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
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);
62 static __u64 obd_max_pages;
63 static __u64 obd_max_alloc;
69 static DEFINE_SPINLOCK(obd_updatemax_lock);
71 /* The following are visible and mutable through /proc/sys/lustre/. */
72 unsigned int obd_alloc_fail_rate = 0;
73 EXPORT_SYMBOL(obd_alloc_fail_rate);
74 unsigned int obd_debug_peer_on_timeout;
75 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
76 unsigned int obd_dump_on_timeout;
77 EXPORT_SYMBOL(obd_dump_on_timeout);
78 unsigned int obd_dump_on_eviction;
79 EXPORT_SYMBOL(obd_dump_on_eviction);
80 unsigned long obd_max_dirty_pages;
81 EXPORT_SYMBOL(obd_max_dirty_pages);
82 atomic_long_t obd_dirty_pages;
83 EXPORT_SYMBOL(obd_dirty_pages);
84 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
85 EXPORT_SYMBOL(obd_timeout);
86 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
87 EXPORT_SYMBOL(ldlm_timeout);
88 unsigned int obd_timeout_set;
89 EXPORT_SYMBOL(obd_timeout_set);
90 unsigned int ldlm_timeout_set;
91 EXPORT_SYMBOL(ldlm_timeout_set);
92 /* bulk transfer timeout, give up after 100s by default */
93 unsigned int bulk_timeout = 100; /* seconds */
94 EXPORT_SYMBOL(bulk_timeout);
95 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
96 unsigned int at_min = 0;
97 EXPORT_SYMBOL(at_min);
98 unsigned int at_max = 600;
99 EXPORT_SYMBOL(at_max);
100 unsigned int at_history = 600;
101 EXPORT_SYMBOL(at_history);
102 int at_early_margin = 5;
103 EXPORT_SYMBOL(at_early_margin);
105 EXPORT_SYMBOL(at_extra);
107 atomic_long_t obd_dirty_transit_pages;
108 EXPORT_SYMBOL(obd_dirty_transit_pages);
110 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
111 EXPORT_SYMBOL(obd_jobid_var);
114 struct lprocfs_stats *obd_memory = NULL;
115 EXPORT_SYMBOL(obd_memory);
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 /* Use process name + fsuid as jobid */
142 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
143 snprintf(jobid, LUSTRE_JOBID_SIZE, "%s.%u",
145 from_kuid(&init_user_ns, current_fsuid()));
149 rc = cfs_get_environ(obd_jobid_var, 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",
175 EXPORT_SYMBOL(lustre_get_jobid);
177 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
178 size_t size, const char *file, int line)
181 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
182 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
183 ptr ? "force " :"", type, name, (__u64)size, file,
185 CERROR(LPU64" total bytes and "LPU64" total pages "
186 "("LPU64" bytes) allocated by Lustre, "
187 "%d total bytes by LNET\n",
189 obd_pages_sum() << PAGE_CACHE_SHIFT,
191 atomic_read(&libcfs_kmemory));
196 EXPORT_SYMBOL(obd_alloc_fail);
198 static int class_resolve_dev_name(__u32 len, const char *name)
205 CERROR("No name passed,!\n");
206 GOTO(out, rc = -EINVAL);
208 if (name[len - 1] != 0) {
209 CERROR("Name not nul terminated!\n");
210 GOTO(out, rc = -EINVAL);
213 CDEBUG(D_IOCTL, "device name %s\n", name);
214 dev = class_name2dev(name);
216 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
217 GOTO(out, rc = -EINVAL);
220 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
227 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
230 struct obd_ioctl_data *data;
231 struct libcfs_debug_ioctl_data *debug_data;
232 struct obd_device *obd = NULL;
233 int err = 0, len = 0;
236 /* only for debugging */
237 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
238 debug_data = (struct libcfs_debug_ioctl_data*)arg;
239 libcfs_subsystem_debug = debug_data->subs;
240 libcfs_debug = debug_data->debug;
244 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
245 if (obd_ioctl_getdata(&buf, &len, (void __user *)arg)) {
246 CERROR("OBD ioctl: data error\n");
249 data = (struct obd_ioctl_data *)buf;
252 case OBD_IOC_PROCESS_CFG: {
253 struct lustre_cfg *lcfg;
255 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
256 CERROR("No config buffer passed!\n");
257 GOTO(out, err = -EINVAL);
259 OBD_ALLOC(lcfg, data->ioc_plen1);
261 GOTO(out, err = -ENOMEM);
262 err = copy_from_user(lcfg, data->ioc_pbuf1,
265 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
267 err = class_process_config(lcfg);
269 OBD_FREE(lcfg, data->ioc_plen1);
273 case OBD_GET_VERSION:
274 if (!data->ioc_inlbuf1) {
275 CERROR("No buffer passed in ioctl\n");
276 GOTO(out, err = -EINVAL);
279 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
280 CERROR("ioctl buffer too small to hold version\n");
281 GOTO(out, err = -EINVAL);
284 memcpy(data->ioc_bulk, BUILD_VERSION,
285 strlen(BUILD_VERSION) + 1);
287 err = obd_ioctl_popdata((void __user *)arg, data, len);
292 case OBD_IOC_NAME2DEV: {
293 /* Resolve a device name. This does not change the
294 * currently selected device.
298 dev = class_resolve_dev_name(data->ioc_inllen1,
302 GOTO(out, err = -EINVAL);
304 err = obd_ioctl_popdata((void __user *)arg, data,
311 case OBD_IOC_UUID2DEV: {
312 /* Resolve a device uuid. This does not change the
313 * currently selected device.
316 struct obd_uuid uuid;
318 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
319 CERROR("No UUID passed!\n");
320 GOTO(out, err = -EINVAL);
322 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
323 CERROR("UUID not NUL terminated!\n");
324 GOTO(out, err = -EINVAL);
327 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
328 obd_str2uuid(&uuid, data->ioc_inlbuf1);
329 dev = class_uuid2dev(&uuid);
332 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
334 GOTO(out, err = -EINVAL);
337 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
339 err = obd_ioctl_popdata((void __user *)arg, data,
346 case OBD_IOC_GETDEVICE: {
347 int index = data->ioc_count;
350 if (!data->ioc_inlbuf1) {
351 CERROR("No buffer passed in ioctl\n");
352 GOTO(out, err = -EINVAL);
354 if (data->ioc_inllen1 < 128) {
355 CERROR("ioctl buffer too small to hold version\n");
356 GOTO(out, err = -EINVAL);
359 obd = class_num2obd(index);
361 GOTO(out, err = -ENOENT);
363 if (obd->obd_stopping)
365 else if (obd->obd_set_up)
367 else if (obd->obd_attached)
371 str = (char *)data->ioc_bulk;
372 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
373 (int)index, status, obd->obd_type->typ_name,
374 obd->obd_name, obd->obd_uuid.uuid,
375 atomic_read(&obd->obd_refcount));
376 err = obd_ioctl_popdata((void __user *)arg, data, len);
383 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
384 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
385 GOTO(out, err = -EINVAL);
386 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
387 GOTO(out, err = -EINVAL);
388 obd = class_name2obd(data->ioc_inlbuf4);
389 } else if (data->ioc_dev < class_devno_max()) {
390 obd = class_num2obd(data->ioc_dev);
392 CERROR("OBD ioctl: No device\n");
393 GOTO(out, err = -EINVAL);
397 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
398 GOTO(out, err = -EINVAL);
400 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
402 if (!obd->obd_set_up || obd->obd_stopping) {
403 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
404 GOTO(out, err = -EINVAL);
408 case OBD_IOC_NO_TRANSNO: {
409 if (!obd->obd_attached) {
410 CERROR("Device %d not attached\n", obd->obd_minor);
411 GOTO(out, err = -ENODEV);
413 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
415 obd->obd_no_transno = 1;
420 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
424 err = obd_ioctl_popdata((void __user *)arg, data, len);
433 obd_ioctl_freedata(buf, len);
435 } /* class_handle_ioctl */
437 #define OBD_INIT_CHECK
438 #ifdef OBD_INIT_CHECK
439 static int obd_init_checks(void)
441 __u64 u64val, div64val;
445 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
447 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
449 u64val = OBD_OBJECT_EOF;
450 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
451 if (u64val != OBD_OBJECT_EOF) {
452 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
453 u64val, (int)sizeof(u64val));
456 len = snprintf(buf, sizeof(buf), LPX64, u64val);
458 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
462 div64val = OBD_OBJECT_EOF;
463 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
464 if (u64val != OBD_OBJECT_EOF) {
465 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
466 u64val, (int)sizeof(u64val));
469 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
470 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
471 u64val, (int)sizeof(u64val));
474 if (do_div(div64val, 256) != (u64val & 255)) {
475 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
478 if (u64val >> 8 != div64val) {
479 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
480 u64val, div64val, u64val >> 8);
483 len = snprintf(buf, sizeof(buf), LPX64, u64val);
485 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
488 len = snprintf(buf, sizeof(buf), LPU64, u64val);
490 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
493 len = snprintf(buf, sizeof(buf), LPD64, u64val);
495 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
498 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
499 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
500 (__u64)PAGE_CACHE_SIZE);
507 #define obd_init_checks() do {} while(0)
510 static int __init init_obdclass(void)
514 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
515 INIT_LIST_HEAD(&capa_list[i]);
517 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
519 spin_lock_init(&obd_types_lock);
520 obd_zombie_impexp_init();
522 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
523 LPROCFS_STATS_FLAG_NONE |
524 LPROCFS_STATS_FLAG_IRQ_SAFE);
525 if (obd_memory == NULL) {
526 CERROR("kmalloc of 'obd_memory' failed\n");
530 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
531 LPROCFS_CNTR_AVGMINMAX,
533 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
534 LPROCFS_CNTR_AVGMINMAX,
535 "pagesused", "pages");
537 err = obd_init_checks();
538 if (err == -EOVERFLOW)
541 class_init_uuidlist();
542 err = class_handle_init();
546 INIT_LIST_HEAD(&obd_types);
548 err = misc_register(&obd_psdev);
550 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
554 /* This struct is already zeroed for us (static global) */
555 for (i = 0; i < class_devno_max(); i++)
558 /* Default the dirty page cache cap to 1/2 of system memory.
559 * For clients with less memory, a larger fraction is needed
560 * for other purposes (mostly for BGL). */
561 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
562 obd_max_dirty_pages = totalram_pages / 4;
564 obd_max_dirty_pages = totalram_pages / 2;
566 err = obd_init_caches();
569 err = class_procfs_init();
573 err = lu_global_init();
577 err = lu_capainfo_init();
581 err = cl_global_init();
585 #ifdef HAVE_SERVER_SUPPORT
586 err = dt_global_init();
590 err = lu_ucred_global_init();
593 #endif /* HAVE_SERVER_SUPPORT */
595 err = llog_info_init();
599 err = lustre_register_fs();
604 void obd_update_maxusage(void)
608 max1 = obd_pages_sum();
609 max2 = obd_memory_sum();
611 spin_lock(&obd_updatemax_lock);
612 if (max1 > obd_max_pages)
613 obd_max_pages = max1;
614 if (max2 > obd_max_alloc)
615 obd_max_alloc = max2;
616 spin_unlock(&obd_updatemax_lock);
618 EXPORT_SYMBOL(obd_update_maxusage);
621 __u64 obd_memory_max(void)
625 spin_lock(&obd_updatemax_lock);
627 spin_unlock(&obd_updatemax_lock);
631 EXPORT_SYMBOL(obd_memory_max);
633 __u64 obd_pages_max(void)
637 spin_lock(&obd_updatemax_lock);
639 spin_unlock(&obd_updatemax_lock);
643 EXPORT_SYMBOL(obd_pages_max);
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_LICENSE("GPL");
696 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);