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 (c) 2002, 2010, Oracle and/or its affiliates. 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 /* The echo objid needs to be below 2^32, because regular FID numbers are
57 * limited to 2^32 objects in f_oid for the FID_SEQ_ECHO range. b=23335 */
58 #define ECHO_INIT_OID 0x10000000ULL
59 #define ECHO_HANDLE_MAGIC 0xabcd0123fedc9876ULL
61 #define ECHO_PERSISTENT_PAGES (ECHO_PERSISTENT_SIZE >> CFS_PAGE_SHIFT)
62 static cfs_page_t *echo_persistent_pages[ECHO_PERSISTENT_PAGES];
65 LPROC_ECHO_READ_BYTES = 1,
66 LPROC_ECHO_WRITE_BYTES = 2,
67 LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES +1
70 static int echo_connect(const struct lu_env *env,
71 struct obd_export **exp, struct obd_device *obd,
72 struct obd_uuid *cluuid, struct obd_connect_data *data,
75 struct lustre_handle conn = { 0 };
78 data->ocd_connect_flags &= ECHO_CONNECT_SUPPORTED;
79 rc = class_connect(&conn, obd, cluuid);
81 CERROR("can't connect %d\n", rc);
84 *exp = class_conn2export(&conn);
89 static int echo_disconnect(struct obd_export *exp)
91 LASSERT (exp != NULL);
93 return server_disconnect_export(exp);
96 static int echo_init_export(struct obd_export *exp)
98 return ldlm_init_export(exp);
101 static int echo_destroy_export(struct obd_export *exp)
105 target_destroy_export(exp);
106 ldlm_destroy_export(exp);
111 static __u64 echo_next_id(struct obd_device *obddev)
115 cfs_spin_lock(&obddev->u.echo.eo_lock);
116 id = ++obddev->u.echo.eo_lastino;
117 cfs_spin_unlock(&obddev->u.echo.eo_lock);
122 int echo_create(struct obd_export *exp, struct obdo *oa,
123 struct lov_stripe_md **ea, struct obd_trans_info *oti)
125 struct obd_device *obd = class_exp2obd(exp);
128 CERROR("invalid client cookie "LPX64"\n",
129 exp->exp_handle.h_cookie);
133 if (!(oa->o_mode && S_IFMT)) {
134 CERROR("echo obd: no type!\n");
138 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
139 CERROR("invalid o_valid "LPX64"\n", oa->o_valid);
143 oa->o_id = echo_next_id(obd);
144 oa->o_valid = OBD_MD_FLID;
149 int echo_destroy(struct obd_export *exp, struct obdo *oa,
150 struct lov_stripe_md *ea, struct obd_trans_info *oti,
151 struct obd_export *md_exp, void *capa)
153 struct obd_device *obd = class_exp2obd(exp);
157 CERROR("invalid client cookie "LPX64"\n",
158 exp->exp_handle.h_cookie);
162 if (!(oa->o_valid & OBD_MD_FLID)) {
163 CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid);
167 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OID) {
168 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
175 static int echo_getattr(struct obd_export *exp, struct obd_info *oinfo)
177 struct obd_device *obd = class_exp2obd(exp);
178 obd_id id = oinfo->oi_oa->o_id;
182 CERROR("invalid client cookie "LPX64"\n",
183 exp->exp_handle.h_cookie);
187 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
188 CERROR("obdo missing FLID valid flag: "LPX64"\n",
189 oinfo->oi_oa->o_valid);
193 obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid);
194 oinfo->oi_oa->o_id = id;
199 static int echo_setattr(struct obd_export *exp, struct obd_info *oinfo,
200 struct obd_trans_info *oti)
202 struct obd_device *obd = class_exp2obd(exp);
206 CERROR("invalid client cookie "LPX64"\n",
207 exp->exp_handle.h_cookie);
211 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
212 CERROR("obdo missing FLID valid flag: "LPX64"\n",
213 oinfo->oi_oa->o_valid);
217 memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa));
219 if (oinfo->oi_oa->o_id & 4) {
220 /* Save lock to force ACKed reply */
221 ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL);
222 oti->oti_ack_locks[0].mode = LCK_NL;
223 oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock;
230 echo_page_debug_setup(cfs_page_t *page, int rw, obd_id id,
231 __u64 offset, int len)
233 int page_offset = offset & ~CFS_PAGE_MASK;
234 char *addr = ((char *)cfs_kmap(page)) + page_offset;
236 if (len % OBD_ECHO_BLOCK_SIZE != 0)
237 CERROR("Unexpected block size %d\n", len);
240 if (rw & OBD_BRW_READ)
241 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
244 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
245 0xecc0ecc0ecc0ecc0ULL,
246 0xecc0ecc0ecc0ecc0ULL);
248 addr += OBD_ECHO_BLOCK_SIZE;
249 offset += OBD_ECHO_BLOCK_SIZE;
250 len -= OBD_ECHO_BLOCK_SIZE;
257 echo_page_debug_check(cfs_page_t *page, obd_id id,
258 __u64 offset, int len)
260 int page_offset = offset & ~CFS_PAGE_MASK;
261 char *addr = ((char *)cfs_kmap(page)) + page_offset;
265 if (len % OBD_ECHO_BLOCK_SIZE != 0)
266 CERROR("Unexpected block size %d\n", len);
269 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
272 if (rc2 != 0 && rc == 0)
275 addr += OBD_ECHO_BLOCK_SIZE;
276 offset += OBD_ECHO_BLOCK_SIZE;
277 len -= OBD_ECHO_BLOCK_SIZE;
285 /* This allows us to verify that desc_private is passed unmolested */
286 #define DESC_PRIV 0x10293847
288 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
289 struct niobuf_remote *nb, int *pages,
290 struct niobuf_local *lb, int cmd, int *left)
292 int gfp_mask = (obj->ioo_id & 1) ? CFS_ALLOC_HIGHUSER : CFS_ALLOC_STD;
293 int ispersistent = obj->ioo_id == ECHO_PERSISTENT_OBJID;
294 int debug_setup = (!ispersistent &&
295 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
296 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
297 struct niobuf_local *res = lb;
298 obd_off offset = nb->offset;
302 int plen = CFS_PAGE_SIZE - (offset & (CFS_PAGE_SIZE-1));
306 /* check for local buf overflow */
310 res->offset = offset;
312 LASSERT((res->offset & ~CFS_PAGE_MASK) + res->len <= CFS_PAGE_SIZE);
316 (res->offset >> CFS_PAGE_SHIFT) < ECHO_PERSISTENT_PAGES) {
317 res->page = echo_persistent_pages[res->offset >>
319 /* Take extra ref so __free_pages() can be called OK */
320 cfs_get_page (res->page);
322 OBD_PAGE_ALLOC(res->page, gfp_mask);
323 if (res->page == NULL) {
324 CERROR("can't get page for id " LPU64"\n",
330 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
331 res->page, res->offset, res->len);
333 if (cmd & OBD_BRW_READ)
337 echo_page_debug_setup(res->page, cmd, obj->ioo_id,
338 res->offset, res->len);
351 static int echo_finalize_lb(struct obdo *oa, struct obd_ioobj *obj,
352 struct niobuf_remote *rb, int *pgs,
353 struct niobuf_local *lb, int verify)
355 struct niobuf_local *res = lb;
356 obd_off start = rb->offset >> CFS_PAGE_SHIFT;
357 obd_off end = (rb->offset + rb->len + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
358 int count = (int)(end - start);
362 for (i = 0; i < count; i++, (*pgs) ++, res++) {
363 cfs_page_t *page = res->page;
367 CERROR("null page objid "LPU64":%p, buf %d/%d\n",
368 obj->ioo_id, page, i, obj->ioo_bufcnt);
372 addr = cfs_kmap(page);
374 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
375 res->page, addr, res->offset);
378 int vrc = echo_page_debug_check(page, obj->ioo_id,
379 res->offset, res->len);
380 /* check all the pages always */
381 if (vrc != 0 && rc == 0)
386 /* NB see comment above regarding persistent pages */
393 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
394 int objcount, struct obd_ioobj *obj, struct niobuf_remote *nb,
395 int *pages, struct niobuf_local *res,
396 struct obd_trans_info *oti, struct lustre_capa *unused)
398 struct obd_device *obd;
404 obd = export->exp_obd;
408 /* Temp fix to stop falling foul of osc_announce_cached() */
409 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
411 memset(res, 0, sizeof(*res) * *pages);
413 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
414 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
417 oti->oti_handle = (void *)DESC_PRIV;
422 for (i = 0; i < objcount; i++, obj++) {
425 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++) {
427 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
428 res + *pages, cmd, &left);
430 GOTO(preprw_cleanup, rc);
432 tot_bytes += nb->len;
436 cfs_atomic_add(*pages, &obd->u.echo.eo_prep);
438 if (cmd & OBD_BRW_READ)
439 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
442 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
445 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
446 cfs_atomic_read(&obd->u.echo.eo_prep));
451 /* It is possible that we would rather handle errors by allow
452 * any already-set-up pages to complete, rather than tearing them
453 * all down again. I believe that this is what the in-kernel
454 * prep/commit operations do.
456 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
457 for (i = 0; i < *pages; i++) {
458 cfs_kunmap(res[i].page);
459 /* NB if this is a persistent page, __free_pages will just
460 * lose the extra ref gained above */
461 OBD_PAGE_FREE(res[i].page);
463 cfs_atomic_dec(&obd->u.echo.eo_prep);
469 int echo_commitrw(int cmd, struct obd_export *export, struct obdo *oa,
470 int objcount, struct obd_ioobj *obj,
471 struct niobuf_remote *rb, int niocount,
472 struct niobuf_local *res, struct obd_trans_info *oti, int rc)
474 struct obd_device *obd;
479 obd = export->exp_obd;
484 GOTO(commitrw_cleanup, rc);
486 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
487 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
490 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
494 if (niocount && res == NULL) {
495 CERROR("NULL res niobuf with niocount %d\n", niocount);
499 LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV);
501 for (i = 0; i < objcount; i++, obj++) {
502 int verify = (rc == 0 &&
503 obj->ioo_id != ECHO_PERSISTENT_OBJID &&
504 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
505 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
508 for (j = 0 ; j < obj->ioo_bufcnt ; j++, rb++) {
509 int vrc = echo_finalize_lb(oa, obj, rb, &pgs, &res[pgs], verify);
515 GOTO(commitrw_cleanup, rc = vrc);
523 cfs_atomic_sub(pgs, &obd->u.echo.eo_prep);
525 CDEBUG(D_PAGE, "%d pages remain after commit\n",
526 cfs_atomic_read(&obd->u.echo.eo_prep));
530 cfs_atomic_sub(pgs, &obd->u.echo.eo_prep);
532 CERROR("cleaning up %d pages (%d obdos)\n",
533 niocount - pgs - 1, objcount);
535 while (pgs < niocount) {
536 cfs_page_t *page = res[pgs++].page;
541 /* NB see comment above regarding persistent pages */
543 cfs_atomic_dec(&obd->u.echo.eo_prep);
548 static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
550 struct lprocfs_static_vars lvars;
553 struct ldlm_res_id res_id = {.name = {1}};
557 cfs_spin_lock_init(&obd->u.echo.eo_lock);
558 obd->u.echo.eo_lastino = ECHO_INIT_OID;
560 sprintf(ns_name, "echotgt-%s", obd->obd_uuid.uuid);
561 obd->obd_namespace = ldlm_namespace_new(obd, ns_name,
562 LDLM_NAMESPACE_SERVER,
563 LDLM_NAMESPACE_MODEST);
564 if (obd->obd_namespace == NULL) {
569 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
570 NULL, LCK_NL, &lock_flags, NULL,
571 ldlm_completion_ast, NULL, NULL, 0, NULL,
572 &obd->u.echo.eo_nl_lock);
573 LASSERT (rc == ELDLM_OK);
575 lprocfs_echo_init_vars(&lvars);
576 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
577 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
578 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
579 LPROCFS_CNTR_AVGMINMAX,
580 "read_bytes", "bytes");
581 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
582 LPROCFS_CNTR_AVGMINMAX,
583 "write_bytes", "bytes");
586 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
587 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
591 static int echo_cleanup(struct obd_device *obd)
596 lprocfs_obd_cleanup(obd);
597 lprocfs_free_obd_stats(obd);
599 ldlm_lock_decref(&obd->u.echo.eo_nl_lock, LCK_NL);
601 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
602 * happened before calling ldlm_namespace_free() */
603 cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT, cfs_time_seconds(1));
605 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
606 obd->obd_namespace = NULL;
608 leaked = cfs_atomic_read(&obd->u.echo.eo_prep);
610 CERROR("%d prep/commitrw pages leaked\n", leaked);
615 static struct obd_ops echo_obd_ops = {
616 .o_owner = THIS_MODULE,
617 .o_connect = echo_connect,
618 .o_disconnect = echo_disconnect,
619 .o_init_export = echo_init_export,
620 .o_destroy_export = echo_destroy_export,
621 .o_create = echo_create,
622 .o_destroy = echo_destroy,
623 .o_getattr = echo_getattr,
624 .o_setattr = echo_setattr,
625 .o_preprw = echo_preprw,
626 .o_commitrw = echo_commitrw,
627 .o_setup = echo_setup,
628 .o_cleanup = echo_cleanup
631 extern int echo_client_init(void);
632 extern void echo_client_exit(void);
635 echo_persistent_pages_fini (void)
639 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
640 if (echo_persistent_pages[i] != NULL) {
641 OBD_PAGE_FREE(echo_persistent_pages[i]);
642 echo_persistent_pages[i] = NULL;
647 echo_persistent_pages_init (void)
652 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
653 int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
654 CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
656 OBD_PAGE_ALLOC(pg, gfp_mask);
658 echo_persistent_pages_fini ();
662 memset (cfs_kmap (pg), 0, CFS_PAGE_SIZE);
665 echo_persistent_pages[i] = pg;
671 static int __init obdecho_init(void)
673 struct lprocfs_static_vars lvars;
677 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
679 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
681 lprocfs_echo_init_vars(&lvars);
683 rc = echo_persistent_pages_init ();
687 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
688 LUSTRE_ECHO_NAME, NULL);
692 rc = echo_client_init();
696 class_unregister_type(LUSTRE_ECHO_NAME);
698 echo_persistent_pages_fini ();
703 static void /*__exit*/ obdecho_exit(void)
706 class_unregister_type(LUSTRE_ECHO_NAME);
707 echo_persistent_pages_fini ();
710 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
711 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
712 MODULE_LICENSE("GPL");
714 cfs_module(obdecho, "1.0.0", obdecho_init, obdecho_exit);