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 struct list_head obd_types;
58 DEFINE_RWLOCK(obd_dev_lock);
61 static __u64 obd_max_pages;
62 static __u64 obd_max_alloc;
68 static DEFINE_SPINLOCK(obd_updatemax_lock);
70 /* The following are visible and mutable through /proc/sys/lustre/. */
71 unsigned int obd_alloc_fail_rate = 0;
72 EXPORT_SYMBOL(obd_alloc_fail_rate);
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 /* Get jobid of current process by reading the environment variable
117 * stored in between the "env_start" & "env_end" of task struct.
120 * It's better to cache the jobid for later use if there is any
121 * efficient way, the cl_env code probably could be reused for this
124 * If some job scheduler doesn't store jobid in the "env_start/end",
125 * then an upcall could be issued here to get the jobid by utilizing
126 * the userspace tools/api. Then, the jobid must be cached.
128 int lustre_get_jobid(char *jobid)
130 int jobid_len = LUSTRE_JOBID_SIZE;
134 memset(jobid, 0, LUSTRE_JOBID_SIZE);
135 /* Jobstats isn't enabled */
136 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
139 /* Use process name + fsuid as jobid */
140 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
141 snprintf(jobid, LUSTRE_JOBID_SIZE, "%s.%u",
143 from_kuid(&init_user_ns, current_fsuid()));
147 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
149 if (rc == -EOVERFLOW) {
150 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
151 * variable length strings instead of just numbers), it
152 * might make sense to keep the unique parts for JobID,
153 * instead of just returning an error. That means a
154 * larger temp buffer for cfs_get_environ(), then
155 * truncating the string at some separator to fit into
156 * the specified jobid_len. Fix later if needed. */
158 if (unlikely(!printed)) {
159 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
160 "for JobID buffer (%d)\n",
161 obd_jobid_var, jobid_len);
165 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
166 rc == -EDEADLK) ? D_INFO : D_ERROR,
167 "Get jobid for (%s) failed: rc = %d\n",
173 EXPORT_SYMBOL(lustre_get_jobid);
175 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
176 size_t size, const char *file, int line)
179 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
180 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
181 ptr ? "force " :"", type, name, (__u64)size, file,
183 CERROR(LPU64" total bytes and "LPU64" total pages "
184 "("LPU64" bytes) allocated by Lustre, "
185 "%d total bytes by LNET\n",
187 obd_pages_sum() << PAGE_CACHE_SHIFT,
189 atomic_read(&libcfs_kmemory));
194 EXPORT_SYMBOL(obd_alloc_fail);
196 static int class_resolve_dev_name(__u32 len, const char *name)
203 CERROR("No name passed,!\n");
204 GOTO(out, rc = -EINVAL);
206 if (name[len - 1] != 0) {
207 CERROR("Name not nul terminated!\n");
208 GOTO(out, rc = -EINVAL);
211 CDEBUG(D_IOCTL, "device name %s\n", name);
212 dev = class_name2dev(name);
214 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
215 GOTO(out, rc = -EINVAL);
218 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
225 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
228 struct obd_ioctl_data *data;
229 struct libcfs_debug_ioctl_data *debug_data;
230 struct obd_device *obd = NULL;
231 int err = 0, len = 0;
234 /* only for debugging */
235 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
236 debug_data = (struct libcfs_debug_ioctl_data*)arg;
237 libcfs_subsystem_debug = debug_data->subs;
238 libcfs_debug = debug_data->debug;
242 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
243 if (obd_ioctl_getdata(&buf, &len, (void __user *)arg)) {
244 CERROR("OBD ioctl: data error\n");
247 data = (struct obd_ioctl_data *)buf;
250 case OBD_IOC_PROCESS_CFG: {
251 struct lustre_cfg *lcfg;
253 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
254 CERROR("No config buffer passed!\n");
255 GOTO(out, err = -EINVAL);
257 OBD_ALLOC(lcfg, data->ioc_plen1);
259 GOTO(out, err = -ENOMEM);
260 err = copy_from_user(lcfg, data->ioc_pbuf1,
263 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
265 err = class_process_config(lcfg);
267 OBD_FREE(lcfg, data->ioc_plen1);
271 case OBD_GET_VERSION:
272 if (!data->ioc_inlbuf1) {
273 CERROR("No buffer passed in ioctl\n");
274 GOTO(out, err = -EINVAL);
277 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
278 CERROR("ioctl buffer too small to hold version\n");
279 GOTO(out, err = -EINVAL);
282 memcpy(data->ioc_bulk, BUILD_VERSION,
283 strlen(BUILD_VERSION) + 1);
285 err = obd_ioctl_popdata((void __user *)arg, data, len);
290 case OBD_IOC_NAME2DEV: {
291 /* Resolve a device name. This does not change the
292 * currently selected device.
296 dev = class_resolve_dev_name(data->ioc_inllen1,
300 GOTO(out, err = -EINVAL);
302 err = obd_ioctl_popdata((void __user *)arg, data,
309 case OBD_IOC_UUID2DEV: {
310 /* Resolve a device uuid. This does not change the
311 * currently selected device.
314 struct obd_uuid uuid;
316 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
317 CERROR("No UUID passed!\n");
318 GOTO(out, err = -EINVAL);
320 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
321 CERROR("UUID not NUL terminated!\n");
322 GOTO(out, err = -EINVAL);
325 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
326 obd_str2uuid(&uuid, data->ioc_inlbuf1);
327 dev = class_uuid2dev(&uuid);
330 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
332 GOTO(out, err = -EINVAL);
335 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
337 err = obd_ioctl_popdata((void __user *)arg, data,
344 case OBD_IOC_GETDEVICE: {
345 int index = data->ioc_count;
348 if (!data->ioc_inlbuf1) {
349 CERROR("No buffer passed in ioctl\n");
350 GOTO(out, err = -EINVAL);
352 if (data->ioc_inllen1 < 128) {
353 CERROR("ioctl buffer too small to hold version\n");
354 GOTO(out, err = -EINVAL);
357 obd = class_num2obd(index);
359 GOTO(out, err = -ENOENT);
361 if (obd->obd_stopping)
363 else if (obd->obd_set_up)
365 else if (obd->obd_attached)
369 str = (char *)data->ioc_bulk;
370 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
371 (int)index, status, obd->obd_type->typ_name,
372 obd->obd_name, obd->obd_uuid.uuid,
373 atomic_read(&obd->obd_refcount));
374 err = obd_ioctl_popdata((void __user *)arg, data, len);
381 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
382 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
383 GOTO(out, err = -EINVAL);
384 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
385 GOTO(out, err = -EINVAL);
386 obd = class_name2obd(data->ioc_inlbuf4);
387 } else if (data->ioc_dev < class_devno_max()) {
388 obd = class_num2obd(data->ioc_dev);
390 CERROR("OBD ioctl: No device\n");
391 GOTO(out, err = -EINVAL);
395 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
396 GOTO(out, err = -EINVAL);
398 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
400 if (!obd->obd_set_up || obd->obd_stopping) {
401 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
402 GOTO(out, err = -EINVAL);
406 case OBD_IOC_NO_TRANSNO: {
407 if (!obd->obd_attached) {
408 CERROR("Device %d not attached\n", obd->obd_minor);
409 GOTO(out, err = -ENODEV);
411 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
413 obd->obd_no_transno = 1;
418 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
422 err = obd_ioctl_popdata((void __user *)arg, data, len);
431 obd_ioctl_freedata(buf, len);
433 } /* class_handle_ioctl */
435 #define OBD_INIT_CHECK
436 #ifdef OBD_INIT_CHECK
437 static int obd_init_checks(void)
439 __u64 u64val, div64val;
443 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
445 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
447 u64val = 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 len = snprintf(buf, sizeof(buf), LPX64, u64val);
456 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
460 div64val = OBD_OBJECT_EOF;
461 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
462 if (u64val != OBD_OBJECT_EOF) {
463 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
464 u64val, (int)sizeof(u64val));
467 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
468 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
469 u64val, (int)sizeof(u64val));
472 if (do_div(div64val, 256) != (u64val & 255)) {
473 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
476 if (u64val >> 8 != div64val) {
477 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
478 u64val, div64val, u64val >> 8);
481 len = snprintf(buf, sizeof(buf), LPX64, u64val);
483 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
486 len = snprintf(buf, sizeof(buf), LPU64, u64val);
488 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
491 len = snprintf(buf, sizeof(buf), LPD64, u64val);
493 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
496 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
497 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
498 (__u64)PAGE_CACHE_SIZE);
505 #define obd_init_checks() do {} while(0)
508 static int __init init_obdclass(void)
512 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
513 INIT_LIST_HEAD(&capa_list[i]);
515 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
517 spin_lock_init(&obd_types_lock);
518 obd_zombie_impexp_init();
519 #ifdef CONFIG_PROC_FS
520 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
521 LPROCFS_STATS_FLAG_NONE |
522 LPROCFS_STATS_FLAG_IRQ_SAFE);
523 if (obd_memory == NULL) {
524 CERROR("kmalloc of 'obd_memory' failed\n");
528 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
529 LPROCFS_CNTR_AVGMINMAX,
531 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
532 LPROCFS_CNTR_AVGMINMAX,
533 "pagesused", "pages");
535 err = obd_init_checks();
536 if (err == -EOVERFLOW)
539 class_init_uuidlist();
540 err = class_handle_init();
544 INIT_LIST_HEAD(&obd_types);
546 err = misc_register(&obd_psdev);
548 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
552 /* This struct is already zeroed for us (static global) */
553 for (i = 0; i < class_devno_max(); i++)
556 /* Default the dirty page cache cap to 1/2 of system memory.
557 * For clients with less memory, a larger fraction is needed
558 * for other purposes (mostly for BGL). */
559 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
560 obd_max_dirty_pages = totalram_pages / 4;
562 obd_max_dirty_pages = totalram_pages / 2;
564 err = obd_init_caches();
567 err = class_procfs_init();
571 err = lu_global_init();
575 err = lu_capainfo_init();
579 err = cl_global_init();
583 #ifdef HAVE_SERVER_SUPPORT
584 err = dt_global_init();
588 err = lu_ucred_global_init();
591 #endif /* HAVE_SERVER_SUPPORT */
593 err = llog_info_init();
597 err = lustre_register_fs();
602 void obd_update_maxusage(void)
606 max1 = obd_pages_sum();
607 max2 = obd_memory_sum();
609 spin_lock(&obd_updatemax_lock);
610 if (max1 > obd_max_pages)
611 obd_max_pages = max1;
612 if (max2 > obd_max_alloc)
613 obd_max_alloc = max2;
614 spin_unlock(&obd_updatemax_lock);
616 EXPORT_SYMBOL(obd_update_maxusage);
618 #ifdef CONFIG_PROC_FS
619 __u64 obd_memory_max(void)
623 spin_lock(&obd_updatemax_lock);
625 spin_unlock(&obd_updatemax_lock);
630 __u64 obd_pages_max(void)
634 spin_lock(&obd_updatemax_lock);
636 spin_unlock(&obd_updatemax_lock);
640 #endif /* CONFIG_PROC_FS */
642 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
643 * ifdef to the end of the file to cover module and versioning goo.*/
644 static void cleanup_obdclass(void)
646 __u64 memory_leaked, pages_leaked;
647 __u64 memory_max, pages_max;
650 lustre_unregister_fs();
652 misc_deregister(&obd_psdev);
654 #ifdef HAVE_SERVER_SUPPORT
655 lu_ucred_global_fini();
657 #endif /* HAVE_SERVER_SUPPORT */
662 obd_cleanup_caches();
665 class_procfs_clean();
667 class_handle_cleanup();
668 class_exit_uuidlist();
669 obd_zombie_impexp_stop();
671 memory_leaked = obd_memory_sum();
672 pages_leaked = obd_pages_sum();
674 memory_max = obd_memory_max();
675 pages_max = obd_pages_max();
677 lprocfs_free_stats(&obd_memory);
678 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
679 "obd_memory max: "LPU64", leaked: "LPU64"\n",
680 memory_max, memory_leaked);
681 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
682 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
683 pages_max, pages_leaked);
688 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
689 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
690 MODULE_LICENSE("GPL");
692 cfs_module(obdclass, LUSTRE_VERSION_STRING, init_obdclass, cleanup_obdclass);