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, 2012, Intel Corporation.
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
38 # include <asm/atomic.h>
40 #include "../include/obd_support.h"
41 #include "../include/obd_class.h"
42 #include "../../include/linux/lnet/lnetctl.h"
43 #include "../include/lustre_debug.h"
44 #include "../include/lprocfs_status.h"
45 #include "../include/lustre/lustre_build_version.h"
46 #include <linux/list.h>
47 #include "../include/cl_object.h"
48 #include "llog_internal.h"
51 struct obd_device *obd_devs[MAX_OBD_DEVICES];
52 EXPORT_SYMBOL(obd_devs);
53 struct list_head obd_types;
54 DEFINE_RWLOCK(obd_dev_lock);
56 __u64 obd_max_pages = 0;
57 EXPORT_SYMBOL(obd_max_pages);
58 __u64 obd_max_alloc = 0;
59 EXPORT_SYMBOL(obd_max_alloc);
61 EXPORT_SYMBOL(obd_alloc);
63 EXPORT_SYMBOL(obd_pages);
64 DEFINE_SPINLOCK(obd_updatemax_lock);
66 /* The following are visible and mutable through /proc/sys/lustre/. */
67 unsigned int obd_alloc_fail_rate = 0;
68 EXPORT_SYMBOL(obd_alloc_fail_rate);
69 unsigned int obd_debug_peer_on_timeout;
70 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
71 unsigned int obd_dump_on_timeout;
72 EXPORT_SYMBOL(obd_dump_on_timeout);
73 unsigned int obd_dump_on_eviction;
74 EXPORT_SYMBOL(obd_dump_on_eviction);
75 unsigned int obd_max_dirty_pages = 256;
76 EXPORT_SYMBOL(obd_max_dirty_pages);
77 atomic_t obd_dirty_pages;
78 EXPORT_SYMBOL(obd_dirty_pages);
79 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
80 EXPORT_SYMBOL(obd_timeout);
81 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
82 EXPORT_SYMBOL(ldlm_timeout);
83 unsigned int obd_timeout_set;
84 EXPORT_SYMBOL(obd_timeout_set);
85 unsigned int ldlm_timeout_set;
86 EXPORT_SYMBOL(ldlm_timeout_set);
87 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
88 unsigned int at_min = 0;
89 EXPORT_SYMBOL(at_min);
90 unsigned int at_max = 600;
91 EXPORT_SYMBOL(at_max);
92 unsigned int at_history = 600;
93 EXPORT_SYMBOL(at_history);
94 int at_early_margin = 5;
95 EXPORT_SYMBOL(at_early_margin);
97 EXPORT_SYMBOL(at_extra);
99 atomic_t obd_dirty_transit_pages;
100 EXPORT_SYMBOL(obd_dirty_transit_pages);
102 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
103 EXPORT_SYMBOL(obd_jobid_var);
105 char obd_jobid_node[JOBSTATS_JOBID_SIZE + 1];
107 /* Get jobid of current process from stored variable or calculate
108 * it from pid and user_id.
110 * Historically this was also done by reading the environment variable
111 * stored in between the "env_start" & "env_end" of task struct.
112 * This is now deprecated.
114 int lustre_get_jobid(char *jobid)
116 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
117 /* Jobstats isn't enabled */
118 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
121 /* Use process name + fsuid as jobid */
122 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
123 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
125 from_kuid(&init_user_ns, current_fsuid()));
129 /* Whole node dedicated to single job */
130 if (strcmp(obd_jobid_var, JOBSTATS_NODELOCAL) == 0) {
131 strcpy(jobid, obd_jobid_node);
137 EXPORT_SYMBOL(lustre_get_jobid);
139 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
140 size_t size, const char *file, int line)
143 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
144 CERROR("%s%salloc of %s (%llu bytes) failed at %s:%d\n",
145 ptr ? "force " :"", type, name, (__u64)size, file,
147 CERROR("%llu total bytes and %llu total pages "
148 "(%llu bytes) allocated by Lustre, "
149 "%d total bytes by LNET\n",
151 obd_pages_sum() << PAGE_CACHE_SHIFT,
153 atomic_read(&libcfs_kmemory));
158 EXPORT_SYMBOL(obd_alloc_fail);
160 static inline void obd_data2conn(struct lustre_handle *conn,
161 struct obd_ioctl_data *data)
163 memset(conn, 0, sizeof(*conn));
164 conn->cookie = data->ioc_cookie;
167 static inline void obd_conn2data(struct obd_ioctl_data *data,
168 struct lustre_handle *conn)
170 data->ioc_cookie = conn->cookie;
173 int class_resolve_dev_name(__u32 len, const char *name)
179 CERROR("No name passed,!\n");
180 GOTO(out, rc = -EINVAL);
182 if (name[len - 1] != 0) {
183 CERROR("Name not nul terminated!\n");
184 GOTO(out, rc = -EINVAL);
187 CDEBUG(D_IOCTL, "device name %s\n", name);
188 dev = class_name2dev(name);
190 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
191 GOTO(out, rc = -EINVAL);
194 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
201 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
204 struct obd_ioctl_data *data;
205 struct libcfs_debug_ioctl_data *debug_data;
206 struct obd_device *obd = NULL;
207 int err = 0, len = 0;
209 /* only for debugging */
210 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
211 debug_data = (struct libcfs_debug_ioctl_data*)arg;
212 libcfs_subsystem_debug = debug_data->subs;
213 libcfs_debug = debug_data->debug;
217 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
218 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
219 CERROR("OBD ioctl: data error\n");
222 data = (struct obd_ioctl_data *)buf;
225 case OBD_IOC_PROCESS_CFG: {
226 struct lustre_cfg *lcfg;
228 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
229 CERROR("No config buffer passed!\n");
230 GOTO(out, err = -EINVAL);
232 OBD_ALLOC(lcfg, data->ioc_plen1);
234 GOTO(out, err = -ENOMEM);
235 err = copy_from_user(lcfg, data->ioc_pbuf1,
238 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
240 err = class_process_config(lcfg);
242 OBD_FREE(lcfg, data->ioc_plen1);
246 case OBD_GET_VERSION:
247 if (!data->ioc_inlbuf1) {
248 CERROR("No buffer passed in ioctl\n");
249 GOTO(out, err = -EINVAL);
252 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
253 CERROR("ioctl buffer too small to hold version\n");
254 GOTO(out, err = -EINVAL);
257 memcpy(data->ioc_bulk, BUILD_VERSION,
258 strlen(BUILD_VERSION) + 1);
260 err = obd_ioctl_popdata((void *)arg, data, len);
265 case OBD_IOC_NAME2DEV: {
266 /* Resolve a device name. This does not change the
267 * currently selected device.
271 dev = class_resolve_dev_name(data->ioc_inllen1,
275 GOTO(out, err = -EINVAL);
277 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
283 case OBD_IOC_UUID2DEV: {
284 /* Resolve a device uuid. This does not change the
285 * currently selected device.
288 struct obd_uuid uuid;
290 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
291 CERROR("No UUID passed!\n");
292 GOTO(out, err = -EINVAL);
294 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
295 CERROR("UUID not NUL terminated!\n");
296 GOTO(out, err = -EINVAL);
299 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
300 obd_str2uuid(&uuid, data->ioc_inlbuf1);
301 dev = class_uuid2dev(&uuid);
304 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
306 GOTO(out, err = -EINVAL);
309 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
311 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
317 case OBD_IOC_CLOSE_UUID: {
318 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
323 case OBD_IOC_GETDEVICE: {
324 int index = data->ioc_count;
327 if (!data->ioc_inlbuf1) {
328 CERROR("No buffer passed in ioctl\n");
329 GOTO(out, err = -EINVAL);
331 if (data->ioc_inllen1 < 128) {
332 CERROR("ioctl buffer too small to hold version\n");
333 GOTO(out, err = -EINVAL);
336 obd = class_num2obd(index);
338 GOTO(out, err = -ENOENT);
340 if (obd->obd_stopping)
342 else if (obd->obd_set_up)
344 else if (obd->obd_attached)
348 str = (char *)data->ioc_bulk;
349 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
350 (int)index, status, obd->obd_type->typ_name,
351 obd->obd_name, obd->obd_uuid.uuid,
352 atomic_read(&obd->obd_refcount));
353 err = obd_ioctl_popdata((void *)arg, data, len);
360 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
361 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
362 GOTO(out, err = -EINVAL);
363 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
364 GOTO(out, err = -EINVAL);
365 obd = class_name2obd(data->ioc_inlbuf4);
366 } else if (data->ioc_dev < class_devno_max()) {
367 obd = class_num2obd(data->ioc_dev);
369 CERROR("OBD ioctl: No device\n");
370 GOTO(out, err = -EINVAL);
374 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
375 GOTO(out, err = -EINVAL);
377 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
379 if (!obd->obd_set_up || obd->obd_stopping) {
380 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
381 GOTO(out, err = -EINVAL);
385 case OBD_IOC_NO_TRANSNO: {
386 if (!obd->obd_attached) {
387 CERROR("Device %d not attached\n", obd->obd_minor);
388 GOTO(out, err = -ENODEV);
390 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
392 obd->obd_no_transno = 1;
397 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
401 err = obd_ioctl_popdata((void *)arg, data, len);
410 obd_ioctl_freedata(buf, len);
412 } /* class_handle_ioctl */
414 extern struct miscdevice obd_psdev;
416 #define OBD_INIT_CHECK
417 int obd_init_checks(void)
419 __u64 u64val, div64val;
423 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", "%llu", "%lld", "%#llx");
425 CDEBUG(D_INFO, "OBD_OBJECT_EOF = %#llx\n", (__u64)OBD_OBJECT_EOF);
427 u64val = OBD_OBJECT_EOF;
428 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = %#llx\n", u64val);
429 if (u64val != OBD_OBJECT_EOF) {
430 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
431 u64val, (int)sizeof(u64val));
434 len = snprintf(buf, sizeof(buf), "%#llx", u64val);
436 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
440 div64val = OBD_OBJECT_EOF;
441 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = %#llx\n", u64val);
442 if (u64val != OBD_OBJECT_EOF) {
443 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
444 u64val, (int)sizeof(u64val));
447 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
448 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
449 u64val, (int)sizeof(u64val));
452 if (do_div(div64val, 256) != (u64val & 255)) {
453 CERROR("do_div(%#llx,256) != %llu\n", u64val, u64val &255);
456 if (u64val >> 8 != div64val) {
457 CERROR("do_div(%#llx,256) %llu != %llu\n",
458 u64val, div64val, u64val >> 8);
461 len = snprintf(buf, sizeof(buf), "%#llx", u64val);
463 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
466 len = snprintf(buf, sizeof(buf), "%llu", u64val);
468 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
471 len = snprintf(buf, sizeof(buf), "%lld", u64val);
473 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
476 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
477 CWARN("mask failed: u64val %llu >= %llu\n", u64val,
478 (__u64)PAGE_CACHE_SIZE);
485 extern spinlock_t obd_types_lock;
486 #if defined (CONFIG_PROC_FS)
487 extern int class_procfs_init(void);
488 extern int class_procfs_clean(void);
490 static inline int class_procfs_init(void)
492 static inline int class_procfs_clean(void)
496 static int __init init_obdclass(void)
499 int lustre_register_fs(void);
501 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
502 INIT_LIST_HEAD(&capa_list[i]);
504 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
506 spin_lock_init(&obd_types_lock);
507 obd_zombie_impexp_init();
509 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
510 LPROCFS_STATS_FLAG_NONE |
511 LPROCFS_STATS_FLAG_IRQ_SAFE);
512 if (obd_memory == NULL) {
513 CERROR("kmalloc of 'obd_memory' failed\n");
517 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
518 LPROCFS_CNTR_AVGMINMAX,
520 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
521 LPROCFS_CNTR_AVGMINMAX,
522 "pagesused", "pages");
524 err = obd_init_checks();
525 if (err == -EOVERFLOW)
528 class_init_uuidlist();
529 err = class_handle_init();
533 INIT_LIST_HEAD(&obd_types);
535 err = misc_register(&obd_psdev);
537 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
541 /* This struct is already zeroed for us (static global) */
542 for (i = 0; i < class_devno_max(); i++)
545 /* Default the dirty page cache cap to 1/2 of system memory.
546 * For clients with less memory, a larger fraction is needed
547 * for other purposes (mostly for BGL). */
548 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
549 obd_max_dirty_pages = totalram_pages / 4;
551 obd_max_dirty_pages = totalram_pages / 2;
553 err = obd_init_caches();
559 err = class_procfs_init();
563 err = lu_global_init();
567 err = cl_global_init();
572 err = llog_info_init();
576 err = lustre_register_fs();
581 void obd_update_maxusage(void)
585 max1 = obd_pages_sum();
586 max2 = obd_memory_sum();
588 spin_lock(&obd_updatemax_lock);
589 if (max1 > obd_max_pages)
590 obd_max_pages = max1;
591 if (max2 > obd_max_alloc)
592 obd_max_alloc = max2;
593 spin_unlock(&obd_updatemax_lock);
595 EXPORT_SYMBOL(obd_update_maxusage);
597 #if defined (CONFIG_PROC_FS)
598 __u64 obd_memory_max(void)
602 spin_lock(&obd_updatemax_lock);
604 spin_unlock(&obd_updatemax_lock);
608 EXPORT_SYMBOL(obd_memory_max);
610 __u64 obd_pages_max(void)
614 spin_lock(&obd_updatemax_lock);
616 spin_unlock(&obd_updatemax_lock);
620 EXPORT_SYMBOL(obd_pages_max);
623 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
624 * ifdef to the end of the file to cover module and versioning goo.*/
625 static void cleanup_obdclass(void)
628 int lustre_unregister_fs(void);
629 __u64 memory_leaked, pages_leaked;
630 __u64 memory_max, pages_max;
632 lustre_unregister_fs();
634 misc_deregister(&obd_psdev);
635 for (i = 0; i < class_devno_max(); i++) {
636 struct obd_device *obd = class_num2obd(i);
637 if (obd && obd->obd_set_up &&
638 OBT(obd) && OBP(obd, detach)) {
639 /* XXX should this call generic detach otherwise? */
640 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
641 OBP(obd, detach)(obd);
648 obd_cleanup_caches();
651 class_procfs_clean();
653 class_handle_cleanup();
654 class_exit_uuidlist();
655 obd_zombie_impexp_stop();
657 memory_leaked = obd_memory_sum();
658 pages_leaked = obd_pages_sum();
660 memory_max = obd_memory_max();
661 pages_max = obd_pages_max();
663 lprocfs_free_stats(&obd_memory);
664 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
665 "obd_memory max: %llu, leaked: %llu\n",
666 memory_max, memory_leaked);
667 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
668 "obd_memory_pages max: %llu, leaked: %llu\n",
669 pages_max, pages_leaked);
672 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
673 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
674 MODULE_LICENSE("GPL");
675 MODULE_VERSION(LUSTRE_VERSION_STRING);
677 module_init(init_obdclass);
678 module_exit(cleanup_obdclass);