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/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 create_misc_device(struct dlm_ls *ls, char *name)
348 len = strlen(name) + strlen(name_prefix) + 2;
349 ls->ls_device.name = kzalloc(len, GFP_KERNEL);
350 if (!ls->ls_device.name)
353 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
355 ls->ls_device.fops = &device_fops;
356 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
358 error = misc_register(&ls->ls_device);
360 kfree(ls->ls_device.name);
366 static int device_user_purge(struct dlm_user_proc *proc,
367 struct dlm_purge_params *params)
372 ls = dlm_find_lockspace_local(proc->lockspace);
376 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
378 dlm_put_lockspace(ls);
382 static int device_create_lockspace(struct dlm_lspace_params *params)
384 dlm_lockspace_t *lockspace;
388 if (!capable(CAP_SYS_ADMIN))
391 error = dlm_new_lockspace(params->name, strlen(params->name),
392 &lockspace, params->flags, DLM_USER_LVB_LEN);
396 ls = dlm_find_lockspace_local(lockspace);
400 error = create_misc_device(ls, params->name);
401 dlm_put_lockspace(ls);
404 dlm_release_lockspace(lockspace, 0);
406 error = ls->ls_device.minor;
411 static int device_remove_lockspace(struct dlm_lspace_params *params)
413 dlm_lockspace_t *lockspace;
415 int error, force = 0;
417 if (!capable(CAP_SYS_ADMIN))
420 ls = dlm_find_lockspace_device(params->minor);
424 /* Deregister the misc device first, so we don't have
425 * a device that's not attached to a lockspace. If
426 * dlm_release_lockspace fails then we can recreate it
428 error = misc_deregister(&ls->ls_device);
430 dlm_put_lockspace(ls);
433 kfree(ls->ls_device.name);
435 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
438 lockspace = ls->ls_local_handle;
440 /* dlm_release_lockspace waits for references to go to zero,
441 so all processes will need to close their device for the ls
442 before the release will procede */
444 dlm_put_lockspace(ls);
445 error = dlm_release_lockspace(lockspace, force);
447 create_misc_device(ls, ls->ls_name);
452 /* Check the user's version matches ours */
453 static int check_version(struct dlm_write_request *req)
455 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
456 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
457 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
459 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
460 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
462 task_pid_nr(current),
466 DLM_DEVICE_VERSION_MAJOR,
467 DLM_DEVICE_VERSION_MINOR,
468 DLM_DEVICE_VERSION_PATCH);
478 * dlm_user_request -> request_lock
479 * dlm_user_convert -> convert_lock
482 * dlm_user_unlock -> unlock_lock
483 * dlm_user_cancel -> cancel_lock
485 * device_create_lockspace
488 * device_remove_lockspace
489 * dlm_release_lockspace
492 /* a write to a lockspace device is a lock or unlock request, a write
493 to the control device is to create/remove a lockspace */
495 static ssize_t device_write(struct file *file, const char __user *buf,
496 size_t count, loff_t *ppos)
498 struct dlm_user_proc *proc = file->private_data;
499 struct dlm_write_request *kbuf;
500 sigset_t tmpsig, allsigs;
504 if (count < sizeof(struct dlm_write_request32))
506 if (count < sizeof(struct dlm_write_request))
510 kbuf = kzalloc(count + 1, GFP_KERNEL);
514 if (copy_from_user(kbuf, buf, count)) {
519 if (check_version(kbuf)) {
525 if (!kbuf->is64bit) {
526 struct dlm_write_request32 *k32buf;
527 k32buf = (struct dlm_write_request32 *)kbuf;
528 kbuf = kmalloc(count + 1 + (sizeof(struct dlm_write_request) -
529 sizeof(struct dlm_write_request32)), GFP_KERNEL);
536 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
537 compat_input(kbuf, k32buf, count + 1);
542 /* do we really need this? can a write happen after a close? */
543 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
544 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
549 sigfillset(&allsigs);
550 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
558 log_print("no locking on control device");
561 error = device_user_lock(proc, &kbuf->i.lock);
564 case DLM_USER_UNLOCK:
566 log_print("no locking on control device");
569 error = device_user_unlock(proc, &kbuf->i.lock);
572 case DLM_USER_DEADLOCK:
574 log_print("no locking on control device");
577 error = device_user_deadlock(proc, &kbuf->i.lock);
580 case DLM_USER_CREATE_LOCKSPACE:
582 log_print("create/remove only on control device");
585 error = device_create_lockspace(&kbuf->i.lspace);
588 case DLM_USER_REMOVE_LOCKSPACE:
590 log_print("create/remove only on control device");
593 error = device_remove_lockspace(&kbuf->i.lspace);
598 log_print("no locking on control device");
601 error = device_user_purge(proc, &kbuf->i.purge);
605 log_print("Unknown command passed to DLM device : %d\n",
610 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
617 /* Every process that opens the lockspace device has its own "proc" structure
618 hanging off the open file that's used to keep track of locks owned by the
619 process and asts that need to be delivered to the process. */
621 static int device_open(struct inode *inode, struct file *file)
623 struct dlm_user_proc *proc;
627 ls = dlm_find_lockspace_device(iminor(inode));
633 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
635 dlm_put_lockspace(ls);
640 proc->lockspace = ls->ls_local_handle;
641 INIT_LIST_HEAD(&proc->asts);
642 INIT_LIST_HEAD(&proc->locks);
643 INIT_LIST_HEAD(&proc->unlocking);
644 spin_lock_init(&proc->asts_spin);
645 spin_lock_init(&proc->locks_spin);
646 init_waitqueue_head(&proc->wait);
647 file->private_data = proc;
653 static int device_close(struct inode *inode, struct file *file)
655 struct dlm_user_proc *proc = file->private_data;
657 sigset_t tmpsig, allsigs;
659 ls = dlm_find_lockspace_local(proc->lockspace);
663 sigfillset(&allsigs);
664 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
666 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
668 dlm_clear_proc_locks(ls, proc);
670 /* at this point no more lkb's should exist for this lockspace,
671 so there's no chance of dlm_user_add_ast() being called and
672 looking for lkb->ua->proc */
675 file->private_data = NULL;
677 dlm_put_lockspace(ls);
678 dlm_put_lockspace(ls); /* for the find in device_open() */
680 /* FIXME: AUTOFREE: if this ls is no longer used do
681 device_remove_lockspace() */
683 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
689 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
690 int bmode, char __user *buf, size_t count)
693 struct dlm_lock_result32 result32;
695 struct dlm_lock_result result;
701 memset(&result, 0, sizeof(struct dlm_lock_result));
702 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
703 result.version[1] = DLM_DEVICE_VERSION_MINOR;
704 result.version[2] = DLM_DEVICE_VERSION_PATCH;
705 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
706 result.user_lksb = ua->user_lksb;
708 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
709 in a conversion unless the conversion is successful. See code
710 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
711 notes that a new blocking AST address and parameter are set even if
712 the conversion fails, so maybe we should just do that. */
714 if (type == AST_BAST) {
715 result.user_astaddr = ua->bastaddr;
716 result.user_astparam = ua->bastparam;
717 result.bast_mode = bmode;
719 result.user_astaddr = ua->castaddr;
720 result.user_astparam = ua->castparam;
725 len = sizeof(struct dlm_lock_result32);
728 len = sizeof(struct dlm_lock_result);
731 /* copy lvb to userspace if there is one, it's been updated, and
732 the user buffer has space for it */
734 if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
735 count >= len + DLM_USER_LVB_LEN) {
736 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
742 result.lvb_offset = len;
743 len += DLM_USER_LVB_LEN;
750 compat_output(&result, &result32);
751 resultptr = &result32;
755 if (copy_to_user(buf, resultptr, struct_len))
763 static int copy_version_to_user(char __user *buf, size_t count)
765 struct dlm_device_version ver;
767 memset(&ver, 0, sizeof(struct dlm_device_version));
768 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
769 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
770 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
772 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
774 return sizeof(struct dlm_device_version);
777 /* a read returns a single ast described in a struct dlm_lock_result */
779 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
782 struct dlm_user_proc *proc = file->private_data;
784 DECLARE_WAITQUEUE(wait, current);
785 int error, type=0, bmode=0, removed = 0;
787 if (count == sizeof(struct dlm_device_version)) {
788 error = copy_version_to_user(buf, count);
793 log_print("non-version read from control device %zu", count);
798 if (count < sizeof(struct dlm_lock_result32))
800 if (count < sizeof(struct dlm_lock_result))
804 /* do we really need this? can a read happen after a close? */
805 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
808 spin_lock(&proc->asts_spin);
809 if (list_empty(&proc->asts)) {
810 if (file->f_flags & O_NONBLOCK) {
811 spin_unlock(&proc->asts_spin);
815 add_wait_queue(&proc->wait, &wait);
818 set_current_state(TASK_INTERRUPTIBLE);
819 if (list_empty(&proc->asts) && !signal_pending(current)) {
820 spin_unlock(&proc->asts_spin);
822 spin_lock(&proc->asts_spin);
825 set_current_state(TASK_RUNNING);
826 remove_wait_queue(&proc->wait, &wait);
828 if (signal_pending(current)) {
829 spin_unlock(&proc->asts_spin);
834 /* there may be both completion and blocking asts to return for
835 the lkb, don't remove lkb from asts list unless no asts remain */
837 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
839 if (lkb->lkb_ast_type & AST_COMP) {
840 lkb->lkb_ast_type &= ~AST_COMP;
842 } else if (lkb->lkb_ast_type & AST_BAST) {
843 lkb->lkb_ast_type &= ~AST_BAST;
845 bmode = lkb->lkb_bastmode;
848 if (!lkb->lkb_ast_type) {
849 list_del(&lkb->lkb_astqueue);
852 spin_unlock(&proc->asts_spin);
854 error = copy_result_to_user(lkb->lkb_ua,
855 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
856 type, bmode, buf, count);
858 /* removes reference for the proc->asts lists added by
859 dlm_user_add_ast() and may result in the lkb being freed */
866 static unsigned int device_poll(struct file *file, poll_table *wait)
868 struct dlm_user_proc *proc = file->private_data;
870 poll_wait(file, &proc->wait, wait);
872 spin_lock(&proc->asts_spin);
873 if (!list_empty(&proc->asts)) {
874 spin_unlock(&proc->asts_spin);
875 return POLLIN | POLLRDNORM;
877 spin_unlock(&proc->asts_spin);
881 static int ctl_device_open(struct inode *inode, struct file *file)
884 file->private_data = NULL;
888 static int ctl_device_close(struct inode *inode, struct file *file)
893 static const struct file_operations device_fops = {
895 .release = device_close,
897 .write = device_write,
899 .owner = THIS_MODULE,
902 static const struct file_operations ctl_device_fops = {
903 .open = ctl_device_open,
904 .release = ctl_device_close,
906 .write = device_write,
907 .owner = THIS_MODULE,
910 static struct miscdevice ctl_device = {
911 .name = "dlm-control",
912 .fops = &ctl_device_fops,
913 .minor = MISC_DYNAMIC_MINOR,
916 int __init dlm_user_init(void)
920 error = misc_register(&ctl_device);
922 log_print("misc_register failed for control device");
927 void dlm_user_exit(void)
929 misc_deregister(&ctl_device);