4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_CLASS
39 # include <liblustre.h>
41 # include <asm/atomic.h>
44 #include <obd_support.h>
45 #include <obd_class.h>
46 #include <lnet/lnetctl.h>
47 #include <lustre_debug.h>
48 #include <lprocfs_status.h>
49 #include <lustre/lustre_build_version.h>
50 #include <libcfs/list.h>
51 #include "llog_internal.h"
54 /* liblustre workaround */
55 cfs_atomic_t libcfs_kmemory = {0};
58 struct obd_device *obd_devs[MAX_OBD_DEVICES];
59 EXPORT_SYMBOL(obd_devs);
61 DEFINE_RWLOCK(obd_dev_lock);
64 __u64 obd_max_pages = 0;
65 __u64 obd_max_alloc = 0;
70 /* The following are visible and mutable through /proc/sys/lustre/. */
71 unsigned int obd_debug_peer_on_timeout;
72 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
73 unsigned int obd_dump_on_timeout;
74 EXPORT_SYMBOL(obd_dump_on_timeout);
75 unsigned int obd_dump_on_eviction;
76 EXPORT_SYMBOL(obd_dump_on_eviction);
77 unsigned int obd_max_dirty_pages = 256;
78 EXPORT_SYMBOL(obd_max_dirty_pages);
79 cfs_atomic_t obd_dirty_pages;
80 EXPORT_SYMBOL(obd_dirty_pages);
81 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
82 EXPORT_SYMBOL(obd_timeout);
83 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
84 EXPORT_SYMBOL(ldlm_timeout);
85 unsigned int obd_timeout_set;
86 EXPORT_SYMBOL(obd_timeout_set);
87 unsigned int ldlm_timeout_set;
88 EXPORT_SYMBOL(ldlm_timeout_set);
89 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
90 unsigned int at_min = 0;
91 EXPORT_SYMBOL(at_min);
92 unsigned int at_max = 600;
93 EXPORT_SYMBOL(at_max);
94 unsigned int at_history = 600;
95 EXPORT_SYMBOL(at_history);
96 int at_early_margin = 5;
97 EXPORT_SYMBOL(at_early_margin);
99 EXPORT_SYMBOL(at_extra);
101 cfs_atomic_t obd_dirty_transit_pages;
102 EXPORT_SYMBOL(obd_dirty_transit_pages);
104 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
105 EXPORT_SYMBOL(obd_jobid_var);
107 /* Get jobid of current process by reading the environment variable
108 * stored in between the "env_start" & "env_end" of task struct.
111 * It's better to cache the jobid for later use if there is any
112 * efficient way, the cl_env code probably could be reused for this
115 * If some job scheduler doesn't store jobid in the "env_start/end",
116 * then an upcall could be issued here to get the jobid by utilizing
117 * the userspace tools/api. Then, the jobid must be cached.
119 int lustre_get_jobid(char *jobid)
122 int jobid_len = JOBSTATS_JOBID_SIZE;
127 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
128 /* Jobstats isn't enabled */
129 if (!memcmp(obd_jobid_var, JOBSTATS_DISABLE,
130 strlen(JOBSTATS_DISABLE))) {
134 /* Use process name + fsuid as jobid */
135 if (!memcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID,
136 strlen(JOBSTATS_PROCNAME_UID))) {
137 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s_%u",
138 cfs_curproc_comm(), cfs_curproc_fsuid());
143 ret = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
145 CDEBUG((ret != -ENOENT && ret != -EINVAL && ret != -EDEADLK) ?
146 D_ERROR : D_INFO, "Get jobid for (%s) failed: rc = %d\n",
152 EXPORT_SYMBOL(lustre_get_jobid);
154 static inline void obd_data2conn(struct lustre_handle *conn,
155 struct obd_ioctl_data *data)
157 memset(conn, 0, sizeof *conn);
158 conn->cookie = data->ioc_cookie;
161 static inline void obd_conn2data(struct obd_ioctl_data *data,
162 struct lustre_handle *conn)
164 data->ioc_cookie = conn->cookie;
167 int class_resolve_dev_name(__u32 len, const char *name)
174 CERROR("No name passed,!\n");
175 GOTO(out, rc = -EINVAL);
177 if (name[len - 1] != 0) {
178 CERROR("Name not nul terminated!\n");
179 GOTO(out, rc = -EINVAL);
182 CDEBUG(D_IOCTL, "device name %s\n", name);
183 dev = class_name2dev(name);
185 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
186 GOTO(out, rc = -EINVAL);
189 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
196 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
199 struct obd_ioctl_data *data;
200 struct libcfs_debug_ioctl_data *debug_data;
201 struct obd_device *obd = NULL;
202 int err = 0, len = 0;
205 /* only for debugging */
206 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
207 debug_data = (struct libcfs_debug_ioctl_data*)arg;
208 libcfs_subsystem_debug = debug_data->subs;
209 libcfs_debug = debug_data->debug;
213 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
214 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
215 CERROR("OBD ioctl: data error\n");
218 data = (struct obd_ioctl_data *)buf;
221 case OBD_IOC_PROCESS_CFG: {
222 struct lustre_cfg *lcfg;
224 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
225 CERROR("No config buffer passed!\n");
226 GOTO(out, err = -EINVAL);
228 OBD_ALLOC(lcfg, data->ioc_plen1);
230 GOTO(out, err = -ENOMEM);
231 err = cfs_copy_from_user(lcfg, data->ioc_pbuf1,
234 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
236 err = class_process_config(lcfg);
238 OBD_FREE(lcfg, data->ioc_plen1);
242 case OBD_GET_VERSION:
243 if (!data->ioc_inlbuf1) {
244 CERROR("No buffer passed in ioctl\n");
245 GOTO(out, err = -EINVAL);
248 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
249 CERROR("ioctl buffer too small to hold version\n");
250 GOTO(out, err = -EINVAL);
253 memcpy(data->ioc_bulk, BUILD_VERSION,
254 strlen(BUILD_VERSION) + 1);
256 err = obd_ioctl_popdata((void *)arg, data, len);
261 case OBD_IOC_NAME2DEV: {
262 /* Resolve a device name. This does not change the
263 * currently selected device.
267 dev = class_resolve_dev_name(data->ioc_inllen1,
271 GOTO(out, err = -EINVAL);
273 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
279 case OBD_IOC_UUID2DEV: {
280 /* Resolve a device uuid. This does not change the
281 * currently selected device.
284 struct obd_uuid uuid;
286 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
287 CERROR("No UUID passed!\n");
288 GOTO(out, err = -EINVAL);
290 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
291 CERROR("UUID not NUL terminated!\n");
292 GOTO(out, err = -EINVAL);
295 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
296 obd_str2uuid(&uuid, data->ioc_inlbuf1);
297 dev = class_uuid2dev(&uuid);
300 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
302 GOTO(out, err = -EINVAL);
305 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
307 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
313 case OBD_IOC_CLOSE_UUID: {
314 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
319 case OBD_IOC_GETDEVICE: {
320 int index = data->ioc_count;
323 if (!data->ioc_inlbuf1) {
324 CERROR("No buffer passed in ioctl\n");
325 GOTO(out, err = -EINVAL);
327 if (data->ioc_inllen1 < 128) {
328 CERROR("ioctl buffer too small to hold version\n");
329 GOTO(out, err = -EINVAL);
332 obd = class_num2obd(index);
334 GOTO(out, err = -ENOENT);
336 if (obd->obd_stopping)
338 else if (obd->obd_set_up)
340 else if (obd->obd_attached)
344 str = (char *)data->ioc_bulk;
345 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
346 (int)index, status, obd->obd_type->typ_name,
347 obd->obd_name, obd->obd_uuid.uuid,
348 cfs_atomic_read(&obd->obd_refcount));
349 err = obd_ioctl_popdata((void *)arg, data, len);
356 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
357 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
358 GOTO(out, err = -EINVAL);
359 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
360 GOTO(out, err = -EINVAL);
361 obd = class_name2obd(data->ioc_inlbuf4);
362 } else if (data->ioc_dev < class_devno_max()) {
363 obd = class_num2obd(data->ioc_dev);
365 CERROR("OBD ioctl: No device\n");
366 GOTO(out, err = -EINVAL);
370 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
371 GOTO(out, err = -EINVAL);
373 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
375 if (!obd->obd_set_up || obd->obd_stopping) {
376 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
377 GOTO(out, err = -EINVAL);
381 case OBD_IOC_NO_TRANSNO: {
382 if (!obd->obd_attached) {
383 CERROR("Device %d not attached\n", obd->obd_minor);
384 GOTO(out, err = -ENODEV);
386 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
388 obd->obd_no_transno = 1;
393 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
397 err = obd_ioctl_popdata((void *)arg, data, len);
406 obd_ioctl_freedata(buf, len);
408 } /* class_handle_ioctl */
411 extern cfs_psdev_t obd_psdev;
413 void *obd_psdev = NULL;
416 #define OBD_INIT_CHECK
417 #ifdef OBD_INIT_CHECK
418 int obd_init_checks(void)
420 __u64 u64val, div64val;
424 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
426 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
428 u64val = OBD_OBJECT_EOF;
429 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
430 if (u64val != OBD_OBJECT_EOF) {
431 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
432 u64val, (int)sizeof(u64val));
435 len = snprintf(buf, sizeof(buf), LPX64, u64val);
437 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
441 div64val = OBD_OBJECT_EOF;
442 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
443 if (u64val != OBD_OBJECT_EOF) {
444 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
445 u64val, (int)sizeof(u64val));
448 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
449 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
450 u64val, (int)sizeof(u64val));
453 if (do_div(div64val, 256) != (u64val & 255)) {
454 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
457 if (u64val >> 8 != div64val) {
458 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
459 u64val, div64val, u64val >> 8);
462 len = snprintf(buf, sizeof(buf), LPX64, u64val);
464 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
467 len = snprintf(buf, sizeof(buf), LPU64, u64val);
469 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
472 len = snprintf(buf, sizeof(buf), LPD64, u64val);
474 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
477 if ((u64val & ~CFS_PAGE_MASK) >= CFS_PAGE_SIZE) {
478 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
479 (__u64)CFS_PAGE_SIZE);
486 #define obd_init_checks() do {} while(0)
489 extern cfs_spinlock_t obd_types_lock;
490 extern int class_procfs_init(void);
491 extern int class_procfs_clean(void);
494 static int __init init_obdclass(void)
496 int init_obdclass(void)
501 int lustre_register_fs(void);
503 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
504 CFS_INIT_LIST_HEAD(&capa_list[i]);
507 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
509 cfs_spin_lock_init(&obd_types_lock);
510 obd_zombie_impexp_init();
512 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
513 LPROCFS_STATS_FLAG_NONE |
514 LPROCFS_STATS_FLAG_IRQ_SAFE);
515 if (obd_memory == NULL) {
516 CERROR("kmalloc of 'obd_memory' failed\n");
520 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
521 LPROCFS_CNTR_AVGMINMAX,
523 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
524 LPROCFS_CNTR_AVGMINMAX,
525 "pagesused", "pages");
527 err = obd_init_checks();
528 if (err == -EOVERFLOW)
531 class_init_uuidlist();
532 err = class_handle_init();
536 CFS_INIT_LIST_HEAD(&obd_types);
538 err = cfs_psdev_register(&obd_psdev);
540 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
544 /* This struct is already zeroed for us (static global) */
545 for (i = 0; i < class_devno_max(); i++)
548 /* Default the dirty page cache cap to 1/2 of system memory.
549 * For clients with less memory, a larger fraction is needed
550 * for other purposes (mostly for BGL). */
551 if (cfs_num_physpages <= 512 << (20 - CFS_PAGE_SHIFT))
552 obd_max_dirty_pages = cfs_num_physpages / 4;
554 obd_max_dirty_pages = cfs_num_physpages / 2;
556 err = obd_init_caches();
560 err = class_procfs_init();
565 err = lu_global_init();
569 err = llog_info_init();
574 err = lustre_register_fs();
580 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
581 * ifdef to the end of the file to cover module and versioning goo.*/
583 static void cleanup_obdclass(void)
586 int lustre_unregister_fs(void);
587 __u64 memory_leaked, pages_leaked;
588 __u64 memory_max, pages_max;
591 lustre_unregister_fs();
593 cfs_psdev_deregister(&obd_psdev);
594 for (i = 0; i < class_devno_max(); i++) {
595 struct obd_device *obd = class_num2obd(i);
596 if (obd && obd->obd_set_up &&
597 OBT(obd) && OBP(obd, detach)) {
598 /* XXX should this call generic detach otherwise? */
599 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
600 OBP(obd, detach)(obd);
606 obd_cleanup_caches();
609 class_procfs_clean();
611 class_handle_cleanup();
612 class_exit_uuidlist();
613 obd_zombie_impexp_stop();
615 memory_leaked = obd_memory_sum();
616 pages_leaked = obd_pages_sum();
618 memory_max = obd_memory_max();
619 pages_max = obd_pages_max();
621 lprocfs_free_stats(&obd_memory);
622 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
623 "obd_memory max: "LPU64", leaked: "LPU64"\n",
624 memory_max, memory_leaked);
625 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
626 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
627 pages_max, pages_leaked);
632 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
633 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
634 MODULE_LICENSE("GPL");
636 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);