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 EXPORT_SYMBOL(obd_max_pages);
58 __u64 obd_max_alloc = 0;
59 EXPORT_SYMBOL(obd_max_alloc);
61 EXPORT_SYMBOL(obd_alloc);
63 EXPORT_SYMBOL(obd_pages);
64 DEFINE_SPINLOCK(obd_updatemax_lock);
66 /* The following are visible and mutable through /proc/sys/lustre/. */
67 unsigned int obd_alloc_fail_rate = 0;
68 EXPORT_SYMBOL(obd_alloc_fail_rate);
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 int obd_max_dirty_pages = 256;
76 EXPORT_SYMBOL(obd_max_dirty_pages);
77 atomic_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 /* Adaptive timeout defs here instead of ptlrpc module for /proc/sys/ access */
88 unsigned int at_min = 0;
89 EXPORT_SYMBOL(at_min);
90 unsigned int at_max = 600;
91 EXPORT_SYMBOL(at_max);
92 unsigned int at_history = 600;
93 EXPORT_SYMBOL(at_history);
94 int at_early_margin = 5;
95 EXPORT_SYMBOL(at_early_margin);
97 EXPORT_SYMBOL(at_extra);
99 atomic_t obd_dirty_transit_pages;
100 EXPORT_SYMBOL(obd_dirty_transit_pages);
102 char obd_jobid_var[JOBSTATS_JOBID_VAR_MAX_LEN + 1] = JOBSTATS_DISABLE;
103 EXPORT_SYMBOL(obd_jobid_var);
105 /* Get jobid of current process by reading the environment variable
106 * stored in between the "env_start" & "env_end" of task struct.
109 * It's better to cache the jobid for later use if there is any
110 * efficient way, the cl_env code probably could be reused for this
113 * If some job scheduler doesn't store jobid in the "env_start/end",
114 * then an upcall could be issued here to get the jobid by utilizing
115 * the userspace tools/api. Then, the jobid must be cached.
117 int lustre_get_jobid(char *jobid)
119 int jobid_len = JOBSTATS_JOBID_SIZE;
122 memset(jobid, 0, JOBSTATS_JOBID_SIZE);
123 /* Jobstats isn't enabled */
124 if (strcmp(obd_jobid_var, JOBSTATS_DISABLE) == 0)
127 /* Use process name + fsuid as jobid */
128 if (strcmp(obd_jobid_var, JOBSTATS_PROCNAME_UID) == 0) {
129 snprintf(jobid, JOBSTATS_JOBID_SIZE, "%s.%u",
131 from_kuid(&init_user_ns, current_fsuid()));
135 rc = cfs_get_environ(obd_jobid_var, jobid, &jobid_len);
137 if (rc == -EOVERFLOW) {
138 /* For the PBS_JOBID and LOADL_STEP_ID keys (which are
139 * variable length strings instead of just numbers), it
140 * might make sense to keep the unique parts for JobID,
141 * instead of just returning an error. That means a
142 * larger temp buffer for cfs_get_environ(), then
143 * truncating the string at some separator to fit into
144 * the specified jobid_len. Fix later if needed. */
146 if (unlikely(!printed)) {
147 LCONSOLE_ERROR_MSG(0x16b, "%s value too large "
148 "for JobID buffer (%d)\n",
149 obd_jobid_var, jobid_len);
153 CDEBUG((rc == -ENOENT || rc == -EINVAL ||
154 rc == -EDEADLK) ? D_INFO : D_ERROR,
155 "Get jobid for (%s) failed: rc = %d\n",
161 EXPORT_SYMBOL(lustre_get_jobid);
163 int obd_alloc_fail(const void *ptr, const char *name, const char *type,
164 size_t size, const char *file, int line)
167 (cfs_rand() & OBD_ALLOC_FAIL_MASK) < obd_alloc_fail_rate) {
168 CERROR("%s%salloc of %s ("LPU64" bytes) failed at %s:%d\n",
169 ptr ? "force " :"", type, name, (__u64)size, file,
171 CERROR(LPU64" total bytes and "LPU64" total pages "
172 "("LPU64" bytes) allocated by Lustre, "
173 "%d total bytes by LNET\n",
175 obd_pages_sum() << PAGE_CACHE_SHIFT,
177 atomic_read(&libcfs_kmemory));
182 EXPORT_SYMBOL(obd_alloc_fail);
184 static inline void obd_data2conn(struct lustre_handle *conn,
185 struct obd_ioctl_data *data)
187 memset(conn, 0, sizeof(*conn));
188 conn->cookie = data->ioc_cookie;
191 static inline void obd_conn2data(struct obd_ioctl_data *data,
192 struct lustre_handle *conn)
194 data->ioc_cookie = conn->cookie;
197 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;
233 /* only for debugging */
234 if (cmd == LIBCFS_IOC_DEBUG_MASK) {
235 debug_data = (struct libcfs_debug_ioctl_data*)arg;
236 libcfs_subsystem_debug = debug_data->subs;
237 libcfs_debug = debug_data->debug;
241 CDEBUG(D_IOCTL, "cmd = %x\n", cmd);
242 if (obd_ioctl_getdata(&buf, &len, (void *)arg)) {
243 CERROR("OBD ioctl: data error\n");
246 data = (struct obd_ioctl_data *)buf;
249 case OBD_IOC_PROCESS_CFG: {
250 struct lustre_cfg *lcfg;
252 if (!data->ioc_plen1 || !data->ioc_pbuf1) {
253 CERROR("No config buffer passed!\n");
254 GOTO(out, err = -EINVAL);
256 OBD_ALLOC(lcfg, data->ioc_plen1);
258 GOTO(out, err = -ENOMEM);
259 err = copy_from_user(lcfg, data->ioc_pbuf1,
262 err = lustre_cfg_sanity_check(lcfg, data->ioc_plen1);
264 err = class_process_config(lcfg);
266 OBD_FREE(lcfg, data->ioc_plen1);
270 case OBD_GET_VERSION:
271 if (!data->ioc_inlbuf1) {
272 CERROR("No buffer passed in ioctl\n");
273 GOTO(out, err = -EINVAL);
276 if (strlen(BUILD_VERSION) + 1 > data->ioc_inllen1) {
277 CERROR("ioctl buffer too small to hold version\n");
278 GOTO(out, err = -EINVAL);
281 memcpy(data->ioc_bulk, BUILD_VERSION,
282 strlen(BUILD_VERSION) + 1);
284 err = obd_ioctl_popdata((void *)arg, data, len);
289 case OBD_IOC_NAME2DEV: {
290 /* Resolve a device name. This does not change the
291 * currently selected device.
295 dev = class_resolve_dev_name(data->ioc_inllen1,
299 GOTO(out, err = -EINVAL);
301 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
307 case OBD_IOC_UUID2DEV: {
308 /* Resolve a device uuid. This does not change the
309 * currently selected device.
312 struct obd_uuid uuid;
314 if (!data->ioc_inllen1 || !data->ioc_inlbuf1) {
315 CERROR("No UUID passed!\n");
316 GOTO(out, err = -EINVAL);
318 if (data->ioc_inlbuf1[data->ioc_inllen1 - 1] != 0) {
319 CERROR("UUID not NUL terminated!\n");
320 GOTO(out, err = -EINVAL);
323 CDEBUG(D_IOCTL, "device name %s\n", data->ioc_inlbuf1);
324 obd_str2uuid(&uuid, data->ioc_inlbuf1);
325 dev = class_uuid2dev(&uuid);
328 CDEBUG(D_IOCTL, "No device for UUID %s!\n",
330 GOTO(out, err = -EINVAL);
333 CDEBUG(D_IOCTL, "device name %s, dev %d\n", data->ioc_inlbuf1,
335 err = obd_ioctl_popdata((void *)arg, data, sizeof(*data));
341 case OBD_IOC_CLOSE_UUID: {
342 CDEBUG(D_IOCTL, "closing all connections to uuid %s (NOOP)\n",
347 case OBD_IOC_GETDEVICE: {
348 int index = data->ioc_count;
351 if (!data->ioc_inlbuf1) {
352 CERROR("No buffer passed in ioctl\n");
353 GOTO(out, err = -EINVAL);
355 if (data->ioc_inllen1 < 128) {
356 CERROR("ioctl buffer too small to hold version\n");
357 GOTO(out, err = -EINVAL);
360 obd = class_num2obd(index);
362 GOTO(out, err = -ENOENT);
364 if (obd->obd_stopping)
366 else if (obd->obd_set_up)
368 else if (obd->obd_attached)
372 str = (char *)data->ioc_bulk;
373 snprintf(str, len - sizeof(*data), "%3d %s %s %s %s %d",
374 (int)index, status, obd->obd_type->typ_name,
375 obd->obd_name, obd->obd_uuid.uuid,
376 atomic_read(&obd->obd_refcount));
377 err = obd_ioctl_popdata((void *)arg, data, len);
384 if (data->ioc_dev == OBD_DEV_BY_DEVNAME) {
385 if (data->ioc_inllen4 <= 0 || data->ioc_inlbuf4 == NULL)
386 GOTO(out, err = -EINVAL);
387 if (strnlen(data->ioc_inlbuf4, MAX_OBD_NAME) >= MAX_OBD_NAME)
388 GOTO(out, err = -EINVAL);
389 obd = class_name2obd(data->ioc_inlbuf4);
390 } else if (data->ioc_dev < class_devno_max()) {
391 obd = class_num2obd(data->ioc_dev);
393 CERROR("OBD ioctl: No device\n");
394 GOTO(out, err = -EINVAL);
398 CERROR("OBD ioctl : No Device %d\n", data->ioc_dev);
399 GOTO(out, err = -EINVAL);
401 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
403 if (!obd->obd_set_up || obd->obd_stopping) {
404 CERROR("OBD ioctl: device not setup %d \n", data->ioc_dev);
405 GOTO(out, err = -EINVAL);
409 case OBD_IOC_NO_TRANSNO: {
410 if (!obd->obd_attached) {
411 CERROR("Device %d not attached\n", obd->obd_minor);
412 GOTO(out, err = -ENODEV);
414 CDEBUG(D_HA, "%s: disabling committed-transno notification\n",
416 obd->obd_no_transno = 1;
421 err = obd_iocontrol(cmd, obd->obd_self_export, len, data, NULL);
425 err = obd_ioctl_popdata((void *)arg, data, len);
434 obd_ioctl_freedata(buf, len);
436 } /* class_handle_ioctl */
438 extern struct miscdevice obd_psdev;
440 #define OBD_INIT_CHECK
441 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 & ~CFS_PAGE_MASK) >= PAGE_CACHE_SIZE) {
501 CWARN("mask failed: u64val "LPU64" >= "LPU64"\n", u64val,
502 (__u64)PAGE_CACHE_SIZE);
509 extern spinlock_t obd_types_lock;
511 extern int class_procfs_init(void);
512 extern int class_procfs_clean(void);
514 static inline int class_procfs_init(void)
516 static inline int class_procfs_clean(void)
520 static int __init init_obdclass(void)
523 int lustre_register_fs(void);
525 for (i = CAPA_SITE_CLIENT; i < CAPA_SITE_MAX; i++)
526 INIT_LIST_HEAD(&capa_list[i]);
528 LCONSOLE_INFO("Lustre: Build Version: "BUILD_VERSION"\n");
530 spin_lock_init(&obd_types_lock);
531 obd_zombie_impexp_init();
533 obd_memory = lprocfs_alloc_stats(OBD_STATS_NUM,
534 LPROCFS_STATS_FLAG_NONE |
535 LPROCFS_STATS_FLAG_IRQ_SAFE);
536 if (obd_memory == NULL) {
537 CERROR("kmalloc of 'obd_memory' failed\n");
541 lprocfs_counter_init(obd_memory, OBD_MEMORY_STAT,
542 LPROCFS_CNTR_AVGMINMAX,
544 lprocfs_counter_init(obd_memory, OBD_MEMORY_PAGES_STAT,
545 LPROCFS_CNTR_AVGMINMAX,
546 "pagesused", "pages");
548 err = obd_init_checks();
549 if (err == -EOVERFLOW)
552 class_init_uuidlist();
553 err = class_handle_init();
557 INIT_LIST_HEAD(&obd_types);
559 err = misc_register(&obd_psdev);
561 CERROR("cannot register %d err %d\n", OBD_DEV_MINOR, err);
565 /* This struct is already zeroed for us (static global) */
566 for (i = 0; i < class_devno_max(); i++)
569 /* Default the dirty page cache cap to 1/2 of system memory.
570 * For clients with less memory, a larger fraction is needed
571 * for other purposes (mostly for BGL). */
572 if (totalram_pages <= 512 << (20 - PAGE_CACHE_SHIFT))
573 obd_max_dirty_pages = totalram_pages / 4;
575 obd_max_dirty_pages = totalram_pages / 2;
577 err = obd_init_caches();
583 err = class_procfs_init();
587 err = lu_global_init();
591 err = cl_global_init();
596 err = llog_info_init();
600 err = lustre_register_fs();
605 void obd_update_maxusage(void)
609 max1 = obd_pages_sum();
610 max2 = obd_memory_sum();
612 spin_lock(&obd_updatemax_lock);
613 if (max1 > obd_max_pages)
614 obd_max_pages = max1;
615 if (max2 > obd_max_alloc)
616 obd_max_alloc = max2;
617 spin_unlock(&obd_updatemax_lock);
619 EXPORT_SYMBOL(obd_update_maxusage);
622 __u64 obd_memory_max(void)
626 spin_lock(&obd_updatemax_lock);
628 spin_unlock(&obd_updatemax_lock);
632 EXPORT_SYMBOL(obd_memory_max);
634 __u64 obd_pages_max(void)
638 spin_lock(&obd_updatemax_lock);
640 spin_unlock(&obd_updatemax_lock);
644 EXPORT_SYMBOL(obd_pages_max);
647 /* liblustre doesn't call cleanup_obdclass, apparently. we carry on in this
648 * ifdef to the end of the file to cover module and versioning goo.*/
649 static void cleanup_obdclass(void)
652 int lustre_unregister_fs(void);
653 __u64 memory_leaked, pages_leaked;
654 __u64 memory_max, pages_max;
656 lustre_unregister_fs();
658 misc_deregister(&obd_psdev);
659 for (i = 0; i < class_devno_max(); i++) {
660 struct obd_device *obd = class_num2obd(i);
661 if (obd && obd->obd_set_up &&
662 OBT(obd) && OBP(obd, detach)) {
663 /* XXX should this call generic detach otherwise? */
664 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
665 OBP(obd, detach)(obd);
672 obd_cleanup_caches();
675 class_procfs_clean();
677 class_handle_cleanup();
678 class_exit_uuidlist();
679 obd_zombie_impexp_stop();
681 memory_leaked = obd_memory_sum();
682 pages_leaked = obd_pages_sum();
684 memory_max = obd_memory_max();
685 pages_max = obd_pages_max();
687 lprocfs_free_stats(&obd_memory);
688 CDEBUG((memory_leaked) ? D_ERROR : D_INFO,
689 "obd_memory max: "LPU64", leaked: "LPU64"\n",
690 memory_max, memory_leaked);
691 CDEBUG((pages_leaked) ? D_ERROR : D_INFO,
692 "obd_memory_pages max: "LPU64", leaked: "LPU64"\n",
693 pages_max, pages_leaked);
696 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
697 MODULE_DESCRIPTION("Lustre Class Driver Build Version: " BUILD_VERSION);
698 MODULE_LICENSE("GPL");
699 MODULE_VERSION(LUSTRE_VERSION_STRING);
701 module_init(init_obdclass);
702 module_exit(cleanup_obdclass);