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, 2014, 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_pages;
66 static __u64 obd_max_alloc;
72 static DEFINE_SPINLOCK(obd_updatemax_lock);
74 /* The following are visible and mutable through /proc/sys/lustre/. */
75 unsigned int obd_alloc_fail_rate = 0;
76 EXPORT_SYMBOL(obd_alloc_fail_rate);
77 unsigned int obd_debug_peer_on_timeout;
78 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
79 unsigned int obd_dump_on_timeout;
80 EXPORT_SYMBOL(obd_dump_on_timeout);
81 unsigned int obd_dump_on_eviction;
82 EXPORT_SYMBOL(obd_dump_on_eviction);
83 unsigned long obd_max_dirty_pages;
84 EXPORT_SYMBOL(obd_max_dirty_pages);
85 atomic_long_t obd_dirty_pages;
86 EXPORT_SYMBOL(obd_dirty_pages);
87 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
88 EXPORT_SYMBOL(obd_timeout);
89 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
90 EXPORT_SYMBOL(ldlm_timeout);
91 unsigned int obd_timeout_set;
92 EXPORT_SYMBOL(obd_timeout_set);
93 unsigned int ldlm_timeout_set;
94 EXPORT_SYMBOL(ldlm_timeout_set);
95 /* bulk transfer timeout, give up after 100s by default */
96 unsigned int bulk_timeout = 100; /* seconds */
97 EXPORT_SYMBOL(bulk_timeout);
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_long_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;
115 #ifdef CONFIG_PROC_FS
116 struct lprocfs_stats *obd_memory = NULL;
117 EXPORT_SYMBOL(obd_memory);
120 /* Get jobid of current process by reading the environment variable
121 * stored in between the "env_start" & "env_end" of task struct.
124 * It's better to cache the jobid for later use if there is any
125 * efficient way, the cl_env code probably could be reused for this
128 * If some job scheduler doesn't store jobid in the "env_start/end",
129 * then an upcall could be issued here to get the jobid by utilizing
130 * the userspace tools/api. Then, the jobid must be cached.
132 int lustre_get_jobid(char *jobid)
134 int jobid_len = LUSTRE_JOBID_SIZE;
138 memset(jobid, 0, LUSTRE_JOBID_SIZE);
139 /* Jobstats isn't enabled */
140 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
143 /* Use process name + fsuid as jobid */
144 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
145 snprintf(jobid, LUSTRE_JOBID_SIZE, "%s.%u",
147 from_kuid(&init_user_ns, current_fsuid()));
151 rc = cfs_get_environ(obd_jobid_var, 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",
177 EXPORT_SYMBOL(lustre_get_jobid);
179 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
180 size_t size, const char *file, int line)
183 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
184 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
185 ptr ? "force " :"", type, name, (__u64)size, file,
187 CERROR(LPU64" total bytes and "LPU64" total pages "
188 "("LPU64" bytes) allocated by Lustre, "
189 "%d total bytes by LNET\n",
191 obd_pages_sum() << PAGE_CACHE_SHIFT,
193 atomic_read(&libcfs_kmemory));
198 EXPORT_SYMBOL(obd_alloc_fail);
200 static int class_resolve_dev_name(__u32 len, const char *name)
207 CERROR("No name passed,!\n");
208 GOTO(out, rc = -EINVAL);
210 if (name[len - 1] != 0) {
211 CERROR("Name not nul terminated!\n");
212 GOTO(out, rc = -EINVAL);
215 CDEBUG(D_IOCTL, "device name %s\n", name);
216 dev = class_name2dev(name);
218 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
219 GOTO(out, rc = -EINVAL);
222 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
229 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
232 struct obd_ioctl_data *data;
233 struct libcfs_debug_ioctl_data *debug_data;
234 struct obd_device *obd = NULL;
235 int err = 0, len = 0;
238 /* only for debugging */
239 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
240 debug_data = (struct libcfs_debug_ioctl_data*)arg;
241 libcfs_subsystem_debug = debug_data->subs;
242 libcfs_debug = debug_data->debug;
246 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
247 if (obd_ioctl_getdata(&buf, &len, (void __user *)arg)) {
248 CERROR("OBD ioctl: data error\n");
251 data = (struct obd_ioctl_data *)buf;
254 case OBD_IOC_PROCESS_CFG: {
255 struct lustre_cfg *lcfg;
257 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
258 CERROR("No config buffer passed!\n");
259 GOTO(out, err = -EINVAL);
261 OBD_ALLOC(lcfg, data->ioc_plen1);
263 GOTO(out, err = -ENOMEM);
264 err = copy_from_user(lcfg, data->ioc_pbuf1,
267 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
269 err = class_process_config(lcfg);
271 OBD_FREE(lcfg, data->ioc_plen1);
275 case OBD_GET_VERSION:
276 if (!data->ioc_inlbuf1) {
277 CERROR("No buffer passed in ioctl\n");
278 GOTO(out, err = -EINVAL);
281 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
282 CERROR("ioctl buffer too small to hold version\n");
283 GOTO(out, err = -EINVAL);
286 memcpy(data->ioc_bulk, BUILD_VERSION,
287 strlen(BUILD_VERSION) + 1);
289 err = obd_ioctl_popdata((void __user *)arg, data, len);
294 case OBD_IOC_NAME2DEV: {
295 /* Resolve a device name. This does not change the
296 * currently selected device.
300 dev = class_resolve_dev_name(data->ioc_inllen1,
304 GOTO(out, err = -EINVAL);
306 err = obd_ioctl_popdata((void __user *)arg, 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 __user *)arg, data,
348 case OBD_IOC_GETDEVICE: {
349 int index = data->ioc_count;
352 if (!data->ioc_inlbuf1) {
353 CERROR("No buffer passed in ioctl\n");
354 GOTO(out, err = -EINVAL);
356 if (data->ioc_inllen1 < 128) {
357 CERROR("ioctl buffer too small to hold version\n");
358 GOTO(out, err = -EINVAL);
361 obd = class_num2obd(index);
363 GOTO(out, err = -ENOENT);
365 if (obd->obd_stopping)
367 else if (obd->obd_set_up)
369 else if (obd->obd_attached)
373 str = (char *)data->ioc_bulk;
374 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
375 (int)index, status, obd->obd_type->typ_name,
376 obd->obd_name, obd->obd_uuid.uuid,
377 atomic_read(&obd->obd_refcount));
378 err = obd_ioctl_popdata((void __user *)arg, data, len);
385 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
386 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
387 GOTO(out, err = -EINVAL);
388 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
389 GOTO(out, err = -EINVAL);
390 obd = class_name2obd(data->ioc_inlbuf4);
391 } else if (data->ioc_dev < class_devno_max()) {
392 obd = class_num2obd(data->ioc_dev);
394 CERROR("OBD ioctl: No device\n");
395 GOTO(out, err = -EINVAL);
399 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
400 GOTO(out, err = -EINVAL);
402 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
404 if (!obd->obd_set_up || obd->obd_stopping) {
405 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
406 GOTO(out, err = -EINVAL);
410 case OBD_IOC_NO_TRANSNO: {
411 if (!obd->obd_attached) {
412 CERROR("Device %d not attached\n", obd->obd_minor);
413 GOTO(out, err = -ENODEV);
415 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
417 obd->obd_no_transno = 1;
422 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
426 err = obd_ioctl_popdata((void __user *)arg, data, len);
435 obd_ioctl_freedata(buf, len);
437 } /* class_handle_ioctl */
439 #define OBD_INIT_CHECK
440 #ifdef OBD_INIT_CHECK
441 static int obd_init_checks(void)
443 __u64 u64val, div64val;
447 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
449 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
451 u64val = OBD_OBJECT_EOF;
452 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
453 if (u64val != OBD_OBJECT_EOF) {
454 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
455 u64val, (int)sizeof(u64val));
458 len = snprintf(buf, sizeof(buf), LPX64, u64val);
460 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
464 div64val = OBD_OBJECT_EOF;
465 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
466 if (u64val != OBD_OBJECT_EOF) {
467 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
468 u64val, (int)sizeof(u64val));
471 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
472 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
473 u64val, (int)sizeof(u64val));
476 if (do_div(div64val, 256) != (u64val & 255)) {
477 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
480 if (u64val >> 8 != div64val) {
481 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
482 u64val, div64val, u64val >> 8);
485 len = snprintf(buf, sizeof(buf), LPX64, u64val);
487 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
490 len = snprintf(buf, sizeof(buf), LPU64, u64val);
492 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
495 len = snprintf(buf, sizeof(buf), LPD64, u64val);
497 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
500 if ((u64val & ~PAGE_CACHE_MASK) >= PAGE_CACHE_SIZE) {
501 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
502 (__u64)PAGE_CACHE_SIZE);
509 #define obd_init_checks() do {} while(0)
512 static int __init init_obdclass(void)
516 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
517 INIT_LIST_HEAD(&capa_list[i]);
519 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
521 spin_lock_init(&obd_types_lock);
522 obd_zombie_impexp_init();
523 #ifdef CONFIG_PROC_FS
524 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
525 LPROCFS_STATS_FLAG_NONE |
526 LPROCFS_STATS_FLAG_IRQ_SAFE);
527 if (obd_memory == NULL) {
528 CERROR("kmalloc of 'obd_memory' failed\n");
532 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
533 LPROCFS_CNTR_AVGMINMAX,
535 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
536 LPROCFS_CNTR_AVGMINMAX,
537 "pagesused", "pages");
539 err = obd_init_checks();
540 if (err == -EOVERFLOW)
543 class_init_uuidlist();
544 err = class_handle_init();
548 INIT_LIST_HEAD(&obd_types);
550 err = misc_register(&obd_psdev);
552 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
556 /* This struct is already zeroed for us (static global) */
557 for (i = 0; i < class_devno_max(); i++)
560 /* Default the dirty page cache cap to 1/2 of system memory.
561 * For clients with less memory, a larger fraction is needed
562 * for other purposes (mostly for BGL). */
563 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
564 obd_max_dirty_pages = totalram_pages / 4;
566 obd_max_dirty_pages = totalram_pages / 2;
568 err = obd_init_caches();
571 err = class_procfs_init();
575 err = lu_global_init();
579 err = lu_capainfo_init();
583 err = cl_global_init();
587 #ifdef HAVE_SERVER_SUPPORT
588 err = dt_global_init();
592 err = lu_ucred_global_init();
595 #endif /* HAVE_SERVER_SUPPORT */
597 err = llog_info_init();
601 err = lustre_register_fs();
606 void obd_update_maxusage(void)
610 max1 = obd_pages_sum();
611 max2 = obd_memory_sum();
613 spin_lock(&obd_updatemax_lock);
614 if (max1 > obd_max_pages)
615 obd_max_pages = max1;
616 if (max2 > obd_max_alloc)
617 obd_max_alloc = max2;
618 spin_unlock(&obd_updatemax_lock);
620 EXPORT_SYMBOL(obd_update_maxusage);
622 #ifdef CONFIG_PROC_FS
623 __u64 obd_memory_max(void)
627 spin_lock(&obd_updatemax_lock);
629 spin_unlock(&obd_updatemax_lock);
634 __u64 obd_pages_max(void)
638 spin_lock(&obd_updatemax_lock);
640 spin_unlock(&obd_updatemax_lock);
644 #endif /* CONFIG_PROC_FS */
646 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
647 * ifdef to the end of the file to cover module and versioning goo.*/
648 static void cleanup_obdclass(void)
650 __u64 memory_leaked, pages_leaked;
651 __u64 memory_max, pages_max;
654 lustre_unregister_fs();
656 misc_deregister(&obd_psdev);
658 #ifdef HAVE_SERVER_SUPPORT
659 lu_ucred_global_fini();
661 #endif /* HAVE_SERVER_SUPPORT */
666 obd_cleanup_caches();
669 class_procfs_clean();
671 class_handle_cleanup();
672 class_exit_uuidlist();
673 obd_zombie_impexp_stop();
675 memory_leaked = obd_memory_sum();
676 pages_leaked = obd_pages_sum();
678 memory_max = obd_memory_max();
679 pages_max = obd_pages_max();
681 lprocfs_free_stats(&obd_memory);
682 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
683 "obd_memory max: "LPU64", leaked: "LPU64"\n",
684 memory_max, memory_leaked);
685 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
686 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
687 pages_max, pages_leaked);
692 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
693 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
694 MODULE_VERSION(LUSTRE_VERSION_STRING);
695 MODULE_LICENSE("GPL");
697 module_init(init_obdclass);
698 module_exit(cleanup_obdclass);