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 <lustre_ioctl.h>
57 #include "llog_internal.h"
60 /* liblustre workaround */
61 atomic_t libcfs_kmemory = {0};
64 struct obd_device *obd_devs[MAX_OBD_DEVICES];
65 EXPORT_SYMBOL(obd_devs);
67 DEFINE_RWLOCK(obd_dev_lock);
69 __u64 obd_max_pages = 0;
70 __u64 obd_max_alloc = 0;
75 DEFINE_SPINLOCK(obd_updatemax_lock);
77 /* The following are visible and mutable through /proc/sys/lustre/. */
78 unsigned int obd_alloc_fail_rate = 0;
79 EXPORT_SYMBOL(obd_alloc_fail_rate);
80 unsigned int obd_debug_peer_on_timeout;
81 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
82 unsigned int obd_dump_on_timeout;
83 EXPORT_SYMBOL(obd_dump_on_timeout);
84 unsigned int obd_dump_on_eviction;
85 EXPORT_SYMBOL(obd_dump_on_eviction);
86 unsigned int obd_max_dirty_pages = 256;
87 EXPORT_SYMBOL(obd_max_dirty_pages);
88 atomic_t obd_unstable_pages;
89 EXPORT_SYMBOL(obd_unstable_pages);
90 atomic_t obd_dirty_pages;
91 EXPORT_SYMBOL(obd_dirty_pages);
92 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
93 EXPORT_SYMBOL(obd_timeout);
94 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
95 EXPORT_SYMBOL(ldlm_timeout);
96 unsigned int obd_timeout_set;
97 EXPORT_SYMBOL(obd_timeout_set);
98 unsigned int ldlm_timeout_set;
99 EXPORT_SYMBOL(ldlm_timeout_set);
100 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
101 unsigned int at_min = 0;
102 EXPORT_SYMBOL(at_min);
103 unsigned int at_max = 600;
104 EXPORT_SYMBOL(at_max);
105 unsigned int at_history = 600;
106 EXPORT_SYMBOL(at_history);
107 int at_early_margin = 5;
108 EXPORT_SYMBOL(at_early_margin);
110 EXPORT_SYMBOL(at_extra);
112 atomic_t obd_dirty_transit_pages;
113 EXPORT_SYMBOL(obd_dirty_transit_pages);
115 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
116 EXPORT_SYMBOL(obd_jobid_var);
119 struct lprocfs_stats *obd_memory = NULL;
120 EXPORT_SYMBOL(obd_memory);
123 /* Get jobid of current process by reading the environment variable
124 * stored in between the "env_start" & "env_end" of task struct.
127 * It's better to cache the jobid for later use if there is any
128 * efficient way, the cl_env code probably could be reused for this
131 * If some job scheduler doesn't store jobid in the "env_start/end",
132 * then an upcall could be issued here to get the jobid by utilizing
133 * the userspace tools/api. Then, the jobid must be cached.
135 int lustre_get_jobid(char *jobid)
137 int jobid_len = JOBSTATS_JOBID_SIZE;
141 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
142 /* Jobstats isn't enabled */
143 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
146 /* Use process name + fsuid as jobid */
147 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
148 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
150 from_kuid(&init_user_ns, current_fsuid()));
154 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
156 if (rc == -EOVERFLOW) {
157 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
158 * variable length strings instead of just numbers), it
159 * might make sense to keep the unique parts for JobID,
160 * instead of just returning an error. That means a
161 * larger temp buffer for cfs_get_environ(), then
162 * truncating the string at some separator to fit into
163 * the specified jobid_len. Fix later if needed. */
165 if (unlikely(!printed)) {
166 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
167 "for JobID buffer (%d)\n",
168 obd_jobid_var, jobid_len);
172 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
173 rc == -EDEADLK) ? D_INFO : D_ERROR,
174 "Get jobid for (%s) failed: rc = %d\n",
180 EXPORT_SYMBOL(lustre_get_jobid);
182 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
183 size_t size, const char *file, int line)
186 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
187 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
188 ptr ? "force " :"", type, name, (__u64)size, file,
190 CERROR(LPU64" total bytes and "LPU64" total pages "
191 "("LPU64" bytes) allocated by Lustre, "
192 "%d total bytes by LNET\n",
194 obd_pages_sum() << PAGE_CACHE_SHIFT,
196 atomic_read(&libcfs_kmemory));
201 EXPORT_SYMBOL(obd_alloc_fail);
203 int class_resolve_dev_name(__u32 len, const char *name)
210 CERROR("No name passed,!\n");
211 GOTO(out, rc = -EINVAL);
213 if (name[len - 1] != 0) {
214 CERROR("Name not nul terminated!\n");
215 GOTO(out, rc = -EINVAL);
218 CDEBUG(D_IOCTL, "device name %s\n", name);
219 dev = class_name2dev(name);
221 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
222 GOTO(out, rc = -EINVAL);
225 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
232 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
235 struct obd_ioctl_data *data;
236 struct libcfs_debug_ioctl_data *debug_data;
237 struct obd_device *obd = NULL;
238 int err = 0, len = 0;
241 /* only for debugging */
242 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
243 debug_data = (struct libcfs_debug_ioctl_data*)arg;
244 libcfs_subsystem_debug = debug_data->subs;
245 libcfs_debug = debug_data->debug;
249 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
250 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
251 CERROR("OBD ioctl: data error\n");
254 data = (struct obd_ioctl_data *)buf;
257 case OBD_IOC_PROCESS_CFG: {
258 struct lustre_cfg *lcfg;
260 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
261 CERROR("No config buffer passed!\n");
262 GOTO(out, err = -EINVAL);
264 OBD_ALLOC(lcfg, data->ioc_plen1);
266 GOTO(out, err = -ENOMEM);
267 err = copy_from_user(lcfg, data->ioc_pbuf1,
270 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
272 err = class_process_config(lcfg);
274 OBD_FREE(lcfg, data->ioc_plen1);
278 case OBD_GET_VERSION:
279 if (!data->ioc_inlbuf1) {
280 CERROR("No buffer passed in ioctl\n");
281 GOTO(out, err = -EINVAL);
284 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
285 CERROR("ioctl buffer too small to hold version\n");
286 GOTO(out, err = -EINVAL);
289 memcpy(data->ioc_bulk, BUILD_VERSION,
290 strlen(BUILD_VERSION) + 1);
292 err = obd_ioctl_popdata((void *)arg, data, len);
297 case OBD_IOC_NAME2DEV: {
298 /* Resolve a device name. This does not change the
299 * currently selected device.
303 dev = class_resolve_dev_name(data->ioc_inllen1,
307 GOTO(out, err = -EINVAL);
309 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
315 case OBD_IOC_UUID2DEV: {
316 /* Resolve a device uuid. This does not change the
317 * currently selected device.
320 struct obd_uuid uuid;
322 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
323 CERROR("No UUID passed!\n");
324 GOTO(out, err = -EINVAL);
326 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
327 CERROR("UUID not NUL terminated!\n");
328 GOTO(out, err = -EINVAL);
331 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
332 obd_str2uuid(&uuid, data->ioc_inlbuf1);
333 dev = class_uuid2dev(&uuid);
336 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
338 GOTO(out, err = -EINVAL);
341 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
343 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
349 case OBD_IOC_GETDEVICE: {
350 int index = data->ioc_count;
353 if (!data->ioc_inlbuf1) {
354 CERROR("No buffer passed in ioctl\n");
355 GOTO(out, err = -EINVAL);
357 if (data->ioc_inllen1 < 128) {
358 CERROR("ioctl buffer too small to hold version\n");
359 GOTO(out, err = -EINVAL);
362 obd = class_num2obd(index);
364 GOTO(out, err = -ENOENT);
366 if (obd->obd_stopping)
368 else if (obd->obd_set_up)
370 else if (obd->obd_attached)
374 str = (char *)data->ioc_bulk;
375 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
376 (int)index, status, obd->obd_type->typ_name,
377 obd->obd_name, obd->obd_uuid.uuid,
378 atomic_read(&obd->obd_refcount));
379 err = obd_ioctl_popdata((void *)arg, data, len);
386 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
387 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
388 GOTO(out, err = -EINVAL);
389 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
390 GOTO(out, err = -EINVAL);
391 obd = class_name2obd(data->ioc_inlbuf4);
392 } else if (data->ioc_dev < class_devno_max()) {
393 obd = class_num2obd(data->ioc_dev);
395 CERROR("OBD ioctl: No device\n");
396 GOTO(out, err = -EINVAL);
400 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
401 GOTO(out, err = -EINVAL);
403 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
405 if (!obd->obd_set_up || obd->obd_stopping) {
406 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
407 GOTO(out, err = -EINVAL);
411 case OBD_IOC_NO_TRANSNO: {
412 if (!obd->obd_attached) {
413 CERROR("Device %d not attached\n", obd->obd_minor);
414 GOTO(out, err = -ENODEV);
416 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
418 obd->obd_no_transno = 1;
423 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
427 err = obd_ioctl_popdata((void *)arg, data, len);
436 obd_ioctl_freedata(buf, len);
438 } /* class_handle_ioctl */
441 extern struct miscdevice obd_psdev;
443 struct miscdevice obd_psdev;
446 #define OBD_INIT_CHECK
447 #ifdef OBD_INIT_CHECK
448 int obd_init_checks(void)
450 __u64 u64val, div64val;
454 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
456 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
458 u64val = OBD_OBJECT_EOF;
459 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
460 if (u64val != OBD_OBJECT_EOF) {
461 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
462 u64val, (int)sizeof(u64val));
465 len = snprintf(buf, sizeof(buf), LPX64, u64val);
467 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
471 div64val = OBD_OBJECT_EOF;
472 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
473 if (u64val != OBD_OBJECT_EOF) {
474 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
475 u64val, (int)sizeof(u64val));
478 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
479 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
480 u64val, (int)sizeof(u64val));
483 if (do_div(div64val, 256) != (u64val & 255)) {
484 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
487 if (u64val >> 8 != div64val) {
488 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
489 u64val, div64val, u64val >> 8);
492 len = snprintf(buf, sizeof(buf), LPX64, u64val);
494 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
497 len = snprintf(buf, sizeof(buf), LPU64, u64val);
499 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
502 len = snprintf(buf, sizeof(buf), LPD64, u64val);
504 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
507 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
508 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
509 (__u64)PAGE_CACHE_SIZE);
516 #define obd_init_checks() do {} while(0)
519 extern spinlock_t obd_types_lock;
520 extern int class_procfs_init(void);
521 extern int class_procfs_clean(void);
524 static int __init init_obdclass(void)
526 int init_obdclass(void)
531 int lustre_register_fs(void);
533 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
534 CFS_INIT_LIST_HEAD(&capa_list[i]);
537 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
539 spin_lock_init(&obd_types_lock);
540 obd_zombie_impexp_init();
542 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
543 LPROCFS_STATS_FLAG_NONE |
544 LPROCFS_STATS_FLAG_IRQ_SAFE);
545 if (obd_memory == NULL) {
546 CERROR("kmalloc of 'obd_memory' failed\n");
550 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
551 LPROCFS_CNTR_AVGMINMAX,
553 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
554 LPROCFS_CNTR_AVGMINMAX,
555 "pagesused", "pages");
557 err = obd_init_checks();
558 if (err == -EOVERFLOW)
561 class_init_uuidlist();
562 err = class_handle_init();
566 CFS_INIT_LIST_HEAD(&obd_types);
568 err = misc_register(&obd_psdev);
570 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
574 /* This struct is already zeroed for us (static global) */
575 for (i = 0; i < class_devno_max(); i++)
578 /* Default the dirty page cache cap to 1/2 of system memory.
579 * For clients with less memory, a larger fraction is needed
580 * for other purposes (mostly for BGL). */
581 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
582 obd_max_dirty_pages = totalram_pages / 4;
584 obd_max_dirty_pages = totalram_pages / 2;
586 err = obd_init_caches();
590 err = class_procfs_init();
595 err = lu_global_init();
599 err = lu_capainfo_init();
603 err = cl_global_init();
607 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
608 err = dt_global_init();
612 err = lu_ucred_global_init();
617 err = llog_info_init();
622 err = lustre_register_fs();
628 void obd_update_maxusage(void)
632 max1 = obd_pages_sum();
633 max2 = obd_memory_sum();
635 spin_lock(&obd_updatemax_lock);
636 if (max1 > obd_max_pages)
637 obd_max_pages = max1;
638 if (max2 > obd_max_alloc)
639 obd_max_alloc = max2;
640 spin_unlock(&obd_updatemax_lock);
642 EXPORT_SYMBOL(obd_update_maxusage);
645 __u64 obd_memory_max(void)
649 spin_lock(&obd_updatemax_lock);
651 spin_unlock(&obd_updatemax_lock);
655 EXPORT_SYMBOL(obd_memory_max);
657 __u64 obd_pages_max(void)
661 spin_lock(&obd_updatemax_lock);
663 spin_unlock(&obd_updatemax_lock);
667 EXPORT_SYMBOL(obd_pages_max);
670 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
671 * ifdef to the end of the file to cover module and versioning goo.*/
673 static void cleanup_obdclass(void)
675 int lustre_unregister_fs(void);
676 __u64 memory_leaked, pages_leaked;
677 __u64 memory_max, pages_max;
680 lustre_unregister_fs();
682 misc_deregister(&obd_psdev);
684 #ifdef HAVE_SERVER_SUPPORT
685 lu_ucred_global_fini();
692 obd_cleanup_caches();
695 class_procfs_clean();
697 class_handle_cleanup();
698 class_exit_uuidlist();
699 obd_zombie_impexp_stop();
701 memory_leaked = obd_memory_sum();
702 pages_leaked = obd_pages_sum();
704 memory_max = obd_memory_max();
705 pages_max = obd_pages_max();
707 lprocfs_free_stats(&obd_memory);
708 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
709 "obd_memory max: "LPU64", leaked: "LPU64"\n",
710 memory_max, memory_leaked);
711 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
712 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
713 pages_max, pages_leaked);
718 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
719 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
720 MODULE_LICENSE("GPL");
722 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);