2 * Copyright (C) 2006-2007 Red Hat, Inc. All rights reserved.
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License v.2.
9 #include <linux/miscdevice.h>
10 #include <linux/init.h>
11 #include <linux/wait.h>
12 #include <linux/module.h>
13 #include <linux/file.h>
15 #include <linux/poll.h>
16 #include <linux/signal.h>
17 #include <linux/spinlock.h>
18 #include <linux/dlm.h>
19 #include <linux/dlm_device.h>
21 #include "dlm_internal.h"
22 #include "lockspace.h"
24 #include "lvb_table.h"
27 static const char name_prefix[] = "dlm";
28 static const struct file_operations device_fops;
32 struct dlm_lock_params32 {
46 char lvb[DLM_USER_LVB_LEN];
50 struct dlm_write_request32 {
57 struct dlm_lock_params32 lock;
58 struct dlm_lspace_params lspace;
59 struct dlm_purge_params purge;
70 struct dlm_lock_result32 {
76 struct dlm_lksb32 lksb;
79 /* Offsets may be zero if no data is present */
83 static void compat_input(struct dlm_write_request *kb,
84 struct dlm_write_request32 *kb32,
87 kb->version[0] = kb32->version[0];
88 kb->version[1] = kb32->version[1];
89 kb->version[2] = kb32->version[2];
92 kb->is64bit = kb32->is64bit;
93 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
94 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
95 kb->i.lspace.flags = kb32->i.lspace.flags;
96 kb->i.lspace.minor = kb32->i.lspace.minor;
97 memcpy(kb->i.lspace.name, kb32->i.lspace.name, count -
98 offsetof(struct dlm_write_request32, i.lspace.name));
99 } else if (kb->cmd == DLM_USER_PURGE) {
100 kb->i.purge.nodeid = kb32->i.purge.nodeid;
101 kb->i.purge.pid = kb32->i.purge.pid;
103 kb->i.lock.mode = kb32->i.lock.mode;
104 kb->i.lock.namelen = kb32->i.lock.namelen;
105 kb->i.lock.flags = kb32->i.lock.flags;
106 kb->i.lock.lkid = kb32->i.lock.lkid;
107 kb->i.lock.parent = kb32->i.lock.parent;
108 kb->i.lock.xid = kb32->i.lock.xid;
109 kb->i.lock.timeout = kb32->i.lock.timeout;
110 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
111 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
112 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
113 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
114 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
115 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
116 memcpy(kb->i.lock.name, kb32->i.lock.name, count -
117 offsetof(struct dlm_write_request32, i.lock.name));
121 static void compat_output(struct dlm_lock_result *res,
122 struct dlm_lock_result32 *res32)
124 res32->version[0] = res->version[0];
125 res32->version[1] = res->version[1];
126 res32->version[2] = res->version[2];
128 res32->user_astaddr = (__u32)(long)res->user_astaddr;
129 res32->user_astparam = (__u32)(long)res->user_astparam;
130 res32->user_lksb = (__u32)(long)res->user_lksb;
131 res32->bast_mode = res->bast_mode;
133 res32->lvb_offset = res->lvb_offset;
134 res32->length = res->length;
136 res32->lksb.sb_status = res->lksb.sb_status;
137 res32->lksb.sb_flags = res->lksb.sb_flags;
138 res32->lksb.sb_lkid = res->lksb.sb_lkid;
139 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
143 /* Figure out if this lock is at the end of its life and no longer
144 available for the application to use. The lkb still exists until
145 the final ast is read. A lock becomes EOL in three situations:
146 1. a noqueue request fails with EAGAIN
147 2. an unlock completes with EUNLOCK
148 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
149 An EOL lock needs to be removed from the process's list of locks.
150 And we can't allow any new operation on an EOL lock. This is
151 not related to the lifetime of the lkb struct which is managed
152 entirely by refcount. */
154 static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
162 if (lkb->lkb_grmode == DLM_LOCK_IV)
166 if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
173 /* we could possibly check if the cancel of an orphan has resulted in the lkb
174 being removed and then remove that lkb from the orphans list and free it */
176 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
179 struct dlm_user_args *ua;
180 struct dlm_user_proc *proc;
181 int eol = 0, ast_type;
183 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
186 ls = lkb->lkb_resource->res_ls;
187 mutex_lock(&ls->ls_clear_proc_locks);
189 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
190 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
191 lkb->ua so we can't try to use it. This second check is necessary
192 for cases where a completion ast is received for an operation that
193 began before clear_proc_locks did its cancel/unlock. */
195 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
198 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
202 if (type == AST_BAST && ua->bastaddr == NULL)
205 spin_lock(&proc->asts_spin);
207 ast_type = lkb->lkb_ast_type;
208 lkb->lkb_ast_type |= type;
211 kref_get(&lkb->lkb_ref);
212 list_add_tail(&lkb->lkb_astqueue, &proc->asts);
213 wake_up_interruptible(&proc->wait);
215 if (type == AST_COMP && (ast_type & AST_COMP))
216 log_debug(ls, "ast overlap %x status %x %x",
217 lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
219 eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
221 lkb->lkb_ast_type &= ~AST_BAST;
222 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
225 /* We want to copy the lvb to userspace when the completion
226 ast is read if the status is 0, the lock has an lvb and
227 lvb_ops says we should. We could probably have set_lvb_lock()
228 set update_user_lvb instead and not need old_mode */
230 if ((lkb->lkb_ast_type & AST_COMP) &&
231 (lkb->lkb_lksb->sb_status == 0) &&
232 lkb->lkb_lksb->sb_lvbptr &&
233 dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
234 ua->update_user_lvb = 1;
236 ua->update_user_lvb = 0;
238 spin_unlock(&proc->asts_spin);
241 spin_lock(&proc->locks_spin);
242 if (!list_empty(&lkb->lkb_ownqueue)) {
243 list_del_init(&lkb->lkb_ownqueue);
246 spin_unlock(&proc->locks_spin);
249 mutex_unlock(&ls->ls_clear_proc_locks);
252 static int device_user_lock(struct dlm_user_proc *proc,
253 struct dlm_lock_params *params)
256 struct dlm_user_args *ua;
259 ls = dlm_find_lockspace_local(proc->lockspace);
263 if (!params->castaddr || !params->lksb) {
268 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
272 ua->user_lksb = params->lksb;
273 ua->castparam = params->castparam;
274 ua->castaddr = params->castaddr;
275 ua->bastparam = params->bastparam;
276 ua->bastaddr = params->bastaddr;
277 ua->xid = params->xid;
279 if (params->flags & DLM_LKF_CONVERT)
280 error = dlm_user_convert(ls, ua,
281 params->mode, params->flags,
282 params->lkid, params->lvb,
283 (unsigned long) params->timeout);
285 error = dlm_user_request(ls, ua,
286 params->mode, params->flags,
287 params->name, params->namelen,
288 (unsigned long) params->timeout);
290 error = ua->lksb.sb_lkid;
293 dlm_put_lockspace(ls);
297 static int device_user_unlock(struct dlm_user_proc *proc,
298 struct dlm_lock_params *params)
301 struct dlm_user_args *ua;
304 ls = dlm_find_lockspace_local(proc->lockspace);
308 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
312 ua->user_lksb = params->lksb;
313 ua->castparam = params->castparam;
314 ua->castaddr = params->castaddr;
316 if (params->flags & DLM_LKF_CANCEL)
317 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
319 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
322 dlm_put_lockspace(ls);
326 static int device_user_deadlock(struct dlm_user_proc *proc,
327 struct dlm_lock_params *params)
332 ls = dlm_find_lockspace_local(proc->lockspace);
336 error = dlm_user_deadlock(ls, params->flags, params->lkid);
338 dlm_put_lockspace(ls);
342 static int create_misc_device(struct dlm_ls *ls, char *name)
347 len = strlen(name) + strlen(name_prefix) + 2;
348 ls->ls_device.name = kzalloc(len, GFP_KERNEL);
349 if (!ls->ls_device.name)
352 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
354 ls->ls_device.fops = &device_fops;
355 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
357 error = misc_register(&ls->ls_device);
359 kfree(ls->ls_device.name);
365 static int device_user_purge(struct dlm_user_proc *proc,
366 struct dlm_purge_params *params)
371 ls = dlm_find_lockspace_local(proc->lockspace);
375 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
377 dlm_put_lockspace(ls);
381 static int device_create_lockspace(struct dlm_lspace_params *params)
383 dlm_lockspace_t *lockspace;
387 if (!capable(CAP_SYS_ADMIN))
390 error = dlm_new_lockspace(params->name, strlen(params->name),
391 &lockspace, params->flags, DLM_USER_LVB_LEN);
395 ls = dlm_find_lockspace_local(lockspace);
399 error = create_misc_device(ls, params->name);
400 dlm_put_lockspace(ls);
403 dlm_release_lockspace(lockspace, 0);
405 error = ls->ls_device.minor;
410 static int device_remove_lockspace(struct dlm_lspace_params *params)
412 dlm_lockspace_t *lockspace;
414 int error, force = 0;
416 if (!capable(CAP_SYS_ADMIN))
419 ls = dlm_find_lockspace_device(params->minor);
423 /* Deregister the misc device first, so we don't have
424 * a device that's not attached to a lockspace. If
425 * dlm_release_lockspace fails then we can recreate it
427 error = misc_deregister(&ls->ls_device);
429 dlm_put_lockspace(ls);
432 kfree(ls->ls_device.name);
434 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
437 lockspace = ls->ls_local_handle;
439 /* dlm_release_lockspace waits for references to go to zero,
440 so all processes will need to close their device for the ls
441 before the release will procede */
443 dlm_put_lockspace(ls);
444 error = dlm_release_lockspace(lockspace, force);
446 create_misc_device(ls, ls->ls_name);
451 /* Check the user's version matches ours */
452 static int check_version(struct dlm_write_request *req)
454 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
455 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
456 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
458 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
459 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
461 task_pid_nr(current),
465 DLM_DEVICE_VERSION_MAJOR,
466 DLM_DEVICE_VERSION_MINOR,
467 DLM_DEVICE_VERSION_PATCH);
477 * dlm_user_request -> request_lock
478 * dlm_user_convert -> convert_lock
481 * dlm_user_unlock -> unlock_lock
482 * dlm_user_cancel -> cancel_lock
484 * device_create_lockspace
487 * device_remove_lockspace
488 * dlm_release_lockspace
491 /* a write to a lockspace device is a lock or unlock request, a write
492 to the control device is to create/remove a lockspace */
494 static ssize_t device_write(struct file *file, const char __user *buf,
495 size_t count, loff_t *ppos)
497 struct dlm_user_proc *proc = file->private_data;
498 struct dlm_write_request *kbuf;
499 sigset_t tmpsig, allsigs;
503 if (count < sizeof(struct dlm_write_request32))
505 if (count < sizeof(struct dlm_write_request))
509 kbuf = kzalloc(count + 1, GFP_KERNEL);
513 if (copy_from_user(kbuf, buf, count)) {
518 if (check_version(kbuf)) {
524 if (!kbuf->is64bit) {
525 struct dlm_write_request32 *k32buf;
526 k32buf = (struct dlm_write_request32 *)kbuf;
527 kbuf = kmalloc(count + 1 + (sizeof(struct dlm_write_request) -
528 sizeof(struct dlm_write_request32)), GFP_KERNEL);
533 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
534 compat_input(kbuf, k32buf, count + 1);
539 /* do we really need this? can a write happen after a close? */
540 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
541 test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
544 sigfillset(&allsigs);
545 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
553 log_print("no locking on control device");
556 error = device_user_lock(proc, &kbuf->i.lock);
559 case DLM_USER_UNLOCK:
561 log_print("no locking on control device");
564 error = device_user_unlock(proc, &kbuf->i.lock);
567 case DLM_USER_DEADLOCK:
569 log_print("no locking on control device");
572 error = device_user_deadlock(proc, &kbuf->i.lock);
575 case DLM_USER_CREATE_LOCKSPACE:
577 log_print("create/remove only on control device");
580 error = device_create_lockspace(&kbuf->i.lspace);
583 case DLM_USER_REMOVE_LOCKSPACE:
585 log_print("create/remove only on control device");
588 error = device_remove_lockspace(&kbuf->i.lspace);
593 log_print("no locking on control device");
596 error = device_user_purge(proc, &kbuf->i.purge);
600 log_print("Unknown command passed to DLM device : %d\n",
605 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
612 /* Every process that opens the lockspace device has its own "proc" structure
613 hanging off the open file that's used to keep track of locks owned by the
614 process and asts that need to be delivered to the process. */
616 static int device_open(struct inode *inode, struct file *file)
618 struct dlm_user_proc *proc;
621 ls = dlm_find_lockspace_device(iminor(inode));
625 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
627 dlm_put_lockspace(ls);
631 proc->lockspace = ls->ls_local_handle;
632 INIT_LIST_HEAD(&proc->asts);
633 INIT_LIST_HEAD(&proc->locks);
634 INIT_LIST_HEAD(&proc->unlocking);
635 spin_lock_init(&proc->asts_spin);
636 spin_lock_init(&proc->locks_spin);
637 init_waitqueue_head(&proc->wait);
638 file->private_data = proc;
643 static int device_close(struct inode *inode, struct file *file)
645 struct dlm_user_proc *proc = file->private_data;
647 sigset_t tmpsig, allsigs;
649 ls = dlm_find_lockspace_local(proc->lockspace);
653 sigfillset(&allsigs);
654 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
656 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
658 dlm_clear_proc_locks(ls, proc);
660 /* at this point no more lkb's should exist for this lockspace,
661 so there's no chance of dlm_user_add_ast() being called and
662 looking for lkb->ua->proc */
665 file->private_data = NULL;
667 dlm_put_lockspace(ls);
668 dlm_put_lockspace(ls); /* for the find in device_open() */
670 /* FIXME: AUTOFREE: if this ls is no longer used do
671 device_remove_lockspace() */
673 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
679 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
680 int bmode, char __user *buf, size_t count)
683 struct dlm_lock_result32 result32;
685 struct dlm_lock_result result;
691 memset(&result, 0, sizeof(struct dlm_lock_result));
692 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
693 result.version[1] = DLM_DEVICE_VERSION_MINOR;
694 result.version[2] = DLM_DEVICE_VERSION_PATCH;
695 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
696 result.user_lksb = ua->user_lksb;
698 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
699 in a conversion unless the conversion is successful. See code
700 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
701 notes that a new blocking AST address and parameter are set even if
702 the conversion fails, so maybe we should just do that. */
704 if (type == AST_BAST) {
705 result.user_astaddr = ua->bastaddr;
706 result.user_astparam = ua->bastparam;
707 result.bast_mode = bmode;
709 result.user_astaddr = ua->castaddr;
710 result.user_astparam = ua->castparam;
715 len = sizeof(struct dlm_lock_result32);
718 len = sizeof(struct dlm_lock_result);
721 /* copy lvb to userspace if there is one, it's been updated, and
722 the user buffer has space for it */
724 if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
725 count >= len + DLM_USER_LVB_LEN) {
726 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
732 result.lvb_offset = len;
733 len += DLM_USER_LVB_LEN;
740 compat_output(&result, &result32);
741 resultptr = &result32;
745 if (copy_to_user(buf, resultptr, struct_len))
753 static int copy_version_to_user(char __user *buf, size_t count)
755 struct dlm_device_version ver;
757 memset(&ver, 0, sizeof(struct dlm_device_version));
758 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
759 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
760 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
762 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
764 return sizeof(struct dlm_device_version);
767 /* a read returns a single ast described in a struct dlm_lock_result */
769 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
772 struct dlm_user_proc *proc = file->private_data;
774 DECLARE_WAITQUEUE(wait, current);
775 int error, type=0, bmode=0, removed = 0;
777 if (count == sizeof(struct dlm_device_version)) {
778 error = copy_version_to_user(buf, count);
783 log_print("non-version read from control device %zu", count);
788 if (count < sizeof(struct dlm_lock_result32))
790 if (count < sizeof(struct dlm_lock_result))
794 /* do we really need this? can a read happen after a close? */
795 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
798 spin_lock(&proc->asts_spin);
799 if (list_empty(&proc->asts)) {
800 if (file->f_flags & O_NONBLOCK) {
801 spin_unlock(&proc->asts_spin);
805 add_wait_queue(&proc->wait, &wait);
808 set_current_state(TASK_INTERRUPTIBLE);
809 if (list_empty(&proc->asts) && !signal_pending(current)) {
810 spin_unlock(&proc->asts_spin);
812 spin_lock(&proc->asts_spin);
815 set_current_state(TASK_RUNNING);
816 remove_wait_queue(&proc->wait, &wait);
818 if (signal_pending(current)) {
819 spin_unlock(&proc->asts_spin);
824 /* there may be both completion and blocking asts to return for
825 the lkb, don't remove lkb from asts list unless no asts remain */
827 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
829 if (lkb->lkb_ast_type & AST_COMP) {
830 lkb->lkb_ast_type &= ~AST_COMP;
832 } else if (lkb->lkb_ast_type & AST_BAST) {
833 lkb->lkb_ast_type &= ~AST_BAST;
835 bmode = lkb->lkb_bastmode;
838 if (!lkb->lkb_ast_type) {
839 list_del(&lkb->lkb_astqueue);
842 spin_unlock(&proc->asts_spin);
844 error = copy_result_to_user(lkb->lkb_ua,
845 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
846 type, bmode, buf, count);
848 /* removes reference for the proc->asts lists added by
849 dlm_user_add_ast() and may result in the lkb being freed */
856 static unsigned int device_poll(struct file *file, poll_table *wait)
858 struct dlm_user_proc *proc = file->private_data;
860 poll_wait(file, &proc->wait, wait);
862 spin_lock(&proc->asts_spin);
863 if (!list_empty(&proc->asts)) {
864 spin_unlock(&proc->asts_spin);
865 return POLLIN | POLLRDNORM;
867 spin_unlock(&proc->asts_spin);
871 static int ctl_device_open(struct inode *inode, struct file *file)
873 file->private_data = NULL;
877 static int ctl_device_close(struct inode *inode, struct file *file)
882 static const struct file_operations device_fops = {
884 .release = device_close,
886 .write = device_write,
888 .owner = THIS_MODULE,
891 static const struct file_operations ctl_device_fops = {
892 .open = ctl_device_open,
893 .release = ctl_device_close,
895 .write = device_write,
896 .owner = THIS_MODULE,
899 static struct miscdevice ctl_device = {
900 .name = "dlm-control",
901 .fops = &ctl_device_fops,
902 .minor = MISC_DYNAMIC_MINOR,
905 int __init dlm_user_init(void)
909 error = misc_register(&ctl_device);
911 log_print("misc_register failed for control device");
916 void dlm_user_exit(void)
918 misc_deregister(&ctl_device);