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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_CLASS
35 #include <linux/miscdevice.h>
36 #include <linux/user_namespace.h>
37 #ifdef HAVE_UIDGID_HEADER
38 # include <linux/uidgid.h>
40 #include <linux/atomic.h>
41 #include <linux/list.h>
43 #include <obd_support.h>
44 #include <obd_class.h>
45 #include <lnet/lnetctl.h>
46 #include <lustre_debug.h>
47 #include <lustre_kernelcomm.h>
48 #include <lprocfs_status.h>
49 #include <lustre_ver.h>
50 #include <cl_object.h>
51 #ifdef HAVE_SERVER_SUPPORT
52 # include <dt_object.h>
53 # include <md_object.h>
54 #endif /* HAVE_SERVER_SUPPORT */
55 #include <uapi/linux/lustre_ioctl.h>
56 #include "llog_internal.h"
59 static __u64 obd_max_alloc;
64 static DEFINE_SPINLOCK(obd_updatemax_lock);
66 /* The following are visible and mutable through /proc/sys/lustre/. */
67 unsigned int obd_debug_peer_on_timeout;
68 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
69 unsigned int obd_dump_on_timeout;
70 EXPORT_SYMBOL(obd_dump_on_timeout);
71 unsigned int obd_dump_on_eviction;
72 EXPORT_SYMBOL(obd_dump_on_eviction);
73 unsigned long obd_max_dirty_pages;
74 EXPORT_SYMBOL(obd_max_dirty_pages);
75 atomic_long_t obd_dirty_pages;
76 EXPORT_SYMBOL(obd_dirty_pages);
77 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
78 EXPORT_SYMBOL(obd_timeout);
79 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
80 EXPORT_SYMBOL(ldlm_timeout);
81 unsigned int obd_timeout_set;
82 EXPORT_SYMBOL(obd_timeout_set);
83 unsigned int ldlm_timeout_set;
84 EXPORT_SYMBOL(ldlm_timeout_set);
85 /* bulk transfer timeout, give up after 100s by default */
86 unsigned int bulk_timeout = 100; /* seconds */
87 EXPORT_SYMBOL(bulk_timeout);
88 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
89 unsigned int at_min = 0;
90 EXPORT_SYMBOL(at_min);
91 unsigned int at_max = 600;
92 EXPORT_SYMBOL(at_max);
93 unsigned int at_history = 600;
94 EXPORT_SYMBOL(at_history);
95 int at_early_margin = 5;
96 EXPORT_SYMBOL(at_early_margin);
98 EXPORT_SYMBOL(at_extra);
100 atomic_long_t obd_dirty_transit_pages;
101 EXPORT_SYMBOL(obd_dirty_transit_pages);
103 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
105 #ifdef CONFIG_PROC_FS
106 struct lprocfs_stats *obd_memory = NULL;
107 EXPORT_SYMBOL(obd_memory);
110 char obd_jobid_node[LUSTRE_JOBID_SIZE + 1];
112 /* Get jobid of current process by reading the environment variable
113 * stored in between the "env_start" & "env_end" of task struct.
116 * It's better to cache the jobid for later use if there is any
117 * efficient way, the cl_env code probably could be reused for this
120 * If some job scheduler doesn't store jobid in the "env_start/end",
121 * then an upcall could be issued here to get the jobid by utilizing
122 * the userspace tools/api. Then, the jobid must be cached.
124 int lustre_get_jobid(char *jobid)
126 int jobid_len = LUSTRE_JOBID_SIZE;
127 char tmp_jobid[LUSTRE_JOBID_SIZE] = { 0 };
131 /* Jobstats isn't enabled */
132 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
135 /* Whole node dedicated to single job */
136 if (strcmp(obd_jobid_var, JOBSTATS_NODELOCAL) == 0) {
137 memcpy(tmp_jobid, obd_jobid_node, LUSTRE_JOBID_SIZE);
141 /* Use process name + fsuid as jobid */
142 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
143 snprintf(tmp_jobid, LUSTRE_JOBID_SIZE, "%s.%u",
145 from_kuid(&init_user_ns, current_fsuid()));
149 rc = cfs_get_environ(obd_jobid_var, tmp_jobid, &jobid_len);
151 if (rc == -EOVERFLOW) {
152 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
153 * variable length strings instead of just numbers), it
154 * might make sense to keep the unique parts for JobID,
155 * instead of just returning an error. That means a
156 * larger temp buffer for cfs_get_environ(), then
157 * truncating the string at some separator to fit into
158 * the specified jobid_len. Fix later if needed. */
160 if (unlikely(!printed)) {
161 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
162 "for JobID buffer (%d)\n",
163 obd_jobid_var, jobid_len);
167 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
168 rc == -EDEADLK) ? D_INFO : D_ERROR,
169 "Get jobid for (%s) failed: rc = %d\n",
178 /* Only replace the job ID if it changed. */
179 if (strcmp(jobid, tmp_jobid) != 0)
180 memcpy(jobid, tmp_jobid, jobid_len);
184 EXPORT_SYMBOL(lustre_get_jobid);
186 static int class_resolve_dev_name(__u32 len, const char *name)
193 CERROR("No name passed,!\n");
194 GOTO(out, rc = -EINVAL);
196 if (name[len - 1] != 0) {
197 CERROR("Name not nul terminated!\n");
198 GOTO(out, rc = -EINVAL);
201 CDEBUG(D_IOCTL, "device name %s\n", name);
202 dev = class_name2dev(name);
204 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
205 GOTO(out, rc = -EINVAL);
208 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
215 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
218 struct obd_ioctl_data *data;
219 struct libcfs_debug_ioctl_data *debug_data;
220 struct obd_device *obd = NULL;
221 int err = 0, len = 0;
224 /* only for debugging */
225 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
226 debug_data = (struct libcfs_debug_ioctl_data*)arg;
227 libcfs_subsystem_debug = debug_data->subs;
228 libcfs_debug = debug_data->debug;
232 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
233 if (obd_ioctl_getdata(&buf, &len, (void __user *)arg)) {
234 CERROR("OBD ioctl: data error\n");
237 data = (struct obd_ioctl_data *)buf;
240 case OBD_IOC_PROCESS_CFG: {
241 struct lustre_cfg *lcfg;
243 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
244 CERROR("No config buffer passed!\n");
245 GOTO(out, err = -EINVAL);
247 OBD_ALLOC(lcfg, data->ioc_plen1);
249 GOTO(out, err = -ENOMEM);
250 err = copy_from_user(lcfg, data->ioc_pbuf1,
253 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
255 err = class_process_config(lcfg);
257 OBD_FREE(lcfg, data->ioc_plen1);
261 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
262 case OBD_GET_VERSION: {
265 if (!data->ioc_inlbuf1) {
266 CERROR("No buffer passed in ioctl\n");
267 GOTO(out, err = -EINVAL);
270 if (strlen(LUSTRE_VERSION_STRING) + 1 > data->ioc_inllen1) {
271 CERROR("ioctl buffer too small to hold version\n");
272 GOTO(out, err = -EINVAL);
277 CWARN("%s: ioctl(OBD_GET_VERSION) is deprecated, "
278 "use llapi_get_version_string() and/or relink\n",
281 memcpy(data->ioc_bulk, LUSTRE_VERSION_STRING,
282 strlen(LUSTRE_VERSION_STRING) + 1);
284 if (copy_to_user((void __user *)arg, data, len))
289 case OBD_IOC_NAME2DEV: {
290 /* Resolve a device name. This does not change the
291 * currently selected device.
295 dev = class_resolve_dev_name(data->ioc_inllen1,
299 GOTO(out, err = -EINVAL);
301 if (copy_to_user((void __user *)arg, data, sizeof(*data)))
306 case OBD_IOC_UUID2DEV: {
307 /* Resolve a device uuid. This does not change the
308 * currently selected device.
311 struct obd_uuid uuid;
313 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
314 CERROR("No UUID passed!\n");
315 GOTO(out, err = -EINVAL);
317 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
318 CERROR("UUID not NUL terminated!\n");
319 GOTO(out, err = -EINVAL);
322 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
323 obd_str2uuid(&uuid, data->ioc_inlbuf1);
324 dev = class_uuid2dev(&uuid);
327 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
329 GOTO(out, err = -EINVAL);
332 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
334 if (copy_to_user((void __user *)arg, data, sizeof(*data)))
339 case OBD_IOC_GETDEVICE: {
340 int index = data->ioc_count;
343 if (!data->ioc_inlbuf1) {
344 CERROR("No buffer passed in ioctl\n");
345 GOTO(out, err = -EINVAL);
347 if (data->ioc_inllen1 < 128) {
348 CERROR("ioctl buffer too small to hold version\n");
349 GOTO(out, err = -EINVAL);
352 obd = class_num2obd(index);
354 GOTO(out, err = -ENOENT);
356 if (obd->obd_stopping)
358 else if (obd->obd_set_up)
360 else if (obd->obd_attached)
364 str = (char *)data->ioc_bulk;
365 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
366 (int)index, status, obd->obd_type->typ_name,
367 obd->obd_name, obd->obd_uuid.uuid,
368 atomic_read(&obd->obd_refcount));
370 if (copy_to_user((void __user *)arg, data, len))
378 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
379 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
380 GOTO(out, err = -EINVAL);
381 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
382 GOTO(out, err = -EINVAL);
383 obd = class_name2obd(data->ioc_inlbuf4);
384 } else if (data->ioc_dev < class_devno_max()) {
385 obd = class_num2obd(data->ioc_dev);
387 CERROR("OBD ioctl: No device\n");
388 GOTO(out, err = -EINVAL);
392 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
393 GOTO(out, err = -EINVAL);
395 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
397 if (!obd->obd_set_up || obd->obd_stopping) {
398 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
399 GOTO(out, err = -EINVAL);
403 case OBD_IOC_NO_TRANSNO: {
404 if (!obd->obd_attached) {
405 CERROR("Device %d not attached\n", obd->obd_minor);
406 GOTO(out, err = -ENODEV);
408 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
410 obd->obd_no_transno = 1;
415 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
419 if (copy_to_user((void __user *)arg, data, len))
426 OBD_FREE_LARGE(buf, len);
428 } /* class_handle_ioctl */
430 #define OBD_INIT_CHECK
431 #ifdef OBD_INIT_CHECK
432 static int obd_init_checks(void)
434 __u64 u64val, div64val;
438 CDEBUG(D_INFO, "OBD_OBJECT_EOF = %#llx\n", (__u64)OBD_OBJECT_EOF);
440 u64val = OBD_OBJECT_EOF;
441 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = %#llx\n", u64val);
442 if (u64val != OBD_OBJECT_EOF) {
443 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
444 u64val, (int)sizeof(u64val));
447 len = snprintf(buf, sizeof(buf), "%#llx", u64val);
449 CWARN("u64 hex wrong length! strlen(%s)=%d != 18\n", buf, len);
453 div64val = OBD_OBJECT_EOF;
454 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = %#llx\n", u64val);
455 if (u64val != OBD_OBJECT_EOF) {
456 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
457 u64val, (int)sizeof(u64val));
460 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
461 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
462 u64val, (int)sizeof(u64val));
465 if (do_div(div64val, 256) != (u64val & 255)) {
466 CERROR("do_div(%#llx,256) != %llu\n", u64val, u64val & 255);
469 if (u64val >> 8 != div64val) {
470 CERROR("do_div(%#llx,256) %llu != %llu\n",
471 u64val, div64val, u64val >> 8);
474 len = snprintf(buf, sizeof(buf), "%#llx", u64val);
476 CWARN("u64 hex wrong length! strlen(%s)=%d != 18\n", buf, len);
479 len = snprintf(buf, sizeof(buf), "%llu", u64val);
481 CWARN("u64 wrong length! strlen(%s)=%d != 20\n", buf, len);
484 len = snprintf(buf, sizeof(buf), "%lld", u64val);
486 CWARN("s64 wrong length! strlen(%s)=%d != 2\n", buf, len);
489 if ((u64val & ~PAGE_MASK) >= PAGE_SIZE) {
490 CWARN("mask failed: u64val %llu >= %llu\n", u64val,
498 #define obd_init_checks() do {} while(0)
501 static int __init obdclass_init(void)
505 LCONSOLE_INFO("Lustre: Build Version: "LUSTRE_VERSION_STRING"\n");
509 err = obd_init_checks();
510 if (err == -EOVERFLOW)
513 #ifdef CONFIG_PROC_FS
514 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
515 LPROCFS_STATS_FLAG_NONE |
516 LPROCFS_STATS_FLAG_IRQ_SAFE);
517 if (obd_memory == NULL) {
518 CERROR("kmalloc of 'obd_memory' failed\n");
522 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
523 LPROCFS_CNTR_AVGMINMAX,
526 err = obd_zombie_impexp_init();
528 goto cleanup_obd_memory;
530 err = class_handle_init();
532 goto cleanup_zombie_impexp;
534 err = misc_register(&obd_psdev);
536 CERROR("cannot register OBD miscdevice: err = %d\n", err);
537 goto cleanup_class_handle;
540 /* Default the dirty page cache cap to 1/2 of system memory.
541 * For clients with less memory, a larger fraction is needed
542 * for other purposes (mostly for BGL). */
543 if (totalram_pages <= 512 << (20 - PAGE_SHIFT))
544 obd_max_dirty_pages = totalram_pages / 4;
546 obd_max_dirty_pages = totalram_pages / 2;
548 err = obd_init_caches();
550 goto cleanup_deregister;
552 err = class_procfs_init();
556 err = lu_global_init();
558 goto cleanup_class_procfs;
560 err = cl_global_init();
562 goto cleanup_lu_global;
564 #ifdef HAVE_SERVER_SUPPORT
565 err = dt_global_init();
567 goto cleanup_cl_global;
569 err = lu_ucred_global_init();
571 goto cleanup_dt_global;
572 #endif /* HAVE_SERVER_SUPPORT */
574 err = llog_info_init();
576 #ifdef HAVE_SERVER_SUPPORT
577 goto cleanup_lu_ucred_global;
578 #else /* !HAVE_SERVER_SUPPORT */
579 goto cleanup_cl_global;
580 #endif /* HAVE_SERVER_SUPPORT */
582 err = lustre_register_fs();
584 /* simulate a late OOM situation now to require all
585 * alloc'ed/initialized resources to be freed */
586 if (OBD_FAIL_CHECK(OBD_FAIL_OBDCLASS_MODULE_LOAD)) {
587 /* fake error but filesystem has been registered */
588 lustre_unregister_fs();
589 /* force error to ensure module will be unloaded/cleaned */
594 goto cleanup_llog_info;
601 #ifdef HAVE_SERVER_SUPPORT
602 cleanup_lu_ucred_global:
603 lu_ucred_global_fini();
607 #endif /* HAVE_SERVER_SUPPORT */
615 cleanup_class_procfs:
617 class_procfs_clean();
620 obd_cleanup_caches();
623 misc_deregister(&obd_psdev);
625 cleanup_class_handle:
626 class_handle_cleanup();
628 cleanup_zombie_impexp:
629 obd_zombie_impexp_stop();
632 #ifdef CONFIG_PROC_FS
633 lprocfs_free_stats(&obd_memory);
639 void obd_update_maxusage(void)
643 max = obd_memory_sum();
645 spin_lock(&obd_updatemax_lock);
646 if (max > obd_max_alloc)
648 spin_unlock(&obd_updatemax_lock);
650 EXPORT_SYMBOL(obd_update_maxusage);
652 #ifdef CONFIG_PROC_FS
653 __u64 obd_memory_max(void)
657 obd_update_maxusage();
658 spin_lock(&obd_updatemax_lock);
660 spin_unlock(&obd_updatemax_lock);
664 #endif /* CONFIG_PROC_FS */
666 static void __exit obdclass_exit(void)
668 #ifdef CONFIG_PROC_FS
671 #endif /* CONFIG_PROC_FS */
674 lustre_unregister_fs();
676 misc_deregister(&obd_psdev);
678 #ifdef HAVE_SERVER_SUPPORT
679 lu_ucred_global_fini();
681 #endif /* HAVE_SERVER_SUPPORT */
685 obd_cleanup_caches();
688 class_procfs_clean();
690 class_handle_cleanup();
691 class_del_uuid(NULL); /* Delete all UUIDs. */
692 obd_zombie_impexp_stop();
694 #ifdef CONFIG_PROC_FS
695 memory_leaked = obd_memory_sum();
696 memory_max = obd_memory_max();
698 lprocfs_free_stats(&obd_memory);
699 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
700 "obd_memory max: %llu, leaked: %llu\n",
701 memory_max, memory_leaked);
702 #endif /* CONFIG_PROC_FS */
707 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
708 MODULE_DESCRIPTION("Lustre Class Driver");
709 MODULE_VERSION(LUSTRE_VERSION_STRING);
710 MODULE_LICENSE("GPL");
712 module_init(obdclass_init);
713 module_exit(obdclass_exit);