1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdecho/echo.c
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Andreas Dilger <adilger@clusterfs.com>
43 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_ECHO
48 #include <obd_support.h>
49 #include <obd_class.h>
50 #include <lustre_debug.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
54 #include "echo_internal.h"
56 #define ECHO_INIT_OBJID 0x1000000000000000ULL
57 #define ECHO_HANDLE_MAGIC 0xabcd0123fedc9876ULL
59 #define ECHO_PERSISTENT_PAGES (ECHO_PERSISTENT_SIZE >> CFS_PAGE_SHIFT)
60 static cfs_page_t *echo_persistent_pages[ECHO_PERSISTENT_PAGES];
63 LPROC_ECHO_READ_BYTES = 1,
64 LPROC_ECHO_WRITE_BYTES = 2,
65 LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES +1
68 static int echo_connect(const struct lu_env *env,
69 struct obd_export **exp, struct obd_device *obd,
70 struct obd_uuid *cluuid, struct obd_connect_data *data,
73 struct lustre_handle conn = { 0 };
76 data->ocd_connect_flags &= ECHO_CONNECT_SUPPORTED;
77 rc = class_connect(&conn, obd, cluuid);
79 CERROR("can't connect %d\n", rc);
82 *exp = class_conn2export(&conn);
87 static int echo_disconnect(struct obd_export *exp)
89 LASSERT (exp != NULL);
91 ldlm_cancel_locks_for_export(exp);
93 /* complete all outstanding replies */
94 spin_lock(&exp->exp_lock);
95 while (!list_empty(&exp->exp_outstanding_replies)) {
96 struct ptlrpc_reply_state *rs =
97 list_entry(exp->exp_outstanding_replies.next,
98 struct ptlrpc_reply_state, rs_exp_list);
99 struct ptlrpc_service *svc = rs->rs_service;
101 spin_lock(&svc->srv_lock);
102 list_del_init(&rs->rs_exp_list);
103 spin_lock(&rs->rs_lock);
104 ptlrpc_schedule_difficult_reply(rs);
105 spin_unlock(&rs->rs_lock);
106 spin_unlock(&svc->srv_lock);
108 spin_unlock(&exp->exp_lock);
110 return class_disconnect(exp);
113 static int echo_init_export(struct obd_export *exp)
115 return ldlm_init_export(exp);
118 static int echo_destroy_export(struct obd_export *exp)
122 target_destroy_export(exp);
123 ldlm_destroy_export(exp);
128 static __u64 echo_next_id(struct obd_device *obddev)
132 spin_lock(&obddev->u.echo.eo_lock);
133 id = ++obddev->u.echo.eo_lastino;
134 spin_unlock(&obddev->u.echo.eo_lock);
139 int echo_create(struct obd_export *exp, struct obdo *oa,
140 struct lov_stripe_md **ea, struct obd_trans_info *oti)
142 struct obd_device *obd = class_exp2obd(exp);
145 CERROR("invalid client cookie "LPX64"\n",
146 exp->exp_handle.h_cookie);
150 if (!(oa->o_mode && S_IFMT)) {
151 CERROR("echo obd: no type!\n");
155 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
156 CERROR("invalid o_valid "LPX64"\n", oa->o_valid);
160 oa->o_id = echo_next_id(obd);
161 oa->o_valid = OBD_MD_FLID;
166 int echo_destroy(struct obd_export *exp, struct obdo *oa,
167 struct lov_stripe_md *ea, struct obd_trans_info *oti,
168 struct obd_export *md_exp, void *capa)
170 struct obd_device *obd = class_exp2obd(exp);
174 CERROR("invalid client cookie "LPX64"\n",
175 exp->exp_handle.h_cookie);
179 if (!(oa->o_valid & OBD_MD_FLID)) {
180 CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid);
184 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OBJID) {
185 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
192 static int echo_getattr(struct obd_export *exp, struct obd_info *oinfo)
194 struct obd_device *obd = class_exp2obd(exp);
195 obd_id id = oinfo->oi_oa->o_id;
199 CERROR("invalid client cookie "LPX64"\n",
200 exp->exp_handle.h_cookie);
204 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
205 CERROR("obdo missing FLID valid flag: "LPX64"\n",
206 oinfo->oi_oa->o_valid);
210 obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid);
211 oinfo->oi_oa->o_id = id;
216 static int echo_setattr(struct obd_export *exp, struct obd_info *oinfo,
217 struct obd_trans_info *oti)
219 struct obd_device *obd = class_exp2obd(exp);
223 CERROR("invalid client cookie "LPX64"\n",
224 exp->exp_handle.h_cookie);
228 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
229 CERROR("obdo missing FLID valid flag: "LPX64"\n",
230 oinfo->oi_oa->o_valid);
234 memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa));
236 if (oinfo->oi_oa->o_id & 4) {
237 /* Save lock to force ACKed reply */
238 ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL);
239 oti->oti_ack_locks[0].mode = LCK_NL;
240 oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock;
247 echo_page_debug_setup(cfs_page_t *page, int rw, obd_id id,
248 __u64 offset, int len)
250 int page_offset = offset & ~CFS_PAGE_MASK;
251 char *addr = ((char *)cfs_kmap(page)) + page_offset;
253 if (len % OBD_ECHO_BLOCK_SIZE != 0)
254 CERROR("Unexpected block size %d\n", len);
257 if (rw & OBD_BRW_READ)
258 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
261 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
262 0xecc0ecc0ecc0ecc0ULL,
263 0xecc0ecc0ecc0ecc0ULL);
265 addr += OBD_ECHO_BLOCK_SIZE;
266 offset += OBD_ECHO_BLOCK_SIZE;
267 len -= OBD_ECHO_BLOCK_SIZE;
274 echo_page_debug_check(cfs_page_t *page, obd_id id,
275 __u64 offset, int len)
277 int page_offset = offset & ~CFS_PAGE_MASK;
278 char *addr = ((char *)cfs_kmap(page)) + page_offset;
282 if (len % OBD_ECHO_BLOCK_SIZE != 0)
283 CERROR("Unexpected block size %d\n", len);
286 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
289 if (rc2 != 0 && rc == 0)
292 addr += OBD_ECHO_BLOCK_SIZE;
293 offset += OBD_ECHO_BLOCK_SIZE;
294 len -= OBD_ECHO_BLOCK_SIZE;
302 /* This allows us to verify that desc_private is passed unmolested */
303 #define DESC_PRIV 0x10293847
305 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
306 struct niobuf_remote *nb, int *pages,
307 struct niobuf_local *lb, int cmd, int *left)
309 int gfp_mask = (obj->ioo_id & 1) ? CFS_ALLOC_HIGHUSER : CFS_ALLOC_STD;
310 int ispersistent = obj->ioo_id == ECHO_PERSISTENT_OBJID;
311 int debug_setup = (!ispersistent &&
312 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
313 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
314 struct niobuf_local *res = lb;
315 obd_off offset = nb->offset;
319 int plen = CFS_PAGE_SIZE - (offset & (CFS_PAGE_SIZE-1));
323 /* check for local buf overflow */
327 res->offset = offset;
329 LASSERT((res->offset & ~CFS_PAGE_MASK) + res->len <= CFS_PAGE_SIZE);
333 (res->offset >> CFS_PAGE_SHIFT) < ECHO_PERSISTENT_PAGES) {
334 res->page = echo_persistent_pages[res->offset >>
336 /* Take extra ref so __free_pages() can be called OK */
337 cfs_get_page (res->page);
339 OBD_PAGE_ALLOC(res->page, gfp_mask);
340 if (res->page == NULL) {
341 CERROR("can't get page for id " LPU64"\n",
347 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
348 res->page, res->offset, res->len);
350 if (cmd & OBD_BRW_READ)
354 echo_page_debug_setup(res->page, cmd, obj->ioo_id,
355 res->offset, res->len);
368 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
369 int objcount, struct obd_ioobj *obj, struct niobuf_remote *nb,
370 int *pages, struct niobuf_local *res,
371 struct obd_trans_info *oti, struct lustre_capa *unused)
373 struct obd_device *obd;
374 struct niobuf_local *r = res;
380 obd = export->exp_obd;
384 /* Temp fix to stop falling foul of osc_announce_cached() */
385 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
387 memset(res, 0, sizeof(*res) * *pages);
389 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
390 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
393 oti->oti_handle = (void *)DESC_PRIV;
398 for (i = 0; i < objcount; i++, obj++) {
401 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++, r++) {
403 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
404 res + *pages, cmd, &left);
406 GOTO(preprw_cleanup, rc);
408 tot_bytes += nb->len;
412 atomic_add(*pages, &obd->u.echo.eo_prep);
414 if (cmd & OBD_BRW_READ)
415 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
418 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
421 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
422 atomic_read(&obd->u.echo.eo_prep));
427 /* It is possible that we would rather handle errors by allow
428 * any already-set-up pages to complete, rather than tearing them
429 * all down again. I believe that this is what the in-kernel
430 * prep/commit operations do.
432 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
433 for (i = 0; i < *pages; i++) {
434 cfs_kunmap(res[i].page);
435 /* NB if this is a persistent page, __free_pages will just
436 * lose the extra ref gained above */
437 OBD_PAGE_FREE(res[i].page);
439 atomic_dec(&obd->u.echo.eo_prep);
445 int echo_commitrw(int cmd, struct obd_export *export, struct obdo *oa,
446 int objcount, struct obd_ioobj *obj,
447 struct niobuf_remote *rb, int niocount,
448 struct niobuf_local *res, struct obd_trans_info *oti, int rc)
450 struct obd_device *obd;
451 struct niobuf_local *r = res;
455 obd = export->exp_obd;
460 GOTO(commitrw_cleanup, rc);
462 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
463 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
466 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
470 if (niocount && !r) {
471 CERROR("NULL res niobuf with niocount %d\n", niocount);
475 LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV);
477 for (i = 0; i < objcount; i++, obj++) {
478 int verify = (rc == 0 &&
479 obj->ioo_id != ECHO_PERSISTENT_OBJID &&
480 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
481 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
484 for (j = 0 ; j < obj->ioo_bufcnt ; j++, r++) {
485 cfs_page_t *page = r->page;
489 CERROR("null page objid "LPU64":%p, buf %d/%d\n",
490 obj->ioo_id, page, j, obj->ioo_bufcnt);
491 GOTO(commitrw_cleanup, rc = -EFAULT);
494 addr = cfs_kmap(page);
496 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
497 r->page, addr, r->offset);
500 vrc = echo_page_debug_check(page, obj->ioo_id,
502 /* check all the pages always */
503 if (vrc != 0 && rc == 0)
508 /* NB see comment above regarding persistent pages */
510 atomic_dec(&obd->u.echo.eo_prep);
513 CDEBUG(D_PAGE, "%d pages remain after commit\n",
514 atomic_read(&obd->u.echo.eo_prep));
518 CERROR("cleaning up %ld pages (%d obdos)\n",
519 niocount - (long)(r - res) - 1, objcount);
520 while (++r < res + niocount) {
521 cfs_page_t *page = r->page;
523 /* NB see comment above regarding persistent pages */
525 atomic_dec(&obd->u.echo.eo_prep);
530 static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
532 struct lprocfs_static_vars lvars;
535 struct ldlm_res_id res_id = {.name = {1}};
538 spin_lock_init(&obd->u.echo.eo_lock);
539 obd->u.echo.eo_lastino = ECHO_INIT_OBJID;
541 obd->obd_namespace = ldlm_namespace_new(obd, "echo-tgt",
542 LDLM_NAMESPACE_SERVER,
543 LDLM_NAMESPACE_MODEST);
544 if (obd->obd_namespace == NULL) {
549 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
550 NULL, LCK_NL, &lock_flags, NULL,
551 ldlm_completion_ast, NULL, NULL,
552 0, NULL, NULL, &obd->u.echo.eo_nl_lock);
553 LASSERT (rc == ELDLM_OK);
555 lprocfs_echo_init_vars(&lvars);
556 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
557 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
558 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
559 LPROCFS_CNTR_AVGMINMAX,
560 "read_bytes", "bytes");
561 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
562 LPROCFS_CNTR_AVGMINMAX,
563 "write_bytes", "bytes");
566 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
567 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
571 static int echo_cleanup(struct obd_device *obd)
576 lprocfs_obd_cleanup(obd);
577 lprocfs_free_obd_stats(obd);
579 ldlm_lock_decref (&obd->u.echo.eo_nl_lock, LCK_NL);
581 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
582 * happened before calling ldlm_namespace_free() */
583 cfs_schedule_timeout (CFS_TASK_UNINT, cfs_time_seconds(1));
585 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
586 obd->obd_namespace = NULL;
588 leaked = atomic_read(&obd->u.echo.eo_prep);
590 CERROR("%d prep/commitrw pages leaked\n", leaked);
595 static struct obd_ops echo_obd_ops = {
596 .o_owner = THIS_MODULE,
597 .o_connect = echo_connect,
598 .o_disconnect = echo_disconnect,
599 .o_init_export = echo_init_export,
600 .o_destroy_export = echo_destroy_export,
601 .o_create = echo_create,
602 .o_destroy = echo_destroy,
603 .o_getattr = echo_getattr,
604 .o_setattr = echo_setattr,
605 .o_preprw = echo_preprw,
606 .o_commitrw = echo_commitrw,
607 .o_setup = echo_setup,
608 .o_cleanup = echo_cleanup
611 extern int echo_client_init(void);
612 extern void echo_client_exit(void);
615 echo_persistent_pages_fini (void)
619 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
620 if (echo_persistent_pages[i] != NULL) {
621 OBD_PAGE_FREE(echo_persistent_pages[i]);
622 echo_persistent_pages[i] = NULL;
627 echo_persistent_pages_init (void)
632 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
633 int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
634 CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
636 OBD_PAGE_ALLOC(pg, gfp_mask);
638 echo_persistent_pages_fini ();
642 memset (cfs_kmap (pg), 0, CFS_PAGE_SIZE);
645 echo_persistent_pages[i] = pg;
651 static int __init obdecho_init(void)
653 struct lprocfs_static_vars lvars;
657 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
659 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
661 lprocfs_echo_init_vars(&lvars);
663 rc = echo_persistent_pages_init ();
667 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
668 LUSTRE_ECHO_NAME, NULL);
672 rc = echo_client_init();
676 class_unregister_type(LUSTRE_ECHO_NAME);
678 echo_persistent_pages_fini ();
683 static void /*__exit*/ obdecho_exit(void)
686 class_unregister_type(LUSTRE_ECHO_NAME);
687 echo_persistent_pages_fini ();
690 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
691 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
692 MODULE_LICENSE("GPL");
694 cfs_module(obdecho, "1.0.0", obdecho_init, obdecho_exit);