd_id last_object_id;
1381 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1383 struct lov_stripe_md *ulsm = _ulsm;
1386 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1390 if (copy_to_user (ulsm, lsm, sizeof(*ulsm)))
1393 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1394 if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1395 sizeof(lsm->lsm_oinfo[0])))
1402 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1403 void *ulsm, int ulsm_nob)
1405 struct echo_client_obd *ec = ed->ed_ec;
1408 if (ulsm_nob < sizeof (*lsm))
1411 if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
1414 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1415 lsm->lsm_magic != LOV_MAGIC ||
1416 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1417 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1421 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1422 if (copy_from_user(lsm->lsm_oinfo[i],
1423 ((struct lov_stripe_md *)ulsm)-> \
1425 sizeof(lsm->lsm_oinfo[0])))
1431 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1434 sprintf(name, LPU64, id);
1435 lname->ln_name = name;
1436 lname->ln_namelen = strlen(name);
1439 /* similar to mdt_attr_get_complex */
1440 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1443 struct echo_thread_info *info = echo_env_info(env);
1446 LASSERT(ma->ma_lmm_size > 0);
1448 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1452 /* big_lmm may need to be grown */
1453 if (info->eti_big_lmmsize < rc) {
1454 int size = size_roundup_power2(rc);
1456 if (info->eti_big_lmmsize > 0) {
1457 /* free old buffer */
1458 LASSERT(info->eti_big_lmm);
1459 OBD_FREE_LARGE(info->eti_big_lmm,
1460 info->eti_big_lmmsize);
1461 info->eti_big_lmm = NULL;
1462 info->eti_big_lmmsize = 0;
1465 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1466 if (info->eti_big_lmm == NULL)
1468 info->eti_big_lmmsize = size;
1470 LASSERT(info->eti_big_lmmsize >= rc);
1472 info->eti_buf.lb_buf = info->eti_big_lmm;
1473 info->eti_buf.lb_len = info->eti_big_lmmsize;
1474 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1478 ma->ma_valid |= MA_LOV;
1479 ma->ma_lmm = info->eti_big_lmm;
1480 ma->ma_lmm_size = rc;
1485 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1488 struct echo_thread_info *info = echo_env_info(env);
1489 struct lu_buf *buf = &info->eti_buf;
1490 umode_t mode = lu_object_attr(&next->mo_lu);
1491 int need = ma->ma_need;
1496 if (need & MA_INODE) {
1497 ma->ma_need = MA_INODE;
1498 rc = mo_attr_get(env, next, ma);
1501 ma->ma_valid |= MA_INODE;
1504 if (need & MA_LOV) {
1505 if (S_ISREG(mode) || S_ISDIR(mode)) {
1506 LASSERT(ma->ma_lmm_size > 0);
1507 buf->lb_buf = ma->ma_lmm;
1508 buf->lb_len = ma->ma_lmm_size;
1509 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1511 ma->ma_lmm_size = rc2;
1512 ma->ma_valid |= MA_LOV;
1513 } else if (rc2 == -ENODATA) {
1515 ma->ma_lmm_size = 0;
1516 } else if (rc2 == -ERANGE) {
1517 rc2 = echo_big_lmm_get(env, next, ma);
1519 GOTO(out, rc = rc2);
1521 GOTO(out, rc = rc2);
1526 #ifdef CONFIG_FS_POSIX_ACL
1527 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1528 buf->lb_buf = ma->ma_acl;
1529 buf->lb_len = ma->ma_acl_size;
1530 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1532 ma->ma_acl_size = rc2;
1533 ma->ma_valid |= MA_ACL_DEF;
1534 } else if (rc2 == -ENODATA) {
1536 ma->ma_acl_size = 0;
1538 GOTO(out, rc = rc2);
1544 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1545 rc, ma->ma_valid, ma->ma_lmm);
1550 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1551 struct md_object *parent, struct lu_fid *fid,
1552 struct lu_name *lname, struct md_op_spec *spec,
1555 struct lu_object *ec_child, *child;
1556 struct lu_device *ld = ed->ed_next;
1557 struct echo_thread_info *info = echo_env_info(env);
1558 struct lu_fid *fid2 = &info->eti_fid2;
1559 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1562 rc = mdo_lookup(env, parent, lname, fid2, spec);
1565 else if (rc != -ENOENT)
1568 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1570 if (IS_ERR(ec_child)) {
1571 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1573 return PTR_ERR(ec_child);
1576 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1577 if (child == NULL) {
1578 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1579 GOTO(out_put, rc = -EINVAL);
1582 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1583 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1586 * Do not perform lookup sanity check. We know that name does not exist.
1588 spec->sp_cr_lookup = 0;
1589 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1591 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1594 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1595 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1597 lu_object_put(env, ec_child);
1601 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1604 struct echo_thread_info *info = echo_env_info(env);
1606 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1607 ma->ma_lmm = (void *)&info->eti_lmm;
1608 ma->ma_lmm_size = sizeof(info->eti_lmm);
1610 LASSERT(info->eti_big_lmmsize);
1611 ma->ma_lmm = info->eti_big_lmm;
1612 ma->ma_lmm_size = info->eti_big_lmmsize;
1618 static int echo_create_md_object(const struct lu_env *env,
1619 struct echo_device *ed,
1620 struct lu_object *ec_parent,
1622 char *name, int namelen,
1623 __u64 id, __u32 mode, int count,
1624 int stripe_count, int stripe_offset)
1626 struct lu_object *parent;
1627 struct echo_thread_info *info = echo_env_info(env);
1628 struct lu_name *lname = &info->eti_lname;
1629 struct md_op_spec *spec = &info->eti_spec;
1630 struct md_attr *ma = &info->eti_ma;
1631 struct lu_device *ld = ed->ed_next;
1635 if (ec_parent == NULL)
1637 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1641 memset(ma, 0, sizeof(*ma));
1642 memset(spec, 0, sizeof(*spec));
1643 if (stripe_count != 0) {
1644 spec->sp_cr_flags |= FMODE_WRITE;
1645 echo_set_lmm_size(env, ld, ma);
1646 if (stripe_count != -1) {
1647 struct lov_user_md_v3 *lum = &info->eti_lum;
1649 lum->lmm_magic = LOV_USER_MAGIC_V3;
1650 lum->lmm_stripe_count = stripe_count;
1651 lum->lmm_stripe_offset = stripe_offset;
1652 lum->lmm_pattern = 0;
1653 spec->u.sp_ea.eadata = lum;
1654 spec->u.sp_ea.eadatalen = sizeof(*lum);
1655 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1659 ma->ma_attr.la_mode = mode;
1660 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1661 ma->ma_attr.la_ctime = cfs_time_current_64();
1664 lname->ln_name = name;
1665 lname->ln_namelen = namelen;
1666 /* If name is specified, only create one object by name */
1667 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1672 /* Create multiple object sequenced by id */
1673 for (i = 0; i < count; i++) {
1674 char *tmp_name = info->eti_name;
1676 echo_md_build_name(lname, tmp_name, id);
1678 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1681 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1692 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1693 struct echo_device *ed,
1694 struct md_object *parent,
1695 struct lu_name *lname)
1697 struct echo_thread_info *info = echo_env_info(env);
1698 struct lu_fid *fid = &info->eti_fid;
1699 struct lu_object *child;
1702 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1704 rc = mdo_lookup(env, parent, lname, fid, NULL);
1706 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1710 /* In the function below, .hs_keycmp resolves to
1711 * lu_obj_hop_keycmp() */
1712 /* coverity[overrun-buffer-val] */
1713 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1718 static int echo_setattr_object(const struct lu_env *env,
1719 struct echo_device *ed,
1720 struct lu_object *ec_parent,
1721 __u64 id, int count)
1723 struct lu_object *parent;
1724 struct echo_thread_info *info = echo_env_info(env);
1725 struct lu_name *lname = &info->eti_lname;
1726 char *name = info->eti_name;
1727 struct lu_device *ld = ed->ed_next;
1728 struct lu_buf *buf = &info->eti_buf;
1732 if (ec_parent == NULL)
1734 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1738 for (i = 0; i < count; i++) {
1739 struct lu_object *ec_child, *child;
1741 echo_md_build_name(lname, name, id);
1743 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1744 if (IS_ERR(ec_child)) {
1745 CERROR("Can't find child %s: rc = %ld\n",
1746 lname->ln_name, PTR_ERR(ec_child));
1747 return PTR_ERR(ec_child);
1750 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1751 if (child == NULL) {
1752 CERROR("Can not locate the child %s\n", lname->ln_name);
1753 lu_object_put(env, ec_child);
1758 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1759 PFID(lu_object_fid(child)));
1761 buf->lb_buf = info->eti_xattr_buf;
1762 buf->lb_len = sizeof(info->eti_xattr_buf);
1764 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1765 rc = mo_xattr_set(env, lu2md(child), buf, name,
1768 CERROR("Can not setattr child "DFID": rc = %d\n",
1769 PFID(lu_object_fid(child)), rc);
1770 lu_object_put(env, ec_child);
1773 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1774 PFID(lu_object_fid(child)));
1776 lu_object_put(env, ec_child);
1781 static int echo_getattr_object(const struct lu_env *env,
1782 struct echo_device *ed,
1783 struct lu_object *ec_parent,
1784 __u64 id, int count)
1786 struct lu_object *parent;
1787 struct echo_thread_info *info = echo_env_info(env);
1788 struct lu_name *lname = &info->eti_lname;
1789 char *name = info->eti_name;
1790 struct md_attr *ma = &info->eti_ma;
1791 struct lu_device *ld = ed->ed_next;
1795 if (ec_parent == NULL)
1797 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1801 memset(ma, 0, sizeof(*ma));
1802 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1803 ma->ma_acl = info->eti_xattr_buf;
1804 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1806 for (i = 0; i < count; i++) {
1807 struct lu_object *ec_child, *child;
1810 echo_md_build_name(lname, name, id);
1811 echo_set_lmm_size(env, ld, ma);
1813 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1814 if (IS_ERR(ec_child)) {
1815 CERROR("Can't find child %s: rc = %ld\n",
1816 lname->ln_name, PTR_ERR(ec_child));
1817 return PTR_ERR(ec_child);
1820 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1821 if (child == NULL) {
1822 CERROR("Can not locate the child %s\n", lname->ln_name);
1823 lu_object_put(env, ec_child);
1827 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1828 PFID(lu_object_fid(child)));
1829 rc = echo_attr_get_complex(env, lu2md(child), ma);
1831 CERROR("Can not getattr child "DFID": rc = %d\n",
1832 PFID(lu_object_fid(child)), rc);
1833 lu_object_put(env, ec_child);
1836 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1837 PFID(lu_object_fid(child)));
1839 lu_object_put(env, ec_child);
1845 static int echo_lookup_object(const struct lu_env *env,
1846 struct echo_device *ed,
1847 struct lu_object *ec_parent,
1848 __u64 id, int count)
1850 struct lu_object *parent;
1851 struct echo_thread_info *info = echo_env_info(env);
1852 struct lu_name *lname = &info->eti_lname;
1853 char *name = info->eti_name;
1854 struct lu_fid *fid = &info->eti_fid;
1855 struct lu_device *ld = ed->ed_next;
1859 if (ec_parent == NULL)
1861 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1865 /*prepare the requests*/
1866 for (i = 0; i < count; i++) {
1867 echo_md_build_name(lname, name, id);
1869 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1870 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1872 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1874 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1877 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1878 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1885 static int echo_md_destroy_internal(const struct lu_env *env,
1886 struct echo_device *ed,
1887 struct md_object *parent,
1888 struct lu_name *lname,
1891 struct lu_device *ld = ed->ed_next;
1892 struct lu_object *ec_child;
1893 struct lu_object *child;
1896 ec_child = echo_md_lookup(env, ed, parent, lname);
1897 if (IS_ERR(ec_child)) {
1898 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1900 return PTR_ERR(ec_child);
1903 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1904 if (child == NULL) {
1905 CERROR("Can not locate the child %s\n", lname->ln_name);
1906 GOTO(out_put, rc = -EINVAL);
1909 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1910 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1912 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1914 CERROR("Can not unlink child %s: rc = %d\n",
1915 lname->ln_name, rc);
1918 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1919 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1921 lu_object_put(env, ec_child);
1925 static int echo_destroy_object(const struct lu_env *env,
1926 struct echo_device *ed,
1927 struct lu_object *ec_parent,
1928 char *name, int namelen,
1929 __u64 id, __u32 mode,
1932 struct echo_thread_info *info = echo_env_info(env);
1933 struct lu_name *lname = &info->eti_lname;
1934 struct md_attr *ma = &info->eti_ma;
1935 struct lu_device *ld = ed->ed_next;
1936 struct lu_object *parent;
1940 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1944 memset(ma, 0, sizeof(*ma));
1945 ma->ma_attr.la_mode = mode;
1946 ma->ma_attr.la_valid = LA_CTIME;
1947 ma->ma_attr.la_ctime = cfs_time_current_64();
1948 ma->ma_need = MA_INODE;
1952 lname->ln_name = name;
1953 lname->ln_namelen = namelen;
1954 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1959 /*prepare the requests*/
1960 for (i = 0; i < count; i++) {
1961 char *tmp_name = info->eti_name;
1964 echo_md_build_name(lname, tmp_name, id);
1966 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1969 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
1978 static struct lu_object *echo_resolve_path(const struct lu_env *env,
1979 struct echo_device *ed, char *path,
1982 struct lu_device *ld = ed->ed_next;
1983 struct md_device *md = lu2md_dev(ld);
1984 struct echo_thread_info *info = echo_env_info(env);
1985 struct lu_fid *fid = &info->eti_fid;
1986 struct lu_name *lname = &info->eti_lname;
1987 struct lu_object *parent = NULL;
1988 struct lu_object *child = NULL;
1991 /*Only support MDD layer right now*/
1992 rc = md->md_ops->mdo_root_get(env, md, fid);
1994 CERROR("get root error: rc = %d\n", rc);
1998 /* In the function below, .hs_keycmp resolves to
1999 * lu_obj_hop_keycmp() */
2000 /* coverity[overrun-buffer-val] */
2001 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2002 if (IS_ERR(parent)) {
2003 CERROR("Can not find the parent "DFID": rc = %ld\n",
2004 PFID(fid), PTR_ERR(parent));
2009 struct lu_object *ld_parent;
2012 e = strsep(&path, "/");
2017 if (!path || path[0] == '\0')
2023 lname->ln_namelen = strlen(e);
2025 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2026 if (ld_parent == NULL) {
2027 lu_object_put(env, parent);
2032 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2033 lu_object_put(env, parent);
2034 if (IS_ERR(child)) {
2035 rc = (int)PTR_ERR(child);
2036 CERROR("lookup %s under parent "DFID": rc = %d\n",
2037 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2049 static void echo_ucred_init(struct lu_env *env)
2051 struct lu_ucred *ucred = lu_ucred(env);
2053 ucred->uc_valid = UCRED_INVALID;
2055 ucred->uc_suppgids[0] = -1;
2056 ucred->uc_suppgids[1] = -1;
2058 ucred->uc_uid = ucred->uc_o_uid =
2059 from_kuid(&init_user_ns, current_uid());
2060 ucred->uc_gid = ucred->uc_o_gid =
2061 from_kgid(&init_user_ns, current_gid());
2062 ucred->uc_fsuid = ucred->uc_o_fsuid =
2063 from_kuid(&init_user_ns, current_fsuid());
2064 ucred->uc_fsgid = ucred->uc_o_fsgid =
2065 from_kgid(&init_user_ns, current_fsgid());
2066 ucred->uc_cap = cfs_curproc_cap_pack();
2068 /* remove fs privilege for non-root user. */
2069 if (ucred->uc_fsuid)
2070 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2071 ucred->uc_valid = UCRED_NEW;
2074 static void echo_ucred_fini(struct lu_env *env)
2076 struct lu_ucred *ucred = lu_ucred(env);
2077 ucred->uc_valid = UCRED_INIT;
2080 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2081 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2082 static int echo_md_handler(struct echo_device *ed, int command,
2083 char *path, int path_len, __u64 id, int count,
2084 struct obd_ioctl_data *data)
2086 struct echo_thread_info *info;
2087 struct lu_device *ld = ed->ed_next;
2090 struct lu_object *parent;
2092 int namelen = data->ioc_plen2;
2096 CERROR("MD echo client is not being initialized properly\n");
2100 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2101 CERROR("Only support MDD layer right now!\n");
2105 env = cl_env_get(&refcheck);
2107 return PTR_ERR(env);
2109 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2113 /* init big_lmm buffer */
2114 info = echo_env_info(env);
2115 LASSERT(info->eti_big_lmm == NULL);
2116 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2117 if (info->eti_big_lmm == NULL)
2118 GOTO(out_env, rc = -ENOMEM);
2119 info->eti_big_lmmsize = MIN_MD_SIZE;
2121 parent = echo_resolve_path(env, ed, path, path_len);
2122 if (IS_ERR(parent)) {
2123 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2125 GOTO(out_free, rc = PTR_ERR(parent));
2129 OBD_ALLOC(name, namelen + 1);
2131 GOTO(out_put, rc = -ENOMEM);
2132 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2133 GOTO(out_name, rc = -EFAULT);
2136 echo_ucred_init(env);
2139 case ECHO_MD_CREATE:
2140 case ECHO_MD_MKDIR: {
2141 struct echo_thread_info *info = echo_env_info(env);
2142 __u32 mode = data->ioc_obdo2.o_mode;
2143 struct lu_fid *fid = &info->eti_fid;
2144 int stripe_count = (int)data->ioc_obdo2.o_misc;
2145 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2147 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2151 /* In the function below, .hs_keycmp resolves to
2152 * lu_obj_hop_keycmp() */
2153 /* coverity[overrun-buffer-val] */
2154 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2155 id, mode, count, stripe_count,
2159 case ECHO_MD_DESTROY:
2160 case ECHO_MD_RMDIR: {
2161 __u32 mode = data->ioc_obdo2.o_mode;
2163 rc = echo_destroy_object(env, ed, parent, name, namelen,
2167 case ECHO_MD_LOOKUP:
2168 rc = echo_lookup_object(env, ed, parent, id, count);
2170 case ECHO_MD_GETATTR:
2171 rc = echo_getattr_object(env, ed, parent, id, count);
2173 case ECHO_MD_SETATTR:
2174 rc = echo_setattr_object(env, ed, parent, id, count);
2177 CERROR("unknown command %d\n", command);
2181 echo_ucred_fini(env);
2185 OBD_FREE(name, namelen + 1);
2187 lu_object_put(env, parent);
2189 LASSERT(info->eti_big_lmm);
2190 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2191 info->eti_big_lmm = NULL;
2192 info->eti_big_lmmsize = 0;
2194 cl_env_put(env, &refcheck);
2198 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2199 int on_target, struct obdo *oa, void *ulsm,
2200 int ulsm_nob, struct obd_trans_info *oti)
2202 struct echo_object *eco;
2203 struct echo_client_obd *ec = ed->ed_ec;
2204 struct lov_stripe_md *lsm = NULL;
2208 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2209 (on_target || /* set_stripe */
2210 ec->ec_nstripes != 0)) { /* LOV */
2211 CERROR ("No valid oid\n");
2215 rc = echo_alloc_memmd(ed, &lsm);
2217 CERROR("Cannot allocate md: rc = %d\n", rc);
2224 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2228 if (lsm->lsm_stripe_count == 0)
2229 lsm->lsm_stripe_count = ec->ec_nstripes;
2231 if (lsm->lsm_stripe_size == 0)
2232 lsm->lsm_stripe_size = PAGE_CACHE_SIZE;
2236 /* setup stripes: indices + default ids if required */
2237 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2238 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2239 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2241 lsm->lsm_oinfo[i]->loi_ost_idx =
2242 (idx + i) % ec->ec_nstripes;
2246 /* setup object ID here for !on_target and LOV hint */
2247 if (oa->o_valid & OBD_MD_FLID) {
2248 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2249 lsm->lsm_oi = oa->o_oi;
2252 if (ostid_id(&lsm->lsm_oi) == 0)
2253 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2257 /* Only echo objects are allowed to be created */
2258 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2259 (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2260 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2262 CERROR("Cannot create objects: rc = %d\n", rc);
2268 /* See what object ID we were given */
2269 oa->o_oi = lsm->lsm_oi;
2270 oa->o_valid |= OBD_MD_FLID;
2272 eco = cl_echo_object_find(ed, &lsm);
2274 GOTO(failed, rc = PTR_ERR(eco));
2275 cl_echo_object_put(eco);
2277 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2281 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2283 echo_free_memmd(ed, &lsm);
2285 CERROR("create object failed with: rc = %d\n", rc);
2289 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2292 struct lov_stripe_md *lsm = NULL;
2293 struct echo_object *eco;
2296 if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2297 /* disallow use of object id 0 */
2298 CERROR ("No valid oid\n");
2302 rc = echo_alloc_memmd(ed, &lsm);
2306 lsm->lsm_oi = oa->o_oi;
2307 if (!(oa->o_valid & OBD_MD_FLGROUP))
2308 ostid_set_seq_echo(&lsm->lsm_oi);
2311 eco = cl_echo_object_find(ed, &lsm);
2317 echo_free_memmd(ed, &lsm);
2321 static void echo_put_object(struct echo_object *eco)
2323 if (cl_echo_object_put(eco))
2324 CERROR("echo client: drop an object failed");
2328 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2330 unsigned long stripe_count;
2331 unsigned long stripe_size;
2332 unsigned long width;
2333 unsigned long woffset;
2337 if (lsm->lsm_stripe_count <= 1)
2341 stripe_size = lsm->lsm_stripe_size;
2342 stripe_count = lsm->lsm_stripe_count;
2344 /* width = # bytes in all stripes */
2345 width = stripe_size * stripe_count;
2347 /* woffset = offset within a width; offset = whole number of widths */
2348 woffset = do_div (offset, width);
2350 stripe_index = woffset / stripe_size;
2352 *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2353 *offp = offset * stripe_size + woffset % stripe_size;
2357 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2358 struct page *page, int rw, obd_id id,
2359 obd_off offset, obd_off count)
2366 /* no partial pages on the client */
2367 LASSERT(count == PAGE_CACHE_SIZE);
2371 for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2372 if (rw == OBD_BRW_WRITE) {
2373 stripe_off = offset + delta;
2375 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2377 stripe_off = 0xdeadbeef00c0ffeeULL;
2378 stripe_id = 0xdeadbeef00c0ffeeULL;
2380 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2381 stripe_off, stripe_id);
2387 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2388 struct page *page, obd_id id,
2389 obd_off offset, obd_off count)
2398 /* no partial pages on the client */
2399 LASSERT(count == PAGE_CACHE_SIZE);
2403 for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2404 stripe_off = offset + delta;
2406 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2408 rc2 = block_debug_check("test_brw",
2409 addr + delta, OBD_ECHO_BLOCK_SIZE,
2410 stripe_off, stripe_id);
2412 CERROR ("Error in echo object "LPX64"\n", id);
2421 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2422 struct echo_object *eco, obd_off offset,
2423 obd_size count, int async,
2424 struct obd_trans_info *oti)
2426 struct lov_stripe_md *lsm = eco->eo_lsm;
2428 struct brw_page *pga;
2429 struct brw_page *pgp;
2430 struct page **pages;
2438 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2439 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2440 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2442 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2444 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2445 LASSERT(lsm != NULL);
2446 LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2449 (count & (~CFS_PAGE_MASK)) != 0)
2452 /* XXX think again with misaligned I/O */
2453 npages = count >> PAGE_CACHE_SHIFT;
2455 if (rw == OBD_BRW_WRITE)
2456 brw_flags = OBD_BRW_ASYNC;
2458 OBD_ALLOC(pga, npages * sizeof(*pga));
2462 OBD_ALLOC(pages, npages * sizeof(*pages));
2463 if (pages == NULL) {
2464 OBD_FREE(pga, npages * sizeof(*pga));
2468 for (i = 0, pgp = pga, off = offset;
2470 i++, pgp++, off += PAGE_CACHE_SIZE) {
2472 LASSERT (pgp->pg == NULL); /* for cleanup */
2475 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2476 if (pgp->pg == NULL)
2480 pgp->count = PAGE_CACHE_SIZE;
2482 pgp->flag = brw_flags;
2485 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2486 ostid_id(&oa->o_oi), off,
2490 /* brw mode can only be used at client */
2491 LASSERT(ed->ed_next != NULL);
2492 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2495 if (rc != 0 || rw != OBD_BRW_READ)
2498 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2499 if (pgp->pg == NULL)
2504 vrc = echo_client_page_debug_check(lsm, pgp->pg,
2505 ostid_id(&oa->o_oi),
2506 pgp->off, pgp->count);
2507 if (vrc != 0 && rc == 0)
2510 OBD_PAGE_FREE(pgp->pg);
2512 OBD_FREE(pga, npages * sizeof(*pga));
2513 OBD_FREE(pages, npages * sizeof(*pages));
2517 static int echo_client_prep_commit(const struct lu_env *env,
2518 struct obd_export *exp, int rw,
2519 struct obdo *oa, struct echo_object *eco,
2520 obd_off offset, obd_size count,
2521 obd_size batch, struct obd_trans_info *oti,
2524 struct lov_stripe_md *lsm = eco->eo_lsm;
2525 struct obd_ioobj ioo;
2526 struct niobuf_local *lnb;
2527 struct niobuf_remote *rnb;
2529 obd_size npages, tot_pages;
2530 int i, ret = 0, brw_flags = 0;
2532 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2533 (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2536 npages = batch >> PAGE_CACHE_SHIFT;
2537 tot_pages = count >> PAGE_CACHE_SHIFT;
2539 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2540 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2542 if (lnb == NULL || rnb == NULL)
2543 GOTO(out, ret = -ENOMEM);
2545 if (rw == OBD_BRW_WRITE && async)
2546 brw_flags |= OBD_BRW_ASYNC;
2548 obdo_to_ioobj(oa, &ioo);
2552 for(; tot_pages; tot_pages -= npages) {
2555 if (tot_pages < npages)
2558 for (i = 0; i < npages; i++, off += PAGE_CACHE_SIZE) {
2559 rnb[i].offset = off;
2560 rnb[i].len = PAGE_CACHE_SIZE;
2561 rnb[i].flags = brw_flags;
2564 ioo.ioo_bufcnt = npages;
2565 oti->oti_transno = 0;
2568 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2572 LASSERT(lpages == npages);
2574 for (i = 0; i < lpages; i++) {
2575 struct page *page = lnb[i].page;
2577 /* read past eof? */
2578 if (page == NULL && lnb[i].rc == 0)
2582 lnb[i].flags |= OBD_BRW_ASYNC;
2584 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2585 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2586 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2589 if (rw == OBD_BRW_WRITE)
2590 echo_client_page_debug_setup(lsm, page, rw,
2591 ostid_id(&oa->o_oi),
2595 echo_client_page_debug_check(lsm, page,
2596 ostid_id(&oa->o_oi),
2601 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2602 rnb, npages, lnb, oti, ret);
2606 /* Reset oti otherwise it would confuse ldiskfs. */
2607 memset(oti, 0, sizeof(*oti));
2609 /* Reuse env context. */
2610 lu_context_exit((struct lu_context *)&env->le_ctx);
2611 lu_context_enter((struct lu_context *)&env->le_ctx);
2616 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2618 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2622 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2623 struct obd_export *exp,
2624 struct obd_ioctl_data *data,
2625 struct obd_trans_info *dummy_oti)
2627 struct obd_device *obd = class_exp2obd(exp);
2628 struct echo_device *ed = obd2echo_dev(obd);
2629 struct echo_client_obd *ec = ed->ed_ec;
2630 struct obdo *oa = &data->ioc_obdo1;
2631 struct echo_object *eco;
2636 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2638 rc = echo_get_object(&eco, ed, oa);
2642 oa->o_valid &= ~OBD_MD_FLHANDLE;
2644 /* OFD/obdfilter works only via prep/commit */
2645 test_mode = (long)data->ioc_pbuf1;
2649 if (ed->ed_next == NULL && test_mode != 3) {
2651 data->ioc_plen1 = data->ioc_count;
2654 /* Truncate batch size to maximum */
2655 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2656 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2658 switch (test_mode) {
2662 rc = echo_client_kbrw(ed, rw, oa,
2663 eco, data->ioc_offset,
2664 data->ioc_count, async, dummy_oti);
2667 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2668 eco, data->ioc_offset,
2669 data->ioc_count, data->ioc_plen1,
2675 echo_put_object(eco);
2680 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2681 int mode, obd_off offset, obd_size nob)
2683 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2684 struct lustre_handle *ulh = &oa->o_handle;
2685 struct echo_object *eco;
2689 if (ed->ed_next == NULL)
2692 if (!(mode == LCK_PR || mode == LCK_PW))
2695 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2696 (nob & (~CFS_PAGE_MASK)) != 0)
2699 rc = echo_get_object (&eco, ed, oa);
2703 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2704 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2706 oa->o_valid |= OBD_MD_FLHANDLE;
2707 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2709 echo_put_object(eco);
2714 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2716 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2717 __u64 cookie = oa->o_handle.cookie;
2719 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2722 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2723 return cl_echo_cancel(ed, cookie);
2727 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2728 void *karg, void *uarg)
2730 struct obd_device *obd = exp->exp_obd;
2731 struct echo_device *ed = obd2echo_dev(obd);
2732 struct echo_client_obd *ec = ed->ed_ec;
2733 struct echo_object *eco;
2734 struct obd_ioctl_data *data = karg;
2735 struct obd_trans_info dummy_oti;
2737 struct oti_req_ack_lock *ack_lock;
2740 int rw = OBD_BRW_READ;
2744 memset(&dummy_oti, 0, sizeof(dummy_oti));
2746 oa = &data->ioc_obdo1;
2747 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2748 oa->o_valid |= OBD_MD_FLGROUP;
2749 ostid_set_seq_echo(&oa->o_oi);
2752 /* This FID is unpacked just for validation at this point */
2753 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2761 rc = lu_env_init(env, LCT_DT_THREAD);
2763 GOTO(out, rc = -ENOMEM);
2766 case OBD_IOC_CREATE: /* may create echo object */
2767 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2768 GOTO (out, rc = -EPERM);
2770 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2771 data->ioc_plen1, &dummy_oti);
2774 case OBD_IOC_ECHO_MD: {
2781 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2782 GOTO(out, rc = -EPERM);
2784 count = data->ioc_count;
2785 cmd = data->ioc_command;
2787 id = ostid_id(&data->ioc_obdo2.o_oi);
2789 dirlen = data->ioc_plen1;
2790 OBD_ALLOC(dir, dirlen + 1);
2792 GOTO(out, rc = -ENOMEM);
2794 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2795 OBD_FREE(dir, data->ioc_plen1 + 1);
2796 GOTO(out, rc = -EFAULT);
2799 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2800 OBD_FREE(dir, dirlen + 1);
2803 case OBD_IOC_ECHO_ALLOC_SEQ: {
2804 struct lu_env *cl_env;
2809 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2810 GOTO(out, rc = -EPERM);
2812 cl_env = cl_env_get(&refcheck);
2814 GOTO(out, rc = PTR_ERR(cl_env));
2816 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2819 cl_env_put(cl_env, &refcheck);
2823 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2824 cl_env_put(cl_env, &refcheck);
2826 CERROR("%s: Can not alloc seq: rc = %d\n",
2831 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2834 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2835 if (copy_to_user(data->ioc_pbuf2, &max_count,
2840 case OBD_IOC_DESTROY:
2841 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2842 GOTO (out, rc = -EPERM);
2844 rc = echo_get_object(&eco, ed, oa);
2846 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2847 &dummy_oti, NULL, NULL);
2849 eco->eo_deleted = 1;
2850 echo_put_object(eco);
2854 case OBD_IOC_GETATTR:
2855 rc = echo_get_object(&eco, ed, oa);
2857 struct obd_info oinfo = { { { 0 } } };
2858 oinfo.oi_md = eco->eo_lsm;
2860 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2861 echo_put_object(eco);
2865 case OBD_IOC_SETATTR:
2866 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2867 GOTO (out, rc = -EPERM);
2869 rc = echo_get_object(&eco, ed, oa);
2871 struct obd_info oinfo = { { { 0 } } };
2873 oinfo.oi_md = eco->eo_lsm;
2875 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2876 echo_put_object(eco);
2880 case OBD_IOC_BRW_WRITE:
2881 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2882 GOTO (out, rc = -EPERM);
2886 case OBD_IOC_BRW_READ:
2887 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2890 case ECHO_IOC_GET_STRIPE:
2891 rc = echo_get_object(&eco, ed, oa);
2893 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2895 echo_put_object(eco);
2899 case ECHO_IOC_SET_STRIPE:
2900 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2901 GOTO (out, rc = -EPERM);
2903 if (data->ioc_pbuf1 == NULL) { /* unset */
2904 rc = echo_get_object(&eco, ed, oa);
2906 eco->eo_deleted = 1;
2907 echo_put_object(eco);
2910 rc = echo_create_object(env, ed, 0, oa,
2912 data->ioc_plen1, &dummy_oti);
2916 case ECHO_IOC_ENQUEUE:
2917 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2918 GOTO (out, rc = -EPERM);
2920 rc = echo_client_enqueue(exp, oa,
2921 data->ioc_conn1, /* lock mode */
2923 data->ioc_count);/*extent*/
2926 case ECHO_IOC_CANCEL:
2927 rc = echo_client_cancel(exp, oa);
2931 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2932 GOTO (out, rc = -ENOTTY);
2939 /* XXX this should be in a helper also called by target_send_reply */
2940 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2942 if (!ack_lock->mode)
2944 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
2950 static int echo_client_setup(const struct lu_env *env,
2951 struct obd_device *obddev, struct lustre_cfg *lcfg)
2953 struct echo_client_obd *ec = &obddev->u.echo_client;
2954 struct obd_device *tgt;
2955 struct obd_uuid echo_uuid = { "ECHO_UUID" };
2956 struct obd_connect_data *ocd = NULL;
2959 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2960 CERROR("requires a TARGET OBD name\n");
2964 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2965 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2966 CERROR("device not attached or not set up (%s)\n",
2967 lustre_cfg_string(lcfg, 1));
2971 spin_lock_init(&ec->ec_lock);
2972 INIT_LIST_HEAD (&ec->ec_objects);
2973 INIT_LIST_HEAD (&ec->ec_locks);
2975 ec->ec_nstripes = 0;
2977 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2978 lu_context_tags_update(ECHO_MD_CTX_TAG);
2979 lu_session_tags_update(ECHO_MD_SES_TAG);
2983 OBD_ALLOC(ocd, sizeof(*ocd));
2985 CERROR("Can't alloc ocd connecting to %s\n",
2986 lustre_cfg_string(lcfg, 1));
2990 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2991 OBD_CONNECT_BRW_SIZE |
2992 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2993 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2995 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2996 ocd->ocd_version = LUSTRE_VERSION_CODE;
2997 ocd->ocd_group = FID_SEQ_ECHO;
2999 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3001 /* Turn off pinger because it connects to tgt obd directly. */
3002 spin_lock(&tgt->obd_dev_lock);
3003 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3004 spin_unlock(&tgt->obd_dev_lock);
3007 OBD_FREE(ocd, sizeof(*ocd));
3010 CERROR("fail to connect to device %s\n",
3011 lustre_cfg_string(lcfg, 1));
3018 static int echo_client_cleanup(struct obd_device *obddev)
3020 struct echo_device *ed = obd2echo_dev(obddev);
3021 struct echo_client_obd *ec = &obddev->u.echo_client;
3024 /*Do nothing for Metadata echo client*/
3028 if (ed->ed_next_ismd) {
3029 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3030 lu_session_tags_clear(ECHO_MD_SES_TAG);
3034 if (!list_empty(&obddev->obd_exports)) {
3035 CERROR("still has clients!\n");
3039 LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
3040 rc = obd_disconnect(ec->ec_exp);
3042 CERROR("fail to disconnect device: %d\n", rc);
3047 static int echo_client_connect(const struct lu_env *env,
3048 struct obd_export **exp,
3049 struct obd_device *src, struct obd_uuid *cluuid,
3050 struct obd_connect_data *data, void *localdata)
3053 struct lustre_handle conn = { 0 };
3055 rc = class_connect(&conn, src, cluuid);
3057 *exp = class_conn2export(&conn);
3063 static int echo_client_disconnect(struct obd_export *exp)
3066 struct obd_device *obd;
3067 struct echo_client_obd *ec;
3068 struct ec_lock *ecl;
3073 GOTO(out, rc = -EINVAL);
3077 ec = &obd->u.echo_client;
3079 /* no more contention on export's lock list */
3080 while (!list_empty (&exp->exp_ec_data.eced_locks)) {
3081 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
3082 struct ec_lock, ecl_exp_chain);
3083 list_del (&ecl->ecl_exp_chain);
3085 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3086 ecl->ecl_mode, &ecl->ecl_lock_handle);
3088 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3089 "(%d)\n", ecl->ecl_object->eco_id, rc);
3091 echo_put_object (ecl->ecl_object);
3092 OBD_FREE (ecl, sizeof (*ecl));
3096 rc = class_disconnect(exp);
3102 static struct obd_ops echo_client_obd_ops = {
3103 .o_owner = THIS_MODULE,
3106 .o_setup = echo_client_setup,
3107 .o_cleanup = echo_client_cleanup,
3110 .o_iocontrol = echo_client_iocontrol,
3111 .o_connect = echo_client_connect,
3112 .o_disconnect = echo_client_disconnect
3115 int echo_client_init(void)
3117 struct lprocfs_static_vars lvars = { 0 };
3120 lprocfs_echo_init_vars(&lvars);
3122 rc = lu_kmem_init(echo_caches);
3124 rc = class_register_type(&echo_client_obd_ops, NULL,
3126 LUSTRE_ECHO_CLIENT_NAME,
3129 lu_kmem_fini(echo_caches);
3134 void echo_client_exit(void)
3136 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3137 lu_kmem_fini(echo_caches);
3140 static int __init obdecho_init(void)
3142 struct lprocfs_static_vars lvars;
3145 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3147 LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3149 lprocfs_echo_init_vars(&lvars);
3152 rc = echo_client_init();
3157 static void /*__exit*/ obdecho_exit(void)
3163 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3164 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3165 MODULE_LICENSE("GPL");
3166 MODULE_VERSION(LUSTRE_VERSION_STRING);
3168 module_init(obdecho_init);
3169 module_exit(obdecho_exit);
3171 /** @} echo_client */