4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_ECHO
38 #include <linux/libcfs/libcfs.h>
41 #include <obd_support.h>
42 #include <obd_class.h>
43 #include <lustre_debug.h>
44 #include <lprocfs_status.h>
45 #include <cl_object.h>
46 #include <lustre_fid.h>
47 #include <lustre_acl.h>
48 #include <lustre_net.h>
51 #include "echo_internal.h"
53 /** \defgroup echo_client Echo Client
58 struct cl_device ed_cl;
59 struct echo_client_obd *ed_ec;
61 struct cl_site ed_site_myself;
62 struct cl_site *ed_site;
63 struct lu_device *ed_next;
66 struct lu_client_seq *ed_cl_seq;
70 struct cl_object eo_cl;
71 struct cl_object_header eo_hdr;
73 struct echo_device *eo_dev;
74 struct list_head eo_obj_chain;
75 struct lov_stripe_md *eo_lsm;
80 struct echo_object_conf {
81 struct cl_object_conf eoc_cl;
82 struct lov_stripe_md **eoc_md;
86 struct cl_page_slice ep_cl;
88 struct page *ep_vmpage;
92 struct cl_lock_slice el_cl;
93 struct list_head el_chain;
94 struct echo_object *el_object;
100 struct cl_io_slice ei_cl;
105 struct cl_req_slice er_cl;
109 static int echo_client_setup(const struct lu_env *env,
110 struct obd_device *obddev,
111 struct lustre_cfg *lcfg);
112 static int echo_client_cleanup(struct obd_device *obddev);
115 /** \defgroup echo_helpers Helper functions
118 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
120 return container_of0(dev, struct echo_device, ed_cl);
123 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
128 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
130 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
133 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
138 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
140 return container_of(o, struct echo_object, eo_cl);
143 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
145 return container_of(s, struct echo_page, ep_cl);
148 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
150 return container_of(s, struct echo_lock, el_cl);
153 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
155 return ecl->el_cl.cls_lock;
158 static struct lu_context_key echo_thread_key;
159 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
161 struct echo_thread_info *info;
162 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
163 LASSERT(info != NULL);
168 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
170 return container_of(c, struct echo_object_conf, eoc_cl);
173 /** @} echo_helpers */
175 static struct echo_object *cl_echo_object_find(struct echo_device *d,
176 struct lov_stripe_md **lsm);
177 static int cl_echo_object_put(struct echo_object *eco);
178 static int cl_echo_enqueue (struct echo_object *eco, obd_off start,
179 obd_off end, int mode, __u64 *cookie);
180 static int cl_echo_cancel (struct echo_device *d, __u64 cookie);
181 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
182 struct page **pages, int npages, int async);
184 static struct echo_thread_info *echo_env_info(const struct lu_env *env);
186 struct echo_thread_info {
187 struct echo_object_conf eti_conf;
188 struct lustre_md eti_md;
190 struct cl_2queue eti_queue;
192 struct cl_lock_descr eti_descr;
193 struct lu_fid eti_fid;
194 struct lu_fid eti_fid2;
195 struct md_op_spec eti_spec;
196 struct lov_mds_md_v3 eti_lmm;
197 struct lov_user_md_v3 eti_lum;
198 struct md_attr eti_ma;
199 struct lu_name eti_lname;
200 /* per-thread values, can be re-used */
204 struct lu_buf eti_buf;
205 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
208 /* No session used right now */
209 struct echo_session_info {
213 static struct kmem_cache *echo_lock_kmem;
214 static struct kmem_cache *echo_object_kmem;
215 static struct kmem_cache *echo_thread_kmem;
216 static struct kmem_cache *echo_session_kmem;
217 //static struct kmem_cache *echo_req_kmem;
219 static struct lu_kmem_descr echo_caches[] = {
221 .ckd_cache = &echo_lock_kmem,
222 .ckd_name = "echo_lock_kmem",
223 .ckd_size = sizeof (struct echo_lock)
226 .ckd_cache = &echo_object_kmem,
227 .ckd_name = "echo_object_kmem",
228 .ckd_size = sizeof (struct echo_object)
231 .ckd_cache = &echo_thread_kmem,
232 .ckd_name = "echo_thread_kmem",
233 .ckd_size = sizeof (struct echo_thread_info)
236 .ckd_cache = &echo_session_kmem,
237 .ckd_name = "echo_session_kmem",
238 .ckd_size = sizeof (struct echo_session_info)
242 .ckd_cache = &echo_req_kmem,
243 .ckd_name = "echo_req_kmem",
244 .ckd_size = sizeof (struct echo_req)
252 /** \defgroup echo_page Page operations
254 * Echo page operations.
258 static struct page *echo_page_vmpage(const struct lu_env *env,
259 const struct cl_page_slice *slice)
261 return cl2echo_page(slice)->ep_vmpage;
264 static int echo_page_own(const struct lu_env *env,
265 const struct cl_page_slice *slice,
266 struct cl_io *io, int nonblock)
268 struct echo_page *ep = cl2echo_page(slice);
271 mutex_lock(&ep->ep_lock);
272 else if (!mutex_trylock(&ep->ep_lock))
277 static void echo_page_disown(const struct lu_env *env,
278 const struct cl_page_slice *slice,
281 struct echo_page *ep = cl2echo_page(slice);
283 LASSERT(mutex_is_locked(&ep->ep_lock));
284 mutex_unlock(&ep->ep_lock);
287 static void echo_page_discard(const struct lu_env *env,
288 const struct cl_page_slice *slice,
289 struct cl_io *unused)
291 cl_page_delete(env, slice->cpl_page);
294 static int echo_page_is_vmlocked(const struct lu_env *env,
295 const struct cl_page_slice *slice)
297 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
302 static void echo_page_completion(const struct lu_env *env,
303 const struct cl_page_slice *slice,
306 LASSERT(slice->cpl_page->cp_sync_io != NULL);
309 static void echo_page_fini(const struct lu_env *env,
310 struct cl_page_slice *slice)
312 struct echo_page *ep = cl2echo_page(slice);
313 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
314 struct page *vmpage = ep->ep_vmpage;
317 atomic_dec(&eco->eo_npages);
318 page_cache_release(vmpage);
322 static int echo_page_prep(const struct lu_env *env,
323 const struct cl_page_slice *slice,
324 struct cl_io *unused)
329 static int echo_page_print(const struct lu_env *env,
330 const struct cl_page_slice *slice,
331 void *cookie, lu_printer_t printer)
333 struct echo_page *ep = cl2echo_page(slice);
335 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
336 ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage);
340 static const struct cl_page_operations echo_page_ops = {
341 .cpo_own = echo_page_own,
342 .cpo_disown = echo_page_disown,
343 .cpo_discard = echo_page_discard,
344 .cpo_vmpage = echo_page_vmpage,
345 .cpo_fini = echo_page_fini,
346 .cpo_print = echo_page_print,
347 .cpo_is_vmlocked = echo_page_is_vmlocked,
350 .cpo_prep = echo_page_prep,
351 .cpo_completion = echo_page_completion,
354 .cpo_prep = echo_page_prep,
355 .cpo_completion = echo_page_completion,
361 /** \defgroup echo_lock Locking
363 * echo lock operations
367 static void echo_lock_fini(const struct lu_env *env,
368 struct cl_lock_slice *slice)
370 struct echo_lock *ecl = cl2echo_lock(slice);
372 LASSERT(list_empty(&ecl->el_chain));
373 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
376 static void echo_lock_delete(const struct lu_env *env,
377 const struct cl_lock_slice *slice)
379 struct echo_lock *ecl = cl2echo_lock(slice);
381 LASSERT(list_empty(&ecl->el_chain));
384 static int echo_lock_fits_into(const struct lu_env *env,
385 const struct cl_lock_slice *slice,
386 const struct cl_lock_descr *need,
387 const struct cl_io *unused)
392 static struct cl_lock_operations echo_lock_ops = {
393 .clo_fini = echo_lock_fini,
394 .clo_delete = echo_lock_delete,
395 .clo_fits_into = echo_lock_fits_into
400 /** \defgroup echo_cl_ops cl_object operations
402 * operations for cl_object
406 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
407 struct cl_page *page, struct page *vmpage)
409 struct echo_page *ep = cl_object_page_slice(obj, page);
410 struct echo_object *eco = cl2echo_obj(obj);
413 ep->ep_vmpage = vmpage;
414 page_cache_get(vmpage);
415 mutex_init(&ep->ep_lock);
416 cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops);
417 atomic_inc(&eco->eo_npages);
421 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
427 static int echo_lock_init(const struct lu_env *env,
428 struct cl_object *obj, struct cl_lock *lock,
429 const struct cl_io *unused)
431 struct echo_lock *el;
434 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, __GFP_IO);
436 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
437 el->el_object = cl2echo_obj(obj);
438 INIT_LIST_HEAD(&el->el_chain);
439 atomic_set(&el->el_refcount, 0);
441 RETURN(el == NULL ? -ENOMEM : 0);
444 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
445 const struct cl_object_conf *conf)
450 static const struct cl_object_operations echo_cl_obj_ops = {
451 .coo_page_init = echo_page_init,
452 .coo_lock_init = echo_lock_init,
453 .coo_io_init = echo_io_init,
454 .coo_conf_set = echo_conf_set
456 /** @} echo_cl_ops */
458 /** \defgroup echo_lu_ops lu_object operations
460 * operations for echo lu object.
464 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
465 const struct lu_object_conf *conf)
467 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
468 struct echo_client_obd *ec = ed->ed_ec;
469 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
473 struct lu_object *below;
474 struct lu_device *under;
477 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
481 lu_object_add(obj, below);
484 if (!ed->ed_next_ismd) {
485 const struct cl_object_conf *cconf = lu2cl_conf(conf);
486 struct echo_object_conf *econf = cl2echo_conf(cconf);
488 LASSERT(econf->eoc_md);
489 eco->eo_lsm = *econf->eoc_md;
490 /* clear the lsm pointer so that it won't get freed. */
491 *econf->eoc_md = NULL;
497 atomic_set(&eco->eo_npages, 0);
498 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
500 spin_lock(&ec->ec_lock);
501 list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
502 spin_unlock(&ec->ec_lock);
507 /* taken from osc_unpackmd() */
508 static int echo_alloc_memmd(struct echo_device *ed,
509 struct lov_stripe_md **lsmp)
515 /* If export is lov/osc then use their obd method */
516 if (ed->ed_next != NULL)
517 return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp);
518 /* OFD has no unpackmd method, do everything here */
519 lsm_size = lov_stripe_md_size(1);
521 LASSERT(*lsmp == NULL);
522 OBD_ALLOC(*lsmp, lsm_size);
526 OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
527 if ((*lsmp)->lsm_oinfo[0] == NULL) {
528 OBD_FREE(*lsmp, lsm_size);
532 loi_init((*lsmp)->lsm_oinfo[0]);
533 (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
534 ostid_set_seq_echo(&(*lsmp)->lsm_oi);
539 static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp)
545 /* If export is lov/osc then use their obd method */
546 if (ed->ed_next != NULL)
547 return obd_free_memmd(ed->ed_ec->ec_exp, lsmp);
548 /* OFD has no unpackmd method, do everything here */
549 lsm_size = lov_stripe_md_size(1);
551 LASSERT(*lsmp != NULL);
552 OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
553 OBD_FREE(*lsmp, lsm_size);
558 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
560 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
561 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
564 LASSERT(atomic_read(&eco->eo_npages) == 0);
566 spin_lock(&ec->ec_lock);
567 list_del_init(&eco->eo_obj_chain);
568 spin_unlock(&ec->ec_lock);
571 lu_object_header_fini(obj->lo_header);
574 echo_free_memmd(eco->eo_dev, &eco->eo_lsm);
575 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
579 static int echo_object_print(const struct lu_env *env, void *cookie,
580 lu_printer_t p, const struct lu_object *o)
582 struct echo_object *obj = cl2echo_obj(lu2cl(o));
584 return (*p)(env, cookie, "echoclient-object@%p", obj);
587 static const struct lu_object_operations echo_lu_obj_ops = {
588 .loo_object_init = echo_object_init,
589 .loo_object_delete = NULL,
590 .loo_object_release = NULL,
591 .loo_object_free = echo_object_free,
592 .loo_object_print = echo_object_print,
593 .loo_object_invariant = NULL
595 /** @} echo_lu_ops */
597 /** \defgroup echo_lu_dev_ops lu_device operations
599 * Operations for echo lu device.
603 static struct lu_object *echo_object_alloc(const struct lu_env *env,
604 const struct lu_object_header *hdr,
605 struct lu_device *dev)
607 struct echo_object *eco;
608 struct lu_object *obj = NULL;
611 /* we're the top dev. */
612 LASSERT(hdr == NULL);
613 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, __GFP_IO);
615 struct cl_object_header *hdr = &eco->eo_hdr;
617 obj = &echo_obj2cl(eco)->co_lu;
618 cl_object_header_init(hdr);
619 lu_object_init(obj, &hdr->coh_lu, dev);
620 lu_object_add_top(&hdr->coh_lu, obj);
622 eco->eo_cl.co_ops = &echo_cl_obj_ops;
623 obj->lo_ops = &echo_lu_obj_ops;
628 static struct lu_device_operations echo_device_lu_ops = {
629 .ldo_object_alloc = echo_object_alloc,
632 /** @} echo_lu_dev_ops */
634 static struct cl_device_operations echo_device_cl_ops = {
637 /** \defgroup echo_init Setup and teardown
639 * Init and fini functions for echo client.
643 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
645 struct cl_site *site = &ed->ed_site_myself;
648 /* initialize site */
649 rc = cl_site_init(site, &ed->ed_cl);
651 CERROR("Cannot initilize site for echo client(%d)\n", rc);
655 rc = lu_site_init_finish(&site->cs_lu);
663 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
666 if (!ed->ed_next_ismd)
667 cl_site_fini(ed->ed_site);
672 static void *echo_thread_key_init(const struct lu_context *ctx,
673 struct lu_context_key *key)
675 struct echo_thread_info *info;
677 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, __GFP_IO);
679 info = ERR_PTR(-ENOMEM);
683 static void echo_thread_key_fini(const struct lu_context *ctx,
684 struct lu_context_key *key, void *data)
686 struct echo_thread_info *info = data;
687 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
690 static void echo_thread_key_exit(const struct lu_context *ctx,
691 struct lu_context_key *key, void *data)
695 static struct lu_context_key echo_thread_key = {
696 .lct_tags = LCT_CL_THREAD,
697 .lct_init = echo_thread_key_init,
698 .lct_fini = echo_thread_key_fini,
699 .lct_exit = echo_thread_key_exit
702 static void *echo_session_key_init(const struct lu_context *ctx,
703 struct lu_context_key *key)
705 struct echo_session_info *session;
707 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, __GFP_IO);
709 session = ERR_PTR(-ENOMEM);
713 static void echo_session_key_fini(const struct lu_context *ctx,
714 struct lu_context_key *key, void *data)
716 struct echo_session_info *session = data;
717 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
720 static void echo_session_key_exit(const struct lu_context *ctx,
721 struct lu_context_key *key, void *data)
725 static struct lu_context_key echo_session_key = {
726 .lct_tags = LCT_SESSION,
727 .lct_init = echo_session_key_init,
728 .lct_fini = echo_session_key_fini,
729 .lct_exit = echo_session_key_exit
732 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
734 #define ECHO_SEQ_WIDTH 0xffffffff
735 static int echo_fid_init(struct echo_device *ed, char *obd_name,
736 struct seq_server_site *ss)
742 OBD_ALLOC_PTR(ed->ed_cl_seq);
743 if (ed->ed_cl_seq == NULL)
746 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
748 GOTO(out_free_seq, rc = -ENOMEM);
750 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
752 /* Init client side sequence-manager */
753 rc = seq_client_init(ed->ed_cl_seq, NULL,
755 prefix, ss->ss_server_seq);
756 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
757 OBD_FREE(prefix, MAX_OBD_NAME + 5);
759 GOTO(out_free_seq, rc);
764 OBD_FREE_PTR(ed->ed_cl_seq);
765 ed->ed_cl_seq = NULL;
769 static int echo_fid_fini(struct obd_device *obddev)
771 struct echo_device *ed = obd2echo_dev(obddev);
774 if (ed->ed_cl_seq != NULL) {
775 seq_client_fini(ed->ed_cl_seq);
776 OBD_FREE_PTR(ed->ed_cl_seq);
777 ed->ed_cl_seq = NULL;
783 static struct lu_device *echo_device_alloc(const struct lu_env *env,
784 struct lu_device_type *t,
785 struct lustre_cfg *cfg)
787 struct lu_device *next;
788 struct echo_device *ed;
789 struct cl_device *cd;
790 struct obd_device *obd = NULL; /* to keep compiler happy */
791 struct obd_device *tgt;
792 const char *tgt_type_name;
799 GOTO(out, rc = -ENOMEM);
803 rc = cl_device_init(cd, t);
807 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
808 cd->cd_ops = &echo_device_cl_ops;
811 obd = class_name2obd(lustre_cfg_string(cfg, 0));
812 LASSERT(obd != NULL);
813 LASSERT(env != NULL);
815 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
817 CERROR("Can not find tgt device %s\n",
818 lustre_cfg_string(cfg, 1));
819 GOTO(out, rc = -ENODEV);
822 next = tgt->obd_lu_dev;
823 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
824 ed->ed_next_ismd = 1;
826 ed->ed_next_ismd = 0;
827 rc = echo_site_init(env, ed);
833 rc = echo_client_setup(env, obd, cfg);
837 ed->ed_ec = &obd->u.echo_client;
840 if (ed->ed_next_ismd) {
841 /* Suppose to connect to some Metadata layer */
843 struct lu_device *ld;
847 CERROR("%s is not lu device type!\n",
848 lustre_cfg_string(cfg, 1));
849 GOTO(out, rc = -EINVAL);
852 tgt_type_name = lustre_cfg_string(cfg, 2);
853 if (!tgt_type_name) {
854 CERROR("%s no type name for echo %s setup\n",
855 lustre_cfg_string(cfg, 1),
856 tgt->obd_type->typ_name);
857 GOTO(out, rc = -EINVAL);
862 spin_lock(&ls->ls_ld_lock);
863 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
864 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
869 spin_unlock(&ls->ls_ld_lock);
872 CERROR("%s is not lu device type!\n",
873 lustre_cfg_string(cfg, 1));
874 GOTO(out, rc = -EINVAL);
878 /* For MD echo client, it will use the site in MDS stack */
879 ed->ed_site_myself.cs_lu = *ls;
880 ed->ed_site = &ed->ed_site_myself;
881 ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu;
882 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
884 CERROR("echo fid init error %d\n", rc);
888 /* if echo client is to be stacked upon ost device, the next is
889 * NULL since ost is not a clio device so far */
890 if (next != NULL && !lu_device_is_cl(next))
893 tgt_type_name = tgt->obd_type->typ_name;
895 LASSERT(next != NULL);
896 if (next->ld_site != NULL)
897 GOTO(out, rc = -EBUSY);
899 next->ld_site = &ed->ed_site->cs_lu;
900 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
901 next->ld_type->ldt_name,
906 /* Tricky case, I have to determine the obd type since
907 * CLIO uses the different parameters to initialize
908 * objects for lov & osc. */
909 if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0)
910 ed->ed_next_islov = 1;
912 LASSERT(strcmp(tgt_type_name,
913 LUSTRE_OSC_NAME) == 0);
915 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
919 RETURN(&cd->cd_lu_dev);
924 rc2 = echo_client_cleanup(obd);
926 CERROR("Cleanup obd device %s error(%d)\n",
931 echo_site_fini(env, ed);
933 cl_device_fini(&ed->ed_cl);
943 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
944 const char *name, struct lu_device *next)
950 static struct lu_device *echo_device_fini(const struct lu_env *env,
953 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
954 struct lu_device *next = ed->ed_next;
956 while (next && !ed->ed_next_ismd)
957 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
961 static void echo_lock_release(const struct lu_env *env,
962 struct echo_lock *ecl,
965 struct cl_lock *clk = echo_lock2cl(ecl);
969 cl_lock_release(env, clk, "ec enqueue", ecl->el_object);
971 cl_lock_mutex_get(env, clk);
972 cl_lock_cancel(env, clk);
973 cl_lock_delete(env, clk);
974 cl_lock_mutex_put(env, clk);
976 cl_lock_put(env, clk);
979 static struct lu_device *echo_device_free(const struct lu_env *env,
982 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
983 struct echo_client_obd *ec = ed->ed_ec;
984 struct echo_object *eco;
985 struct lu_device *next = ed->ed_next;
987 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
990 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
992 /* check if there are objects still alive.
993 * It shouldn't have any object because lu_site_purge would cleanup
994 * all of cached objects. Anyway, probably the echo device is being
995 * parallelly accessed.
997 spin_lock(&ec->ec_lock);
998 list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1000 spin_unlock(&ec->ec_lock);
1003 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1006 "Waiting for the reference of echo object to be dropped\n");
1008 /* Wait for the last reference to be dropped. */
1009 spin_lock(&ec->ec_lock);
1010 while (!list_empty(&ec->ec_objects)) {
1011 spin_unlock(&ec->ec_lock);
1012 CERROR("echo_client still has objects at cleanup time, "
1013 "wait for 1 second\n");
1014 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
1015 cfs_time_seconds(1));
1016 lu_site_purge(env, &ed->ed_site->cs_lu, -1);
1017 spin_lock(&ec->ec_lock);
1019 spin_unlock(&ec->ec_lock);
1021 LASSERT(list_empty(&ec->ec_locks));
1023 CDEBUG(D_INFO, "No object exists, exiting...\n");
1025 echo_client_cleanup(d->ld_obd);
1026 echo_fid_fini(d->ld_obd);
1027 while (next && !ed->ed_next_ismd)
1028 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1030 LASSERT(ed->ed_site == lu2cl_site(d->ld_site));
1031 echo_site_fini(env, ed);
1032 cl_device_fini(&ed->ed_cl);
1038 static const struct lu_device_type_operations echo_device_type_ops = {
1039 .ldto_init = echo_type_init,
1040 .ldto_fini = echo_type_fini,
1042 .ldto_start = echo_type_start,
1043 .ldto_stop = echo_type_stop,
1045 .ldto_device_alloc = echo_device_alloc,
1046 .ldto_device_free = echo_device_free,
1047 .ldto_device_init = echo_device_init,
1048 .ldto_device_fini = echo_device_fini
1051 static struct lu_device_type echo_device_type = {
1052 .ldt_tags = LU_DEVICE_CL,
1053 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1054 .ldt_ops = &echo_device_type_ops,
1055 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1059 /** \defgroup echo_exports Exported operations
1061 * exporting functions to echo client
1066 /* Interfaces to echo client obd device */
1067 static struct echo_object *cl_echo_object_find(struct echo_device *d,
1068 struct lov_stripe_md **lsmp)
1071 struct echo_thread_info *info;
1072 struct echo_object_conf *conf;
1073 struct lov_stripe_md *lsm;
1074 struct echo_object *eco;
1075 struct cl_object *obj;
1084 LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi));
1085 LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n",
1086 POSTID(&lsm->lsm_oi));
1088 /* Never return an object if the obd is to be freed. */
1089 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1090 RETURN(ERR_PTR(-ENODEV));
1092 env = cl_env_get(&refcheck);
1094 RETURN((void *)env);
1096 info = echo_env_info(env);
1097 conf = &info->eti_conf;
1099 if (!d->ed_next_islov) {
1100 struct lov_oinfo *oinfo = lsm->lsm_oinfo[0];
1101 LASSERT(oinfo != NULL);
1102 oinfo->loi_oi = lsm->lsm_oi;
1103 conf->eoc_cl.u.coc_oinfo = oinfo;
1105 struct lustre_md *md;
1107 memset(md, 0, sizeof *md);
1109 conf->eoc_cl.u.coc_md = md;
1112 conf->eoc_md = lsmp;
1114 fid = &info->eti_fid;
1115 rc = ostid_to_fid(fid, &lsm->lsm_oi, 0);
1117 GOTO(out, eco = ERR_PTR(rc));
1119 /* In the function below, .hs_keycmp resolves to
1120 * lu_obj_hop_keycmp() */
1121 /* coverity[overrun-buffer-val] */
1122 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1124 GOTO(out, eco = (void*)obj);
1126 eco = cl2echo_obj(obj);
1127 if (eco->eo_deleted) {
1128 cl_object_put(env, obj);
1129 eco = ERR_PTR(-EAGAIN);
1133 cl_env_put(env, &refcheck);
1137 static int cl_echo_object_put(struct echo_object *eco)
1140 struct cl_object *obj = echo_obj2cl(eco);
1144 env = cl_env_get(&refcheck);
1146 RETURN(PTR_ERR(env));
1148 /* an external function to kill an object? */
1149 if (eco->eo_deleted) {
1150 struct lu_object_header *loh = obj->co_lu.lo_header;
1151 LASSERT(&eco->eo_hdr == luh2coh(loh));
1152 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1155 cl_object_put(env, obj);
1156 cl_env_put(env, &refcheck);
1160 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1161 obd_off start, obd_off end, int mode,
1162 __u64 *cookie , __u32 enqflags)
1165 struct cl_lock *lck;
1166 struct cl_object *obj;
1167 struct cl_lock_descr *descr;
1168 struct echo_thread_info *info;
1172 info = echo_env_info(env);
1174 descr = &info->eti_descr;
1175 obj = echo_obj2cl(eco);
1177 descr->cld_obj = obj;
1178 descr->cld_start = cl_index(obj, start);
1179 descr->cld_end = cl_index(obj, end);
1180 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1181 descr->cld_enq_flags = enqflags;
1184 lck = cl_lock_request(env, io, descr, "ec enqueue", eco);
1186 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1187 struct echo_lock *el;
1189 rc = cl_wait(env, lck);
1191 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1192 spin_lock(&ec->ec_lock);
1193 if (list_empty(&el->el_chain)) {
1194 list_add(&el->el_chain, &ec->ec_locks);
1195 el->el_cookie = ++ec->ec_unique;
1197 atomic_inc(&el->el_refcount);
1198 *cookie = el->el_cookie;
1199 spin_unlock(&ec->ec_lock);
1201 cl_lock_release(env, lck, "ec enqueue", current);
1207 static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end,
1208 int mode, __u64 *cookie)
1210 struct echo_thread_info *info;
1217 env = cl_env_get(&refcheck);
1219 RETURN(PTR_ERR(env));
1221 info = echo_env_info(env);
1224 io->ci_ignore_layout = 1;
1225 result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco));
1228 LASSERT(result == 0);
1230 result = cl_echo_enqueue0(env, eco, start, end, mode, cookie, 0);
1231 cl_io_fini(env, io);
1235 cl_env_put(env, &refcheck);
1239 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1242 struct echo_client_obd *ec = ed->ed_ec;
1243 struct echo_lock *ecl = NULL;
1244 struct list_head *el;
1245 int found = 0, still_used = 0;
1248 LASSERT(ec != NULL);
1249 spin_lock(&ec->ec_lock);
1250 list_for_each (el, &ec->ec_locks) {
1251 ecl = list_entry (el, struct echo_lock, el_chain);
1252 CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie);
1253 found = (ecl->el_cookie == cookie);
1255 if (atomic_dec_and_test(&ecl->el_refcount))
1256 list_del_init(&ecl->el_chain);
1262 spin_unlock(&ec->ec_lock);
1267 echo_lock_release(env, ecl, still_used);
1271 static int cl_echo_cancel(struct echo_device *ed, __u64 cookie)
1278 env = cl_env_get(&refcheck);
1280 RETURN(PTR_ERR(env));
1282 rc = cl_echo_cancel0(env, ed, cookie);
1284 cl_env_put(env, &refcheck);
1288 static int cl_echo_async_brw(const struct lu_env *env, struct cl_io *io,
1289 enum cl_req_type unused, struct cl_2queue *queue)
1291 struct cl_page *clp;
1292 struct cl_page *temp;
1296 cl_page_list_for_each_safe(clp, temp, &queue->c2_qin) {
1298 rc = cl_page_cache_add(env, io, clp, CRT_WRITE);
1301 result = result ?: rc;
1306 static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset,
1307 struct page **pages, int npages, int async)
1310 struct echo_thread_info *info;
1311 struct cl_object *obj = echo_obj2cl(eco);
1312 struct echo_device *ed = eco->eo_dev;
1313 struct cl_2queue *queue;
1315 struct cl_page *clp;
1316 struct lustre_handle lh = { 0 };
1317 int page_size = cl_page_size(obj);
1323 LASSERT((offset & ~CFS_PAGE_MASK) == 0);
1324 LASSERT(ed->ed_next != NULL);
1325 env = cl_env_get(&refcheck);
1327 RETURN(PTR_ERR(env));
1329 info = echo_env_info(env);
1331 queue = &info->eti_queue;
1333 cl_2queue_init(queue);
1335 io->ci_ignore_layout = 1;
1336 rc = cl_io_init(env, io, CIT_MISC, obj);
1342 rc = cl_echo_enqueue0(env, eco, offset,
1343 offset + npages * PAGE_CACHE_SIZE - 1,
1344 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1347 GOTO(error_lock, rc);
1349 for (i = 0; i < npages; i++) {
1351 clp = cl_page_find(env, obj, cl_index(obj, offset),
1352 pages[i], CPT_TRANSIENT);
1357 LASSERT(clp->cp_type == CPT_TRANSIENT);
1359 rc = cl_page_own(env, io, clp);
1361 LASSERT(clp->cp_state == CPS_FREEING);
1362 cl_page_put(env, clp);
1366 cl_2queue_add(queue, clp);
1368 /* drop the reference count for cl_page_find, so that the page
1369 * will be freed in cl_2queue_fini. */
1370 cl_page_put(env, clp);
1371 cl_page_clip(env, clp, 0, page_size);
1373 offset += page_size;
1377 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1379 async = async && (typ == CRT_WRITE);
1381 rc = cl_echo_async_brw(env, io, typ, queue);
1383 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1384 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1385 async ? "async" : "sync", rc);
1388 cl_echo_cancel0(env, ed, lh.cookie);
1391 cl_2queue_discard(env, io, queue);
1392 cl_2queue_disown(env, io, queue);
1393 cl_2queue_fini(env, queue);
1394 cl_io_fini(env, io);
1396 cl_env_put(env, &refcheck);
1399 /** @} echo_exports */
1402 static obd_id last_object_id;
1405 echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob)
1407 struct lov_stripe_md *ulsm = _ulsm;
1410 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
1414 if (copy_to_user (ulsm, lsm, sizeof(ulsm)))
1417 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1418 if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i],
1419 sizeof(lsm->lsm_oinfo[0])))
1426 echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm,
1427 void *ulsm, int ulsm_nob)
1429 struct echo_client_obd *ec = ed->ed_ec;
1432 if (ulsm_nob < sizeof (*lsm))
1435 if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
1438 if (lsm->lsm_stripe_count > ec->ec_nstripes ||
1439 lsm->lsm_magic != LOV_MAGIC ||
1440 (lsm->lsm_stripe_size & (~CFS_PAGE_MASK)) != 0 ||
1441 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
1445 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1446 if (copy_from_user(lsm->lsm_oinfo[i],
1447 ((struct lov_stripe_md *)ulsm)-> \
1449 sizeof(lsm->lsm_oinfo[0])))
1455 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1458 sprintf(name, LPU64, id);
1459 lname->ln_name = name;
1460 lname->ln_namelen = strlen(name);
1463 /* similar to mdt_attr_get_complex */
1464 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1467 struct echo_thread_info *info = echo_env_info(env);
1472 LASSERT(ma->ma_lmm_size > 0);
1474 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1478 /* big_lmm may need to be grown */
1479 if (info->eti_big_lmmsize < rc) {
1480 int size = size_roundup_power2(rc);
1482 if (info->eti_big_lmmsize > 0) {
1483 /* free old buffer */
1484 LASSERT(info->eti_big_lmm);
1485 OBD_FREE_LARGE(info->eti_big_lmm,
1486 info->eti_big_lmmsize);
1487 info->eti_big_lmm = NULL;
1488 info->eti_big_lmmsize = 0;
1491 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1492 if (info->eti_big_lmm == NULL)
1494 info->eti_big_lmmsize = size;
1496 LASSERT(info->eti_big_lmmsize >= rc);
1498 info->eti_buf.lb_buf = info->eti_big_lmm;
1499 info->eti_buf.lb_len = info->eti_big_lmmsize;
1500 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1504 ma->ma_valid |= MA_LOV;
1505 ma->ma_lmm = info->eti_big_lmm;
1506 ma->ma_lmm_size = rc;
1511 int echo_attr_get_complex(const struct lu_env *env, struct md_object *next,
1514 struct echo_thread_info *info = echo_env_info(env);
1515 struct lu_buf *buf = &info->eti_buf;
1516 umode_t mode = lu_object_attr(&next->mo_lu);
1517 int need = ma->ma_need;
1524 if (need & MA_INODE) {
1525 ma->ma_need = MA_INODE;
1526 rc = mo_attr_get(env, next, ma);
1529 ma->ma_valid |= MA_INODE;
1532 if (need & MA_LOV) {
1533 if (S_ISREG(mode) || S_ISDIR(mode)) {
1534 LASSERT(ma->ma_lmm_size > 0);
1535 buf->lb_buf = ma->ma_lmm;
1536 buf->lb_len = ma->ma_lmm_size;
1537 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1539 ma->ma_lmm_size = rc2;
1540 ma->ma_valid |= MA_LOV;
1541 } else if (rc2 == -ENODATA) {
1543 ma->ma_lmm_size = 0;
1544 } else if (rc2 == -ERANGE) {
1545 rc2 = echo_big_lmm_get(env, next, ma);
1547 GOTO(out, rc = rc2);
1549 GOTO(out, rc = rc2);
1554 #ifdef CONFIG_FS_POSIX_ACL
1555 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1556 buf->lb_buf = ma->ma_acl;
1557 buf->lb_len = ma->ma_acl_size;
1558 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1560 ma->ma_acl_size = rc2;
1561 ma->ma_valid |= MA_ACL_DEF;
1562 } else if (rc2 == -ENODATA) {
1564 ma->ma_acl_size = 0;
1566 GOTO(out, rc = rc2);
1572 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
1573 rc, ma->ma_valid, ma->ma_lmm);
1578 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1579 struct md_object *parent, struct lu_fid *fid,
1580 struct lu_name *lname, struct md_op_spec *spec,
1583 struct lu_object *ec_child, *child;
1584 struct lu_device *ld = ed->ed_next;
1585 struct echo_thread_info *info = echo_env_info(env);
1586 struct lu_fid *fid2 = &info->eti_fid2;
1587 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1592 rc = mdo_lookup(env, parent, lname, fid2, spec);
1595 else if (rc != -ENOENT)
1598 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1600 if (IS_ERR(ec_child)) {
1601 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1603 RETURN(PTR_ERR(ec_child));
1606 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1607 if (child == NULL) {
1608 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1609 GOTO(out_put, rc = -EINVAL);
1612 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1613 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1616 * Do not perform lookup sanity check. We know that name does not exist.
1618 spec->sp_cr_lookup = 0;
1619 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1621 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1624 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1625 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1628 lu_object_put(env, ec_child);
1632 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1635 struct echo_thread_info *info = echo_env_info(env);
1637 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1638 ma->ma_lmm = (void *)&info->eti_lmm;
1639 ma->ma_lmm_size = sizeof(info->eti_lmm);
1641 LASSERT(info->eti_big_lmmsize);
1642 ma->ma_lmm = info->eti_big_lmm;
1643 ma->ma_lmm_size = info->eti_big_lmmsize;
1649 static int echo_create_md_object(const struct lu_env *env,
1650 struct echo_device *ed,
1651 struct lu_object *ec_parent,
1653 char *name, int namelen,
1654 __u64 id, __u32 mode, int count,
1655 int stripe_count, int stripe_offset)
1657 struct lu_object *parent;
1658 struct echo_thread_info *info = echo_env_info(env);
1659 struct lu_name *lname = &info->eti_lname;
1660 struct md_op_spec *spec = &info->eti_spec;
1661 struct md_attr *ma = &info->eti_ma;
1662 struct lu_device *ld = ed->ed_next;
1668 if (ec_parent == NULL)
1670 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1674 memset(ma, 0, sizeof(*ma));
1675 memset(spec, 0, sizeof(*spec));
1676 if (stripe_count != 0) {
1677 spec->sp_cr_flags |= FMODE_WRITE;
1678 echo_set_lmm_size(env, ld, ma);
1679 if (stripe_count != -1) {
1680 struct lov_user_md_v3 *lum = &info->eti_lum;
1682 lum->lmm_magic = LOV_USER_MAGIC_V3;
1683 lum->lmm_stripe_count = stripe_count;
1684 lum->lmm_stripe_offset = stripe_offset;
1685 lum->lmm_pattern = 0;
1686 spec->u.sp_ea.eadata = lum;
1687 spec->u.sp_ea.eadatalen = sizeof(*lum);
1688 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1692 ma->ma_attr.la_mode = mode;
1693 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1694 ma->ma_attr.la_ctime = cfs_time_current_64();
1697 lname->ln_name = name;
1698 lname->ln_namelen = namelen;
1699 /* If name is specified, only create one object by name */
1700 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1705 /* Create multiple object sequenced by id */
1706 for (i = 0; i < count; i++) {
1707 char *tmp_name = info->eti_name;
1709 echo_md_build_name(lname, tmp_name, id);
1711 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1714 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1725 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1726 struct echo_device *ed,
1727 struct md_object *parent,
1728 struct lu_name *lname)
1730 struct echo_thread_info *info = echo_env_info(env);
1731 struct lu_fid *fid = &info->eti_fid;
1732 struct lu_object *child;
1736 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1738 rc = mdo_lookup(env, parent, lname, fid, NULL);
1740 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1741 RETURN(ERR_PTR(rc));
1744 /* In the function below, .hs_keycmp resolves to
1745 * lu_obj_hop_keycmp() */
1746 /* coverity[overrun-buffer-val] */
1747 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1752 static int echo_setattr_object(const struct lu_env *env,
1753 struct echo_device *ed,
1754 struct lu_object *ec_parent,
1755 __u64 id, int count)
1757 struct lu_object *parent;
1758 struct echo_thread_info *info = echo_env_info(env);
1759 struct lu_name *lname = &info->eti_lname;
1760 char *name = info->eti_name;
1761 struct lu_device *ld = ed->ed_next;
1762 struct lu_buf *buf = &info->eti_buf;
1768 if (ec_parent == NULL)
1770 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1774 for (i = 0; i < count; i++) {
1775 struct lu_object *ec_child, *child;
1777 echo_md_build_name(lname, name, id);
1779 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1780 if (IS_ERR(ec_child)) {
1781 CERROR("Can't find child %s: rc = %ld\n",
1782 lname->ln_name, PTR_ERR(ec_child));
1783 RETURN(PTR_ERR(ec_child));
1786 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1787 if (child == NULL) {
1788 CERROR("Can not locate the child %s\n", lname->ln_name);
1789 lu_object_put(env, ec_child);
1794 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1795 PFID(lu_object_fid(child)));
1797 buf->lb_buf = info->eti_xattr_buf;
1798 buf->lb_len = sizeof(info->eti_xattr_buf);
1800 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1801 rc = mo_xattr_set(env, lu2md(child), buf, name,
1804 CERROR("Can not setattr child "DFID": rc = %d\n",
1805 PFID(lu_object_fid(child)), rc);
1806 lu_object_put(env, ec_child);
1809 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1810 PFID(lu_object_fid(child)));
1812 lu_object_put(env, ec_child);
1817 static int echo_getattr_object(const struct lu_env *env,
1818 struct echo_device *ed,
1819 struct lu_object *ec_parent,
1820 __u64 id, int count)
1822 struct lu_object *parent;
1823 struct echo_thread_info *info = echo_env_info(env);
1824 struct lu_name *lname = &info->eti_lname;
1825 char *name = info->eti_name;
1826 struct md_attr *ma = &info->eti_ma;
1827 struct lu_device *ld = ed->ed_next;
1833 if (ec_parent == NULL)
1835 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1839 memset(ma, 0, sizeof(*ma));
1840 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1841 ma->ma_acl = info->eti_xattr_buf;
1842 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1844 for (i = 0; i < count; i++) {
1845 struct lu_object *ec_child, *child;
1848 echo_md_build_name(lname, name, id);
1849 echo_set_lmm_size(env, ld, ma);
1851 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1852 if (IS_ERR(ec_child)) {
1853 CERROR("Can't find child %s: rc = %ld\n",
1854 lname->ln_name, PTR_ERR(ec_child));
1855 RETURN(PTR_ERR(ec_child));
1858 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1859 if (child == NULL) {
1860 CERROR("Can not locate the child %s\n", lname->ln_name);
1861 lu_object_put(env, ec_child);
1865 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1866 PFID(lu_object_fid(child)));
1867 rc = echo_attr_get_complex(env, lu2md(child), ma);
1869 CERROR("Can not getattr child "DFID": rc = %d\n",
1870 PFID(lu_object_fid(child)), rc);
1871 lu_object_put(env, ec_child);
1874 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1875 PFID(lu_object_fid(child)));
1877 lu_object_put(env, ec_child);
1883 static int echo_lookup_object(const struct lu_env *env,
1884 struct echo_device *ed,
1885 struct lu_object *ec_parent,
1886 __u64 id, int count)
1888 struct lu_object *parent;
1889 struct echo_thread_info *info = echo_env_info(env);
1890 struct lu_name *lname = &info->eti_lname;
1891 char *name = info->eti_name;
1892 struct lu_fid *fid = &info->eti_fid;
1893 struct lu_device *ld = ed->ed_next;
1897 if (ec_parent == NULL)
1899 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1903 /*prepare the requests*/
1904 for (i = 0; i < count; i++) {
1905 echo_md_build_name(lname, name, id);
1907 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1908 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1910 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1912 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1915 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1916 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1923 static int echo_md_destroy_internal(const struct lu_env *env,
1924 struct echo_device *ed,
1925 struct md_object *parent,
1926 struct lu_name *lname,
1929 struct lu_device *ld = ed->ed_next;
1930 struct lu_object *ec_child;
1931 struct lu_object *child;
1936 ec_child = echo_md_lookup(env, ed, parent, lname);
1937 if (IS_ERR(ec_child)) {
1938 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1940 RETURN(PTR_ERR(ec_child));
1943 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1944 if (child == NULL) {
1945 CERROR("Can not locate the child %s\n", lname->ln_name);
1946 GOTO(out_put, rc = -EINVAL);
1949 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1950 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1952 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1954 CERROR("Can not unlink child %s: rc = %d\n",
1955 lname->ln_name, rc);
1958 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1959 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1961 lu_object_put(env, ec_child);
1965 static int echo_destroy_object(const struct lu_env *env,
1966 struct echo_device *ed,
1967 struct lu_object *ec_parent,
1968 char *name, int namelen,
1969 __u64 id, __u32 mode,
1972 struct echo_thread_info *info = echo_env_info(env);
1973 struct lu_name *lname = &info->eti_lname;
1974 struct md_attr *ma = &info->eti_ma;
1975 struct lu_device *ld = ed->ed_next;
1976 struct lu_object *parent;
1981 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1985 memset(ma, 0, sizeof(*ma));
1986 ma->ma_attr.la_mode = mode;
1987 ma->ma_attr.la_valid = LA_CTIME;
1988 ma->ma_attr.la_ctime = cfs_time_current_64();
1989 ma->ma_need = MA_INODE;
1993 lname->ln_name = name;
1994 lname->ln_namelen = namelen;
1995 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2000 /*prepare the requests*/
2001 for (i = 0; i < count; i++) {
2002 char *tmp_name = info->eti_name;
2005 echo_md_build_name(lname, tmp_name, id);
2007 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
2010 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2019 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2020 struct echo_device *ed, char *path,
2023 struct lu_device *ld = ed->ed_next;
2024 struct md_device *md = lu2md_dev(ld);
2025 struct echo_thread_info *info = echo_env_info(env);
2026 struct lu_fid *fid = &info->eti_fid;
2027 struct lu_name *lname = &info->eti_lname;
2028 struct lu_object *parent = NULL;
2029 struct lu_object *child = NULL;
2033 /*Only support MDD layer right now*/
2034 rc = md->md_ops->mdo_root_get(env, md, fid);
2036 CERROR("get root error: rc = %d\n", rc);
2037 RETURN(ERR_PTR(rc));
2040 /* In the function below, .hs_keycmp resolves to
2041 * lu_obj_hop_keycmp() */
2042 /* coverity[overrun-buffer-val] */
2043 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2044 if (IS_ERR(parent)) {
2045 CERROR("Can not find the parent "DFID": rc = %ld\n",
2046 PFID(fid), PTR_ERR(parent));
2051 struct lu_object *ld_parent;
2054 e = strsep(&path, "/");
2059 if (!path || path[0] == '\0')
2065 lname->ln_namelen = strlen(e);
2067 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2068 if (ld_parent == NULL) {
2069 lu_object_put(env, parent);
2074 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2075 lu_object_put(env, parent);
2076 if (IS_ERR(child)) {
2077 rc = (int)PTR_ERR(child);
2078 CERROR("lookup %s under parent "DFID": rc = %d\n",
2079 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2086 RETURN(ERR_PTR(rc));
2091 static void echo_ucred_init(struct lu_env *env)
2093 struct lu_ucred *ucred = lu_ucred(env);
2095 ucred->uc_valid = UCRED_INVALID;
2097 ucred->uc_suppgids[0] = -1;
2098 ucred->uc_suppgids[1] = -1;
2100 ucred->uc_uid = ucred->uc_o_uid =
2101 from_kuid(&init_user_ns, current_uid());
2102 ucred->uc_gid = ucred->uc_o_gid =
2103 from_kgid(&init_user_ns, current_gid());
2104 ucred->uc_fsuid = ucred->uc_o_fsuid =
2105 from_kuid(&init_user_ns, current_fsuid());
2106 ucred->uc_fsgid = ucred->uc_o_fsgid =
2107 from_kgid(&init_user_ns, current_fsgid());
2108 ucred->uc_cap = cfs_curproc_cap_pack();
2110 /* remove fs privilege for non-root user. */
2111 if (ucred->uc_fsuid)
2112 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2113 ucred->uc_valid = UCRED_NEW;
2116 static void echo_ucred_fini(struct lu_env *env)
2118 struct lu_ucred *ucred = lu_ucred(env);
2119 ucred->uc_valid = UCRED_INIT;
2122 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2123 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION)
2124 static int echo_md_handler(struct echo_device *ed, int command,
2125 char *path, int path_len, __u64 id, int count,
2126 struct obd_ioctl_data *data)
2128 struct echo_thread_info *info;
2129 struct lu_device *ld = ed->ed_next;
2132 struct lu_object *parent;
2134 int namelen = data->ioc_plen2;
2139 CERROR("MD echo client is not being initialized properly\n");
2143 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2144 CERROR("Only support MDD layer right now!\n");
2148 env = cl_env_get(&refcheck);
2150 RETURN(PTR_ERR(env));
2152 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2156 /* init big_lmm buffer */
2157 info = echo_env_info(env);
2158 LASSERT(info->eti_big_lmm == NULL);
2159 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2160 if (info->eti_big_lmm == NULL)
2161 GOTO(out_env, rc = -ENOMEM);
2162 info->eti_big_lmmsize = MIN_MD_SIZE;
2164 parent = echo_resolve_path(env, ed, path, path_len);
2165 if (IS_ERR(parent)) {
2166 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2168 GOTO(out_free, rc = PTR_ERR(parent));
2172 OBD_ALLOC(name, namelen + 1);
2174 GOTO(out_put, rc = -ENOMEM);
2175 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2176 GOTO(out_name, rc = -EFAULT);
2179 echo_ucred_init(env);
2182 case ECHO_MD_CREATE:
2183 case ECHO_MD_MKDIR: {
2184 struct echo_thread_info *info = echo_env_info(env);
2185 __u32 mode = data->ioc_obdo2.o_mode;
2186 struct lu_fid *fid = &info->eti_fid;
2187 int stripe_count = (int)data->ioc_obdo2.o_misc;
2188 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2190 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2194 /* In the function below, .hs_keycmp resolves to
2195 * lu_obj_hop_keycmp() */
2196 /* coverity[overrun-buffer-val] */
2197 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2198 id, mode, count, stripe_count,
2202 case ECHO_MD_DESTROY:
2203 case ECHO_MD_RMDIR: {
2204 __u32 mode = data->ioc_obdo2.o_mode;
2206 rc = echo_destroy_object(env, ed, parent, name, namelen,
2210 case ECHO_MD_LOOKUP:
2211 rc = echo_lookup_object(env, ed, parent, id, count);
2213 case ECHO_MD_GETATTR:
2214 rc = echo_getattr_object(env, ed, parent, id, count);
2216 case ECHO_MD_SETATTR:
2217 rc = echo_setattr_object(env, ed, parent, id, count);
2220 CERROR("unknown command %d\n", command);
2224 echo_ucred_fini(env);
2228 OBD_FREE(name, namelen + 1);
2230 lu_object_put(env, parent);
2232 LASSERT(info->eti_big_lmm);
2233 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2234 info->eti_big_lmm = NULL;
2235 info->eti_big_lmmsize = 0;
2237 cl_env_put(env, &refcheck);
2241 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2242 int on_target, struct obdo *oa, void *ulsm,
2243 int ulsm_nob, struct obd_trans_info *oti)
2245 struct echo_object *eco;
2246 struct echo_client_obd *ec = ed->ed_ec;
2247 struct lov_stripe_md *lsm = NULL;
2252 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
2253 (on_target || /* set_stripe */
2254 ec->ec_nstripes != 0)) { /* LOV */
2255 CERROR ("No valid oid\n");
2259 rc = echo_alloc_memmd(ed, &lsm);
2261 CERROR("Cannot allocate md: rc = %d\n", rc);
2268 rc = echo_copyin_lsm (ed, lsm, ulsm, ulsm_nob);
2272 if (lsm->lsm_stripe_count == 0)
2273 lsm->lsm_stripe_count = ec->ec_nstripes;
2275 if (lsm->lsm_stripe_size == 0)
2276 lsm->lsm_stripe_size = PAGE_CACHE_SIZE;
2280 /* setup stripes: indices + default ids if required */
2281 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2282 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0)
2283 lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi;
2285 lsm->lsm_oinfo[i]->loi_ost_idx =
2286 (idx + i) % ec->ec_nstripes;
2290 /* setup object ID here for !on_target and LOV hint */
2291 if (oa->o_valid & OBD_MD_FLID) {
2292 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2293 lsm->lsm_oi = oa->o_oi;
2296 if (ostid_id(&lsm->lsm_oi) == 0)
2297 ostid_set_id(&lsm->lsm_oi, ++last_object_id);
2301 /* Only echo objects are allowed to be created */
2302 LASSERT((oa->o_valid & OBD_MD_FLGROUP) &&
2303 (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO));
2304 rc = obd_create(env, ec->ec_exp, oa, &lsm, oti);
2306 CERROR("Cannot create objects: rc = %d\n", rc);
2312 /* See what object ID we were given */
2313 oa->o_oi = lsm->lsm_oi;
2314 oa->o_valid |= OBD_MD_FLID;
2316 eco = cl_echo_object_find(ed, &lsm);
2318 GOTO(failed, rc = PTR_ERR(eco));
2319 cl_echo_object_put(eco);
2321 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2326 obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL);
2328 echo_free_memmd(ed, &lsm);
2330 CERROR("create object failed with: rc = %d\n", rc);
2334 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2337 struct lov_stripe_md *lsm = NULL;
2338 struct echo_object *eco;
2342 if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) {
2343 /* disallow use of object id 0 */
2344 CERROR ("No valid oid\n");
2348 rc = echo_alloc_memmd(ed, &lsm);
2352 lsm->lsm_oi = oa->o_oi;
2353 if (!(oa->o_valid & OBD_MD_FLGROUP))
2354 ostid_set_seq_echo(&lsm->lsm_oi);
2357 eco = cl_echo_object_find(ed, &lsm);
2363 echo_free_memmd(ed, &lsm);
2367 static void echo_put_object(struct echo_object *eco)
2369 if (cl_echo_object_put(eco))
2370 CERROR("echo client: drop an object failed");
2374 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
2376 unsigned long stripe_count;
2377 unsigned long stripe_size;
2378 unsigned long width;
2379 unsigned long woffset;
2383 if (lsm->lsm_stripe_count <= 1)
2387 stripe_size = lsm->lsm_stripe_size;
2388 stripe_count = lsm->lsm_stripe_count;
2390 /* width = # bytes in all stripes */
2391 width = stripe_size * stripe_count;
2393 /* woffset = offset within a width; offset = whole number of widths */
2394 woffset = do_div (offset, width);
2396 stripe_index = woffset / stripe_size;
2398 *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi);
2399 *offp = offset * stripe_size + woffset % stripe_size;
2403 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
2404 struct page *page, int rw, obd_id id,
2405 obd_off offset, obd_off count)
2412 /* no partial pages on the client */
2413 LASSERT(count == PAGE_CACHE_SIZE);
2417 for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2418 if (rw == OBD_BRW_WRITE) {
2419 stripe_off = offset + delta;
2421 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
2423 stripe_off = 0xdeadbeef00c0ffeeULL;
2424 stripe_id = 0xdeadbeef00c0ffeeULL;
2426 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2427 stripe_off, stripe_id);
2433 static int echo_client_page_debug_check(struct lov_stripe_md *lsm,
2434 struct page *page, obd_id id,
2435 obd_off offset, obd_off count)
2444 /* no partial pages on the client */
2445 LASSERT(count == PAGE_CACHE_SIZE);
2449 for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2450 stripe_off = offset + delta;
2452 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
2454 rc2 = block_debug_check("test_brw",
2455 addr + delta, OBD_ECHO_BLOCK_SIZE,
2456 stripe_off, stripe_id);
2458 CERROR ("Error in echo object "LPX64"\n", id);
2467 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2468 struct echo_object *eco, obd_off offset,
2469 obd_size count, int async,
2470 struct obd_trans_info *oti)
2472 struct lov_stripe_md *lsm = eco->eo_lsm;
2474 struct brw_page *pga;
2475 struct brw_page *pgp;
2476 struct page **pages;
2485 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2486 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2487 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2489 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_IOFS : GFP_HIGHUSER;
2491 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2492 LASSERT(lsm != NULL);
2493 LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi));
2496 (count & (~CFS_PAGE_MASK)) != 0)
2499 /* XXX think again with misaligned I/O */
2500 npages = count >> PAGE_CACHE_SHIFT;
2502 if (rw == OBD_BRW_WRITE)
2503 brw_flags = OBD_BRW_ASYNC;
2505 OBD_ALLOC(pga, npages * sizeof(*pga));
2509 OBD_ALLOC(pages, npages * sizeof(*pages));
2510 if (pages == NULL) {
2511 OBD_FREE(pga, npages * sizeof(*pga));
2515 for (i = 0, pgp = pga, off = offset;
2517 i++, pgp++, off += PAGE_CACHE_SIZE) {
2519 LASSERT (pgp->pg == NULL); /* for cleanup */
2522 OBD_PAGE_ALLOC(pgp->pg, gfp_mask);
2523 if (pgp->pg == NULL)
2527 pgp->count = PAGE_CACHE_SIZE;
2529 pgp->flag = brw_flags;
2532 echo_client_page_debug_setup(lsm, pgp->pg, rw,
2533 ostid_id(&oa->o_oi), off,
2537 /* brw mode can only be used at client */
2538 LASSERT(ed->ed_next != NULL);
2539 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2542 if (rc != 0 || rw != OBD_BRW_READ)
2545 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2546 if (pgp->pg == NULL)
2551 vrc = echo_client_page_debug_check(lsm, pgp->pg,
2552 ostid_id(&oa->o_oi),
2553 pgp->off, pgp->count);
2554 if (vrc != 0 && rc == 0)
2557 OBD_PAGE_FREE(pgp->pg);
2559 OBD_FREE(pga, npages * sizeof(*pga));
2560 OBD_FREE(pages, npages * sizeof(*pages));
2564 static int echo_client_prep_commit(const struct lu_env *env,
2565 struct obd_export *exp, int rw,
2566 struct obdo *oa, struct echo_object *eco,
2567 obd_off offset, obd_size count,
2568 obd_size batch, struct obd_trans_info *oti,
2571 struct lov_stripe_md *lsm = eco->eo_lsm;
2572 struct obd_ioobj ioo;
2573 struct niobuf_local *lnb;
2574 struct niobuf_remote *rnb;
2576 obd_size npages, tot_pages;
2577 int i, ret = 0, brw_flags = 0;
2581 if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 ||
2582 (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi)))
2585 npages = batch >> PAGE_CACHE_SHIFT;
2586 tot_pages = count >> PAGE_CACHE_SHIFT;
2588 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
2589 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
2591 if (lnb == NULL || rnb == NULL)
2592 GOTO(out, ret = -ENOMEM);
2594 if (rw == OBD_BRW_WRITE && async)
2595 brw_flags |= OBD_BRW_ASYNC;
2597 obdo_to_ioobj(oa, &ioo);
2601 for(; tot_pages; tot_pages -= npages) {
2604 if (tot_pages < npages)
2607 for (i = 0; i < npages; i++, off += PAGE_CACHE_SIZE) {
2608 rnb[i].offset = off;
2609 rnb[i].len = PAGE_CACHE_SIZE;
2610 rnb[i].flags = brw_flags;
2613 ioo.ioo_bufcnt = npages;
2614 oti->oti_transno = 0;
2617 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages,
2621 LASSERT(lpages == npages);
2623 for (i = 0; i < lpages; i++) {
2624 struct page *page = lnb[i].page;
2626 /* read past eof? */
2627 if (page == NULL && lnb[i].rc == 0)
2631 lnb[i].flags |= OBD_BRW_ASYNC;
2633 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2634 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2635 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2638 if (rw == OBD_BRW_WRITE)
2639 echo_client_page_debug_setup(lsm, page, rw,
2640 ostid_id(&oa->o_oi),
2644 echo_client_page_debug_check(lsm, page,
2645 ostid_id(&oa->o_oi),
2650 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo,
2651 rnb, npages, lnb, oti, ret);
2655 /* Reset oti otherwise it would confuse ldiskfs. */
2656 memset(oti, 0, sizeof(*oti));
2658 /* Reuse env context. */
2659 lu_context_exit((struct lu_context *)&env->le_ctx);
2660 lu_context_enter((struct lu_context *)&env->le_ctx);
2665 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
2667 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
2671 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2672 struct obd_export *exp,
2673 struct obd_ioctl_data *data,
2674 struct obd_trans_info *dummy_oti)
2676 struct obd_device *obd = class_exp2obd(exp);
2677 struct echo_device *ed = obd2echo_dev(obd);
2678 struct echo_client_obd *ec = ed->ed_ec;
2679 struct obdo *oa = &data->ioc_obdo1;
2680 struct echo_object *eco;
2686 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2688 rc = echo_get_object(&eco, ed, oa);
2692 oa->o_valid &= ~OBD_MD_FLHANDLE;
2694 /* OFD/obdfilter works only via prep/commit */
2695 test_mode = (long)data->ioc_pbuf1;
2699 if (ed->ed_next == NULL && test_mode != 3) {
2701 data->ioc_plen1 = data->ioc_count;
2704 /* Truncate batch size to maximum */
2705 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2706 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2708 switch (test_mode) {
2712 rc = echo_client_kbrw(ed, rw, oa,
2713 eco, data->ioc_offset,
2714 data->ioc_count, async, dummy_oti);
2717 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa,
2718 eco, data->ioc_offset,
2719 data->ioc_count, data->ioc_plen1,
2725 echo_put_object(eco);
2730 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
2731 int mode, obd_off offset, obd_size nob)
2733 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2734 struct lustre_handle *ulh = &oa->o_handle;
2735 struct echo_object *eco;
2740 if (ed->ed_next == NULL)
2741 RETURN(-EOPNOTSUPP);
2743 if (!(mode == LCK_PR || mode == LCK_PW))
2746 if ((offset & (~CFS_PAGE_MASK)) != 0 ||
2747 (nob & (~CFS_PAGE_MASK)) != 0)
2750 rc = echo_get_object (&eco, ed, oa);
2754 end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
2755 rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie);
2757 oa->o_valid |= OBD_MD_FLHANDLE;
2758 CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie);
2760 echo_put_object(eco);
2765 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
2767 struct echo_device *ed = obd2echo_dev(exp->exp_obd);
2768 __u64 cookie = oa->o_handle.cookie;
2770 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
2773 CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie);
2774 return cl_echo_cancel(ed, cookie);
2778 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2779 void *karg, void *uarg)
2781 struct obd_device *obd = exp->exp_obd;
2782 struct echo_device *ed = obd2echo_dev(obd);
2783 struct echo_client_obd *ec = ed->ed_ec;
2784 struct echo_object *eco;
2785 struct obd_ioctl_data *data = karg;
2786 struct obd_trans_info dummy_oti;
2788 struct oti_req_ack_lock *ack_lock;
2791 int rw = OBD_BRW_READ;
2796 memset(&dummy_oti, 0, sizeof(dummy_oti));
2798 oa = &data->ioc_obdo1;
2799 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2800 oa->o_valid |= OBD_MD_FLGROUP;
2801 ostid_set_seq_echo(&oa->o_oi);
2804 /* This FID is unpacked just for validation at this point */
2805 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2813 rc = lu_env_init(env, LCT_DT_THREAD);
2815 GOTO(out, rc = -ENOMEM);
2818 case OBD_IOC_CREATE: /* may create echo object */
2819 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2820 GOTO (out, rc = -EPERM);
2822 rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1,
2823 data->ioc_plen1, &dummy_oti);
2826 case OBD_IOC_ECHO_MD: {
2833 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2834 GOTO(out, rc = -EPERM);
2836 count = data->ioc_count;
2837 cmd = data->ioc_command;
2839 id = ostid_id(&data->ioc_obdo2.o_oi);
2841 dirlen = data->ioc_plen1;
2842 OBD_ALLOC(dir, dirlen + 1);
2844 GOTO(out, rc = -ENOMEM);
2846 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2847 OBD_FREE(dir, data->ioc_plen1 + 1);
2848 GOTO(out, rc = -EFAULT);
2851 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2852 OBD_FREE(dir, dirlen + 1);
2855 case OBD_IOC_ECHO_ALLOC_SEQ: {
2856 struct lu_env *cl_env;
2861 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2862 GOTO(out, rc = -EPERM);
2864 cl_env = cl_env_get(&refcheck);
2866 GOTO(out, rc = PTR_ERR(cl_env));
2868 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2871 cl_env_put(cl_env, &refcheck);
2875 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2876 cl_env_put(cl_env, &refcheck);
2878 CERROR("%s: Can not alloc seq: rc = %d\n",
2883 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2886 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2887 if (copy_to_user(data->ioc_pbuf2, &max_count,
2892 case OBD_IOC_DESTROY:
2893 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2894 GOTO (out, rc = -EPERM);
2896 rc = echo_get_object(&eco, ed, oa);
2898 rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm,
2899 &dummy_oti, NULL, NULL);
2901 eco->eo_deleted = 1;
2902 echo_put_object(eco);
2906 case OBD_IOC_GETATTR:
2907 rc = echo_get_object(&eco, ed, oa);
2909 struct obd_info oinfo = { { { 0 } } };
2910 oinfo.oi_md = eco->eo_lsm;
2912 rc = obd_getattr(env, ec->ec_exp, &oinfo);
2913 echo_put_object(eco);
2917 case OBD_IOC_SETATTR:
2918 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2919 GOTO (out, rc = -EPERM);
2921 rc = echo_get_object(&eco, ed, oa);
2923 struct obd_info oinfo = { { { 0 } } };
2925 oinfo.oi_md = eco->eo_lsm;
2927 rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL);
2928 echo_put_object(eco);
2932 case OBD_IOC_BRW_WRITE:
2933 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2934 GOTO (out, rc = -EPERM);
2938 case OBD_IOC_BRW_READ:
2939 rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti);
2942 case ECHO_IOC_GET_STRIPE:
2943 rc = echo_get_object(&eco, ed, oa);
2945 rc = echo_copyout_lsm(eco->eo_lsm, data->ioc_pbuf1,
2947 echo_put_object(eco);
2951 case ECHO_IOC_SET_STRIPE:
2952 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2953 GOTO (out, rc = -EPERM);
2955 if (data->ioc_pbuf1 == NULL) { /* unset */
2956 rc = echo_get_object(&eco, ed, oa);
2958 eco->eo_deleted = 1;
2959 echo_put_object(eco);
2962 rc = echo_create_object(env, ed, 0, oa,
2964 data->ioc_plen1, &dummy_oti);
2968 case ECHO_IOC_ENQUEUE:
2969 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2970 GOTO (out, rc = -EPERM);
2972 rc = echo_client_enqueue(exp, oa,
2973 data->ioc_conn1, /* lock mode */
2975 data->ioc_count);/*extent*/
2978 case ECHO_IOC_CANCEL:
2979 rc = echo_client_cancel(exp, oa);
2983 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2984 GOTO (out, rc = -ENOTTY);
2992 /* XXX this should be in a helper also called by target_send_reply */
2993 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
2995 if (!ack_lock->mode)
2997 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
3003 static int echo_client_setup(const struct lu_env *env,
3004 struct obd_device *obddev, struct lustre_cfg *lcfg)
3006 struct echo_client_obd *ec = &obddev->u.echo_client;
3007 struct obd_device *tgt;
3008 struct obd_uuid echo_uuid = { "ECHO_UUID" };
3009 struct obd_connect_data *ocd = NULL;
3013 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
3014 CERROR("requires a TARGET OBD name\n");
3018 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
3019 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
3020 CERROR("device not attached or not set up (%s)\n",
3021 lustre_cfg_string(lcfg, 1));
3025 spin_lock_init(&ec->ec_lock);
3026 INIT_LIST_HEAD (&ec->ec_objects);
3027 INIT_LIST_HEAD (&ec->ec_locks);
3029 ec->ec_nstripes = 0;
3031 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
3032 lu_context_tags_update(ECHO_MD_CTX_TAG);
3033 lu_session_tags_update(ECHO_MD_SES_TAG);
3037 OBD_ALLOC(ocd, sizeof(*ocd));
3039 CERROR("Can't alloc ocd connecting to %s\n",
3040 lustre_cfg_string(lcfg, 1));
3044 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
3045 OBD_CONNECT_BRW_SIZE |
3046 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
3047 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
3049 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
3050 ocd->ocd_version = LUSTRE_VERSION_CODE;
3051 ocd->ocd_group = FID_SEQ_ECHO;
3053 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
3055 /* Turn off pinger because it connects to tgt obd directly. */
3056 spin_lock(&tgt->obd_dev_lock);
3057 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
3058 spin_unlock(&tgt->obd_dev_lock);
3061 OBD_FREE(ocd, sizeof(*ocd));
3064 CERROR("fail to connect to device %s\n",
3065 lustre_cfg_string(lcfg, 1));
3072 static int echo_client_cleanup(struct obd_device *obddev)
3074 struct echo_device *ed = obd2echo_dev(obddev);
3075 struct echo_client_obd *ec = &obddev->u.echo_client;
3079 /*Do nothing for Metadata echo client*/
3083 if (ed->ed_next_ismd) {
3084 lu_context_tags_clear(ECHO_MD_CTX_TAG);
3085 lu_session_tags_clear(ECHO_MD_SES_TAG);
3089 if (!list_empty(&obddev->obd_exports)) {
3090 CERROR("still has clients!\n");
3094 LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
3095 rc = obd_disconnect(ec->ec_exp);
3097 CERROR("fail to disconnect device: %d\n", rc);
3102 static int echo_client_connect(const struct lu_env *env,
3103 struct obd_export **exp,
3104 struct obd_device *src, struct obd_uuid *cluuid,
3105 struct obd_connect_data *data, void *localdata)
3108 struct lustre_handle conn = { 0 };
3111 rc = class_connect(&conn, src, cluuid);
3113 *exp = class_conn2export(&conn);
3119 static int echo_client_disconnect(struct obd_export *exp)
3122 struct obd_device *obd;
3123 struct echo_client_obd *ec;
3124 struct ec_lock *ecl;
3130 GOTO(out, rc = -EINVAL);
3134 ec = &obd->u.echo_client;
3136 /* no more contention on export's lock list */
3137 while (!list_empty (&exp->exp_ec_data.eced_locks)) {
3138 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
3139 struct ec_lock, ecl_exp_chain);
3140 list_del (&ecl->ecl_exp_chain);
3142 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
3143 ecl->ecl_mode, &ecl->ecl_lock_handle);
3145 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
3146 "(%d)\n", ecl->ecl_object->eco_id, rc);
3148 echo_put_object (ecl->ecl_object);
3149 OBD_FREE (ecl, sizeof (*ecl));
3153 rc = class_disconnect(exp);
3159 static struct obd_ops echo_client_obd_ops = {
3160 .o_owner = THIS_MODULE,
3163 .o_setup = echo_client_setup,
3164 .o_cleanup = echo_client_cleanup,
3167 .o_iocontrol = echo_client_iocontrol,
3168 .o_connect = echo_client_connect,
3169 .o_disconnect = echo_client_disconnect
3172 int echo_client_init(void)
3174 struct lprocfs_static_vars lvars = { 0 };
3177 lprocfs_echo_init_vars(&lvars);
3179 rc = lu_kmem_init(echo_caches);
3181 rc = class_register_type(&echo_client_obd_ops, NULL,
3183 LUSTRE_ECHO_CLIENT_NAME,
3186 lu_kmem_fini(echo_caches);
3191 void echo_client_exit(void)
3193 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3194 lu_kmem_fini(echo_caches);
3197 static int __init obdecho_init(void)
3199 struct lprocfs_static_vars lvars;
3203 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3205 LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3207 lprocfs_echo_init_vars(&lvars);
3210 rc = echo_client_init();
3215 static void /*__exit*/ obdecho_exit(void)
3221 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3222 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
3223 MODULE_LICENSE("GPL");
3225 cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit);
3227 /** @} echo_client */