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
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 <cl_object.h>
52 #ifdef HAVE_SERVER_SUPPORT
53 # include <dt_object.h>
54 # include <md_object.h>
55 #endif /* HAVE_SERVER_SUPPORT */
56 #include "llog_internal.h"
59 /* liblustre workaround */
60 cfs_atomic_t libcfs_kmemory = {0};
63 struct obd_device *obd_devs[MAX_OBD_DEVICES];
64 EXPORT_SYMBOL(obd_devs);
66 DEFINE_RWLOCK(obd_dev_lock);
68 __u64 obd_max_pages = 0;
69 __u64 obd_max_alloc = 0;
74 DEFINE_SPINLOCK(obd_updatemax_lock);
76 /* The following are visible and mutable through /proc/sys/lustre/. */
77 unsigned int obd_alloc_fail_rate = 0;
78 EXPORT_SYMBOL(obd_alloc_fail_rate);
79 unsigned int obd_debug_peer_on_timeout;
80 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
81 unsigned int obd_dump_on_timeout;
82 EXPORT_SYMBOL(obd_dump_on_timeout);
83 unsigned int obd_dump_on_eviction;
84 EXPORT_SYMBOL(obd_dump_on_eviction);
85 unsigned int obd_max_dirty_pages = 256;
86 EXPORT_SYMBOL(obd_max_dirty_pages);
87 cfs_atomic_t obd_dirty_pages;
88 EXPORT_SYMBOL(obd_dirty_pages);
89 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
90 EXPORT_SYMBOL(obd_timeout);
91 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
92 EXPORT_SYMBOL(ldlm_timeout);
93 unsigned int obd_timeout_set;
94 EXPORT_SYMBOL(obd_timeout_set);
95 unsigned int ldlm_timeout_set;
96 EXPORT_SYMBOL(ldlm_timeout_set);
97 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
98 unsigned int at_min = 0;
99 EXPORT_SYMBOL(at_min);
100 unsigned int at_max = 600;
101 EXPORT_SYMBOL(at_max);
102 unsigned int at_history = 600;
103 EXPORT_SYMBOL(at_history);
104 int at_early_margin = 5;
105 EXPORT_SYMBOL(at_early_margin);
107 EXPORT_SYMBOL(at_extra);
109 cfs_atomic_t obd_dirty_transit_pages;
110 EXPORT_SYMBOL(obd_dirty_transit_pages);
112 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
113 EXPORT_SYMBOL(obd_jobid_var);
116 struct lprocfs_stats *obd_memory = NULL;
117 EXPORT_SYMBOL(obd_memory);
120 /* Get jobid of current process by reading the environment variable
121 * stored in between the "env_start" & "env_end" of task struct.
124 * It's better to cache the jobid for later use if there is any
125 * efficient way, the cl_env code probably could be reused for this
128 * If some job scheduler doesn't store jobid in the "env_start/end",
129 * then an upcall could be issued here to get the jobid by utilizing
130 * the userspace tools/api. Then, the jobid must be cached.
132 int lustre_get_jobid(char *jobid)
134 int jobid_len = JOBSTATS_JOBID_SIZE;
138 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
139 /* Jobstats isn't enabled */
140 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
143 /* Use process name + fsuid as jobid */
144 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
145 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
146 current_comm(), current_fsuid());
150 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
152 if (rc == -EOVERFLOW) {
153 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
154 * variable length strings instead of just numbers), it
155 * might make sense to keep the unique parts for JobID,
156 * instead of just returning an error. That means a
157 * larger temp buffer for cfs_get_environ(), then
158 * truncating the string at some separator to fit into
159 * the specified jobid_len. Fix later if needed. */
161 if (unlikely(!printed)) {
162 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
163 "for JobID buffer (%d)\n",
164 obd_jobid_var, jobid_len);
168 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
169 rc == -EDEADLK) ? D_INFO : D_ERROR,
170 "Get jobid for (%s) failed: rc = %d\n",
176 EXPORT_SYMBOL(lustre_get_jobid);
178 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
179 size_t size, const char *file, int line)
182 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
183 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
184 ptr ? "force " :"", type, name, (__u64)size, file,
186 CERROR(LPU64" total bytes and "LPU64" total pages "
187 "("LPU64" bytes) allocated by Lustre, "
188 "%d total bytes by LNET\n",
190 obd_pages_sum() << PAGE_CACHE_SHIFT,
192 cfs_atomic_read(&libcfs_kmemory));
197 EXPORT_SYMBOL(obd_alloc_fail);
199 static inline void obd_data2conn(struct lustre_handle *conn,
200 struct obd_ioctl_data *data)
202 memset(conn, 0, sizeof *conn);
203 conn->cookie = data->ioc_cookie;
206 static inline void obd_conn2data(struct obd_ioctl_data *data,
207 struct lustre_handle *conn)
209 data->ioc_cookie = conn->cookie;
212 int class_resolve_dev_name(__u32 len, const char *name)
219 CERROR("No name passed,!\n");
220 GOTO(out, rc = -EINVAL);
222 if (name[len - 1] != 0) {
223 CERROR("Name not nul terminated!\n");
224 GOTO(out, rc = -EINVAL);
227 CDEBUG(D_IOCTL, "device name %s\n", name);
228 dev = class_name2dev(name);
230 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
231 GOTO(out, rc = -EINVAL);
234 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
241 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
244 struct obd_ioctl_data *data;
245 struct libcfs_debug_ioctl_data *debug_data;
246 struct obd_device *obd = NULL;
247 int err = 0, len = 0;
250 /* only for debugging */
251 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
252 debug_data = (struct libcfs_debug_ioctl_data*)arg;
253 libcfs_subsystem_debug = debug_data->subs;
254 libcfs_debug = debug_data->debug;
258 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
259 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
260 CERROR("OBD ioctl: data error\n");
263 data = (struct obd_ioctl_data *)buf;
266 case OBD_IOC_PROCESS_CFG: {
267 struct lustre_cfg *lcfg;
269 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
270 CERROR("No config buffer passed!\n");
271 GOTO(out, err = -EINVAL);
273 OBD_ALLOC(lcfg, data->ioc_plen1);
275 GOTO(out, err = -ENOMEM);
276 err = copy_from_user(lcfg, data->ioc_pbuf1,
279 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
281 err = class_process_config(lcfg);
283 OBD_FREE(lcfg, data->ioc_plen1);
287 case OBD_GET_VERSION:
288 if (!data->ioc_inlbuf1) {
289 CERROR("No buffer passed in ioctl\n");
290 GOTO(out, err = -EINVAL);
293 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
294 CERROR("ioctl buffer too small to hold version\n");
295 GOTO(out, err = -EINVAL);
298 memcpy(data->ioc_bulk, BUILD_VERSION,
299 strlen(BUILD_VERSION) + 1);
301 err = obd_ioctl_popdata((void *)arg, data, len);
306 case OBD_IOC_NAME2DEV: {
307 /* Resolve a device name. This does not change the
308 * currently selected device.
312 dev = class_resolve_dev_name(data->ioc_inllen1,
316 GOTO(out, err = -EINVAL);
318 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
324 case OBD_IOC_UUID2DEV: {
325 /* Resolve a device uuid. This does not change the
326 * currently selected device.
329 struct obd_uuid uuid;
331 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
332 CERROR("No UUID passed!\n");
333 GOTO(out, err = -EINVAL);
335 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
336 CERROR("UUID not NUL terminated!\n");
337 GOTO(out, err = -EINVAL);
340 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
341 obd_str2uuid(&uuid, data->ioc_inlbuf1);
342 dev = class_uuid2dev(&uuid);
345 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
347 GOTO(out, err = -EINVAL);
350 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
352 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
358 case OBD_IOC_CLOSE_UUID: {
359 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
364 case OBD_IOC_GETDEVICE: {
365 int index = data->ioc_count;
368 if (!data->ioc_inlbuf1) {
369 CERROR("No buffer passed in ioctl\n");
370 GOTO(out, err = -EINVAL);
372 if (data->ioc_inllen1 < 128) {
373 CERROR("ioctl buffer too small to hold version\n");
374 GOTO(out, err = -EINVAL);
377 obd = class_num2obd(index);
379 GOTO(out, err = -ENOENT);
381 if (obd->obd_stopping)
383 else if (obd->obd_set_up)
385 else if (obd->obd_attached)
389 str = (char *)data->ioc_bulk;
390 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
391 (int)index, status, obd->obd_type->typ_name,
392 obd->obd_name, obd->obd_uuid.uuid,
393 cfs_atomic_read(&obd->obd_refcount));
394 err = obd_ioctl_popdata((void *)arg, data, len);
401 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
402 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
403 GOTO(out, err = -EINVAL);
404 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
405 GOTO(out, err = -EINVAL);
406 obd = class_name2obd(data->ioc_inlbuf4);
407 } else if (data->ioc_dev < class_devno_max()) {
408 obd = class_num2obd(data->ioc_dev);
410 CERROR("OBD ioctl: No device\n");
411 GOTO(out, err = -EINVAL);
415 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
416 GOTO(out, err = -EINVAL);
418 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
420 if (!obd->obd_set_up || obd->obd_stopping) {
421 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
422 GOTO(out, err = -EINVAL);
426 case OBD_IOC_NO_TRANSNO: {
427 if (!obd->obd_attached) {
428 CERROR("Device %d not attached\n", obd->obd_minor);
429 GOTO(out, err = -ENODEV);
431 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
433 obd->obd_no_transno = 1;
438 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
442 err = obd_ioctl_popdata((void *)arg, data, len);
451 obd_ioctl_freedata(buf, len);
453 } /* class_handle_ioctl */
456 extern struct miscdevice obd_psdev;
458 struct miscdevice obd_psdev;
461 #define OBD_INIT_CHECK
462 #ifdef OBD_INIT_CHECK
463 int obd_init_checks(void)
465 __u64 u64val, div64val;
469 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
471 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
473 u64val = OBD_OBJECT_EOF;
474 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
475 if (u64val != OBD_OBJECT_EOF) {
476 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
477 u64val, (int)sizeof(u64val));
480 len = snprintf(buf, sizeof(buf), LPX64, u64val);
482 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
486 div64val = OBD_OBJECT_EOF;
487 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
488 if (u64val != OBD_OBJECT_EOF) {
489 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
490 u64val, (int)sizeof(u64val));
493 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
494 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
495 u64val, (int)sizeof(u64val));
498 if (do_div(div64val, 256) != (u64val & 255)) {
499 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
502 if (u64val >> 8 != div64val) {
503 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
504 u64val, div64val, u64val >> 8);
507 len = snprintf(buf, sizeof(buf), LPX64, u64val);
509 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
512 len = snprintf(buf, sizeof(buf), LPU64, u64val);
514 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
517 len = snprintf(buf, sizeof(buf), LPD64, u64val);
519 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
522 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
523 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
524 (__u64)PAGE_CACHE_SIZE);
531 #define obd_init_checks() do {} while(0)
534 extern spinlock_t obd_types_lock;
535 extern int class_procfs_init(void);
536 extern int class_procfs_clean(void);
539 static int __init init_obdclass(void)
541 int init_obdclass(void)
546 int lustre_register_fs(void);
548 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
549 CFS_INIT_LIST_HEAD(&capa_list[i]);
552 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
554 spin_lock_init(&obd_types_lock);
555 obd_zombie_impexp_init();
557 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
558 LPROCFS_STATS_FLAG_NONE |
559 LPROCFS_STATS_FLAG_IRQ_SAFE);
560 if (obd_memory == NULL) {
561 CERROR("kmalloc of 'obd_memory' failed\n");
565 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
566 LPROCFS_CNTR_AVGMINMAX,
568 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
569 LPROCFS_CNTR_AVGMINMAX,
570 "pagesused", "pages");
572 err = obd_init_checks();
573 if (err == -EOVERFLOW)
576 class_init_uuidlist();
577 err = class_handle_init();
581 CFS_INIT_LIST_HEAD(&obd_types);
583 err = misc_register(&obd_psdev);
585 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
589 /* This struct is already zeroed for us (static global) */
590 for (i = 0; i < class_devno_max(); i++)
593 /* Default the dirty page cache cap to 1/2 of system memory.
594 * For clients with less memory, a larger fraction is needed
595 * for other purposes (mostly for BGL). */
596 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
597 obd_max_dirty_pages = totalram_pages / 4;
599 obd_max_dirty_pages = totalram_pages / 2;
601 err = obd_init_caches();
605 err = class_procfs_init();
610 err = lu_global_init();
614 err = lu_capainfo_init();
618 err = cl_global_init();
622 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
623 err = dt_global_init();
627 err = lu_ucred_global_init();
632 err = llog_info_init();
637 err = lustre_register_fs();
643 void obd_update_maxusage(void)
647 max1 = obd_pages_sum();
648 max2 = obd_memory_sum();
650 spin_lock(&obd_updatemax_lock);
651 if (max1 > obd_max_pages)
652 obd_max_pages = max1;
653 if (max2 > obd_max_alloc)
654 obd_max_alloc = max2;
655 spin_unlock(&obd_updatemax_lock);
657 EXPORT_SYMBOL(obd_update_maxusage);
660 __u64 obd_memory_max(void)
664 spin_lock(&obd_updatemax_lock);
666 spin_unlock(&obd_updatemax_lock);
670 EXPORT_SYMBOL(obd_memory_max);
672 __u64 obd_pages_max(void)
676 spin_lock(&obd_updatemax_lock);
678 spin_unlock(&obd_updatemax_lock);
682 EXPORT_SYMBOL(obd_pages_max);
685 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
686 * ifdef to the end of the file to cover module and versioning goo.*/
688 static void cleanup_obdclass(void)
691 int lustre_unregister_fs(void);
692 __u64 memory_leaked, pages_leaked;
693 __u64 memory_max, pages_max;
696 lustre_unregister_fs();
698 misc_deregister(&obd_psdev);
699 for (i = 0; i < class_devno_max(); i++) {
700 struct obd_device *obd = class_num2obd(i);
701 if (obd && obd->obd_set_up &&
702 OBT(obd) && OBP(obd, detach)) {
703 /* XXX should this call generic detach otherwise? */
704 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
705 OBP(obd, detach)(obd);
709 #ifdef HAVE_SERVER_SUPPORT
710 lu_ucred_global_fini();
717 obd_cleanup_caches();
720 class_procfs_clean();
722 class_handle_cleanup();
723 class_exit_uuidlist();
724 obd_zombie_impexp_stop();
726 memory_leaked = obd_memory_sum();
727 pages_leaked = obd_pages_sum();
729 memory_max = obd_memory_max();
730 pages_max = obd_pages_max();
732 lprocfs_free_stats(&obd_memory);
733 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
734 "obd_memory max: "LPU64", leaked: "LPU64"\n",
735 memory_max, memory_leaked);
736 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
737 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
738 pages_max, pages_leaked);
743 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
744 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
745 MODULE_LICENSE("GPL");
747 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);