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, 2015, 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 <linux/user_namespace.h>
40 #ifdef HAVE_UIDGID_HEADER
41 # include <linux/uidgid.h>
43 #include <linux/atomic.h>
45 #include <obd_support.h>
46 #include <obd_class.h>
47 #include <lnet/lnetctl.h>
48 #include <lustre_debug.h>
49 #include <lprocfs_status.h>
50 #include <lustre/lustre_build_version.h>
51 #include <libcfs/list.h>
52 #include <cl_object.h>
53 #ifdef HAVE_SERVER_SUPPORT
54 # include <dt_object.h>
55 # include <md_object.h>
56 #endif /* HAVE_SERVER_SUPPORT */
57 #include <lustre_ioctl.h>
58 #include "llog_internal.h"
60 struct obd_device *obd_devs[MAX_OBD_DEVICES];
61 struct list_head obd_types;
62 DEFINE_RWLOCK(obd_dev_lock);
65 static __u64 obd_max_alloc;
70 static DEFINE_SPINLOCK(obd_updatemax_lock);
72 /* The following are visible and mutable through /proc/sys/lustre/. */
73 unsigned int obd_debug_peer_on_timeout;
74 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
75 unsigned int obd_dump_on_timeout;
76 EXPORT_SYMBOL(obd_dump_on_timeout);
77 unsigned int obd_dump_on_eviction;
78 EXPORT_SYMBOL(obd_dump_on_eviction);
79 unsigned long obd_max_dirty_pages;
80 EXPORT_SYMBOL(obd_max_dirty_pages);
81 atomic_long_t obd_dirty_pages;
82 EXPORT_SYMBOL(obd_dirty_pages);
83 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
84 EXPORT_SYMBOL(obd_timeout);
85 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
86 EXPORT_SYMBOL(ldlm_timeout);
87 unsigned int obd_timeout_set;
88 EXPORT_SYMBOL(obd_timeout_set);
89 unsigned int ldlm_timeout_set;
90 EXPORT_SYMBOL(ldlm_timeout_set);
91 /* bulk transfer timeout, give up after 100s by default */
92 unsigned int bulk_timeout = 100; /* seconds */
93 EXPORT_SYMBOL(bulk_timeout);
94 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
95 unsigned int at_min = 0;
96 EXPORT_SYMBOL(at_min);
97 unsigned int at_max = 600;
98 EXPORT_SYMBOL(at_max);
99 unsigned int at_history = 600;
100 EXPORT_SYMBOL(at_history);
101 int at_early_margin = 5;
102 EXPORT_SYMBOL(at_early_margin);
104 EXPORT_SYMBOL(at_extra);
106 atomic_long_t obd_dirty_transit_pages;
107 EXPORT_SYMBOL(obd_dirty_transit_pages);
109 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
111 #ifdef CONFIG_PROC_FS
112 struct lprocfs_stats *obd_memory = NULL;
113 EXPORT_SYMBOL(obd_memory);
116 char obd_jobid_node[LUSTRE_JOBID_SIZE + 1];
118 /* Get jobid of current process by reading the environment variable
119 * stored in between the "env_start" & "env_end" of task struct.
122 * It's better to cache the jobid for later use if there is any
123 * efficient way, the cl_env code probably could be reused for this
126 * If some job scheduler doesn't store jobid in the "env_start/end",
127 * then an upcall could be issued here to get the jobid by utilizing
128 * the userspace tools/api. Then, the jobid must be cached.
130 int lustre_get_jobid(char *jobid)
132 int jobid_len = LUSTRE_JOBID_SIZE;
136 memset(jobid, 0, LUSTRE_JOBID_SIZE);
137 /* Jobstats isn't enabled */
138 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
141 /* Whole node dedicated to single job */
142 if (strcmp(obd_jobid_var, JOBSTATS_NODELOCAL) == 0) {
143 memcpy(jobid, obd_jobid_node, LUSTRE_JOBID_SIZE);
147 /* Use process name + fsuid as jobid */
148 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
149 snprintf(jobid, LUSTRE_JOBID_SIZE, "%s.%u",
151 from_kuid(&init_user_ns, current_fsuid()));
155 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
157 if (rc == -EOVERFLOW) {
158 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
159 * variable length strings instead of just numbers), it
160 * might make sense to keep the unique parts for JobID,
161 * instead of just returning an error. That means a
162 * larger temp buffer for cfs_get_environ(), then
163 * truncating the string at some separator to fit into
164 * the specified jobid_len. Fix later if needed. */
166 if (unlikely(!printed)) {
167 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
168 "for JobID buffer (%d)\n",
169 obd_jobid_var, jobid_len);
173 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
174 rc == -EDEADLK) ? D_INFO : D_ERROR,
175 "Get jobid for (%s) failed: rc = %d\n",
181 EXPORT_SYMBOL(lustre_get_jobid);
183 static int class_resolve_dev_name(__u32 len, const char *name)
190 CERROR("No name passed,!\n");
191 GOTO(out, rc = -EINVAL);
193 if (name[len - 1] != 0) {
194 CERROR("Name not nul terminated!\n");
195 GOTO(out, rc = -EINVAL);
198 CDEBUG(D_IOCTL, "device name %s\n", name);
199 dev = class_name2dev(name);
201 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
202 GOTO(out, rc = -EINVAL);
205 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
212 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
215 struct obd_ioctl_data *data;
216 struct libcfs_debug_ioctl_data *debug_data;
217 struct obd_device *obd = NULL;
218 int err = 0, len = 0;
221 /* only for debugging */
222 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
223 debug_data = (struct libcfs_debug_ioctl_data*)arg;
224 libcfs_subsystem_debug = debug_data->subs;
225 libcfs_debug = debug_data->debug;
229 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
230 if (obd_ioctl_getdata(&buf, &len, (void __user *)arg)) {
231 CERROR("OBD ioctl: data error\n");
234 data = (struct obd_ioctl_data *)buf;
237 case OBD_IOC_PROCESS_CFG: {
238 struct lustre_cfg *lcfg;
240 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
241 CERROR("No config buffer passed!\n");
242 GOTO(out, err = -EINVAL);
244 OBD_ALLOC(lcfg, data->ioc_plen1);
246 GOTO(out, err = -ENOMEM);
247 err = copy_from_user(lcfg, data->ioc_pbuf1,
250 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
252 err = class_process_config(lcfg);
254 OBD_FREE(lcfg, data->ioc_plen1);
258 case OBD_GET_VERSION:
259 if (!data->ioc_inlbuf1) {
260 CERROR("No buffer passed in ioctl\n");
261 GOTO(out, err = -EINVAL);
264 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
265 CERROR("ioctl buffer too small to hold version\n");
266 GOTO(out, err = -EINVAL);
269 memcpy(data->ioc_bulk, BUILD_VERSION,
270 strlen(BUILD_VERSION) + 1);
272 err = obd_ioctl_popdata((void __user *)arg, data, len);
277 case OBD_IOC_NAME2DEV: {
278 /* Resolve a device name. This does not change the
279 * currently selected device.
283 dev = class_resolve_dev_name(data->ioc_inllen1,
287 GOTO(out, err = -EINVAL);
289 err = obd_ioctl_popdata((void __user *)arg, data,
296 case OBD_IOC_UUID2DEV: {
297 /* Resolve a device uuid. This does not change the
298 * currently selected device.
301 struct obd_uuid uuid;
303 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
304 CERROR("No UUID passed!\n");
305 GOTO(out, err = -EINVAL);
307 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
308 CERROR("UUID not NUL terminated!\n");
309 GOTO(out, err = -EINVAL);
312 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
313 obd_str2uuid(&uuid, data->ioc_inlbuf1);
314 dev = class_uuid2dev(&uuid);
317 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
319 GOTO(out, err = -EINVAL);
322 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
324 err = obd_ioctl_popdata((void __user *)arg, data,
331 case OBD_IOC_GETDEVICE: {
332 int index = data->ioc_count;
335 if (!data->ioc_inlbuf1) {
336 CERROR("No buffer passed in ioctl\n");
337 GOTO(out, err = -EINVAL);
339 if (data->ioc_inllen1 < 128) {
340 CERROR("ioctl buffer too small to hold version\n");
341 GOTO(out, err = -EINVAL);
344 obd = class_num2obd(index);
346 GOTO(out, err = -ENOENT);
348 if (obd->obd_stopping)
350 else if (obd->obd_set_up)
352 else if (obd->obd_attached)
356 str = (char *)data->ioc_bulk;
357 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
358 (int)index, status, obd->obd_type->typ_name,
359 obd->obd_name, obd->obd_uuid.uuid,
360 atomic_read(&obd->obd_refcount));
361 err = obd_ioctl_popdata((void __user *)arg, data, len);
368 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
369 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
370 GOTO(out, err = -EINVAL);
371 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
372 GOTO(out, err = -EINVAL);
373 obd = class_name2obd(data->ioc_inlbuf4);
374 } else if (data->ioc_dev < class_devno_max()) {
375 obd = class_num2obd(data->ioc_dev);
377 CERROR("OBD ioctl: No device\n");
378 GOTO(out, err = -EINVAL);
382 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
383 GOTO(out, err = -EINVAL);
385 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
387 if (!obd->obd_set_up || obd->obd_stopping) {
388 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
389 GOTO(out, err = -EINVAL);
393 case OBD_IOC_NO_TRANSNO: {
394 if (!obd->obd_attached) {
395 CERROR("Device %d not attached\n", obd->obd_minor);
396 GOTO(out, err = -ENODEV);
398 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
400 obd->obd_no_transno = 1;
405 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
409 err = obd_ioctl_popdata((void __user *)arg, data, len);
418 obd_ioctl_freedata(buf, len);
420 } /* class_handle_ioctl */
422 #define OBD_INIT_CHECK
423 #ifdef OBD_INIT_CHECK
424 static int obd_init_checks(void)
426 __u64 u64val, div64val;
430 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
432 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
434 u64val = OBD_OBJECT_EOF;
435 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
436 if (u64val != OBD_OBJECT_EOF) {
437 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
438 u64val, (int)sizeof(u64val));
441 len = snprintf(buf, sizeof(buf), LPX64, u64val);
443 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
447 div64val = OBD_OBJECT_EOF;
448 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
449 if (u64val != OBD_OBJECT_EOF) {
450 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
451 u64val, (int)sizeof(u64val));
454 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
455 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
456 u64val, (int)sizeof(u64val));
459 if (do_div(div64val, 256) != (u64val & 255)) {
460 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
463 if (u64val >> 8 != div64val) {
464 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
465 u64val, div64val, u64val >> 8);
468 len = snprintf(buf, sizeof(buf), LPX64, u64val);
470 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
473 len = snprintf(buf, sizeof(buf), LPU64, u64val);
475 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
478 len = snprintf(buf, sizeof(buf), LPD64, u64val);
480 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
483 if ((u64val & ~PAGE_CACHE_MASK) >= PAGE_CACHE_SIZE) {
484 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
485 (__u64)PAGE_CACHE_SIZE);
492 #define obd_init_checks() do {} while(0)
495 static int __init obdclass_init(void)
499 spin_lock_init(&obd_stale_export_lock);
500 INIT_LIST_HEAD(&obd_stale_exports);
501 atomic_set(&obd_stale_export_num, 0);
503 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
505 spin_lock_init(&obd_types_lock);
506 obd_zombie_impexp_init();
507 #ifdef CONFIG_PROC_FS
508 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
509 LPROCFS_STATS_FLAG_NONE |
510 LPROCFS_STATS_FLAG_IRQ_SAFE);
511 if (obd_memory == NULL) {
512 CERROR("kmalloc of 'obd_memory' failed\n");
516 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
517 LPROCFS_CNTR_AVGMINMAX,
520 err = obd_init_checks();
521 if (err == -EOVERFLOW)
524 class_init_uuidlist();
525 err = class_handle_init();
529 INIT_LIST_HEAD(&obd_types);
531 err = misc_register(&obd_psdev);
533 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
537 /* This struct is already zeroed for us (static global) */
538 for (i = 0; i < class_devno_max(); i++)
541 /* Default the dirty page cache cap to 1/2 of system memory.
542 * For clients with less memory, a larger fraction is needed
543 * for other purposes (mostly for BGL). */
544 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
545 obd_max_dirty_pages = totalram_pages / 4;
547 obd_max_dirty_pages = totalram_pages / 2;
549 err = obd_init_caches();
552 err = class_procfs_init();
556 err = lu_global_init();
560 err = cl_global_init();
564 #ifdef HAVE_SERVER_SUPPORT
565 err = dt_global_init();
569 err = lu_ucred_global_init();
572 #endif /* HAVE_SERVER_SUPPORT */
574 err = llog_info_init();
578 err = lustre_register_fs();
583 void obd_update_maxusage(void)
587 max = obd_memory_sum();
589 spin_lock(&obd_updatemax_lock);
590 if (max > obd_max_alloc)
592 spin_unlock(&obd_updatemax_lock);
594 EXPORT_SYMBOL(obd_update_maxusage);
596 #ifdef CONFIG_PROC_FS
597 __u64 obd_memory_max(void)
601 obd_update_maxusage();
602 spin_lock(&obd_updatemax_lock);
604 spin_unlock(&obd_updatemax_lock);
608 #endif /* CONFIG_PROC_FS */
610 static void __exit obdclass_exit(void)
616 lustre_unregister_fs();
618 misc_deregister(&obd_psdev);
620 #ifdef HAVE_SERVER_SUPPORT
621 lu_ucred_global_fini();
623 #endif /* HAVE_SERVER_SUPPORT */
627 obd_cleanup_caches();
630 class_procfs_clean();
632 class_handle_cleanup();
633 class_exit_uuidlist();
634 obd_zombie_impexp_stop();
635 LASSERT(list_empty(&obd_stale_exports));
637 memory_leaked = obd_memory_sum();
639 memory_max = obd_memory_max();
641 lprocfs_free_stats(&obd_memory);
642 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
643 "obd_memory max: "LPU64", leaked: "LPU64"\n",
644 memory_max, memory_leaked);
649 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
650 MODULE_DESCRIPTION("Lustre Class Driver");
651 MODULE_VERSION(LUSTRE_VERSION_STRING);
652 MODULE_LICENSE("GPL");
654 module_init(obdclass_init);
655 module_exit(obdclass_exit);