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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_ECHO
35 #include <linux/user_namespace.h>
36 #ifdef HAVE_UIDGID_HEADER
37 # include <linux/uidgid.h>
39 #include <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <obd_class.h>
44 #include <lustre_debug.h>
45 #include <lprocfs_status.h>
46 #include <cl_object.h>
47 #include <lustre_fid.h>
48 #include <lustre_lmv.h>
49 #include <lustre_acl.h>
50 #include <uapi/linux/lustre/lustre_ioctl.h>
51 #include <lustre_net.h>
52 #ifdef HAVE_SERVER_SUPPORT
53 # include <md_object.h>
55 #define ETI_NAME_LEN 20
57 #endif /* HAVE_SERVER_SUPPORT */
59 #include "echo_internal.h"
61 /** \defgroup echo_client Echo Client
66 struct cl_device ed_cl;
67 struct echo_client_obd *ed_ec;
69 struct cl_site ed_site_myself;
70 struct lu_site *ed_site;
71 struct lu_device *ed_next;
73 struct lu_client_seq *ed_cl_seq;
74 #ifdef HAVE_SERVER_SUPPORT
75 struct local_oid_storage *ed_los;
76 struct lu_fid ed_root_fid;
77 #endif /* HAVE_SERVER_SUPPORT */
81 struct cl_object eo_cl;
82 struct cl_object_header eo_hdr;
83 struct echo_device *eo_dev;
84 struct list_head eo_obj_chain;
85 struct lov_oinfo *eo_oinfo;
90 struct echo_object_conf {
91 struct cl_object_conf eoc_cl;
92 struct lov_oinfo **eoc_oinfo;
96 struct cl_page_slice ep_cl;
101 struct cl_lock_slice el_cl;
102 struct list_head el_chain;
103 struct echo_object *el_object;
105 atomic_t el_refcount;
108 #ifdef HAVE_SERVER_SUPPORT
109 static const char echo_md_root_dir_name[] = "ROOT_ECHO";
112 * In order to use the values of members in struct mdd_device,
113 * we define an alias structure here.
115 struct echo_md_device {
116 struct md_device emd_md_dev;
117 struct obd_export *emd_child_exp;
118 struct dt_device *emd_child;
119 struct dt_device *emd_bottom;
120 struct lu_fid emd_root_fid;
121 struct lu_fid emd_local_root_fid;
123 #endif /* HAVE_SERVER_SUPPORT */
125 static int echo_client_setup(const struct lu_env *env,
126 struct obd_device *obddev,
127 struct lustre_cfg *lcfg);
128 static int echo_client_cleanup(struct obd_device *obddev);
131 /** \defgroup echo_helpers Helper functions
134 static inline struct echo_device *cl2echo_dev(const struct cl_device *dev)
136 return container_of0(dev, struct echo_device, ed_cl);
139 static inline struct cl_device *echo_dev2cl(struct echo_device *d)
144 static inline struct echo_device *obd2echo_dev(const struct obd_device *obd)
146 return cl2echo_dev(lu2cl_dev(obd->obd_lu_dev));
149 static inline struct cl_object *echo_obj2cl(struct echo_object *eco)
154 static inline struct echo_object *cl2echo_obj(const struct cl_object *o)
156 return container_of(o, struct echo_object, eo_cl);
159 static inline struct echo_page *cl2echo_page(const struct cl_page_slice *s)
161 return container_of(s, struct echo_page, ep_cl);
164 static inline struct echo_lock *cl2echo_lock(const struct cl_lock_slice *s)
166 return container_of(s, struct echo_lock, el_cl);
169 static inline struct cl_lock *echo_lock2cl(const struct echo_lock *ecl)
171 return ecl->el_cl.cls_lock;
174 static struct lu_context_key echo_thread_key;
175 static inline struct echo_thread_info *echo_env_info(const struct lu_env *env)
177 struct echo_thread_info *info;
178 info = lu_context_key_get(&env->le_ctx, &echo_thread_key);
179 LASSERT(info != NULL);
184 struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c)
186 return container_of(c, struct echo_object_conf, eoc_cl);
189 #ifdef HAVE_SERVER_SUPPORT
190 static inline struct echo_md_device *lu2emd_dev(struct lu_device *d)
192 return container_of0(d, struct echo_md_device, emd_md_dev.md_lu_dev);
195 static inline struct lu_device *emd2lu_dev(struct echo_md_device *d)
197 return &d->emd_md_dev.md_lu_dev;
200 static inline struct seq_server_site *echo_md_seq_site(struct echo_md_device *d)
202 return emd2lu_dev(d)->ld_site->ld_seq_site;
205 static inline struct obd_device *emd2obd_dev(struct echo_md_device *d)
207 return d->emd_md_dev.md_lu_dev.ld_obd;
209 #endif /* HAVE_SERVER_SUPPORT */
211 /** @} echo_helpers */
213 static int cl_echo_object_put(struct echo_object *eco);
214 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
215 struct page **pages, int npages, int async);
217 struct echo_thread_info {
218 struct echo_object_conf eti_conf;
219 struct lustre_md eti_md;
221 struct cl_2queue eti_queue;
223 struct cl_lock eti_lock;
224 struct lu_fid eti_fid;
225 struct lu_fid eti_fid2;
226 #ifdef HAVE_SERVER_SUPPORT
227 struct md_op_spec eti_spec;
228 struct lov_mds_md_v3 eti_lmm;
229 struct lov_user_md_v3 eti_lum;
230 struct md_attr eti_ma;
231 struct lu_name eti_lname;
232 /* per-thread values, can be re-used */
233 void *eti_big_lmm; /* may be vmalloc'd */
235 char eti_name[ETI_NAME_LEN];
236 struct lu_buf eti_buf;
237 /* If we want to test large ACL, then need to enlarge the buffer. */
238 char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE_OLD];
242 /* No session used right now */
243 struct echo_session_info {
247 static struct kmem_cache *echo_lock_kmem;
248 static struct kmem_cache *echo_object_kmem;
249 static struct kmem_cache *echo_thread_kmem;
250 static struct kmem_cache *echo_session_kmem;
251 /* static struct kmem_cache *echo_req_kmem; */
253 static struct lu_kmem_descr echo_caches[] = {
255 .ckd_cache = &echo_lock_kmem,
256 .ckd_name = "echo_lock_kmem",
257 .ckd_size = sizeof (struct echo_lock)
260 .ckd_cache = &echo_object_kmem,
261 .ckd_name = "echo_object_kmem",
262 .ckd_size = sizeof (struct echo_object)
265 .ckd_cache = &echo_thread_kmem,
266 .ckd_name = "echo_thread_kmem",
267 .ckd_size = sizeof (struct echo_thread_info)
270 .ckd_cache = &echo_session_kmem,
271 .ckd_name = "echo_session_kmem",
272 .ckd_size = sizeof (struct echo_session_info)
279 /** \defgroup echo_page Page operations
281 * Echo page operations.
285 static int echo_page_own(const struct lu_env *env,
286 const struct cl_page_slice *slice,
287 struct cl_io *io, int nonblock)
289 struct echo_page *ep = cl2echo_page(slice);
292 mutex_lock(&ep->ep_lock);
293 else if (!mutex_trylock(&ep->ep_lock))
298 static void echo_page_disown(const struct lu_env *env,
299 const struct cl_page_slice *slice,
302 struct echo_page *ep = cl2echo_page(slice);
304 LASSERT(mutex_is_locked(&ep->ep_lock));
305 mutex_unlock(&ep->ep_lock);
308 static void echo_page_discard(const struct lu_env *env,
309 const struct cl_page_slice *slice,
310 struct cl_io *unused)
312 cl_page_delete(env, slice->cpl_page);
315 static int echo_page_is_vmlocked(const struct lu_env *env,
316 const struct cl_page_slice *slice)
318 if (mutex_is_locked(&cl2echo_page(slice)->ep_lock))
323 static void echo_page_completion(const struct lu_env *env,
324 const struct cl_page_slice *slice,
327 LASSERT(slice->cpl_page->cp_sync_io != NULL);
330 static void echo_page_fini(const struct lu_env *env,
331 struct cl_page_slice *slice,
332 struct pagevec *pvec)
334 struct echo_object *eco = cl2echo_obj(slice->cpl_obj);
337 atomic_dec(&eco->eo_npages);
338 put_page(slice->cpl_page->cp_vmpage);
342 static int echo_page_prep(const struct lu_env *env,
343 const struct cl_page_slice *slice,
344 struct cl_io *unused)
349 static int echo_page_print(const struct lu_env *env,
350 const struct cl_page_slice *slice,
351 void *cookie, lu_printer_t printer)
353 struct echo_page *ep = cl2echo_page(slice);
355 (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n",
356 ep, mutex_is_locked(&ep->ep_lock),
357 slice->cpl_page->cp_vmpage);
361 static const struct cl_page_operations echo_page_ops = {
362 .cpo_own = echo_page_own,
363 .cpo_disown = echo_page_disown,
364 .cpo_discard = echo_page_discard,
365 .cpo_fini = echo_page_fini,
366 .cpo_print = echo_page_print,
367 .cpo_is_vmlocked = echo_page_is_vmlocked,
370 .cpo_prep = echo_page_prep,
371 .cpo_completion = echo_page_completion,
374 .cpo_prep = echo_page_prep,
375 .cpo_completion = echo_page_completion,
381 /** \defgroup echo_lock Locking
383 * echo lock operations
387 static void echo_lock_fini(const struct lu_env *env,
388 struct cl_lock_slice *slice)
390 struct echo_lock *ecl = cl2echo_lock(slice);
392 LASSERT(list_empty(&ecl->el_chain));
393 OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem);
396 static struct cl_lock_operations echo_lock_ops = {
397 .clo_fini = echo_lock_fini,
402 /** \defgroup echo_cl_ops cl_object operations
404 * operations for cl_object
408 static int echo_page_init(const struct lu_env *env, struct cl_object *obj,
409 struct cl_page *page, pgoff_t index)
411 struct echo_page *ep = cl_object_page_slice(obj, page);
412 struct echo_object *eco = cl2echo_obj(obj);
415 get_page(page->cp_vmpage);
416 mutex_init(&ep->ep_lock);
417 cl_page_slice_add(page, &ep->ep_cl, obj, index, &echo_page_ops);
418 atomic_inc(&eco->eo_npages);
422 static int echo_io_init(const struct lu_env *env, struct cl_object *obj,
428 static int echo_lock_init(const struct lu_env *env,
429 struct cl_object *obj, struct cl_lock *lock,
430 const struct cl_io *unused)
432 struct echo_lock *el;
435 OBD_SLAB_ALLOC_PTR_GFP(el, echo_lock_kmem, GFP_NOFS);
437 cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops);
438 el->el_object = cl2echo_obj(obj);
439 INIT_LIST_HEAD(&el->el_chain);
440 atomic_set(&el->el_refcount, 0);
442 RETURN(el == NULL ? -ENOMEM : 0);
445 static int echo_conf_set(const struct lu_env *env, struct cl_object *obj,
446 const struct cl_object_conf *conf)
451 static const struct cl_object_operations echo_cl_obj_ops = {
452 .coo_page_init = echo_page_init,
453 .coo_lock_init = echo_lock_init,
454 .coo_io_init = echo_io_init,
455 .coo_conf_set = echo_conf_set
457 /** @} echo_cl_ops */
459 /** \defgroup echo_lu_ops lu_object operations
461 * operations for echo lu object.
465 static int echo_object_init(const struct lu_env *env, struct lu_object *obj,
466 const struct lu_object_conf *conf)
468 struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev));
469 struct echo_client_obd *ec = ed->ed_ec;
470 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
474 struct lu_object *below;
475 struct lu_device *under;
478 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header,
482 lu_object_add(obj, below);
485 if (!ed->ed_next_ismd) {
486 const struct cl_object_conf *cconf = lu2cl_conf(conf);
487 struct echo_object_conf *econf = cl2echo_conf(cconf);
489 LASSERT(econf->eoc_oinfo != NULL);
491 /* Transfer the oinfo pointer to eco that it won't be
493 eco->eo_oinfo = *econf->eoc_oinfo;
494 *econf->eoc_oinfo = NULL;
496 eco->eo_oinfo = NULL;
500 atomic_set(&eco->eo_npages, 0);
501 cl_object_page_init(lu2cl(obj), sizeof(struct echo_page));
503 spin_lock(&ec->ec_lock);
504 list_add_tail(&eco->eo_obj_chain, &ec->ec_objects);
505 spin_unlock(&ec->ec_lock);
510 static void echo_object_free(const struct lu_env *env, struct lu_object *obj)
512 struct echo_object *eco = cl2echo_obj(lu2cl(obj));
513 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
516 LASSERT(atomic_read(&eco->eo_npages) == 0);
518 spin_lock(&ec->ec_lock);
519 list_del_init(&eco->eo_obj_chain);
520 spin_unlock(&ec->ec_lock);
523 lu_object_header_fini(obj->lo_header);
525 if (eco->eo_oinfo != NULL)
526 OBD_FREE_PTR(eco->eo_oinfo);
528 OBD_SLAB_FREE_PTR(eco, echo_object_kmem);
532 static int echo_object_print(const struct lu_env *env, void *cookie,
533 lu_printer_t p, const struct lu_object *o)
535 struct echo_object *obj = cl2echo_obj(lu2cl(o));
537 return (*p)(env, cookie, "echoclient-object@%p", obj);
540 static const struct lu_object_operations echo_lu_obj_ops = {
541 .loo_object_init = echo_object_init,
542 .loo_object_delete = NULL,
543 .loo_object_release = NULL,
544 .loo_object_free = echo_object_free,
545 .loo_object_print = echo_object_print,
546 .loo_object_invariant = NULL
548 /** @} echo_lu_ops */
550 /** \defgroup echo_lu_dev_ops lu_device operations
552 * Operations for echo lu device.
556 static struct lu_object *echo_object_alloc(const struct lu_env *env,
557 const struct lu_object_header *hdr,
558 struct lu_device *dev)
560 struct echo_object *eco;
561 struct lu_object *obj = NULL;
564 /* we're the top dev. */
565 LASSERT(hdr == NULL);
566 OBD_SLAB_ALLOC_PTR_GFP(eco, echo_object_kmem, GFP_NOFS);
568 struct cl_object_header *hdr = &eco->eo_hdr;
570 obj = &echo_obj2cl(eco)->co_lu;
571 cl_object_header_init(hdr);
572 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
574 lu_object_init(obj, &hdr->coh_lu, dev);
575 lu_object_add_top(&hdr->coh_lu, obj);
577 eco->eo_cl.co_ops = &echo_cl_obj_ops;
578 obj->lo_ops = &echo_lu_obj_ops;
583 static struct lu_device_operations echo_device_lu_ops = {
584 .ldo_object_alloc = echo_object_alloc,
587 /** @} echo_lu_dev_ops */
589 /** \defgroup echo_init Setup and teardown
591 * Init and fini functions for echo client.
595 static int echo_site_init(const struct lu_env *env, struct echo_device *ed)
597 struct cl_site *site = &ed->ed_site_myself;
600 /* initialize site */
601 rc = cl_site_init(site, &ed->ed_cl);
603 CERROR("Cannot initialize site for echo client(%d)\n", rc);
607 rc = lu_site_init_finish(&site->cs_lu);
613 ed->ed_site = &site->cs_lu;
617 static void echo_site_fini(const struct lu_env *env, struct echo_device *ed)
620 if (!ed->ed_next_ismd)
621 lu_site_fini(ed->ed_site);
626 static void *echo_thread_key_init(const struct lu_context *ctx,
627 struct lu_context_key *key)
629 struct echo_thread_info *info;
631 OBD_SLAB_ALLOC_PTR_GFP(info, echo_thread_kmem, GFP_NOFS);
633 info = ERR_PTR(-ENOMEM);
637 static void echo_thread_key_fini(const struct lu_context *ctx,
638 struct lu_context_key *key, void *data)
640 struct echo_thread_info *info = data;
641 OBD_SLAB_FREE_PTR(info, echo_thread_kmem);
644 static struct lu_context_key echo_thread_key = {
645 .lct_tags = LCT_CL_THREAD,
646 .lct_init = echo_thread_key_init,
647 .lct_fini = echo_thread_key_fini,
650 static void *echo_session_key_init(const struct lu_context *ctx,
651 struct lu_context_key *key)
653 struct echo_session_info *session;
655 OBD_SLAB_ALLOC_PTR_GFP(session, echo_session_kmem, GFP_NOFS);
657 session = ERR_PTR(-ENOMEM);
661 static void echo_session_key_fini(const struct lu_context *ctx,
662 struct lu_context_key *key, void *data)
664 struct echo_session_info *session = data;
665 OBD_SLAB_FREE_PTR(session, echo_session_kmem);
668 static struct lu_context_key echo_session_key = {
669 .lct_tags = LCT_SESSION,
670 .lct_init = echo_session_key_init,
671 .lct_fini = echo_session_key_fini,
674 LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key);
676 #ifdef HAVE_SERVER_SUPPORT
677 # define ECHO_SEQ_WIDTH 0xffffffff
678 static int echo_fid_init(struct echo_device *ed, char *obd_name,
679 struct seq_server_site *ss)
685 OBD_ALLOC_PTR(ed->ed_cl_seq);
686 if (ed->ed_cl_seq == NULL)
689 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
691 GOTO(out_free_seq, rc = -ENOMEM);
693 snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name);
695 /* Init client side sequence-manager */
696 rc = seq_client_init(ed->ed_cl_seq, NULL,
698 prefix, ss->ss_server_seq);
699 ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH;
700 OBD_FREE(prefix, MAX_OBD_NAME + 5);
702 GOTO(out_free_seq, rc);
707 OBD_FREE_PTR(ed->ed_cl_seq);
708 ed->ed_cl_seq = NULL;
712 static int echo_fid_fini(struct obd_device *obddev)
714 struct echo_device *ed = obd2echo_dev(obddev);
717 if (ed->ed_cl_seq != NULL) {
718 seq_client_fini(ed->ed_cl_seq);
719 OBD_FREE_PTR(ed->ed_cl_seq);
720 ed->ed_cl_seq = NULL;
726 static void echo_ed_los_fini(const struct lu_env *env, struct echo_device *ed)
730 if (ed != NULL && ed->ed_next_ismd && ed->ed_los != NULL) {
731 local_oid_storage_fini(env, ed->ed_los);
737 echo_md_local_file_create(const struct lu_env *env, struct echo_md_device *emd,
738 struct local_oid_storage *los,
739 const struct lu_fid *pfid, const char *name,
740 __u32 mode, struct lu_fid *fid)
742 struct dt_object *parent = NULL;
743 struct dt_object *dto = NULL;
747 LASSERT(!fid_is_zero(pfid));
748 parent = dt_locate(env, emd->emd_bottom, pfid);
749 if (unlikely(IS_ERR(parent)))
750 RETURN(PTR_ERR(parent));
752 /* create local file with @fid */
753 dto = local_file_find_or_create_with_fid(env, emd->emd_bottom, fid,
756 GOTO(out_put, rc = PTR_ERR(dto));
758 *fid = *lu_object_fid(&dto->do_lu);
759 /* since stack is not fully set up the local_storage uses own stack
760 * and we should drop its object from cache */
761 dt_object_put_nocache(env, dto);
765 dt_object_put(env, parent);
770 echo_md_root_get(const struct lu_env *env, struct echo_md_device *emd,
771 struct echo_device *ed)
777 /* Setup local dirs */
778 fid.f_seq = FID_SEQ_LOCAL_NAME;
781 rc = local_oid_storage_init(env, emd->emd_bottom, &fid, &ed->ed_los);
785 lu_echo_root_fid(&fid);
786 if (echo_md_seq_site(emd)->ss_node_id == 0) {
787 rc = echo_md_local_file_create(env, emd, ed->ed_los,
788 &emd->emd_local_root_fid,
789 echo_md_root_dir_name, S_IFDIR |
790 S_IRUGO | S_IWUSR | S_IXUGO,
793 CERROR("%s: create md echo root fid failed: rc = %d\n",
794 emd2obd_dev(emd)->obd_name, rc);
798 ed->ed_root_fid = fid;
802 echo_ed_los_fini(env, ed);
806 #endif /* HAVE_SERVER_SUPPORT */
808 static struct lu_device *echo_device_alloc(const struct lu_env *env,
809 struct lu_device_type *t,
810 struct lustre_cfg *cfg)
812 struct lu_device *next;
813 struct echo_device *ed;
814 struct cl_device *cd;
815 struct obd_device *obd = NULL; /* to keep compiler happy */
816 struct obd_device *tgt;
817 const char *tgt_type_name;
824 GOTO(out, rc = -ENOMEM);
828 rc = cl_device_init(cd, t);
832 cd->cd_lu_dev.ld_ops = &echo_device_lu_ops;
835 obd = class_name2obd(lustre_cfg_string(cfg, 0));
836 LASSERT(obd != NULL);
837 LASSERT(env != NULL);
839 tgt = class_name2obd(lustre_cfg_string(cfg, 1));
841 CERROR("Can not find tgt device %s\n",
842 lustre_cfg_string(cfg, 1));
843 GOTO(out, rc = -ENODEV);
846 next = tgt->obd_lu_dev;
848 if (strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME) == 0) {
849 ed->ed_next_ismd = 1;
850 } else if (strcmp(tgt->obd_type->typ_name, LUSTRE_OST_NAME) == 0 ||
851 strcmp(tgt->obd_type->typ_name, LUSTRE_OSC_NAME) == 0) {
852 ed->ed_next_ismd = 0;
853 rc = echo_site_init(env, ed);
857 GOTO(out, rc = -EINVAL);
862 rc = echo_client_setup(env, obd, cfg);
866 ed->ed_ec = &obd->u.echo_client;
869 if (ed->ed_next_ismd) {
870 #ifdef HAVE_SERVER_SUPPORT
871 /* Suppose to connect to some Metadata layer */
872 struct lu_site *ls = NULL;
873 struct lu_device *ld = NULL;
874 struct md_device *md = NULL;
875 struct echo_md_device *emd = NULL;
879 CERROR("%s is not lu device type!\n",
880 lustre_cfg_string(cfg, 1));
881 GOTO(out, rc = -EINVAL);
884 tgt_type_name = lustre_cfg_string(cfg, 2);
885 if (!tgt_type_name) {
886 CERROR("%s no type name for echo %s setup\n",
887 lustre_cfg_string(cfg, 1),
888 tgt->obd_type->typ_name);
889 GOTO(out, rc = -EINVAL);
894 spin_lock(&ls->ls_ld_lock);
895 list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) {
896 if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) {
901 spin_unlock(&ls->ls_ld_lock);
904 CERROR("%s is not lu device type!\n",
905 lustre_cfg_string(cfg, 1));
906 GOTO(out, rc = -EINVAL);
910 /* For MD echo client, it will use the site in MDS stack */
912 ed->ed_cl.cd_lu_dev.ld_site = ls;
913 rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls));
915 CERROR("echo fid init error %d\n", rc);
919 md = lu2md_dev(next);
920 emd = lu2emd_dev(&md->md_lu_dev);
921 rc = echo_md_root_get(env, emd, ed);
923 CERROR("%s: get root error: rc = %d\n",
924 emd2obd_dev(emd)->obd_name, rc);
927 #else /* !HAVE_SERVER_SUPPORT */
928 CERROR("Local operations are NOT supported on client side. "
929 "Only remote operations are supported. Metadata client "
930 "must be run on server side.\n");
931 GOTO(out, rc = -EOPNOTSUPP);
932 #endif /* HAVE_SERVER_SUPPORT */
934 /* if echo client is to be stacked upon ost device, the next is
935 * NULL since ost is not a clio device so far */
936 if (next != NULL && !lu_device_is_cl(next))
939 tgt_type_name = tgt->obd_type->typ_name;
941 LASSERT(next != NULL);
942 if (next->ld_site != NULL)
943 GOTO(out, rc = -EBUSY);
945 next->ld_site = ed->ed_site;
946 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
947 next->ld_type->ldt_name,
952 LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0);
956 RETURN(&cd->cd_lu_dev);
961 rc2 = echo_client_cleanup(obd);
963 CERROR("Cleanup obd device %s error(%d)\n",
968 echo_site_fini(env, ed);
970 cl_device_fini(&ed->ed_cl);
980 static int echo_device_init(const struct lu_env *env, struct lu_device *d,
981 const char *name, struct lu_device *next)
987 static struct lu_device *echo_device_fini(const struct lu_env *env,
990 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
991 struct lu_device *next = ed->ed_next;
993 while (next && !ed->ed_next_ismd)
994 next = next->ld_type->ldt_ops->ldto_device_fini(env, next);
998 static void echo_lock_release(const struct lu_env *env,
999 struct echo_lock *ecl,
1002 struct cl_lock *clk = echo_lock2cl(ecl);
1004 cl_lock_release(env, clk);
1007 static struct lu_device *echo_device_free(const struct lu_env *env,
1008 struct lu_device *d)
1010 struct echo_device *ed = cl2echo_dev(lu2cl_dev(d));
1011 struct echo_client_obd *ec = ed->ed_ec;
1012 struct echo_object *eco;
1013 struct lu_device *next = ed->ed_next;
1015 CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n",
1018 lu_site_purge(env, ed->ed_site, -1);
1020 /* check if there are objects still alive.
1021 * It shouldn't have any object because lu_site_purge would cleanup
1022 * all of cached objects. Anyway, probably the echo device is being
1023 * parallelly accessed.
1025 spin_lock(&ec->ec_lock);
1026 list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain)
1027 eco->eo_deleted = 1;
1028 spin_unlock(&ec->ec_lock);
1031 lu_site_purge(env, ed->ed_site, -1);
1034 "Waiting for the reference of echo object to be dropped\n");
1036 /* Wait for the last reference to be dropped. */
1037 spin_lock(&ec->ec_lock);
1038 while (!list_empty(&ec->ec_objects)) {
1039 spin_unlock(&ec->ec_lock);
1040 CERROR("echo_client still has objects at cleanup time, "
1041 "wait for 1 second\n");
1042 set_current_state(TASK_UNINTERRUPTIBLE);
1043 schedule_timeout(cfs_time_seconds(1));
1044 lu_site_purge(env, ed->ed_site, -1);
1045 spin_lock(&ec->ec_lock);
1047 spin_unlock(&ec->ec_lock);
1049 LASSERT(list_empty(&ec->ec_locks));
1051 CDEBUG(D_INFO, "No object exists, exiting...\n");
1053 echo_client_cleanup(d->ld_obd);
1054 #ifdef HAVE_SERVER_SUPPORT
1055 echo_fid_fini(d->ld_obd);
1056 echo_ed_los_fini(env, ed);
1058 while (next && !ed->ed_next_ismd)
1059 next = next->ld_type->ldt_ops->ldto_device_free(env, next);
1061 LASSERT(ed->ed_site == d->ld_site);
1062 echo_site_fini(env, ed);
1063 cl_device_fini(&ed->ed_cl);
1066 cl_env_cache_purge(~0);
1071 static const struct lu_device_type_operations echo_device_type_ops = {
1072 .ldto_init = echo_type_init,
1073 .ldto_fini = echo_type_fini,
1075 .ldto_start = echo_type_start,
1076 .ldto_stop = echo_type_stop,
1078 .ldto_device_alloc = echo_device_alloc,
1079 .ldto_device_free = echo_device_free,
1080 .ldto_device_init = echo_device_init,
1081 .ldto_device_fini = echo_device_fini
1084 static struct lu_device_type echo_device_type = {
1085 .ldt_tags = LU_DEVICE_CL,
1086 .ldt_name = LUSTRE_ECHO_CLIENT_NAME,
1087 .ldt_ops = &echo_device_type_ops,
1088 .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD,
1092 /** \defgroup echo_exports Exported operations
1094 * exporting functions to echo client
1099 /* Interfaces to echo client obd device */
1100 static struct echo_object *
1101 cl_echo_object_find(struct echo_device *d, const struct ost_id *oi)
1104 struct echo_thread_info *info;
1105 struct echo_object_conf *conf;
1106 struct echo_object *eco;
1107 struct cl_object *obj;
1108 struct lov_oinfo *oinfo = NULL;
1114 LASSERTF(ostid_id(oi) != 0, DOSTID"\n", POSTID(oi));
1115 LASSERTF(ostid_seq(oi) == FID_SEQ_ECHO, DOSTID"\n", POSTID(oi));
1117 /* Never return an object if the obd is to be freed. */
1118 if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping)
1119 RETURN(ERR_PTR(-ENODEV));
1121 env = cl_env_get(&refcheck);
1123 RETURN((void *)env);
1125 info = echo_env_info(env);
1126 conf = &info->eti_conf;
1128 OBD_ALLOC_PTR(oinfo);
1130 GOTO(out, eco = ERR_PTR(-ENOMEM));
1132 oinfo->loi_oi = *oi;
1133 conf->eoc_cl.u.coc_oinfo = oinfo;
1136 /* If echo_object_init() is successful then ownership of oinfo
1137 * is transferred to the object. */
1138 conf->eoc_oinfo = &oinfo;
1140 fid = &info->eti_fid;
1141 rc = ostid_to_fid(fid, oi, 0);
1143 GOTO(out, eco = ERR_PTR(rc));
1145 /* In the function below, .hs_keycmp resolves to
1146 * lu_obj_hop_keycmp() */
1147 /* coverity[overrun-buffer-val] */
1148 obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl);
1150 GOTO(out, eco = (void*)obj);
1152 eco = cl2echo_obj(obj);
1153 if (eco->eo_deleted) {
1154 cl_object_put(env, obj);
1155 eco = ERR_PTR(-EAGAIN);
1160 OBD_FREE_PTR(oinfo);
1162 cl_env_put(env, &refcheck);
1166 static int cl_echo_object_put(struct echo_object *eco)
1169 struct cl_object *obj = echo_obj2cl(eco);
1173 env = cl_env_get(&refcheck);
1175 RETURN(PTR_ERR(env));
1177 /* an external function to kill an object? */
1178 if (eco->eo_deleted) {
1179 struct lu_object_header *loh = obj->co_lu.lo_header;
1180 LASSERT(&eco->eo_hdr == luh2coh(loh));
1181 set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags);
1184 cl_object_put(env, obj);
1185 cl_env_put(env, &refcheck);
1189 static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco,
1190 u64 start, u64 end, int mode,
1191 __u64 *cookie , __u32 enqflags)
1194 struct cl_lock *lck;
1195 struct cl_object *obj;
1196 struct cl_lock_descr *descr;
1197 struct echo_thread_info *info;
1201 info = echo_env_info(env);
1203 lck = &info->eti_lock;
1204 obj = echo_obj2cl(eco);
1206 memset(lck, 0, sizeof(*lck));
1207 descr = &lck->cll_descr;
1208 descr->cld_obj = obj;
1209 descr->cld_start = cl_index(obj, start);
1210 descr->cld_end = cl_index(obj, end);
1211 descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ;
1212 descr->cld_enq_flags = enqflags;
1215 rc = cl_lock_request(env, io, lck);
1217 struct echo_client_obd *ec = eco->eo_dev->ed_ec;
1218 struct echo_lock *el;
1220 el = cl2echo_lock(cl_lock_at(lck, &echo_device_type));
1221 spin_lock(&ec->ec_lock);
1222 if (list_empty(&el->el_chain)) {
1223 list_add(&el->el_chain, &ec->ec_locks);
1224 el->el_cookie = ++ec->ec_unique;
1226 atomic_inc(&el->el_refcount);
1227 *cookie = el->el_cookie;
1228 spin_unlock(&ec->ec_lock);
1233 static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed,
1236 struct echo_client_obd *ec = ed->ed_ec;
1237 struct echo_lock *ecl = NULL;
1238 struct list_head *el;
1239 int found = 0, still_used = 0;
1242 LASSERT(ec != NULL);
1243 spin_lock(&ec->ec_lock);
1244 list_for_each(el, &ec->ec_locks) {
1245 ecl = list_entry(el, struct echo_lock, el_chain);
1246 CDEBUG(D_INFO, "ecl: %p, cookie: %#llx\n", ecl, ecl->el_cookie);
1247 found = (ecl->el_cookie == cookie);
1249 if (atomic_dec_and_test(&ecl->el_refcount))
1250 list_del_init(&ecl->el_chain);
1256 spin_unlock(&ec->ec_lock);
1261 echo_lock_release(env, ecl, still_used);
1265 static void echo_commit_callback(const struct lu_env *env, struct cl_io *io,
1266 struct cl_page *page)
1268 struct echo_thread_info *info;
1269 struct cl_2queue *queue;
1271 info = echo_env_info(env);
1272 LASSERT(io == &info->eti_io);
1274 queue = &info->eti_queue;
1275 cl_page_list_add(&queue->c2_qout, page);
1278 static int cl_echo_object_brw(struct echo_object *eco, int rw, u64 offset,
1279 struct page **pages, int npages, int async)
1282 struct echo_thread_info *info;
1283 struct cl_object *obj = echo_obj2cl(eco);
1284 struct echo_device *ed = eco->eo_dev;
1285 struct cl_2queue *queue;
1287 struct cl_page *clp;
1288 struct lustre_handle lh = { 0 };
1289 int page_size = cl_page_size(obj);
1295 LASSERT((offset & ~PAGE_MASK) == 0);
1296 LASSERT(ed->ed_next != NULL);
1297 env = cl_env_get(&refcheck);
1299 RETURN(PTR_ERR(env));
1301 info = echo_env_info(env);
1303 queue = &info->eti_queue;
1305 cl_2queue_init(queue);
1307 io->ci_ignore_layout = 1;
1308 rc = cl_io_init(env, io, CIT_MISC, obj);
1314 rc = cl_echo_enqueue0(env, eco, offset,
1315 offset + npages * PAGE_SIZE - 1,
1316 rw == READ ? LCK_PR : LCK_PW, &lh.cookie,
1319 GOTO(error_lock, rc);
1321 for (i = 0; i < npages; i++) {
1323 clp = cl_page_find(env, obj, cl_index(obj, offset),
1324 pages[i], CPT_TRANSIENT);
1329 LASSERT(clp->cp_type == CPT_TRANSIENT);
1331 rc = cl_page_own(env, io, clp);
1333 LASSERT(clp->cp_state == CPS_FREEING);
1334 cl_page_put(env, clp);
1338 cl_2queue_add(queue, clp);
1340 /* drop the reference count for cl_page_find, so that the page
1341 * will be freed in cl_2queue_fini. */
1342 cl_page_put(env, clp);
1343 cl_page_clip(env, clp, 0, page_size);
1345 offset += page_size;
1349 enum cl_req_type typ = rw == READ ? CRT_READ : CRT_WRITE;
1351 async = async && (typ == CRT_WRITE);
1353 rc = cl_io_commit_async(env, io, &queue->c2_qin,
1355 echo_commit_callback);
1357 rc = cl_io_submit_sync(env, io, typ, queue, 0);
1358 CDEBUG(D_INFO, "echo_client %s write returns %d\n",
1359 async ? "async" : "sync", rc);
1362 cl_echo_cancel0(env, ed, lh.cookie);
1365 cl_2queue_discard(env, io, queue);
1366 cl_2queue_disown(env, io, queue);
1367 cl_2queue_fini(env, queue);
1368 cl_io_fini(env, io);
1370 cl_env_put(env, &refcheck);
1373 /** @} echo_exports */
1376 static u64 last_object_id;
1378 #ifdef HAVE_SERVER_SUPPORT
1379 static inline void echo_md_build_name(struct lu_name *lname, char *name,
1382 snprintf(name, ETI_NAME_LEN, "%llu", id);
1383 lname->ln_name = name;
1384 lname->ln_namelen = strlen(name);
1387 /* similar to mdt_attr_get_complex */
1388 static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o,
1391 struct echo_thread_info *info = echo_env_info(env);
1396 LASSERT(ma->ma_lmm_size > 0);
1398 LASSERT(ma->ma_need & (MA_LOV | MA_LMV));
1399 if (ma->ma_need & MA_LOV)
1400 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV);
1402 rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LMV);
1407 /* big_lmm may need to be grown */
1408 if (info->eti_big_lmmsize < rc) {
1409 int size = size_roundup_power2(rc);
1411 if (info->eti_big_lmmsize > 0) {
1412 /* free old buffer */
1413 LASSERT(info->eti_big_lmm);
1414 OBD_FREE_LARGE(info->eti_big_lmm,
1415 info->eti_big_lmmsize);
1416 info->eti_big_lmm = NULL;
1417 info->eti_big_lmmsize = 0;
1420 OBD_ALLOC_LARGE(info->eti_big_lmm, size);
1421 if (info->eti_big_lmm == NULL)
1423 info->eti_big_lmmsize = size;
1425 LASSERT(info->eti_big_lmmsize >= rc);
1427 info->eti_buf.lb_buf = info->eti_big_lmm;
1428 info->eti_buf.lb_len = info->eti_big_lmmsize;
1429 if (ma->ma_need & MA_LOV)
1430 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV);
1432 rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LMV);
1436 if (ma->ma_need & MA_LOV)
1437 ma->ma_valid |= MA_LOV;
1439 ma->ma_valid |= MA_LMV;
1441 ma->ma_lmm = info->eti_big_lmm;
1442 ma->ma_lmm_size = rc;
1447 static int echo_attr_get_complex(const struct lu_env *env,
1448 struct md_object *next,
1451 struct echo_thread_info *info = echo_env_info(env);
1452 struct lu_buf *buf = &info->eti_buf;
1453 umode_t mode = lu_object_attr(&next->mo_lu);
1460 if (ma->ma_need & MA_INODE) {
1461 rc = mo_attr_get(env, next, ma);
1464 ma->ma_valid |= MA_INODE;
1467 if ((ma->ma_need & MA_LOV) && (S_ISREG(mode) || S_ISDIR(mode))) {
1468 LASSERT(ma->ma_lmm_size > 0);
1469 buf->lb_buf = ma->ma_lmm;
1470 buf->lb_len = ma->ma_lmm_size;
1471 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV);
1473 ma->ma_lmm_size = rc2;
1474 ma->ma_valid |= MA_LOV;
1475 } else if (rc2 == -ENODATA) {
1477 ma->ma_lmm_size = 0;
1478 } else if (rc2 == -ERANGE) {
1479 rc2 = echo_big_lmm_get(env, next, ma);
1481 GOTO(out, rc = rc2);
1483 GOTO(out, rc = rc2);
1487 if ((ma->ma_need & MA_LMV) && S_ISDIR(mode)) {
1488 LASSERT(ma->ma_lmm_size > 0);
1489 buf->lb_buf = ma->ma_lmm;
1490 buf->lb_len = ma->ma_lmm_size;
1491 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
1493 ma->ma_lmm_size = rc2;
1494 ma->ma_valid |= MA_LMV;
1495 } else if (rc2 == -ENODATA) {
1497 ma->ma_lmm_size = 0;
1498 } else if (rc2 == -ERANGE) {
1499 rc2 = echo_big_lmm_get(env, next, ma);
1501 GOTO(out, rc = rc2);
1503 GOTO(out, rc = rc2);
1507 #ifdef CONFIG_FS_POSIX_ACL
1508 if ((ma->ma_need & MA_ACL_DEF) && S_ISDIR(mode)) {
1509 buf->lb_buf = ma->ma_acl;
1510 buf->lb_len = ma->ma_acl_size;
1511 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1513 ma->ma_acl_size = rc2;
1514 ma->ma_valid |= MA_ACL_DEF;
1515 } else if (rc2 == -ENODATA) {
1517 ma->ma_acl_size = 0;
1519 GOTO(out, rc = rc2);
1524 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1525 rc, ma->ma_valid, ma->ma_lmm);
1530 echo_md_create_internal(const struct lu_env *env, struct echo_device *ed,
1531 struct md_object *parent, struct lu_fid *fid,
1532 struct lu_name *lname, struct md_op_spec *spec,
1535 struct lu_object *ec_child, *child;
1536 struct lu_device *ld = ed->ed_next;
1537 struct echo_thread_info *info = echo_env_info(env);
1538 struct lu_fid *fid2 = &info->eti_fid2;
1539 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1544 rc = mdo_lookup(env, parent, lname, fid2, spec);
1547 else if (rc != -ENOENT)
1550 ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev,
1552 if (IS_ERR(ec_child)) {
1553 CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid),
1555 RETURN(PTR_ERR(ec_child));
1558 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1559 if (child == NULL) {
1560 CERROR("Can not locate the child "DFID"\n", PFID(fid));
1561 GOTO(out_put, rc = -EINVAL);
1564 CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n",
1565 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
1568 * Do not perform lookup sanity check. We know that name does not exist.
1570 spec->sp_cr_lookup = 0;
1571 rc = mdo_create(env, parent, lname, lu2md(child), spec, ma);
1573 CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc);
1576 CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n",
1577 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc);
1580 lu_object_put(env, ec_child);
1584 static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld,
1587 struct echo_thread_info *info = echo_env_info(env);
1589 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
1590 ma->ma_lmm = (void *)&info->eti_lmm;
1591 ma->ma_lmm_size = sizeof(info->eti_lmm);
1593 LASSERT(info->eti_big_lmmsize);
1594 ma->ma_lmm = info->eti_big_lmm;
1595 ma->ma_lmm_size = info->eti_big_lmmsize;
1602 echo_md_dir_stripe_choose(const struct lu_env *env, struct echo_device *ed,
1603 struct lu_object *obj, const char *name,
1604 unsigned int namelen, __u64 id,
1605 struct lu_object **new_parent)
1607 struct echo_thread_info *info = echo_env_info(env);
1608 struct md_attr *ma = &info->eti_ma;
1609 struct lmv_mds_md_v1 *lmv;
1610 struct lu_device *ld = ed->ed_next;
1612 struct lu_name tmp_ln_name;
1613 struct lu_fid stripe_fid;
1614 struct lu_object *stripe_obj;
1617 LASSERT(obj != NULL);
1618 LASSERT(S_ISDIR(obj->lo_header->loh_attr));
1620 memset(ma, 0, sizeof(*ma));
1621 echo_set_lmm_size(env, ld, ma);
1622 ma->ma_need = MA_LMV;
1623 rc = echo_attr_get_complex(env, lu2md(obj), ma);
1625 CERROR("Can not getattr child "DFID": rc = %d\n",
1626 PFID(lu_object_fid(obj)), rc);
1630 if (!(ma->ma_valid & MA_LMV)) {
1635 lmv = (struct lmv_mds_md_v1 *)ma->ma_lmm;
1636 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1) {
1638 CERROR("Invalid mds md magic %x "DFID": rc = %d\n",
1639 le32_to_cpu(lmv->lmv_magic), PFID(lu_object_fid(obj)),
1645 tmp_ln_name.ln_name = name;
1646 tmp_ln_name.ln_namelen = namelen;
1649 echo_md_build_name(&tmp_ln_name, info->eti_name, id);
1652 idx = lmv_name_to_stripe_index(LMV_HASH_TYPE_FNV_1A_64,
1653 le32_to_cpu(lmv->lmv_stripe_count),
1654 tmp_ln_name.ln_name, tmp_ln_name.ln_namelen);
1656 LASSERT(idx < le32_to_cpu(lmv->lmv_stripe_count));
1657 fid_le_to_cpu(&stripe_fid, &lmv->lmv_stripe_fids[idx]);
1659 stripe_obj = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, &stripe_fid,
1661 if (IS_ERR(stripe_obj)) {
1662 rc = PTR_ERR(stripe_obj);
1663 CERROR("Can not find the parent "DFID": rc = %d\n",
1664 PFID(&stripe_fid), rc);
1668 *new_parent = lu_object_locate(stripe_obj->lo_header, ld->ld_type);
1669 if (*new_parent == NULL) {
1670 lu_object_put(env, stripe_obj);
1677 static int echo_create_md_object(const struct lu_env *env,
1678 struct echo_device *ed,
1679 struct lu_object *ec_parent,
1681 char *name, int namelen,
1682 __u64 id, __u32 mode, int count,
1683 int stripe_count, int stripe_offset)
1685 struct lu_object *parent;
1686 struct lu_object *new_parent;
1687 struct echo_thread_info *info = echo_env_info(env);
1688 struct lu_name *lname = &info->eti_lname;
1689 struct md_op_spec *spec = &info->eti_spec;
1690 struct md_attr *ma = &info->eti_ma;
1691 struct lu_device *ld = ed->ed_next;
1697 if (ec_parent == NULL)
1699 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1703 rc = echo_md_dir_stripe_choose(env, ed, parent, name, namelen,
1708 LASSERT(new_parent != NULL);
1709 memset(ma, 0, sizeof(*ma));
1710 memset(spec, 0, sizeof(*spec));
1711 echo_set_lmm_size(env, ld, ma);
1712 if (stripe_count != 0) {
1713 spec->sp_cr_flags |= MDS_FMODE_WRITE;
1714 if (stripe_count != -1) {
1715 if (S_ISDIR(mode)) {
1716 struct lmv_user_md *lmu;
1718 lmu = (struct lmv_user_md *)&info->eti_lum;
1719 lmu->lum_magic = LMV_USER_MAGIC;
1720 lmu->lum_stripe_offset = stripe_offset;
1721 lmu->lum_stripe_count = stripe_count;
1722 lmu->lum_hash_type = LMV_HASH_TYPE_FNV_1A_64;
1723 spec->u.sp_ea.eadata = lmu;
1724 spec->u.sp_ea.eadatalen = sizeof(*lmu);
1726 struct lov_user_md_v3 *lum = &info->eti_lum;
1728 lum->lmm_magic = LOV_USER_MAGIC_V3;
1729 lum->lmm_stripe_count = stripe_count;
1730 lum->lmm_stripe_offset = stripe_offset;
1731 lum->lmm_pattern = LOV_PATTERN_NONE;
1732 spec->u.sp_ea.eadata = lum;
1733 spec->u.sp_ea.eadatalen = sizeof(*lum);
1735 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1739 ma->ma_attr.la_mode = mode;
1740 ma->ma_attr.la_valid = LA_CTIME | LA_MODE;
1741 ma->ma_attr.la_ctime = ktime_get_real_seconds();
1744 lname->ln_name = name;
1745 lname->ln_namelen = namelen;
1746 /* If name is specified, only create one object by name */
1747 rc = echo_md_create_internal(env, ed, lu2md(new_parent), fid,
1752 /* Create multiple object sequenced by id */
1753 for (i = 0; i < count; i++) {
1754 char *tmp_name = info->eti_name;
1756 echo_md_build_name(lname, tmp_name, id);
1758 rc = echo_md_create_internal(env, ed, lu2md(new_parent),
1759 fid, lname, spec, ma);
1761 CERROR("Can not create child %s: rc = %d\n", tmp_name,
1770 if (new_parent != parent)
1771 lu_object_put(env, new_parent);
1776 static struct lu_object *echo_md_lookup(const struct lu_env *env,
1777 struct echo_device *ed,
1778 struct md_object *parent,
1779 struct lu_name *lname)
1781 struct echo_thread_info *info = echo_env_info(env);
1782 struct lu_fid *fid = &info->eti_fid;
1783 struct lu_object *child;
1787 CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name,
1790 rc = mdo_lookup(env, parent, lname, fid, NULL);
1792 CERROR("lookup %s: rc = %d\n", lname->ln_name, rc);
1793 RETURN(ERR_PTR(rc));
1796 /* In the function below, .hs_keycmp resolves to
1797 * lu_obj_hop_keycmp() */
1798 /* coverity[overrun-buffer-val] */
1799 child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
1804 static int echo_setattr_object(const struct lu_env *env,
1805 struct echo_device *ed,
1806 struct lu_object *ec_parent,
1807 __u64 id, int count)
1809 struct lu_object *parent;
1810 struct lu_object *new_parent;
1811 struct echo_thread_info *info = echo_env_info(env);
1812 struct lu_name *lname = &info->eti_lname;
1813 char *name = info->eti_name;
1814 struct lu_device *ld = ed->ed_next;
1815 struct lu_buf *buf = &info->eti_buf;
1821 if (ec_parent == NULL)
1823 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1827 rc = echo_md_dir_stripe_choose(env, ed, parent, NULL, 0, id,
1832 for (i = 0; i < count; i++) {
1833 struct lu_object *ec_child, *child;
1835 echo_md_build_name(lname, name, id);
1837 ec_child = echo_md_lookup(env, ed, lu2md(new_parent), lname);
1838 if (IS_ERR(ec_child)) {
1839 rc = PTR_ERR(ec_child);
1840 CERROR("Can't find child %s: rc = %d\n",
1841 lname->ln_name, rc);
1845 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1846 if (child == NULL) {
1847 CERROR("Can not locate the child %s\n", lname->ln_name);
1848 lu_object_put(env, ec_child);
1853 CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n",
1854 PFID(lu_object_fid(child)));
1856 buf->lb_buf = info->eti_xattr_buf;
1857 buf->lb_len = sizeof(info->eti_xattr_buf);
1859 sprintf(name, "%s.test1", XATTR_USER_PREFIX);
1860 rc = mo_xattr_set(env, lu2md(child), buf, name,
1863 CERROR("Can not setattr child "DFID": rc = %d\n",
1864 PFID(lu_object_fid(child)), rc);
1865 lu_object_put(env, ec_child);
1868 CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n",
1869 PFID(lu_object_fid(child)));
1871 lu_object_put(env, ec_child);
1874 if (new_parent != parent)
1875 lu_object_put(env, new_parent);
1880 static int echo_getattr_object(const struct lu_env *env,
1881 struct echo_device *ed,
1882 struct lu_object *ec_parent,
1883 __u64 id, int count)
1885 struct lu_object *parent;
1886 struct lu_object *new_parent;
1887 struct echo_thread_info *info = echo_env_info(env);
1888 struct lu_name *lname = &info->eti_lname;
1889 char *name = info->eti_name;
1890 struct md_attr *ma = &info->eti_ma;
1891 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 rc = echo_md_dir_stripe_choose(env, ed, parent, NULL, 0, id,
1908 memset(ma, 0, sizeof(*ma));
1909 ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF;
1910 ma->ma_acl = info->eti_xattr_buf;
1911 ma->ma_acl_size = sizeof(info->eti_xattr_buf);
1913 for (i = 0; i < count; i++) {
1914 struct lu_object *ec_child, *child;
1917 echo_md_build_name(lname, name, id);
1918 echo_set_lmm_size(env, ld, ma);
1920 ec_child = echo_md_lookup(env, ed, lu2md(new_parent), lname);
1921 if (IS_ERR(ec_child)) {
1922 CERROR("Can't find child %s: rc = %ld\n",
1923 lname->ln_name, PTR_ERR(ec_child));
1924 RETURN(PTR_ERR(ec_child));
1927 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
1928 if (child == NULL) {
1929 CERROR("Can not locate the child %s\n", lname->ln_name);
1930 lu_object_put(env, ec_child);
1934 CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n",
1935 PFID(lu_object_fid(child)));
1936 rc = echo_attr_get_complex(env, lu2md(child), ma);
1938 CERROR("Can not getattr child "DFID": rc = %d\n",
1939 PFID(lu_object_fid(child)), rc);
1940 lu_object_put(env, ec_child);
1943 CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n",
1944 PFID(lu_object_fid(child)));
1946 lu_object_put(env, ec_child);
1949 if (new_parent != parent)
1950 lu_object_put(env, new_parent);
1955 static int echo_lookup_object(const struct lu_env *env,
1956 struct echo_device *ed,
1957 struct lu_object *ec_parent,
1958 __u64 id, int count)
1960 struct lu_object *parent;
1961 struct lu_object *new_parent;
1962 struct echo_thread_info *info = echo_env_info(env);
1963 struct lu_name *lname = &info->eti_lname;
1964 char *name = info->eti_name;
1965 struct lu_fid *fid = &info->eti_fid;
1966 struct lu_device *ld = ed->ed_next;
1970 if (ec_parent == NULL)
1972 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
1976 rc = echo_md_dir_stripe_choose(env, ed, parent, NULL, 0, id,
1981 /*prepare the requests*/
1982 for (i = 0; i < count; i++) {
1983 echo_md_build_name(lname, name, id);
1985 CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n",
1986 PFID(lu_object_fid(new_parent)), lname->ln_name,
1989 rc = mdo_lookup(env, lu2md(new_parent), lname, fid, NULL);
1991 CERROR("Can not lookup child %s: rc = %d\n", name, rc);
1995 CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n",
1996 PFID(lu_object_fid(new_parent)), lname->ln_name,
2002 if (new_parent != parent)
2003 lu_object_put(env, new_parent);
2008 static int echo_md_destroy_internal(const struct lu_env *env,
2009 struct echo_device *ed,
2010 struct md_object *parent,
2011 struct lu_name *lname,
2014 struct lu_device *ld = ed->ed_next;
2015 struct lu_object *ec_child;
2016 struct lu_object *child;
2021 ec_child = echo_md_lookup(env, ed, parent, lname);
2022 if (IS_ERR(ec_child)) {
2023 CERROR("Can't find child %s: rc = %ld\n", lname->ln_name,
2025 RETURN(PTR_ERR(ec_child));
2028 child = lu_object_locate(ec_child->lo_header, ld->ld_type);
2029 if (child == NULL) {
2030 CERROR("Can not locate the child %s\n", lname->ln_name);
2031 GOTO(out_put, rc = -EINVAL);
2034 if (lu_object_remote(child)) {
2035 CERROR("Can not destroy remote object %s: rc = %d\n",
2036 lname->ln_name, -EPERM);
2037 GOTO(out_put, rc = -EPERM);
2039 CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n",
2040 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
2042 rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0);
2044 CERROR("Can not unlink child %s: rc = %d\n",
2045 lname->ln_name, rc);
2048 CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n",
2049 PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent);
2051 lu_object_put(env, ec_child);
2055 static int echo_destroy_object(const struct lu_env *env,
2056 struct echo_device *ed,
2057 struct lu_object *ec_parent,
2058 char *name, int namelen,
2059 __u64 id, __u32 mode,
2062 struct echo_thread_info *info = echo_env_info(env);
2063 struct lu_name *lname = &info->eti_lname;
2064 struct md_attr *ma = &info->eti_ma;
2065 struct lu_device *ld = ed->ed_next;
2066 struct lu_object *parent;
2067 struct lu_object *new_parent;
2072 parent = lu_object_locate(ec_parent->lo_header, ld->ld_type);
2076 rc = echo_md_dir_stripe_choose(env, ed, parent, name, namelen,
2081 memset(ma, 0, sizeof(*ma));
2082 ma->ma_attr.la_mode = mode;
2083 ma->ma_attr.la_valid = LA_CTIME;
2084 ma->ma_attr.la_ctime = ktime_get_real_seconds();
2085 ma->ma_need = MA_INODE;
2089 lname->ln_name = name;
2090 lname->ln_namelen = namelen;
2091 rc = echo_md_destroy_internal(env, ed, lu2md(new_parent), lname,
2096 /*prepare the requests*/
2097 for (i = 0; i < count; i++) {
2098 char *tmp_name = info->eti_name;
2101 echo_md_build_name(lname, tmp_name, id);
2103 rc = echo_md_destroy_internal(env, ed, lu2md(new_parent), lname,
2106 CERROR("Can not unlink child %s: rc = %d\n", name, rc);
2113 if (new_parent != parent)
2114 lu_object_put(env, new_parent);
2119 static struct lu_object *echo_resolve_path(const struct lu_env *env,
2120 struct echo_device *ed, char *path,
2123 struct lu_device *ld = ed->ed_next;
2124 struct echo_thread_info *info = echo_env_info(env);
2125 struct lu_fid *fid = &info->eti_fid;
2126 struct lu_name *lname = &info->eti_lname;
2127 struct lu_object *parent = NULL;
2128 struct lu_object *child = NULL;
2132 *fid = ed->ed_root_fid;
2134 /* In the function below, .hs_keycmp resolves to
2135 * lu_obj_hop_keycmp() */
2136 /* coverity[overrun-buffer-val] */
2137 parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL);
2138 if (IS_ERR(parent)) {
2139 CERROR("Can not find the parent "DFID": rc = %ld\n",
2140 PFID(fid), PTR_ERR(parent));
2145 struct lu_object *ld_parent;
2148 e = strsep(&path, "/");
2153 if (!path || path[0] == '\0')
2159 lname->ln_namelen = strlen(e);
2161 ld_parent = lu_object_locate(parent->lo_header, ld->ld_type);
2162 if (ld_parent == NULL) {
2163 lu_object_put(env, parent);
2168 child = echo_md_lookup(env, ed, lu2md(ld_parent), lname);
2169 lu_object_put(env, parent);
2170 if (IS_ERR(child)) {
2171 rc = (int)PTR_ERR(child);
2172 CERROR("lookup %s under parent "DFID": rc = %d\n",
2173 lname->ln_name, PFID(lu_object_fid(ld_parent)),
2180 RETURN(ERR_PTR(rc));
2185 static void echo_ucred_init(struct lu_env *env)
2187 struct lu_ucred *ucred = lu_ucred(env);
2189 ucred->uc_valid = UCRED_INVALID;
2191 ucred->uc_suppgids[0] = -1;
2192 ucred->uc_suppgids[1] = -1;
2194 ucred->uc_uid = ucred->uc_o_uid =
2195 from_kuid(&init_user_ns, current_uid());
2196 ucred->uc_gid = ucred->uc_o_gid =
2197 from_kgid(&init_user_ns, current_gid());
2198 ucred->uc_fsuid = ucred->uc_o_fsuid =
2199 from_kuid(&init_user_ns, current_fsuid());
2200 ucred->uc_fsgid = ucred->uc_o_fsgid =
2201 from_kgid(&init_user_ns, current_fsgid());
2202 ucred->uc_cap = cfs_curproc_cap_pack();
2204 /* remove fs privilege for non-root user. */
2205 if (ucred->uc_fsuid)
2206 ucred->uc_cap &= ~CFS_CAP_FS_MASK;
2207 ucred->uc_valid = UCRED_NEW;
2210 static void echo_ucred_fini(struct lu_env *env)
2212 struct lu_ucred *ucred = lu_ucred(env);
2213 ucred->uc_valid = UCRED_INIT;
2216 #define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD)
2217 #define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION | LCT_SERVER_SESSION)
2218 static int echo_md_handler(struct echo_device *ed, int command,
2219 char *path, int path_len, __u64 id, int count,
2220 struct obd_ioctl_data *data)
2222 struct echo_thread_info *info;
2223 struct lu_device *ld = ed->ed_next;
2226 struct lu_object *parent;
2228 int namelen = data->ioc_plen2;
2233 CERROR("MD echo client is not being initialized properly\n");
2237 if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) {
2238 CERROR("Only support MDD layer right now!\n");
2242 env = cl_env_get(&refcheck);
2244 RETURN(PTR_ERR(env));
2246 rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG);
2250 /* init big_lmm buffer */
2251 info = echo_env_info(env);
2252 LASSERT(info->eti_big_lmm == NULL);
2253 OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE);
2254 if (info->eti_big_lmm == NULL)
2255 GOTO(out_env, rc = -ENOMEM);
2256 info->eti_big_lmmsize = MIN_MD_SIZE;
2258 parent = echo_resolve_path(env, ed, path, path_len);
2259 if (IS_ERR(parent)) {
2260 CERROR("Can not resolve the path %s: rc = %ld\n", path,
2262 GOTO(out_free, rc = PTR_ERR(parent));
2266 OBD_ALLOC(name, namelen + 1);
2268 GOTO(out_put, rc = -ENOMEM);
2269 if (copy_from_user(name, data->ioc_pbuf2, namelen))
2270 GOTO(out_name, rc = -EFAULT);
2273 echo_ucred_init(env);
2276 case ECHO_MD_CREATE:
2277 case ECHO_MD_MKDIR: {
2278 struct echo_thread_info *info = echo_env_info(env);
2279 __u32 mode = data->ioc_obdo2.o_mode;
2280 struct lu_fid *fid = &info->eti_fid;
2281 int stripe_count = (int)data->ioc_obdo2.o_misc;
2282 int stripe_index = (int)data->ioc_obdo2.o_stripe_idx;
2284 rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0);
2288 /* In the function below, .hs_keycmp resolves to
2289 * lu_obj_hop_keycmp() */
2290 /* coverity[overrun-buffer-val] */
2291 rc = echo_create_md_object(env, ed, parent, fid, name, namelen,
2292 id, mode, count, stripe_count,
2296 case ECHO_MD_DESTROY:
2297 case ECHO_MD_RMDIR: {
2298 __u32 mode = data->ioc_obdo2.o_mode;
2300 rc = echo_destroy_object(env, ed, parent, name, namelen,
2304 case ECHO_MD_LOOKUP:
2305 rc = echo_lookup_object(env, ed, parent, id, count);
2307 case ECHO_MD_GETATTR:
2308 rc = echo_getattr_object(env, ed, parent, id, count);
2310 case ECHO_MD_SETATTR:
2311 rc = echo_setattr_object(env, ed, parent, id, count);
2314 CERROR("unknown command %d\n", command);
2318 echo_ucred_fini(env);
2322 OBD_FREE(name, namelen + 1);
2324 lu_object_put(env, parent);
2326 LASSERT(info->eti_big_lmm);
2327 OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize);
2328 info->eti_big_lmm = NULL;
2329 info->eti_big_lmmsize = 0;
2331 cl_env_put(env, &refcheck);
2334 #endif /* HAVE_SERVER_SUPPORT */
2336 static int echo_create_object(const struct lu_env *env, struct echo_device *ed,
2339 struct echo_object *eco;
2340 struct echo_client_obd *ec = ed->ed_ec;
2345 if (!(oa->o_valid & OBD_MD_FLID) ||
2346 !(oa->o_valid & OBD_MD_FLGROUP) ||
2347 !fid_seq_is_echo(ostid_seq(&oa->o_oi))) {
2348 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2352 if (ostid_id(&oa->o_oi) == 0) {
2353 rc = ostid_set_id(&oa->o_oi, ++last_object_id);
2358 rc = obd_create(env, ec->ec_exp, oa);
2360 CERROR("Cannot create objects: rc = %d\n", rc);
2366 oa->o_valid |= OBD_MD_FLID;
2368 eco = cl_echo_object_find(ed, &oa->o_oi);
2370 GOTO(failed, rc = PTR_ERR(eco));
2371 cl_echo_object_put(eco);
2373 CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi));
2377 if (created && rc != 0)
2378 obd_destroy(env, ec->ec_exp, oa);
2381 CERROR("create object failed with: rc = %d\n", rc);
2386 static int echo_get_object(struct echo_object **ecop, struct echo_device *ed,
2389 struct echo_object *eco;
2393 if (!(oa->o_valid & OBD_MD_FLID) ||
2394 !(oa->o_valid & OBD_MD_FLGROUP) ||
2395 ostid_id(&oa->o_oi) == 0) {
2396 CERROR("invalid oid "DOSTID"\n", POSTID(&oa->o_oi));
2401 eco = cl_echo_object_find(ed, &oa->o_oi);
2410 static void echo_put_object(struct echo_object *eco)
2414 rc = cl_echo_object_put(eco);
2416 CERROR("%s: echo client drop an object failed: rc = %d\n",
2417 eco->eo_dev->ed_ec->ec_exp->exp_obd->obd_name, rc);
2420 static void echo_client_page_debug_setup(struct page *page, int rw, u64 id,
2421 u64 offset, u64 count)
2428 /* no partial pages on the client */
2429 LASSERT(count == PAGE_SIZE);
2433 for (delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2434 if (rw == OBD_BRW_WRITE) {
2435 stripe_off = offset + delta;
2438 stripe_off = 0xdeadbeef00c0ffeeULL;
2439 stripe_id = 0xdeadbeef00c0ffeeULL;
2441 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
2442 stripe_off, stripe_id);
2449 echo_client_page_debug_check(struct page *page, u64 id, u64 offset, u64 count)
2458 /* no partial pages on the client */
2459 LASSERT(count == PAGE_SIZE);
2463 for (rc = delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
2464 stripe_off = offset + delta;
2467 rc2 = block_debug_check("test_brw",
2468 addr + delta, OBD_ECHO_BLOCK_SIZE,
2469 stripe_off, stripe_id);
2471 CERROR("Error in echo object %#llx\n", id);
2480 static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa,
2481 struct echo_object *eco, u64 offset,
2482 u64 count, int async)
2485 struct brw_page *pga;
2486 struct brw_page *pgp;
2487 struct page **pages;
2496 verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID &&
2497 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
2498 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
2500 gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
2502 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
2504 if ((count & (~PAGE_MASK)) != 0)
2507 /* XXX think again with misaligned I/O */
2508 npages = count >> PAGE_SHIFT;
2510 if (rw == OBD_BRW_WRITE)
2511 brw_flags = OBD_BRW_ASYNC;
2513 OBD_ALLOC(pga, npages * sizeof(*pga));
2517 OBD_ALLOC(pages, npages * sizeof(*pages));
2518 if (pages == NULL) {
2519 OBD_FREE(pga, npages * sizeof(*pga));
2523 for (i = 0, pgp = pga, off = offset;
2525 i++, pgp++, off += PAGE_SIZE) {
2527 LASSERT(pgp->pg == NULL); /* for cleanup */
2530 pgp->pg = alloc_page(gfp_mask);
2531 if (pgp->pg == NULL)
2535 pgp->count = PAGE_SIZE;
2537 pgp->flag = brw_flags;
2540 echo_client_page_debug_setup(pgp->pg, rw,
2541 ostid_id(&oa->o_oi), off,
2545 /* brw mode can only be used at client */
2546 LASSERT(ed->ed_next != NULL);
2547 rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async);
2550 if (rc != 0 || rw != OBD_BRW_READ)
2553 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
2554 if (pgp->pg == NULL)
2559 vrc = echo_client_page_debug_check(pgp->pg,
2560 ostid_id(&oa->o_oi),
2561 pgp->off, pgp->count);
2562 if (vrc != 0 && rc == 0)
2565 __free_page(pgp->pg);
2567 OBD_FREE(pga, npages * sizeof(*pga));
2568 OBD_FREE(pages, npages * sizeof(*pages));
2572 static int echo_client_prep_commit(const struct lu_env *env,
2573 struct obd_export *exp, int rw,
2574 struct obdo *oa, struct echo_object *eco,
2575 u64 offset, u64 count,
2576 u64 batch, int async)
2578 struct obd_ioobj ioo;
2579 struct niobuf_local *lnb;
2580 struct niobuf_remote rnb;
2582 u64 npages, tot_pages, apc;
2583 int i, ret = 0, brw_flags = 0;
2587 if (count <= 0 || (count & ~PAGE_MASK) != 0)
2590 apc = npages = batch >> PAGE_SHIFT;
2591 tot_pages = count >> PAGE_SHIFT;
2593 OBD_ALLOC_LARGE(lnb, apc * sizeof(struct niobuf_local));
2597 if (rw == OBD_BRW_WRITE && async)
2598 brw_flags |= OBD_BRW_ASYNC;
2600 obdo_to_ioobj(oa, &ioo);
2604 for (; tot_pages > 0; tot_pages -= npages) {
2607 if (tot_pages < npages)
2610 rnb.rnb_offset = off;
2611 rnb.rnb_len = npages * PAGE_SIZE;
2612 rnb.rnb_flags = brw_flags;
2614 off += npages * PAGE_SIZE;
2617 ret = obd_preprw(env, rw, exp, oa, 1, &ioo, &rnb, &lpages, lnb);
2621 for (i = 0; i < lpages; i++) {
2622 struct page *page = lnb[i].lnb_page;
2624 /* read past eof? */
2625 if (page == NULL && lnb[i].lnb_rc == 0)
2629 lnb[i].lnb_flags |= OBD_BRW_ASYNC;
2631 if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID ||
2632 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
2633 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
2636 if (rw == OBD_BRW_WRITE)
2637 echo_client_page_debug_setup(page, rw,
2638 ostid_id(&oa->o_oi),
2639 lnb[i].lnb_file_offset,
2642 echo_client_page_debug_check(page,
2643 ostid_id(&oa->o_oi),
2644 lnb[i].lnb_file_offset,
2648 ret = obd_commitrw(env, rw, exp, oa, 1, &ioo, &rnb, npages, lnb,
2653 /* Reuse env context. */
2654 lu_context_exit((struct lu_context *)&env->le_ctx);
2655 lu_context_enter((struct lu_context *)&env->le_ctx);
2659 OBD_FREE_LARGE(lnb, apc * sizeof(struct niobuf_local));
2664 static int echo_client_brw_ioctl(const struct lu_env *env, int rw,
2665 struct obd_export *exp,
2666 struct obd_ioctl_data *data)
2668 struct obd_device *obd = class_exp2obd(exp);
2669 struct echo_device *ed = obd2echo_dev(obd);
2670 struct echo_client_obd *ec = ed->ed_ec;
2671 struct obdo *oa = &data->ioc_obdo1;
2672 struct echo_object *eco;
2678 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
2680 rc = echo_get_object(&eco, ed, oa);
2684 oa->o_valid &= ~OBD_MD_FLHANDLE;
2686 /* OFD/obdfilter works only via prep/commit */
2687 test_mode = (long)data->ioc_pbuf1;
2688 if (ed->ed_next == NULL && test_mode != 3) {
2690 data->ioc_plen1 = data->ioc_count;
2696 /* Truncate batch size to maximum */
2697 if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE)
2698 data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE;
2700 switch (test_mode) {
2704 rc = echo_client_kbrw(ed, rw, oa, eco, data->ioc_offset,
2705 data->ioc_count, async);
2708 rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa, eco,
2709 data->ioc_offset, data->ioc_count,
2710 data->ioc_plen1, async);
2716 echo_put_object(eco);
2722 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2723 void *karg, void __user *uarg)
2725 #ifdef HAVE_SERVER_SUPPORT
2726 struct tgt_session_info *tsi;
2728 struct obd_device *obd = exp->exp_obd;
2729 struct echo_device *ed = obd2echo_dev(obd);
2730 struct echo_client_obd *ec = ed->ed_ec;
2731 struct echo_object *eco;
2732 struct obd_ioctl_data *data = karg;
2736 int rw = OBD_BRW_READ;
2738 #ifdef HAVE_SERVER_SUPPORT
2739 struct lu_context echo_session;
2743 oa = &data->ioc_obdo1;
2744 if (!(oa->o_valid & OBD_MD_FLGROUP)) {
2745 oa->o_valid |= OBD_MD_FLGROUP;
2746 ostid_set_seq_echo(&oa->o_oi);
2749 /* This FID is unpacked just for validation at this point */
2750 rc = ostid_to_fid(&fid, &oa->o_oi, 0);
2758 rc = lu_env_init(env, LCT_DT_THREAD);
2760 GOTO(out_alloc, rc = -ENOMEM);
2762 #ifdef HAVE_SERVER_SUPPORT
2763 env->le_ses = &echo_session;
2764 rc = lu_context_init(env->le_ses, LCT_SERVER_SESSION | LCT_NOREF);
2765 if (unlikely(rc < 0))
2767 lu_context_enter(env->le_ses);
2769 tsi = tgt_ses_info(env);
2770 tsi->tsi_exp = ec->ec_exp;
2771 tsi->tsi_jobid = NULL;
2774 case OBD_IOC_CREATE: /* may create echo object */
2775 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2776 GOTO (out, rc = -EPERM);
2778 rc = echo_create_object(env, ed, oa);
2781 #ifdef HAVE_SERVER_SUPPORT
2782 case OBD_IOC_ECHO_MD: {
2789 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2790 GOTO(out, rc = -EPERM);
2792 count = data->ioc_count;
2793 cmd = data->ioc_command;
2795 id = data->ioc_obdo2.o_oi.oi.oi_id;
2796 dirlen = data->ioc_plen1;
2797 OBD_ALLOC(dir, dirlen + 1);
2799 GOTO(out, rc = -ENOMEM);
2801 if (copy_from_user(dir, data->ioc_pbuf1, dirlen)) {
2802 OBD_FREE(dir, data->ioc_plen1 + 1);
2803 GOTO(out, rc = -EFAULT);
2806 rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data);
2807 OBD_FREE(dir, dirlen + 1);
2810 case OBD_IOC_ECHO_ALLOC_SEQ: {
2811 struct lu_env *cl_env;
2816 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2817 GOTO(out, rc = -EPERM);
2819 cl_env = cl_env_get(&refcheck);
2821 GOTO(out, rc = PTR_ERR(cl_env));
2823 rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG,
2826 cl_env_put(cl_env, &refcheck);
2830 rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq);
2831 cl_env_put(cl_env, &refcheck);
2833 CERROR("%s: Can not alloc seq: rc = %d\n",
2838 if (copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1))
2841 max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH;
2842 if (copy_to_user(data->ioc_pbuf2, &max_count,
2847 #endif /* HAVE_SERVER_SUPPORT */
2848 case OBD_IOC_DESTROY:
2849 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2850 GOTO (out, rc = -EPERM);
2852 rc = echo_get_object(&eco, ed, oa);
2854 rc = obd_destroy(env, ec->ec_exp, oa);
2856 eco->eo_deleted = 1;
2857 echo_put_object(eco);
2861 case OBD_IOC_GETATTR:
2862 rc = echo_get_object(&eco, ed, oa);
2864 rc = obd_getattr(env, ec->ec_exp, oa);
2865 echo_put_object(eco);
2869 case OBD_IOC_SETATTR:
2870 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2871 GOTO (out, rc = -EPERM);
2873 rc = echo_get_object(&eco, ed, oa);
2875 rc = obd_setattr(env, ec->ec_exp, oa);
2876 echo_put_object(eco);
2880 case OBD_IOC_BRW_WRITE:
2881 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2882 GOTO (out, rc = -EPERM);
2886 case OBD_IOC_BRW_READ:
2887 rc = echo_client_brw_ioctl(env, rw, exp, data);
2891 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
2892 GOTO (out, rc = -ENOTTY);
2897 #ifdef HAVE_SERVER_SUPPORT
2898 lu_context_exit(env->le_ses);
2899 lu_context_fini(env->le_ses);
2909 static int echo_client_setup(const struct lu_env *env,
2910 struct obd_device *obddev, struct lustre_cfg *lcfg)
2912 struct echo_client_obd *ec = &obddev->u.echo_client;
2913 struct obd_device *tgt;
2914 struct obd_uuid echo_uuid = { "ECHO_UUID" };
2915 struct obd_connect_data *ocd = NULL;
2919 if (lcfg->lcfg_bufcount < 2 || LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
2920 CERROR("requires a TARGET OBD name\n");
2924 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
2925 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
2926 CERROR("device not attached or not set up (%s)\n",
2927 lustre_cfg_string(lcfg, 1));
2931 spin_lock_init(&ec->ec_lock);
2932 INIT_LIST_HEAD(&ec->ec_objects);
2933 INIT_LIST_HEAD(&ec->ec_locks);
2936 if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) {
2937 #ifdef HAVE_SERVER_SUPPORT
2938 lu_context_tags_update(ECHO_MD_CTX_TAG);
2939 lu_session_tags_update(ECHO_MD_SES_TAG);
2941 CERROR("Local operations are NOT supported on client side. "
2942 "Only remote operations are supported. Metadata client "
2943 "must be run on server side.\n");
2948 OBD_ALLOC(ocd, sizeof(*ocd));
2950 CERROR("Can't alloc ocd connecting to %s\n",
2951 lustre_cfg_string(lcfg, 1));
2955 ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL |
2956 OBD_CONNECT_BRW_SIZE |
2957 OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 |
2958 OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE |
2960 ocd->ocd_brw_size = DT_MAX_BRW_SIZE;
2961 ocd->ocd_version = LUSTRE_VERSION_CODE;
2962 ocd->ocd_group = FID_SEQ_ECHO;
2964 rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL);
2966 /* Turn off pinger because it connects to tgt obd directly. */
2967 spin_lock(&tgt->obd_dev_lock);
2968 list_del_init(&ec->ec_exp->exp_obd_chain_timed);
2969 spin_unlock(&tgt->obd_dev_lock);
2972 OBD_FREE(ocd, sizeof(*ocd));
2975 CERROR("fail to connect to device %s\n",
2976 lustre_cfg_string(lcfg, 1));
2983 static int echo_client_cleanup(struct obd_device *obddev)
2985 struct echo_device *ed = obd2echo_dev(obddev);
2986 struct echo_client_obd *ec = &obddev->u.echo_client;
2990 /*Do nothing for Metadata echo client*/
2994 if (ed->ed_next_ismd) {
2995 #ifdef HAVE_SERVER_SUPPORT
2996 lu_context_tags_clear(ECHO_MD_CTX_TAG);
2997 lu_session_tags_clear(ECHO_MD_SES_TAG);
2999 CERROR("This is client-side only module, does not support "
3000 "metadata echo client.\n");
3005 if (!list_empty(&obddev->obd_exports)) {
3006 CERROR("still has clients!\n");
3010 LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0);
3011 rc = obd_disconnect(ec->ec_exp);
3013 CERROR("fail to disconnect device: %d\n", rc);
3018 static int echo_client_connect(const struct lu_env *env,
3019 struct obd_export **exp,
3020 struct obd_device *src, struct obd_uuid *cluuid,
3021 struct obd_connect_data *data, void *localdata)
3024 struct lustre_handle conn = { 0 };
3027 rc = class_connect(&conn, src, cluuid);
3029 *exp = class_conn2export(&conn);
3035 static int echo_client_disconnect(struct obd_export *exp)
3041 GOTO(out, rc = -EINVAL);
3043 rc = class_disconnect(exp);
3049 static struct obd_ops echo_client_obd_ops = {
3050 .o_owner = THIS_MODULE,
3051 .o_iocontrol = echo_client_iocontrol,
3052 .o_connect = echo_client_connect,
3053 .o_disconnect = echo_client_disconnect
3056 static int __init obdecho_init(void)
3061 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
3063 LASSERT(PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
3065 # ifdef HAVE_SERVER_SUPPORT
3066 rc = echo_persistent_pages_init();
3070 rc = class_register_type(&echo_obd_ops, NULL, true, NULL,
3071 LUSTRE_ECHO_NAME, &echo_srv_type);
3076 rc = lu_kmem_init(echo_caches);
3078 rc = class_register_type(&echo_client_obd_ops, NULL, false,
3079 NULL, LUSTRE_ECHO_CLIENT_NAME,
3082 lu_kmem_fini(echo_caches);
3085 # ifdef HAVE_SERVER_SUPPORT
3089 class_unregister_type(LUSTRE_ECHO_NAME);
3091 echo_persistent_pages_fini();
3097 static void __exit obdecho_exit(void)
3099 class_unregister_type(LUSTRE_ECHO_CLIENT_NAME);
3100 lu_kmem_fini(echo_caches);
3102 #ifdef HAVE_SERVER_SUPPORT
3103 class_unregister_type(LUSTRE_ECHO_NAME);
3104 echo_persistent_pages_fini();
3108 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3109 MODULE_DESCRIPTION("Lustre Echo Client test driver");
3110 MODULE_VERSION(LUSTRE_VERSION_STRING);
3111 MODULE_LICENSE("GPL");
3113 module_init(obdecho_init);
3114 module_exit(obdecho_exit);
3116 /** @} echo_client */