2 * Copyright (C) 2006-2008 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/smp_lock.h>
19 #include <linux/dlm.h>
20 #include <linux/dlm_device.h>
22 #include "dlm_internal.h"
23 #include "lockspace.h"
25 #include "lvb_table.h"
28 static const char name_prefix[] = "dlm";
29 static const struct file_operations device_fops;
33 struct dlm_lock_params32 {
47 char lvb[DLM_USER_LVB_LEN];
51 struct dlm_write_request32 {
58 struct dlm_lock_params32 lock;
59 struct dlm_lspace_params lspace;
60 struct dlm_purge_params purge;
71 struct dlm_lock_result32 {
77 struct dlm_lksb32 lksb;
80 /* Offsets may be zero if no data is present */
84 static void compat_input(struct dlm_write_request *kb,
85 struct dlm_write_request32 *kb32,
88 kb->version[0] = kb32->version[0];
89 kb->version[1] = kb32->version[1];
90 kb->version[2] = kb32->version[2];
93 kb->is64bit = kb32->is64bit;
94 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
95 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
96 kb->i.lspace.flags = kb32->i.lspace.flags;
97 kb->i.lspace.minor = kb32->i.lspace.minor;
98 memcpy(kb->i.lspace.name, kb32->i.lspace.name, count -
99 offsetof(struct dlm_write_request32, i.lspace.name));
100 } else if (kb->cmd == DLM_USER_PURGE) {
101 kb->i.purge.nodeid = kb32->i.purge.nodeid;
102 kb->i.purge.pid = kb32->i.purge.pid;
104 kb->i.lock.mode = kb32->i.lock.mode;
105 kb->i.lock.namelen = kb32->i.lock.namelen;
106 kb->i.lock.flags = kb32->i.lock.flags;
107 kb->i.lock.lkid = kb32->i.lock.lkid;
108 kb->i.lock.parent = kb32->i.lock.parent;
109 kb->i.lock.xid = kb32->i.lock.xid;
110 kb->i.lock.timeout = kb32->i.lock.timeout;
111 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
112 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
113 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
114 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
115 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
116 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
117 memcpy(kb->i.lock.name, kb32->i.lock.name, count -
118 offsetof(struct dlm_write_request32, i.lock.name));
122 static void compat_output(struct dlm_lock_result *res,
123 struct dlm_lock_result32 *res32)
125 res32->version[0] = res->version[0];
126 res32->version[1] = res->version[1];
127 res32->version[2] = res->version[2];
129 res32->user_astaddr = (__u32)(long)res->user_astaddr;
130 res32->user_astparam = (__u32)(long)res->user_astparam;
131 res32->user_lksb = (__u32)(long)res->user_lksb;
132 res32->bast_mode = res->bast_mode;
134 res32->lvb_offset = res->lvb_offset;
135 res32->length = res->length;
137 res32->lksb.sb_status = res->lksb.sb_status;
138 res32->lksb.sb_flags = res->lksb.sb_flags;
139 res32->lksb.sb_lkid = res->lksb.sb_lkid;
140 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
144 /* Figure out if this lock is at the end of its life and no longer
145 available for the application to use. The lkb still exists until
146 the final ast is read. A lock becomes EOL in three situations:
147 1. a noqueue request fails with EAGAIN
148 2. an unlock completes with EUNLOCK
149 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
150 An EOL lock needs to be removed from the process's list of locks.
151 And we can't allow any new operation on an EOL lock. This is
152 not related to the lifetime of the lkb struct which is managed
153 entirely by refcount. */
155 static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
163 if (lkb->lkb_grmode == DLM_LOCK_IV)
167 if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
174 /* we could possibly check if the cancel of an orphan has resulted in the lkb
175 being removed and then remove that lkb from the orphans list and free it */
177 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
180 struct dlm_user_args *ua;
181 struct dlm_user_proc *proc;
182 int eol = 0, ast_type;
184 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
187 ls = lkb->lkb_resource->res_ls;
188 mutex_lock(&ls->ls_clear_proc_locks);
190 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
191 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
192 lkb->ua so we can't try to use it. This second check is necessary
193 for cases where a completion ast is received for an operation that
194 began before clear_proc_locks did its cancel/unlock. */
196 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
199 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
203 if (type == AST_BAST && ua->bastaddr == NULL)
206 spin_lock(&proc->asts_spin);
208 ast_type = lkb->lkb_ast_type;
209 lkb->lkb_ast_type |= type;
212 kref_get(&lkb->lkb_ref);
213 list_add_tail(&lkb->lkb_astqueue, &proc->asts);
214 wake_up_interruptible(&proc->wait);
216 if (type == AST_COMP && (ast_type & AST_COMP))
217 log_debug(ls, "ast overlap %x status %x %x",
218 lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
220 eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
222 lkb->lkb_ast_type &= ~AST_BAST;
223 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
226 /* We want to copy the lvb to userspace when the completion
227 ast is read if the status is 0, the lock has an lvb and
228 lvb_ops says we should. We could probably have set_lvb_lock()
229 set update_user_lvb instead and not need old_mode */
231 if ((lkb->lkb_ast_type & AST_COMP) &&
232 (lkb->lkb_lksb->sb_status == 0) &&
233 lkb->lkb_lksb->sb_lvbptr &&
234 dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
235 ua->update_user_lvb = 1;
237 ua->update_user_lvb = 0;
239 spin_unlock(&proc->asts_spin);
242 spin_lock(&proc->locks_spin);
243 if (!list_empty(&lkb->lkb_ownqueue)) {
244 list_del_init(&lkb->lkb_ownqueue);
247 spin_unlock(&proc->locks_spin);
250 mutex_unlock(&ls->ls_clear_proc_locks);
253 static int device_user_lock(struct dlm_user_proc *proc,
254 struct dlm_lock_params *params)
257 struct dlm_user_args *ua;
260 ls = dlm_find_lockspace_local(proc->lockspace);
264 if (!params->castaddr || !params->lksb) {
269 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
273 ua->user_lksb = params->lksb;
274 ua->castparam = params->castparam;
275 ua->castaddr = params->castaddr;
276 ua->bastparam = params->bastparam;
277 ua->bastaddr = params->bastaddr;
278 ua->xid = params->xid;
280 if (params->flags & DLM_LKF_CONVERT)
281 error = dlm_user_convert(ls, ua,
282 params->mode, params->flags,
283 params->lkid, params->lvb,
284 (unsigned long) params->timeout);
286 error = dlm_user_request(ls, ua,
287 params->mode, params->flags,
288 params->name, params->namelen,
289 (unsigned long) params->timeout);
291 error = ua->lksb.sb_lkid;
294 dlm_put_lockspace(ls);
298 static int device_user_unlock(struct dlm_user_proc *proc,
299 struct dlm_lock_params *params)
302 struct dlm_user_args *ua;
305 ls = dlm_find_lockspace_local(proc->lockspace);
309 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
313 ua->user_lksb = params->lksb;
314 ua->castparam = params->castparam;
315 ua->castaddr = params->castaddr;
317 if (params->flags & DLM_LKF_CANCEL)
318 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
320 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
323 dlm_put_lockspace(ls);
327 static int device_user_deadlock(struct dlm_user_proc *proc,
328 struct dlm_lock_params *params)
333 ls = dlm_find_lockspace_local(proc->lockspace);
337 error = dlm_user_deadlock(ls, params->flags, params->lkid);
339 dlm_put_lockspace(ls);
343 static int dlm_device_register(struct dlm_ls *ls, char *name)
347 /* The device is already registered. This happens when the
348 lockspace is created multiple times from userspace. */
349 if (ls->ls_device.name)
353 len = strlen(name) + strlen(name_prefix) + 2;
354 ls->ls_device.name = kzalloc(len, GFP_KERNEL);
355 if (!ls->ls_device.name)
358 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
360 ls->ls_device.fops = &device_fops;
361 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
363 error = misc_register(&ls->ls_device);
365 kfree(ls->ls_device.name);
371 int dlm_device_deregister(struct dlm_ls *ls)
375 /* The device is not registered. This happens when the lockspace
376 was never used from userspace, or when device_create_lockspace()
377 calls dlm_release_lockspace() after the register fails. */
378 if (!ls->ls_device.name)
381 error = misc_deregister(&ls->ls_device);
383 kfree(ls->ls_device.name);
387 static int device_user_purge(struct dlm_user_proc *proc,
388 struct dlm_purge_params *params)
393 ls = dlm_find_lockspace_local(proc->lockspace);
397 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
399 dlm_put_lockspace(ls);
403 static int device_create_lockspace(struct dlm_lspace_params *params)
405 dlm_lockspace_t *lockspace;
409 if (!capable(CAP_SYS_ADMIN))
412 error = dlm_new_lockspace(params->name, strlen(params->name),
413 &lockspace, params->flags, DLM_USER_LVB_LEN);
417 ls = dlm_find_lockspace_local(lockspace);
421 error = dlm_device_register(ls, params->name);
422 dlm_put_lockspace(ls);
425 dlm_release_lockspace(lockspace, 0);
427 error = ls->ls_device.minor;
432 static int device_remove_lockspace(struct dlm_lspace_params *params)
434 dlm_lockspace_t *lockspace;
436 int error, force = 0;
438 if (!capable(CAP_SYS_ADMIN))
441 ls = dlm_find_lockspace_device(params->minor);
445 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
448 lockspace = ls->ls_local_handle;
449 dlm_put_lockspace(ls);
451 /* The final dlm_release_lockspace waits for references to go to
452 zero, so all processes will need to close their device for the
453 ls before the release will proceed. release also calls the
454 device_deregister above. Converting a positive return value
455 from release to zero means that userspace won't know when its
456 release was the final one, but it shouldn't need to know. */
458 error = dlm_release_lockspace(lockspace, force);
464 /* Check the user's version matches ours */
465 static int check_version(struct dlm_write_request *req)
467 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
468 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
469 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
471 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
472 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
474 task_pid_nr(current),
478 DLM_DEVICE_VERSION_MAJOR,
479 DLM_DEVICE_VERSION_MINOR,
480 DLM_DEVICE_VERSION_PATCH);
490 * dlm_user_request -> request_lock
491 * dlm_user_convert -> convert_lock
494 * dlm_user_unlock -> unlock_lock
495 * dlm_user_cancel -> cancel_lock
497 * device_create_lockspace
500 * device_remove_lockspace
501 * dlm_release_lockspace
504 /* a write to a lockspace device is a lock or unlock request, a write
505 to the control device is to create/remove a lockspace */
507 static ssize_t device_write(struct file *file, const char __user *buf,
508 size_t count, loff_t *ppos)
510 struct dlm_user_proc *proc = file->private_data;
511 struct dlm_write_request *kbuf;
512 sigset_t tmpsig, allsigs;
516 if (count < sizeof(struct dlm_write_request32))
518 if (count < sizeof(struct dlm_write_request))
522 kbuf = kzalloc(count + 1, GFP_KERNEL);
526 if (copy_from_user(kbuf, buf, count)) {
531 if (check_version(kbuf)) {
537 if (!kbuf->is64bit) {
538 struct dlm_write_request32 *k32buf;
539 k32buf = (struct dlm_write_request32 *)kbuf;
540 kbuf = kmalloc(count + 1 + (sizeof(struct dlm_write_request) -
541 sizeof(struct dlm_write_request32)), GFP_KERNEL);
548 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
549 compat_input(kbuf, k32buf, count + 1);
554 /* do we really need this? can a write happen after a close? */
555 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
556 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
561 sigfillset(&allsigs);
562 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
570 log_print("no locking on control device");
573 error = device_user_lock(proc, &kbuf->i.lock);
576 case DLM_USER_UNLOCK:
578 log_print("no locking on control device");
581 error = device_user_unlock(proc, &kbuf->i.lock);
584 case DLM_USER_DEADLOCK:
586 log_print("no locking on control device");
589 error = device_user_deadlock(proc, &kbuf->i.lock);
592 case DLM_USER_CREATE_LOCKSPACE:
594 log_print("create/remove only on control device");
597 error = device_create_lockspace(&kbuf->i.lspace);
600 case DLM_USER_REMOVE_LOCKSPACE:
602 log_print("create/remove only on control device");
605 error = device_remove_lockspace(&kbuf->i.lspace);
610 log_print("no locking on control device");
613 error = device_user_purge(proc, &kbuf->i.purge);
617 log_print("Unknown command passed to DLM device : %d\n",
622 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
629 /* Every process that opens the lockspace device has its own "proc" structure
630 hanging off the open file that's used to keep track of locks owned by the
631 process and asts that need to be delivered to the process. */
633 static int device_open(struct inode *inode, struct file *file)
635 struct dlm_user_proc *proc;
639 ls = dlm_find_lockspace_device(iminor(inode));
645 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
647 dlm_put_lockspace(ls);
652 proc->lockspace = ls->ls_local_handle;
653 INIT_LIST_HEAD(&proc->asts);
654 INIT_LIST_HEAD(&proc->locks);
655 INIT_LIST_HEAD(&proc->unlocking);
656 spin_lock_init(&proc->asts_spin);
657 spin_lock_init(&proc->locks_spin);
658 init_waitqueue_head(&proc->wait);
659 file->private_data = proc;
665 static int device_close(struct inode *inode, struct file *file)
667 struct dlm_user_proc *proc = file->private_data;
669 sigset_t tmpsig, allsigs;
671 ls = dlm_find_lockspace_local(proc->lockspace);
675 sigfillset(&allsigs);
676 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
678 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
680 dlm_clear_proc_locks(ls, proc);
682 /* at this point no more lkb's should exist for this lockspace,
683 so there's no chance of dlm_user_add_ast() being called and
684 looking for lkb->ua->proc */
687 file->private_data = NULL;
689 dlm_put_lockspace(ls);
690 dlm_put_lockspace(ls); /* for the find in device_open() */
692 /* FIXME: AUTOFREE: if this ls is no longer used do
693 device_remove_lockspace() */
695 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
701 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
702 int bmode, char __user *buf, size_t count)
705 struct dlm_lock_result32 result32;
707 struct dlm_lock_result result;
713 memset(&result, 0, sizeof(struct dlm_lock_result));
714 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
715 result.version[1] = DLM_DEVICE_VERSION_MINOR;
716 result.version[2] = DLM_DEVICE_VERSION_PATCH;
717 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
718 result.user_lksb = ua->user_lksb;
720 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
721 in a conversion unless the conversion is successful. See code
722 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
723 notes that a new blocking AST address and parameter are set even if
724 the conversion fails, so maybe we should just do that. */
726 if (type == AST_BAST) {
727 result.user_astaddr = ua->bastaddr;
728 result.user_astparam = ua->bastparam;
729 result.bast_mode = bmode;
731 result.user_astaddr = ua->castaddr;
732 result.user_astparam = ua->castparam;
737 len = sizeof(struct dlm_lock_result32);
740 len = sizeof(struct dlm_lock_result);
743 /* copy lvb to userspace if there is one, it's been updated, and
744 the user buffer has space for it */
746 if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
747 count >= len + DLM_USER_LVB_LEN) {
748 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
754 result.lvb_offset = len;
755 len += DLM_USER_LVB_LEN;
762 compat_output(&result, &result32);
763 resultptr = &result32;
767 if (copy_to_user(buf, resultptr, struct_len))
775 static int copy_version_to_user(char __user *buf, size_t count)
777 struct dlm_device_version ver;
779 memset(&ver, 0, sizeof(struct dlm_device_version));
780 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
781 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
782 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
784 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
786 return sizeof(struct dlm_device_version);
789 /* a read returns a single ast described in a struct dlm_lock_result */
791 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
794 struct dlm_user_proc *proc = file->private_data;
796 DECLARE_WAITQUEUE(wait, current);
797 int error, type=0, bmode=0, removed = 0;
799 if (count == sizeof(struct dlm_device_version)) {
800 error = copy_version_to_user(buf, count);
805 log_print("non-version read from control device %zu", count);
810 if (count < sizeof(struct dlm_lock_result32))
812 if (count < sizeof(struct dlm_lock_result))
816 /* do we really need this? can a read happen after a close? */
817 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
820 spin_lock(&proc->asts_spin);
821 if (list_empty(&proc->asts)) {
822 if (file->f_flags & O_NONBLOCK) {
823 spin_unlock(&proc->asts_spin);
827 add_wait_queue(&proc->wait, &wait);
830 set_current_state(TASK_INTERRUPTIBLE);
831 if (list_empty(&proc->asts) && !signal_pending(current)) {
832 spin_unlock(&proc->asts_spin);
834 spin_lock(&proc->asts_spin);
837 set_current_state(TASK_RUNNING);
838 remove_wait_queue(&proc->wait, &wait);
840 if (signal_pending(current)) {
841 spin_unlock(&proc->asts_spin);
846 /* there may be both completion and blocking asts to return for
847 the lkb, don't remove lkb from asts list unless no asts remain */
849 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
851 if (lkb->lkb_ast_type & AST_COMP) {
852 lkb->lkb_ast_type &= ~AST_COMP;
854 } else if (lkb->lkb_ast_type & AST_BAST) {
855 lkb->lkb_ast_type &= ~AST_BAST;
857 bmode = lkb->lkb_bastmode;
860 if (!lkb->lkb_ast_type) {
861 list_del(&lkb->lkb_astqueue);
864 spin_unlock(&proc->asts_spin);
866 error = copy_result_to_user(lkb->lkb_ua,
867 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
868 type, bmode, buf, count);
870 /* removes reference for the proc->asts lists added by
871 dlm_user_add_ast() and may result in the lkb being freed */
878 static unsigned int device_poll(struct file *file, poll_table *wait)
880 struct dlm_user_proc *proc = file->private_data;
882 poll_wait(file, &proc->wait, wait);
884 spin_lock(&proc->asts_spin);
885 if (!list_empty(&proc->asts)) {
886 spin_unlock(&proc->asts_spin);
887 return POLLIN | POLLRDNORM;
889 spin_unlock(&proc->asts_spin);
893 static int ctl_device_open(struct inode *inode, struct file *file)
896 file->private_data = NULL;
900 static int ctl_device_close(struct inode *inode, struct file *file)
905 static const struct file_operations device_fops = {
907 .release = device_close,
909 .write = device_write,
911 .owner = THIS_MODULE,
914 static const struct file_operations ctl_device_fops = {
915 .open = ctl_device_open,
916 .release = ctl_device_close,
918 .write = device_write,
919 .owner = THIS_MODULE,
922 static struct miscdevice ctl_device = {
923 .name = "dlm-control",
924 .fops = &ctl_device_fops,
925 .minor = MISC_DYNAMIC_MINOR,
928 int __init dlm_user_init(void)
932 error = misc_register(&ctl_device);
934 log_print("misc_register failed for control device");
939 void dlm_user_exit(void)
941 misc_deregister(&ctl_device);