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>
51 #include <lustre_debug.h>
52 #include <lustre_dlm.h>
53 #include <lprocfs_status.h>
55 /* The echo objid needs to be below 2^32, because regular FID numbers are
56 * limited to 2^32 objects in f_oid for the FID_SEQ_ECHO range. b=23335 */
57 #define ECHO_INIT_OID 0x10000000ULL
58 #define ECHO_HANDLE_MAGIC 0xabcd0123fedc9876ULL
60 #define ECHO_PERSISTENT_PAGES (ECHO_PERSISTENT_SIZE >> CFS_PAGE_SHIFT)
61 static cfs_page_t *echo_persistent_pages[ECHO_PERSISTENT_PAGES];
64 LPROC_ECHO_READ_BYTES = 1,
65 LPROC_ECHO_WRITE_BYTES = 2,
66 LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES +1
69 static int echo_connect(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 return server_disconnect_export(exp);
84 static int echo_init_export(struct obd_export *exp)
86 return ldlm_init_export(exp);
89 static int echo_destroy_export(struct obd_export *exp)
93 target_destroy_export(exp);
94 ldlm_destroy_export(exp);
99 static __u64 echo_next_id(struct obd_device *obddev)
103 spin_lock(&obddev->u.echo.eo_lock);
104 id = ++obddev->u.echo.eo_lastino;
105 spin_unlock(&obddev->u.echo.eo_lock);
110 int echo_create(struct obd_export *exp, struct obdo *oa,
111 struct lov_stripe_md **ea, struct obd_trans_info *oti)
113 struct obd_device *obd = class_exp2obd(exp);
116 CERROR("invalid client cookie "LPX64"\n",
117 exp->exp_handle.h_cookie);
121 if (!(oa->o_mode && S_IFMT)) {
122 CERROR("echo obd: no type!\n");
126 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
127 CERROR("invalid o_valid "LPX64"\n", oa->o_valid);
131 oa->o_id = echo_next_id(obd);
132 oa->o_valid = OBD_MD_FLID;
137 int echo_destroy(struct obd_export *exp, struct obdo *oa,
138 struct lov_stripe_md *ea, struct obd_trans_info *oti,
139 struct obd_export *md_exp)
141 struct obd_device *obd = class_exp2obd(exp);
145 CERROR("invalid client cookie "LPX64"\n",
146 exp->exp_handle.h_cookie);
150 if (!(oa->o_valid & OBD_MD_FLID)) {
151 CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid);
155 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OID) {
156 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
163 static int echo_getattr(struct obd_export *exp, struct obd_info *oinfo)
165 struct obd_device *obd = class_exp2obd(exp);
166 obd_id id = oinfo->oi_oa->o_id;
170 CERROR("invalid client cookie "LPX64"\n",
171 exp->exp_handle.h_cookie);
175 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
176 CERROR("obdo missing FLID valid flag: "LPX64"\n",
177 oinfo->oi_oa->o_valid);
181 obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid);
182 oinfo->oi_oa->o_id = id;
187 static int echo_setattr(struct obd_export *exp, struct obd_info *oinfo,
188 struct obd_trans_info *oti)
190 struct obd_device *obd = class_exp2obd(exp);
194 CERROR("invalid client cookie "LPX64"\n",
195 exp->exp_handle.h_cookie);
199 if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) {
200 CERROR("obdo missing FLID valid flag: "LPX64"\n",
201 oinfo->oi_oa->o_valid);
205 memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa));
207 if (oinfo->oi_oa->o_id & 4) {
208 /* Save lock to force ACKed reply */
209 ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL);
210 oti->oti_ack_locks[0].mode = LCK_NL;
211 oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock;
218 echo_page_debug_setup(cfs_page_t *page, int rw, obd_id id,
219 __u64 offset, int len)
221 int page_offset = offset & ~CFS_PAGE_MASK;
222 char *addr = ((char *)cfs_kmap(page)) + page_offset;
224 if (len % OBD_ECHO_BLOCK_SIZE != 0)
225 CERROR("Unexpected block size %d\n", len);
228 if (rw & OBD_BRW_READ)
229 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
232 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
233 0xecc0ecc0ecc0ecc0ULL,
234 0xecc0ecc0ecc0ecc0ULL);
236 addr += OBD_ECHO_BLOCK_SIZE;
237 offset += OBD_ECHO_BLOCK_SIZE;
238 len -= OBD_ECHO_BLOCK_SIZE;
245 echo_page_debug_check(cfs_page_t *page, obd_id id,
246 __u64 offset, int len)
248 int page_offset = offset & ~CFS_PAGE_MASK;
249 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 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
260 if (rc2 != 0 && rc == 0)
263 addr += OBD_ECHO_BLOCK_SIZE;
264 offset += OBD_ECHO_BLOCK_SIZE;
265 len -= OBD_ECHO_BLOCK_SIZE;
273 /* This allows us to verify that desc_private is passed unmolested */
274 #define DESC_PRIV 0x10293847
276 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
277 struct niobuf_remote *nb, int *pages,
278 struct niobuf_local *lb, int cmd, int *left)
280 int gfp_mask = (obj->ioo_id & 1) ? CFS_ALLOC_HIGHUSER : CFS_ALLOC_STD;
281 int ispersistent = obj->ioo_id == ECHO_PERSISTENT_OBJID;
282 int debug_setup = (!ispersistent &&
283 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
284 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
285 struct niobuf_local *res = lb;
286 obd_off offset = nb->offset;
290 int plen = CFS_PAGE_SIZE - (offset & (CFS_PAGE_SIZE-1));
294 /* check for local buf overflow */
298 res->offset = offset;
300 LASSERT((res->offset & ~CFS_PAGE_MASK) + res->len <= CFS_PAGE_SIZE);
304 (res->offset >> CFS_PAGE_SHIFT) < ECHO_PERSISTENT_PAGES) {
305 res->page = echo_persistent_pages[res->offset >>
307 /* Take extra ref so __free_pages() can be called OK */
308 cfs_get_page (res->page);
310 OBD_PAGE_ALLOC(res->page, gfp_mask);
311 if (res->page == NULL) {
312 CERROR("can't get page for id " LPU64"\n",
318 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
319 res->page, res->offset, res->len);
321 if (cmd & OBD_BRW_READ)
325 echo_page_debug_setup(res->page, cmd, obj->ioo_id,
326 res->offset, res->len);
339 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
340 int objcount, struct obd_ioobj *obj, struct niobuf_remote *nb,
341 int *pages, struct niobuf_local *res,
342 struct obd_trans_info *oti)
344 struct obd_device *obd;
345 struct niobuf_local *r = res;
351 obd = export->exp_obd;
355 /* Temp fix to stop falling foul of osc_announce_cached() */
356 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
358 memset(res, 0, sizeof(*res) * *pages);
360 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
361 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
364 oti->oti_handle = (void *)DESC_PRIV;
369 for (i = 0; i < objcount; i++, obj++) {
372 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++, r++) {
374 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
375 res + *pages, cmd, &left);
377 GOTO(preprw_cleanup, rc);
379 tot_bytes += nb->len;
383 atomic_add(*pages, &obd->u.echo.eo_prep);
385 if (cmd & OBD_BRW_READ)
386 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
389 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
392 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
393 atomic_read(&obd->u.echo.eo_prep));
398 /* It is possible that we would rather handle errors by allow
399 * any already-set-up pages to complete, rather than tearing them
400 * all down again. I believe that this is what the in-kernel
401 * prep/commit operations do.
403 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
404 for (i = 0; i < *pages; i++) {
405 cfs_kunmap(res[i].page);
406 /* NB if this is a persistent page, __free_pages will just
407 * lose the extra ref gained above */
408 OBD_PAGE_FREE(res[i].page);
410 atomic_dec(&obd->u.echo.eo_prep);
416 int echo_commitrw(int cmd, struct obd_export *export, struct obdo *oa,
417 int objcount, struct obd_ioobj *obj,
418 struct niobuf_remote *rb, int niocount,
419 struct niobuf_local *res, struct obd_trans_info *oti, int rc)
421 struct obd_device *obd;
422 struct niobuf_local *r = res;
423 struct niobuf_remote *rbc = rb;
427 obd = export->exp_obd;
432 GOTO(commitrw_cleanup, rc);
434 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
435 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
438 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
442 if (niocount && !r) {
443 CERROR("NULL res niobuf with niocount %d\n", niocount);
447 LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV);
449 for (i = 0; i < objcount; i++, obj++) {
450 int verify = (rc == 0 &&
451 obj->ioo_id != ECHO_PERSISTENT_OBJID &&
452 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
453 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
456 for (j = 0 ; j < obj->ioo_bufcnt ; j++, rbc++) {
458 for (l = rbc->len; l>0; l-=r->len, r++) {
459 cfs_page_t *page = r->page;
463 CERROR("null page objid "LPU64":%p, buf %d/%d\n",
464 obj->ioo_id, page, j, obj->ioo_bufcnt);
465 GOTO(commitrw_cleanup, rc = -EFAULT);
468 addr = cfs_kmap(page);
470 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
471 r->page, addr, r->offset);
474 vrc = echo_page_debug_check(page, obj->ioo_id,
476 /* check all the pages always */
477 if (vrc != 0 && rc == 0)
482 /* NB see comment above regarding persistent pages */
484 atomic_dec(&obd->u.echo.eo_prep);
488 CDEBUG(D_PAGE, "%d pages remain after commit\n",
489 atomic_read(&obd->u.echo.eo_prep));
493 CERROR("cleaning up %ld pages (%d obdos)\n",
494 niocount - (long)(r - res) - 1, objcount);
495 while (++r < res + niocount) {
496 cfs_page_t *page = r->page;
498 /* NB see comment above regarding persistent pages */
500 atomic_dec(&obd->u.echo.eo_prep);
505 static int echo_setup(struct obd_device *obd, obd_count len, void *buf)
507 struct lprocfs_static_vars lvars;
510 struct ldlm_res_id res_id = {.name = {1}};
514 spin_lock_init(&obd->u.echo.eo_lock);
515 obd->u.echo.eo_lastino = ECHO_INIT_OID;
517 sprintf(ns_name, "echotgt-%s", obd->obd_uuid.uuid);
518 obd->obd_namespace = ldlm_namespace_new(obd, ns_name,
519 LDLM_NAMESPACE_SERVER,
520 LDLM_NAMESPACE_GREEDY);
521 if (obd->obd_namespace == NULL) {
526 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
527 NULL, LCK_NL, &lock_flags, NULL,
528 ldlm_completion_ast, NULL, NULL,
529 0, NULL, &obd->u.echo.eo_nl_lock);
530 LASSERT (rc == ELDLM_OK);
532 lprocfs_echo_init_vars(&lvars);
533 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0 &&
534 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
535 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
536 LPROCFS_CNTR_AVGMINMAX,
537 "read_bytes", "bytes");
538 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
539 LPROCFS_CNTR_AVGMINMAX,
540 "write_bytes", "bytes");
543 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
544 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
548 static int echo_cleanup(struct obd_device *obd)
553 lprocfs_obd_cleanup(obd);
554 lprocfs_free_obd_stats(obd);
556 ldlm_lock_decref (&obd->u.echo.eo_nl_lock, LCK_NL);
558 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
559 * happened before calling ldlm_namespace_free() */
560 cfs_schedule_timeout (CFS_TASK_UNINT, cfs_time_seconds(1));
562 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
563 obd->obd_namespace = NULL;
565 leaked = atomic_read(&obd->u.echo.eo_prep);
567 CERROR("%d prep/commitrw pages leaked\n", leaked);
572 static struct obd_ops echo_obd_ops = {
573 .o_owner = THIS_MODULE,
574 .o_connect = echo_connect,
575 .o_disconnect = echo_disconnect,
576 .o_init_export = echo_init_export,
577 .o_destroy_export = echo_destroy_export,
578 .o_create = echo_create,
579 .o_destroy = echo_destroy,
580 .o_getattr = echo_getattr,
581 .o_setattr = echo_setattr,
582 .o_preprw = echo_preprw,
583 .o_commitrw = echo_commitrw,
584 .o_setup = echo_setup,
585 .o_cleanup = echo_cleanup
588 extern int echo_client_init(void);
589 extern void echo_client_exit(void);
592 echo_persistent_pages_fini (void)
596 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
597 if (echo_persistent_pages[i] != NULL) {
598 OBD_PAGE_FREE(echo_persistent_pages[i]);
599 echo_persistent_pages[i] = NULL;
604 echo_persistent_pages_init (void)
609 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
610 int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
611 CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER;
613 OBD_PAGE_ALLOC(pg, gfp_mask);
615 echo_persistent_pages_fini ();
619 memset (cfs_kmap (pg), 0, CFS_PAGE_SIZE);
622 echo_persistent_pages[i] = pg;
628 static int __init obdecho_init(void)
630 struct lprocfs_static_vars lvars;
634 printk(KERN_INFO "Lustre: Echo OBD driver; http://www.lustre.org/\n");
636 LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0);
638 lprocfs_echo_init_vars(&lvars);
640 rc = echo_persistent_pages_init ();
644 rc = class_register_type(&echo_obd_ops, lvars.module_vars,
649 rc = echo_client_init();
653 class_unregister_type(LUSTRE_ECHO_NAME);
655 echo_persistent_pages_fini ();
660 static void /*__exit*/ obdecho_exit(void)
663 class_unregister_type(LUSTRE_ECHO_NAME);
664 echo_persistent_pages_fini ();
667 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
668 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
669 MODULE_LICENSE("GPL");
671 cfs_module(obdecho, "1.0.0", obdecho_init, obdecho_exit);