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/user_namespace.h>
36 #ifdef HAVE_UIDGID_HEADER
37 # include <linux/uidgid.h>
39 #include <linux/atomic.h>
40 #include <linux/list.h>
42 #include <obd_support.h>
43 #include <obd_class.h>
44 #include <lnet/lnetctl.h>
45 #include <lustre_debug.h>
46 #include <lprocfs_status.h>
47 #include <lustre_ver.h>
48 #include <cl_object.h>
49 #ifdef HAVE_SERVER_SUPPORT
50 # include <dt_object.h>
51 # include <md_object.h>
52 #endif /* HAVE_SERVER_SUPPORT */
53 #include <lustre_ioctl.h>
54 #include "llog_internal.h"
56 struct obd_device *obd_devs[MAX_OBD_DEVICES];
57 struct list_head obd_types;
58 DEFINE_RWLOCK(obd_dev_lock);
61 static __u64 obd_max_alloc;
66 static DEFINE_SPINLOCK(obd_updatemax_lock);
68 /* The following are visible and mutable through /proc/sys/lustre/. */
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 long obd_max_dirty_pages;
76 EXPORT_SYMBOL(obd_max_dirty_pages);
77 atomic_long_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 /* bulk transfer timeout, give up after 100s by default */
88 unsigned int bulk_timeout = 100; /* seconds */
89 EXPORT_SYMBOL(bulk_timeout);
90 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
91 unsigned int at_min = 0;
92 EXPORT_SYMBOL(at_min);
93 unsigned int at_max = 600;
94 EXPORT_SYMBOL(at_max);
95 unsigned int at_history = 600;
96 EXPORT_SYMBOL(at_history);
97 int at_early_margin = 5;
98 EXPORT_SYMBOL(at_early_margin);
100 EXPORT_SYMBOL(at_extra);
102 atomic_long_t obd_dirty_transit_pages;
103 EXPORT_SYMBOL(obd_dirty_transit_pages);
105 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
107 #ifdef CONFIG_PROC_FS
108 struct lprocfs_stats *obd_memory = NULL;
109 EXPORT_SYMBOL(obd_memory);
112 char obd_jobid_node[LUSTRE_JOBID_SIZE + 1];
114 /* Get jobid of current process by reading the environment variable
115 * stored in between the "env_start" & "env_end" of task struct.
118 * It's better to cache the jobid for later use if there is any
119 * efficient way, the cl_env code probably could be reused for this
122 * If some job scheduler doesn't store jobid in the "env_start/end",
123 * then an upcall could be issued here to get the jobid by utilizing
124 * the userspace tools/api. Then, the jobid must be cached.
126 int lustre_get_jobid(char *jobid)
128 int jobid_len = LUSTRE_JOBID_SIZE;
129 char tmp_jobid[LUSTRE_JOBID_SIZE] = { 0 };
133 /* Jobstats isn't enabled */
134 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
137 /* Whole node dedicated to single job */
138 if (strcmp(obd_jobid_var, JOBSTATS_NODELOCAL) == 0) {
139 memcpy(tmp_jobid, obd_jobid_node, LUSTRE_JOBID_SIZE);
143 /* Use process name + fsuid as jobid */
144 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
145 snprintf(tmp_jobid, LUSTRE_JOBID_SIZE, "%s.%u",
147 from_kuid(&init_user_ns, current_fsuid()));
151 rc = cfs_get_environ(obd_jobid_var, tmp_jobid, &jobid_len);
153 if (rc == -EOVERFLOW) {
154 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
155 * variable length strings instead of just numbers), it
156 * might make sense to keep the unique parts for JobID,
157 * instead of just returning an error. That means a
158 * larger temp buffer for cfs_get_environ(), then
159 * truncating the string at some separator to fit into
160 * the specified jobid_len. Fix later if needed. */
162 if (unlikely(!printed)) {
163 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
164 "for JobID buffer (%d)\n",
165 obd_jobid_var, jobid_len);
169 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
170 rc == -EDEADLK) ? D_INFO : D_ERROR,
171 "Get jobid for (%s) failed: rc = %d\n",
180 /* Only replace the job ID if it changed. */
181 if (strcmp(jobid, tmp_jobid) != 0)
182 memcpy(jobid, tmp_jobid, jobid_len);
186 EXPORT_SYMBOL(lustre_get_jobid);
188 static int class_resolve_dev_name(__u32 len, const char *name)
195 CERROR("No name passed,!\n");
196 GOTO(out, rc = -EINVAL);
198 if (name[len - 1] != 0) {
199 CERROR("Name not nul terminated!\n");
200 GOTO(out, rc = -EINVAL);
203 CDEBUG(D_IOCTL, "device name %s\n", name);
204 dev = class_name2dev(name);
206 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
207 GOTO(out, rc = -EINVAL);
210 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
217 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
220 struct obd_ioctl_data *data;
221 struct libcfs_debug_ioctl_data *debug_data;
222 struct obd_device *obd = NULL;
223 int err = 0, len = 0;
226 /* only for debugging */
227 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
228 debug_data = (struct libcfs_debug_ioctl_data*)arg;
229 libcfs_subsystem_debug = debug_data->subs;
230 libcfs_debug = debug_data->debug;
234 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
235 if (obd_ioctl_getdata(&buf, &len, (void __user *)arg)) {
236 CERROR("OBD ioctl: data error\n");
239 data = (struct obd_ioctl_data *)buf;
242 case OBD_IOC_PROCESS_CFG: {
243 struct lustre_cfg *lcfg;
245 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
246 CERROR("No config buffer passed!\n");
247 GOTO(out, err = -EINVAL);
249 OBD_ALLOC(lcfg, data->ioc_plen1);
251 GOTO(out, err = -ENOMEM);
252 err = copy_from_user(lcfg, data->ioc_pbuf1,
255 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
257 err = class_process_config(lcfg);
259 OBD_FREE(lcfg, data->ioc_plen1);
263 case OBD_GET_VERSION:
264 if (!data->ioc_inlbuf1) {
265 CERROR("No buffer passed in ioctl\n");
266 GOTO(out, err = -EINVAL);
269 if (strlen(LUSTRE_VERSION_STRING) + 1 > data->ioc_inllen1) {
270 CERROR("ioctl buffer too small to hold version\n");
271 GOTO(out, err = -EINVAL);
274 memcpy(data->ioc_bulk, LUSTRE_VERSION_STRING,
275 strlen(LUSTRE_VERSION_STRING) + 1);
277 err = obd_ioctl_popdata((void __user *)arg, data, len);
282 case OBD_IOC_NAME2DEV: {
283 /* Resolve a device name. This does not change the
284 * currently selected device.
288 dev = class_resolve_dev_name(data->ioc_inllen1,
292 GOTO(out, err = -EINVAL);
294 err = obd_ioctl_popdata((void __user *)arg, data,
301 case OBD_IOC_UUID2DEV: {
302 /* Resolve a device uuid. This does not change the
303 * currently selected device.
306 struct obd_uuid uuid;
308 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
309 CERROR("No UUID passed!\n");
310 GOTO(out, err = -EINVAL);
312 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
313 CERROR("UUID not NUL terminated!\n");
314 GOTO(out, err = -EINVAL);
317 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
318 obd_str2uuid(&uuid, data->ioc_inlbuf1);
319 dev = class_uuid2dev(&uuid);
322 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
324 GOTO(out, err = -EINVAL);
327 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
329 err = obd_ioctl_popdata((void __user *)arg, 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 __user *)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 __user *)arg, data, len);
423 obd_ioctl_freedata(buf, len);
425 } /* class_handle_ioctl */
427 #define OBD_INIT_CHECK
428 #ifdef OBD_INIT_CHECK
429 static int obd_init_checks(void)
431 __u64 u64val, div64val;
435 CDEBUG(D_INFO, "OBD_OBJECT_EOF = %#llx\n", (__u64)OBD_OBJECT_EOF);
437 u64val = OBD_OBJECT_EOF;
438 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = %#llx\n", u64val);
439 if (u64val != OBD_OBJECT_EOF) {
440 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
441 u64val, (int)sizeof(u64val));
444 len = snprintf(buf, sizeof(buf), "%#llx", u64val);
446 CWARN("u64 hex wrong length! strlen(%s)=%d != 18\n", buf, len);
450 div64val = OBD_OBJECT_EOF;
451 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = %#llx\n", u64val);
452 if (u64val != OBD_OBJECT_EOF) {
453 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
454 u64val, (int)sizeof(u64val));
457 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
458 CERROR("__u64 %#llx(%d) != 0xffffffffffffffff\n",
459 u64val, (int)sizeof(u64val));
462 if (do_div(div64val, 256) != (u64val & 255)) {
463 CERROR("do_div(%#llx,256) != %llu\n", u64val, u64val & 255);
466 if (u64val >> 8 != div64val) {
467 CERROR("do_div(%#llx,256) %llu != %llu\n",
468 u64val, div64val, u64val >> 8);
471 len = snprintf(buf, sizeof(buf), "%#llx", u64val);
473 CWARN("u64 hex wrong length! strlen(%s)=%d != 18\n", buf, len);
476 len = snprintf(buf, sizeof(buf), "%llu", u64val);
478 CWARN("u64 wrong length! strlen(%s)=%d != 20\n", buf, len);
481 len = snprintf(buf, sizeof(buf), "%lld", u64val);
483 CWARN("s64 wrong length! strlen(%s)=%d != 2\n", buf, len);
486 if ((u64val & ~PAGE_MASK) >= PAGE_SIZE) {
487 CWARN("mask failed: u64val %llu >= %llu\n", u64val,
495 #define obd_init_checks() do {} while(0)
498 static int __init obdclass_init(void)
502 spin_lock_init(&obd_stale_export_lock);
503 INIT_LIST_HEAD(&obd_stale_exports);
504 atomic_set(&obd_stale_export_num, 0);
506 LCONSOLE_INFO("Lustre: Build Version: "LUSTRE_VERSION_STRING"\n");
508 spin_lock_init(&obd_types_lock);
509 obd_zombie_impexp_init();
510 #ifdef CONFIG_PROC_FS
511 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
512 LPROCFS_STATS_FLAG_NONE |
513 LPROCFS_STATS_FLAG_IRQ_SAFE);
514 if (obd_memory == NULL) {
515 CERROR("kmalloc of 'obd_memory' failed\n");
519 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
520 LPROCFS_CNTR_AVGMINMAX,
523 err = obd_init_checks();
524 if (err == -EOVERFLOW)
527 class_init_uuidlist();
528 err = class_handle_init();
532 INIT_LIST_HEAD(&obd_types);
534 err = misc_register(&obd_psdev);
536 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
540 /* This struct is already zeroed for us (static global) */
541 for (i = 0; i < class_devno_max(); i++)
544 /* Default the dirty page cache cap to 1/2 of system memory.
545 * For clients with less memory, a larger fraction is needed
546 * for other purposes (mostly for BGL). */
547 if (totalram_pages <= 512 << (20 - PAGE_SHIFT))
548 obd_max_dirty_pages = totalram_pages / 4;
550 obd_max_dirty_pages = totalram_pages / 2;
552 err = obd_init_caches();
555 err = class_procfs_init();
559 err = lu_global_init();
563 err = cl_global_init();
567 #ifdef HAVE_SERVER_SUPPORT
568 err = dt_global_init();
572 err = lu_ucred_global_init();
575 #endif /* HAVE_SERVER_SUPPORT */
577 err = llog_info_init();
581 err = lustre_register_fs();
586 void obd_update_maxusage(void)
590 max = obd_memory_sum();
592 spin_lock(&obd_updatemax_lock);
593 if (max > obd_max_alloc)
595 spin_unlock(&obd_updatemax_lock);
597 EXPORT_SYMBOL(obd_update_maxusage);
599 #ifdef CONFIG_PROC_FS
600 __u64 obd_memory_max(void)
604 obd_update_maxusage();
605 spin_lock(&obd_updatemax_lock);
607 spin_unlock(&obd_updatemax_lock);
611 #endif /* CONFIG_PROC_FS */
613 static void __exit obdclass_exit(void)
619 lustre_unregister_fs();
621 misc_deregister(&obd_psdev);
623 #ifdef HAVE_SERVER_SUPPORT
624 lu_ucred_global_fini();
626 #endif /* HAVE_SERVER_SUPPORT */
630 obd_cleanup_caches();
633 class_procfs_clean();
635 class_handle_cleanup();
636 class_exit_uuidlist();
637 obd_zombie_impexp_stop();
638 LASSERT(list_empty(&obd_stale_exports));
640 memory_leaked = obd_memory_sum();
642 memory_max = obd_memory_max();
644 lprocfs_free_stats(&obd_memory);
645 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
646 "obd_memory max: %llu, leaked: %llu\n",
647 memory_max, memory_leaked);
652 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
653 MODULE_DESCRIPTION("Lustre Class Driver");
654 MODULE_VERSION(LUSTRE_VERSION_STRING);
655 MODULE_LICENSE("GPL");
657 module_init(obdclass_init);
658 module_exit(obdclass_exit);