1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
5 * Author: Peter Braam <braam@clusterfs.com>
6 * Author: Andreas Dilger <adilger@clusterfs.com>
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 #include <linux/version.h>
27 #include <linux/module.h>
29 #include <linux/highmem.h>
31 #include <linux/stat.h>
32 #include <linux/sched.h>
33 #include <linux/smp_lock.h>
34 #include <linux/ext2_fs.h>
35 #include <linux/quotaops.h>
36 #include <linux/proc_fs.h>
37 #include <linux/init.h>
38 #include <asm/unistd.h>
40 #define DEBUG_SUBSYSTEM S_ECHO
42 #include <linux/obd_support.h>
43 #include <linux/obd_class.h>
44 #include <linux/obd_echo.h>
45 #include <linux/lustre_debug.h>
46 #include <linux/lustre_dlm.h>
47 #include <linux/lprocfs_status.h>
49 #define ECHO_INIT_OBJID 0x1000000000000000ULL
50 #define ECHO_HANDLE_MAGIC 0xabcd0123fedc9876ULL
52 #define ECHO_OBJECT0_NPAGES 16
53 static struct page *echo_object0_pages[ECHO_OBJECT0_NPAGES];
56 LPROC_ECHO_READ_BYTES = 1,
57 LPROC_ECHO_WRITE_BYTES = 2,
58 LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES +1
61 static int echo_connect(struct lustre_handle *conn, struct obd_device *obd,
62 struct obd_uuid *cluuid)
64 return class_connect(conn, obd, cluuid);
67 static int echo_disconnect(struct lustre_handle *conn, int failover)
69 struct obd_export *exp = class_conn2export(conn);
71 LASSERT (exp != NULL);
73 ldlm_cancel_locks_for_export(exp);
74 class_export_put(exp);
75 return (class_disconnect(conn, failover));
78 static __u64 echo_next_id(struct obd_device *obddev)
82 spin_lock(&obddev->u.echo.eo_lock);
83 id = ++obddev->u.echo.eo_lastino;
84 spin_unlock(&obddev->u.echo.eo_lock);
89 int echo_create(struct lustre_handle *conn, struct obdo *oa,
90 struct lov_stripe_md **ea, struct obd_trans_info *oti)
92 struct obd_device *obd = class_conn2obd(conn);
95 CERROR("invalid client cookie "LPX64"\n", conn->cookie);
99 if (!(oa->o_mode && S_IFMT)) {
100 CERROR("echo obd: no type!\n");
104 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
105 CERROR("invalid o_valid %08x\n", oa->o_valid);
109 oa->o_id = echo_next_id(obd);
110 oa->o_valid = OBD_MD_FLID;
111 atomic_inc(&obd->u.echo.eo_create);
116 int echo_destroy(struct lustre_handle *conn, struct obdo *oa,
117 struct lov_stripe_md *ea, struct obd_trans_info *oti)
119 struct obd_device *obd = class_conn2obd(conn);
122 CERROR("invalid client cookie "LPX64"\n", conn->cookie);
126 if (!(oa->o_valid & OBD_MD_FLID)) {
127 CERROR("obdo missing FLID valid flag: %08x\n", oa->o_valid);
131 if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OBJID) {
132 CERROR("bad destroy objid: "LPX64"\n", oa->o_id);
136 atomic_inc(&obd->u.echo.eo_destroy);
141 static int echo_open(struct lustre_handle *conn, struct obdo *oa,
142 struct lov_stripe_md *md, struct obd_trans_info *oti,
143 struct obd_client_handle *och)
145 struct lustre_handle *fh = obdo_handle (oa);
146 struct obd_device *obd = class_conn2obd (conn);
149 CERROR("invalid client cookie "LPX64"\n", conn->cookie);
153 if (!(oa->o_valid & OBD_MD_FLID)) {
154 CERROR ("obdo missing FLID valid flag: %08x\n", oa->o_valid);
158 fh->cookie = ECHO_HANDLE_MAGIC;
160 oa->o_valid |= OBD_MD_FLHANDLE;
164 static int echo_close(struct lustre_handle *conn, struct obdo *oa,
165 struct lov_stripe_md *md, struct obd_trans_info *oti)
167 struct lustre_handle *fh = obdo_handle (oa);
168 struct obd_device *obd = class_conn2obd(conn);
171 CERROR("invalid client cookie "LPX64"\n", conn->cookie);
175 if (!(oa->o_valid & OBD_MD_FLHANDLE)) {
176 CERROR("obdo missing FLHANDLE valid flag: %08x\n", oa->o_valid);
180 if (fh->cookie != ECHO_HANDLE_MAGIC) {
181 CERROR ("invalid file handle on close: "LPX64"\n", fh->cookie);
188 static int echo_getattr(struct lustre_handle *conn, struct obdo *oa,
189 struct lov_stripe_md *md)
191 struct obd_device *obd = class_conn2obd(conn);
192 obd_id id = oa->o_id;
195 CERROR("invalid client cookie "LPX64"\n", conn->cookie);
199 if (!(oa->o_valid & OBD_MD_FLID)) {
200 CERROR("obdo missing FLID valid flag: %08x\n", oa->o_valid);
204 obdo_cpy_md(oa, &obd->u.echo.oa, oa->o_valid);
210 static int echo_setattr(struct lustre_handle *conn, struct obdo *oa,
211 struct lov_stripe_md *md, struct obd_trans_info *oti)
213 struct obd_device *obd = class_conn2obd(conn);
216 CERROR("invalid client cookie "LPX64"\n", conn->cookie);
220 if (!(oa->o_valid & OBD_MD_FLID)) {
221 CERROR("obdo missing FLID valid flag: %08x\n", oa->o_valid);
225 memcpy(&obd->u.echo.oa, oa, sizeof(*oa));
227 atomic_inc(&obd->u.echo.eo_setattr);
232 /* This allows us to verify that desc_private is passed unmolested */
233 #define DESC_PRIV 0x10293847
235 int echo_preprw(int cmd, struct obd_export *export, struct obdo *oa,
236 int objcount, struct obd_ioobj *obj, int niocount,
237 struct niobuf_remote *nb, struct niobuf_local *res,
238 void **desc_private, struct obd_trans_info *oti)
240 struct obd_device *obd;
241 struct niobuf_local *r = res;
247 obd = export->exp_obd;
251 memset(res, 0, sizeof(*res) * niocount);
253 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
254 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, niocount);
256 *desc_private = (void *)DESC_PRIV;
258 for (i = 0; i < objcount; i++, obj++) {
259 int gfp_mask = (obj->ioo_id & 1) ? GFP_HIGHUSER : GFP_KERNEL;
260 int isobj0 = obj->ioo_id == 0;
261 int verify = !isobj0;
264 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++, r++) {
267 (nb->offset >> PAGE_SHIFT) < ECHO_OBJECT0_NPAGES) {
268 r->page = echo_object0_pages[nb->offset >>
270 /* Take extra ref so __free_pages() can be called OK */
273 r->page = alloc_pages(gfp_mask, 0);
274 if (r->page == NULL) {
275 CERROR("can't get page %u/%u for id "
277 j, obj->ioo_bufcnt, obj->ioo_id);
278 GOTO(preprw_cleanup, rc = -ENOMEM);
284 atomic_inc(&obd->u.echo.eo_prep);
286 r->offset = nb->offset;
288 LASSERT ((r->offset & (PAGE_SIZE - 1)) + r->len <= PAGE_SIZE);
290 CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n",
291 r->page, r->offset, r->len);
293 if (cmd & OBD_BRW_READ) {
296 page_debug_setup(kmap (r->page), r->len,
297 r->offset,obj->ioo_id);
303 page_debug_setup(kmap (r->page), r->len,
311 if (cmd & OBD_BRW_READ)
312 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
315 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
318 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
319 atomic_read(&obd->u.echo.eo_prep));
324 /* It is possible that we would rather handle errors by allow
325 * any already-set-up pages to complete, rather than tearing them
326 * all down again. I believe that this is what the in-kernel
327 * prep/commit operations do.
329 CERROR("cleaning up %ld pages (%d obdos)\n", (long)(r - res), objcount);
332 /* NB if this is an 'object0' page, __free_pages will just
333 * lose the extra ref gained above */
334 __free_pages(r->page, 0);
335 atomic_dec(&obd->u.echo.eo_prep);
337 memset(res, 0, sizeof(*res) * niocount);
342 int echo_commitrw(int cmd, struct obd_export *export, int objcount,
343 struct obd_ioobj *obj, int niocount, struct niobuf_local *res,
344 void *desc_private, struct obd_trans_info *oti)
346 struct obd_device *obd;
347 struct niobuf_local *r = res;
348 int i, vrc = 0, rc = 0;
351 obd = export->exp_obd;
355 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
356 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
359 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
363 if (niocount && !r) {
364 CERROR("NULL res niobuf with niocount %d\n", niocount);
368 LASSERT(desc_private == (void *)DESC_PRIV);
370 for (i = 0; i < objcount; i++, obj++) {
371 int verify = obj->ioo_id != 0;
374 for (j = 0 ; j < obj->ioo_bufcnt ; j++, r++) {
375 struct page *page = r->page;
378 if (!page || !(addr = kmap(page)) ||
379 !kern_addr_valid(addr)) {
381 CERROR("bad page objid "LPU64":%p, buf %d/%d\n",
382 obj->ioo_id, page, j, obj->ioo_bufcnt);
384 GOTO(commitrw_cleanup, rc = -EFAULT);
387 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n",
388 r->page, addr, r->offset);
391 vrc = page_debug_check("echo", addr, r->len,
392 r->offset, obj->ioo_id);
393 /* check all the pages always */
394 if (vrc != 0 && rc == 0)
399 /* NB see comment above regarding object0 pages */
400 __free_pages(page, 0);
401 atomic_dec(&obd->u.echo.eo_prep);
404 CDEBUG(D_PAGE, "%d pages remain after commit\n",
405 atomic_read(&obd->u.echo.eo_prep));
409 CERROR("cleaning up %ld pages (%d obdos)\n",
410 niocount - (long)(r - res) - 1, objcount);
411 while (++r < res + niocount) {
412 struct page *page = r->page;
414 /* NB see comment above regarding object0 pages */
415 __free_pages(page, 0);
416 atomic_dec(&obd->u.echo.eo_prep);
421 static int echo_setup(struct obd_device *obddev, obd_count len, void *buf)
425 spin_lock_init(&obddev->u.echo.eo_lock);
426 obddev->u.echo.eo_lastino = ECHO_INIT_OBJID;
428 obddev->obd_namespace =
429 ldlm_namespace_new("echo-tgt", LDLM_NAMESPACE_SERVER);
430 if (obddev->obd_namespace == NULL) {
435 ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
436 "echo_ldlm_cb_client", &obddev->obd_ldlm_client);
440 static int echo_cleanup(struct obd_device *obddev, int force, int failover)
444 ldlm_namespace_free(obddev->obd_namespace);
445 CERROR("%d prep/commitrw pages leaked\n",
446 atomic_read(&obddev->u.echo.eo_prep));
451 int echo_attach(struct obd_device *obd, obd_count len, void *data)
453 struct lprocfs_static_vars lvars;
456 lprocfs_init_vars(&lvars);
457 rc = lprocfs_obd_attach(obd, lvars.obd_vars);
460 rc = lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST);
464 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
465 LPROCFS_CNTR_AVGMINMAX, "read_bytes", "bytes");
466 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
467 LPROCFS_CNTR_AVGMINMAX, "write_bytes", "bytes");
471 int echo_detach(struct obd_device *dev)
473 lprocfs_free_obd_stats(dev);
474 return lprocfs_obd_detach(dev);
477 static struct obd_ops echo_obd_ops = {
478 o_owner: THIS_MODULE,
479 o_attach: echo_attach,
480 o_detach: echo_detach,
481 o_connect: echo_connect,
482 o_disconnect: echo_disconnect,
483 o_create: echo_create,
484 o_destroy: echo_destroy,
487 o_getattr: echo_getattr,
488 o_setattr: echo_setattr,
489 o_preprw: echo_preprw,
490 o_commitrw: echo_commitrw,
492 o_cleanup: echo_cleanup
495 extern int echo_client_init(void);
496 extern void echo_client_cleanup(void);
499 echo_object0_pages_fini (void)
503 for (i = 0; i < ECHO_OBJECT0_NPAGES; i++)
504 if (echo_object0_pages[i] != NULL) {
505 __free_pages (echo_object0_pages[i], 0);
506 echo_object0_pages[i] = NULL;
511 echo_object0_pages_init (void)
516 for (i = 0; i < ECHO_OBJECT0_NPAGES; i++) {
517 int gfp_mask = (i < ECHO_OBJECT0_NPAGES/2) ?
518 GFP_KERNEL : GFP_HIGHUSER;
520 pg = alloc_pages (gfp_mask, 0);
522 echo_object0_pages_fini ();
526 memset (kmap (pg), 0, PAGE_SIZE);
529 echo_object0_pages[i] = pg;
535 static int __init obdecho_init(void)
537 struct lprocfs_static_vars lvars;
540 printk(KERN_INFO "Lustre Echo OBD driver; info@clusterfs.com\n");
542 lprocfs_init_vars(&lvars);
544 rc = echo_object0_pages_init ();
548 rc = class_register_type(&echo_obd_ops, lvars.module_vars,
549 OBD_ECHO_DEVICENAME);
553 rc = echo_client_init();
557 class_unregister_type(OBD_ECHO_DEVICENAME);
559 echo_object0_pages_fini ();
564 static void __exit obdecho_exit(void)
566 echo_client_cleanup();
567 class_unregister_type(OBD_ECHO_DEVICENAME);
568 echo_object0_pages_fini ();
571 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
572 MODULE_DESCRIPTION("Lustre Testing Echo OBD driver");
573 MODULE_LICENSE("GPL");
575 module_init(obdecho_init);
576 module_exit(obdecho_exit);