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, 2012, 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 <linux/lnet/lnetctl.h>
43 #include <lustre_debug.h>
44 #include <lprocfs_status.h>
45 #include <lustre/lustre_build_version.h>
46 #include <linux/list.h>
47 #include <cl_object.h>
48 #include "llog_internal.h"
51 struct obd_device *obd_devs[MAX_OBD_DEVICES];
52 EXPORT_SYMBOL(obd_devs);
53 struct list_head obd_types;
54 DEFINE_RWLOCK(obd_dev_lock);
56 __u64 obd_max_pages = 0;
57 __u64 obd_max_alloc = 0;
58 DEFINE_SPINLOCK(obd_updatemax_lock);
60 /* The following are visible and mutable through /proc/sys/lustre/. */
61 unsigned int obd_alloc_fail_rate = 0;
62 EXPORT_SYMBOL(obd_alloc_fail_rate);
63 unsigned int obd_debug_peer_on_timeout;
64 EXPORT_SYMBOL(obd_debug_peer_on_timeout);
65 unsigned int obd_dump_on_timeout;
66 EXPORT_SYMBOL(obd_dump_on_timeout);
67 unsigned int obd_dump_on_eviction;
68 EXPORT_SYMBOL(obd_dump_on_eviction);
69 unsigned int obd_max_dirty_pages = 256;
70 EXPORT_SYMBOL(obd_max_dirty_pages);
71 atomic_t obd_dirty_pages;
72 EXPORT_SYMBOL(obd_dirty_pages);
73 unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */
74 EXPORT_SYMBOL(obd_timeout);
75 unsigned int ldlm_timeout = LDLM_TIMEOUT_DEFAULT; /* seconds */
76 EXPORT_SYMBOL(ldlm_timeout);
77 unsigned int obd_timeout_set;
78 EXPORT_SYMBOL(obd_timeout_set);
79 unsigned int ldlm_timeout_set;
80 EXPORT_SYMBOL(ldlm_timeout_set);
81 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
82 unsigned int at_min = 0;
83 EXPORT_SYMBOL(at_min);
84 unsigned int at_max = 600;
85 EXPORT_SYMBOL(at_max);
86 unsigned int at_history = 600;
87 EXPORT_SYMBOL(at_history);
88 int at_early_margin = 5;
89 EXPORT_SYMBOL(at_early_margin);
91 EXPORT_SYMBOL(at_extra);
93 atomic_t obd_dirty_transit_pages;
94 EXPORT_SYMBOL(obd_dirty_transit_pages);
96 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
97 EXPORT_SYMBOL(obd_jobid_var);
99 /* Get jobid of current process by reading the environment variable
100 * stored in between the "env_start" & "env_end" of task struct.
103 * It's better to cache the jobid for later use if there is any
104 * efficient way, the cl_env code probably could be reused for this
107 * If some job scheduler doesn't store jobid in the "env_start/end",
108 * then an upcall could be issued here to get the jobid by utilizing
109 * the userspace tools/api. Then, the jobid must be cached.
111 int lustre_get_jobid(char *jobid)
113 int jobid_len = JOBSTATS_JOBID_SIZE;
116 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
117 /* Jobstats isn't enabled */
118 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
121 /* Use process name + fsuid as jobid */
122 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
123 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
125 from_kuid(&init_user_ns, current_fsuid()));
129 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
131 if (rc == -EOVERFLOW) {
132 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
133 * variable length strings instead of just numbers), it
134 * might make sense to keep the unique parts for JobID,
135 * instead of just returning an error. That means a
136 * larger temp buffer for cfs_get_environ(), then
137 * truncating the string at some separator to fit into
138 * the specified jobid_len. Fix later if needed. */
140 if (unlikely(!printed)) {
141 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
142 "for JobID buffer (%d)\n",
143 obd_jobid_var, jobid_len);
147 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
148 rc == -EDEADLK) ? D_INFO : D_ERROR,
149 "Get jobid for (%s) failed: rc = %d\n",
155 EXPORT_SYMBOL(lustre_get_jobid);
157 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
158 size_t size, const char *file, int line)
161 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
162 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
163 ptr ? "force " :"", type, name, (__u64)size, file,
165 CERROR(LPU64" total bytes and "LPU64" total pages "
166 "("LPU64" bytes) allocated by Lustre, "
167 "%d total bytes by LNET\n",
169 obd_pages_sum() << PAGE_CACHE_SHIFT,
171 atomic_read(&libcfs_kmemory));
176 EXPORT_SYMBOL(obd_alloc_fail);
178 static inline void obd_data2conn(struct lustre_handle *conn,
179 struct obd_ioctl_data *data)
181 memset(conn, 0, sizeof(*conn));
182 conn->cookie = data->ioc_cookie;
185 static inline void obd_conn2data(struct obd_ioctl_data *data,
186 struct lustre_handle *conn)
188 data->ioc_cookie = conn->cookie;
191 int class_resolve_dev_name(__u32 len, const char *name)
197 CERROR("No name passed,!\n");
198 GOTO(out, rc = -EINVAL);
200 if (name[len - 1] != 0) {
201 CERROR("Name not nul terminated!\n");
202 GOTO(out, rc = -EINVAL);
205 CDEBUG(D_IOCTL, "device name %s\n", name);
206 dev = class_name2dev(name);
208 CDEBUG(D_IOCTL, "No device for name %s!\n", name);
209 GOTO(out, rc = -EINVAL);
212 CDEBUG(D_IOCTL, "device name %s, dev %d\n", name, dev);
219 int class_handle_ioctl(unsigned int cmd, unsigned long arg)
222 struct obd_ioctl_data *data;
223 struct libcfs_debug_ioctl_data *debug_data;
224 struct obd_device *obd = NULL;
225 int err = 0, len = 0;
227 /* only for debugging */
228 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
229 debug_data = (struct libcfs_debug_ioctl_data*)arg;
230 libcfs_subsystem_debug = debug_data->subs;
231 libcfs_debug = debug_data->debug;
235 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
236 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
237 CERROR("OBD ioctl: data error\n");
240 data = (struct obd_ioctl_data *)buf;
243 case OBD_IOC_PROCESS_CFG: {
244 struct lustre_cfg *lcfg;
246 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
247 CERROR("No config buffer passed!\n");
248 GOTO(out, err = -EINVAL);
250 OBD_ALLOC(lcfg, data->ioc_plen1);
252 GOTO(out, err = -ENOMEM);
253 err = copy_from_user(lcfg, data->ioc_pbuf1,
256 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
258 err = class_process_config(lcfg);
260 OBD_FREE(lcfg, data->ioc_plen1);
264 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(BUILD_VERSION) + 1 > data->ioc_inllen1) {
271 CERROR("ioctl buffer too small to hold version\n");
272 GOTO(out, err = -EINVAL);
275 memcpy(data->ioc_bulk, BUILD_VERSION,
276 strlen(BUILD_VERSION) + 1);
278 err = obd_ioctl_popdata((void *)arg, data, len);
283 case OBD_IOC_NAME2DEV: {
284 /* Resolve a device name. This does not change the
285 * currently selected device.
289 dev = class_resolve_dev_name(data->ioc_inllen1,
293 GOTO(out, err = -EINVAL);
295 err = obd_ioctl_popdata((void *)arg, data, sizeof(*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 *)arg, data, sizeof(*data));
335 case OBD_IOC_CLOSE_UUID: {
336 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
341 case OBD_IOC_GETDEVICE: {
342 int index = data->ioc_count;
345 if (!data->ioc_inlbuf1) {
346 CERROR("No buffer passed in ioctl\n");
347 GOTO(out, err = -EINVAL);
349 if (data->ioc_inllen1 < 128) {
350 CERROR("ioctl buffer too small to hold version\n");
351 GOTO(out, err = -EINVAL);
354 obd = class_num2obd(index);
356 GOTO(out, err = -ENOENT);
358 if (obd->obd_stopping)
360 else if (obd->obd_set_up)
362 else if (obd->obd_attached)
366 str = (char *)data->ioc_bulk;
367 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
368 (int)index, status, obd->obd_type->typ_name,
369 obd->obd_name, obd->obd_uuid.uuid,
370 atomic_read(&obd->obd_refcount));
371 err = obd_ioctl_popdata((void *)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 err = obd_ioctl_popdata((void *)arg, data, len);
428 obd_ioctl_freedata(buf, len);
430 } /* class_handle_ioctl */
432 extern struct miscdevice obd_psdev;
434 #define OBD_INIT_CHECK
435 int obd_init_checks(void)
437 __u64 u64val, div64val;
441 CDEBUG(D_INFO, "LPU64=%s, LPD64=%s, LPX64=%s\n", LPU64, LPD64, LPX64);
443 CDEBUG(D_INFO, "OBD_OBJECT_EOF = "LPX64"\n", (__u64)OBD_OBJECT_EOF);
445 u64val = OBD_OBJECT_EOF;
446 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
447 if (u64val != OBD_OBJECT_EOF) {
448 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
449 u64val, (int)sizeof(u64val));
452 len = snprintf(buf, sizeof(buf), LPX64, u64val);
454 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
458 div64val = OBD_OBJECT_EOF;
459 CDEBUG(D_INFO, "u64val OBD_OBJECT_EOF = "LPX64"\n", u64val);
460 if (u64val != OBD_OBJECT_EOF) {
461 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
462 u64val, (int)sizeof(u64val));
465 if (u64val >> 8 != OBD_OBJECT_EOF >> 8) {
466 CERROR("__u64 "LPX64"(%d) != 0xffffffffffffffff\n",
467 u64val, (int)sizeof(u64val));
470 if (do_div(div64val, 256) != (u64val & 255)) {
471 CERROR("do_div("LPX64",256) != "LPU64"\n", u64val, u64val &255);
474 if (u64val >> 8 != div64val) {
475 CERROR("do_div("LPX64",256) "LPU64" != "LPU64"\n",
476 u64val, div64val, u64val >> 8);
479 len = snprintf(buf, sizeof(buf), LPX64, u64val);
481 CWARN("LPX64 wrong length! strlen(%s)=%d != 18\n", buf, len);
484 len = snprintf(buf, sizeof(buf), LPU64, u64val);
486 CWARN("LPU64 wrong length! strlen(%s)=%d != 20\n", buf, len);
489 len = snprintf(buf, sizeof(buf), LPD64, u64val);
491 CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len);
494 if ((u64val & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
495 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
496 (__u64)PAGE_CACHE_SIZE);
503 extern spinlock_t obd_types_lock;
504 extern int class_procfs_init(void);
505 extern int class_procfs_clean(void);
507 static int __init init_obdclass(void)
510 int lustre_register_fs(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();
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();
570 err = class_procfs_init();
574 err = lu_global_init();
578 err = cl_global_init();
583 err = llog_info_init();
587 err = lustre_register_fs();
592 void obd_update_maxusage(void)
596 max1 = obd_pages_sum();
597 max2 = obd_memory_sum();
599 spin_lock(&obd_updatemax_lock);
600 if (max1 > obd_max_pages)
601 obd_max_pages = max1;
602 if (max2 > obd_max_alloc)
603 obd_max_alloc = max2;
604 spin_unlock(&obd_updatemax_lock);
606 EXPORT_SYMBOL(obd_update_maxusage);
609 __u64 obd_memory_max(void)
613 spin_lock(&obd_updatemax_lock);
615 spin_unlock(&obd_updatemax_lock);
619 EXPORT_SYMBOL(obd_memory_max);
621 __u64 obd_pages_max(void)
625 spin_lock(&obd_updatemax_lock);
627 spin_unlock(&obd_updatemax_lock);
631 EXPORT_SYMBOL(obd_pages_max);
634 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
635 * ifdef to the end of the file to cover module and versioning goo.*/
636 static void cleanup_obdclass(void)
639 int lustre_unregister_fs(void);
640 __u64 memory_leaked, pages_leaked;
641 __u64 memory_max, pages_max;
643 lustre_unregister_fs();
645 misc_deregister(&obd_psdev);
646 for (i = 0; i < class_devno_max(); i++) {
647 struct obd_device *obd = class_num2obd(i);
648 if (obd && obd->obd_set_up &&
649 OBT(obd) && OBP(obd, detach)) {
650 /* XXX should this call generic detach otherwise? */
651 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
652 OBP(obd, detach)(obd);
659 obd_cleanup_caches();
662 class_procfs_clean();
664 class_handle_cleanup();
665 class_exit_uuidlist();
666 obd_zombie_impexp_stop();
668 memory_leaked = obd_memory_sum();
669 pages_leaked = obd_pages_sum();
671 memory_max = obd_memory_max();
672 pages_max = obd_pages_max();
674 lprocfs_free_stats(&obd_memory);
675 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
676 "obd_memory max: "LPU64", leaked: "LPU64"\n",
677 memory_max, memory_leaked);
678 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
679 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
680 pages_max, pages_leaked);
683 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
684 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
685 MODULE_LICENSE("GPL");
686 MODULE_VERSION(LUSTRE_VERSION_STRING);
688 module_init(init_obdclass);
689 module_exit(cleanup_obdclass);