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
39 #include <libcfs/libcfs.h>
41 #include <liblustre.h>
45 #include <obd_support.h>
46 #include <obd_class.h>
47 #include <lustre_debug.h>
48 #include <lprocfs_status.h>
49 #include <cl_object.h>
50 #include <lustre_fid.h>
51 #include <lustre_acl.h>
52 #include <lustre_net.h>
55 #include "echo_internal.h"
57 /** \defgroup echo_client Echo Client
62 struct cl_device ed_cl;
63 struct echo_client_obd *ed_ec;
65 struct cl_site ed_site_myself;
66 struct cl_site *ed_site;
67 struct lu_device *ed_next;
70 struct lu_client_seq *ed_cl_seq;
74 struct cl_object eo_cl;
75 struct cl_object_header eo_hdr;
77 struct echo_device *eo_dev;
78 cfs_list_t eo_obj_chain;
79 struct lov_stripe_md *eo_lsm;
80 cfs_atomic_t eo_npages;
84 struct echo_object_conf {
85 struct cl_object_conf eoc_cl;
86 struct lov_stripe_md **eoc_md;
90 struct cl_page_slice ep_cl;
92 cfs_page_t *ep_vmpage;
96 struct cl_lock_slice el_cl;
98 struct echo_object *el_object;
100 cfs_atomic_t el_refcount;
104 struct cl_io_slice ei_cl;
109 struct cl_req_slice er_cl;
113 static int echo_client_setup(const struct lu_env *env,
114 struct obd_device *obddev,
115 struct lustre_cfg *lcfg);
116 static int echo_client_cleanup(struct obd_device *obddev);
119 /** \defgroup echo_helpers Helper functions
122 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
124 return container_of0(dev, struct echo_device, ed_cl);
127 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
132 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
134 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
137 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
142 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
144 return container_of(o, struct echo_object, eo_cl);
147 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
149 return container_of(s, struct echo_page, ep_cl);
152 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
154 return container_of(s, struct echo_lock, el_cl);
157 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
159 return ecl->el_cl.cls_lock;
162 static struct lu_context_key echo_thread_key;
163 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
165 struct echo_thread_info *info;
166 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
167 LASSERT(info != NULL);
172 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
174 return container_of(c, struct echo_object_conf, eoc_cl);
177 static inline void lsm2fid(struct lov_stripe_md *lsm, struct lu_fid *fid)
180 fid->f_seq = FID_SEQ_ECHO;
181 /* truncated to 32 bits by assignment */
182 fid->f_oid = lsm->lsm_object_id;
183 fid->f_ver = lsm->lsm_object_id >> 32;
185 /** @} echo_helpers */
187 static struct echo_object *cl_echo_object_find(struct echo_device *d,
188 struct lov_stripe_md **lsm);
189 static int cl_echo_object_put(struct echo_object *eco);
190 static int cl_echo_enqueue (struct echo_object *eco, obd_off start,
191 obd_off end, int mode, __u64 *cookie);
192 static int cl_echo_cancel (struct echo_device *d, __u64 cookie);
193 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
194 cfs_page_t **pages, int npages, int async);
196 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
198 struct echo_thread_info {
199 struct echo_object_conf eti_conf;
200 struct lustre_md eti_md;
202 struct cl_2queue eti_queue;
204 struct cl_lock_descr eti_descr;
205 struct lu_fid eti_fid;
206 struct lu_fid eti_fid2;
207 struct md_op_spec eti_spec;
208 struct lov_mds_md_v3 eti_lmm;
209 struct lov_user_md_v3 eti_lum;
210 struct md_attr eti_ma;
211 struct lu_name eti_lname;
212 /* per-thread values, can be re-used */
216 struct lu_buf eti_buf;
217 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
220 /* No session used right now */
221 struct echo_session_info {
225 static cfs_mem_cache_t *echo_lock_kmem;
226 static cfs_mem_cache_t *echo_object_kmem;
227 static cfs_mem_cache_t *echo_thread_kmem;
228 static cfs_mem_cache_t *echo_session_kmem;
229 //static cfs_mem_cache_t *echo_req_kmem;
231 static struct lu_kmem_descr echo_caches[] = {
233 .ckd_cache = &echo_lock_kmem,
234 .ckd_name = "echo_lock_kmem",
235 .ckd_size = sizeof (struct echo_lock)
238 .ckd_cache = &echo_object_kmem,
239 .ckd_name = "echo_object_kmem",
240 .ckd_size = sizeof (struct echo_object)
243 .ckd_cache = &echo_thread_kmem,
244 .ckd_name = "echo_thread_kmem",
245 .ckd_size = sizeof (struct echo_thread_info)
248 .ckd_cache = &echo_session_kmem,
249 .ckd_name = "echo_session_kmem",
250 .ckd_size = sizeof (struct echo_session_info)
254 .ckd_cache = &echo_req_kmem,
255 .ckd_name = "echo_req_kmem",
256 .ckd_size = sizeof (struct echo_req)
264 /** \defgroup echo_page Page operations
266 * Echo page operations.
270 static cfs_page_t *echo_page_vmpage(const struct lu_env *env,
271 const struct cl_page_slice *slice)
273 return cl2echo_page(slice)->ep_vmpage;
276 static int echo_page_own(const struct lu_env *env,
277 const struct cl_page_slice *slice,
278 struct cl_io *io, int nonblock)
280 struct echo_page *ep = cl2echo_page(slice);
283 mutex_lock(&ep->ep_lock);
284 else if (!mutex_trylock(&ep->ep_lock))
289 static void echo_page_disown(const struct lu_env *env,
290 const struct cl_page_slice *slice,
293 struct echo_page *ep = cl2echo_page(slice);
295 LASSERT(mutex_is_locked(&ep->ep_lock));
296 mutex_unlock(&ep->ep_lock);
299 static void echo_page_discard(const struct lu_env *env,
300 const struct cl_page_slice *slice,
301 struct cl_io *unused)
303 cl_page_delete(env, slice->cpl_page);
306 static int echo_page_is_vmlocked(const struct lu_env *env,
307 const struct cl_page_slice *slice)
309 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
314 static void echo_page_completion(const struct lu_env *env,
315 const struct cl_page_slice *slice,
318 LASSERT(slice->cpl_page->cp_sync_io != NULL);
321 static void echo_page_fini(const struct lu_env *env,
322 struct cl_page_slice *slice)
324 struct echo_page *ep = cl2echo_page(slice);
325 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
326 cfs_page_t *vmpage = ep->ep_vmpage;
329 cfs_atomic_dec(&eco->eo_npages);
330 page_cache_release(vmpage);
334 static int echo_page_prep(const struct lu_env *env,
335 const struct cl_page_slice *slice,
336 struct cl_io *unused)
341 static int echo_page_print(const struct lu_env *env,
342 const struct cl_page_slice *slice,
343 void *cookie, lu_printer_t printer)
345 struct echo_page *ep = cl2echo_page(slice);
347 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
348 ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
352 static const struct cl_page_operations echo_page_ops = {
353 .cpo_own = echo_page_own,
354 .cpo_disown = echo_page_disown,
355 .cpo_discard = echo_page_discard,
356 .cpo_vmpage = echo_page_vmpage,
357 .cpo_fini = echo_page_fini,
358 .cpo_print = echo_page_print,
359 .cpo_is_vmlocked = echo_page_is_vmlocked,
362 .cpo_prep = echo_page_prep,
363 .cpo_completion = echo_page_completion,
366 .cpo_prep = echo_page_prep,
367 .cpo_completion = echo_page_completion,
373 /** \defgroup echo_lock Locking
375 * echo lock operations
379 static void echo_lock_fini(const struct lu_env *env,
380 struct cl_lock_slice *slice)
382 struct echo_lock *ecl = cl2echo_lock(slice);
384 LASSERT(cfs_list_empty(&ecl->el_chain));
385 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
388 static void echo_lock_delete(const struct lu_env *env,
389 const struct cl_lock_slice *slice)
391 struct echo_lock *ecl = cl2echo_lock(slice);
393 LASSERT(cfs_list_empty(&ecl->el_chain));
396 static int echo_lock_fits_into(const struct lu_env *env,
397 const struct cl_lock_slice *slice,
398 const struct cl_lock_descr *need,
399 const struct cl_io *unused)
404 static struct cl_lock_operations echo_lock_ops = {
405 .clo_fini = echo_lock_fini,
406 .clo_delete = echo_lock_delete,
407 .clo_fits_into = echo_lock_fits_into
412 /** \defgroup echo_cl_ops cl_object operations
414 * operations for cl_object
418 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
419 struct cl_page *page, cfs_page_t *vmpage)
421 struct echo_page *ep = cl_object_page_slice(obj, page);
422 struct echo_object *eco = cl2echo_obj(obj);
425 ep->ep_vmpage = vmpage;
426 page_cache_get(vmpage);
427 mutex_init(&ep->ep_lock);
428 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
429 cfs_atomic_inc(&eco->eo_npages);
433 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
439 static int echo_lock_init(const struct lu_env *env,
440 struct cl_object *obj, struct cl_lock *lock,
441 const struct cl_io *unused)
443 struct echo_lock *el;
446 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, CFS_ALLOC_IO);
448 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
449 el->el_object = cl2echo_obj(obj);
450 CFS_INIT_LIST_HEAD(&el->el_chain);
451 cfs_atomic_set(&el->el_refcount, 0);
453 RETURN(el == NULL ? -ENOMEM : 0);
456 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
457 const struct cl_object_conf *conf)
462 static const struct cl_object_operations echo_cl_obj_ops = {
463 .coo_page_init = echo_page_init,
464 .coo_lock_init = echo_lock_init,
465 .coo_io_init = echo_io_init,
466 .coo_conf_set = echo_conf_set
468 /** @} echo_cl_ops */
470 /** \defgroup echo_lu_ops lu_object operations
472 * operations for echo lu object.
476 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
477 const struct lu_object_conf *conf)
479 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
480 struct echo_client_obd *ec = ed->ed_ec;
481 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
485 struct lu_object *below;
486 struct lu_device *under;
489 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
493 lu_object_add(obj, below);
496 if (!ed->ed_next_ismd) {
497 const struct cl_object_conf *cconf = lu2cl_conf(conf);
498 struct echo_object_conf *econf = cl2echo_conf(cconf);
500 LASSERT(econf->eoc_md);
501 eco->eo_lsm = *econf->eoc_md;
502 /* clear the lsm pointer so that it won't get freed. */
503 *econf->eoc_md = NULL;
509 cfs_atomic_set(&eco->eo_npages, 0);
510 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
512 spin_lock(&ec->ec_lock);
513 cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
514 spin_unlock(&ec->ec_lock);
519 /* taken from osc_unpackmd() */
520 static int echo_alloc_memmd(struct echo_device *ed,
521 struct lov_stripe_md **lsmp)
527 /* If export is lov/osc then use their obd method */
528 if (ed->ed_next != NULL)
529 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
530 /* OFD has no unpackmd method, do everything here */
531 lsm_size = lov_stripe_md_size(1);
533 LASSERT(*lsmp == NULL);
534 OBD_ALLOC(*lsmp, lsm_size);
538 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
539 if ((*lsmp)->lsm_oinfo[0] == NULL) {
540 OBD_FREE(*lsmp, lsm_size);
544 loi_init((*lsmp)->lsm_oinfo[0]);
545 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
550 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
556 /* If export is lov/osc then use their obd method */
557 if (ed->ed_next != NULL)
558 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
559 /* OFD has no unpackmd method, do everything here */
560 lsm_size = lov_stripe_md_size(1);
562 LASSERT(*lsmp != NULL);
563 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
564 OBD_FREE(*lsmp, lsm_size);
569 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
571 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
572 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
575 LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
577 spin_lock(&ec->ec_lock);
578 cfs_list_del_init(&eco->eo_obj_chain);
579 spin_unlock(&ec->ec_lock);
582 lu_object_header_fini(obj->lo_header);
585 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
586 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
590 static int echo_object_print(const struct lu_env *env, void *cookie,
591 lu_printer_t p, const struct lu_object *o)
593 struct echo_object *obj = cl2echo_obj(lu2cl(o));
595 return (*p)(env, cookie, "echoclient-object@%p", obj);
598 static const struct lu_object_operations echo_lu_obj_ops = {
599 .loo_object_init = echo_object_init,
600 .loo_object_delete = NULL,
601 .loo_object_release = NULL,
602 .loo_object_free = echo_object_free,
603 .loo_object_print = echo_object_print,
604 .loo_object_invariant = NULL
606 /** @} echo_lu_ops */
608 /** \defgroup echo_lu_dev_ops lu_device operations
610 * Operations for echo lu device.
614 static struct lu_object *echo_object_alloc(const struct lu_env *env,
615 const struct lu_object_header *hdr,
616 struct lu_device *dev)
618 struct echo_object *eco;
619 struct lu_object *obj = NULL;
622 /* we're the top dev. */
623 LASSERT(hdr == NULL);
624 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, CFS_ALLOC_IO);
626 struct cl_object_header *hdr = &eco->eo_hdr;
628 obj = &echo_obj2cl(eco)->co_lu;
629 cl_object_header_init(hdr);
630 lu_object_init(obj, &hdr->coh_lu, dev);
631 lu_object_add_top(&hdr->coh_lu, obj);
633 eco->eo_cl.co_ops = &echo_cl_obj_ops;
634 obj->lo_ops = &echo_lu_obj_ops;
639 static struct lu_device_operations echo_device_lu_ops = {
640 .ldo_object_alloc = echo_object_alloc,
643 /** @} echo_lu_dev_ops */
645 static struct cl_device_operations echo_device_cl_ops = {
648 /** \defgroup echo_init Setup and teardown
650 * Init and fini functions for echo client.
654 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
656 struct cl_site *site = &ed->ed_site_myself;
659 /* initialize site */
660 rc = cl_site_init(site, &ed->ed_cl);
662 CERROR("Cannot initilize site for echo client(%d)\n", rc);
666 rc = lu_site_init_finish(&site->cs_lu);
674 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
677 if (!ed->ed_next_ismd)
678 cl_site_fini(ed->ed_site);
683 static void *echo_thread_key_init(const struct lu_context *ctx,
684 struct lu_context_key *key)
686 struct echo_thread_info *info;
688 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, CFS_ALLOC_IO);
690 info = ERR_PTR(-ENOMEM);
694 static void echo_thread_key_fini(const struct lu_context *ctx,
695 struct lu_context_key *key, void *data)
697 struct echo_thread_info *info = data;
698 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
701 static void echo_thread_key_exit(const struct lu_context *ctx,
702 struct lu_context_key *key, void *data)
706 static struct lu_context_key echo_thread_key = {
707 .lct_tags = LCT_CL_THREAD,
708 .lct_init = echo_thread_key_init,
709 .lct_fini = echo_thread_key_fini,
710 .lct_exit = echo_thread_key_exit
713 static void *echo_session_key_init(const struct lu_context *ctx,
714 struct lu_context_key *key)
716 struct echo_session_info *session;
718 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, CFS_ALLOC_IO);
720 session = ERR_PTR(-ENOMEM);
724 static void echo_session_key_fini(const struct lu_context *ctx,
725 struct lu_context_key *key, void *data)
727 struct echo_session_info *session = data;
728 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
731 static void echo_session_key_exit(const struct lu_context *ctx,
732 struct lu_context_key *key, void *data)
736 static struct lu_context_key echo_session_key = {
737 .lct_tags = LCT_SESSION,
738 .lct_init = echo_session_key_init,
739 .lct_fini = echo_session_key_fini,
740 .lct_exit = echo_session_key_exit
743 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
745 #define ECHO_SEQ_WIDTH 0xffffffff
746 static int echo_fid_init(struct echo_device *ed, char *obd_name,
747 struct seq_server_site *ss)
753 OBD_ALLOC_PTR(ed->ed_cl_seq);
754 if (ed->ed_cl_seq == NULL)
757 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
759 GOTO(out_free_seq, rc = -ENOMEM);
761 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
763 /* Init client side sequence-manager */
764 rc = seq_client_init(ed->ed_cl_seq, NULL,
766 prefix, ss->ss_server_seq);
767 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
768 OBD_FREE(prefix, MAX_OBD_NAME + 5);
770 GOTO(out_free_seq, rc);
775 OBD_FREE_PTR(ed->ed_cl_seq);
776 ed->ed_cl_seq = NULL;
780 static int echo_fid_fini(struct obd_device *obddev)
782 struct echo_device *ed = obd2echo_dev(obddev);
785 if (ed->ed_cl_seq != NULL) {
786 seq_client_fini(ed->ed_cl_seq);
787 OBD_FREE_PTR(ed->ed_cl_seq);
788 ed->ed_cl_seq = NULL;
794 static struct lu_device *echo_device_alloc(const struct lu_env *env,
795 struct lu_device_type *t,
796 struct lustre_cfg *cfg)
798 struct lu_device *next;
799 struct echo_device *ed;
800 struct cl_device *cd;
801 struct obd_device *obd = NULL; /* to keep compiler happy */
802 struct obd_device *tgt;
803 const char *tgt_type_name;
810 GOTO(out, rc = -ENOMEM);
814 rc = cl_device_init(cd, t);
818 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
819 cd->cd_ops = &echo_device_cl_ops;
822 obd = class_name2obd(lustre_cfg_string(cfg, 0));
823 LASSERT(obd != NULL);
824 LASSERT(env != NULL);
826 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
828 CERROR("Can not find tgt device %s\n",
829 lustre_cfg_string(cfg, 1));
830 GOTO(out, rc = -ENODEV);
833 next = tgt->obd_lu_dev;
834 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
835 ed->ed_next_ismd = 1;
837 ed->ed_next_ismd = 0;
838 rc = echo_site_init(env, ed);
844 rc = echo_client_setup(env, obd, cfg);
848 ed->ed_ec = &obd->u.echo_client;
851 if (ed->ed_next_ismd) {
852 /* Suppose to connect to some Metadata layer */
854 struct lu_device *ld;
858 CERROR("%s is not lu device type!\n",
859 lustre_cfg_string(cfg, 1));
860 GOTO(out, rc = -EINVAL);
863 tgt_type_name = lustre_cfg_string(cfg, 2);
864 if (!tgt_type_name) {
865 CERROR("%s no type name for echo %s setup\n",
866 lustre_cfg_string(cfg, 1),
867 tgt->obd_type->typ_name);
868 GOTO(out, rc = -EINVAL);
873 spin_lock(&ls->ls_ld_lock);
874 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
875 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
880 spin_unlock(&ls->ls_ld_lock);
883 CERROR("%s is not lu device type!\n",
884 lustre_cfg_string(cfg, 1));
885 GOTO(out, rc = -EINVAL);
889 /* For MD echo client, it will use the site in MDS stack */
890 ed->ed_site_myself.cs_lu = *ls;
891 ed->ed_site = &ed->ed_site_myself;
892 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
893 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
895 CERROR("echo fid init error %d\n", rc);
899 /* if echo client is to be stacked upon ost device, the next is
900 * NULL since ost is not a clio device so far */
901 if (next != NULL && !lu_device_is_cl(next))
904 tgt_type_name = tgt->obd_type->typ_name;
906 LASSERT(next != NULL);
907 if (next->ld_site != NULL)
908 GOTO(out, rc = -EBUSY);
910 next->ld_site = &ed->ed_site->cs_lu;
911 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
912 next->ld_type->ldt_name,
917 /* Tricky case, I have to determine the obd type since
918 * CLIO uses the different parameters to initialize
919 * objects for lov & osc. */
920 if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
921 ed->ed_next_islov = 1;
923 LASSERT(strcmp(tgt_type_name,
924 LUSTRE_OSC_NAME) == 0);
926 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
930 RETURN(&cd->cd_lu_dev);
935 rc2 = echo_client_cleanup(obd);
937 CERROR("Cleanup obd device %s error(%d)\n",
942 echo_site_fini(env, ed);
944 cl_device_fini(&ed->ed_cl);
954 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
955 const char *name, struct lu_device *next)
961 static struct lu_device *echo_device_fini(const struct lu_env *env,
964 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
965 struct lu_device *next = ed->ed_next;
967 while (next && !ed->ed_next_ismd)
968 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
972 static void echo_lock_release(const struct lu_env *env,
973 struct echo_lock *ecl,
976 struct cl_lock *clk = echo_lock2cl(ecl);
980 cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
982 cl_lock_mutex_get(env, clk);
983 cl_lock_cancel(env, clk);
984 cl_lock_delete(env, clk);
985 cl_lock_mutex_put(env, clk);
987 cl_lock_put(env, clk);
990 static struct lu_device *echo_device_free(const struct lu_env *env,
993 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
994 struct echo_client_obd *ec = ed->ed_ec;
995 struct echo_object *eco;
996 struct lu_device *next = ed->ed_next;
998 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1001 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1003 /* check if there are objects still alive.
1004 * It shouldn't have any object because lu_site_purge would cleanup
1005 * all of cached objects. Anyway, probably the echo device is being
1006 * parallelly accessed.
1008 spin_lock(&ec->ec_lock);
1009 cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1010 eco->eo_deleted = 1;
1011 spin_unlock(&ec->ec_lock);
1014 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1017 "Waiting for the reference of echo object to be dropped\n");
1019 /* Wait for the last reference to be dropped. */
1020 spin_lock(&ec->ec_lock);
1021 while (!cfs_list_empty(&ec->ec_objects)) {
1022 spin_unlock(&ec->ec_lock);
1023 CERROR("echo_client still has objects at cleanup time, "
1024 "wait for 1 second\n");
1025 cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT,
1026 cfs_time_seconds(1));
1027 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1028 spin_lock(&ec->ec_lock);
1030 spin_unlock(&ec->ec_lock);
1032 LASSERT(cfs_list_empty(&ec->ec_locks));
1034 CDEBUG(D_INFO, "No object exists, exiting...\n");
1036 echo_client_cleanup(d->ld_obd);
1037 echo_fid_fini(d->ld_obd);
1038 while (next && !ed->ed_next_ismd)
1039 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1041 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1042 echo_site_fini(env, ed);
1043 cl_device_fini(&ed->ed_cl);
1049 static const struct lu_device_type_operations echo_device_type_ops = {
1050 .ldto_init = echo_type_init,
1051 .ldto_fini = echo_type_fini,
1053 .ldto_start = echo_type_start,
1054 .ldto_stop = echo_type_stop,
1056 .ldto_device_alloc = echo_device_alloc,
1057 .ldto_device_free = echo_device_free,
1058 .ldto_device_init = echo_device_init,
1059 .ldto_device_fini = echo_device_fini
1062 static struct lu_device_type echo_device_type = {
1063 .ldt_tags = LU_DEVICE_CL,
1064 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1065 .ldt_ops = &echo_device_type_ops,
1066 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1070 /** \defgroup echo_exports Exported operations
1072 * exporting functions to echo client
1077 /* Interfaces to echo client obd device */
1078 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1079 struct lov_stripe_md **lsmp)
1082 struct echo_thread_info *info;
1083 struct echo_object_conf *conf;
1084 struct lov_stripe_md *lsm;
1085 struct echo_object *eco;
1086 struct cl_object *obj;
1094 LASSERT(lsm->lsm_object_id);
1096 /* Never return an object if the obd is to be freed. */
1097 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1098 RETURN(ERR_PTR(-ENODEV));
1100 env = cl_env_get(&refcheck);
1102 RETURN((void *)env);
1104 info = echo_env_info(env);
1105 conf = &info->eti_conf;
1107 if (!d->ed_next_islov) {
1108 struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1109 LASSERT(oinfo != NULL);
1110 oinfo->loi_id = lsm->lsm_object_id;
1111 oinfo->loi_seq = lsm->lsm_object_seq;
1112 conf->eoc_cl.u.coc_oinfo = oinfo;
1114 struct lustre_md *md;
1116 memset(md, 0, sizeof *md);
1118 conf->eoc_cl.u.coc_md = md;
1121 conf->eoc_md = lsmp;
1123 fid = &info->eti_fid;
1126 /* In the function below, .hs_keycmp resolves to
1127 * lu_obj_hop_keycmp() */
1128 /* coverity[overrun-buffer-val] */
1129 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1131 GOTO(out, eco = (void*)obj);
1133 eco = cl2echo_obj(obj);
1134 if (eco->eo_deleted) {
1135 cl_object_put(env, obj);
1136 eco = ERR_PTR(-EAGAIN);
1140 cl_env_put(env, &refcheck);
1144 static int cl_echo_object_put(struct echo_object *eco)
1147 struct cl_object *obj = echo_obj2cl(eco);
1151 env = cl_env_get(&refcheck);
1153 RETURN(PTR_ERR(env));
1155 /* an external function to kill an object? */
1156 if (eco->eo_deleted) {
1157 struct lu_object_header *loh = obj->co_lu.lo_header;
1158 LASSERT(&eco->eo_hdr == luh2coh(loh));
1159 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1162 cl_object_put(env, obj);
1163 cl_env_put(env, &refcheck);
1167 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1168 obd_off start, obd_off end, int mode,
1169 __u64 *cookie , __u32 enqflags)
1172 struct cl_lock *lck;
1173 struct cl_object *obj;
1174 struct cl_lock_descr *descr;
1175 struct echo_thread_info *info;
1179 info = echo_env_info(env);
1181 descr = &info->eti_descr;
1182 obj = echo_obj2cl(eco);
1184 descr->cld_obj = obj;
1185 descr->cld_start = cl_index(obj, start);
1186 descr->cld_end = cl_index(obj, end);
1187 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1188 descr->cld_enq_flags = enqflags;
1191 lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1193 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1194 struct echo_lock *el;
1196 rc = cl_wait(env, lck);
1198 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1199 spin_lock(&ec->ec_lock);
1200 if (cfs_list_empty(&el->el_chain)) {
1201 cfs_list_add(&el->el_chain, &ec->ec_locks);
1202 el->el_cookie = ++ec->ec_unique;
1204 cfs_atomic_inc(&el->el_refcount);
1205 *cookie = el->el_cookie;
1206 spin_unlock(&ec->ec_lock);
1208 cl_lock_release(env, lck, "ec enqueue", cfs_current());
1214 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1215 int mode, __u64 *cookie)
1217 struct echo_thread_info *info;
1224 env = cl_env_get(&refcheck);
1226 RETURN(PTR_ERR(env));
1228 info = echo_env_info(env);
1231 io->ci_ignore_layout = 1;
1232 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1235 LASSERT(result == 0);
1237 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1238 cl_io_fini(env, io);
1242 cl_env_put(env, &refcheck);
1246 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1249 struct echo_client_obd *ec = ed->ed_ec;
1250 struct echo_lock *ecl = NULL;
1252 int found = 0, still_used = 0;
1255 LASSERT(ec != NULL);
1256 spin_lock(&ec->ec_lock);
1257 cfs_list_for_each (el, &ec->ec_locks) {
1258 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1259 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1260 found = (ecl->el_cookie == cookie);
1262 if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1263 cfs_list_del_init(&ecl->el_chain);
1269 spin_unlock(&ec->ec_lock);
1274 echo_lock_release(env, ecl, still_used);
1278 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1285 env = cl_env_get(&refcheck);
1287 RETURN(PTR_ERR(env));
1289 rc = cl_echo_cancel0(env, ed, cookie);
1291 cl_env_put(env, &refcheck);
1295 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1296 enum cl_req_type unused, struct cl_2queue *queue)
1298 struct cl_page *clp;
1299 struct cl_page *temp;
1303 cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1305 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1308 result = result ?: rc;
1313 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1314 cfs_page_t **pages, int npages, int async)
1317 struct echo_thread_info *info;
1318 struct cl_object *obj = echo_obj2cl(eco);
1319 struct echo_device *ed = eco->eo_dev;
1320 struct cl_2queue *queue;
1322 struct cl_page *clp;
1323 struct lustre_handle lh = { 0 };
1324 int page_size = cl_page_size(obj);
1330 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1331 LASSERT(ed->ed_next != NULL);
1332 env = cl_env_get(&refcheck);
1334 RETURN(PTR_ERR(env));
1336 info = echo_env_info(env);
1338 queue = &info->eti_queue;
1340 cl_2queue_init(queue);
1342 io->ci_ignore_layout = 1;
1343 rc = cl_io_init(env, io, CIT_MISC, obj);
1349 rc = cl_echo_enqueue0(env, eco, offset,
1350 offset + npages * CFS_PAGE_SIZE - 1,
1351 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1354 GOTO(error_lock, rc);
1356 for (i = 0; i < npages; i++) {
1358 clp = cl_page_find(env, obj, cl_index(obj, offset),
1359 pages[i], CPT_TRANSIENT);
1364 LASSERT(clp->cp_type == CPT_TRANSIENT);
1366 rc = cl_page_own(env, io, clp);
1368 LASSERT(clp->cp_state == CPS_FREEING);
1369 cl_page_put(env, clp);
1373 cl_2queue_add(queue, clp);
1375 /* drop the reference count for cl_page_find, so that the page
1376 * will be freed in cl_2queue_fini. */
1377 cl_page_put(env, clp);
1378 cl_page_clip(env, clp, 0, page_size);
1380 offset += page_size;
1384 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1386 async = async && (typ == CRT_WRITE);
1388 rc = cl_echo_async_brw(env, io, typ, queue);
1390 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1391 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1392 async ? "async" : "sync", rc);
1395 cl_echo_cancel0(env, ed, lh.cookie);
1398 cl_2queue_discard(env, io, queue);
1399 cl_2queue_disown(env, io, queue);
1400 cl_2queue_fini(env, queue);
1401 cl_io_fini(env, io);
1403 cl_env_put(env, &refcheck);
1406 /** @} echo_exports */
1409 static obd_id last_object_id;
1412 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1414 struct lov_stripe_md *ulsm = _ulsm;
1417 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1421 if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm)))
1424 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1425 if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1426 sizeof(lsm->lsm_oinfo[0])))
1433 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1434 void *ulsm, int ulsm_nob)
1436 struct echo_client_obd *ec = ed->ed_ec;
1439 if (ulsm_nob < sizeof (*lsm))
1442 if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm)))
1445 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1446 lsm->lsm_magic != LOV_MAGIC ||
1447 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1448 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1452 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1453 if (cfs_copy_from_user(lsm->lsm_oinfo[i],
1454 ((struct lov_stripe_md *)ulsm)-> \
1456 sizeof(lsm->lsm_oinfo[0])))
1462 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1465 sprintf(name, LPU64, id);
1466 lname->ln_name = name;
1467 lname->ln_namelen = strlen(name);
1470 /* similar to mdt_attr_get_complex */
1471 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1474 struct echo_thread_info *info = echo_env_info(env);
1479 LASSERT(ma->ma_lmm_size > 0);
1481 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1485 /* big_lmm may need to be grown */
1486 if (info->eti_big_lmmsize < rc) {
1487 int size = size_roundup_power2(rc);
1489 if (info->eti_big_lmmsize > 0) {
1490 /* free old buffer */
1491 LASSERT(info->eti_big_lmm);
1492 OBD_FREE_LARGE(info->eti_big_lmm,
1493 info->eti_big_lmmsize);
1494 info->eti_big_lmm = NULL;
1495 info->eti_big_lmmsize = 0;
1498 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1499 if (info->eti_big_lmm == NULL)
1501 info->eti_big_lmmsize = size;
1503 LASSERT(info->eti_big_lmmsize >= rc);
1505 info->eti_buf.lb_buf = info->eti_big_lmm;
1506 info->eti_buf.lb_len = info->eti_big_lmmsize;
1507 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1511 ma->ma_valid |= MA_LOV;
1512 ma->ma_lmm = info->eti_big_lmm;
1513 ma->ma_lmm_size = rc;
1518 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1521 struct echo_thread_info *info = echo_env_info(env);
1522 struct lu_buf *buf = &info->eti_buf;
1523 cfs_umode_t mode = lu_object_attr(&next->mo_lu);
1524 int need = ma->ma_need;
1531 if (need & MA_INODE) {
1532 ma->ma_need = MA_INODE;
1533 rc = mo_attr_get(env, next, ma);
1536 ma->ma_valid |= MA_INODE;
1539 if (need & MA_LOV) {
1540 if (S_ISREG(mode) || S_ISDIR(mode)) {
1541 LASSERT(ma->ma_lmm_size > 0);
1542 buf->lb_buf = ma->ma_lmm;
1543 buf->lb_len = ma->ma_lmm_size;
1544 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1546 ma->ma_lmm_size = rc2;
1547 ma->ma_valid |= MA_LOV;
1548 } else if (rc2 == -ENODATA) {
1550 ma->ma_lmm_size = 0;
1551 } else if (rc2 == -ERANGE) {
1552 rc2 = echo_big_lmm_get(env, next, ma);
1554 GOTO(out, rc = rc2);
1556 GOTO(out, rc = rc2);
1561 #ifdef CONFIG_FS_POSIX_ACL
1562 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1563 buf->lb_buf = ma->ma_acl;
1564 buf->lb_len = ma->ma_acl_size;
1565 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1567 ma->ma_acl_size = rc2;
1568 ma->ma_valid |= MA_ACL_DEF;
1569 } else if (rc2 == -ENODATA) {
1571 ma->ma_acl_size = 0;
1573 GOTO(out, rc = rc2);
1579 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1580 rc, ma->ma_valid, ma->ma_lmm);
1585 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1586 struct md_object *parent, struct lu_fid *fid,
1587 struct lu_name *lname, struct md_op_spec *spec,
1590 struct lu_object *ec_child, *child;
1591 struct lu_device *ld = ed->ed_next;
1592 struct echo_thread_info *info = echo_env_info(env);
1593 struct lu_fid *fid2 = &info->eti_fid2;
1594 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1599 rc = mdo_lookup(env, parent, lname, fid2, spec);
1602 else if (rc != -ENOENT)
1605 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1607 if (IS_ERR(ec_child)) {
1608 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1610 RETURN(PTR_ERR(ec_child));
1613 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1614 if (child == NULL) {
1615 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1616 GOTO(out_put, rc = -EINVAL);
1619 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1620 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1623 * Do not perform lookup sanity check. We know that name does not exist.
1625 spec->sp_cr_lookup = 0;
1626 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1628 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1631 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1632 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1635 lu_object_put(env, ec_child);
1639 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1642 struct echo_thread_info *info = echo_env_info(env);
1644 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1645 ma->ma_lmm = (void *)&info->eti_lmm;
1646 ma->ma_lmm_size = sizeof(info->eti_lmm);
1648 LASSERT(info->eti_big_lmmsize);
1649 ma->ma_lmm = info->eti_big_lmm;
1650 ma->ma_lmm_size = info->eti_big_lmmsize;
1656 static int echo_create_md_object(const struct lu_env *env,
1657 struct echo_device *ed,
1658 struct lu_object *ec_parent,
1660 char *name, int namelen,
1661 __u64 id, __u32 mode, int count,
1662 int stripe_count, int stripe_offset)
1664 struct lu_object *parent;
1665 struct echo_thread_info *info = echo_env_info(env);
1666 struct lu_name *lname = &info->eti_lname;
1667 struct md_op_spec *spec = &info->eti_spec;
1668 struct md_attr *ma = &info->eti_ma;
1669 struct lu_device *ld = ed->ed_next;
1675 if (ec_parent == NULL)
1677 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1681 memset(ma, 0, sizeof(*ma));
1682 memset(spec, 0, sizeof(*spec));
1683 if (stripe_count != 0) {
1684 spec->sp_cr_flags |= FMODE_WRITE;
1685 echo_set_lmm_size(env, ld, ma);
1686 if (stripe_count != -1) {
1687 struct lov_user_md_v3 *lum = &info->eti_lum;
1689 lum->lmm_magic = LOV_USER_MAGIC_V3;
1690 lum->lmm_stripe_count = stripe_count;
1691 lum->lmm_stripe_offset = stripe_offset;
1692 lum->lmm_pattern = 0;
1693 spec->u.sp_ea.eadata = lum;
1694 spec->u.sp_ea.eadatalen = sizeof(*lum);
1695 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1699 ma->ma_attr.la_mode = mode;
1700 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1701 ma->ma_attr.la_ctime = cfs_time_current_64();
1704 lname->ln_name = name;
1705 lname->ln_namelen = namelen;
1706 /* If name is specified, only create one object by name */
1707 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1712 /* Create multiple object sequenced by id */
1713 for (i = 0; i < count; i++) {
1714 char *tmp_name = info->eti_name;
1716 echo_md_build_name(lname, tmp_name, id);
1718 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1721 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1732 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1733 struct echo_device *ed,
1734 struct md_object *parent,
1735 struct lu_name *lname)
1737 struct echo_thread_info *info = echo_env_info(env);
1738 struct lu_fid *fid = &info->eti_fid;
1739 struct lu_object *child;
1743 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1745 rc = mdo_lookup(env, parent, lname, fid, NULL);
1747 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1748 RETURN(ERR_PTR(rc));
1751 /* In the function below, .hs_keycmp resolves to
1752 * lu_obj_hop_keycmp() */
1753 /* coverity[overrun-buffer-val] */
1754 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1759 static int echo_setattr_object(const struct lu_env *env,
1760 struct echo_device *ed,
1761 struct lu_object *ec_parent,
1762 __u64 id, int count)
1764 struct lu_object *parent;
1765 struct echo_thread_info *info = echo_env_info(env);
1766 struct lu_name *lname = &info->eti_lname;
1767 char *name = info->eti_name;
1768 struct lu_device *ld = ed->ed_next;
1769 struct lu_buf *buf = &info->eti_buf;
1775 if (ec_parent == NULL)
1777 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1781 for (i = 0; i < count; i++) {
1782 struct lu_object *ec_child, *child;
1784 echo_md_build_name(lname, name, id);
1786 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1787 if (IS_ERR(ec_child)) {
1788 CERROR("Can't find child %s: rc = %ld\n",
1789 lname->ln_name, PTR_ERR(ec_child));
1790 RETURN(PTR_ERR(ec_child));
1793 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1794 if (child == NULL) {
1795 CERROR("Can not locate the child %s\n", lname->ln_name);
1796 lu_object_put(env, ec_child);
1801 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1802 PFID(lu_object_fid(child)));
1804 buf->lb_buf = info->eti_xattr_buf;
1805 buf->lb_len = sizeof(info->eti_xattr_buf);
1807 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1808 rc = mo_xattr_set(env, lu2md(child), buf, name,
1811 CERROR("Can not setattr child "DFID": rc = %d\n",
1812 PFID(lu_object_fid(child)), rc);
1813 lu_object_put(env, ec_child);
1816 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1817 PFID(lu_object_fid(child)));
1819 lu_object_put(env, ec_child);
1824 static int echo_getattr_object(const struct lu_env *env,
1825 struct echo_device *ed,
1826 struct lu_object *ec_parent,
1827 __u64 id, int count)
1829 struct lu_object *parent;
1830 struct echo_thread_info *info = echo_env_info(env);
1831 struct lu_name *lname = &info->eti_lname;
1832 char *name = info->eti_name;
1833 struct md_attr *ma = &info->eti_ma;
1834 struct lu_device *ld = ed->ed_next;
1840 if (ec_parent == NULL)
1842 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1846 memset(ma, 0, sizeof(*ma));
1847 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1848 ma->ma_acl = info->eti_xattr_buf;
1849 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1851 for (i = 0; i < count; i++) {
1852 struct lu_object *ec_child, *child;
1855 echo_md_build_name(lname, name, id);
1856 echo_set_lmm_size(env, ld, ma);
1858 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1859 if (IS_ERR(ec_child)) {
1860 CERROR("Can't find child %s: rc = %ld\n",
1861 lname->ln_name, PTR_ERR(ec_child));
1862 RETURN(PTR_ERR(ec_child));
1865 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1866 if (child == NULL) {
1867 CERROR("Can not locate the child %s\n", lname->ln_name);
1868 lu_object_put(env, ec_child);
1872 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1873 PFID(lu_object_fid(child)));
1874 rc = echo_attr_get_complex(env, lu2md(child), ma);
1876 CERROR("Can not getattr child "DFID": rc = %d\n",
1877 PFID(lu_object_fid(child)), rc);
1878 lu_object_put(env, ec_child);
1881 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1882 PFID(lu_object_fid(child)));
1884 lu_object_put(env, ec_child);
1890 static int echo_lookup_object(const struct lu_env *env,
1891 struct echo_device *ed,
1892 struct lu_object *ec_parent,
1893 __u64 id, int count)
1895 struct lu_object *parent;
1896 struct echo_thread_info *info = echo_env_info(env);
1897 struct lu_name *lname = &info->eti_lname;
1898 char *name = info->eti_name;
1899 struct lu_fid *fid = &info->eti_fid;
1900 struct lu_device *ld = ed->ed_next;
1904 if (ec_parent == NULL)
1906 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1910 /*prepare the requests*/
1911 for (i = 0; i < count; i++) {
1912 echo_md_build_name(lname, name, id);
1914 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1915 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1917 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1919 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1922 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1923 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1930 static int echo_md_destroy_internal(const struct lu_env *env,
1931 struct echo_device *ed,
1932 struct md_object *parent,
1933 struct lu_name *lname,
1936 struct lu_device *ld = ed->ed_next;
1937 struct lu_object *ec_child;
1938 struct lu_object *child;
1943 ec_child = echo_md_lookup(env, ed, parent, lname);
1944 if (IS_ERR(ec_child)) {
1945 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1947 RETURN(PTR_ERR(ec_child));
1950 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1951 if (child == NULL) {
1952 CERROR("Can not locate the child %s\n", lname->ln_name);
1953 GOTO(out_put, rc = -EINVAL);
1956 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1957 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1959 rc = mdo_unlink(env, parent, lu2md(child), lname, ma);
1961 CERROR("Can not unlink child %s: rc = %d\n",
1962 lname->ln_name, rc);
1965 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1966 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1968 lu_object_put(env, ec_child);
1972 static int echo_destroy_object(const struct lu_env *env,
1973 struct echo_device *ed,
1974 struct lu_object *ec_parent,
1975 char *name, int namelen,
1976 __u64 id, __u32 mode,
1979 struct echo_thread_info *info = echo_env_info(env);
1980 struct lu_name *lname = &info->eti_lname;
1981 struct md_attr *ma = &info->eti_ma;
1982 struct lu_device *ld = ed->ed_next;
1983 struct lu_object *parent;
1988 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1992 memset(ma, 0, sizeof(*ma));
1993 ma->ma_attr.la_mode = mode;
1994 ma->ma_attr.la_valid = LA_CTIME;
1995 ma->ma_attr.la_ctime = cfs_time_current_64();
1996 ma->ma_need = MA_INODE;
2000 lname->ln_name = name;
2001 lname->ln_namelen = namelen;
2002 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2007 /*prepare the requests*/
2008 for (i = 0; i < count; i++) {
2009 char *tmp_name = info->eti_name;
2012 echo_md_build_name(lname, tmp_name, id);
2014 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2017 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2026 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2027 struct echo_device *ed, char *path,
2030 struct lu_device *ld = ed->ed_next;
2031 struct md_device *md = lu2md_dev(ld);
2032 struct echo_thread_info *info = echo_env_info(env);
2033 struct lu_fid *fid = &info->eti_fid;
2034 struct lu_name *lname = &info->eti_lname;
2035 struct lu_object *parent = NULL;
2036 struct lu_object *child = NULL;
2040 /*Only support MDD layer right now*/
2041 rc = md->md_ops->mdo_root_get(env, md, fid);
2043 CERROR("get root error: rc = %d\n", rc);
2044 RETURN(ERR_PTR(rc));
2047 /* In the function below, .hs_keycmp resolves to
2048 * lu_obj_hop_keycmp() */
2049 /* coverity[overrun-buffer-val] */
2050 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2051 if (IS_ERR(parent)) {
2052 CERROR("Can not find the parent "DFID": rc = %ld\n",
2053 PFID(fid), PTR_ERR(parent));
2058 struct lu_object *ld_parent;
2061 e = strsep(&path, "/");
2066 if (!path || path[0] == '\0')
2072 lname->ln_namelen = strlen(e);
2074 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2075 if (ld_parent == NULL) {
2076 lu_object_put(env, parent);
2081 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2082 lu_object_put(env, parent);
2083 if (IS_ERR(child)) {
2084 rc = (int)PTR_ERR(child);
2085 CERROR("lookup %s under parent "DFID": rc = %d\n",
2086 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2093 RETURN(ERR_PTR(rc));
2098 static void echo_ucred_init(struct lu_env *env)
2100 struct lu_ucred *ucred = lu_ucred(env);
2102 ucred->uc_valid = UCRED_INVALID;
2104 ucred->uc_suppgids[0] = -1;
2105 ucred->uc_suppgids[1] = -1;
2107 ucred->uc_uid = ucred->uc_o_uid = cfs_curproc_uid();
2108 ucred->uc_gid = ucred->uc_o_gid = cfs_curproc_gid();
2109 ucred->uc_fsuid = ucred->uc_o_fsuid = cfs_curproc_fsuid();
2110 ucred->uc_fsgid = ucred->uc_o_fsgid = cfs_curproc_fsgid();
2111 ucred->uc_cap = cfs_curproc_cap_pack();
2113 /* remove fs privilege for non-root user. */
2114 if (ucred->uc_fsuid)
2115 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2116 ucred->uc_valid = UCRED_NEW;
2119 static void echo_ucred_fini(struct lu_env *env)
2121 struct lu_ucred *ucred = lu_ucred(env);
2122 ucred->uc_valid = UCRED_INIT;
2125 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2126 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2127 static int echo_md_handler(struct echo_device *ed, int command,
2128 char *path, int path_len, int id, int count,
2129 struct obd_ioctl_data *data)
2131 struct echo_thread_info *info;
2132 struct lu_device *ld = ed->ed_next;
2135 struct lu_object *parent;
2137 int namelen = data->ioc_plen2;
2142 CERROR("MD echo client is not being initialized properly\n");
2146 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2147 CERROR("Only support MDD layer right now!\n");
2151 env = cl_env_get(&refcheck);
2153 RETURN(PTR_ERR(env));
2155 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2159 /* init big_lmm buffer */
2160 info = echo_env_info(env);
2161 LASSERT(info->eti_big_lmm == NULL);
2162 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2163 if (info->eti_big_lmm == NULL)
2164 GOTO(out_env, rc = -ENOMEM);
2165 info->eti_big_lmmsize = MIN_MD_SIZE;
2167 parent = echo_resolve_path(env, ed, path, path_len);
2168 if (IS_ERR(parent)) {
2169 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2171 GOTO(out_free, rc = PTR_ERR(parent));
2175 OBD_ALLOC(name, namelen + 1);
2177 GOTO(out_put, rc = -ENOMEM);
2178 if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen))
2179 GOTO(out_name, rc = -EFAULT);
2182 echo_ucred_init(env);
2185 case ECHO_MD_CREATE:
2186 case ECHO_MD_MKDIR: {
2187 struct echo_thread_info *info = echo_env_info(env);
2188 __u32 mode = data->ioc_obdo2.o_mode;
2189 struct lu_fid *fid = &info->eti_fid;
2190 int stripe_count = (int)data->ioc_obdo2.o_misc;
2191 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2193 fid->f_seq = data->ioc_obdo1.o_seq;
2194 fid->f_oid = (__u32)data->ioc_obdo1.o_id;
2196 /* In the function below, .hs_keycmp resolves to
2197 * lu_obj_hop_keycmp() */
2198 /* coverity[overrun-buffer-val] */
2199 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2200 id, mode, count, stripe_count,
2204 case ECHO_MD_DESTROY:
2205 case ECHO_MD_RMDIR: {
2206 __u32 mode = data->ioc_obdo2.o_mode;
2208 rc = echo_destroy_object(env, ed, parent, name, namelen,
2212 case ECHO_MD_LOOKUP:
2213 rc = echo_lookup_object(env, ed, parent, id, count);
2215 case ECHO_MD_GETATTR:
2216 rc = echo_getattr_object(env, ed, parent, id, count);
2218 case ECHO_MD_SETATTR:
2219 rc = echo_setattr_object(env, ed, parent, id, count);
2222 CERROR("unknown command %d\n", command);
2226 echo_ucred_fini(env);
2230 OBD_FREE(name, namelen + 1);
2232 lu_object_put(env, parent);
2234 LASSERT(info->eti_big_lmm);
2235 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2236 info->eti_big_lmm = NULL;
2237 info->eti_big_lmmsize = 0;
2239 cl_env_put(env, &refcheck);
2243 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2244 int on_target, struct obdo *oa, void *ulsm,
2245 int ulsm_nob, struct obd_trans_info *oti)
2247 struct echo_object *eco;
2248 struct echo_client_obd *ec = ed->ed_ec;
2249 struct lov_stripe_md *lsm = NULL;
2254 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2255 (on_target || /* set_stripe */
2256 ec->ec_nstripes != 0)) { /* LOV */
2257 CERROR ("No valid oid\n");
2261 rc = echo_alloc_memmd(ed, &lsm);
2263 CERROR("Cannot allocate md: rc = %d\n", rc);
2270 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2274 if (lsm->lsm_stripe_count == 0)
2275 lsm->lsm_stripe_count = ec->ec_nstripes;
2277 if (lsm->lsm_stripe_size == 0)
2278 lsm->lsm_stripe_size = CFS_PAGE_SIZE;
2282 /* setup stripes: indices + default ids if required */
2283 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2284 if (lsm->lsm_oinfo[i]->loi_id == 0)
2285 lsm->lsm_oinfo[i]->loi_id = lsm->lsm_object_id;
2287 lsm->lsm_oinfo[i]->loi_ost_idx =
2288 (idx + i) % ec->ec_nstripes;
2292 /* setup object ID here for !on_target and LOV hint */
2293 if (oa->o_valid & OBD_MD_FLID)
2294 lsm->lsm_object_id = oa->o_id;
2296 if (lsm->lsm_object_id == 0)
2297 lsm->lsm_object_id = ++last_object_id;
2301 /* Only echo objects are allowed to be created */
2302 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2303 (oa->o_seq == FID_SEQ_ECHO));
2304 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2306 CERROR("Cannot create objects: rc = %d\n", rc);
2312 /* See what object ID we were given */
2313 oa->o_id = lsm->lsm_object_id;
2314 oa->o_valid |= OBD_MD_FLID;
2316 eco = cl_echo_object_find(ed, &lsm);
2318 GOTO(failed, rc = PTR_ERR(eco));
2319 cl_echo_object_put(eco);
2321 CDEBUG(D_INFO, "oa->o_id = %lx\n", (long)oa->o_id);
2326 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2328 echo_free_memmd(ed, &lsm);
2330 CERROR("create object failed with: rc = %d\n", rc);
2334 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2337 struct lov_stripe_md *lsm = NULL;
2338 struct echo_object *eco;
2342 if ((oa->o_valid & OBD_MD_FLID) == 0 ||
2343 oa->o_id == 0) /* disallow use of object id 0 */
2345 CERROR ("No valid oid\n");
2349 rc = echo_alloc_memmd(ed, &lsm);
2353 lsm->lsm_object_id = oa->o_id;
2354 if (oa->o_valid & OBD_MD_FLGROUP)
2355 lsm->lsm_object_seq = oa->o_seq;
2357 lsm->lsm_object_seq = FID_SEQ_ECHO;
2360 eco = cl_echo_object_find(ed, &lsm);
2366 echo_free_memmd(ed, &lsm);
2370 static void echo_put_object(struct echo_object *eco)
2372 if (cl_echo_object_put(eco))
2373 CERROR("echo client: drop an object failed");
2377 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2379 unsigned long stripe_count;
2380 unsigned long stripe_size;
2381 unsigned long width;
2382 unsigned long woffset;
2386 if (lsm->lsm_stripe_count <= 1)
2390 stripe_size = lsm->lsm_stripe_size;
2391 stripe_count = lsm->lsm_stripe_count;
2393 /* width = # bytes in all stripes */
2394 width = stripe_size * stripe_count;
2396 /* woffset = offset within a width; offset = whole number of widths */
2397 woffset = do_div (offset, width);
2399 stripe_index = woffset / stripe_size;
2401 *idp = lsm->lsm_oinfo[stripe_index]->loi_id;
2402 *offp = offset * stripe_size + woffset % stripe_size;
2406 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2407 cfs_page_t *page, int rw, obd_id id,
2408 obd_off offset, obd_off count)
2415 /* no partial pages on the client */
2416 LASSERT(count == CFS_PAGE_SIZE);
2418 addr = cfs_kmap(page);
2420 for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2421 if (rw == OBD_BRW_WRITE) {
2422 stripe_off = offset + delta;
2424 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2426 stripe_off = 0xdeadbeef00c0ffeeULL;
2427 stripe_id = 0xdeadbeef00c0ffeeULL;
2429 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2430 stripe_off, stripe_id);
2436 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2437 cfs_page_t *page, obd_id id,
2438 obd_off offset, obd_off count)
2447 /* no partial pages on the client */
2448 LASSERT(count == CFS_PAGE_SIZE);
2450 addr = cfs_kmap(page);
2452 for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2453 stripe_off = offset + delta;
2455 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2457 rc2 = block_debug_check("test_brw",
2458 addr + delta, OBD_ECHO_BLOCK_SIZE,
2459 stripe_off, stripe_id);
2461 CERROR ("Error in echo object "LPX64"\n", id);
2470 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2471 struct echo_object *eco, obd_off offset,
2472 obd_size count, int async,
2473 struct obd_trans_info *oti)
2475 struct lov_stripe_md *lsm = eco->eo_lsm;
2477 struct brw_page *pga;
2478 struct brw_page *pgp;
2488 verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
2489 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2490 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2492 gfp_mask = ((oa->o_id & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
2494 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2495 LASSERT(lsm != NULL);
2496 LASSERT(lsm->lsm_object_id == oa->o_id);
2499 (count & (~CFS_PAGE_MASK)) != 0)
2502 /* XXX think again with misaligned I/O */
2503 npages = count >> CFS_PAGE_SHIFT;
2505 if (rw == OBD_BRW_WRITE)
2506 brw_flags = OBD_BRW_ASYNC;
2508 OBD_ALLOC(pga, npages * sizeof(*pga));
2512 OBD_ALLOC(pages, npages * sizeof(*pages));
2513 if (pages == NULL) {
2514 OBD_FREE(pga, npages * sizeof(*pga));
2518 for (i = 0, pgp = pga, off = offset;
2520 i++, pgp++, off += CFS_PAGE_SIZE) {
2522 LASSERT (pgp->pg == NULL); /* for cleanup */
2525 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2526 if (pgp->pg == NULL)
2530 pgp->count = CFS_PAGE_SIZE;
2532 pgp->flag = brw_flags;
2535 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2536 oa->o_id, off, pgp->count);
2539 /* brw mode can only be used at client */
2540 LASSERT(ed->ed_next != NULL);
2541 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2544 if (rc != 0 || rw != OBD_BRW_READ)
2547 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2548 if (pgp->pg == NULL)
2553 vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
2554 pgp->off, pgp->count);
2555 if (vrc != 0 && rc == 0)
2558 OBD_PAGE_FREE(pgp->pg);
2560 OBD_FREE(pga, npages * sizeof(*pga));
2561 OBD_FREE(pages, npages * sizeof(*pages));
2565 static int echo_client_prep_commit(const struct lu_env *env,
2566 struct obd_export *exp, int rw,
2567 struct obdo *oa, struct echo_object *eco,
2568 obd_off offset, obd_size count,
2569 obd_size batch, struct obd_trans_info *oti,
2572 struct lov_stripe_md *lsm = eco->eo_lsm;
2573 struct obd_ioobj ioo;
2574 struct niobuf_local *lnb;
2575 struct niobuf_remote *rnb;
2577 obd_size npages, tot_pages;
2578 int i, ret = 0, brw_flags = 0;
2582 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2583 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
2586 npages = batch >> CFS_PAGE_SHIFT;
2587 tot_pages = count >> CFS_PAGE_SHIFT;
2589 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2590 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2592 if (lnb == NULL || rnb == NULL)
2593 GOTO(out, ret = -ENOMEM);
2595 if (rw == OBD_BRW_WRITE && async)
2596 brw_flags |= OBD_BRW_ASYNC;
2598 obdo_to_ioobj(oa, &ioo);
2602 for(; tot_pages; tot_pages -= npages) {
2605 if (tot_pages < npages)
2608 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
2609 rnb[i].offset = off;
2610 rnb[i].len = CFS_PAGE_SIZE;
2611 rnb[i].flags = brw_flags;
2614 ioo.ioo_bufcnt = npages;
2615 oti->oti_transno = 0;
2618 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2622 LASSERT(lpages == npages);
2624 for (i = 0; i < lpages; i++) {
2625 cfs_page_t *page = lnb[i].page;
2627 /* read past eof? */
2628 if (page == NULL && lnb[i].rc == 0)
2632 lnb[i].flags |= OBD_BRW_ASYNC;
2634 if (oa->o_id == ECHO_PERSISTENT_OBJID ||
2635 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2636 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2639 if (rw == OBD_BRW_WRITE)
2640 echo_client_page_debug_setup(lsm, page, rw,
2645 echo_client_page_debug_check(lsm, page,
2651 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2652 rnb, npages, lnb, oti, ret);
2656 /* Reset oti otherwise it would confuse ldiskfs. */
2657 memset(oti, 0, sizeof(*oti));
2662 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2664 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2668 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2669 struct obd_export *exp,
2670 struct obd_ioctl_data *data,
2671 struct obd_trans_info *dummy_oti)
2673 struct obd_device *obd = class_exp2obd(exp);
2674 struct echo_device *ed = obd2echo_dev(obd);
2675 struct echo_client_obd *ec = ed->ed_ec;
2676 struct obdo *oa = &data->ioc_obdo1;
2677 struct echo_object *eco;
2683 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2685 rc = echo_get_object(&eco, ed, oa);
2689 oa->o_valid &= ~OBD_MD_FLHANDLE;
2691 /* OFD/obdfilter works only via prep/commit */
2692 test_mode = (long)data->ioc_pbuf1;
2696 if (ed->ed_next == NULL && test_mode != 3) {
2698 data->ioc_plen1 = data->ioc_count;
2701 /* Truncate batch size to maximum */
2702 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2703 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2705 switch (test_mode) {
2709 rc = echo_client_kbrw(ed, rw, oa,
2710 eco, data->ioc_offset,
2711 data->ioc_count, async, dummy_oti);
2714 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2715 eco, data->ioc_offset,
2716 data->ioc_count, data->ioc_plen1,
2722 echo_put_object(eco);
2727 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2728 int mode, obd_off offset, obd_size nob)
2730 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2731 struct lustre_handle *ulh = &oa->o_handle;
2732 struct echo_object *eco;
2737 if (ed->ed_next == NULL)
2738 RETURN(-EOPNOTSUPP);
2740 if (!(mode == LCK_PR || mode == LCK_PW))
2743 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2744 (nob & (~CFS_PAGE_MASK)) != 0)
2747 rc = echo_get_object (&eco, ed, oa);
2751 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2752 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2754 oa->o_valid |= OBD_MD_FLHANDLE;
2755 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2757 echo_put_object(eco);
2762 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2764 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2765 __u64 cookie = oa->o_handle.cookie;
2767 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2770 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2771 return cl_echo_cancel(ed, cookie);
2775 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2776 void *karg, void *uarg)
2778 struct obd_device *obd = exp->exp_obd;
2779 struct echo_device *ed = obd2echo_dev(obd);
2780 struct echo_client_obd *ec = ed->ed_ec;
2781 struct echo_object *eco;
2782 struct obd_ioctl_data *data = karg;
2783 struct obd_trans_info dummy_oti;
2785 struct oti_req_ack_lock *ack_lock;
2788 int rw = OBD_BRW_READ;
2793 memset(&dummy_oti, 0, sizeof(dummy_oti));
2795 oa = &data->ioc_obdo1;
2796 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2797 oa->o_valid |= OBD_MD_FLGROUP;
2798 oa->o_seq = FID_SEQ_ECHO;
2801 /* This FID is unpacked just for validation at this point */
2802 rc = fid_ostid_unpack(&fid, &oa->o_oi, 0);
2810 rc = lu_env_init(env, LCT_DT_THREAD);
2812 GOTO(out, rc = -ENOMEM);
2815 case OBD_IOC_CREATE: /* may create echo object */
2816 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2817 GOTO (out, rc = -EPERM);
2819 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2820 data->ioc_plen1, &dummy_oti);
2823 case OBD_IOC_ECHO_MD: {
2830 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2831 GOTO(out, rc = -EPERM);
2833 count = data->ioc_count;
2834 cmd = data->ioc_command;
2836 id = data->ioc_obdo2.o_id;
2838 dirlen = data->ioc_plen1;
2839 OBD_ALLOC(dir, dirlen + 1);
2841 GOTO(out, rc = -ENOMEM);
2843 if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2844 OBD_FREE(dir, data->ioc_plen1 + 1);
2845 GOTO(out, rc = -EFAULT);
2848 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2849 OBD_FREE(dir, dirlen + 1);
2852 case OBD_IOC_ECHO_ALLOC_SEQ: {
2853 struct lu_env *cl_env;
2858 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2859 GOTO(out, rc = -EPERM);
2861 cl_env = cl_env_get(&refcheck);
2863 GOTO(out, rc = PTR_ERR(cl_env));
2865 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2868 cl_env_put(cl_env, &refcheck);
2872 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2873 cl_env_put(cl_env, &refcheck);
2875 CERROR("%s: Can not alloc seq: rc = %d\n",
2880 if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2883 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2884 if (cfs_copy_to_user(data->ioc_pbuf2, &max_count,
2889 case OBD_IOC_DESTROY:
2890 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2891 GOTO (out, rc = -EPERM);
2893 rc = echo_get_object(&eco, ed, oa);
2895 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2896 &dummy_oti, NULL, NULL);
2898 eco->eo_deleted = 1;
2899 echo_put_object(eco);
2903 case OBD_IOC_GETATTR:
2904 rc = echo_get_object(&eco, ed, oa);
2906 struct obd_info oinfo = { { { 0 } } };
2907 oinfo.oi_md = eco->eo_lsm;
2909 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2910 echo_put_object(eco);
2914 case OBD_IOC_SETATTR:
2915 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2916 GOTO (out, rc = -EPERM);
2918 rc = echo_get_object(&eco, ed, oa);
2920 struct obd_info oinfo = { { { 0 } } };
2922 oinfo.oi_md = eco->eo_lsm;
2924 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2925 echo_put_object(eco);
2929 case OBD_IOC_BRW_WRITE:
2930 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2931 GOTO (out, rc = -EPERM);
2935 case OBD_IOC_BRW_READ:
2936 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2939 case ECHO_IOC_GET_STRIPE:
2940 rc = echo_get_object(&eco, ed, oa);
2942 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2944 echo_put_object(eco);
2948 case ECHO_IOC_SET_STRIPE:
2949 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2950 GOTO (out, rc = -EPERM);
2952 if (data->ioc_pbuf1 == NULL) { /* unset */
2953 rc = echo_get_object(&eco, ed, oa);
2955 eco->eo_deleted = 1;
2956 echo_put_object(eco);
2959 rc = echo_create_object(env, ed, 0, oa,
2961 data->ioc_plen1, &dummy_oti);
2965 case ECHO_IOC_ENQUEUE:
2966 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2967 GOTO (out, rc = -EPERM);
2969 rc = echo_client_enqueue(exp, oa,
2970 data->ioc_conn1, /* lock mode */
2972 data->ioc_count);/*extent*/
2975 case ECHO_IOC_CANCEL:
2976 rc = echo_client_cancel(exp, oa);
2980 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2981 GOTO (out, rc = -ENOTTY);
2989 /* XXX this should be in a helper also called by target_send_reply */
2990 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2992 if (!ack_lock->mode)
2994 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
3000 static int echo_client_setup(const struct lu_env *env,
3001 struct obd_device *obddev, struct lustre_cfg *lcfg)
3003 struct echo_client_obd *ec = &obddev->u.echo_client;
3004 struct obd_device *tgt;
3005 struct obd_uuid echo_uuid = { "ECHO_UUID" };
3006 struct obd_connect_data *ocd = NULL;
3010 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3011 CERROR("requires a TARGET OBD name\n");
3015 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3016 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3017 CERROR("device not attached or not set up (%s)\n",
3018 lustre_cfg_string(lcfg, 1));
3022 spin_lock_init(&ec->ec_lock);
3023 CFS_INIT_LIST_HEAD (&ec->ec_objects);
3024 CFS_INIT_LIST_HEAD (&ec->ec_locks);
3026 ec->ec_nstripes = 0;
3028 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3029 lu_context_tags_update(ECHO_MD_CTX_TAG);
3030 lu_session_tags_update(ECHO_MD_SES_TAG);
3034 OBD_ALLOC(ocd, sizeof(*ocd));
3036 CERROR("Can't alloc ocd connecting to %s\n",
3037 lustre_cfg_string(lcfg, 1));
3041 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3042 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3043 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE;
3044 ocd->ocd_version = LUSTRE_VERSION_CODE;
3045 ocd->ocd_group = FID_SEQ_ECHO;
3047 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3049 /* Turn off pinger because it connects to tgt obd directly. */
3050 spin_lock(&tgt->obd_dev_lock);
3051 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3052 spin_unlock(&tgt->obd_dev_lock);
3055 OBD_FREE(ocd, sizeof(*ocd));
3058 CERROR("fail to connect to device %s\n",
3059 lustre_cfg_string(lcfg, 1));
3066 static int echo_client_cleanup(struct obd_device *obddev)
3068 struct echo_device *ed = obd2echo_dev(obddev);
3069 struct echo_client_obd *ec = &obddev->u.echo_client;
3073 /*Do nothing for Metadata echo client*/
3077 if (ed->ed_next_ismd) {
3078 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3079 lu_session_tags_clear(ECHO_MD_SES_TAG);
3083 if (!cfs_list_empty(&obddev->obd_exports)) {
3084 CERROR("still has clients!\n");
3088 LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3089 rc = obd_disconnect(ec->ec_exp);
3091 CERROR("fail to disconnect device: %d\n", rc);
3096 static int echo_client_connect(const struct lu_env *env,
3097 struct obd_export **exp,
3098 struct obd_device *src, struct obd_uuid *cluuid,
3099 struct obd_connect_data *data, void *localdata)
3102 struct lustre_handle conn = { 0 };
3105 rc = class_connect(&conn, src, cluuid);
3107 *exp = class_conn2export(&conn);
3113 static int echo_client_disconnect(struct obd_export *exp)
3116 struct obd_device *obd;
3117 struct echo_client_obd *ec;
3118 struct ec_lock *ecl;
3124 GOTO(out, rc = -EINVAL);
3128 ec = &obd->u.echo_client;
3130 /* no more contention on export's lock list */
3131 while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
3132 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
3133 struct ec_lock, ecl_exp_chain);
3134 cfs_list_del (&ecl->ecl_exp_chain);
3136 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3137 ecl->ecl_mode, &ecl->ecl_lock_handle);
3139 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3140 "(%d)\n", ecl->ecl_object->eco_id, rc);
3142 echo_put_object (ecl->ecl_object);
3143 OBD_FREE (ecl, sizeof (*ecl));
3147 rc = class_disconnect(exp);
3153 static struct obd_ops echo_client_obd_ops = {
3154 .o_owner = THIS_MODULE,
3157 .o_setup = echo_client_setup,
3158 .o_cleanup = echo_client_cleanup,
3161 .o_iocontrol = echo_client_iocontrol,
3162 .o_connect = echo_client_connect,
3163 .o_disconnect = echo_client_disconnect
3166 int echo_client_init(void)
3168 struct lprocfs_static_vars lvars = { 0 };
3171 lprocfs_echo_init_vars(&lvars);
3173 rc = lu_kmem_init(echo_caches);
3175 rc = class_register_type(&echo_client_obd_ops, NULL,
3177 LUSTRE_ECHO_CLIENT_NAME,
3180 lu_kmem_fini(echo_caches);
3185 void echo_client_exit(void)
3187 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3188 lu_kmem_fini(echo_caches);
3192 static int __init obdecho_init(void)
3194 struct lprocfs_static_vars lvars;
3198 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3200 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3202 lprocfs_echo_init_vars(&lvars);
3204 # ifdef HAVE_SERVER_SUPPORT
3205 rc = echo_persistent_pages_init();
3209 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3210 LUSTRE_ECHO_NAME, NULL);
3215 rc = echo_client_init();
3217 # ifdef HAVE_SERVER_SUPPORT
3221 class_unregister_type(LUSTRE_ECHO_NAME);
3223 echo_persistent_pages_fini();
3229 static void /*__exit*/ obdecho_exit(void)
3233 # ifdef HAVE_SERVER_SUPPORT
3234 class_unregister_type(LUSTRE_ECHO_NAME);
3235 echo_persistent_pages_fini();
3239 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3240 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3241 MODULE_LICENSE("GPL");
3243 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3244 #endif /* __KERNEL__ */
3246 /** @} echo_client */