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 return server_disconnect_export(exp);
94 static int echo_init_export(struct obd_export *exp)
96 return ldlm_init_export(exp);
99 static int echo_destroy_export(struct obd_export *exp)
103 target_destroy_export(exp);
104 ldlm_destroy_export(exp);
109 static __u64 echo_next_id(struct obd_device *obddev)
113 spin_lock(&obddev->u.echo.eo_lock);
114 id = ++obddev->u.echo.eo_lastino;
115 spin_unlock(&obddev->u.echo.eo_lock);
120 int echo_create(struct obd_export *exp, struct obdo *oa,
121 struct lov_stripe_md **ea, struct obd_trans_info *oti)
123 struct obd_device *obd = class_exp2obd(exp);
126 CERROR("invalid client cookie "LPX64"\n",
127 exp->exp_handle.h_cookie);
131 if (!(oa->o_mode && S_IFMT)) {
132 CERROR("echo obd: no type!\n");
136 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
137 CERROR("invalid o_valid "LPX64"\n", oa->o_valid);
141 oa->o_id = echo_next_id(obd);
142 oa->o_valid = OBD_MD_FLID;
147 int echo_destroy(struct obd_export *exp, struct obdo *oa,
148 struct lov_stripe_md *ea, struct obd_trans_info *oti,
149 struct obd_export *md_exp, void *capa)
151 struct obd_device *obd = class_exp2obd(exp);
155 CERROR("invalid client cookie "LPX64"\n",
156 exp->exp_handle.h_cookie);
160 if (!(oa->o_valid & OBD_MD_FLID)) {
161 CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid);
165 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OBJID) {
166 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
173 static int echo_getattr(struct obd_export *exp, struct obd_info *oinfo)
175 struct obd_device *obd = class_exp2obd(exp);
176 obd_id id = oinfo->oi_oa->o_id;
180 CERROR("invalid client cookie "LPX64"\n",
181 exp->exp_handle.h_cookie);
185 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
186 CERROR("obdo missing FLID valid flag: "LPX64"\n",
187 oinfo->oi_oa->o_valid);
191 obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid);
192 oinfo->oi_oa->o_id = id;
197 static int echo_setattr(struct obd_export *exp, struct obd_info *oinfo,
198 struct obd_trans_info *oti)
200 struct obd_device *obd = class_exp2obd(exp);
204 CERROR("invalid client cookie "LPX64"\n",
205 exp->exp_handle.h_cookie);
209 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
210 CERROR("obdo missing FLID valid flag: "LPX64"\n",
211 oinfo->oi_oa->o_valid);
215 memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa));
217 if (oinfo->oi_oa->o_id & 4) {
218 /* Save lock to force ACKed reply */
219 ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL);
220 oti->oti_ack_locks[0].mode = LCK_NL;
221 oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock;
228 echo_page_debug_setup(cfs_page_t *page, int rw, obd_id id,
229 __u64 offset, int len)
231 int page_offset = offset & ~CFS_PAGE_MASK;
232 char *addr = ((char *)cfs_kmap(page)) + page_offset;
234 if (len % OBD_ECHO_BLOCK_SIZE != 0)
235 CERROR("Unexpected block size %d\n", len);
238 if (rw & OBD_BRW_READ)
239 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
242 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
243 0xecc0ecc0ecc0ecc0ULL,
244 0xecc0ecc0ecc0ecc0ULL);
246 addr += OBD_ECHO_BLOCK_SIZE;
247 offset += OBD_ECHO_BLOCK_SIZE;
248 len -= OBD_ECHO_BLOCK_SIZE;
255 echo_page_debug_check(cfs_page_t *page, obd_id id,
256 __u64 offset, int len)
258 int page_offset = offset & ~CFS_PAGE_MASK;
259 char *addr = ((char *)cfs_kmap(page)) + page_offset;
263 if (len % OBD_ECHO_BLOCK_SIZE != 0)
264 CERROR("Unexpected block size %d\n", len);
267 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
270 if (rc2 != 0 && rc == 0)
273 addr += OBD_ECHO_BLOCK_SIZE;
274 offset += OBD_ECHO_BLOCK_SIZE;
275 len -= OBD_ECHO_BLOCK_SIZE;
283 /* This allows us to verify that desc_private is passed unmolested */
284 #define DESC_PRIV 0x10293847
286 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
287 struct niobuf_remote *nb, int *pages,
288 struct niobuf_local *lb, int cmd, int *left)
290 int gfp_mask = (obj->ioo_id & 1) ? CFS_ALLOC_HIGHUSER : CFS_ALLOC_STD;
291 int ispersistent = obj->ioo_id == ECHO_PERSISTENT_OBJID;
292 int debug_setup = (!ispersistent &&
293 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
294 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
295 struct niobuf_local *res = lb;
296 obd_off offset = nb->offset;
300 int plen = CFS_PAGE_SIZE - (offset & (CFS_PAGE_SIZE-1));
304 /* check for local buf overflow */
308 res->offset = offset;
310 LASSERT((res->offset & ~CFS_PAGE_MASK) + res->len <= CFS_PAGE_SIZE);
314 (res->offset >> CFS_PAGE_SHIFT) < ECHO_PERSISTENT_PAGES) {
315 res->page = echo_persistent_pages[res->offset >>
317 /* Take extra ref so __free_pages() can be called OK */
318 cfs_get_page (res->page);
320 OBD_PAGE_ALLOC(res->page, gfp_mask);
321 if (res->page == NULL) {
322 CERROR("can't get page for id " LPU64"\n",
328 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
329 res->page, res->offset, res->len);
331 if (cmd & OBD_BRW_READ)
335 echo_page_debug_setup(res->page, cmd, obj->ioo_id,
336 res->offset, res->len);
349 static int echo_finalize_lb(struct obdo *oa, struct obd_ioobj *obj,
350 struct niobuf_remote *rb, int *pgs,
351 struct niobuf_local *lb, int verify)
353 struct niobuf_local *res = lb;
354 obd_off start = rb->offset >> CFS_PAGE_SHIFT;
355 obd_off end = (rb->offset + rb->len + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
356 int count = (int)(end - start);
360 for (i = 0; i < count; i++, (*pgs) ++, res++) {
361 cfs_page_t *page = res->page;
365 CERROR("null page objid "LPU64":%p, buf %d/%d\n",
366 obj->ioo_id, page, i, obj->ioo_bufcnt);
370 addr = cfs_kmap(page);
372 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
373 res->page, addr, res->offset);
376 int vrc = echo_page_debug_check(page, obj->ioo_id,
377 res->offset, res->len);
378 /* check all the pages always */
379 if (vrc != 0 && rc == 0)
384 /* NB see comment above regarding persistent pages */
391 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
392 int objcount, struct obd_ioobj *obj, struct niobuf_remote *nb,
393 int *pages, struct niobuf_local *res,
394 struct obd_trans_info *oti, struct lustre_capa *unused)
396 struct obd_device *obd;
402 obd = export->exp_obd;
406 /* Temp fix to stop falling foul of osc_announce_cached() */
407 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
409 memset(res, 0, sizeof(*res) * *pages);
411 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
412 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
415 oti->oti_handle = (void *)DESC_PRIV;
420 for (i = 0; i < objcount; i++, obj++) {
423 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++) {
425 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
426 res + *pages, cmd, &left);
428 GOTO(preprw_cleanup, rc);
430 tot_bytes += nb->len;
434 atomic_add(*pages, &obd->u.echo.eo_prep);
436 if (cmd & OBD_BRW_READ)
437 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
440 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
443 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
444 atomic_read(&obd->u.echo.eo_prep));
449 /* It is possible that we would rather handle errors by allow
450 * any already-set-up pages to complete, rather than tearing them
451 * all down again. I believe that this is what the in-kernel
452 * prep/commit operations do.
454 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
455 for (i = 0; i < *pages; i++) {
456 cfs_kunmap(res[i].page);
457 /* NB if this is a persistent page, __free_pages will just
458 * lose the extra ref gained above */
459 OBD_PAGE_FREE(res[i].page);
461 atomic_dec(&obd->u.echo.eo_prep);
467 int echo_commitrw(int cmd, struct obd_export *export, struct obdo *oa,
468 int objcount, struct obd_ioobj *obj,
469 struct niobuf_remote *rb, int niocount,
470 struct niobuf_local *res, struct obd_trans_info *oti, int rc)
472 struct obd_device *obd;
477 obd = export->exp_obd;
482 GOTO(commitrw_cleanup, rc);
484 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
485 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
488 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
492 if (niocount && res == NULL) {
493 CERROR("NULL res niobuf with niocount %d\n", niocount);
497 LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV);
499 for (i = 0; i < objcount; i++, obj++) {
500 int verify = (rc == 0 &&
501 obj->ioo_id != ECHO_PERSISTENT_OBJID &&
502 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
503 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
506 for (j = 0 ; j < obj->ioo_bufcnt ; j++, rb++) {
507 int vrc = echo_finalize_lb(oa, obj, rb, &pgs, &res[pgs], verify);
513 GOTO(commitrw_cleanup, rc = vrc);
521 atomic_sub(pgs, &obd->u.echo.eo_prep);
523 CDEBUG(D_PAGE, "%d pages remain after commit\n",
524 atomic_read(&obd->u.echo.eo_prep));
528 atomic_sub(pgs, &obd->u.echo.eo_prep);
530 CERROR("cleaning up %d pages (%d obdos)\n",
531 niocount - pgs - 1, objcount);
533 while (pgs ++ < niocount) {
534 cfs_page_t *page = res[pgs].page;
539 /* NB see comment above regarding persistent pages */
541 atomic_dec(&obd->u.echo.eo_prep);
546 static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
548 struct lprocfs_static_vars lvars;
551 struct ldlm_res_id res_id = {.name = {1}};
555 spin_lock_init(&obd->u.echo.eo_lock);
556 obd->u.echo.eo_lastino = ECHO_INIT_OBJID;
558 sprintf(ns_name, "echotgt-%s", obd->obd_uuid.uuid);
559 obd->obd_namespace = ldlm_namespace_new(obd, ns_name,
560 LDLM_NAMESPACE_SERVER,
561 LDLM_NAMESPACE_MODEST);
562 if (obd->obd_namespace == NULL) {
567 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
568 NULL, LCK_NL, &lock_flags, NULL,
569 ldlm_completion_ast, NULL, NULL, 0, NULL,
570 &obd->u.echo.eo_nl_lock);
571 LASSERT (rc == ELDLM_OK);
573 lprocfs_echo_init_vars(&lvars);
574 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
575 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
576 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
577 LPROCFS_CNTR_AVGMINMAX,
578 "read_bytes", "bytes");
579 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
580 LPROCFS_CNTR_AVGMINMAX,
581 "write_bytes", "bytes");
584 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
585 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
589 static int echo_cleanup(struct obd_device *obd)
594 lprocfs_obd_cleanup(obd);
595 lprocfs_free_obd_stats(obd);
597 ldlm_lock_decref (&obd->u.echo.eo_nl_lock, LCK_NL);
599 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
600 * happened before calling ldlm_namespace_free() */
601 cfs_schedule_timeout (CFS_TASK_UNINT, cfs_time_seconds(1));
603 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
604 obd->obd_namespace = NULL;
606 leaked = atomic_read(&obd->u.echo.eo_prep);
608 CERROR("%d prep/commitrw pages leaked\n", leaked);
613 static struct obd_ops echo_obd_ops = {
614 .o_owner = THIS_MODULE,
615 .o_connect = echo_connect,
616 .o_disconnect = echo_disconnect,
617 .o_init_export = echo_init_export,
618 .o_destroy_export = echo_destroy_export,
619 .o_create = echo_create,
620 .o_destroy = echo_destroy,
621 .o_getattr = echo_getattr,
622 .o_setattr = echo_setattr,
623 .o_preprw = echo_preprw,
624 .o_commitrw = echo_commitrw,
625 .o_setup = echo_setup,
626 .o_cleanup = echo_cleanup
629 extern int echo_client_init(void);
630 extern void echo_client_exit(void);
633 echo_persistent_pages_fini (void)
637 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
638 if (echo_persistent_pages[i] != NULL) {
639 OBD_PAGE_FREE(echo_persistent_pages[i]);
640 echo_persistent_pages[i] = NULL;
645 echo_persistent_pages_init (void)
650 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
651 int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
652 CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
654 OBD_PAGE_ALLOC(pg, gfp_mask);
656 echo_persistent_pages_fini ();
660 memset (cfs_kmap (pg), 0, CFS_PAGE_SIZE);
663 echo_persistent_pages[i] = pg;
669 static int __init obdecho_init(void)
671 struct lprocfs_static_vars lvars;
675 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
677 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
679 lprocfs_echo_init_vars(&lvars);
681 rc = echo_persistent_pages_init ();
685 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
686 LUSTRE_ECHO_NAME, NULL);
690 rc = echo_client_init();
694 class_unregister_type(LUSTRE_ECHO_NAME);
696 echo_persistent_pages_fini ();
701 static void /*__exit*/ obdecho_exit(void)
704 class_unregister_type(LUSTRE_ECHO_NAME);
705 echo_persistent_pages_fini ();
708 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
709 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
710 MODULE_LICENSE("GPL");
712 cfs_module(obdecho, "1.0.0", obdecho_init, obdecho_exit);