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, Whamcloud, Inc.
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_page_kmem;
226 static cfs_mem_cache_t *echo_lock_kmem;
227 static cfs_mem_cache_t *echo_object_kmem;
228 static cfs_mem_cache_t *echo_thread_kmem;
229 static cfs_mem_cache_t *echo_session_kmem;
230 //static cfs_mem_cache_t *echo_req_kmem;
232 static struct lu_kmem_descr echo_caches[] = {
234 .ckd_cache = &echo_page_kmem,
235 .ckd_name = "echo_page_kmem",
236 .ckd_size = sizeof (struct echo_page)
239 .ckd_cache = &echo_lock_kmem,
240 .ckd_name = "echo_lock_kmem",
241 .ckd_size = sizeof (struct echo_lock)
244 .ckd_cache = &echo_object_kmem,
245 .ckd_name = "echo_object_kmem",
246 .ckd_size = sizeof (struct echo_object)
249 .ckd_cache = &echo_thread_kmem,
250 .ckd_name = "echo_thread_kmem",
251 .ckd_size = sizeof (struct echo_thread_info)
254 .ckd_cache = &echo_session_kmem,
255 .ckd_name = "echo_session_kmem",
256 .ckd_size = sizeof (struct echo_session_info)
260 .ckd_cache = &echo_req_kmem,
261 .ckd_name = "echo_req_kmem",
262 .ckd_size = sizeof (struct echo_req)
270 /** \defgroup echo_page Page operations
272 * Echo page operations.
276 static cfs_page_t *echo_page_vmpage(const struct lu_env *env,
277 const struct cl_page_slice *slice)
279 return cl2echo_page(slice)->ep_vmpage;
282 static int echo_page_own(const struct lu_env *env,
283 const struct cl_page_slice *slice,
284 struct cl_io *io, int nonblock)
286 struct echo_page *ep = cl2echo_page(slice);
289 cfs_mutex_lock(&ep->ep_lock);
290 else if (!cfs_mutex_trylock(&ep->ep_lock))
295 static void echo_page_disown(const struct lu_env *env,
296 const struct cl_page_slice *slice,
299 struct echo_page *ep = cl2echo_page(slice);
301 LASSERT(cfs_mutex_is_locked(&ep->ep_lock));
302 cfs_mutex_unlock(&ep->ep_lock);
305 static void echo_page_discard(const struct lu_env *env,
306 const struct cl_page_slice *slice,
307 struct cl_io *unused)
309 cl_page_delete(env, slice->cpl_page);
312 static int echo_page_is_vmlocked(const struct lu_env *env,
313 const struct cl_page_slice *slice)
315 if (cfs_mutex_is_locked(&cl2echo_page(slice)->ep_lock))
320 static void echo_page_completion(const struct lu_env *env,
321 const struct cl_page_slice *slice,
324 LASSERT(slice->cpl_page->cp_sync_io != NULL);
327 static void echo_page_fini(const struct lu_env *env,
328 struct cl_page_slice *slice)
330 struct echo_page *ep = cl2echo_page(slice);
331 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
332 cfs_page_t *vmpage = ep->ep_vmpage;
335 cfs_atomic_dec(&eco->eo_npages);
336 page_cache_release(vmpage);
337 OBD_SLAB_FREE_PTR(ep, echo_page_kmem);
341 static int echo_page_prep(const struct lu_env *env,
342 const struct cl_page_slice *slice,
343 struct cl_io *unused)
348 static int echo_page_print(const struct lu_env *env,
349 const struct cl_page_slice *slice,
350 void *cookie, lu_printer_t printer)
352 struct echo_page *ep = cl2echo_page(slice);
354 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
355 ep, cfs_mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
359 static const struct cl_page_operations echo_page_ops = {
360 .cpo_own = echo_page_own,
361 .cpo_disown = echo_page_disown,
362 .cpo_discard = echo_page_discard,
363 .cpo_vmpage = echo_page_vmpage,
364 .cpo_fini = echo_page_fini,
365 .cpo_print = echo_page_print,
366 .cpo_is_vmlocked = echo_page_is_vmlocked,
369 .cpo_prep = echo_page_prep,
370 .cpo_completion = echo_page_completion,
373 .cpo_prep = echo_page_prep,
374 .cpo_completion = echo_page_completion,
380 /** \defgroup echo_lock Locking
382 * echo lock operations
386 static void echo_lock_fini(const struct lu_env *env,
387 struct cl_lock_slice *slice)
389 struct echo_lock *ecl = cl2echo_lock(slice);
391 LASSERT(cfs_list_empty(&ecl->el_chain));
392 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
395 static void echo_lock_delete(const struct lu_env *env,
396 const struct cl_lock_slice *slice)
398 struct echo_lock *ecl = cl2echo_lock(slice);
400 LASSERT(cfs_list_empty(&ecl->el_chain));
403 static int echo_lock_fits_into(const struct lu_env *env,
404 const struct cl_lock_slice *slice,
405 const struct cl_lock_descr *need,
406 const struct cl_io *unused)
411 static struct cl_lock_operations echo_lock_ops = {
412 .clo_fini = echo_lock_fini,
413 .clo_delete = echo_lock_delete,
414 .clo_fits_into = echo_lock_fits_into
419 /** \defgroup echo_cl_ops cl_object operations
421 * operations for cl_object
425 static struct cl_page *echo_page_init(const struct lu_env *env,
426 struct cl_object *obj,
427 struct cl_page *page, cfs_page_t *vmpage)
429 struct echo_page *ep;
432 OBD_SLAB_ALLOC_PTR_GFP(ep, echo_page_kmem, CFS_ALLOC_IO);
434 struct echo_object *eco = cl2echo_obj(obj);
435 ep->ep_vmpage = vmpage;
436 page_cache_get(vmpage);
437 cfs_mutex_init(&ep->ep_lock);
438 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
439 cfs_atomic_inc(&eco->eo_npages);
441 RETURN(ERR_PTR(ep ? 0 : -ENOMEM));
444 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
450 static int echo_lock_init(const struct lu_env *env,
451 struct cl_object *obj, struct cl_lock *lock,
452 const struct cl_io *unused)
454 struct echo_lock *el;
457 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, CFS_ALLOC_IO);
459 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
460 el->el_object = cl2echo_obj(obj);
461 CFS_INIT_LIST_HEAD(&el->el_chain);
462 cfs_atomic_set(&el->el_refcount, 0);
464 RETURN(el == NULL ? -ENOMEM : 0);
467 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
468 const struct cl_object_conf *conf)
473 static const struct cl_object_operations echo_cl_obj_ops = {
474 .coo_page_init = echo_page_init,
475 .coo_lock_init = echo_lock_init,
476 .coo_io_init = echo_io_init,
477 .coo_conf_set = echo_conf_set
479 /** @} echo_cl_ops */
481 /** \defgroup echo_lu_ops lu_object operations
483 * operations for echo lu object.
487 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
488 const struct lu_object_conf *conf)
490 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
491 struct echo_client_obd *ec = ed->ed_ec;
492 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
496 struct lu_object *below;
497 struct lu_device *under;
500 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
504 lu_object_add(obj, below);
507 if (!ed->ed_next_ismd) {
508 const struct cl_object_conf *cconf = lu2cl_conf(conf);
509 struct echo_object_conf *econf = cl2echo_conf(cconf);
511 LASSERT(econf->eoc_md);
512 eco->eo_lsm = *econf->eoc_md;
513 /* clear the lsm pointer so that it won't get freed. */
514 *econf->eoc_md = NULL;
520 cfs_atomic_set(&eco->eo_npages, 0);
522 cfs_spin_lock(&ec->ec_lock);
523 cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
524 cfs_spin_unlock(&ec->ec_lock);
529 /* taken from osc_unpackmd() */
530 static int echo_alloc_memmd(struct echo_device *ed,
531 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_alloc_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_ALLOC(*lsmp, lsm_size);
548 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
549 if ((*lsmp)->lsm_oinfo[0] == NULL) {
550 OBD_FREE(*lsmp, lsm_size);
554 loi_init((*lsmp)->lsm_oinfo[0]);
555 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
560 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
566 /* If export is lov/osc then use their obd method */
567 if (ed->ed_next != NULL)
568 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
569 /* OFD has no unpackmd method, do everything here */
570 lsm_size = lov_stripe_md_size(1);
572 LASSERT(*lsmp != NULL);
573 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
574 OBD_FREE(*lsmp, lsm_size);
579 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
581 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
582 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
585 LASSERT(cfs_atomic_read(&eco->eo_npages) == 0);
587 cfs_spin_lock(&ec->ec_lock);
588 cfs_list_del_init(&eco->eo_obj_chain);
589 cfs_spin_unlock(&ec->ec_lock);
592 lu_object_header_fini(obj->lo_header);
595 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
596 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
600 static int echo_object_print(const struct lu_env *env, void *cookie,
601 lu_printer_t p, const struct lu_object *o)
603 struct echo_object *obj = cl2echo_obj(lu2cl(o));
605 return (*p)(env, cookie, "echoclient-object@%p", obj);
608 static const struct lu_object_operations echo_lu_obj_ops = {
609 .loo_object_init = echo_object_init,
610 .loo_object_delete = NULL,
611 .loo_object_release = NULL,
612 .loo_object_free = echo_object_free,
613 .loo_object_print = echo_object_print,
614 .loo_object_invariant = NULL
616 /** @} echo_lu_ops */
618 /** \defgroup echo_lu_dev_ops lu_device operations
620 * Operations for echo lu device.
624 static struct lu_object *echo_object_alloc(const struct lu_env *env,
625 const struct lu_object_header *hdr,
626 struct lu_device *dev)
628 struct echo_object *eco;
629 struct lu_object *obj = NULL;
632 /* we're the top dev. */
633 LASSERT(hdr == NULL);
634 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, CFS_ALLOC_IO);
636 struct cl_object_header *hdr = &eco->eo_hdr;
638 obj = &echo_obj2cl(eco)->co_lu;
639 cl_object_header_init(hdr);
640 lu_object_init(obj, &hdr->coh_lu, dev);
641 lu_object_add_top(&hdr->coh_lu, obj);
643 eco->eo_cl.co_ops = &echo_cl_obj_ops;
644 obj->lo_ops = &echo_lu_obj_ops;
649 static struct lu_device_operations echo_device_lu_ops = {
650 .ldo_object_alloc = echo_object_alloc,
653 /** @} echo_lu_dev_ops */
655 static struct cl_device_operations echo_device_cl_ops = {
658 /** \defgroup echo_init Setup and teardown
660 * Init and fini functions for echo client.
664 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
666 struct cl_site *site = &ed->ed_site_myself;
669 /* initialize site */
670 rc = cl_site_init(site, &ed->ed_cl);
672 CERROR("Cannot initilize site for echo client(%d)\n", rc);
676 rc = lu_site_init_finish(&site->cs_lu);
684 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
687 if (!ed->ed_next_ismd)
688 cl_site_fini(ed->ed_site);
693 static void *echo_thread_key_init(const struct lu_context *ctx,
694 struct lu_context_key *key)
696 struct echo_thread_info *info;
698 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, CFS_ALLOC_IO);
700 info = ERR_PTR(-ENOMEM);
704 static void echo_thread_key_fini(const struct lu_context *ctx,
705 struct lu_context_key *key, void *data)
707 struct echo_thread_info *info = data;
708 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
711 static void echo_thread_key_exit(const struct lu_context *ctx,
712 struct lu_context_key *key, void *data)
716 static struct lu_context_key echo_thread_key = {
717 .lct_tags = LCT_CL_THREAD,
718 .lct_init = echo_thread_key_init,
719 .lct_fini = echo_thread_key_fini,
720 .lct_exit = echo_thread_key_exit
723 static void *echo_session_key_init(const struct lu_context *ctx,
724 struct lu_context_key *key)
726 struct echo_session_info *session;
728 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, CFS_ALLOC_IO);
730 session = ERR_PTR(-ENOMEM);
734 static void echo_session_key_fini(const struct lu_context *ctx,
735 struct lu_context_key *key, void *data)
737 struct echo_session_info *session = data;
738 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
741 static void echo_session_key_exit(const struct lu_context *ctx,
742 struct lu_context_key *key, void *data)
746 static struct lu_context_key echo_session_key = {
747 .lct_tags = LCT_SESSION,
748 .lct_init = echo_session_key_init,
749 .lct_fini = echo_session_key_fini,
750 .lct_exit = echo_session_key_exit
753 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
755 #define ECHO_SEQ_WIDTH 0xffffffff
756 static int echo_fid_init(struct echo_device *ed, char *obd_name,
763 OBD_ALLOC_PTR(ed->ed_cl_seq);
764 if (ed->ed_cl_seq == NULL)
767 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
769 GOTO(out_free_seq, rc = -ENOMEM);
771 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
773 /* Init client side sequence-manager */
774 rc = seq_client_init(ed->ed_cl_seq, NULL,
776 prefix, ms->ms_server_seq);
777 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
778 OBD_FREE(prefix, MAX_OBD_NAME + 5);
780 GOTO(out_free_seq, rc);
785 OBD_FREE_PTR(ed->ed_cl_seq);
786 ed->ed_cl_seq = NULL;
790 static int echo_fid_fini(struct obd_device *obddev)
792 struct echo_device *ed = obd2echo_dev(obddev);
795 if (ed->ed_cl_seq != NULL) {
796 seq_client_fini(ed->ed_cl_seq);
797 OBD_FREE_PTR(ed->ed_cl_seq);
798 ed->ed_cl_seq = NULL;
804 static struct lu_device *echo_device_alloc(const struct lu_env *env,
805 struct lu_device_type *t,
806 struct lustre_cfg *cfg)
808 struct lu_device *next;
809 struct echo_device *ed;
810 struct cl_device *cd;
811 struct obd_device *obd = NULL; /* to keep compiler happy */
812 struct obd_device *tgt;
813 const char *tgt_type_name;
820 GOTO(out, rc = -ENOMEM);
824 rc = cl_device_init(cd, t);
828 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
829 cd->cd_ops = &echo_device_cl_ops;
832 obd = class_name2obd(lustre_cfg_string(cfg, 0));
833 LASSERT(obd != NULL);
834 LASSERT(env != NULL);
836 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
838 CERROR("Can not find tgt device %s\n",
839 lustre_cfg_string(cfg, 1));
840 GOTO(out, rc = -ENODEV);
843 next = tgt->obd_lu_dev;
844 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
845 ed->ed_next_ismd = 1;
847 ed->ed_next_ismd = 0;
848 rc = echo_site_init(env, ed);
854 rc = echo_client_setup(env, obd, cfg);
858 ed->ed_ec = &obd->u.echo_client;
861 if (ed->ed_next_ismd) {
862 /* Suppose to connect to some Metadata layer */
864 struct lu_device *ld;
868 CERROR("%s is not lu device type!\n",
869 lustre_cfg_string(cfg, 1));
870 GOTO(out, rc = -EINVAL);
873 tgt_type_name = lustre_cfg_string(cfg, 2);
874 if (!tgt_type_name) {
875 CERROR("%s no type name for echo %s setup\n",
876 lustre_cfg_string(cfg, 1),
877 tgt->obd_type->typ_name);
878 GOTO(out, rc = -EINVAL);
883 cfs_spin_lock(&ls->ls_ld_lock);
884 cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
885 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
890 cfs_spin_unlock(&ls->ls_ld_lock);
893 CERROR("%s is not lu device type!\n",
894 lustre_cfg_string(cfg, 1));
895 GOTO(out, rc = -EINVAL);
899 /* For MD echo client, it will use the site in MDS stack */
900 ed->ed_site_myself.cs_lu = *ls;
901 ed->ed_site = &ed->ed_site_myself;
902 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
903 rc = echo_fid_init(ed, obd->obd_name, lu_site2md(ls));
905 CERROR("echo fid init error %d\n", rc);
909 /* if echo client is to be stacked upon ost device, the next is
910 * NULL since ost is not a clio device so far */
911 if (next != NULL && !lu_device_is_cl(next))
914 tgt_type_name = tgt->obd_type->typ_name;
916 LASSERT(next != NULL);
917 if (next->ld_site != NULL)
918 GOTO(out, rc = -EBUSY);
920 next->ld_site = &ed->ed_site->cs_lu;
921 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
922 next->ld_type->ldt_name,
927 /* Tricky case, I have to determine the obd type since
928 * CLIO uses the different parameters to initialize
929 * objects for lov & osc. */
930 if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
931 ed->ed_next_islov = 1;
933 LASSERT(strcmp(tgt_type_name,
934 LUSTRE_OSC_NAME) == 0);
936 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
940 RETURN(&cd->cd_lu_dev);
945 rc2 = echo_client_cleanup(obd);
947 CERROR("Cleanup obd device %s error(%d)\n",
952 echo_site_fini(env, ed);
954 cl_device_fini(&ed->ed_cl);
964 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
965 const char *name, struct lu_device *next)
971 static struct lu_device *echo_device_fini(const struct lu_env *env,
974 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
975 struct lu_device *next = ed->ed_next;
977 while (next && !ed->ed_next_ismd)
978 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
982 static void echo_lock_release(const struct lu_env *env,
983 struct echo_lock *ecl,
986 struct cl_lock *clk = echo_lock2cl(ecl);
990 cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
992 cl_lock_mutex_get(env, clk);
993 cl_lock_cancel(env, clk);
994 cl_lock_delete(env, clk);
995 cl_lock_mutex_put(env, clk);
997 cl_lock_put(env, clk);
1000 static struct lu_device *echo_device_free(const struct lu_env *env,
1001 struct lu_device *d)
1003 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
1004 struct echo_client_obd *ec = ed->ed_ec;
1005 struct echo_object *eco;
1006 struct lu_device *next = ed->ed_next;
1008 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1011 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1013 /* check if there are objects still alive.
1014 * It shouldn't have any object because lu_site_purge would cleanup
1015 * all of cached objects. Anyway, probably the echo device is being
1016 * parallelly accessed.
1018 cfs_spin_lock(&ec->ec_lock);
1019 cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1020 eco->eo_deleted = 1;
1021 cfs_spin_unlock(&ec->ec_lock);
1024 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1027 "Waiting for the reference of echo object to be dropped\n");
1029 /* Wait for the last reference to be dropped. */
1030 cfs_spin_lock(&ec->ec_lock);
1031 while (!cfs_list_empty(&ec->ec_objects)) {
1032 cfs_spin_unlock(&ec->ec_lock);
1033 CERROR("echo_client still has objects at cleanup time, "
1034 "wait for 1 second\n");
1035 cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT,
1036 cfs_time_seconds(1));
1037 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1038 cfs_spin_lock(&ec->ec_lock);
1040 cfs_spin_unlock(&ec->ec_lock);
1042 LASSERT(cfs_list_empty(&ec->ec_locks));
1044 CDEBUG(D_INFO, "No object exists, exiting...\n");
1046 echo_client_cleanup(d->ld_obd);
1047 echo_fid_fini(d->ld_obd);
1048 while (next && !ed->ed_next_ismd)
1049 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1051 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1052 echo_site_fini(env, ed);
1053 cl_device_fini(&ed->ed_cl);
1059 static const struct lu_device_type_operations echo_device_type_ops = {
1060 .ldto_init = echo_type_init,
1061 .ldto_fini = echo_type_fini,
1063 .ldto_start = echo_type_start,
1064 .ldto_stop = echo_type_stop,
1066 .ldto_device_alloc = echo_device_alloc,
1067 .ldto_device_free = echo_device_free,
1068 .ldto_device_init = echo_device_init,
1069 .ldto_device_fini = echo_device_fini
1072 static struct lu_device_type echo_device_type = {
1073 .ldt_tags = LU_DEVICE_CL,
1074 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1075 .ldt_ops = &echo_device_type_ops,
1076 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1080 /** \defgroup echo_exports Exported operations
1082 * exporting functions to echo client
1087 /* Interfaces to echo client obd device */
1088 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1089 struct lov_stripe_md **lsmp)
1092 struct echo_thread_info *info;
1093 struct echo_object_conf *conf;
1094 struct lov_stripe_md *lsm;
1095 struct echo_object *eco;
1096 struct cl_object *obj;
1104 LASSERT(lsm->lsm_object_id);
1106 /* Never return an object if the obd is to be freed. */
1107 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1108 RETURN(ERR_PTR(-ENODEV));
1110 env = cl_env_get(&refcheck);
1112 RETURN((void *)env);
1114 info = echo_env_info(env);
1115 conf = &info->eti_conf;
1117 if (!d->ed_next_islov) {
1118 struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1119 LASSERT(oinfo != NULL);
1120 oinfo->loi_id = lsm->lsm_object_id;
1121 oinfo->loi_seq = lsm->lsm_object_seq;
1122 conf->eoc_cl.u.coc_oinfo = oinfo;
1124 struct lustre_md *md;
1126 memset(md, 0, sizeof *md);
1128 conf->eoc_cl.u.coc_md = md;
1131 conf->eoc_md = lsmp;
1133 fid = &info->eti_fid;
1136 /* In the function below, .hs_keycmp resolves to
1137 * lu_obj_hop_keycmp() */
1138 /* coverity[overrun-buffer-val] */
1139 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1141 GOTO(out, eco = (void*)obj);
1143 eco = cl2echo_obj(obj);
1144 if (eco->eo_deleted) {
1145 cl_object_put(env, obj);
1146 eco = ERR_PTR(-EAGAIN);
1150 cl_env_put(env, &refcheck);
1154 static int cl_echo_object_put(struct echo_object *eco)
1157 struct cl_object *obj = echo_obj2cl(eco);
1161 env = cl_env_get(&refcheck);
1163 RETURN(PTR_ERR(env));
1165 /* an external function to kill an object? */
1166 if (eco->eo_deleted) {
1167 struct lu_object_header *loh = obj->co_lu.lo_header;
1168 LASSERT(&eco->eo_hdr == luh2coh(loh));
1169 cfs_set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1172 cl_object_put(env, obj);
1173 cl_env_put(env, &refcheck);
1177 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1178 obd_off start, obd_off end, int mode,
1179 __u64 *cookie , __u32 enqflags)
1182 struct cl_lock *lck;
1183 struct cl_object *obj;
1184 struct cl_lock_descr *descr;
1185 struct echo_thread_info *info;
1189 info = echo_env_info(env);
1191 descr = &info->eti_descr;
1192 obj = echo_obj2cl(eco);
1194 descr->cld_obj = obj;
1195 descr->cld_start = cl_index(obj, start);
1196 descr->cld_end = cl_index(obj, end);
1197 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1198 descr->cld_enq_flags = enqflags;
1201 lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1203 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1204 struct echo_lock *el;
1206 rc = cl_wait(env, lck);
1208 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1209 cfs_spin_lock(&ec->ec_lock);
1210 if (cfs_list_empty(&el->el_chain)) {
1211 cfs_list_add(&el->el_chain, &ec->ec_locks);
1212 el->el_cookie = ++ec->ec_unique;
1214 cfs_atomic_inc(&el->el_refcount);
1215 *cookie = el->el_cookie;
1216 cfs_spin_unlock(&ec->ec_lock);
1218 cl_lock_release(env, lck, "ec enqueue", cfs_current());
1223 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1224 int mode, __u64 *cookie)
1226 struct echo_thread_info *info;
1233 env = cl_env_get(&refcheck);
1235 RETURN(PTR_ERR(env));
1237 info = echo_env_info(env);
1240 io->ci_ignore_layout = 1;
1241 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1244 LASSERT(result == 0);
1246 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1247 cl_io_fini(env, io);
1251 cl_env_put(env, &refcheck);
1255 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1258 struct echo_client_obd *ec = ed->ed_ec;
1259 struct echo_lock *ecl = NULL;
1261 int found = 0, still_used = 0;
1264 LASSERT(ec != NULL);
1265 cfs_spin_lock (&ec->ec_lock);
1266 cfs_list_for_each (el, &ec->ec_locks) {
1267 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1268 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1269 found = (ecl->el_cookie == cookie);
1271 if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1272 cfs_list_del_init(&ecl->el_chain);
1278 cfs_spin_unlock (&ec->ec_lock);
1283 echo_lock_release(env, ecl, still_used);
1287 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1294 env = cl_env_get(&refcheck);
1296 RETURN(PTR_ERR(env));
1298 rc = cl_echo_cancel0(env, ed, cookie);
1300 cl_env_put(env, &refcheck);
1304 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1305 enum cl_req_type unused, struct cl_2queue *queue)
1307 struct cl_page *clp;
1308 struct cl_page *temp;
1312 cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1314 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1317 result = result ?: rc;
1322 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1323 cfs_page_t **pages, int npages, int async)
1326 struct echo_thread_info *info;
1327 struct cl_object *obj = echo_obj2cl(eco);
1328 struct echo_device *ed = eco->eo_dev;
1329 struct cl_2queue *queue;
1331 struct cl_page *clp;
1332 struct lustre_handle lh = { 0 };
1333 int page_size = cl_page_size(obj);
1339 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1340 LASSERT(ed->ed_next != NULL);
1341 env = cl_env_get(&refcheck);
1343 RETURN(PTR_ERR(env));
1345 info = echo_env_info(env);
1347 queue = &info->eti_queue;
1349 cl_2queue_init(queue);
1351 io->ci_ignore_layout = 1;
1352 rc = cl_io_init(env, io, CIT_MISC, obj);
1358 rc = cl_echo_enqueue0(env, eco, offset,
1359 offset + npages * CFS_PAGE_SIZE - 1,
1360 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1363 GOTO(error_lock, rc);
1365 for (i = 0; i < npages; i++) {
1367 clp = cl_page_find(env, obj, cl_index(obj, offset),
1368 pages[i], CPT_TRANSIENT);
1373 LASSERT(clp->cp_type == CPT_TRANSIENT);
1375 rc = cl_page_own(env, io, clp);
1377 LASSERT(clp->cp_state == CPS_FREEING);
1378 cl_page_put(env, clp);
1382 cl_2queue_add(queue, clp);
1384 /* drop the reference count for cl_page_find, so that the page
1385 * will be freed in cl_2queue_fini. */
1386 cl_page_put(env, clp);
1387 cl_page_clip(env, clp, 0, page_size);
1389 offset += page_size;
1393 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1395 async = async && (typ == CRT_WRITE);
1397 rc = cl_echo_async_brw(env, io, typ, queue);
1399 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1400 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1401 async ? "async" : "sync", rc);
1404 cl_echo_cancel0(env, ed, lh.cookie);
1407 cl_2queue_discard(env, io, queue);
1408 cl_2queue_disown(env, io, queue);
1409 cl_2queue_fini(env, queue);
1410 cl_io_fini(env, io);
1412 cl_env_put(env, &refcheck);
1415 /** @} echo_exports */
1418 static obd_id last_object_id;
1421 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1423 struct lov_stripe_md *ulsm = _ulsm;
1426 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1430 if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm)))
1433 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1434 if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1435 sizeof(lsm->lsm_oinfo[0])))
1442 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1443 void *ulsm, int ulsm_nob)
1445 struct echo_client_obd *ec = ed->ed_ec;
1448 if (ulsm_nob < sizeof (*lsm))
1451 if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm)))
1454 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1455 lsm->lsm_magic != LOV_MAGIC ||
1456 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1457 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1461 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1462 if (cfs_copy_from_user(lsm->lsm_oinfo[i],
1463 ((struct lov_stripe_md *)ulsm)-> \
1465 sizeof(lsm->lsm_oinfo[0])))
1471 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1474 sprintf(name, LPU64, id);
1475 lname->ln_name = name;
1476 lname->ln_namelen = strlen(name);
1479 /* similar to mdt_attr_get_complex */
1480 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1483 struct echo_thread_info *info = echo_env_info(env);
1488 LASSERT(ma->ma_lmm_size > 0);
1490 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1494 /* big_lmm may need to be grown */
1495 if (info->eti_big_lmmsize < rc) {
1496 int size = size_roundup_power2(rc);
1498 if (info->eti_big_lmmsize > 0) {
1499 /* free old buffer */
1500 LASSERT(info->eti_big_lmm);
1501 OBD_FREE_LARGE(info->eti_big_lmm,
1502 info->eti_big_lmmsize);
1503 info->eti_big_lmm = NULL;
1504 info->eti_big_lmmsize = 0;
1507 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1508 if (info->eti_big_lmm == NULL)
1510 info->eti_big_lmmsize = size;
1512 LASSERT(info->eti_big_lmmsize >= rc);
1514 info->eti_buf.lb_buf = info->eti_big_lmm;
1515 info->eti_buf.lb_len = info->eti_big_lmmsize;
1516 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1520 ma->ma_valid |= MA_LOV;
1521 ma->ma_lmm = info->eti_big_lmm;
1522 ma->ma_lmm_size = rc;
1527 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1530 struct echo_thread_info *info = echo_env_info(env);
1531 struct lu_buf *buf = &info->eti_buf;
1532 cfs_umode_t mode = lu_object_attr(&next->mo_lu);
1533 int need = ma->ma_need;
1540 if (need & MA_INODE) {
1541 ma->ma_need = MA_INODE;
1542 rc = mo_attr_get(env, next, ma);
1545 ma->ma_valid |= MA_INODE;
1548 if (need & MA_LOV) {
1549 if (S_ISREG(mode) || S_ISDIR(mode)) {
1550 LASSERT(ma->ma_lmm_size > 0);
1551 buf->lb_buf = ma->ma_lmm;
1552 buf->lb_len = ma->ma_lmm_size;
1553 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1555 ma->ma_lmm_size = rc2;
1556 ma->ma_valid |= MA_LOV;
1557 } else if (rc2 == -ENODATA) {
1559 ma->ma_lmm_size = 0;
1560 } else if (rc2 == -ERANGE) {
1561 rc2 = echo_big_lmm_get(env, next, ma);
1563 GOTO(out, rc = rc2);
1565 GOTO(out, rc = rc2);
1570 #ifdef CONFIG_FS_POSIX_ACL
1571 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1572 buf->lb_buf = ma->ma_acl;
1573 buf->lb_len = ma->ma_acl_size;
1574 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1576 ma->ma_acl_size = rc2;
1577 ma->ma_valid |= MA_ACL_DEF;
1578 } else if (rc2 == -ENODATA) {
1580 ma->ma_acl_size = 0;
1582 GOTO(out, rc = rc2);
1588 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1589 rc, ma->ma_valid, ma->ma_lmm);
1594 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1595 struct md_object *parent, struct lu_fid *fid,
1596 struct lu_name *lname, struct md_op_spec *spec,
1599 struct lu_object *ec_child, *child;
1600 struct lu_device *ld = ed->ed_next;
1601 struct echo_thread_info *info = echo_env_info(env);
1602 struct lu_fid *fid2 = &info->eti_fid2;
1603 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1608 rc = mdo_lookup(env, parent, lname, fid2, spec);
1611 else if (rc != -ENOENT)
1614 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1616 if (IS_ERR(ec_child)) {
1617 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1619 RETURN(PTR_ERR(ec_child));
1622 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1623 if (child == NULL) {
1624 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1625 GOTO(out_put, rc = -EINVAL);
1628 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1629 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1632 * Do not perform lookup sanity check. We know that name does not exist.
1634 spec->sp_cr_lookup = 0;
1635 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1637 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1640 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1641 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1644 lu_object_put(env, ec_child);
1648 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1651 struct echo_thread_info *info = echo_env_info(env);
1653 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1654 ma->ma_lmm = (void *)&info->eti_lmm;
1655 ma->ma_lmm_size = sizeof(info->eti_lmm);
1657 LASSERT(info->eti_big_lmmsize);
1658 ma->ma_lmm = info->eti_big_lmm;
1659 ma->ma_lmm_size = info->eti_big_lmmsize;
1665 static int echo_create_md_object(const struct lu_env *env,
1666 struct echo_device *ed,
1667 struct lu_object *ec_parent,
1669 char *name, int namelen,
1670 __u64 id, __u32 mode, int count,
1671 int stripe_count, int stripe_offset)
1673 struct lu_object *parent;
1674 struct echo_thread_info *info = echo_env_info(env);
1675 struct lu_name *lname = &info->eti_lname;
1676 struct md_op_spec *spec = &info->eti_spec;
1677 struct md_attr *ma = &info->eti_ma;
1678 struct lu_device *ld = ed->ed_next;
1684 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1688 memset(ma, 0, sizeof(*ma));
1689 memset(spec, 0, sizeof(*spec));
1690 if (stripe_count != 0) {
1691 spec->sp_cr_flags |= FMODE_WRITE;
1692 echo_set_lmm_size(env, ld, ma);
1693 if (stripe_count != -1) {
1694 struct lov_user_md_v3 *lum = &info->eti_lum;
1696 lum->lmm_magic = LOV_USER_MAGIC_V3;
1697 lum->lmm_stripe_count = stripe_count;
1698 lum->lmm_stripe_offset = stripe_offset;
1699 lum->lmm_pattern = 0;
1700 spec->u.sp_ea.eadata = lum;
1701 spec->u.sp_ea.eadatalen = sizeof(*lum);
1702 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1706 ma->ma_attr.la_mode = mode;
1707 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1708 ma->ma_attr.la_ctime = cfs_time_current_64();
1711 lname->ln_name = name;
1712 lname->ln_namelen = namelen;
1713 /* If name is specified, only create one object by name */
1714 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1719 /* Create multiple object sequenced by id */
1720 for (i = 0; i < count; i++) {
1721 char *tmp_name = info->eti_name;
1723 echo_md_build_name(lname, tmp_name, id);
1725 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1728 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1739 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1740 struct echo_device *ed,
1741 struct md_object *parent,
1742 struct lu_name *lname)
1744 struct echo_thread_info *info = echo_env_info(env);
1745 struct lu_fid *fid = &info->eti_fid;
1746 struct lu_object *child;
1750 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1752 rc = mdo_lookup(env, parent, lname, fid, NULL);
1754 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1755 RETURN(ERR_PTR(rc));
1758 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1763 static int echo_setattr_object(const struct lu_env *env,
1764 struct echo_device *ed,
1765 struct lu_object *ec_parent,
1766 __u64 id, int count)
1768 struct lu_object *parent;
1769 struct echo_thread_info *info = echo_env_info(env);
1770 struct lu_name *lname = &info->eti_lname;
1771 char *name = info->eti_name;
1772 struct lu_device *ld = ed->ed_next;
1773 struct lu_buf *buf = &info->eti_buf;
1779 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1783 for (i = 0; i < count; i++) {
1784 struct lu_object *ec_child, *child;
1786 echo_md_build_name(lname, name, id);
1788 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1789 if (IS_ERR(ec_child)) {
1790 CERROR("Can't find child %s: rc = %ld\n",
1791 lname->ln_name, PTR_ERR(ec_child));
1792 RETURN(PTR_ERR(ec_child));
1795 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1796 if (child == NULL) {
1797 CERROR("Can not locate the child %s\n", lname->ln_name);
1798 lu_object_put(env, ec_child);
1803 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1804 PFID(lu_object_fid(child)));
1806 buf->lb_buf = info->eti_xattr_buf;
1807 buf->lb_len = sizeof(info->eti_xattr_buf);
1809 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1810 rc = mo_xattr_set(env, lu2md(child), buf, name,
1813 CERROR("Can not setattr child "DFID": rc = %d\n",
1814 PFID(lu_object_fid(child)), rc);
1815 lu_object_put(env, ec_child);
1818 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1819 PFID(lu_object_fid(child)));
1821 lu_object_put(env, ec_child);
1826 static int echo_getattr_object(const struct lu_env *env,
1827 struct echo_device *ed,
1828 struct lu_object *ec_parent,
1829 __u64 id, int count)
1831 struct lu_object *parent;
1832 struct echo_thread_info *info = echo_env_info(env);
1833 struct lu_name *lname = &info->eti_lname;
1834 char *name = info->eti_name;
1835 struct md_attr *ma = &info->eti_ma;
1836 struct lu_device *ld = ed->ed_next;
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 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1908 /*prepare the requests*/
1909 for (i = 0; i < count; i++) {
1910 echo_md_build_name(lname, name, id);
1912 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1913 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1915 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1917 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1920 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1921 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1928 static int echo_md_destroy_internal(const struct lu_env *env,
1929 struct echo_device *ed,
1930 struct md_object *parent,
1931 struct lu_name *lname,
1934 struct lu_device *ld = ed->ed_next;
1935 struct lu_object *ec_child;
1936 struct lu_object *child;
1941 ec_child = echo_md_lookup(env, ed, parent, lname);
1942 if (IS_ERR(ec_child)) {
1943 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1945 RETURN(PTR_ERR(ec_child));
1948 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1949 if (child == NULL) {
1950 CERROR("Can not locate the child %s\n", lname->ln_name);
1951 GOTO(out_put, rc = -EINVAL);
1954 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1955 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1957 rc = mdo_unlink(env, parent, lu2md(child), lname, ma);
1959 CERROR("Can not unlink child %s: rc = %d\n",
1960 lname->ln_name, rc);
1963 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1964 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1966 lu_object_put(env, ec_child);
1970 static int echo_destroy_object(const struct lu_env *env,
1971 struct echo_device *ed,
1972 struct lu_object *ec_parent,
1973 char *name, int namelen,
1974 __u64 id, __u32 mode,
1977 struct echo_thread_info *info = echo_env_info(env);
1978 struct lu_name *lname = &info->eti_lname;
1979 struct md_attr *ma = &info->eti_ma;
1980 struct lu_device *ld = ed->ed_next;
1981 struct lu_object *parent;
1986 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1990 memset(ma, 0, sizeof(*ma));
1991 ma->ma_attr.la_mode = mode;
1992 ma->ma_attr.la_valid = LA_CTIME;
1993 ma->ma_attr.la_ctime = cfs_time_current_64();
1994 ma->ma_need = MA_INODE;
1998 lname->ln_name = name;
1999 lname->ln_namelen = namelen;
2000 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2005 /*prepare the requests*/
2006 for (i = 0; i < count; i++) {
2007 char *tmp_name = info->eti_name;
2010 echo_md_build_name(lname, tmp_name, id);
2012 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2015 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2024 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2025 struct echo_device *ed, char *path,
2028 struct lu_device *ld = ed->ed_next;
2029 struct md_device *md = lu2md_dev(ld);
2030 struct echo_thread_info *info = echo_env_info(env);
2031 struct lu_fid *fid = &info->eti_fid;
2032 struct lu_name *lname = &info->eti_lname;
2033 struct lu_object *parent = NULL;
2034 struct lu_object *child = NULL;
2038 /*Only support MDD layer right now*/
2039 rc = md->md_ops->mdo_root_get(env, md, fid);
2041 CERROR("get root error: rc = %d\n", rc);
2042 RETURN(ERR_PTR(rc));
2045 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2046 if (IS_ERR(parent)) {
2047 CERROR("Can not find the parent "DFID": rc = %ld\n",
2048 PFID(fid), PTR_ERR(parent));
2053 struct lu_object *ld_parent;
2056 e = strsep(&path, "/");
2061 if (!path || path[0] == '\0')
2067 lname->ln_namelen = strlen(e);
2069 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2070 if (ld_parent == NULL) {
2071 lu_object_put(env, parent);
2076 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2077 lu_object_put(env, parent);
2078 if (IS_ERR(child)) {
2079 rc = (int)PTR_ERR(child);
2080 CERROR("lookup %s under parent "DFID": rc = %d\n",
2081 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2088 RETURN(ERR_PTR(rc));
2093 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2094 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2095 static int echo_md_handler(struct echo_device *ed, int command,
2096 char *path, int path_len, int id, int count,
2097 struct obd_ioctl_data *data)
2099 struct echo_thread_info *info;
2100 struct lu_device *ld = ed->ed_next;
2103 struct lu_object *parent;
2105 int namelen = data->ioc_plen2;
2110 CERROR("MD echo client is not being initialized properly\n");
2114 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2115 CERROR("Only support MDD layer right now!\n");
2119 env = cl_env_get(&refcheck);
2121 RETURN(PTR_ERR(env));
2123 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2127 /* init big_lmm buffer */
2128 info = echo_env_info(env);
2129 LASSERT(info->eti_big_lmm == NULL);
2130 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2131 if (info->eti_big_lmm == NULL)
2132 GOTO(out_env, rc = -ENOMEM);
2133 info->eti_big_lmmsize = MIN_MD_SIZE;
2135 parent = echo_resolve_path(env, ed, path, path_len);
2136 if (IS_ERR(parent)) {
2137 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2139 GOTO(out_free, rc = PTR_ERR(parent));
2143 OBD_ALLOC(name, namelen + 1);
2145 GOTO(out_put, rc = -ENOMEM);
2146 if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen))
2147 GOTO(out_name, rc = -EFAULT);
2151 case ECHO_MD_CREATE:
2152 case ECHO_MD_MKDIR: {
2153 struct echo_thread_info *info = echo_env_info(env);
2154 __u32 mode = data->ioc_obdo2.o_mode;
2155 struct lu_fid *fid = &info->eti_fid;
2156 int stripe_count = (int)data->ioc_obdo2.o_misc;
2157 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2159 fid->f_seq = data->ioc_obdo1.o_seq;
2160 fid->f_oid = (__u32)data->ioc_obdo1.o_id;
2162 /* In the function below, .hs_keycmp resolves to
2163 * lu_obj_hop_keycmp() */
2164 /* coverity[overrun-buffer-val] */
2165 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2166 id, mode, count, stripe_count,
2170 case ECHO_MD_DESTROY:
2171 case ECHO_MD_RMDIR: {
2172 __u32 mode = data->ioc_obdo2.o_mode;
2174 rc = echo_destroy_object(env, ed, parent, name, namelen,
2178 case ECHO_MD_LOOKUP:
2179 rc = echo_lookup_object(env, ed, parent, id, count);
2181 case ECHO_MD_GETATTR:
2182 rc = echo_getattr_object(env, ed, parent, id, count);
2184 case ECHO_MD_SETATTR:
2185 rc = echo_setattr_object(env, ed, parent, id, count);
2188 CERROR("unknown command %d\n", command);
2194 OBD_FREE(name, namelen + 1);
2196 lu_object_put(env, parent);
2198 LASSERT(info->eti_big_lmm);
2199 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2200 info->eti_big_lmm = NULL;
2201 info->eti_big_lmmsize = 0;
2203 cl_env_put(env, &refcheck);
2207 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2208 int on_target, struct obdo *oa, void *ulsm,
2209 int ulsm_nob, struct obd_trans_info *oti)
2211 struct echo_object *eco;
2212 struct echo_client_obd *ec = ed->ed_ec;
2213 struct lov_stripe_md *lsm = NULL;
2218 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2219 (on_target || /* set_stripe */
2220 ec->ec_nstripes != 0)) { /* LOV */
2221 CERROR ("No valid oid\n");
2225 rc = echo_alloc_memmd(ed, &lsm);
2227 CERROR("Cannot allocate md: rc = %d\n", rc);
2234 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2238 if (lsm->lsm_stripe_count == 0)
2239 lsm->lsm_stripe_count = ec->ec_nstripes;
2241 if (lsm->lsm_stripe_size == 0)
2242 lsm->lsm_stripe_size = CFS_PAGE_SIZE;
2246 /* setup stripes: indices + default ids if required */
2247 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2248 if (lsm->lsm_oinfo[i]->loi_id == 0)
2249 lsm->lsm_oinfo[i]->loi_id = lsm->lsm_object_id;
2251 lsm->lsm_oinfo[i]->loi_ost_idx =
2252 (idx + i) % ec->ec_nstripes;
2256 /* setup object ID here for !on_target and LOV hint */
2257 if (oa->o_valid & OBD_MD_FLID)
2258 lsm->lsm_object_id = oa->o_id;
2260 if (lsm->lsm_object_id == 0)
2261 lsm->lsm_object_id = ++last_object_id;
2265 /* Only echo objects are allowed to be created */
2266 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2267 (oa->o_seq == FID_SEQ_ECHO));
2268 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2270 CERROR("Cannot create objects: rc = %d\n", rc);
2276 /* See what object ID we were given */
2277 oa->o_id = lsm->lsm_object_id;
2278 oa->o_valid |= OBD_MD_FLID;
2280 eco = cl_echo_object_find(ed, &lsm);
2282 GOTO(failed, rc = PTR_ERR(eco));
2283 cl_echo_object_put(eco);
2285 CDEBUG(D_INFO, "oa->o_id = %lx\n", (long)oa->o_id);
2290 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2292 echo_free_memmd(ed, &lsm);
2294 CERROR("create object failed with: rc = %d\n", rc);
2298 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2301 struct lov_stripe_md *lsm = NULL;
2302 struct echo_object *eco;
2306 if ((oa->o_valid & OBD_MD_FLID) == 0 ||
2307 oa->o_id == 0) /* disallow use of object id 0 */
2309 CERROR ("No valid oid\n");
2313 rc = echo_alloc_memmd(ed, &lsm);
2317 lsm->lsm_object_id = oa->o_id;
2318 if (oa->o_valid & OBD_MD_FLGROUP)
2319 lsm->lsm_object_seq = oa->o_seq;
2321 lsm->lsm_object_seq = FID_SEQ_ECHO;
2324 eco = cl_echo_object_find(ed, &lsm);
2330 echo_free_memmd(ed, &lsm);
2334 static void echo_put_object(struct echo_object *eco)
2336 if (cl_echo_object_put(eco))
2337 CERROR("echo client: drop an object failed");
2341 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2343 unsigned long stripe_count;
2344 unsigned long stripe_size;
2345 unsigned long width;
2346 unsigned long woffset;
2350 if (lsm->lsm_stripe_count <= 1)
2354 stripe_size = lsm->lsm_stripe_size;
2355 stripe_count = lsm->lsm_stripe_count;
2357 /* width = # bytes in all stripes */
2358 width = stripe_size * stripe_count;
2360 /* woffset = offset within a width; offset = whole number of widths */
2361 woffset = do_div (offset, width);
2363 stripe_index = woffset / stripe_size;
2365 *idp = lsm->lsm_oinfo[stripe_index]->loi_id;
2366 *offp = offset * stripe_size + woffset % stripe_size;
2370 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2371 cfs_page_t *page, int rw, obd_id id,
2372 obd_off offset, obd_off count)
2379 /* no partial pages on the client */
2380 LASSERT(count == CFS_PAGE_SIZE);
2382 addr = cfs_kmap(page);
2384 for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2385 if (rw == OBD_BRW_WRITE) {
2386 stripe_off = offset + delta;
2388 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2390 stripe_off = 0xdeadbeef00c0ffeeULL;
2391 stripe_id = 0xdeadbeef00c0ffeeULL;
2393 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2394 stripe_off, stripe_id);
2400 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2401 cfs_page_t *page, obd_id id,
2402 obd_off offset, obd_off count)
2411 /* no partial pages on the client */
2412 LASSERT(count == CFS_PAGE_SIZE);
2414 addr = cfs_kmap(page);
2416 for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2417 stripe_off = offset + delta;
2419 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2421 rc2 = block_debug_check("test_brw",
2422 addr + delta, OBD_ECHO_BLOCK_SIZE,
2423 stripe_off, stripe_id);
2425 CERROR ("Error in echo object "LPX64"\n", id);
2434 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2435 struct echo_object *eco, obd_off offset,
2436 obd_size count, int async,
2437 struct obd_trans_info *oti)
2439 struct lov_stripe_md *lsm = eco->eo_lsm;
2441 struct brw_page *pga;
2442 struct brw_page *pgp;
2452 verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
2453 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2454 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2456 gfp_mask = ((oa->o_id & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
2458 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2459 LASSERT(lsm != NULL);
2460 LASSERT(lsm->lsm_object_id == oa->o_id);
2463 (count & (~CFS_PAGE_MASK)) != 0)
2466 /* XXX think again with misaligned I/O */
2467 npages = count >> CFS_PAGE_SHIFT;
2469 if (rw == OBD_BRW_WRITE)
2470 brw_flags = OBD_BRW_ASYNC;
2472 OBD_ALLOC(pga, npages * sizeof(*pga));
2476 OBD_ALLOC(pages, npages * sizeof(*pages));
2477 if (pages == NULL) {
2478 OBD_FREE(pga, npages * sizeof(*pga));
2482 for (i = 0, pgp = pga, off = offset;
2484 i++, pgp++, off += CFS_PAGE_SIZE) {
2486 LASSERT (pgp->pg == NULL); /* for cleanup */
2489 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2490 if (pgp->pg == NULL)
2494 pgp->count = CFS_PAGE_SIZE;
2496 pgp->flag = brw_flags;
2499 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2500 oa->o_id, off, pgp->count);
2503 /* brw mode can only be used at client */
2504 LASSERT(ed->ed_next != NULL);
2505 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2508 if (rc != 0 || rw != OBD_BRW_READ)
2511 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2512 if (pgp->pg == NULL)
2517 vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
2518 pgp->off, pgp->count);
2519 if (vrc != 0 && rc == 0)
2522 OBD_PAGE_FREE(pgp->pg);
2524 OBD_FREE(pga, npages * sizeof(*pga));
2525 OBD_FREE(pages, npages * sizeof(*pages));
2529 static int echo_client_prep_commit(const struct lu_env *env,
2530 struct obd_export *exp, int rw,
2531 struct obdo *oa, struct echo_object *eco,
2532 obd_off offset, obd_size count,
2533 obd_size batch, struct obd_trans_info *oti,
2536 struct lov_stripe_md *lsm = eco->eo_lsm;
2537 struct obd_ioobj ioo;
2538 struct niobuf_local *lnb;
2539 struct niobuf_remote *rnb;
2541 obd_size npages, tot_pages;
2542 int i, ret = 0, brw_flags = 0;
2546 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2547 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
2550 npages = batch >> CFS_PAGE_SHIFT;
2551 tot_pages = count >> CFS_PAGE_SHIFT;
2553 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2554 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2556 if (lnb == NULL || rnb == NULL)
2557 GOTO(out, ret = -ENOMEM);
2559 if (rw == OBD_BRW_WRITE && async)
2560 brw_flags |= OBD_BRW_ASYNC;
2562 obdo_to_ioobj(oa, &ioo);
2566 for(; tot_pages; tot_pages -= npages) {
2569 if (tot_pages < npages)
2572 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
2573 rnb[i].offset = off;
2574 rnb[i].len = CFS_PAGE_SIZE;
2575 rnb[i].flags = brw_flags;
2578 ioo.ioo_bufcnt = npages;
2579 oti->oti_transno = 0;
2582 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2586 LASSERT(lpages == npages);
2588 for (i = 0; i < lpages; i++) {
2589 cfs_page_t *page = lnb[i].page;
2591 /* read past eof? */
2592 if (page == NULL && lnb[i].rc == 0)
2596 lnb[i].flags |= OBD_BRW_ASYNC;
2598 if (oa->o_id == ECHO_PERSISTENT_OBJID ||
2599 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2600 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2603 if (rw == OBD_BRW_WRITE)
2604 echo_client_page_debug_setup(lsm, page, rw,
2609 echo_client_page_debug_check(lsm, page,
2615 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2616 rnb, npages, lnb, oti, ret);
2620 /* Reset oti otherwise it would confuse ldiskfs. */
2621 memset(oti, 0, sizeof(*oti));
2626 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2628 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2632 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2633 struct obd_export *exp,
2634 struct obd_ioctl_data *data,
2635 struct obd_trans_info *dummy_oti)
2637 struct obd_device *obd = class_exp2obd(exp);
2638 struct echo_device *ed = obd2echo_dev(obd);
2639 struct echo_client_obd *ec = ed->ed_ec;
2640 struct obdo *oa = &data->ioc_obdo1;
2641 struct echo_object *eco;
2647 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2649 rc = echo_get_object(&eco, ed, oa);
2653 oa->o_valid &= ~OBD_MD_FLHANDLE;
2655 /* OFD/obdfilter works only via prep/commit */
2656 test_mode = (long)data->ioc_pbuf1;
2660 if (ed->ed_next == NULL && test_mode != 3) {
2662 data->ioc_plen1 = data->ioc_count;
2665 /* Truncate batch size to maximum */
2666 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2667 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2669 switch (test_mode) {
2673 rc = echo_client_kbrw(ed, rw, oa,
2674 eco, data->ioc_offset,
2675 data->ioc_count, async, dummy_oti);
2678 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2679 eco, data->ioc_offset,
2680 data->ioc_count, data->ioc_plen1,
2686 echo_put_object(eco);
2691 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2692 int mode, obd_off offset, obd_size nob)
2694 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2695 struct lustre_handle *ulh = &oa->o_handle;
2696 struct echo_object *eco;
2701 if (ed->ed_next == NULL)
2702 RETURN(-EOPNOTSUPP);
2704 if (!(mode == LCK_PR || mode == LCK_PW))
2707 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2708 (nob & (~CFS_PAGE_MASK)) != 0)
2711 rc = echo_get_object (&eco, ed, oa);
2715 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2716 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2718 oa->o_valid |= OBD_MD_FLHANDLE;
2719 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2721 echo_put_object(eco);
2726 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2728 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2729 __u64 cookie = oa->o_handle.cookie;
2731 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2734 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2735 return cl_echo_cancel(ed, cookie);
2739 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2740 void *karg, void *uarg)
2742 struct obd_device *obd = exp->exp_obd;
2743 struct echo_device *ed = obd2echo_dev(obd);
2744 struct echo_client_obd *ec = ed->ed_ec;
2745 struct echo_object *eco;
2746 struct obd_ioctl_data *data = karg;
2747 struct obd_trans_info dummy_oti;
2749 struct oti_req_ack_lock *ack_lock;
2752 int rw = OBD_BRW_READ;
2757 memset(&dummy_oti, 0, sizeof(dummy_oti));
2759 oa = &data->ioc_obdo1;
2760 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2761 oa->o_valid |= OBD_MD_FLGROUP;
2762 oa->o_seq = FID_SEQ_ECHO;
2765 /* This FID is unpacked just for validation at this point */
2766 rc = fid_ostid_unpack(&fid, &oa->o_oi, 0);
2774 rc = lu_env_init(env, LCT_DT_THREAD);
2776 GOTO(out, rc = -ENOMEM);
2779 case OBD_IOC_CREATE: /* may create echo object */
2780 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2781 GOTO (out, rc = -EPERM);
2783 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2784 data->ioc_plen1, &dummy_oti);
2787 case OBD_IOC_ECHO_MD: {
2794 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2795 GOTO(out, rc = -EPERM);
2797 count = data->ioc_count;
2798 cmd = data->ioc_command;
2800 id = data->ioc_obdo2.o_id;
2802 dirlen = data->ioc_plen1;
2803 OBD_ALLOC(dir, dirlen + 1);
2805 GOTO(out, rc = -ENOMEM);
2807 if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2808 OBD_FREE(dir, data->ioc_plen1 + 1);
2809 GOTO(out, rc = -EFAULT);
2812 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2813 OBD_FREE(dir, dirlen + 1);
2816 case OBD_IOC_ECHO_ALLOC_SEQ: {
2817 struct lu_env *cl_env;
2822 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2823 GOTO(out, rc = -EPERM);
2825 cl_env = cl_env_get(&refcheck);
2827 GOTO(out, rc = PTR_ERR(cl_env));
2829 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2832 cl_env_put(cl_env, &refcheck);
2836 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2837 cl_env_put(cl_env, &refcheck);
2839 CERROR("%s: Can not alloc seq: rc = %d\n",
2844 if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2847 max_count = LUSTRE_SEQ_MAX_WIDTH;
2848 if (cfs_copy_to_user(data->ioc_pbuf2, &max_count,
2853 case OBD_IOC_DESTROY:
2854 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2855 GOTO (out, rc = -EPERM);
2857 rc = echo_get_object(&eco, ed, oa);
2859 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2860 &dummy_oti, NULL, NULL);
2862 eco->eo_deleted = 1;
2863 echo_put_object(eco);
2867 case OBD_IOC_GETATTR:
2868 rc = echo_get_object(&eco, ed, oa);
2870 struct obd_info oinfo = { { { 0 } } };
2871 oinfo.oi_md = eco->eo_lsm;
2873 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2874 echo_put_object(eco);
2878 case OBD_IOC_SETATTR:
2879 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2880 GOTO (out, rc = -EPERM);
2882 rc = echo_get_object(&eco, ed, oa);
2884 struct obd_info oinfo = { { { 0 } } };
2886 oinfo.oi_md = eco->eo_lsm;
2888 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2889 echo_put_object(eco);
2893 case OBD_IOC_BRW_WRITE:
2894 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2895 GOTO (out, rc = -EPERM);
2899 case OBD_IOC_BRW_READ:
2900 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2903 case ECHO_IOC_GET_STRIPE:
2904 rc = echo_get_object(&eco, ed, oa);
2906 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2908 echo_put_object(eco);
2912 case ECHO_IOC_SET_STRIPE:
2913 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2914 GOTO (out, rc = -EPERM);
2916 if (data->ioc_pbuf1 == NULL) { /* unset */
2917 rc = echo_get_object(&eco, ed, oa);
2919 eco->eo_deleted = 1;
2920 echo_put_object(eco);
2923 rc = echo_create_object(env, ed, 0, oa,
2925 data->ioc_plen1, &dummy_oti);
2929 case ECHO_IOC_ENQUEUE:
2930 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2931 GOTO (out, rc = -EPERM);
2933 rc = echo_client_enqueue(exp, oa,
2934 data->ioc_conn1, /* lock mode */
2936 data->ioc_count);/*extent*/
2939 case ECHO_IOC_CANCEL:
2940 rc = echo_client_cancel(exp, oa);
2944 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2945 GOTO (out, rc = -ENOTTY);
2953 /* XXX this should be in a helper also called by target_send_reply */
2954 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2956 if (!ack_lock->mode)
2958 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
2964 static int echo_client_setup(const struct lu_env *env,
2965 struct obd_device *obddev, struct lustre_cfg *lcfg)
2967 struct echo_client_obd *ec = &obddev->u.echo_client;
2968 struct obd_device *tgt;
2969 struct obd_uuid echo_uuid = { "ECHO_UUID" };
2970 struct obd_connect_data *ocd = NULL;
2974 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2975 CERROR("requires a TARGET OBD name\n");
2979 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2980 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2981 CERROR("device not attached or not set up (%s)\n",
2982 lustre_cfg_string(lcfg, 1));
2986 cfs_spin_lock_init (&ec->ec_lock);
2987 CFS_INIT_LIST_HEAD (&ec->ec_objects);
2988 CFS_INIT_LIST_HEAD (&ec->ec_locks);
2990 ec->ec_nstripes = 0;
2992 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2993 lu_context_tags_update(ECHO_MD_CTX_TAG);
2994 lu_session_tags_update(ECHO_MD_SES_TAG);
2998 OBD_ALLOC(ocd, sizeof(*ocd));
3000 CERROR("Can't alloc ocd connecting to %s\n",
3001 lustre_cfg_string(lcfg, 1));
3005 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3006 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3007 OBD_CONNECT_64BITHASH;
3008 ocd->ocd_version = LUSTRE_VERSION_CODE;
3009 ocd->ocd_group = FID_SEQ_ECHO;
3011 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3013 /* Turn off pinger because it connects to tgt obd directly. */
3014 cfs_spin_lock(&tgt->obd_dev_lock);
3015 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3016 cfs_spin_unlock(&tgt->obd_dev_lock);
3019 OBD_FREE(ocd, sizeof(*ocd));
3022 CERROR("fail to connect to device %s\n",
3023 lustre_cfg_string(lcfg, 1));
3030 static int echo_client_cleanup(struct obd_device *obddev)
3032 struct echo_device *ed = obd2echo_dev(obddev);
3033 struct echo_client_obd *ec = &obddev->u.echo_client;
3037 /*Do nothing for Metadata echo client*/
3041 if (ed->ed_next_ismd) {
3042 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3043 lu_session_tags_clear(ECHO_MD_SES_TAG);
3047 if (!cfs_list_empty(&obddev->obd_exports)) {
3048 CERROR("still has clients!\n");
3052 LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3053 rc = obd_disconnect(ec->ec_exp);
3055 CERROR("fail to disconnect device: %d\n", rc);
3060 static int echo_client_connect(const struct lu_env *env,
3061 struct obd_export **exp,
3062 struct obd_device *src, struct obd_uuid *cluuid,
3063 struct obd_connect_data *data, void *localdata)
3066 struct lustre_handle conn = { 0 };
3069 rc = class_connect(&conn, src, cluuid);
3071 *exp = class_conn2export(&conn);
3077 static int echo_client_disconnect(struct obd_export *exp)
3080 struct obd_device *obd;
3081 struct echo_client_obd *ec;
3082 struct ec_lock *ecl;
3088 GOTO(out, rc = -EINVAL);
3092 ec = &obd->u.echo_client;
3094 /* no more contention on export's lock list */
3095 while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
3096 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
3097 struct ec_lock, ecl_exp_chain);
3098 cfs_list_del (&ecl->ecl_exp_chain);
3100 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3101 ecl->ecl_mode, &ecl->ecl_lock_handle);
3103 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3104 "(%d)\n", ecl->ecl_object->eco_id, rc);
3106 echo_put_object (ecl->ecl_object);
3107 OBD_FREE (ecl, sizeof (*ecl));
3111 rc = class_disconnect(exp);
3117 static struct obd_ops echo_client_obd_ops = {
3118 .o_owner = THIS_MODULE,
3121 .o_setup = echo_client_setup,
3122 .o_cleanup = echo_client_cleanup,
3125 .o_iocontrol = echo_client_iocontrol,
3126 .o_connect = echo_client_connect,
3127 .o_disconnect = echo_client_disconnect
3130 int echo_client_init(void)
3132 struct lprocfs_static_vars lvars = { 0 };
3135 lprocfs_echo_init_vars(&lvars);
3137 rc = lu_kmem_init(echo_caches);
3139 rc = class_register_type(&echo_client_obd_ops, NULL,
3141 LUSTRE_ECHO_CLIENT_NAME,
3144 lu_kmem_fini(echo_caches);
3149 void echo_client_exit(void)
3151 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3152 lu_kmem_fini(echo_caches);
3156 static int __init obdecho_init(void)
3158 struct lprocfs_static_vars lvars;
3162 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3164 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3166 lprocfs_echo_init_vars(&lvars);
3168 # ifdef HAVE_SERVER_SUPPORT
3169 rc = echo_persistent_pages_init();
3173 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3174 LUSTRE_ECHO_NAME, NULL);
3179 rc = echo_client_init();
3181 # ifdef HAVE_SERVER_SUPPORT
3185 class_unregister_type(LUSTRE_ECHO_NAME);
3187 echo_persistent_pages_fini();
3193 static void /*__exit*/ obdecho_exit(void)
3197 # ifdef HAVE_SERVER_SUPPORT
3198 class_unregister_type(LUSTRE_ECHO_NAME);
3199 echo_persistent_pages_fini();
3203 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3204 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3205 MODULE_LICENSE("GPL");
3207 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3208 #endif /* __KERNEL__ */
3210 /** @} echo_client */