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_dirty_pages;
89 EXPORT_SYMBOL(obd_dirty_pages);
90 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
91 EXPORT_SYMBOL(obd_timeout);
92 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
93 EXPORT_SYMBOL(ldlm_timeout);
94 unsigned int obd_timeout_set;
95 EXPORT_SYMBOL(obd_timeout_set);
96 unsigned int ldlm_timeout_set;
97 EXPORT_SYMBOL(ldlm_timeout_set);
98 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
99 unsigned int at_min = 0;
100 EXPORT_SYMBOL(at_min);
101 unsigned int at_max = 600;
102 EXPORT_SYMBOL(at_max);
103 unsigned int at_history = 600;
104 EXPORT_SYMBOL(at_history);
105 int at_early_margin = 5;
106 EXPORT_SYMBOL(at_early_margin);
108 EXPORT_SYMBOL(at_extra);
110 atomic_t obd_dirty_transit_pages;
111 EXPORT_SYMBOL(obd_dirty_transit_pages);
113 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
114 EXPORT_SYMBOL(obd_jobid_var);
117 struct lprocfs_stats *obd_memory = NULL;
118 EXPORT_SYMBOL(obd_memory);
121 /* Get jobid of current process by reading the environment variable
122 * stored in between the "env_start" & "env_end" of task struct.
125 * It's better to cache the jobid for later use if there is any
126 * efficient way, the cl_env code probably could be reused for this
129 * If some job scheduler doesn't store jobid in the "env_start/end",
130 * then an upcall could be issued here to get the jobid by utilizing
131 * the userspace tools/api. Then, the jobid must be cached.
133 int lustre_get_jobid(char *jobid)
135 int jobid_len = JOBSTATS_JOBID_SIZE;
139 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
140 /* Jobstats isn't enabled */
141 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
144 /* Use process name + fsuid as jobid */
145 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
146 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
148 from_kuid(&init_user_ns, current_fsuid()));
152 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
154 if (rc == -EOVERFLOW) {
155 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
156 * variable length strings instead of just numbers), it
157 * might make sense to keep the unique parts for JobID,
158 * instead of just returning an error. That means a
159 * larger temp buffer for cfs_get_environ(), then
160 * truncating the string at some separator to fit into
161 * the specified jobid_len. Fix later if needed. */
163 if (unlikely(!printed)) {
164 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
165 "for JobID buffer (%d)\n",
166 obd_jobid_var, jobid_len);
170 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
171 rc == -EDEADLK) ? D_INFO : D_ERROR,
172 "Get jobid for (%s) failed: rc = %d\n",
178 EXPORT_SYMBOL(lustre_get_jobid);
180 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
181 size_t size, const char *file, int line)
184 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
185 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
186 ptr ? "force " :"", type, name, (__u64)size, file,
188 CERROR(LPU64" total bytes and "LPU64" total pages "
189 "("LPU64" bytes) allocated by Lustre, "
190 "%d total bytes by LNET\n",
192 obd_pages_sum() << PAGE_CACHE_SHIFT,
194 atomic_read(&libcfs_kmemory));
199 EXPORT_SYMBOL(obd_alloc_fail);
201 int class_resolve_dev_name(__u32 len, const char *name)
208 CERROR("No name passed,!\n");
209 GOTO(out, rc = -EINVAL);
211 if (name[len - 1] != 0) {
212 CERROR("Name not nul terminated!\n");
213 GOTO(out, rc = -EINVAL);
216 CDEBUG(D_IOCTL, "device name %s\n", name);
217 dev = class_name2dev(name);
219 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
220 GOTO(out, rc = -EINVAL);
223 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
230 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
233 struct obd_ioctl_data *data;
234 struct libcfs_debug_ioctl_data *debug_data;
235 struct obd_device *obd = NULL;
236 int err = 0, len = 0;
239 /* only for debugging */
240 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
241 debug_data = (struct libcfs_debug_ioctl_data*)arg;
242 libcfs_subsystem_debug = debug_data->subs;
243 libcfs_debug = debug_data->debug;
247 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
248 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
249 CERROR("OBD ioctl: data error\n");
252 data = (struct obd_ioctl_data *)buf;
255 case OBD_IOC_PROCESS_CFG: {
256 struct lustre_cfg *lcfg;
258 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
259 CERROR("No config buffer passed!\n");
260 GOTO(out, err = -EINVAL);
262 OBD_ALLOC(lcfg, data->ioc_plen1);
264 GOTO(out, err = -ENOMEM);
265 err = copy_from_user(lcfg, data->ioc_pbuf1,
268 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
270 err = class_process_config(lcfg);
272 OBD_FREE(lcfg, data->ioc_plen1);
276 case OBD_GET_VERSION:
277 if (!data->ioc_inlbuf1) {
278 CERROR("No buffer passed in ioctl\n");
279 GOTO(out, err = -EINVAL);
282 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
283 CERROR("ioctl buffer too small to hold version\n");
284 GOTO(out, err = -EINVAL);
287 memcpy(data->ioc_bulk, BUILD_VERSION,
288 strlen(BUILD_VERSION) + 1);
290 err = obd_ioctl_popdata((void *)arg, data, len);
295 case OBD_IOC_NAME2DEV: {
296 /* Resolve a device name. This does not change the
297 * currently selected device.
301 dev = class_resolve_dev_name(data->ioc_inllen1,
305 GOTO(out, err = -EINVAL);
307 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
313 case OBD_IOC_UUID2DEV: {
314 /* Resolve a device uuid. This does not change the
315 * currently selected device.
318 struct obd_uuid uuid;
320 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
321 CERROR("No UUID passed!\n");
322 GOTO(out, err = -EINVAL);
324 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
325 CERROR("UUID not NUL terminated!\n");
326 GOTO(out, err = -EINVAL);
329 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
330 obd_str2uuid(&uuid, data->ioc_inlbuf1);
331 dev = class_uuid2dev(&uuid);
334 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
336 GOTO(out, err = -EINVAL);
339 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
341 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
347 case OBD_IOC_GETDEVICE: {
348 int index = data->ioc_count;
351 if (!data->ioc_inlbuf1) {
352 CERROR("No buffer passed in ioctl\n");
353 GOTO(out, err = -EINVAL);
355 if (data->ioc_inllen1 < 128) {
356 CERROR("ioctl buffer too small to hold version\n");
357 GOTO(out, err = -EINVAL);
360 obd = class_num2obd(index);
362 GOTO(out, err = -ENOENT);
364 if (obd->obd_stopping)
366 else if (obd->obd_set_up)
368 else if (obd->obd_attached)
372 str = (char *)data->ioc_bulk;
373 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
374 (int)index, status, obd->obd_type->typ_name,
375 obd->obd_name, obd->obd_uuid.uuid,
376 atomic_read(&obd->obd_refcount));
377 err = obd_ioctl_popdata((void *)arg, data, len);
384 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
385 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
386 GOTO(out, err = -EINVAL);
387 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
388 GOTO(out, err = -EINVAL);
389 obd = class_name2obd(data->ioc_inlbuf4);
390 } else if (data->ioc_dev < class_devno_max()) {
391 obd = class_num2obd(data->ioc_dev);
393 CERROR("OBD ioctl: No device\n");
394 GOTO(out, err = -EINVAL);
398 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
399 GOTO(out, err = -EINVAL);
401 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
403 if (!obd->obd_set_up || obd->obd_stopping) {
404 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
405 GOTO(out, err = -EINVAL);
409 case OBD_IOC_NO_TRANSNO: {
410 if (!obd->obd_attached) {
411 CERROR("Device %d not attached\n", obd->obd_minor);
412 GOTO(out, err = -ENODEV);
414 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
416 obd->obd_no_transno = 1;
421 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
425 err = obd_ioctl_popdata((void *)arg, data, len);
434 obd_ioctl_freedata(buf, len);
436 } /* class_handle_ioctl */
439 extern struct miscdevice obd_psdev;
441 struct miscdevice obd_psdev;
444 #define OBD_INIT_CHECK
445 #ifdef OBD_INIT_CHECK
446 int obd_init_checks(void)
448 __u64 u64val, div64val;
452 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
454 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
456 u64val = OBD_OBJECT_EOF;
457 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
458 if (u64val != OBD_OBJECT_EOF) {
459 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
460 u64val, (int)sizeof(u64val));
463 len = snprintf(buf, sizeof(buf), LPX64, u64val);
465 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
469 div64val = OBD_OBJECT_EOF;
470 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
471 if (u64val != OBD_OBJECT_EOF) {
472 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
473 u64val, (int)sizeof(u64val));
476 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
477 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
478 u64val, (int)sizeof(u64val));
481 if (do_div(div64val, 256) != (u64val & 255)) {
482 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
485 if (u64val >> 8 != div64val) {
486 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
487 u64val, div64val, u64val >> 8);
490 len = snprintf(buf, sizeof(buf), LPX64, u64val);
492 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
495 len = snprintf(buf, sizeof(buf), LPU64, u64val);
497 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
500 len = snprintf(buf, sizeof(buf), LPD64, u64val);
502 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
505 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
506 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
507 (__u64)PAGE_CACHE_SIZE);
514 #define obd_init_checks() do {} while(0)
517 extern spinlock_t obd_types_lock;
518 extern int class_procfs_init(void);
519 extern int class_procfs_clean(void);
522 static int __init init_obdclass(void)
524 int init_obdclass(void)
529 int lustre_register_fs(void);
531 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
532 CFS_INIT_LIST_HEAD(&capa_list[i]);
535 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
537 spin_lock_init(&obd_types_lock);
538 obd_zombie_impexp_init();
540 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
541 LPROCFS_STATS_FLAG_NONE |
542 LPROCFS_STATS_FLAG_IRQ_SAFE);
543 if (obd_memory == NULL) {
544 CERROR("kmalloc of 'obd_memory' failed\n");
548 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
549 LPROCFS_CNTR_AVGMINMAX,
551 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
552 LPROCFS_CNTR_AVGMINMAX,
553 "pagesused", "pages");
555 err = obd_init_checks();
556 if (err == -EOVERFLOW)
559 class_init_uuidlist();
560 err = class_handle_init();
564 CFS_INIT_LIST_HEAD(&obd_types);
566 err = misc_register(&obd_psdev);
568 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
572 /* This struct is already zeroed for us (static global) */
573 for (i = 0; i < class_devno_max(); i++)
576 /* Default the dirty page cache cap to 1/2 of system memory.
577 * For clients with less memory, a larger fraction is needed
578 * for other purposes (mostly for BGL). */
579 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
580 obd_max_dirty_pages = totalram_pages / 4;
582 obd_max_dirty_pages = totalram_pages / 2;
584 err = obd_init_caches();
588 err = class_procfs_init();
593 err = lu_global_init();
597 err = lu_capainfo_init();
601 err = cl_global_init();
605 #if defined(__KERNEL__) && defined(HAVE_SERVER_SUPPORT)
606 err = dt_global_init();
610 err = lu_ucred_global_init();
615 err = llog_info_init();
620 err = lustre_register_fs();
626 void obd_update_maxusage(void)
630 max1 = obd_pages_sum();
631 max2 = obd_memory_sum();
633 spin_lock(&obd_updatemax_lock);
634 if (max1 > obd_max_pages)
635 obd_max_pages = max1;
636 if (max2 > obd_max_alloc)
637 obd_max_alloc = max2;
638 spin_unlock(&obd_updatemax_lock);
640 EXPORT_SYMBOL(obd_update_maxusage);
643 __u64 obd_memory_max(void)
647 spin_lock(&obd_updatemax_lock);
649 spin_unlock(&obd_updatemax_lock);
653 EXPORT_SYMBOL(obd_memory_max);
655 __u64 obd_pages_max(void)
659 spin_lock(&obd_updatemax_lock);
661 spin_unlock(&obd_updatemax_lock);
665 EXPORT_SYMBOL(obd_pages_max);
668 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
669 * ifdef to the end of the file to cover module and versioning goo.*/
671 static void cleanup_obdclass(void)
673 int lustre_unregister_fs(void);
674 __u64 memory_leaked, pages_leaked;
675 __u64 memory_max, pages_max;
678 lustre_unregister_fs();
680 misc_deregister(&obd_psdev);
682 #ifdef HAVE_SERVER_SUPPORT
683 lu_ucred_global_fini();
690 obd_cleanup_caches();
693 class_procfs_clean();
695 class_handle_cleanup();
696 class_exit_uuidlist();
697 obd_zombie_impexp_stop();
699 memory_leaked = obd_memory_sum();
700 pages_leaked = obd_pages_sum();
702 memory_max = obd_memory_max();
703 pages_max = obd_pages_max();
705 lprocfs_free_stats(&obd_memory);
706 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
707 "obd_memory max: "LPU64", leaked: "LPU64"\n",
708 memory_max, memory_leaked);
709 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
710 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
711 pages_max, pages_leaked);
716 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
717 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
718 MODULE_LICENSE("GPL");
720 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);