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 spin_lock(&rs->rs_lock);
94 ptlrpc_schedule_difficult_reply(rs);
95 spin_unlock(&rs->rs_lock);
96 spin_unlock(&svc->srv_lock);
98 spin_unlock(&exp->exp_lock);
100 return class_disconnect(exp);
103 static int echo_init_export(struct obd_export *exp)
105 return ldlm_init_export(exp);
108 static int echo_destroy_export(struct obd_export *exp)
112 target_destroy_export(exp);
113 ldlm_destroy_export(exp);
118 static __u64 echo_next_id(struct obd_device *obddev)
122 spin_lock(&obddev->u.echo.eo_lock);
123 id = ++obddev->u.echo.eo_lastino;
124 spin_unlock(&obddev->u.echo.eo_lock);
129 int echo_create(struct obd_export *exp, struct obdo *oa,
130 struct lov_stripe_md **ea, struct obd_trans_info *oti)
132 struct obd_device *obd = class_exp2obd(exp);
135 CERROR("invalid client cookie "LPX64"\n",
136 exp->exp_handle.h_cookie);
140 if (!(oa->o_mode && S_IFMT)) {
141 CERROR("echo obd: no type!\n");
145 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
146 CERROR("invalid o_valid "LPX64"\n", oa->o_valid);
150 oa->o_id = echo_next_id(obd);
151 oa->o_valid = OBD_MD_FLID;
156 int echo_destroy(struct obd_export *exp, struct obdo *oa,
157 struct lov_stripe_md *ea, struct obd_trans_info *oti,
158 struct obd_export *md_exp, void *capa)
160 struct obd_device *obd = class_exp2obd(exp);
164 CERROR("invalid client cookie "LPX64"\n",
165 exp->exp_handle.h_cookie);
169 if (!(oa->o_valid & OBD_MD_FLID)) {
170 CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid);
174 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OBJID) {
175 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
182 static int echo_getattr(struct obd_export *exp, struct obd_info *oinfo)
184 struct obd_device *obd = class_exp2obd(exp);
185 obd_id id = oinfo->oi_oa->o_id;
189 CERROR("invalid client cookie "LPX64"\n",
190 exp->exp_handle.h_cookie);
194 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
195 CERROR("obdo missing FLID valid flag: "LPX64"\n",
196 oinfo->oi_oa->o_valid);
200 obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid);
201 oinfo->oi_oa->o_id = id;
206 static int echo_setattr(struct obd_export *exp, struct obd_info *oinfo,
207 struct obd_trans_info *oti)
209 struct obd_device *obd = class_exp2obd(exp);
213 CERROR("invalid client cookie "LPX64"\n",
214 exp->exp_handle.h_cookie);
218 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
219 CERROR("obdo missing FLID valid flag: "LPX64"\n",
220 oinfo->oi_oa->o_valid);
224 memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa));
226 if (oinfo->oi_oa->o_id & 4) {
227 /* Save lock to force ACKed reply */
228 ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL);
229 oti->oti_ack_locks[0].mode = LCK_NL;
230 oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock;
237 echo_page_debug_setup(cfs_page_t *page, int rw, obd_id id,
238 __u64 offset, int len)
240 int page_offset = offset & ~CFS_PAGE_MASK;
241 char *addr = ((char *)cfs_kmap(page)) + page_offset;
243 if (len % OBD_ECHO_BLOCK_SIZE != 0)
244 CERROR("Unexpected block size %d\n", len);
247 if (rw & OBD_BRW_READ)
248 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
251 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
252 0xecc0ecc0ecc0ecc0ULL,
253 0xecc0ecc0ecc0ecc0ULL);
255 addr += OBD_ECHO_BLOCK_SIZE;
256 offset += OBD_ECHO_BLOCK_SIZE;
257 len -= OBD_ECHO_BLOCK_SIZE;
264 echo_page_debug_check(cfs_page_t *page, obd_id id,
265 __u64 offset, int len)
267 int page_offset = offset & ~CFS_PAGE_MASK;
268 char *addr = ((char *)cfs_kmap(page)) + page_offset;
272 if (len % OBD_ECHO_BLOCK_SIZE != 0)
273 CERROR("Unexpected block size %d\n", len);
276 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
279 if (rc2 != 0 && rc == 0)
282 addr += OBD_ECHO_BLOCK_SIZE;
283 offset += OBD_ECHO_BLOCK_SIZE;
284 len -= OBD_ECHO_BLOCK_SIZE;
292 /* This allows us to verify that desc_private is passed unmolested */
293 #define DESC_PRIV 0x10293847
295 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
296 struct niobuf_remote *nb, int *pages,
297 struct niobuf_local *lb, int cmd, int *left)
299 int gfp_mask = (obj->ioo_id & 1) ? CFS_ALLOC_HIGHUSER : CFS_ALLOC_STD;
300 int ispersistent = obj->ioo_id == ECHO_PERSISTENT_OBJID;
301 int debug_setup = (!ispersistent &&
302 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
303 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
304 struct niobuf_local *res = lb;
305 obd_off offset = nb->offset;
309 int plen = CFS_PAGE_SIZE - (offset & (CFS_PAGE_SIZE-1));
313 /* check for local buf overflow */
317 res->offset = offset;
319 LASSERT((res->offset & ~CFS_PAGE_MASK) + res->len <= CFS_PAGE_SIZE);
323 (res->offset >> CFS_PAGE_SHIFT) < ECHO_PERSISTENT_PAGES) {
324 res->page = echo_persistent_pages[res->offset >>
326 /* Take extra ref so __free_pages() can be called OK */
327 cfs_get_page (res->page);
329 OBD_PAGE_ALLOC(res->page, gfp_mask);
330 if (res->page == NULL) {
331 CERROR("can't get page for id " LPU64"\n",
337 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
338 res->page, res->offset, res->len);
340 if (cmd & OBD_BRW_READ)
344 echo_page_debug_setup(res->page, cmd, obj->ioo_id,
345 res->offset, res->len);
358 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
359 int objcount, struct obd_ioobj *obj, struct niobuf_remote *nb,
360 int *pages, struct niobuf_local *res,
361 struct obd_trans_info *oti, struct lustre_capa *unused)
363 struct obd_device *obd;
364 struct niobuf_local *r = res;
370 obd = export->exp_obd;
374 /* Temp fix to stop falling foul of osc_announce_cached() */
375 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
377 memset(res, 0, sizeof(*res) * *pages);
379 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
380 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
383 oti->oti_handle = (void *)DESC_PRIV;
388 for (i = 0; i < objcount; i++, obj++) {
391 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++, r++) {
393 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
394 res + *pages, cmd, &left);
396 GOTO(preprw_cleanup, rc);
398 tot_bytes += nb->len;
402 atomic_add(*pages, &obd->u.echo.eo_prep);
404 if (cmd & OBD_BRW_READ)
405 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
408 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
411 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
412 atomic_read(&obd->u.echo.eo_prep));
417 /* It is possible that we would rather handle errors by allow
418 * any already-set-up pages to complete, rather than tearing them
419 * all down again. I believe that this is what the in-kernel
420 * prep/commit operations do.
422 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
423 for (i = 0; i < *pages; i++) {
424 cfs_kunmap(res[i].page);
425 /* NB if this is a persistent page, __free_pages will just
426 * lose the extra ref gained above */
427 OBD_PAGE_FREE(res[i].page);
429 atomic_dec(&obd->u.echo.eo_prep);
435 int echo_commitrw(int cmd, struct obd_export *export, struct obdo *oa,
436 int objcount, struct obd_ioobj *obj,
437 struct niobuf_remote *rb, int niocount,
438 struct niobuf_local *res, struct obd_trans_info *oti, int rc)
440 struct obd_device *obd;
441 struct niobuf_local *r = res;
445 obd = export->exp_obd;
450 GOTO(commitrw_cleanup, rc);
452 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
453 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
456 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
460 if (niocount && !r) {
461 CERROR("NULL res niobuf with niocount %d\n", niocount);
465 LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV);
467 for (i = 0; i < objcount; i++, obj++) {
468 int verify = (rc == 0 &&
469 obj->ioo_id != ECHO_PERSISTENT_OBJID &&
470 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
471 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
474 for (j = 0 ; j < obj->ioo_bufcnt ; j++, r++) {
475 cfs_page_t *page = r->page;
479 CERROR("null page objid "LPU64":%p, buf %d/%d\n",
480 obj->ioo_id, page, j, obj->ioo_bufcnt);
481 GOTO(commitrw_cleanup, rc = -EFAULT);
484 addr = cfs_kmap(page);
486 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
487 r->page, addr, r->offset);
490 vrc = echo_page_debug_check(page, obj->ioo_id,
492 /* check all the pages always */
493 if (vrc != 0 && rc == 0)
498 /* NB see comment above regarding persistent pages */
500 atomic_dec(&obd->u.echo.eo_prep);
503 CDEBUG(D_PAGE, "%d pages remain after commit\n",
504 atomic_read(&obd->u.echo.eo_prep));
508 CERROR("cleaning up %ld pages (%d obdos)\n",
509 niocount - (long)(r - res) - 1, objcount);
510 while (++r < res + niocount) {
511 cfs_page_t *page = r->page;
513 /* NB see comment above regarding persistent pages */
515 atomic_dec(&obd->u.echo.eo_prep);
520 static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
522 struct lprocfs_static_vars lvars;
525 struct ldlm_res_id res_id = {.name = {1}};
528 spin_lock_init(&obd->u.echo.eo_lock);
529 obd->u.echo.eo_lastino = ECHO_INIT_OBJID;
531 obd->obd_namespace = ldlm_namespace_new(obd, "echo-tgt",
532 LDLM_NAMESPACE_SERVER,
533 LDLM_NAMESPACE_MODEST);
534 if (obd->obd_namespace == NULL) {
539 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
540 NULL, LCK_NL, &lock_flags, NULL,
541 ldlm_completion_ast, NULL, NULL,
542 0, NULL, NULL, &obd->u.echo.eo_nl_lock);
543 LASSERT (rc == ELDLM_OK);
545 lprocfs_echo_init_vars(&lvars);
546 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
547 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
548 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
549 LPROCFS_CNTR_AVGMINMAX,
550 "read_bytes", "bytes");
551 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
552 LPROCFS_CNTR_AVGMINMAX,
553 "write_bytes", "bytes");
556 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
557 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
561 static int echo_cleanup(struct obd_device *obd)
566 lprocfs_obd_cleanup(obd);
567 lprocfs_free_obd_stats(obd);
569 ldlm_lock_decref (&obd->u.echo.eo_nl_lock, LCK_NL);
571 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
572 * happened before calling ldlm_namespace_free() */
573 cfs_schedule_timeout (CFS_TASK_UNINT, cfs_time_seconds(1));
575 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
576 obd->obd_namespace = NULL;
578 leaked = atomic_read(&obd->u.echo.eo_prep);
580 CERROR("%d prep/commitrw pages leaked\n", leaked);
585 static struct obd_ops echo_obd_ops = {
586 .o_owner = THIS_MODULE,
587 .o_connect = echo_connect,
588 .o_disconnect = echo_disconnect,
589 .o_init_export = echo_init_export,
590 .o_destroy_export = echo_destroy_export,
591 .o_create = echo_create,
592 .o_destroy = echo_destroy,
593 .o_getattr = echo_getattr,
594 .o_setattr = echo_setattr,
595 .o_preprw = echo_preprw,
596 .o_commitrw = echo_commitrw,
597 .o_setup = echo_setup,
598 .o_cleanup = echo_cleanup
601 extern int echo_client_init(void);
602 extern void echo_client_exit(void);
605 echo_persistent_pages_fini (void)
609 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
610 if (echo_persistent_pages[i] != NULL) {
611 OBD_PAGE_FREE(echo_persistent_pages[i]);
612 echo_persistent_pages[i] = NULL;
617 echo_persistent_pages_init (void)
622 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
623 int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
624 CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
626 OBD_PAGE_ALLOC(pg, gfp_mask);
628 echo_persistent_pages_fini ();
632 memset (cfs_kmap (pg), 0, CFS_PAGE_SIZE);
635 echo_persistent_pages[i] = pg;
641 static int __init obdecho_init(void)
643 struct lprocfs_static_vars lvars;
647 LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n");
649 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
651 lprocfs_echo_init_vars(&lvars);
653 rc = echo_persistent_pages_init ();
657 rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
658 LUSTRE_ECHO_NAME, NULL);
662 rc = echo_client_init();
666 class_unregister_type(LUSTRE_ECHO_NAME);
668 echo_persistent_pages_fini ();
673 static void /*__exit*/ obdecho_exit(void)
676 class_unregister_type(LUSTRE_ECHO_NAME);
677 echo_persistent_pages_fini ();
680 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
681 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
682 MODULE_LICENSE("GPL");
684 cfs_module(obdecho, "1.0.0", obdecho_init, obdecho_exit);