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, 2015, 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 <linux/user_namespace.h>
40 #ifdef HAVE_UIDGID_HEADER
41 # include <linux/uidgid.h>
43 #include <libcfs/libcfs.h>
46 #include <obd_support.h>
47 #include <obd_class.h>
48 #include <lustre_debug.h>
49 #include <lprocfs_status.h>
50 #include <cl_object.h>
51 #include <lustre_fid.h>
52 #include <lustre_acl.h>
53 #include <lustre_ioctl.h>
54 #include <lustre_net.h>
55 #ifdef HAVE_SERVER_SUPPORT
56 # include <md_object.h>
58 #define ETI_NAME_LEN 20
60 #endif /* HAVE_SERVER_SUPPORT */
62 #include "echo_internal.h"
64 /** \defgroup echo_client Echo Client
69 struct cl_device ed_cl;
70 struct echo_client_obd *ed_ec;
72 struct cl_site ed_site_myself;
73 struct lu_site *ed_site;
74 struct lu_device *ed_next;
76 struct lu_client_seq *ed_cl_seq;
77 #ifdef HAVE_SERVER_SUPPORT
78 struct local_oid_storage *ed_los;
79 struct lu_fid ed_root_fid;
80 #endif /* HAVE_SERVER_SUPPORT */
84 struct cl_object eo_cl;
85 struct cl_object_header eo_hdr;
86 struct echo_device *eo_dev;
87 struct list_head eo_obj_chain;
88 struct lov_oinfo *eo_oinfo;
93 struct echo_object_conf {
94 struct cl_object_conf eoc_cl;
95 struct lov_oinfo **eoc_oinfo;
99 struct cl_page_slice ep_cl;
100 struct mutex ep_lock;
104 struct cl_lock_slice el_cl;
105 struct list_head el_chain;
106 struct echo_object *el_object;
108 atomic_t el_refcount;
111 #ifdef HAVE_SERVER_SUPPORT
112 static const char echo_md_root_dir_name[] = "ROOT_ECHO";
115 * In order to use the values of members in struct mdd_device,
116 * we define an alias structure here.
118 struct echo_md_device {
119 struct md_device emd_md_dev;
120 struct obd_export *emd_child_exp;
121 struct dt_device *emd_child;
122 struct dt_device *emd_bottom;
123 struct lu_fid emd_root_fid;
124 struct lu_fid emd_local_root_fid;
126 #endif /* HAVE_SERVER_SUPPORT */
128 static int echo_client_setup(const struct lu_env *env,
129 struct obd_device *obddev,
130 struct lustre_cfg *lcfg);
131 static int echo_client_cleanup(struct obd_device *obddev);
134 /** \defgroup echo_helpers Helper functions
137 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
139 return container_of0(dev, struct echo_device, ed_cl);
142 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
147 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
149 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
152 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
157 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
159 return container_of(o, struct echo_object, eo_cl);
162 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
164 return container_of(s, struct echo_page, ep_cl);
167 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
169 return container_of(s, struct echo_lock, el_cl);
172 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
174 return ecl->el_cl.cls_lock;
177 static struct lu_context_key echo_thread_key;
178 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
180 struct echo_thread_info *info;
181 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
182 LASSERT(info != NULL);
187 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
189 return container_of(c, struct echo_object_conf, eoc_cl);
192 #ifdef HAVE_SERVER_SUPPORT
193 static inline struct echo_md_device *lu2emd_dev(struct lu_device *d)
195 return container_of0(d, struct echo_md_device, emd_md_dev.md_lu_dev);
198 static inline struct lu_device *emd2lu_dev(struct echo_md_device *d)
200 return &d->emd_md_dev.md_lu_dev;
203 static inline struct seq_server_site *echo_md_seq_site(struct echo_md_device *d)
205 return emd2lu_dev(d)->ld_site->ld_seq_site;
208 static inline struct obd_device *emd2obd_dev(struct echo_md_device *d)
210 return d->emd_md_dev.md_lu_dev.ld_obd;
212 #endif /* HAVE_SERVER_SUPPORT */
214 /** @} echo_helpers */
216 static int cl_echo_object_put(struct echo_object *eco);
217 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
218 struct page **pages, int npages, int async);
220 struct echo_thread_info {
221 struct echo_object_conf eti_conf;
222 struct lustre_md eti_md;
224 struct cl_2queue eti_queue;
226 struct cl_lock eti_lock;
227 struct lu_fid eti_fid;
228 struct lu_fid eti_fid2;
229 #ifdef HAVE_SERVER_SUPPORT
230 struct md_op_spec eti_spec;
231 struct lov_mds_md_v3 eti_lmm;
232 struct lov_user_md_v3 eti_lum;
233 struct md_attr eti_ma;
234 struct lu_name eti_lname;
235 /* per-thread values, can be re-used */
236 void *eti_big_lmm; /* may be vmalloc'd */
238 char eti_name[ETI_NAME_LEN];
239 struct lu_buf eti_buf;
240 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE];
244 /* No session used right now */
245 struct echo_session_info {
249 static struct kmem_cache *echo_lock_kmem;
250 static struct kmem_cache *echo_object_kmem;
251 static struct kmem_cache *echo_thread_kmem;
252 static struct kmem_cache *echo_session_kmem;
253 /* static struct kmem_cache *echo_req_kmem; */
255 static struct lu_kmem_descr echo_caches[] = {
257 .ckd_cache = &echo_lock_kmem,
258 .ckd_name = "echo_lock_kmem",
259 .ckd_size = sizeof (struct echo_lock)
262 .ckd_cache = &echo_object_kmem,
263 .ckd_name = "echo_object_kmem",
264 .ckd_size = sizeof (struct echo_object)
267 .ckd_cache = &echo_thread_kmem,
268 .ckd_name = "echo_thread_kmem",
269 .ckd_size = sizeof (struct echo_thread_info)
272 .ckd_cache = &echo_session_kmem,
273 .ckd_name = "echo_session_kmem",
274 .ckd_size = sizeof (struct echo_session_info)
281 /** \defgroup echo_page Page operations
283 * Echo page operations.
287 static int echo_page_own(const struct lu_env *env,
288 const struct cl_page_slice *slice,
289 struct cl_io *io, int nonblock)
291 struct echo_page *ep = cl2echo_page(slice);
294 mutex_lock(&ep->ep_lock);
295 else if (!mutex_trylock(&ep->ep_lock))
300 static void echo_page_disown(const struct lu_env *env,
301 const struct cl_page_slice *slice,
304 struct echo_page *ep = cl2echo_page(slice);
306 LASSERT(mutex_is_locked(&ep->ep_lock));
307 mutex_unlock(&ep->ep_lock);
310 static void echo_page_discard(const struct lu_env *env,
311 const struct cl_page_slice *slice,
312 struct cl_io *unused)
314 cl_page_delete(env, slice->cpl_page);
317 static int echo_page_is_vmlocked(const struct lu_env *env,
318 const struct cl_page_slice *slice)
320 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
325 static void echo_page_completion(const struct lu_env *env,
326 const struct cl_page_slice *slice,
329 LASSERT(slice->cpl_page->cp_sync_io != NULL);
332 static void echo_page_fini(const struct lu_env *env,
333 struct cl_page_slice *slice)
335 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
338 atomic_dec(&eco->eo_npages);
339 page_cache_release(slice->cpl_page->cp_vmpage);
343 static int echo_page_prep(const struct lu_env *env,
344 const struct cl_page_slice *slice,
345 struct cl_io *unused)
350 static int echo_page_print(const struct lu_env *env,
351 const struct cl_page_slice *slice,
352 void *cookie, lu_printer_t printer)
354 struct echo_page *ep = cl2echo_page(slice);
356 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
357 ep, mutex_is_locked(&ep->ep_lock),
358 slice->cpl_page->cp_vmpage);
362 static const struct cl_page_operations echo_page_ops = {
363 .cpo_own = echo_page_own,
364 .cpo_disown = echo_page_disown,
365 .cpo_discard = echo_page_discard,
366 .cpo_fini = echo_page_fini,
367 .cpo_print = echo_page_print,
368 .cpo_is_vmlocked = echo_page_is_vmlocked,
371 .cpo_prep = echo_page_prep,
372 .cpo_completion = echo_page_completion,
375 .cpo_prep = echo_page_prep,
376 .cpo_completion = echo_page_completion,
382 /** \defgroup echo_lock Locking
384 * echo lock operations
388 static void echo_lock_fini(const struct lu_env *env,
389 struct cl_lock_slice *slice)
391 struct echo_lock *ecl = cl2echo_lock(slice);
393 LASSERT(list_empty(&ecl->el_chain));
394 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
397 static struct cl_lock_operations echo_lock_ops = {
398 .clo_fini = echo_lock_fini,
403 /** \defgroup echo_cl_ops cl_object operations
405 * operations for cl_object
409 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
410 struct cl_page *page, pgoff_t index)
412 struct echo_page *ep = cl_object_page_slice(obj, page);
413 struct echo_object *eco = cl2echo_obj(obj);
416 page_cache_get(page->cp_vmpage);
417 mutex_init(&ep->ep_lock);
418 cl_page_slice_add(page, &ep->ep_cl, obj, index, &echo_page_ops);
419 atomic_inc(&eco->eo_npages);
423 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
429 static int echo_lock_init(const struct lu_env *env,
430 struct cl_object *obj, struct cl_lock *lock,
431 const struct cl_io *unused)
433 struct echo_lock *el;
436 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, GFP_NOFS);
438 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
439 el->el_object = cl2echo_obj(obj);
440 INIT_LIST_HEAD(&el->el_chain);
441 atomic_set(&el->el_refcount, 0);
443 RETURN(el == NULL ? -ENOMEM : 0);
446 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
447 const struct cl_object_conf *conf)
452 static const struct cl_object_operations echo_cl_obj_ops = {
453 .coo_page_init = echo_page_init,
454 .coo_lock_init = echo_lock_init,
455 .coo_io_init = echo_io_init,
456 .coo_conf_set = echo_conf_set
458 /** @} echo_cl_ops */
460 /** \defgroup echo_lu_ops lu_object operations
462 * operations for echo lu object.
466 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
467 const struct lu_object_conf *conf)
469 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
470 struct echo_client_obd *ec = ed->ed_ec;
471 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
475 struct lu_object *below;
476 struct lu_device *under;
479 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
483 lu_object_add(obj, below);
486 if (!ed->ed_next_ismd) {
487 const struct cl_object_conf *cconf = lu2cl_conf(conf);
488 struct echo_object_conf *econf = cl2echo_conf(cconf);
490 LASSERT(econf->eoc_oinfo != NULL);
492 /* Transfer the oinfo pointer to eco that it won't be
494 eco->eo_oinfo = *econf->eoc_oinfo;
495 *econf->eoc_oinfo = NULL;
497 eco->eo_oinfo = NULL;
501 atomic_set(&eco->eo_npages, 0);
502 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
504 spin_lock(&ec->ec_lock);
505 list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
506 spin_unlock(&ec->ec_lock);
511 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
513 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
514 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
517 LASSERT(atomic_read(&eco->eo_npages) == 0);
519 spin_lock(&ec->ec_lock);
520 list_del_init(&eco->eo_obj_chain);
521 spin_unlock(&ec->ec_lock);
524 lu_object_header_fini(obj->lo_header);
526 if (eco->eo_oinfo != NULL)
527 OBD_FREE_PTR(eco->eo_oinfo);
529 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
533 static int echo_object_print(const struct lu_env *env, void *cookie,
534 lu_printer_t p, const struct lu_object *o)
536 struct echo_object *obj = cl2echo_obj(lu2cl(o));
538 return (*p)(env, cookie, "echoclient-object@%p", obj);
541 static const struct lu_object_operations echo_lu_obj_ops = {
542 .loo_object_init = echo_object_init,
543 .loo_object_delete = NULL,
544 .loo_object_release = NULL,
545 .loo_object_free = echo_object_free,
546 .loo_object_print = echo_object_print,
547 .loo_object_invariant = NULL
549 /** @} echo_lu_ops */
551 /** \defgroup echo_lu_dev_ops lu_device operations
553 * Operations for echo lu device.
557 static struct lu_object *echo_object_alloc(const struct lu_env *env,
558 const struct lu_object_header *hdr,
559 struct lu_device *dev)
561 struct echo_object *eco;
562 struct lu_object *obj = NULL;
565 /* we're the top dev. */
566 LASSERT(hdr == NULL);
567 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, GFP_NOFS);
569 struct cl_object_header *hdr = &eco->eo_hdr;
571 obj = &echo_obj2cl(eco)->co_lu;
572 cl_object_header_init(hdr);
573 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
575 lu_object_init(obj, &hdr->coh_lu, dev);
576 lu_object_add_top(&hdr->coh_lu, obj);
578 eco->eo_cl.co_ops = &echo_cl_obj_ops;
579 obj->lo_ops = &echo_lu_obj_ops;
584 static struct lu_device_operations echo_device_lu_ops = {
585 .ldo_object_alloc = echo_object_alloc,
588 /** @} echo_lu_dev_ops */
590 /** \defgroup echo_init Setup and teardown
592 * Init and fini functions for echo client.
596 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
598 struct cl_site *site = &ed->ed_site_myself;
601 /* initialize site */
602 rc = cl_site_init(site, &ed->ed_cl);
604 CERROR("Cannot initialize site for echo client(%d)\n", rc);
608 rc = lu_site_init_finish(&site->cs_lu);
614 ed->ed_site = &site->cs_lu;
618 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
621 if (!ed->ed_next_ismd)
622 lu_site_fini(ed->ed_site);
627 static void *echo_thread_key_init(const struct lu_context *ctx,
628 struct lu_context_key *key)
630 struct echo_thread_info *info;
632 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, GFP_NOFS);
634 info = ERR_PTR(-ENOMEM);
638 static void echo_thread_key_fini(const struct lu_context *ctx,
639 struct lu_context_key *key, void *data)
641 struct echo_thread_info *info = data;
642 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
645 static struct lu_context_key echo_thread_key = {
646 .lct_tags = LCT_CL_THREAD,
647 .lct_init = echo_thread_key_init,
648 .lct_fini = echo_thread_key_fini,
651 static void *echo_session_key_init(const struct lu_context *ctx,
652 struct lu_context_key *key)
654 struct echo_session_info *session;
656 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, GFP_NOFS);
658 session = ERR_PTR(-ENOMEM);
662 static void echo_session_key_fini(const struct lu_context *ctx,
663 struct lu_context_key *key, void *data)
665 struct echo_session_info *session = data;
666 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
669 static struct lu_context_key echo_session_key = {
670 .lct_tags = LCT_SESSION,
671 .lct_init = echo_session_key_init,
672 .lct_fini = echo_session_key_fini,
675 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
677 #ifdef HAVE_SERVER_SUPPORT
678 # define ECHO_SEQ_WIDTH 0xffffffff
679 static int echo_fid_init(struct echo_device *ed, char *obd_name,
680 struct seq_server_site *ss)
686 OBD_ALLOC_PTR(ed->ed_cl_seq);
687 if (ed->ed_cl_seq == NULL)
690 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
692 GOTO(out_free_seq, rc = -ENOMEM);
694 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
696 /* Init client side sequence-manager */
697 rc = seq_client_init(ed->ed_cl_seq, NULL,
699 prefix, ss->ss_server_seq);
700 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
701 OBD_FREE(prefix, MAX_OBD_NAME + 5);
703 GOTO(out_free_seq, rc);
708 OBD_FREE_PTR(ed->ed_cl_seq);
709 ed->ed_cl_seq = NULL;
713 static int echo_fid_fini(struct obd_device *obddev)
715 struct echo_device *ed = obd2echo_dev(obddev);
718 if (ed->ed_cl_seq != NULL) {
719 seq_client_fini(ed->ed_cl_seq);
720 OBD_FREE_PTR(ed->ed_cl_seq);
721 ed->ed_cl_seq = NULL;
727 static void echo_ed_los_fini(const struct lu_env *env, struct echo_device *ed)
731 if (ed != NULL && ed->ed_next_ismd && ed->ed_los != NULL) {
732 local_oid_storage_fini(env, ed->ed_los);
738 echo_md_local_file_create(const struct lu_env *env, struct echo_md_device *emd,
739 struct local_oid_storage *los,
740 const struct lu_fid *pfid, const char *name,
741 __u32 mode, struct lu_fid *fid)
743 struct dt_object *parent = NULL;
744 struct dt_object *dto = NULL;
748 LASSERT(!fid_is_zero(pfid));
749 parent = dt_locate(env, emd->emd_bottom, pfid);
750 if (unlikely(IS_ERR(parent)))
751 RETURN(PTR_ERR(parent));
753 /* create local file with @fid */
754 dto = local_file_find_or_create_with_fid(env, emd->emd_bottom, fid,
757 GOTO(out_put, rc = PTR_ERR(dto));
759 *fid = *lu_object_fid(&dto->do_lu);
760 /* since stack is not fully set up the local_storage uses own stack
761 * and we should drop its object from cache */
762 lu_object_put_nocache(env, &dto->do_lu);
766 lu_object_put(env, &parent->do_lu);
771 echo_md_root_get(const struct lu_env *env, struct echo_md_device *emd,
772 struct echo_device *ed)
778 /* Setup local dirs */
779 fid.f_seq = FID_SEQ_LOCAL_NAME;
782 rc = local_oid_storage_init(env, emd->emd_bottom, &fid, &ed->ed_los);
786 lu_echo_root_fid(&fid);
787 if (echo_md_seq_site(emd)->ss_node_id == 0) {
788 rc = echo_md_local_file_create(env, emd, ed->ed_los,
789 &emd->emd_local_root_fid,
790 echo_md_root_dir_name, S_IFDIR |
791 S_IRUGO | S_IWUSR | S_IXUGO,
794 CERROR("%s: create md echo root fid failed: rc = %d\n",
795 emd2obd_dev(emd)->obd_name, rc);
799 ed->ed_root_fid = fid;
803 echo_ed_los_fini(env, ed);
807 #endif /* HAVE_SERVER_SUPPORT */
809 static struct lu_device *echo_device_alloc(const struct lu_env *env,
810 struct lu_device_type *t,
811 struct lustre_cfg *cfg)
813 struct lu_device *next;
814 struct echo_device *ed;
815 struct cl_device *cd;
816 struct obd_device *obd = NULL; /* to keep compiler happy */
817 struct obd_device *tgt;
818 const char *tgt_type_name;
825 GOTO(out, rc = -ENOMEM);
829 rc = cl_device_init(cd, t);
833 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
836 obd = class_name2obd(lustre_cfg_string(cfg, 0));
837 LASSERT(obd != NULL);
838 LASSERT(env != NULL);
840 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
842 CERROR("Can not find tgt device %s\n",
843 lustre_cfg_string(cfg, 1));
844 GOTO(out, rc = -ENODEV);
847 next = tgt->obd_lu_dev;
849 if (strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME) == 0) {
850 ed->ed_next_ismd = 1;
851 } else if (strcmp(tgt->obd_type->typ_name, LUSTRE_OST_NAME) == 0 ||
852 strcmp(tgt->obd_type->typ_name, LUSTRE_OSC_NAME) == 0) {
853 ed->ed_next_ismd = 0;
854 rc = echo_site_init(env, ed);
858 GOTO(out, rc = -EINVAL);
863 rc = echo_client_setup(env, obd, cfg);
867 ed->ed_ec = &obd->u.echo_client;
870 if (ed->ed_next_ismd) {
871 #ifdef HAVE_SERVER_SUPPORT
872 /* Suppose to connect to some Metadata layer */
873 struct lu_site *ls = NULL;
874 struct lu_device *ld = NULL;
875 struct md_device *md = NULL;
876 struct echo_md_device *emd = NULL;
880 CERROR("%s is not lu device type!\n",
881 lustre_cfg_string(cfg, 1));
882 GOTO(out, rc = -EINVAL);
885 tgt_type_name = lustre_cfg_string(cfg, 2);
886 if (!tgt_type_name) {
887 CERROR("%s no type name for echo %s setup\n",
888 lustre_cfg_string(cfg, 1),
889 tgt->obd_type->typ_name);
890 GOTO(out, rc = -EINVAL);
895 spin_lock(&ls->ls_ld_lock);
896 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
897 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
902 spin_unlock(&ls->ls_ld_lock);
905 CERROR("%s is not lu device type!\n",
906 lustre_cfg_string(cfg, 1));
907 GOTO(out, rc = -EINVAL);
911 /* For MD echo client, it will use the site in MDS stack */
913 ed->ed_cl.cd_lu_dev.ld_site = ls;
914 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
916 CERROR("echo fid init error %d\n", rc);
920 md = lu2md_dev(next);
921 emd = lu2emd_dev(&md->md_lu_dev);
922 rc = echo_md_root_get(env, emd, ed);
924 CERROR("%s: get root error: rc = %d\n",
925 emd2obd_dev(emd)->obd_name, rc);
928 #else /* !HAVE_SERVER_SUPPORT */
929 CERROR("Local operations are NOT supported on client side. "
930 "Only remote operations are supported. Metadata client "
931 "must be run on server side.\n");
932 GOTO(out, rc = -EOPNOTSUPP);
933 #endif /* HAVE_SERVER_SUPPORT */
935 /* if echo client is to be stacked upon ost device, the next is
936 * NULL since ost is not a clio device so far */
937 if (next != NULL && !lu_device_is_cl(next))
940 tgt_type_name = tgt->obd_type->typ_name;
942 LASSERT(next != NULL);
943 if (next->ld_site != NULL)
944 GOTO(out, rc = -EBUSY);
946 next->ld_site = ed->ed_site;
947 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
948 next->ld_type->ldt_name,
953 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
957 RETURN(&cd->cd_lu_dev);
962 rc2 = echo_client_cleanup(obd);
964 CERROR("Cleanup obd device %s error(%d)\n",
969 echo_site_fini(env, ed);
971 cl_device_fini(&ed->ed_cl);
981 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
982 const char *name, struct lu_device *next)
988 static struct lu_device *echo_device_fini(const struct lu_env *env,
991 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
992 struct lu_device *next = ed->ed_next;
994 while (next && !ed->ed_next_ismd)
995 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
999 static void echo_lock_release(const struct lu_env *env,
1000 struct echo_lock *ecl,
1003 struct cl_lock *clk = echo_lock2cl(ecl);
1005 cl_lock_release(env, clk);
1008 static struct lu_device *echo_device_free(const struct lu_env *env,
1009 struct lu_device *d)
1011 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
1012 struct echo_client_obd *ec = ed->ed_ec;
1013 struct echo_object *eco;
1014 struct lu_device *next = ed->ed_next;
1016 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1019 lu_site_purge(env, ed->ed_site, -1);
1021 /* check if there are objects still alive.
1022 * It shouldn't have any object because lu_site_purge would cleanup
1023 * all of cached objects. Anyway, probably the echo device is being
1024 * parallelly accessed.
1026 spin_lock(&ec->ec_lock);
1027 list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1028 eco->eo_deleted = 1;
1029 spin_unlock(&ec->ec_lock);
1032 lu_site_purge(env, ed->ed_site, -1);
1035 "Waiting for the reference of echo object to be dropped\n");
1037 /* Wait for the last reference to be dropped. */
1038 spin_lock(&ec->ec_lock);
1039 while (!list_empty(&ec->ec_objects)) {
1040 spin_unlock(&ec->ec_lock);
1041 CERROR("echo_client still has objects at cleanup time, "
1042 "wait for 1 second\n");
1043 set_current_state(TASK_UNINTERRUPTIBLE);
1044 schedule_timeout(cfs_time_seconds(1));
1045 lu_site_purge(env, ed->ed_site, -1);
1046 spin_lock(&ec->ec_lock);
1048 spin_unlock(&ec->ec_lock);
1050 LASSERT(list_empty(&ec->ec_locks));
1052 CDEBUG(D_INFO, "No object exists, exiting...\n");
1054 echo_client_cleanup(d->ld_obd);
1055 #ifdef HAVE_SERVER_SUPPORT
1056 echo_fid_fini(d->ld_obd);
1057 echo_ed_los_fini(env, ed);
1059 while (next && !ed->ed_next_ismd)
1060 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1062 LASSERT(ed->ed_site == d->ld_site);
1063 echo_site_fini(env, ed);
1064 cl_device_fini(&ed->ed_cl);
1067 cl_env_cache_purge(~0);
1072 static const struct lu_device_type_operations echo_device_type_ops = {
1073 .ldto_init = echo_type_init,
1074 .ldto_fini = echo_type_fini,
1076 .ldto_start = echo_type_start,
1077 .ldto_stop = echo_type_stop,
1079 .ldto_device_alloc = echo_device_alloc,
1080 .ldto_device_free = echo_device_free,
1081 .ldto_device_init = echo_device_init,
1082 .ldto_device_fini = echo_device_fini
1085 static struct lu_device_type echo_device_type = {
1086 .ldt_tags = LU_DEVICE_CL,
1087 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1088 .ldt_ops = &echo_device_type_ops,
1089 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1093 /** \defgroup echo_exports Exported operations
1095 * exporting functions to echo client
1100 /* Interfaces to echo client obd device */
1101 static struct echo_object *
1102 cl_echo_object_find(struct echo_device *d, const struct ost_id *oi)
1105 struct echo_thread_info *info;
1106 struct echo_object_conf *conf;
1107 struct echo_object *eco;
1108 struct cl_object *obj;
1109 struct lov_oinfo *oinfo = NULL;
1115 LASSERTF(ostid_id(oi) != 0, DOSTID"\n", POSTID(oi));
1116 LASSERTF(ostid_seq(oi) == FID_SEQ_ECHO, DOSTID"\n", POSTID(oi));
1118 /* Never return an object if the obd is to be freed. */
1119 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1120 RETURN(ERR_PTR(-ENODEV));
1122 env = cl_env_get(&refcheck);
1124 RETURN((void *)env);
1126 info = echo_env_info(env);
1127 conf = &info->eti_conf;
1129 OBD_ALLOC_PTR(oinfo);
1131 GOTO(out, eco = ERR_PTR(-ENOMEM));
1133 oinfo->loi_oi = *oi;
1134 conf->eoc_cl.u.coc_oinfo = oinfo;
1137 /* If echo_object_init() is successful then ownership of oinfo
1138 * is transferred to the object. */
1139 conf->eoc_oinfo = &oinfo;
1141 fid = &info->eti_fid;
1142 rc = ostid_to_fid(fid, oi, 0);
1144 GOTO(out, eco = ERR_PTR(rc));
1146 /* In the function below, .hs_keycmp resolves to
1147 * lu_obj_hop_keycmp() */
1148 /* coverity[overrun-buffer-val] */
1149 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1151 GOTO(out, eco = (void*)obj);
1153 eco = cl2echo_obj(obj);
1154 if (eco->eo_deleted) {
1155 cl_object_put(env, obj);
1156 eco = ERR_PTR(-EAGAIN);
1161 OBD_FREE_PTR(oinfo);
1163 cl_env_put(env, &refcheck);
1167 static int cl_echo_object_put(struct echo_object *eco)
1170 struct cl_object *obj = echo_obj2cl(eco);
1174 env = cl_env_get(&refcheck);
1176 RETURN(PTR_ERR(env));
1178 /* an external function to kill an object? */
1179 if (eco->eo_deleted) {
1180 struct lu_object_header *loh = obj->co_lu.lo_header;
1181 LASSERT(&eco->eo_hdr == luh2coh(loh));
1182 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1185 cl_object_put(env, obj);
1186 cl_env_put(env, &refcheck);
1190 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1191 u64 start, u64 end, int mode,
1192 __u64 *cookie , __u32 enqflags)
1195 struct cl_lock *lck;
1196 struct cl_object *obj;
1197 struct cl_lock_descr *descr;
1198 struct echo_thread_info *info;
1202 info = echo_env_info(env);
1204 lck = &info->eti_lock;
1205 obj = echo_obj2cl(eco);
1207 memset(lck, 0, sizeof(*lck));
1208 descr = &lck->cll_descr;
1209 descr->cld_obj = obj;
1210 descr->cld_start = cl_index(obj, start);
1211 descr->cld_end = cl_index(obj, end);
1212 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1213 descr->cld_enq_flags = enqflags;
1216 rc = cl_lock_request(env, io, lck);
1218 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1219 struct echo_lock *el;
1221 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1222 spin_lock(&ec->ec_lock);
1223 if (list_empty(&el->el_chain)) {
1224 list_add(&el->el_chain, &ec->ec_locks);
1225 el->el_cookie = ++ec->ec_unique;
1227 atomic_inc(&el->el_refcount);
1228 *cookie = el->el_cookie;
1229 spin_unlock(&ec->ec_lock);
1234 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1237 struct echo_client_obd *ec = ed->ed_ec;
1238 struct echo_lock *ecl = NULL;
1239 struct list_head *el;
1240 int found = 0, still_used = 0;
1243 LASSERT(ec != NULL);
1244 spin_lock(&ec->ec_lock);
1245 list_for_each(el, &ec->ec_locks) {
1246 ecl = list_entry(el, struct echo_lock, el_chain);
1247 CDEBUG(D_INFO, "ecl: %p, cookie: %#llx\n", ecl, ecl->el_cookie);
1248 found = (ecl->el_cookie == cookie);
1250 if (atomic_dec_and_test(&ecl->el_refcount))
1251 list_del_init(&ecl->el_chain);
1257 spin_unlock(&ec->ec_lock);
1262 echo_lock_release(env, ecl, still_used);
1266 static void echo_commit_callback(const struct lu_env *env, struct cl_io *io,
1267 struct cl_page *page)
1269 struct echo_thread_info *info;
1270 struct cl_2queue *queue;
1272 info = echo_env_info(env);
1273 LASSERT(io == &info->eti_io);
1275 queue = &info->eti_queue;
1276 cl_page_list_add(&queue->c2_qout, page);
1279 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
1280 struct page **pages, int npages, int async)
1283 struct echo_thread_info *info;
1284 struct cl_object *obj = echo_obj2cl(eco);
1285 struct echo_device *ed = eco->eo_dev;
1286 struct cl_2queue *queue;
1288 struct cl_page *clp;
1289 struct lustre_handle lh = { 0 };
1290 int page_size = cl_page_size(obj);
1296 LASSERT((offset & ~PAGE_MASK) == 0);
1297 LASSERT(ed->ed_next != NULL);
1298 env = cl_env_get(&refcheck);
1300 RETURN(PTR_ERR(env));
1302 info = echo_env_info(env);
1304 queue = &info->eti_queue;
1306 cl_2queue_init(queue);
1308 io->ci_ignore_layout = 1;
1309 rc = cl_io_init(env, io, CIT_MISC, obj);
1315 rc = cl_echo_enqueue0(env, eco, offset,
1316 offset + npages * PAGE_CACHE_SIZE - 1,
1317 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1320 GOTO(error_lock, rc);
1322 for (i = 0; i < npages; i++) {
1324 clp = cl_page_find(env, obj, cl_index(obj, offset),
1325 pages[i], CPT_TRANSIENT);
1330 LASSERT(clp->cp_type == CPT_TRANSIENT);
1332 rc = cl_page_own(env, io, clp);
1334 LASSERT(clp->cp_state == CPS_FREEING);
1335 cl_page_put(env, clp);
1339 cl_2queue_add(queue, clp);
1341 /* drop the reference count for cl_page_find, so that the page
1342 * will be freed in cl_2queue_fini. */
1343 cl_page_put(env, clp);
1344 cl_page_clip(env, clp, 0, page_size);
1346 offset += page_size;
1350 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1352 async = async && (typ == CRT_WRITE);
1354 rc = cl_io_commit_async(env, io, &queue->c2_qin,
1356 echo_commit_callback);
1358 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1359 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1360 async ? "async" : "sync", rc);
1363 cl_echo_cancel0(env, ed, lh.cookie);
1366 cl_2queue_discard(env, io, queue);
1367 cl_2queue_disown(env, io, queue);
1368 cl_2queue_fini(env, queue);
1369 cl_io_fini(env, io);
1371 cl_env_put(env, &refcheck);
1374 /** @} echo_exports */
1377 static u64 last_object_id;
1379 #ifdef HAVE_SERVER_SUPPORT
1380 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1383 snprintf(name, ETI_NAME_LEN, "%llu", id);
1384 lname->ln_name = name;
1385 lname->ln_namelen = strlen(name);
1388 /* similar to mdt_attr_get_complex */
1389 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1392 struct echo_thread_info *info = echo_env_info(env);
1397 LASSERT(ma->ma_lmm_size > 0);
1399 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1403 /* big_lmm may need to be grown */
1404 if (info->eti_big_lmmsize < rc) {
1405 int size = size_roundup_power2(rc);
1407 if (info->eti_big_lmmsize > 0) {
1408 /* free old buffer */
1409 LASSERT(info->eti_big_lmm);
1410 OBD_FREE_LARGE(info->eti_big_lmm,
1411 info->eti_big_lmmsize);
1412 info->eti_big_lmm = NULL;
1413 info->eti_big_lmmsize = 0;
1416 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1417 if (info->eti_big_lmm == NULL)
1419 info->eti_big_lmmsize = size;
1421 LASSERT(info->eti_big_lmmsize >= rc);
1423 info->eti_buf.lb_buf = info->eti_big_lmm;
1424 info->eti_buf.lb_len = info->eti_big_lmmsize;
1425 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1429 ma->ma_valid |= MA_LOV;
1430 ma->ma_lmm = info->eti_big_lmm;
1431 ma->ma_lmm_size = rc;
1436 static int echo_attr_get_complex(const struct lu_env *env,
1437 struct md_object *next,
1440 struct echo_thread_info *info = echo_env_info(env);
1441 struct lu_buf *buf = &info->eti_buf;
1442 umode_t mode = lu_object_attr(&next->mo_lu);
1443 int need = ma->ma_need;
1450 if (need & MA_INODE) {
1451 ma->ma_need = MA_INODE;
1452 rc = mo_attr_get(env, next, ma);
1455 ma->ma_valid |= MA_INODE;
1458 if (need & MA_LOV) {
1459 if (S_ISREG(mode) || S_ISDIR(mode)) {
1460 LASSERT(ma->ma_lmm_size > 0);
1461 buf->lb_buf = ma->ma_lmm;
1462 buf->lb_len = ma->ma_lmm_size;
1463 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1465 ma->ma_lmm_size = rc2;
1466 ma->ma_valid |= MA_LOV;
1467 } else if (rc2 == -ENODATA) {
1469 ma->ma_lmm_size = 0;
1470 } else if (rc2 == -ERANGE) {
1471 rc2 = echo_big_lmm_get(env, next, ma);
1473 GOTO(out, rc = rc2);
1475 GOTO(out, rc = rc2);
1480 #ifdef CONFIG_FS_POSIX_ACL
1481 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1482 buf->lb_buf = ma->ma_acl;
1483 buf->lb_len = ma->ma_acl_size;
1484 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1486 ma->ma_acl_size = rc2;
1487 ma->ma_valid |= MA_ACL_DEF;
1488 } else if (rc2 == -ENODATA) {
1490 ma->ma_acl_size = 0;
1492 GOTO(out, rc = rc2);
1498 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1499 rc, ma->ma_valid, ma->ma_lmm);
1504 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1505 struct md_object *parent, struct lu_fid *fid,
1506 struct lu_name *lname, struct md_op_spec *spec,
1509 struct lu_object *ec_child, *child;
1510 struct lu_device *ld = ed->ed_next;
1511 struct echo_thread_info *info = echo_env_info(env);
1512 struct lu_fid *fid2 = &info->eti_fid2;
1513 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1518 rc = mdo_lookup(env, parent, lname, fid2, spec);
1521 else if (rc != -ENOENT)
1524 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1526 if (IS_ERR(ec_child)) {
1527 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1529 RETURN(PTR_ERR(ec_child));
1532 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1533 if (child == NULL) {
1534 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1535 GOTO(out_put, rc = -EINVAL);
1538 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1539 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1542 * Do not perform lookup sanity check. We know that name does not exist.
1544 spec->sp_cr_lookup = 0;
1545 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1547 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1550 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1551 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1554 lu_object_put(env, ec_child);
1558 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1561 struct echo_thread_info *info = echo_env_info(env);
1563 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1564 ma->ma_lmm = (void *)&info->eti_lmm;
1565 ma->ma_lmm_size = sizeof(info->eti_lmm);
1567 LASSERT(info->eti_big_lmmsize);
1568 ma->ma_lmm = info->eti_big_lmm;
1569 ma->ma_lmm_size = info->eti_big_lmmsize;
1575 static int echo_create_md_object(const struct lu_env *env,
1576 struct echo_device *ed,
1577 struct lu_object *ec_parent,
1579 char *name, int namelen,
1580 __u64 id, __u32 mode, int count,
1581 int stripe_count, int stripe_offset)
1583 struct lu_object *parent;
1584 struct echo_thread_info *info = echo_env_info(env);
1585 struct lu_name *lname = &info->eti_lname;
1586 struct md_op_spec *spec = &info->eti_spec;
1587 struct md_attr *ma = &info->eti_ma;
1588 struct lu_device *ld = ed->ed_next;
1594 if (ec_parent == NULL)
1596 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1600 memset(ma, 0, sizeof(*ma));
1601 memset(spec, 0, sizeof(*spec));
1602 if (stripe_count != 0) {
1603 spec->sp_cr_flags |= FMODE_WRITE;
1604 echo_set_lmm_size(env, ld, ma);
1605 if (stripe_count != -1) {
1606 struct lov_user_md_v3 *lum = &info->eti_lum;
1608 lum->lmm_magic = LOV_USER_MAGIC_V3;
1609 lum->lmm_stripe_count = stripe_count;
1610 lum->lmm_stripe_offset = stripe_offset;
1611 lum->lmm_pattern = 0;
1612 spec->u.sp_ea.eadata = lum;
1613 spec->u.sp_ea.eadatalen = sizeof(*lum);
1614 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1618 ma->ma_attr.la_mode = mode;
1619 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1620 ma->ma_attr.la_ctime = cfs_time_current_64();
1623 lname->ln_name = name;
1624 lname->ln_namelen = namelen;
1625 /* If name is specified, only create one object by name */
1626 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1631 /* Create multiple object sequenced by id */
1632 for (i = 0; i < count; i++) {
1633 char *tmp_name = info->eti_name;
1635 echo_md_build_name(lname, tmp_name, id);
1637 rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname,
1640 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1651 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1652 struct echo_device *ed,
1653 struct md_object *parent,
1654 struct lu_name *lname)
1656 struct echo_thread_info *info = echo_env_info(env);
1657 struct lu_fid *fid = &info->eti_fid;
1658 struct lu_object *child;
1662 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1664 rc = mdo_lookup(env, parent, lname, fid, NULL);
1666 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1667 RETURN(ERR_PTR(rc));
1670 /* In the function below, .hs_keycmp resolves to
1671 * lu_obj_hop_keycmp() */
1672 /* coverity[overrun-buffer-val] */
1673 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1678 static int echo_setattr_object(const struct lu_env *env,
1679 struct echo_device *ed,
1680 struct lu_object *ec_parent,
1681 __u64 id, int count)
1683 struct lu_object *parent;
1684 struct echo_thread_info *info = echo_env_info(env);
1685 struct lu_name *lname = &info->eti_lname;
1686 char *name = info->eti_name;
1687 struct lu_device *ld = ed->ed_next;
1688 struct lu_buf *buf = &info->eti_buf;
1694 if (ec_parent == NULL)
1696 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1700 for (i = 0; i < count; i++) {
1701 struct lu_object *ec_child, *child;
1703 echo_md_build_name(lname, name, id);
1705 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1706 if (IS_ERR(ec_child)) {
1707 CERROR("Can't find child %s: rc = %ld\n",
1708 lname->ln_name, PTR_ERR(ec_child));
1709 RETURN(PTR_ERR(ec_child));
1712 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1713 if (child == NULL) {
1714 CERROR("Can not locate the child %s\n", lname->ln_name);
1715 lu_object_put(env, ec_child);
1720 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1721 PFID(lu_object_fid(child)));
1723 buf->lb_buf = info->eti_xattr_buf;
1724 buf->lb_len = sizeof(info->eti_xattr_buf);
1726 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1727 rc = mo_xattr_set(env, lu2md(child), buf, name,
1730 CERROR("Can not setattr child "DFID": rc = %d\n",
1731 PFID(lu_object_fid(child)), rc);
1732 lu_object_put(env, ec_child);
1735 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1736 PFID(lu_object_fid(child)));
1738 lu_object_put(env, ec_child);
1743 static int echo_getattr_object(const struct lu_env *env,
1744 struct echo_device *ed,
1745 struct lu_object *ec_parent,
1746 __u64 id, int count)
1748 struct lu_object *parent;
1749 struct echo_thread_info *info = echo_env_info(env);
1750 struct lu_name *lname = &info->eti_lname;
1751 char *name = info->eti_name;
1752 struct md_attr *ma = &info->eti_ma;
1753 struct lu_device *ld = ed->ed_next;
1759 if (ec_parent == NULL)
1761 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1765 memset(ma, 0, sizeof(*ma));
1766 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1767 ma->ma_acl = info->eti_xattr_buf;
1768 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1770 for (i = 0; i < count; i++) {
1771 struct lu_object *ec_child, *child;
1774 echo_md_build_name(lname, name, id);
1775 echo_set_lmm_size(env, ld, ma);
1777 ec_child = echo_md_lookup(env, ed, lu2md(parent), lname);
1778 if (IS_ERR(ec_child)) {
1779 CERROR("Can't find child %s: rc = %ld\n",
1780 lname->ln_name, PTR_ERR(ec_child));
1781 RETURN(PTR_ERR(ec_child));
1784 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1785 if (child == NULL) {
1786 CERROR("Can not locate the child %s\n", lname->ln_name);
1787 lu_object_put(env, ec_child);
1791 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1792 PFID(lu_object_fid(child)));
1793 rc = echo_attr_get_complex(env, lu2md(child), ma);
1795 CERROR("Can not getattr child "DFID": rc = %d\n",
1796 PFID(lu_object_fid(child)), rc);
1797 lu_object_put(env, ec_child);
1800 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1801 PFID(lu_object_fid(child)));
1803 lu_object_put(env, ec_child);
1809 static int echo_lookup_object(const struct lu_env *env,
1810 struct echo_device *ed,
1811 struct lu_object *ec_parent,
1812 __u64 id, int count)
1814 struct lu_object *parent;
1815 struct echo_thread_info *info = echo_env_info(env);
1816 struct lu_name *lname = &info->eti_lname;
1817 char *name = info->eti_name;
1818 struct lu_fid *fid = &info->eti_fid;
1819 struct lu_device *ld = ed->ed_next;
1823 if (ec_parent == NULL)
1825 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1829 /*prepare the requests*/
1830 for (i = 0; i < count; i++) {
1831 echo_md_build_name(lname, name, id);
1833 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1834 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1836 rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL);
1838 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1841 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1842 PFID(lu_object_fid(parent)), lname->ln_name, parent);
1849 static int echo_md_destroy_internal(const struct lu_env *env,
1850 struct echo_device *ed,
1851 struct md_object *parent,
1852 struct lu_name *lname,
1855 struct lu_device *ld = ed->ed_next;
1856 struct lu_object *ec_child;
1857 struct lu_object *child;
1862 ec_child = echo_md_lookup(env, ed, parent, lname);
1863 if (IS_ERR(ec_child)) {
1864 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
1866 RETURN(PTR_ERR(ec_child));
1869 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1870 if (child == NULL) {
1871 CERROR("Can not locate the child %s\n", lname->ln_name);
1872 GOTO(out_put, rc = -EINVAL);
1875 if (lu_object_remote(child)) {
1876 CERROR("Can not destroy remote object %s: rc = %d\n",
1877 lname->ln_name, -EPERM);
1878 GOTO(out_put, rc = -EPERM);
1880 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
1881 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1883 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
1885 CERROR("Can not unlink child %s: rc = %d\n",
1886 lname->ln_name, rc);
1889 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
1890 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1892 lu_object_put(env, ec_child);
1896 static int echo_destroy_object(const struct lu_env *env,
1897 struct echo_device *ed,
1898 struct lu_object *ec_parent,
1899 char *name, int namelen,
1900 __u64 id, __u32 mode,
1903 struct echo_thread_info *info = echo_env_info(env);
1904 struct lu_name *lname = &info->eti_lname;
1905 struct md_attr *ma = &info->eti_ma;
1906 struct lu_device *ld = ed->ed_next;
1907 struct lu_object *parent;
1912 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1916 memset(ma, 0, sizeof(*ma));
1917 ma->ma_attr.la_mode = mode;
1918 ma->ma_attr.la_valid = LA_CTIME;
1919 ma->ma_attr.la_ctime = cfs_time_current_64();
1920 ma->ma_need = MA_INODE;
1924 lname->ln_name = name;
1925 lname->ln_namelen = namelen;
1926 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1931 /*prepare the requests*/
1932 for (i = 0; i < count; i++) {
1933 char *tmp_name = info->eti_name;
1936 echo_md_build_name(lname, tmp_name, id);
1938 rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname,
1941 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
1950 static struct lu_object *echo_resolve_path(const struct lu_env *env,
1951 struct echo_device *ed, char *path,
1954 struct lu_device *ld = ed->ed_next;
1955 struct echo_thread_info *info = echo_env_info(env);
1956 struct lu_fid *fid = &info->eti_fid;
1957 struct lu_name *lname = &info->eti_lname;
1958 struct lu_object *parent = NULL;
1959 struct lu_object *child = NULL;
1963 *fid = ed->ed_root_fid;
1965 /* In the function below, .hs_keycmp resolves to
1966 * lu_obj_hop_keycmp() */
1967 /* coverity[overrun-buffer-val] */
1968 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1969 if (IS_ERR(parent)) {
1970 CERROR("Can not find the parent "DFID": rc = %ld\n",
1971 PFID(fid), PTR_ERR(parent));
1976 struct lu_object *ld_parent;
1979 e = strsep(&path, "/");
1984 if (!path || path[0] == '\0')
1990 lname->ln_namelen = strlen(e);
1992 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
1993 if (ld_parent == NULL) {
1994 lu_object_put(env, parent);
1999 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2000 lu_object_put(env, parent);
2001 if (IS_ERR(child)) {
2002 rc = (int)PTR_ERR(child);
2003 CERROR("lookup %s under parent "DFID": rc = %d\n",
2004 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2011 RETURN(ERR_PTR(rc));
2016 static void echo_ucred_init(struct lu_env *env)
2018 struct lu_ucred *ucred = lu_ucred(env);
2020 ucred->uc_valid = UCRED_INVALID;
2022 ucred->uc_suppgids[0] = -1;
2023 ucred->uc_suppgids[1] = -1;
2025 ucred->uc_uid = ucred->uc_o_uid =
2026 from_kuid(&init_user_ns, current_uid());
2027 ucred->uc_gid = ucred->uc_o_gid =
2028 from_kgid(&init_user_ns, current_gid());
2029 ucred->uc_fsuid = ucred->uc_o_fsuid =
2030 from_kuid(&init_user_ns, current_fsuid());
2031 ucred->uc_fsgid = ucred->uc_o_fsgid =
2032 from_kgid(&init_user_ns, current_fsgid());
2033 ucred->uc_cap = cfs_curproc_cap_pack();
2035 /* remove fs privilege for non-root user. */
2036 if (ucred->uc_fsuid)
2037 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2038 ucred->uc_valid = UCRED_NEW;
2041 static void echo_ucred_fini(struct lu_env *env)
2043 struct lu_ucred *ucred = lu_ucred(env);
2044 ucred->uc_valid = UCRED_INIT;
2047 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2048 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2049 static int echo_md_handler(struct echo_device *ed, int command,
2050 char *path, int path_len, __u64 id, int count,
2051 struct obd_ioctl_data *data)
2053 struct echo_thread_info *info;
2054 struct lu_device *ld = ed->ed_next;
2057 struct lu_object *parent;
2059 int namelen = data->ioc_plen2;
2064 CERROR("MD echo client is not being initialized properly\n");
2068 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2069 CERROR("Only support MDD layer right now!\n");
2073 env = cl_env_get(&refcheck);
2075 RETURN(PTR_ERR(env));
2077 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2081 /* init big_lmm buffer */
2082 info = echo_env_info(env);
2083 LASSERT(info->eti_big_lmm == NULL);
2084 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2085 if (info->eti_big_lmm == NULL)
2086 GOTO(out_env, rc = -ENOMEM);
2087 info->eti_big_lmmsize = MIN_MD_SIZE;
2089 parent = echo_resolve_path(env, ed, path, path_len);
2090 if (IS_ERR(parent)) {
2091 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2093 GOTO(out_free, rc = PTR_ERR(parent));
2097 OBD_ALLOC(name, namelen + 1);
2099 GOTO(out_put, rc = -ENOMEM);
2100 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2101 GOTO(out_name, rc = -EFAULT);
2104 echo_ucred_init(env);
2107 case ECHO_MD_CREATE:
2108 case ECHO_MD_MKDIR: {
2109 struct echo_thread_info *info = echo_env_info(env);
2110 __u32 mode = data->ioc_obdo2.o_mode;
2111 struct lu_fid *fid = &info->eti_fid;
2112 int stripe_count = (int)data->ioc_obdo2.o_misc;
2113 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2115 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2119 /* In the function below, .hs_keycmp resolves to
2120 * lu_obj_hop_keycmp() */
2121 /* coverity[overrun-buffer-val] */
2122 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2123 id, mode, count, stripe_count,
2127 case ECHO_MD_DESTROY:
2128 case ECHO_MD_RMDIR: {
2129 __u32 mode = data->ioc_obdo2.o_mode;
2131 rc = echo_destroy_object(env, ed, parent, name, namelen,
2135 case ECHO_MD_LOOKUP:
2136 rc = echo_lookup_object(env, ed, parent, id, count);
2138 case ECHO_MD_GETATTR:
2139 rc = echo_getattr_object(env, ed, parent, id, count);
2141 case ECHO_MD_SETATTR:
2142 rc = echo_setattr_object(env, ed, parent, id, count);
2145 CERROR("unknown command %d\n", command);
2149 echo_ucred_fini(env);
2153 OBD_FREE(name, namelen + 1);
2155 lu_object_put(env, parent);
2157 LASSERT(info->eti_big_lmm);
2158 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2159 info->eti_big_lmm = NULL;
2160 info->eti_big_lmmsize = 0;
2162 cl_env_put(env, &refcheck);
2165 #endif /* HAVE_SERVER_SUPPORT */
2167 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2170 struct echo_object *eco;
2171 struct echo_client_obd *ec = ed->ed_ec;
2176 if (!(oa->o_valid & OBD_MD_FLID) ||
2177 !(oa->o_valid & OBD_MD_FLGROUP) ||
2178 !fid_seq_is_echo(ostid_seq(&oa->o_oi))) {
2179 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2183 if (ostid_id(&oa->o_oi) == 0)
2184 ostid_set_id(&oa->o_oi, ++last_object_id);
2186 rc = obd_create(env, ec->ec_exp, oa);
2188 CERROR("Cannot create objects: rc = %d\n", rc);
2194 oa->o_valid |= OBD_MD_FLID;
2196 eco = cl_echo_object_find(ed, &oa->o_oi);
2198 GOTO(failed, rc = PTR_ERR(eco));
2199 cl_echo_object_put(eco);
2201 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2205 if (created && rc != 0)
2206 obd_destroy(env, ec->ec_exp, oa);
2209 CERROR("create object failed with: rc = %d\n", rc);
2214 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2217 struct echo_object *eco;
2221 if (!(oa->o_valid & OBD_MD_FLID) ||
2222 !(oa->o_valid & OBD_MD_FLGROUP) ||
2223 ostid_id(&oa->o_oi) == 0) {
2224 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2229 eco = cl_echo_object_find(ed, &oa->o_oi);
2238 static void echo_put_object(struct echo_object *eco)
2242 rc = cl_echo_object_put(eco);
2244 CERROR("%s: echo client drop an object failed: rc = %d\n",
2245 eco->eo_dev->ed_ec->ec_exp->exp_obd->obd_name, rc);
2248 static void echo_client_page_debug_setup(struct page *page, int rw, u64 id,
2249 u64 offset, u64 count)
2256 /* no partial pages on the client */
2257 LASSERT(count == PAGE_CACHE_SIZE);
2261 for (delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2262 if (rw == OBD_BRW_WRITE) {
2263 stripe_off = offset + delta;
2266 stripe_off = 0xdeadbeef00c0ffeeULL;
2267 stripe_id = 0xdeadbeef00c0ffeeULL;
2269 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2270 stripe_off, stripe_id);
2277 echo_client_page_debug_check(struct page *page, u64 id, u64 offset, u64 count)
2286 /* no partial pages on the client */
2287 LASSERT(count == PAGE_CACHE_SIZE);
2291 for (rc = delta = 0; delta < PAGE_CACHE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2292 stripe_off = offset + delta;
2295 rc2 = block_debug_check("test_brw",
2296 addr + delta, OBD_ECHO_BLOCK_SIZE,
2297 stripe_off, stripe_id);
2299 CERROR("Error in echo object %#llx\n", id);
2308 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2309 struct echo_object *eco, u64 offset,
2310 u64 count, int async)
2313 struct brw_page *pga;
2314 struct brw_page *pgp;
2315 struct page **pages;
2324 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2325 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2326 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2328 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
2330 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2332 if ((count & (~PAGE_MASK)) != 0)
2335 /* XXX think again with misaligned I/O */
2336 npages = count >> PAGE_CACHE_SHIFT;
2338 if (rw == OBD_BRW_WRITE)
2339 brw_flags = OBD_BRW_ASYNC;
2341 OBD_ALLOC(pga, npages * sizeof(*pga));
2345 OBD_ALLOC(pages, npages * sizeof(*pages));
2346 if (pages == NULL) {
2347 OBD_FREE(pga, npages * sizeof(*pga));
2351 for (i = 0, pgp = pga, off = offset;
2353 i++, pgp++, off += PAGE_CACHE_SIZE) {
2355 LASSERT(pgp->pg == NULL); /* for cleanup */
2358 pgp->pg = alloc_page(gfp_mask);
2359 if (pgp->pg == NULL)
2363 pgp->count = PAGE_CACHE_SIZE;
2365 pgp->flag = brw_flags;
2368 echo_client_page_debug_setup(pgp->pg, rw,
2369 ostid_id(&oa->o_oi), off,
2373 /* brw mode can only be used at client */
2374 LASSERT(ed->ed_next != NULL);
2375 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2378 if (rc != 0 || rw != OBD_BRW_READ)
2381 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2382 if (pgp->pg == NULL)
2387 vrc = echo_client_page_debug_check(pgp->pg,
2388 ostid_id(&oa->o_oi),
2389 pgp->off, pgp->count);
2390 if (vrc != 0 && rc == 0)
2393 __free_page(pgp->pg);
2395 OBD_FREE(pga, npages * sizeof(*pga));
2396 OBD_FREE(pages, npages * sizeof(*pages));
2400 static int echo_client_prep_commit(const struct lu_env *env,
2401 struct obd_export *exp, int rw,
2402 struct obdo *oa, struct echo_object *eco,
2403 u64 offset, u64 count,
2404 u64 batch, int async)
2406 struct obd_ioobj ioo;
2407 struct niobuf_local *lnb;
2408 struct niobuf_remote rnb;
2410 u64 npages, tot_pages, apc;
2411 int i, ret = 0, brw_flags = 0;
2415 if (count <= 0 || (count & ~PAGE_CACHE_MASK) != 0)
2418 apc = npages = batch >> PAGE_CACHE_SHIFT;
2419 tot_pages = count >> PAGE_CACHE_SHIFT;
2421 OBD_ALLOC(lnb, apc * sizeof(struct niobuf_local));
2425 if (rw == OBD_BRW_WRITE && async)
2426 brw_flags |= OBD_BRW_ASYNC;
2428 obdo_to_ioobj(oa, &ioo);
2432 for (; tot_pages > 0; tot_pages -= npages) {
2435 if (tot_pages < npages)
2438 rnb.rnb_offset = off;
2439 rnb.rnb_len = npages * PAGE_CACHE_SIZE;
2440 rnb.rnb_flags = brw_flags;
2442 off += npages * PAGE_CACHE_SIZE;
2445 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, &rnb, &lpages, lnb);
2449 for (i = 0; i < lpages; i++) {
2450 struct page *page = lnb[i].lnb_page;
2452 /* read past eof? */
2453 if (page == NULL && lnb[i].lnb_rc == 0)
2457 lnb[i].lnb_flags |= OBD_BRW_ASYNC;
2459 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2460 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2461 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2464 if (rw == OBD_BRW_WRITE)
2465 echo_client_page_debug_setup(page, rw,
2466 ostid_id(&oa->o_oi),
2467 lnb[i].lnb_file_offset,
2470 echo_client_page_debug_check(page,
2471 ostid_id(&oa->o_oi),
2472 lnb[i].lnb_file_offset,
2476 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo, &rnb, npages, lnb,
2481 /* Reuse env context. */
2482 lu_context_exit((struct lu_context *)&env->le_ctx);
2483 lu_context_enter((struct lu_context *)&env->le_ctx);
2487 OBD_FREE(lnb, apc * sizeof(struct niobuf_local));
2492 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2493 struct obd_export *exp,
2494 struct obd_ioctl_data *data)
2496 struct obd_device *obd = class_exp2obd(exp);
2497 struct echo_device *ed = obd2echo_dev(obd);
2498 struct echo_client_obd *ec = ed->ed_ec;
2499 struct obdo *oa = &data->ioc_obdo1;
2500 struct echo_object *eco;
2506 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2508 rc = echo_get_object(&eco, ed, oa);
2512 oa->o_valid &= ~OBD_MD_FLHANDLE;
2514 /* OFD/obdfilter works only via prep/commit */
2515 test_mode = (long)data->ioc_pbuf1;
2516 if (ed->ed_next == NULL && test_mode != 3) {
2518 data->ioc_plen1 = data->ioc_count;
2524 /* Truncate batch size to maximum */
2525 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2526 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2528 switch (test_mode) {
2532 rc = echo_client_kbrw(ed, rw, oa, eco, data->ioc_offset,
2533 data->ioc_count, async);
2536 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa, eco,
2537 data->ioc_offset, data->ioc_count,
2538 data->ioc_plen1, async);
2544 echo_put_object(eco);
2550 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2551 void *karg, void __user *uarg)
2553 #ifdef HAVE_SERVER_SUPPORT
2554 struct tgt_session_info *tsi;
2556 struct obd_device *obd = exp->exp_obd;
2557 struct echo_device *ed = obd2echo_dev(obd);
2558 struct echo_client_obd *ec = ed->ed_ec;
2559 struct echo_object *eco;
2560 struct obd_ioctl_data *data = karg;
2564 int rw = OBD_BRW_READ;
2566 #ifdef HAVE_SERVER_SUPPORT
2567 struct lu_context echo_session;
2571 oa = &data->ioc_obdo1;
2572 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2573 oa->o_valid |= OBD_MD_FLGROUP;
2574 ostid_set_seq_echo(&oa->o_oi);
2577 /* This FID is unpacked just for validation at this point */
2578 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2586 rc = lu_env_init(env, LCT_DT_THREAD);
2588 GOTO(out_alloc, rc = -ENOMEM);
2590 #ifdef HAVE_SERVER_SUPPORT
2591 env->le_ses = &echo_session;
2592 rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2593 if (unlikely(rc < 0))
2595 lu_context_enter(env->le_ses);
2597 tsi = tgt_ses_info(env);
2598 tsi->tsi_exp = ec->ec_exp;
2599 tsi->tsi_jobid = NULL;
2602 case OBD_IOC_CREATE: /* may create echo object */
2603 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2604 GOTO (out, rc = -EPERM);
2606 rc = echo_create_object(env, ed, oa);
2609 #ifdef HAVE_SERVER_SUPPORT
2610 case OBD_IOC_ECHO_MD: {
2617 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2618 GOTO(out, rc = -EPERM);
2620 count = data->ioc_count;
2621 cmd = data->ioc_command;
2623 id = data->ioc_obdo2.o_oi.oi.oi_id;
2624 dirlen = data->ioc_plen1;
2625 OBD_ALLOC(dir, dirlen + 1);
2627 GOTO(out, rc = -ENOMEM);
2629 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2630 OBD_FREE(dir, data->ioc_plen1 + 1);
2631 GOTO(out, rc = -EFAULT);
2634 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2635 OBD_FREE(dir, dirlen + 1);
2638 case OBD_IOC_ECHO_ALLOC_SEQ: {
2639 struct lu_env *cl_env;
2644 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2645 GOTO(out, rc = -EPERM);
2647 cl_env = cl_env_get(&refcheck);
2649 GOTO(out, rc = PTR_ERR(cl_env));
2651 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2654 cl_env_put(cl_env, &refcheck);
2658 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2659 cl_env_put(cl_env, &refcheck);
2661 CERROR("%s: Can not alloc seq: rc = %d\n",
2666 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2669 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2670 if (copy_to_user(data->ioc_pbuf2, &max_count,
2675 #endif /* HAVE_SERVER_SUPPORT */
2676 case OBD_IOC_DESTROY:
2677 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2678 GOTO (out, rc = -EPERM);
2680 rc = echo_get_object(&eco, ed, oa);
2682 rc = obd_destroy(env, ec->ec_exp, oa);
2684 eco->eo_deleted = 1;
2685 echo_put_object(eco);
2689 case OBD_IOC_GETATTR:
2690 rc = echo_get_object(&eco, ed, oa);
2692 rc = obd_getattr(env, ec->ec_exp, oa);
2693 echo_put_object(eco);
2697 case OBD_IOC_SETATTR:
2698 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2699 GOTO (out, rc = -EPERM);
2701 rc = echo_get_object(&eco, ed, oa);
2703 rc = obd_setattr(env, ec->ec_exp, oa);
2704 echo_put_object(eco);
2708 case OBD_IOC_BRW_WRITE:
2709 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2710 GOTO (out, rc = -EPERM);
2714 case OBD_IOC_BRW_READ:
2715 rc = echo_client_brw_ioctl(env, rw, exp, data);
2719 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2720 GOTO (out, rc = -ENOTTY);
2725 #ifdef HAVE_SERVER_SUPPORT
2726 lu_context_exit(env->le_ses);
2727 lu_context_fini(env->le_ses);
2737 static int echo_client_setup(const struct lu_env *env,
2738 struct obd_device *obddev, struct lustre_cfg *lcfg)
2740 struct echo_client_obd *ec = &obddev->u.echo_client;
2741 struct obd_device *tgt;
2742 struct obd_uuid echo_uuid = { "ECHO_UUID" };
2743 struct obd_connect_data *ocd = NULL;
2747 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2748 CERROR("requires a TARGET OBD name\n");
2752 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2753 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2754 CERROR("device not attached or not set up (%s)\n",
2755 lustre_cfg_string(lcfg, 1));
2759 spin_lock_init(&ec->ec_lock);
2760 INIT_LIST_HEAD(&ec->ec_objects);
2761 INIT_LIST_HEAD(&ec->ec_locks);
2764 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2765 #ifdef HAVE_SERVER_SUPPORT
2766 lu_context_tags_update(ECHO_MD_CTX_TAG);
2767 lu_session_tags_update(ECHO_MD_SES_TAG);
2769 CERROR("Local operations are NOT supported on client side. "
2770 "Only remote operations are supported. Metadata client "
2771 "must be run on server side.\n");
2776 OBD_ALLOC(ocd, sizeof(*ocd));
2778 CERROR("Can't alloc ocd connecting to %s\n",
2779 lustre_cfg_string(lcfg, 1));
2783 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2784 OBD_CONNECT_BRW_SIZE |
2785 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2786 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2788 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2789 ocd->ocd_version = LUSTRE_VERSION_CODE;
2790 ocd->ocd_group = FID_SEQ_ECHO;
2792 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2794 /* Turn off pinger because it connects to tgt obd directly. */
2795 spin_lock(&tgt->obd_dev_lock);
2796 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
2797 spin_unlock(&tgt->obd_dev_lock);
2800 OBD_FREE(ocd, sizeof(*ocd));
2803 CERROR("fail to connect to device %s\n",
2804 lustre_cfg_string(lcfg, 1));
2811 static int echo_client_cleanup(struct obd_device *obddev)
2813 struct echo_device *ed = obd2echo_dev(obddev);
2814 struct echo_client_obd *ec = &obddev->u.echo_client;
2818 /*Do nothing for Metadata echo client*/
2822 if (ed->ed_next_ismd) {
2823 #ifdef HAVE_SERVER_SUPPORT
2824 lu_context_tags_clear(ECHO_MD_CTX_TAG);
2825 lu_session_tags_clear(ECHO_MD_SES_TAG);
2827 CERROR("This is client-side only module, does not support "
2828 "metadata echo client.\n");
2833 if (!list_empty(&obddev->obd_exports)) {
2834 CERROR("still has clients!\n");
2838 LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
2839 rc = obd_disconnect(ec->ec_exp);
2841 CERROR("fail to disconnect device: %d\n", rc);
2846 static int echo_client_connect(const struct lu_env *env,
2847 struct obd_export **exp,
2848 struct obd_device *src, struct obd_uuid *cluuid,
2849 struct obd_connect_data *data, void *localdata)
2852 struct lustre_handle conn = { 0 };
2855 rc = class_connect(&conn, src, cluuid);
2857 *exp = class_conn2export(&conn);
2863 static int echo_client_disconnect(struct obd_export *exp)
2869 GOTO(out, rc = -EINVAL);
2871 rc = class_disconnect(exp);
2877 static struct obd_ops echo_client_obd_ops = {
2878 .o_owner = THIS_MODULE,
2879 .o_iocontrol = echo_client_iocontrol,
2880 .o_connect = echo_client_connect,
2881 .o_disconnect = echo_client_disconnect
2884 static int __init obdecho_init(void)
2889 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
2891 LASSERT(PAGE_CACHE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
2893 # ifdef HAVE_SERVER_SUPPORT
2894 rc = echo_persistent_pages_init();
2898 rc = class_register_type(&echo_obd_ops, NULL, true, NULL,
2899 LUSTRE_ECHO_NAME, NULL);
2904 rc = lu_kmem_init(echo_caches);
2906 rc = class_register_type(&echo_client_obd_ops, NULL, true, NULL,
2907 LUSTRE_ECHO_CLIENT_NAME,
2910 lu_kmem_fini(echo_caches);
2913 # ifdef HAVE_SERVER_SUPPORT
2917 class_unregister_type(LUSTRE_ECHO_NAME);
2919 echo_persistent_pages_fini();
2925 static void __exit obdecho_exit(void)
2927 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
2928 lu_kmem_fini(echo_caches);
2930 #ifdef HAVE_SERVER_SUPPORT
2931 class_unregister_type(LUSTRE_ECHO_NAME);
2932 echo_persistent_pages_fini();
2936 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2937 MODULE_DESCRIPTION("Lustre Echo Client test driver");
2938 MODULE_VERSION(LUSTRE_VERSION_STRING);
2939 MODULE_LICENSE("GPL");
2941 module_init(obdecho_init);
2942 module_exit(obdecho_exit);
2944 /** @} echo_client */