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 lustre_handle *conn, struct obd_device *obd,
70 struct obd_uuid *cluuid, struct obd_connect_data *data,
73 data->ocd_connect_flags &= ECHO_CONNECT_SUPPORTED;
74 return class_connect(conn, obd, cluuid);
77 static int echo_disconnect(struct obd_export *exp)
79 LASSERT (exp != NULL);
81 ldlm_cancel_locks_for_export(exp);
83 /* complete all outstanding replies */
84 spin_lock(&exp->exp_lock);
85 while (!list_empty(&exp->exp_outstanding_replies)) {
86 struct ptlrpc_reply_state *rs =
87 list_entry(exp->exp_outstanding_replies.next,
88 struct ptlrpc_reply_state, rs_exp_list);
89 struct ptlrpc_service *svc = rs->rs_service;
91 spin_lock(&svc->srv_lock);
92 list_del_init(&rs->rs_exp_list);
93 ptlrpc_schedule_difficult_reply(rs);
94 spin_unlock(&svc->srv_lock);
96 spin_unlock(&exp->exp_lock);
98 return class_disconnect(exp);
101 static int echo_init_export(struct obd_export *exp)
103 return ldlm_init_export(exp);
106 static int echo_destroy_export(struct obd_export *exp)
110 target_destroy_export(exp);
111 ldlm_destroy_export(exp);
116 static __u64 echo_next_id(struct obd_device *obddev)
120 spin_lock(&obddev->u.echo.eo_lock);
121 id = ++obddev->u.echo.eo_lastino;
122 spin_unlock(&obddev->u.echo.eo_lock);
127 int echo_create(struct obd_export *exp, struct obdo *oa,
128 struct lov_stripe_md **ea, struct obd_trans_info *oti)
130 struct obd_device *obd = class_exp2obd(exp);
133 CERROR("invalid client cookie "LPX64"\n",
134 exp->exp_handle.h_cookie);
138 if (!(oa->o_mode && S_IFMT)) {
139 CERROR("echo obd: no type!\n");
143 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
144 CERROR("invalid o_valid "LPX64"\n", oa->o_valid);
148 oa->o_id = echo_next_id(obd);
149 oa->o_valid = OBD_MD_FLID;
154 int echo_destroy(struct obd_export *exp, struct obdo *oa,
155 struct lov_stripe_md *ea, struct obd_trans_info *oti,
156 struct obd_export *md_exp)
158 struct obd_device *obd = class_exp2obd(exp);
162 CERROR("invalid client cookie "LPX64"\n",
163 exp->exp_handle.h_cookie);
167 if (!(oa->o_valid & OBD_MD_FLID)) {
168 CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid);
172 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OBJID) {
173 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
180 static int echo_getattr(struct obd_export *exp, struct obd_info *oinfo)
182 struct obd_device *obd = class_exp2obd(exp);
183 obd_id id = oinfo->oi_oa->o_id;
187 CERROR("invalid client cookie "LPX64"\n",
188 exp->exp_handle.h_cookie);
192 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
193 CERROR("obdo missing FLID valid flag: "LPX64"\n",
194 oinfo->oi_oa->o_valid);
198 obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid);
199 oinfo->oi_oa->o_id = id;
204 static int echo_setattr(struct obd_export *exp, struct obd_info *oinfo,
205 struct obd_trans_info *oti)
207 struct obd_device *obd = class_exp2obd(exp);
211 CERROR("invalid client cookie "LPX64"\n",
212 exp->exp_handle.h_cookie);
216 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
217 CERROR("obdo missing FLID valid flag: "LPX64"\n",
218 oinfo->oi_oa->o_valid);
222 memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa));
224 if (oinfo->oi_oa->o_id & 4) {
225 /* Save lock to force ACKed reply */
226 ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL);
227 oti->oti_ack_locks[0].mode = LCK_NL;
228 oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock;
235 echo_page_debug_setup(cfs_page_t *page, int rw, obd_id id,
236 __u64 offset, int len)
238 int page_offset = offset & ~CFS_PAGE_MASK;
239 char *addr = ((char *)cfs_kmap(page)) + page_offset;
241 if (len % OBD_ECHO_BLOCK_SIZE != 0)
242 CERROR("Unexpected block size %d\n", len);
245 if (rw & OBD_BRW_READ)
246 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
249 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
250 0xecc0ecc0ecc0ecc0ULL,
251 0xecc0ecc0ecc0ecc0ULL);
253 addr += OBD_ECHO_BLOCK_SIZE;
254 offset += OBD_ECHO_BLOCK_SIZE;
255 len -= OBD_ECHO_BLOCK_SIZE;
262 echo_page_debug_check(cfs_page_t *page, obd_id id,
263 __u64 offset, int len)
265 int page_offset = offset & ~CFS_PAGE_MASK;
266 char *addr = ((char *)cfs_kmap(page)) + page_offset;
270 if (len % OBD_ECHO_BLOCK_SIZE != 0)
271 CERROR("Unexpected block size %d\n", len);
274 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
277 if (rc2 != 0 && rc == 0)
280 addr += OBD_ECHO_BLOCK_SIZE;
281 offset += OBD_ECHO_BLOCK_SIZE;
282 len -= OBD_ECHO_BLOCK_SIZE;
290 /* This allows us to verify that desc_private is passed unmolested */
291 #define DESC_PRIV 0x10293847
293 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
294 struct niobuf_remote *nb, int *pages,
295 struct niobuf_local *lb, int cmd, int *left)
297 int gfp_mask = (obj->ioo_id & 1) ? CFS_ALLOC_HIGHUSER : CFS_ALLOC_STD;
298 int ispersistent = obj->ioo_id == ECHO_PERSISTENT_OBJID;
299 int debug_setup = (!ispersistent &&
300 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
301 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
302 struct niobuf_local *res = lb;
303 obd_off offset = nb->offset;
307 int plen = CFS_PAGE_SIZE - (offset & (CFS_PAGE_SIZE-1));
311 /* check for local buf overflow */
315 res->offset = offset;
317 LASSERT((res->offset & ~CFS_PAGE_MASK) + res->len <= CFS_PAGE_SIZE);
321 (res->offset >> CFS_PAGE_SHIFT) < ECHO_PERSISTENT_PAGES) {
322 res->page = echo_persistent_pages[res->offset >>
324 /* Take extra ref so __free_pages() can be called OK */
325 cfs_get_page (res->page);
327 OBD_PAGE_ALLOC(res->page, gfp_mask);
328 if (res->page == NULL) {
329 CERROR("can't get page for id " LPU64"\n",
335 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
336 res->page, res->offset, res->len);
338 if (cmd & OBD_BRW_READ)
342 echo_page_debug_setup(res->page, cmd, obj->ioo_id,
343 res->offset, res->len);
356 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
357 int objcount, struct obd_ioobj *obj, struct niobuf_remote *nb,
358 int *pages, struct niobuf_local *res,
359 struct obd_trans_info *oti, struct lustre_capa *unused)
361 struct obd_device *obd;
362 struct niobuf_local *r = res;
368 obd = export->exp_obd;
372 /* Temp fix to stop falling foul of osc_announce_cached() */
373 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
375 memset(res, 0, sizeof(*res) * *pages);
377 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
378 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
381 oti->oti_handle = (void *)DESC_PRIV;
386 for (i = 0; i < objcount; i++, obj++) {
389 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++, r++) {
391 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
392 res + *pages, cmd, &left);
394 GOTO(preprw_cleanup, rc);
396 tot_bytes += nb->len;
400 atomic_add(*pages, &obd->u.echo.eo_prep);
402 if (cmd & OBD_BRW_READ)
403 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
406 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
409 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
410 atomic_read(&obd->u.echo.eo_prep));
415 /* It is possible that we would rather handle errors by allow
416 * any already-set-up pages to complete, rather than tearing them
417 * all down again. I believe that this is what the in-kernel
418 * prep/commit operations do.
420 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
421 for (i = 0; i < *pages; i++) {
422 cfs_kunmap(res[i].page);
423 /* NB if this is a persistent page, __free_pages will just
424 * lose the extra ref gained above */
425 OBD_PAGE_FREE(res[i].page);
427 atomic_dec(&obd->u.echo.eo_prep);
433 int echo_commitrw(int cmd, struct obd_export *export, struct obdo *oa,
434 int objcount, struct obd_ioobj *obj,
435 struct niobuf_remote *rb, int niocount,
436 struct niobuf_local *res, struct obd_trans_info *oti, int rc)
438 struct obd_device *obd;
439 struct niobuf_local *r = res;
443 obd = export->exp_obd;
448 GOTO(commitrw_cleanup, rc);
450 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
451 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
454 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
458 if (niocount && !r) {
459 CERROR("NULL res niobuf with niocount %d\n", niocount);
463 LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV);
465 for (i = 0; i < objcount; i++, obj++) {
466 int verify = (rc == 0 &&
467 obj->ioo_id != ECHO_PERSISTENT_OBJID &&
468 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
469 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
472 for (j = 0 ; j < obj->ioo_bufcnt ; j++, r++) {
473 cfs_page_t *page = r->page;
477 CERROR("null page objid "LPU64":%p, buf %d/%d\n",
478 obj->ioo_id, page, j, obj->ioo_bufcnt);
479 GOTO(commitrw_cleanup, rc = -EFAULT);
482 addr = cfs_kmap(page);
484 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
485 r->page, addr, r->offset);
488 vrc = echo_page_debug_check(page, obj->ioo_id,
490 /* check all the pages always */
491 if (vrc != 0 && rc == 0)
496 /* NB see comment above regarding persistent pages */
498 atomic_dec(&obd->u.echo.eo_prep);
501 CDEBUG(D_PAGE, "%d pages remain after commit\n",
502 atomic_read(&obd->u.echo.eo_prep));
506 CERROR("cleaning up %ld pages (%d obdos)\n",
507 niocount - (long)(r - res) - 1, objcount);
508 while (++r < res + niocount) {
509 cfs_page_t *page = r->page;
511 /* NB see comment above regarding persistent pages */
513 atomic_dec(&obd->u.echo.eo_prep);
518 static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
520 struct lprocfs_static_vars lvars;
523 struct ldlm_res_id res_id = {.name = {1}};
526 spin_lock_init(&obd->u.echo.eo_lock);
527 obd->u.echo.eo_lastino = ECHO_INIT_OBJID;
529 obd->obd_namespace = ldlm_namespace_new(obd, "echo-tgt",
530 LDLM_NAMESPACE_SERVER,
531 LDLM_NAMESPACE_MODEST);
532 if (obd->obd_namespace == NULL) {
537 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
538 NULL, LCK_NL, &lock_flags, NULL,
539 ldlm_completion_ast, NULL, NULL,
540 0, NULL, NULL, &obd->u.echo.eo_nl_lock);
541 LASSERT (rc == ELDLM_OK);
543 lprocfs_echo_init_vars(&lvars);
544 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
545 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
546 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
547 LPROCFS_CNTR_AVGMINMAX,
548 "read_bytes", "bytes");
549 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
550 LPROCFS_CNTR_AVGMINMAX,
551 "write_bytes", "bytes");
554 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
555 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
559 static int echo_cleanup(struct obd_device *obd)
564 lprocfs_obd_cleanup(obd);
565 lprocfs_free_obd_stats(obd);
567 ldlm_lock_decref (&obd->u.echo.eo_nl_lock, LCK_NL);
569 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
570 * happened before calling ldlm_namespace_free() */
571 cfs_schedule_timeout (CFS_TASK_UNINT, cfs_time_seconds(1));
573 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
574 obd->obd_namespace = NULL;
576 leaked = atomic_read(&obd->u.echo.eo_prep);
578 CERROR("%d prep/commitrw pages leaked\n", leaked);
583 static struct obd_ops echo_obd_ops = {
584 .o_owner = THIS_MODULE,
585 .o_connect = echo_connect,
586 .o_disconnect = echo_disconnect,
587 .o_init_export = echo_init_export,
588 .o_destroy_export = echo_destroy_export,
589 .o_create = echo_create,
590 .o_destroy = echo_destroy,
591 .o_getattr = echo_getattr,
592 .o_setattr = echo_setattr,
593 .o_preprw = echo_preprw,
594 .o_commitrw = echo_commitrw,
595 .o_setup = echo_setup,
596 .o_cleanup = echo_cleanup
599 extern int echo_client_init(void);
600 extern void echo_client_exit(void);
603 echo_persistent_pages_fini (void)
607 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
608 if (echo_persistent_pages[i] != NULL) {
609 OBD_PAGE_FREE(echo_persistent_pages[i]);
610 echo_persistent_pages[i] = NULL;
615 echo_persistent_pages_init (void)
620 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
621 int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
622 CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
624 OBD_PAGE_ALLOC(pg, gfp_mask);
626 echo_persistent_pages_fini ();
630 memset (cfs_kmap (pg), 0, CFS_PAGE_SIZE);
633 echo_persistent_pages[i] = pg;
639 static int __init obdecho_init(void)
641 struct lprocfs_static_vars lvars;
645 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
647 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
649 lprocfs_echo_init_vars(&lvars);
651 rc = echo_persistent_pages_init ();
655 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
656 LUSTRE_ECHO_NAME, NULL);
660 rc = echo_client_init();
664 class_unregister_type(LUSTRE_ECHO_NAME);
666 echo_persistent_pages_fini ();
671 static void /*__exit*/ obdecho_exit(void)
674 class_unregister_type(LUSTRE_ECHO_NAME);
675 echo_persistent_pages_fini ();
678 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
679 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
680 MODULE_LICENSE("GPL");
682 cfs_module(obdecho, "1.0.0", obdecho_init, obdecho_exit);