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 <obd_support.h>
41 #include <obd_class.h>
42 #include <lnet/lnetctl.h>
43 #include <lustre_debug.h>
44 #include <lprocfs_status.h>
45 #include <lustre/lustre_build_version.h>
46 #include <libcfs/list.h>
47 #include <cl_object.h>
48 #ifdef HAVE_SERVER_SUPPORT
49 # include <dt_object.h>
50 # include <md_object.h>
51 #endif /* HAVE_SERVER_SUPPORT */
52 #include <lustre_ioctl.h>
53 #include "llog_internal.h"
56 struct obd_device *obd_devs[MAX_OBD_DEVICES];
57 EXPORT_SYMBOL(obd_devs);
58 struct list_head obd_types;
59 DEFINE_RWLOCK(obd_dev_lock);
61 __u64 obd_max_pages = 0;
62 __u64 obd_max_alloc = 0;
64 static 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);
106 struct lprocfs_stats *obd_memory = NULL;
107 EXPORT_SYMBOL(obd_memory);
110 /* Get jobid of current process by reading the environment variable
111 * stored in between the "env_start" & "env_end" of task struct.
114 * It's better to cache the jobid for later use if there is any
115 * efficient way, the cl_env code probably could be reused for this
118 * If some job scheduler doesn't store jobid in the "env_start/end",
119 * then an upcall could be issued here to get the jobid by utilizing
120 * the userspace tools/api. Then, the jobid must be cached.
122 int lustre_get_jobid(char *jobid)
124 int jobid_len = JOBSTATS_JOBID_SIZE;
128 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
129 /* Jobstats isn't enabled */
130 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
133 /* Use process name + fsuid as jobid */
134 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
135 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
137 from_kuid(&init_user_ns, current_fsuid()));
141 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
143 if (rc == -EOVERFLOW) {
144 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
145 * variable length strings instead of just numbers), it
146 * might make sense to keep the unique parts for JobID,
147 * instead of just returning an error. That means a
148 * larger temp buffer for cfs_get_environ(), then
149 * truncating the string at some separator to fit into
150 * the specified jobid_len. Fix later if needed. */
152 if (unlikely(!printed)) {
153 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
154 "for JobID buffer (%d)\n",
155 obd_jobid_var, jobid_len);
159 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
160 rc == -EDEADLK) ? D_INFO : D_ERROR,
161 "Get jobid for (%s) failed: rc = %d\n",
167 EXPORT_SYMBOL(lustre_get_jobid);
169 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
170 size_t size, const char *file, int line)
173 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
174 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
175 ptr ? "force " :"", type, name, (__u64)size, file,
177 CERROR(LPU64" total bytes and "LPU64" total pages "
178 "("LPU64" bytes) allocated by Lustre, "
179 "%d total bytes by LNET\n",
181 obd_pages_sum() << PAGE_CACHE_SHIFT,
183 atomic_read(&libcfs_kmemory));
188 EXPORT_SYMBOL(obd_alloc_fail);
190 static int class_resolve_dev_name(__u32 len, const char *name)
197 CERROR("No name passed,!\n");
198 GOTO(out, rc = -EINVAL);
200 if (name[len - 1] != 0) {
201 CERROR("Name not nul terminated!\n");
202 GOTO(out, rc = -EINVAL);
205 CDEBUG(D_IOCTL, "device name %s\n", name);
206 dev = class_name2dev(name);
208 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
209 GOTO(out, rc = -EINVAL);
212 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
219 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
222 struct obd_ioctl_data *data;
223 struct libcfs_debug_ioctl_data *debug_data;
224 struct obd_device *obd = NULL;
225 int err = 0, len = 0;
228 /* only for debugging */
229 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
230 debug_data = (struct libcfs_debug_ioctl_data*)arg;
231 libcfs_subsystem_debug = debug_data->subs;
232 libcfs_debug = debug_data->debug;
236 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
237 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
238 CERROR("OBD ioctl: data error\n");
241 data = (struct obd_ioctl_data *)buf;
244 case OBD_IOC_PROCESS_CFG: {
245 struct lustre_cfg *lcfg;
247 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
248 CERROR("No config buffer passed!\n");
249 GOTO(out, err = -EINVAL);
251 OBD_ALLOC(lcfg, data->ioc_plen1);
253 GOTO(out, err = -ENOMEM);
254 err = copy_from_user(lcfg, data->ioc_pbuf1,
257 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
259 err = class_process_config(lcfg);
261 OBD_FREE(lcfg, data->ioc_plen1);
265 case OBD_GET_VERSION:
266 if (!data->ioc_inlbuf1) {
267 CERROR("No buffer passed in ioctl\n");
268 GOTO(out, err = -EINVAL);
271 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
272 CERROR("ioctl buffer too small to hold version\n");
273 GOTO(out, err = -EINVAL);
276 memcpy(data->ioc_bulk, BUILD_VERSION,
277 strlen(BUILD_VERSION) + 1);
279 err = obd_ioctl_popdata((void *)arg, data, len);
284 case OBD_IOC_NAME2DEV: {
285 /* Resolve a device name. This does not change the
286 * currently selected device.
290 dev = class_resolve_dev_name(data->ioc_inllen1,
294 GOTO(out, err = -EINVAL);
296 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
302 case OBD_IOC_UUID2DEV: {
303 /* Resolve a device uuid. This does not change the
304 * currently selected device.
307 struct obd_uuid uuid;
309 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
310 CERROR("No UUID passed!\n");
311 GOTO(out, err = -EINVAL);
313 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
314 CERROR("UUID not NUL terminated!\n");
315 GOTO(out, err = -EINVAL);
318 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
319 obd_str2uuid(&uuid, data->ioc_inlbuf1);
320 dev = class_uuid2dev(&uuid);
323 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
325 GOTO(out, err = -EINVAL);
328 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
330 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
336 case OBD_IOC_GETDEVICE: {
337 int index = data->ioc_count;
340 if (!data->ioc_inlbuf1) {
341 CERROR("No buffer passed in ioctl\n");
342 GOTO(out, err = -EINVAL);
344 if (data->ioc_inllen1 < 128) {
345 CERROR("ioctl buffer too small to hold version\n");
346 GOTO(out, err = -EINVAL);
349 obd = class_num2obd(index);
351 GOTO(out, err = -ENOENT);
353 if (obd->obd_stopping)
355 else if (obd->obd_set_up)
357 else if (obd->obd_attached)
361 str = (char *)data->ioc_bulk;
362 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
363 (int)index, status, obd->obd_type->typ_name,
364 obd->obd_name, obd->obd_uuid.uuid,
365 atomic_read(&obd->obd_refcount));
366 err = obd_ioctl_popdata((void *)arg, data, len);
373 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
374 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
375 GOTO(out, err = -EINVAL);
376 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
377 GOTO(out, err = -EINVAL);
378 obd = class_name2obd(data->ioc_inlbuf4);
379 } else if (data->ioc_dev < class_devno_max()) {
380 obd = class_num2obd(data->ioc_dev);
382 CERROR("OBD ioctl: No device\n");
383 GOTO(out, err = -EINVAL);
387 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
388 GOTO(out, err = -EINVAL);
390 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
392 if (!obd->obd_set_up || obd->obd_stopping) {
393 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
394 GOTO(out, err = -EINVAL);
398 case OBD_IOC_NO_TRANSNO: {
399 if (!obd->obd_attached) {
400 CERROR("Device %d not attached\n", obd->obd_minor);
401 GOTO(out, err = -ENODEV);
403 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
405 obd->obd_no_transno = 1;
410 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
414 err = obd_ioctl_popdata((void *)arg, data, len);
423 obd_ioctl_freedata(buf, len);
425 } /* class_handle_ioctl */
427 extern struct miscdevice obd_psdev;
429 #define OBD_INIT_CHECK
430 #ifdef OBD_INIT_CHECK
431 static int obd_init_checks(void)
433 __u64 u64val, div64val;
437 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
439 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
441 u64val = 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 len = snprintf(buf, sizeof(buf), LPX64, u64val);
450 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
454 div64val = OBD_OBJECT_EOF;
455 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
456 if (u64val != OBD_OBJECT_EOF) {
457 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
458 u64val, (int)sizeof(u64val));
461 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
462 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
463 u64val, (int)sizeof(u64val));
466 if (do_div(div64val, 256) != (u64val & 255)) {
467 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
470 if (u64val >> 8 != div64val) {
471 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
472 u64val, div64val, u64val >> 8);
475 len = snprintf(buf, sizeof(buf), LPX64, u64val);
477 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
480 len = snprintf(buf, sizeof(buf), LPU64, u64val);
482 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
485 len = snprintf(buf, sizeof(buf), LPD64, u64val);
487 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
490 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
491 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
492 (__u64)PAGE_CACHE_SIZE);
499 #define obd_init_checks() do {} while(0)
502 extern int class_procfs_init(void);
503 extern int class_procfs_clean(void);
505 static int __init init_obdclass(void)
508 int lustre_register_fs(void);
510 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
511 INIT_LIST_HEAD(&capa_list[i]);
513 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
515 spin_lock_init(&obd_types_lock);
516 obd_zombie_impexp_init();
518 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
519 LPROCFS_STATS_FLAG_NONE |
520 LPROCFS_STATS_FLAG_IRQ_SAFE);
521 if (obd_memory == NULL) {
522 CERROR("kmalloc of 'obd_memory' failed\n");
526 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
527 LPROCFS_CNTR_AVGMINMAX,
529 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
530 LPROCFS_CNTR_AVGMINMAX,
531 "pagesused", "pages");
533 err = obd_init_checks();
534 if (err == -EOVERFLOW)
537 class_init_uuidlist();
538 err = class_handle_init();
542 INIT_LIST_HEAD(&obd_types);
544 err = misc_register(&obd_psdev);
546 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
550 /* This struct is already zeroed for us (static global) */
551 for (i = 0; i < class_devno_max(); i++)
554 /* Default the dirty page cache cap to 1/2 of system memory.
555 * For clients with less memory, a larger fraction is needed
556 * for other purposes (mostly for BGL). */
557 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
558 obd_max_dirty_pages = totalram_pages / 4;
560 obd_max_dirty_pages = totalram_pages / 2;
562 err = obd_init_caches();
565 err = class_procfs_init();
569 err = lu_global_init();
573 err = lu_capainfo_init();
577 err = cl_global_init();
581 #ifdef HAVE_SERVER_SUPPORT
582 err = dt_global_init();
586 err = lu_ucred_global_init();
589 #endif /* HAVE_SERVER_SUPPORT */
591 err = llog_info_init();
595 err = lustre_register_fs();
600 void obd_update_maxusage(void)
604 max1 = obd_pages_sum();
605 max2 = obd_memory_sum();
607 spin_lock(&obd_updatemax_lock);
608 if (max1 > obd_max_pages)
609 obd_max_pages = max1;
610 if (max2 > obd_max_alloc)
611 obd_max_alloc = max2;
612 spin_unlock(&obd_updatemax_lock);
614 EXPORT_SYMBOL(obd_update_maxusage);
617 __u64 obd_memory_max(void)
621 spin_lock(&obd_updatemax_lock);
623 spin_unlock(&obd_updatemax_lock);
627 EXPORT_SYMBOL(obd_memory_max);
629 __u64 obd_pages_max(void)
633 spin_lock(&obd_updatemax_lock);
635 spin_unlock(&obd_updatemax_lock);
639 EXPORT_SYMBOL(obd_pages_max);
642 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
643 * ifdef to the end of the file to cover module and versioning goo.*/
644 static void cleanup_obdclass(void)
646 int lustre_unregister_fs(void);
647 __u64 memory_leaked, pages_leaked;
648 __u64 memory_max, pages_max;
651 lustre_unregister_fs();
653 misc_deregister(&obd_psdev);
655 #ifdef HAVE_SERVER_SUPPORT
656 lu_ucred_global_fini();
658 #endif /* HAVE_SERVER_SUPPORT */
663 obd_cleanup_caches();
666 class_procfs_clean();
668 class_handle_cleanup();
669 class_exit_uuidlist();
670 obd_zombie_impexp_stop();
672 memory_leaked = obd_memory_sum();
673 pages_leaked = obd_pages_sum();
675 memory_max = obd_memory_max();
676 pages_max = obd_pages_max();
678 lprocfs_free_stats(&obd_memory);
679 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
680 "obd_memory max: "LPU64", leaked: "LPU64"\n",
681 memory_max, memory_leaked);
682 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
683 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
684 pages_max, pages_leaked);
689 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
690 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
691 MODULE_LICENSE("GPL");
693 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);