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_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 mutex_lock(&ep->ep_lock);
290 else if (!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(mutex_is_locked(&ep->ep_lock));
302 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 (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, 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 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 spin_lock(&ec->ec_lock);
523 cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
524 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 spin_lock(&ec->ec_lock);
588 cfs_list_del_init(&eco->eo_obj_chain);
589 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,
757 struct seq_server_site *ss)
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, ss->ss_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 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 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_site2seq(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 spin_lock(&ec->ec_lock);
1019 cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1020 eco->eo_deleted = 1;
1021 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 spin_lock(&ec->ec_lock);
1031 while (!cfs_list_empty(&ec->ec_objects)) {
1032 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 spin_lock(&ec->ec_lock);
1040 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 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 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 spin_unlock(&ec->ec_lock);
1218 cl_lock_release(env, lck, "ec enqueue", cfs_current());
1224 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1225 int mode, __u64 *cookie)
1227 struct echo_thread_info *info;
1234 env = cl_env_get(&refcheck);
1236 RETURN(PTR_ERR(env));
1238 info = echo_env_info(env);
1241 io->ci_ignore_layout = 1;
1242 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1245 LASSERT(result == 0);
1247 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1248 cl_io_fini(env, io);
1252 cl_env_put(env, &refcheck);
1256 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1259 struct echo_client_obd *ec = ed->ed_ec;
1260 struct echo_lock *ecl = NULL;
1262 int found = 0, still_used = 0;
1265 LASSERT(ec != NULL);
1266 spin_lock(&ec->ec_lock);
1267 cfs_list_for_each (el, &ec->ec_locks) {
1268 ecl = cfs_list_entry (el, struct echo_lock, el_chain);
1269 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1270 found = (ecl->el_cookie == cookie);
1272 if (cfs_atomic_dec_and_test(&ecl->el_refcount))
1273 cfs_list_del_init(&ecl->el_chain);
1279 spin_unlock(&ec->ec_lock);
1284 echo_lock_release(env, ecl, still_used);
1288 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1295 env = cl_env_get(&refcheck);
1297 RETURN(PTR_ERR(env));
1299 rc = cl_echo_cancel0(env, ed, cookie);
1301 cl_env_put(env, &refcheck);
1305 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1306 enum cl_req_type unused, struct cl_2queue *queue)
1308 struct cl_page *clp;
1309 struct cl_page *temp;
1313 cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1315 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1318 result = result ?: rc;
1323 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1324 cfs_page_t **pages, int npages, int async)
1327 struct echo_thread_info *info;
1328 struct cl_object *obj = echo_obj2cl(eco);
1329 struct echo_device *ed = eco->eo_dev;
1330 struct cl_2queue *queue;
1332 struct cl_page *clp;
1333 struct lustre_handle lh = { 0 };
1334 int page_size = cl_page_size(obj);
1340 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1341 LASSERT(ed->ed_next != NULL);
1342 env = cl_env_get(&refcheck);
1344 RETURN(PTR_ERR(env));
1346 info = echo_env_info(env);
1348 queue = &info->eti_queue;
1350 cl_2queue_init(queue);
1352 io->ci_ignore_layout = 1;
1353 rc = cl_io_init(env, io, CIT_MISC, obj);
1359 rc = cl_echo_enqueue0(env, eco, offset,
1360 offset + npages * CFS_PAGE_SIZE - 1,
1361 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1364 GOTO(error_lock, rc);
1366 for (i = 0; i < npages; i++) {
1368 clp = cl_page_find(env, obj, cl_index(obj, offset),
1369 pages[i], CPT_TRANSIENT);
1374 LASSERT(clp->cp_type == CPT_TRANSIENT);
1376 rc = cl_page_own(env, io, clp);
1378 LASSERT(clp->cp_state == CPS_FREEING);
1379 cl_page_put(env, clp);
1383 cl_2queue_add(queue, clp);
1385 /* drop the reference count for cl_page_find, so that the page
1386 * will be freed in cl_2queue_fini. */
1387 cl_page_put(env, clp);
1388 cl_page_clip(env, clp, 0, page_size);
1390 offset += page_size;
1394 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1396 async = async && (typ == CRT_WRITE);
1398 rc = cl_echo_async_brw(env, io, typ, queue);
1400 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1401 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1402 async ? "async" : "sync", rc);
1405 cl_echo_cancel0(env, ed, lh.cookie);
1408 cl_2queue_discard(env, io, queue);
1409 cl_2queue_disown(env, io, queue);
1410 cl_2queue_fini(env, queue);
1411 cl_io_fini(env, io);
1413 cl_env_put(env, &refcheck);
1416 /** @} echo_exports */
1419 static obd_id last_object_id;
1422 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1424 struct lov_stripe_md *ulsm = _ulsm;
1427 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1431 if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm)))
1434 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1435 if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1436 sizeof(lsm->lsm_oinfo[0])))
1443 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1444 void *ulsm, int ulsm_nob)
1446 struct echo_client_obd *ec = ed->ed_ec;
1449 if (ulsm_nob < sizeof (*lsm))
1452 if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm)))
1455 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1456 lsm->lsm_magic != LOV_MAGIC ||
1457 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1458 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1462 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1463 if (cfs_copy_from_user(lsm->lsm_oinfo[i],
1464 ((struct lov_stripe_md *)ulsm)-> \
1466 sizeof(lsm->lsm_oinfo[0])))
1472 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1475 sprintf(name, LPU64, id);
1476 lname->ln_name = name;
1477 lname->ln_namelen = strlen(name);
1480 /* similar to mdt_attr_get_complex */
1481 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1484 struct echo_thread_info *info = echo_env_info(env);
1489 LASSERT(ma->ma_lmm_size > 0);
1491 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1495 /* big_lmm may need to be grown */
1496 if (info->eti_big_lmmsize < rc) {
1497 int size = size_roundup_power2(rc);
1499 if (info->eti_big_lmmsize > 0) {
1500 /* free old buffer */
1501 LASSERT(info->eti_big_lmm);
1502 OBD_FREE_LARGE(info->eti_big_lmm,
1503 info->eti_big_lmmsize);
1504 info->eti_big_lmm = NULL;
1505 info->eti_big_lmmsize = 0;
1508 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1509 if (info->eti_big_lmm == NULL)
1511 info->eti_big_lmmsize = size;
1513 LASSERT(info->eti_big_lmmsize >= rc);
1515 info->eti_buf.lb_buf = info->eti_big_lmm;
1516 info->eti_buf.lb_len = info->eti_big_lmmsize;
1517 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1521 ma->ma_valid |= MA_LOV;
1522 ma->ma_lmm = info->eti_big_lmm;
1523 ma->ma_lmm_size = rc;
1528 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1531 struct echo_thread_info *info = echo_env_info(env);
1532 struct lu_buf *buf = &info->eti_buf;
1533 cfs_umode_t mode = lu_object_attr(&next->mo_lu);
1534 int need = ma->ma_need;
1541 if (need & MA_INODE) {
1542 ma->ma_need = MA_INODE;
1543 rc = mo_attr_get(env, next, ma);
1546 ma->ma_valid |= MA_INODE;
1549 if (need & MA_LOV) {
1550 if (S_ISREG(mode) || S_ISDIR(mode)) {
1551 LASSERT(ma->ma_lmm_size > 0);
1552 buf->lb_buf = ma->ma_lmm;
1553 buf->lb_len = ma->ma_lmm_size;
1554 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1556 ma->ma_lmm_size = rc2;
1557 ma->ma_valid |= MA_LOV;
1558 } else if (rc2 == -ENODATA) {
1560 ma->ma_lmm_size = 0;
1561 } else if (rc2 == -ERANGE) {
1562 rc2 = echo_big_lmm_get(env, next, ma);
1564 GOTO(out, rc = rc2);
1566 GOTO(out, rc = rc2);
1571 #ifdef CONFIG_FS_POSIX_ACL
1572 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1573 buf->lb_buf = ma->ma_acl;
1574 buf->lb_len = ma->ma_acl_size;
1575 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1577 ma->ma_acl_size = rc2;
1578 ma->ma_valid |= MA_ACL_DEF;
1579 } else if (rc2 == -ENODATA) {
1581 ma->ma_acl_size = 0;
1583 GOTO(out, rc = rc2);
1589 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1590 rc, ma->ma_valid, ma->ma_lmm);
1595 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1596 struct md_object *parent, struct lu_fid *fid,
1597 struct lu_name *lname, struct md_op_spec *spec,
1600 struct lu_object *ec_child, *child;
1601 struct lu_device *ld = ed->ed_next;
1602 struct echo_thread_info *info = echo_env_info(env);
1603 struct lu_fid *fid2 = &info->eti_fid2;
1604 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1609 rc = mdo_lookup(env, parent, lname, fid2, spec);
1612 else if (rc != -ENOENT)
1615 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1617 if (IS_ERR(ec_child)) {
1618 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1620 RETURN(PTR_ERR(ec_child));
1623 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1624 if (child == NULL) {
1625 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1626 GOTO(out_put, rc = -EINVAL);
1629 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1630 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1633 * Do not perform lookup sanity check. We know that name does not exist.
1635 spec->sp_cr_lookup = 0;
1636 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1638 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1641 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1642 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1645 lu_object_put(env, ec_child);
1649 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1652 struct echo_thread_info *info = echo_env_info(env);
1654 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1655 ma->ma_lmm = (void *)&info->eti_lmm;
1656 ma->ma_lmm_size = sizeof(info->eti_lmm);
1658 LASSERT(info->eti_big_lmmsize);
1659 ma->ma_lmm = info->eti_big_lmm;
1660 ma->ma_lmm_size = info->eti_big_lmmsize;
1666 static int echo_create_md_object(const struct lu_env *env,
1667 struct echo_device *ed,
1668 struct lu_object *ec_parent,
1670 char *name, int namelen,
1671 __u64 id, __u32 mode, int count,
1672 int stripe_count, int stripe_offset)
1674 struct lu_object *parent;
1675 struct echo_thread_info *info = echo_env_info(env);
1676 struct lu_name *lname = &info->eti_lname;
1677 struct md_op_spec *spec = &info->eti_spec;
1678 struct md_attr *ma = &info->eti_ma;
1679 struct lu_device *ld = ed->ed_next;
1685 if (ec_parent == NULL)
1687 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1691 memset(ma, 0, sizeof(*ma));
1692 memset(spec, 0, sizeof(*spec));
1693 if (stripe_count != 0) {
1694 spec->sp_cr_flags |= FMODE_WRITE;
1695 echo_set_lmm_size(env, ld, ma);
1696 if (stripe_count != -1) {
1697 struct lov_user_md_v3 *lum = &info->eti_lum;
1699 lum->lmm_magic = LOV_USER_MAGIC_V3;
1700 lum->lmm_stripe_count = stripe_count;
1701 lum->lmm_stripe_offset = stripe_offset;
1702 lum->lmm_pattern = 0;
1703 spec->u.sp_ea.eadata = lum;
1704 spec->u.sp_ea.eadatalen = sizeof(*lum);
1705 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1709 ma->ma_attr.la_mode = mode;
1710 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1711 ma->ma_attr.la_ctime = cfs_time_current_64();
1714 lname->ln_name = name;
1715 lname->ln_namelen = namelen;
1716 /* If name is specified, only create one object by name */
1717 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1722 /* Create multiple object sequenced by id */
1723 for (i = 0; i < count; i++) {
1724 char *tmp_name = info->eti_name;
1726 echo_md_build_name(lname, tmp_name, id);
1728 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1731 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1742 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1743 struct echo_device *ed,
1744 struct md_object *parent,
1745 struct lu_name *lname)
1747 struct echo_thread_info *info = echo_env_info(env);
1748 struct lu_fid *fid = &info->eti_fid;
1749 struct lu_object *child;
1753 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1755 rc = mdo_lookup(env, parent, lname, fid, NULL);
1757 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1758 RETURN(ERR_PTR(rc));
1761 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1766 static int echo_setattr_object(const struct lu_env *env,
1767 struct echo_device *ed,
1768 struct lu_object *ec_parent,
1769 __u64 id, int count)
1771 struct lu_object *parent;
1772 struct echo_thread_info *info = echo_env_info(env);
1773 struct lu_name *lname = &info->eti_lname;
1774 char *name = info->eti_name;
1775 struct lu_device *ld = ed->ed_next;
1776 struct lu_buf *buf = &info->eti_buf;
1782 if (ec_parent == NULL)
1784 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1788 for (i = 0; i < count; i++) {
1789 struct lu_object *ec_child, *child;
1791 echo_md_build_name(lname, name, id);
1793 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1794 if (IS_ERR(ec_child)) {
1795 CERROR("Can't find child %s: rc = %ld\n",
1796 lname->ln_name, PTR_ERR(ec_child));
1797 RETURN(PTR_ERR(ec_child));
1800 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1801 if (child == NULL) {
1802 CERROR("Can not locate the child %s\n", lname->ln_name);
1803 lu_object_put(env, ec_child);
1808 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1809 PFID(lu_object_fid(child)));
1811 buf->lb_buf = info->eti_xattr_buf;
1812 buf->lb_len = sizeof(info->eti_xattr_buf);
1814 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1815 rc = mo_xattr_set(env, lu2md(child), buf, name,
1818 CERROR("Can not setattr child "DFID": rc = %d\n",
1819 PFID(lu_object_fid(child)), rc);
1820 lu_object_put(env, ec_child);
1823 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1824 PFID(lu_object_fid(child)));
1826 lu_object_put(env, ec_child);
1831 static int echo_getattr_object(const struct lu_env *env,
1832 struct echo_device *ed,
1833 struct lu_object *ec_parent,
1834 __u64 id, int count)
1836 struct lu_object *parent;
1837 struct echo_thread_info *info = echo_env_info(env);
1838 struct lu_name *lname = &info->eti_lname;
1839 char *name = info->eti_name;
1840 struct md_attr *ma = &info->eti_ma;
1841 struct lu_device *ld = ed->ed_next;
1847 if (ec_parent == NULL)
1849 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1853 memset(ma, 0, sizeof(*ma));
1854 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1855 ma->ma_acl = info->eti_xattr_buf;
1856 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1858 for (i = 0; i < count; i++) {
1859 struct lu_object *ec_child, *child;
1862 echo_md_build_name(lname, name, id);
1863 echo_set_lmm_size(env, ld, ma);
1865 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1866 if (IS_ERR(ec_child)) {
1867 CERROR("Can't find child %s: rc = %ld\n",
1868 lname->ln_name, PTR_ERR(ec_child));
1869 RETURN(PTR_ERR(ec_child));
1872 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1873 if (child == NULL) {
1874 CERROR("Can not locate the child %s\n", lname->ln_name);
1875 lu_object_put(env, ec_child);
1879 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1880 PFID(lu_object_fid(child)));
1881 rc = echo_attr_get_complex(env, lu2md(child), ma);
1883 CERROR("Can not getattr child "DFID": rc = %d\n",
1884 PFID(lu_object_fid(child)), rc);
1885 lu_object_put(env, ec_child);
1888 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1889 PFID(lu_object_fid(child)));
1891 lu_object_put(env, ec_child);
1897 static int echo_lookup_object(const struct lu_env *env,
1898 struct echo_device *ed,
1899 struct lu_object *ec_parent,
1900 __u64 id, int count)
1902 struct lu_object *parent;
1903 struct echo_thread_info *info = echo_env_info(env);
1904 struct lu_name *lname = &info->eti_lname;
1905 char *name = info->eti_name;
1906 struct lu_fid *fid = &info->eti_fid;
1907 struct lu_device *ld = ed->ed_next;
1911 if (ec_parent == NULL)
1913 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1917 /*prepare the requests*/
1918 for (i = 0; i < count; i++) {
1919 echo_md_build_name(lname, name, id);
1921 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1922 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1924 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1926 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1929 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1930 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1937 static int echo_md_destroy_internal(const struct lu_env *env,
1938 struct echo_device *ed,
1939 struct md_object *parent,
1940 struct lu_name *lname,
1943 struct lu_device *ld = ed->ed_next;
1944 struct lu_object *ec_child;
1945 struct lu_object *child;
1950 ec_child = echo_md_lookup(env, ed, parent, lname);
1951 if (IS_ERR(ec_child)) {
1952 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1954 RETURN(PTR_ERR(ec_child));
1957 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1958 if (child == NULL) {
1959 CERROR("Can not locate the child %s\n", lname->ln_name);
1960 GOTO(out_put, rc = -EINVAL);
1963 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1964 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1966 rc = mdo_unlink(env, parent, lu2md(child), lname, ma);
1968 CERROR("Can not unlink child %s: rc = %d\n",
1969 lname->ln_name, rc);
1972 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1973 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1975 lu_object_put(env, ec_child);
1979 static int echo_destroy_object(const struct lu_env *env,
1980 struct echo_device *ed,
1981 struct lu_object *ec_parent,
1982 char *name, int namelen,
1983 __u64 id, __u32 mode,
1986 struct echo_thread_info *info = echo_env_info(env);
1987 struct lu_name *lname = &info->eti_lname;
1988 struct md_attr *ma = &info->eti_ma;
1989 struct lu_device *ld = ed->ed_next;
1990 struct lu_object *parent;
1995 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1999 memset(ma, 0, sizeof(*ma));
2000 ma->ma_attr.la_mode = mode;
2001 ma->ma_attr.la_valid = LA_CTIME;
2002 ma->ma_attr.la_ctime = cfs_time_current_64();
2003 ma->ma_need = MA_INODE;
2007 lname->ln_name = name;
2008 lname->ln_namelen = namelen;
2009 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2014 /*prepare the requests*/
2015 for (i = 0; i < count; i++) {
2016 char *tmp_name = info->eti_name;
2019 echo_md_build_name(lname, tmp_name, id);
2021 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2024 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2033 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2034 struct echo_device *ed, char *path,
2037 struct lu_device *ld = ed->ed_next;
2038 struct md_device *md = lu2md_dev(ld);
2039 struct echo_thread_info *info = echo_env_info(env);
2040 struct lu_fid *fid = &info->eti_fid;
2041 struct lu_name *lname = &info->eti_lname;
2042 struct lu_object *parent = NULL;
2043 struct lu_object *child = NULL;
2047 /*Only support MDD layer right now*/
2048 rc = md->md_ops->mdo_root_get(env, md, fid);
2050 CERROR("get root error: rc = %d\n", rc);
2051 RETURN(ERR_PTR(rc));
2054 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2055 if (IS_ERR(parent)) {
2056 CERROR("Can not find the parent "DFID": rc = %ld\n",
2057 PFID(fid), PTR_ERR(parent));
2062 struct lu_object *ld_parent;
2065 e = strsep(&path, "/");
2070 if (!path || path[0] == '\0')
2076 lname->ln_namelen = strlen(e);
2078 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2079 if (ld_parent == NULL) {
2080 lu_object_put(env, parent);
2085 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2086 lu_object_put(env, parent);
2087 if (IS_ERR(child)) {
2088 rc = (int)PTR_ERR(child);
2089 CERROR("lookup %s under parent "DFID": rc = %d\n",
2090 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2097 RETURN(ERR_PTR(rc));
2102 static void echo_ucred_init(struct lu_env *env)
2104 struct lu_ucred *ucred = lu_ucred(env);
2106 ucred->uc_valid = UCRED_INVALID;
2108 ucred->uc_suppgids[0] = -1;
2109 ucred->uc_suppgids[1] = -1;
2111 ucred->uc_uid = ucred->uc_o_uid = cfs_curproc_uid();
2112 ucred->uc_gid = ucred->uc_o_gid = cfs_curproc_gid();
2113 ucred->uc_fsuid = ucred->uc_o_fsuid = cfs_curproc_fsuid();
2114 ucred->uc_fsgid = ucred->uc_o_fsgid = cfs_curproc_fsgid();
2115 ucred->uc_cap = cfs_curproc_cap_pack();
2117 /* remove fs privilege for non-root user. */
2118 if (ucred->uc_fsuid)
2119 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2120 ucred->uc_valid = UCRED_NEW;
2123 static void echo_ucred_fini(struct lu_env *env)
2125 struct lu_ucred *ucred = lu_ucred(env);
2126 ucred->uc_valid = UCRED_INIT;
2129 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2130 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2131 static int echo_md_handler(struct echo_device *ed, int command,
2132 char *path, int path_len, int id, int count,
2133 struct obd_ioctl_data *data)
2135 struct echo_thread_info *info;
2136 struct lu_device *ld = ed->ed_next;
2139 struct lu_object *parent;
2141 int namelen = data->ioc_plen2;
2146 CERROR("MD echo client is not being initialized properly\n");
2150 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2151 CERROR("Only support MDD layer right now!\n");
2155 env = cl_env_get(&refcheck);
2157 RETURN(PTR_ERR(env));
2159 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2163 /* init big_lmm buffer */
2164 info = echo_env_info(env);
2165 LASSERT(info->eti_big_lmm == NULL);
2166 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2167 if (info->eti_big_lmm == NULL)
2168 GOTO(out_env, rc = -ENOMEM);
2169 info->eti_big_lmmsize = MIN_MD_SIZE;
2171 parent = echo_resolve_path(env, ed, path, path_len);
2172 if (IS_ERR(parent)) {
2173 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2175 GOTO(out_free, rc = PTR_ERR(parent));
2179 OBD_ALLOC(name, namelen + 1);
2181 GOTO(out_put, rc = -ENOMEM);
2182 if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen))
2183 GOTO(out_name, rc = -EFAULT);
2186 echo_ucred_init(env);
2189 case ECHO_MD_CREATE:
2190 case ECHO_MD_MKDIR: {
2191 struct echo_thread_info *info = echo_env_info(env);
2192 __u32 mode = data->ioc_obdo2.o_mode;
2193 struct lu_fid *fid = &info->eti_fid;
2194 int stripe_count = (int)data->ioc_obdo2.o_misc;
2195 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2197 fid->f_seq = data->ioc_obdo1.o_seq;
2198 fid->f_oid = (__u32)data->ioc_obdo1.o_id;
2200 /* In the function below, .hs_keycmp resolves to
2201 * lu_obj_hop_keycmp() */
2202 /* coverity[overrun-buffer-val] */
2203 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2204 id, mode, count, stripe_count,
2208 case ECHO_MD_DESTROY:
2209 case ECHO_MD_RMDIR: {
2210 __u32 mode = data->ioc_obdo2.o_mode;
2212 rc = echo_destroy_object(env, ed, parent, name, namelen,
2216 case ECHO_MD_LOOKUP:
2217 rc = echo_lookup_object(env, ed, parent, id, count);
2219 case ECHO_MD_GETATTR:
2220 rc = echo_getattr_object(env, ed, parent, id, count);
2222 case ECHO_MD_SETATTR:
2223 rc = echo_setattr_object(env, ed, parent, id, count);
2226 CERROR("unknown command %d\n", command);
2230 echo_ucred_fini(env);
2234 OBD_FREE(name, namelen + 1);
2236 lu_object_put(env, parent);
2238 LASSERT(info->eti_big_lmm);
2239 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2240 info->eti_big_lmm = NULL;
2241 info->eti_big_lmmsize = 0;
2243 cl_env_put(env, &refcheck);
2247 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2248 int on_target, struct obdo *oa, void *ulsm,
2249 int ulsm_nob, struct obd_trans_info *oti)
2251 struct echo_object *eco;
2252 struct echo_client_obd *ec = ed->ed_ec;
2253 struct lov_stripe_md *lsm = NULL;
2258 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2259 (on_target || /* set_stripe */
2260 ec->ec_nstripes != 0)) { /* LOV */
2261 CERROR ("No valid oid\n");
2265 rc = echo_alloc_memmd(ed, &lsm);
2267 CERROR("Cannot allocate md: rc = %d\n", rc);
2274 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2278 if (lsm->lsm_stripe_count == 0)
2279 lsm->lsm_stripe_count = ec->ec_nstripes;
2281 if (lsm->lsm_stripe_size == 0)
2282 lsm->lsm_stripe_size = CFS_PAGE_SIZE;
2286 /* setup stripes: indices + default ids if required */
2287 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2288 if (lsm->lsm_oinfo[i]->loi_id == 0)
2289 lsm->lsm_oinfo[i]->loi_id = lsm->lsm_object_id;
2291 lsm->lsm_oinfo[i]->loi_ost_idx =
2292 (idx + i) % ec->ec_nstripes;
2296 /* setup object ID here for !on_target and LOV hint */
2297 if (oa->o_valid & OBD_MD_FLID)
2298 lsm->lsm_object_id = oa->o_id;
2300 if (lsm->lsm_object_id == 0)
2301 lsm->lsm_object_id = ++last_object_id;
2305 /* Only echo objects are allowed to be created */
2306 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2307 (oa->o_seq == FID_SEQ_ECHO));
2308 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2310 CERROR("Cannot create objects: rc = %d\n", rc);
2316 /* See what object ID we were given */
2317 oa->o_id = lsm->lsm_object_id;
2318 oa->o_valid |= OBD_MD_FLID;
2320 eco = cl_echo_object_find(ed, &lsm);
2322 GOTO(failed, rc = PTR_ERR(eco));
2323 cl_echo_object_put(eco);
2325 CDEBUG(D_INFO, "oa->o_id = %lx\n", (long)oa->o_id);
2330 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2332 echo_free_memmd(ed, &lsm);
2334 CERROR("create object failed with: rc = %d\n", rc);
2338 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2341 struct lov_stripe_md *lsm = NULL;
2342 struct echo_object *eco;
2346 if ((oa->o_valid & OBD_MD_FLID) == 0 ||
2347 oa->o_id == 0) /* disallow use of object id 0 */
2349 CERROR ("No valid oid\n");
2353 rc = echo_alloc_memmd(ed, &lsm);
2357 lsm->lsm_object_id = oa->o_id;
2358 if (oa->o_valid & OBD_MD_FLGROUP)
2359 lsm->lsm_object_seq = oa->o_seq;
2361 lsm->lsm_object_seq = FID_SEQ_ECHO;
2364 eco = cl_echo_object_find(ed, &lsm);
2370 echo_free_memmd(ed, &lsm);
2374 static void echo_put_object(struct echo_object *eco)
2376 if (cl_echo_object_put(eco))
2377 CERROR("echo client: drop an object failed");
2381 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2383 unsigned long stripe_count;
2384 unsigned long stripe_size;
2385 unsigned long width;
2386 unsigned long woffset;
2390 if (lsm->lsm_stripe_count <= 1)
2394 stripe_size = lsm->lsm_stripe_size;
2395 stripe_count = lsm->lsm_stripe_count;
2397 /* width = # bytes in all stripes */
2398 width = stripe_size * stripe_count;
2400 /* woffset = offset within a width; offset = whole number of widths */
2401 woffset = do_div (offset, width);
2403 stripe_index = woffset / stripe_size;
2405 *idp = lsm->lsm_oinfo[stripe_index]->loi_id;
2406 *offp = offset * stripe_size + woffset % stripe_size;
2410 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2411 cfs_page_t *page, int rw, obd_id id,
2412 obd_off offset, obd_off count)
2419 /* no partial pages on the client */
2420 LASSERT(count == CFS_PAGE_SIZE);
2422 addr = cfs_kmap(page);
2424 for (delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2425 if (rw == OBD_BRW_WRITE) {
2426 stripe_off = offset + delta;
2428 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2430 stripe_off = 0xdeadbeef00c0ffeeULL;
2431 stripe_id = 0xdeadbeef00c0ffeeULL;
2433 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2434 stripe_off, stripe_id);
2440 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2441 cfs_page_t *page, obd_id id,
2442 obd_off offset, obd_off count)
2451 /* no partial pages on the client */
2452 LASSERT(count == CFS_PAGE_SIZE);
2454 addr = cfs_kmap(page);
2456 for (rc = delta = 0; delta < CFS_PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2457 stripe_off = offset + delta;
2459 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2461 rc2 = block_debug_check("test_brw",
2462 addr + delta, OBD_ECHO_BLOCK_SIZE,
2463 stripe_off, stripe_id);
2465 CERROR ("Error in echo object "LPX64"\n", id);
2474 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2475 struct echo_object *eco, obd_off offset,
2476 obd_size count, int async,
2477 struct obd_trans_info *oti)
2479 struct lov_stripe_md *lsm = eco->eo_lsm;
2481 struct brw_page *pga;
2482 struct brw_page *pgp;
2492 verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
2493 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2494 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2496 gfp_mask = ((oa->o_id & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
2498 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2499 LASSERT(lsm != NULL);
2500 LASSERT(lsm->lsm_object_id == oa->o_id);
2503 (count & (~CFS_PAGE_MASK)) != 0)
2506 /* XXX think again with misaligned I/O */
2507 npages = count >> CFS_PAGE_SHIFT;
2509 if (rw == OBD_BRW_WRITE)
2510 brw_flags = OBD_BRW_ASYNC;
2512 OBD_ALLOC(pga, npages * sizeof(*pga));
2516 OBD_ALLOC(pages, npages * sizeof(*pages));
2517 if (pages == NULL) {
2518 OBD_FREE(pga, npages * sizeof(*pga));
2522 for (i = 0, pgp = pga, off = offset;
2524 i++, pgp++, off += CFS_PAGE_SIZE) {
2526 LASSERT (pgp->pg == NULL); /* for cleanup */
2529 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2530 if (pgp->pg == NULL)
2534 pgp->count = CFS_PAGE_SIZE;
2536 pgp->flag = brw_flags;
2539 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2540 oa->o_id, off, pgp->count);
2543 /* brw mode can only be used at client */
2544 LASSERT(ed->ed_next != NULL);
2545 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2548 if (rc != 0 || rw != OBD_BRW_READ)
2551 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2552 if (pgp->pg == NULL)
2557 vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
2558 pgp->off, pgp->count);
2559 if (vrc != 0 && rc == 0)
2562 OBD_PAGE_FREE(pgp->pg);
2564 OBD_FREE(pga, npages * sizeof(*pga));
2565 OBD_FREE(pages, npages * sizeof(*pages));
2569 static int echo_client_prep_commit(const struct lu_env *env,
2570 struct obd_export *exp, int rw,
2571 struct obdo *oa, struct echo_object *eco,
2572 obd_off offset, obd_size count,
2573 obd_size batch, struct obd_trans_info *oti,
2576 struct lov_stripe_md *lsm = eco->eo_lsm;
2577 struct obd_ioobj ioo;
2578 struct niobuf_local *lnb;
2579 struct niobuf_remote *rnb;
2581 obd_size npages, tot_pages;
2582 int i, ret = 0, brw_flags = 0;
2586 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2587 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
2590 npages = batch >> CFS_PAGE_SHIFT;
2591 tot_pages = count >> CFS_PAGE_SHIFT;
2593 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2594 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2596 if (lnb == NULL || rnb == NULL)
2597 GOTO(out, ret = -ENOMEM);
2599 if (rw == OBD_BRW_WRITE && async)
2600 brw_flags |= OBD_BRW_ASYNC;
2602 obdo_to_ioobj(oa, &ioo);
2606 for(; tot_pages; tot_pages -= npages) {
2609 if (tot_pages < npages)
2612 for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) {
2613 rnb[i].offset = off;
2614 rnb[i].len = CFS_PAGE_SIZE;
2615 rnb[i].flags = brw_flags;
2618 ioo.ioo_bufcnt = npages;
2619 oti->oti_transno = 0;
2622 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2626 LASSERT(lpages == npages);
2628 for (i = 0; i < lpages; i++) {
2629 cfs_page_t *page = lnb[i].page;
2631 /* read past eof? */
2632 if (page == NULL && lnb[i].rc == 0)
2636 lnb[i].flags |= OBD_BRW_ASYNC;
2638 if (oa->o_id == ECHO_PERSISTENT_OBJID ||
2639 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2640 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2643 if (rw == OBD_BRW_WRITE)
2644 echo_client_page_debug_setup(lsm, page, rw,
2649 echo_client_page_debug_check(lsm, page,
2655 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2656 rnb, npages, lnb, oti, ret);
2660 /* Reset oti otherwise it would confuse ldiskfs. */
2661 memset(oti, 0, sizeof(*oti));
2666 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2668 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2672 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2673 struct obd_export *exp,
2674 struct obd_ioctl_data *data,
2675 struct obd_trans_info *dummy_oti)
2677 struct obd_device *obd = class_exp2obd(exp);
2678 struct echo_device *ed = obd2echo_dev(obd);
2679 struct echo_client_obd *ec = ed->ed_ec;
2680 struct obdo *oa = &data->ioc_obdo1;
2681 struct echo_object *eco;
2687 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2689 rc = echo_get_object(&eco, ed, oa);
2693 oa->o_valid &= ~OBD_MD_FLHANDLE;
2695 /* OFD/obdfilter works only via prep/commit */
2696 test_mode = (long)data->ioc_pbuf1;
2700 if (ed->ed_next == NULL && test_mode != 3) {
2702 data->ioc_plen1 = data->ioc_count;
2705 /* Truncate batch size to maximum */
2706 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2707 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2709 switch (test_mode) {
2713 rc = echo_client_kbrw(ed, rw, oa,
2714 eco, data->ioc_offset,
2715 data->ioc_count, async, dummy_oti);
2718 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2719 eco, data->ioc_offset,
2720 data->ioc_count, data->ioc_plen1,
2726 echo_put_object(eco);
2731 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2732 int mode, obd_off offset, obd_size nob)
2734 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2735 struct lustre_handle *ulh = &oa->o_handle;
2736 struct echo_object *eco;
2741 if (ed->ed_next == NULL)
2742 RETURN(-EOPNOTSUPP);
2744 if (!(mode == LCK_PR || mode == LCK_PW))
2747 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2748 (nob & (~CFS_PAGE_MASK)) != 0)
2751 rc = echo_get_object (&eco, ed, oa);
2755 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2756 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2758 oa->o_valid |= OBD_MD_FLHANDLE;
2759 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2761 echo_put_object(eco);
2766 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2768 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2769 __u64 cookie = oa->o_handle.cookie;
2771 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2774 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2775 return cl_echo_cancel(ed, cookie);
2779 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2780 void *karg, void *uarg)
2782 struct obd_device *obd = exp->exp_obd;
2783 struct echo_device *ed = obd2echo_dev(obd);
2784 struct echo_client_obd *ec = ed->ed_ec;
2785 struct echo_object *eco;
2786 struct obd_ioctl_data *data = karg;
2787 struct obd_trans_info dummy_oti;
2789 struct oti_req_ack_lock *ack_lock;
2792 int rw = OBD_BRW_READ;
2797 memset(&dummy_oti, 0, sizeof(dummy_oti));
2799 oa = &data->ioc_obdo1;
2800 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2801 oa->o_valid |= OBD_MD_FLGROUP;
2802 oa->o_seq = FID_SEQ_ECHO;
2805 /* This FID is unpacked just for validation at this point */
2806 rc = fid_ostid_unpack(&fid, &oa->o_oi, 0);
2814 rc = lu_env_init(env, LCT_DT_THREAD);
2816 GOTO(out, rc = -ENOMEM);
2819 case OBD_IOC_CREATE: /* may create echo object */
2820 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2821 GOTO (out, rc = -EPERM);
2823 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2824 data->ioc_plen1, &dummy_oti);
2827 case OBD_IOC_ECHO_MD: {
2834 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2835 GOTO(out, rc = -EPERM);
2837 count = data->ioc_count;
2838 cmd = data->ioc_command;
2840 id = data->ioc_obdo2.o_id;
2842 dirlen = data->ioc_plen1;
2843 OBD_ALLOC(dir, dirlen + 1);
2845 GOTO(out, rc = -ENOMEM);
2847 if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2848 OBD_FREE(dir, data->ioc_plen1 + 1);
2849 GOTO(out, rc = -EFAULT);
2852 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2853 OBD_FREE(dir, dirlen + 1);
2856 case OBD_IOC_ECHO_ALLOC_SEQ: {
2857 struct lu_env *cl_env;
2862 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2863 GOTO(out, rc = -EPERM);
2865 cl_env = cl_env_get(&refcheck);
2867 GOTO(out, rc = PTR_ERR(cl_env));
2869 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2872 cl_env_put(cl_env, &refcheck);
2876 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2877 cl_env_put(cl_env, &refcheck);
2879 CERROR("%s: Can not alloc seq: rc = %d\n",
2884 if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2887 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2888 if (cfs_copy_to_user(data->ioc_pbuf2, &max_count,
2893 case OBD_IOC_DESTROY:
2894 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2895 GOTO (out, rc = -EPERM);
2897 rc = echo_get_object(&eco, ed, oa);
2899 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2900 &dummy_oti, NULL, NULL);
2902 eco->eo_deleted = 1;
2903 echo_put_object(eco);
2907 case OBD_IOC_GETATTR:
2908 rc = echo_get_object(&eco, ed, oa);
2910 struct obd_info oinfo = { { { 0 } } };
2911 oinfo.oi_md = eco->eo_lsm;
2913 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2914 echo_put_object(eco);
2918 case OBD_IOC_SETATTR:
2919 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2920 GOTO (out, rc = -EPERM);
2922 rc = echo_get_object(&eco, ed, oa);
2924 struct obd_info oinfo = { { { 0 } } };
2926 oinfo.oi_md = eco->eo_lsm;
2928 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2929 echo_put_object(eco);
2933 case OBD_IOC_BRW_WRITE:
2934 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2935 GOTO (out, rc = -EPERM);
2939 case OBD_IOC_BRW_READ:
2940 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2943 case ECHO_IOC_GET_STRIPE:
2944 rc = echo_get_object(&eco, ed, oa);
2946 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2948 echo_put_object(eco);
2952 case ECHO_IOC_SET_STRIPE:
2953 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2954 GOTO (out, rc = -EPERM);
2956 if (data->ioc_pbuf1 == NULL) { /* unset */
2957 rc = echo_get_object(&eco, ed, oa);
2959 eco->eo_deleted = 1;
2960 echo_put_object(eco);
2963 rc = echo_create_object(env, ed, 0, oa,
2965 data->ioc_plen1, &dummy_oti);
2969 case ECHO_IOC_ENQUEUE:
2970 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2971 GOTO (out, rc = -EPERM);
2973 rc = echo_client_enqueue(exp, oa,
2974 data->ioc_conn1, /* lock mode */
2976 data->ioc_count);/*extent*/
2979 case ECHO_IOC_CANCEL:
2980 rc = echo_client_cancel(exp, oa);
2984 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2985 GOTO (out, rc = -ENOTTY);
2993 /* XXX this should be in a helper also called by target_send_reply */
2994 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2996 if (!ack_lock->mode)
2998 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
3004 static int echo_client_setup(const struct lu_env *env,
3005 struct obd_device *obddev, struct lustre_cfg *lcfg)
3007 struct echo_client_obd *ec = &obddev->u.echo_client;
3008 struct obd_device *tgt;
3009 struct obd_uuid echo_uuid = { "ECHO_UUID" };
3010 struct obd_connect_data *ocd = NULL;
3014 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3015 CERROR("requires a TARGET OBD name\n");
3019 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3020 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3021 CERROR("device not attached or not set up (%s)\n",
3022 lustre_cfg_string(lcfg, 1));
3026 spin_lock_init(&ec->ec_lock);
3027 CFS_INIT_LIST_HEAD (&ec->ec_objects);
3028 CFS_INIT_LIST_HEAD (&ec->ec_locks);
3030 ec->ec_nstripes = 0;
3032 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3033 lu_context_tags_update(ECHO_MD_CTX_TAG);
3034 lu_session_tags_update(ECHO_MD_SES_TAG);
3038 OBD_ALLOC(ocd, sizeof(*ocd));
3040 CERROR("Can't alloc ocd connecting to %s\n",
3041 lustre_cfg_string(lcfg, 1));
3045 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3046 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3047 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE;
3048 ocd->ocd_version = LUSTRE_VERSION_CODE;
3049 ocd->ocd_group = FID_SEQ_ECHO;
3051 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3053 /* Turn off pinger because it connects to tgt obd directly. */
3054 spin_lock(&tgt->obd_dev_lock);
3055 cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3056 spin_unlock(&tgt->obd_dev_lock);
3059 OBD_FREE(ocd, sizeof(*ocd));
3062 CERROR("fail to connect to device %s\n",
3063 lustre_cfg_string(lcfg, 1));
3070 static int echo_client_cleanup(struct obd_device *obddev)
3072 struct echo_device *ed = obd2echo_dev(obddev);
3073 struct echo_client_obd *ec = &obddev->u.echo_client;
3077 /*Do nothing for Metadata echo client*/
3081 if (ed->ed_next_ismd) {
3082 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3083 lu_session_tags_clear(ECHO_MD_SES_TAG);
3087 if (!cfs_list_empty(&obddev->obd_exports)) {
3088 CERROR("still has clients!\n");
3092 LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0);
3093 rc = obd_disconnect(ec->ec_exp);
3095 CERROR("fail to disconnect device: %d\n", rc);
3100 static int echo_client_connect(const struct lu_env *env,
3101 struct obd_export **exp,
3102 struct obd_device *src, struct obd_uuid *cluuid,
3103 struct obd_connect_data *data, void *localdata)
3106 struct lustre_handle conn = { 0 };
3109 rc = class_connect(&conn, src, cluuid);
3111 *exp = class_conn2export(&conn);
3117 static int echo_client_disconnect(struct obd_export *exp)
3120 struct obd_device *obd;
3121 struct echo_client_obd *ec;
3122 struct ec_lock *ecl;
3128 GOTO(out, rc = -EINVAL);
3132 ec = &obd->u.echo_client;
3134 /* no more contention on export's lock list */
3135 while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) {
3136 ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next,
3137 struct ec_lock, ecl_exp_chain);
3138 cfs_list_del (&ecl->ecl_exp_chain);
3140 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3141 ecl->ecl_mode, &ecl->ecl_lock_handle);
3143 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3144 "(%d)\n", ecl->ecl_object->eco_id, rc);
3146 echo_put_object (ecl->ecl_object);
3147 OBD_FREE (ecl, sizeof (*ecl));
3151 rc = class_disconnect(exp);
3157 static struct obd_ops echo_client_obd_ops = {
3158 .o_owner = THIS_MODULE,
3161 .o_setup = echo_client_setup,
3162 .o_cleanup = echo_client_cleanup,
3165 .o_iocontrol = echo_client_iocontrol,
3166 .o_connect = echo_client_connect,
3167 .o_disconnect = echo_client_disconnect
3170 int echo_client_init(void)
3172 struct lprocfs_static_vars lvars = { 0 };
3175 lprocfs_echo_init_vars(&lvars);
3177 rc = lu_kmem_init(echo_caches);
3179 rc = class_register_type(&echo_client_obd_ops, NULL,
3181 LUSTRE_ECHO_CLIENT_NAME,
3184 lu_kmem_fini(echo_caches);
3189 void echo_client_exit(void)
3191 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3192 lu_kmem_fini(echo_caches);
3196 static int __init obdecho_init(void)
3198 struct lprocfs_static_vars lvars;
3202 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3204 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3206 lprocfs_echo_init_vars(&lvars);
3208 # ifdef HAVE_SERVER_SUPPORT
3209 rc = echo_persistent_pages_init();
3213 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
3214 LUSTRE_ECHO_NAME, NULL);
3219 rc = echo_client_init();
3221 # ifdef HAVE_SERVER_SUPPORT
3225 class_unregister_type(LUSTRE_ECHO_NAME);
3227 echo_persistent_pages_fini();
3233 static void /*__exit*/ obdecho_exit(void)
3237 # ifdef HAVE_SERVER_SUPPORT
3238 class_unregister_type(LUSTRE_ECHO_NAME);
3239 echo_persistent_pages_fini();
3243 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3244 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3245 MODULE_LICENSE("GPL");
3247 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3248 #endif /* __KERNEL__ */
3250 /** @} echo_client */