1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001 Cluster File Systems, Inc. <braam@clusterfs.com>
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 * Basic Lustre library routines.
31 # include <asm/semaphore.h>
33 #include <linux/types.h>
34 #include <linux/portals_lib.h>
35 #include <linux/kp30.h> /* XXX just for LASSERT! */
36 #include <linux/lustre_idl.h>
39 #if BITS_PER_LONG > 32
51 struct ptlrpc_request;
56 #include <linux/lustre_ha.h>
57 #include <linux/lustre_net.h>
60 int target_handle_connect(struct ptlrpc_request *req, svc_handler_t handler);
61 int target_handle_disconnect(struct ptlrpc_request *req);
62 int target_handle_reconnect(struct lustre_handle *conn, struct obd_export *exp,
63 struct obd_uuid *cluuid);
64 int target_revoke_connection(struct recovd_data *rd, int phase);
66 #define OBD_RECOVERY_TIMEOUT (obd_timeout * 5 * HZ / 2) /* *waves hands* */
67 void target_start_recovery_timer(struct obd_device *obd, svc_handler_t handler);
68 void target_abort_recovery(void *data);
69 int target_queue_recovery_request(struct ptlrpc_request *req,
70 struct obd_device *obd);
71 int target_queue_final_reply(struct ptlrpc_request *req, int rc);
74 int client_obd_connect(struct lustre_handle *conn, struct obd_device *obd,
75 struct obd_uuid *cluuid, struct recovd_obd *recovd,
76 ptlrpc_recovery_cb_t recover);
77 int client_obd_disconnect(struct lustre_handle *conn);
78 int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf);
79 int client_sanobd_setup(struct obd_device *obddev, obd_count len, void *buf);
80 int client_obd_cleanup(struct obd_device * obddev);
81 struct client_obd *client_conn2cli(struct lustre_handle *conn);
82 struct obd_device *client_tgtuuid2obd(struct obd_uuid *tgtuuid);
85 int obd_self_statfs(struct obd_device *dev, struct statfs *sfs);
90 struct task_struct *l_owner;
91 struct semaphore l_sem;
95 void l_lock_init(struct lustre_lock *);
96 void l_lock(struct lustre_lock *);
97 void l_unlock(struct lustre_lock *);
98 int l_has_lock(struct lustre_lock *);
100 #define CB_PHASE_START 12
101 #define CB_PHASE_FINISH 13
103 /* This list head doesn't need to be locked, because it's only manipulated by
104 * one thread at a time. */
106 struct list_head brw_desc_head; /* list of ptlrpc_bulk_desc */
107 wait_queue_head_t brw_waitq;
108 atomic_t brw_refcount;
111 int (*brw_callback)(struct obd_brw_set *, int phase);
117 void push_ctxt(struct obd_run_ctxt *save, struct obd_run_ctxt *new_ctx,
118 struct obd_ucred *cred);
119 void pop_ctxt(struct obd_run_ctxt *saved, struct obd_run_ctxt *new_ctx,
120 struct obd_ucred *cred);
121 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode);
122 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode);
123 int lustre_fread(struct file *file, char *str, int len, loff_t *off);
124 int lustre_fwrite(struct file *file, const char *str, int len, loff_t *off);
125 int lustre_fsync(struct file *file);
129 static inline void l_dput(struct dentry *de)
131 if (!de || IS_ERR(de))
133 shrink_dcache_parent(de);
134 LASSERT(atomic_read(&de->d_count) > 0);
138 static inline void ll_sleep(int t)
140 set_current_state(TASK_INTERRUPTIBLE);
141 schedule_timeout(t * HZ);
142 set_current_state(TASK_RUNNING);
146 /* FIXME: This needs to validate pointers and cookies */
147 static inline void *lustre_handle2object(struct lustre_handle *handle)
150 return (void *)(unsigned long)(handle->addr);
154 static inline void ldlm_object2handle(void *object, struct lustre_handle *handle)
156 handle->addr = (__u64)(unsigned long)object;
159 #include <linux/portals_lib.h>
164 #define OBD_IOCTL_VERSION 0x00010002
166 struct obd_ioctl_data {
168 uint32_t ioc_version;
175 struct obdo ioc_obdo1;
176 struct obdo ioc_obdo2;
181 uint32_t ioc_command;
186 /* buffers the kernel will treat as user pointers */
192 /* two inline buffers */
193 uint32_t ioc_inllen1;
195 uint32_t ioc_inllen2;
197 uint32_t ioc_inllen3;
203 struct obd_ioctl_hdr {
205 uint32_t ioc_version;
208 static inline int obd_ioctl_packlen(struct obd_ioctl_data *data)
210 int len = size_round(sizeof(struct obd_ioctl_data));
211 len += size_round(data->ioc_inllen1);
212 len += size_round(data->ioc_inllen2);
213 len += size_round(data->ioc_inllen3);
218 static inline int obd_ioctl_is_invalid(struct obd_ioctl_data *data)
220 if (data->ioc_len > (1<<30)) {
221 printk("OBD ioctl: ioc_len larger than 1<<30\n");
224 if (data->ioc_inllen1 > (1<<30)) {
225 printk("OBD ioctl: ioc_inllen1 larger than 1<<30\n");
228 if (data->ioc_inllen2 > (1<<30)) {
229 printk("OBD ioctl: ioc_inllen2 larger than 1<<30\n");
233 if (data->ioc_inllen3 > (1<<30)) {
234 printk("OBD ioctl: ioc_inllen3 larger than 1<<30\n");
237 if (data->ioc_inlbuf1 && !data->ioc_inllen1) {
238 printk("OBD ioctl: inlbuf1 pointer but 0 length\n");
241 if (data->ioc_inlbuf2 && !data->ioc_inllen2) {
242 printk("OBD ioctl: inlbuf2 pointer but 0 length\n");
245 if (data->ioc_inlbuf3 && !data->ioc_inllen3) {
246 printk("OBD ioctl: inlbuf3 pointer but 0 length\n");
249 if (data->ioc_pbuf1 && !data->ioc_plen1) {
250 printk("OBD ioctl: pbuf1 pointer but 0 length\n");
253 if (data->ioc_pbuf2 && !data->ioc_plen2) {
254 printk("OBD ioctl: pbuf2 pointer but 0 length\n");
258 if (data->ioc_inllen1 && !data->ioc_inlbuf1) {
259 printk("OBD ioctl: inllen1 set but NULL pointer\n");
262 if (data->ioc_inllen2 && !data->ioc_inlbuf2) {
263 printk("OBD ioctl: inllen2 set but NULL pointer\n");
266 if (data->ioc_inllen3 && !data->ioc_inlbuf3) {
267 printk("OBD ioctl: inllen3 set but NULL pointer\n");
271 if (data->ioc_plen1 && !data->ioc_pbuf1) {
272 printk("OBD ioctl: plen1 set but NULL pointer\n");
275 if (data->ioc_plen2 && !data->ioc_pbuf2) {
276 printk("OBD ioctl: plen2 set but NULL pointer\n");
279 if (obd_ioctl_packlen(data) != data->ioc_len) {
280 printk("OBD ioctl: packlen exceeds ioc_len (%d != %d)\n",
281 obd_ioctl_packlen(data), data->ioc_len);
285 if (data->ioc_inllen1 &&
286 data->ioc_bulk[data->ioc_inllen1 - 1] != '\0') {
287 printk("OBD ioctl: inlbuf1 not 0 terminated\n");
290 if (data->ioc_inllen2 &&
291 data->ioc_bulk[size_round(data->ioc_inllen1) + data->ioc_inllen2 - 1] != '\0') {
292 printk("OBD ioctl: inlbuf2 not 0 terminated\n");
295 if (data->ioc_inllen3 &&
296 data->ioc_bulk[size_round(data->ioc_inllen1) + size_round(data->ioc_inllen2)
297 + data->ioc_inllen3 - 1] != '\0') {
298 printk("OBD ioctl: inlbuf3 not 0 terminated\n");
306 static inline int obd_ioctl_pack(struct obd_ioctl_data *data, char **pbuf,
310 struct obd_ioctl_data *overlay;
311 data->ioc_len = obd_ioctl_packlen(data);
312 data->ioc_version = OBD_IOCTL_VERSION;
314 if (*pbuf && data->ioc_len > max)
317 *pbuf = malloc(data->ioc_len);
321 overlay = (struct obd_ioctl_data *)*pbuf;
322 memcpy(*pbuf, data, sizeof(*data));
324 ptr = overlay->ioc_bulk;
325 if (data->ioc_inlbuf1)
326 LOGL(data->ioc_inlbuf1, data->ioc_inllen1, ptr);
327 if (data->ioc_inlbuf2)
328 LOGL(data->ioc_inlbuf2, data->ioc_inllen2, ptr);
329 if (data->ioc_inlbuf3)
330 LOGL(data->ioc_inlbuf3, data->ioc_inllen3, ptr);
331 if (obd_ioctl_is_invalid(overlay))
337 static inline int obd_ioctl_unpack(struct obd_ioctl_data *data, char *pbuf,
341 struct obd_ioctl_data *overlay;
345 overlay = (struct obd_ioctl_data *)pbuf;
347 /* Preserve the caller's buffer pointers */
348 overlay->ioc_inlbuf1 = data->ioc_inlbuf1;
349 overlay->ioc_inlbuf2 = data->ioc_inlbuf2;
350 overlay->ioc_inlbuf3 = data->ioc_inlbuf3;
352 memcpy(data, pbuf, sizeof(*data));
354 ptr = overlay->ioc_bulk;
355 if (data->ioc_inlbuf1)
356 LOGU(data->ioc_inlbuf1, data->ioc_inllen1, ptr);
357 if (data->ioc_inlbuf2)
358 LOGU(data->ioc_inlbuf2, data->ioc_inllen2, ptr);
359 if (data->ioc_inlbuf3)
360 LOGU(data->ioc_inlbuf3, data->ioc_inllen3, ptr);
366 #include <linux/obd_support.h>
368 /* buffer MUST be at least the size of obd_ioctl_hdr */
369 static inline int obd_ioctl_getdata(char **buf, int *len, void *arg)
371 struct obd_ioctl_hdr hdr;
372 struct obd_ioctl_data *data;
376 err = copy_from_user(&hdr, (void *)arg, sizeof(hdr));
382 if (hdr.ioc_version != OBD_IOCTL_VERSION) {
383 printk("OBD: version mismatch kernel vs application\n");
387 if (hdr.ioc_len > 8192) {
388 printk("OBD: user buffer exceeds 8192 max buffer\n");
392 if (hdr.ioc_len < sizeof(struct obd_ioctl_data)) {
393 printk("OBD: user buffer too small for ioctl\n");
397 OBD_ALLOC(*buf, hdr.ioc_len);
399 CERROR("Cannot allocate control buffer of len %d\n",
404 data = (struct obd_ioctl_data *)*buf;
406 err = copy_from_user(*buf, (void *)arg, hdr.ioc_len);
412 if (obd_ioctl_is_invalid(data)) {
413 printk("OBD: ioctl not correctly formatted\n");
417 if (data->ioc_inllen1) {
418 data->ioc_inlbuf1 = &data->ioc_bulk[0];
421 if (data->ioc_inllen2) {
422 data->ioc_inlbuf2 = &data->ioc_bulk[0] +
423 size_round(data->ioc_inllen1);
426 if (data->ioc_inllen3) {
427 data->ioc_inlbuf3 = &data->ioc_bulk[0] +
428 size_round(data->ioc_inllen1) +
429 size_round(data->ioc_inllen2);
436 #define OBD_IOC_CREATE _IOR ('f', 101, long)
437 #define OBD_IOC_SETUP _IOW ('f', 102, long)
438 #define OBD_IOC_CLEANUP _IO ('f', 103 )
439 #define OBD_IOC_DESTROY _IOW ('f', 104, long)
440 #define OBD_IOC_PREALLOCATE _IOWR('f', 105, long)
442 #define OBD_IOC_SETATTR _IOW ('f', 107, long)
443 #define OBD_IOC_GETATTR _IOR ('f', 108, long)
444 #define OBD_IOC_READ _IOWR('f', 109, long)
445 #define OBD_IOC_WRITE _IOWR('f', 110, long)
446 #define OBD_IOC_CONNECT _IOR ('f', 111, long)
447 #define OBD_IOC_DISCONNECT _IOW ('f', 112, long)
448 #define OBD_IOC_STATFS _IOWR('f', 113, long)
449 #define OBD_IOC_SYNC _IOR ('f', 114, long)
450 #define OBD_IOC_READ2 _IOWR('f', 115, long)
451 #define OBD_IOC_FORMAT _IOWR('f', 116, long)
452 #define OBD_IOC_PARTITION _IOWR('f', 117, long)
453 #define OBD_IOC_ATTACH _IOWR('f', 118, long)
454 #define OBD_IOC_DETACH _IOWR('f', 119, long)
455 #define OBD_IOC_COPY _IOWR('f', 120, long)
456 #define OBD_IOC_MIGR _IOWR('f', 121, long)
457 #define OBD_IOC_PUNCH _IOWR('f', 122, long)
458 #define OBD_IOC_DEVICE _IOWR('f', 123, long)
459 #define OBD_IOC_MODULE_DEBUG _IOWR('f', 124, long)
460 #define OBD_IOC_BRW_READ _IOWR('f', 125, long)
461 #define OBD_IOC_BRW_WRITE _IOWR('f', 126, long)
462 #define OBD_IOC_NAME2DEV _IOWR('f', 127, long)
463 #define OBD_IOC_NEWDEV _IOWR('f', 128, long)
464 #define OBD_IOC_LIST _IOWR('f', 129, long)
465 #define OBD_IOC_UUID2DEV _IOWR('f', 130, long)
467 #define OBD_IOC_RECOVD_NEWCONN _IOWR('f', 131, long)
468 #define OBD_IOC_LOV_SET_CONFIG _IOWR('f', 132, long)
469 #define OBD_IOC_LOV_GET_CONFIG _IOWR('f', 133, long)
470 #define OBD_IOC_LOV_CONFIG OBD_IOC_LOV_SET_CONFIG
472 #define OBD_IOC_OPEN _IOWR('f', 134, long)
473 #define OBD_IOC_CLOSE _IOWR('f', 135, long)
475 #define OBD_IOC_RECOVD_FAILCONN _IOWR('f', 136, long)
477 #define OBD_IOC_DEC_FS_USE_COUNT _IO ('f', 139 )
478 #define OBD_IOC_NO_TRANSNO _IOW ('f', 140, long)
479 #define OBD_IOC_SET_READONLY _IOW ('f', 141, long)
481 #define OBD_GET_VERSION _IOWR ('f', 144, long)
483 #define OBD_IOC_ADD_UUID _IOWR ('f', 145, long)
484 #define OBD_IOC_DEL_UUID _IOWR ('f', 146, long)
485 #define OBD_IOC_CLOSE_UUID _IOWR ('f', 147, long)
487 #define ECHO_IOC_GET_STRIPE _IOWR('f', 200, long)
488 #define ECHO_IOC_SET_STRIPE _IOWR('f', 201, long)
489 #define ECHO_IOC_ENQUEUE _IOWR('f', 202, long)
490 #define ECHO_IOC_CANCEL _IOWR('f', 203, long)
493 #define CHECKSUM_BULK 0
496 static inline void ost_checksum(__u64 *cksum, void *addr, int len)
498 unsigned char *ptr = (unsigned char *)addr;
501 /* very stupid, but means I don't have to think about byte order */
505 *cksum = (*cksum << 2) + sum;
508 #define ost_checksum(cksum, addr, len) do {} while (0)
512 * l_wait_event is a flexible sleeping function, permitting simple caller
513 * configuration of interrupt and timeout sensitivity along with actions to
514 * be performed in the event of either exception.
516 * Common usage looks like this:
518 * struct l_wait_info lwi = LWI_TIMEOUT_INTR(timeout, timeout_handler,
519 * intr_handler, callback_data);
520 * rc = l_wait_event(waitq, condition, &lwi);
522 * (LWI_TIMEOUT and LWI_INTR macros are available for timeout- and
523 * interrupt-only variants, respectively.)
525 * If a timeout is specified, the timeout_handler will be invoked in the event
526 * that the timeout expires before the process is awakened. (Note that any
527 * waking of the process will restart the timeout, even if the condition is
528 * not satisfied and the process immediately returns to sleep. This might be
529 * considered a bug.) If the timeout_handler returns non-zero, l_wait_event
530 * will return -ETIMEDOUT and the caller will continue. If the handler returns
531 * zero instead, the process will go back to sleep until it is awakened by the
532 * waitq or some similar mechanism, or an interrupt occurs (if the caller has
533 * asked for interrupts to be detected). The timeout will only fire once, so
534 * callers should take care that a timeout_handler which returns zero will take
535 * future steps to awaken the process. N.B. that these steps must include
536 * making the provided condition become true.
538 * If the interrupt flag (lwi_signals) is non-zero, then the process will be
539 * interruptible, and will be awakened by any "killable" signal (SIGTERM,
540 * SIGKILL or SIGINT). If a timeout is also specified, then the process will
541 * only become interruptible _after_ the timeout has expired, though it can be
542 * awakened by a signal that was delivered before the timeout and is still
543 * pending when the timeout expires. If a timeout is not specified, the process
544 * will be interruptible at all times during l_wait_event.
549 int (*lwi_on_timeout)(void *);
551 int (*lwi_on_signal)(void *); /* XXX return is ignored for now */
555 #define LWI_TIMEOUT(time, cb, data) \
556 ((struct l_wait_info) { \
558 lwi_on_timeout: cb, \
562 #define LWI_INTR(cb, data) \
563 ((struct l_wait_info) { \
569 #define LWI_TIMEOUT_INTR(time, time_cb, sig_cb, data) \
570 ((struct l_wait_info) { \
572 lwi_on_timeout: time_cb, \
574 lwi_on_signal: sig_cb, \
579 #define l_sigismember sigismember
581 #define l_sigismember(a,b) (*(a) & b)
584 /* XXX this should be one mask-check */
585 #define l_killable_pending(task) \
586 (l_sigismember(&(task->pending.signal), SIGKILL) || \
587 l_sigismember(&(task->pending.signal), SIGINT) || \
588 l_sigismember(&(task->pending.signal), SIGTERM))
590 #define __l_wait_event(wq, condition, info, ret) \
592 wait_queue_t __wait; \
594 int __timed_out = 0; \
595 init_waitqueue_entry(&__wait, current); \
597 add_wait_queue(&wq, &__wait); \
598 if (info->lwi_signals && !info->lwi_timeout) \
599 __state = TASK_INTERRUPTIBLE; \
601 __state = TASK_UNINTERRUPTIBLE; \
603 set_current_state(__state); \
606 if (__state == TASK_INTERRUPTIBLE && l_killable_pending(current)) {\
607 if (info->lwi_on_signal) \
608 info->lwi_on_signal(info->lwi_cb_data); \
612 if (info->lwi_timeout && !__timed_out) { \
613 if (schedule_timeout(info->lwi_timeout) == 0) { \
615 if (!info->lwi_on_timeout || \
616 info->lwi_on_timeout(info->lwi_cb_data)) { \
620 /* We'll take signals after a timeout. */ \
621 if (info->lwi_signals) { \
622 __state = TASK_INTERRUPTIBLE; \
623 /* Check for a pending interrupt. */ \
624 if (info->lwi_signals && l_killable_pending(current)) {\
625 if (info->lwi_on_signal) \
626 info->lwi_on_signal(info->lwi_cb_data); \
636 current->state = TASK_RUNNING; \
637 remove_wait_queue(&wq, &__wait); \
640 #define l_wait_event(wq, condition, info) \
643 struct l_wait_info *__info = (info); \
645 __l_wait_event(wq, condition, __info, __ret); \
649 #endif /* _LUSTRE_LIB_H */