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 struct miscdevice ctl_device;
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)
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 strcpy(kb->i.lspace.name, kb32->i.lspace.name);
98 } else if (kb->cmd == DLM_USER_PURGE) {
99 kb->i.purge.nodeid = kb32->i.purge.nodeid;
100 kb->i.purge.pid = kb32->i.purge.pid;
102 kb->i.lock.mode = kb32->i.lock.mode;
103 kb->i.lock.namelen = kb32->i.lock.namelen;
104 kb->i.lock.flags = kb32->i.lock.flags;
105 kb->i.lock.lkid = kb32->i.lock.lkid;
106 kb->i.lock.parent = kb32->i.lock.parent;
107 kb->i.lock.xid = kb32->i.lock.xid;
108 kb->i.lock.timeout = kb32->i.lock.timeout;
109 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
110 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
111 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
112 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
113 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
114 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
115 memcpy(kb->i.lock.name, kb32->i.lock.name, kb->i.lock.namelen);
119 static void compat_output(struct dlm_lock_result *res,
120 struct dlm_lock_result32 *res32)
122 res32->version[0] = res->version[0];
123 res32->version[1] = res->version[1];
124 res32->version[2] = res->version[2];
126 res32->user_astaddr = (__u32)(long)res->user_astaddr;
127 res32->user_astparam = (__u32)(long)res->user_astparam;
128 res32->user_lksb = (__u32)(long)res->user_lksb;
129 res32->bast_mode = res->bast_mode;
131 res32->lvb_offset = res->lvb_offset;
132 res32->length = res->length;
134 res32->lksb.sb_status = res->lksb.sb_status;
135 res32->lksb.sb_flags = res->lksb.sb_flags;
136 res32->lksb.sb_lkid = res->lksb.sb_lkid;
137 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
141 /* Figure out if this lock is at the end of its life and no longer
142 available for the application to use. The lkb still exists until
143 the final ast is read. A lock becomes EOL in three situations:
144 1. a noqueue request fails with EAGAIN
145 2. an unlock completes with EUNLOCK
146 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
147 An EOL lock needs to be removed from the process's list of locks.
148 And we can't allow any new operation on an EOL lock. This is
149 not related to the lifetime of the lkb struct which is managed
150 entirely by refcount. */
152 static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
160 if (lkb->lkb_grmode == DLM_LOCK_IV)
164 if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
171 /* we could possibly check if the cancel of an orphan has resulted in the lkb
172 being removed and then remove that lkb from the orphans list and free it */
174 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
177 struct dlm_user_args *ua;
178 struct dlm_user_proc *proc;
179 int eol = 0, ast_type;
181 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
184 ls = lkb->lkb_resource->res_ls;
185 mutex_lock(&ls->ls_clear_proc_locks);
187 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
188 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
189 lkb->ua so we can't try to use it. This second check is necessary
190 for cases where a completion ast is received for an operation that
191 began before clear_proc_locks did its cancel/unlock. */
193 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
196 DLM_ASSERT(lkb->lkb_astparam, dlm_print_lkb(lkb););
197 ua = (struct dlm_user_args *)lkb->lkb_astparam;
200 if (type == AST_BAST && ua->bastaddr == NULL)
203 spin_lock(&proc->asts_spin);
205 ast_type = lkb->lkb_ast_type;
206 lkb->lkb_ast_type |= type;
209 kref_get(&lkb->lkb_ref);
210 list_add_tail(&lkb->lkb_astqueue, &proc->asts);
211 wake_up_interruptible(&proc->wait);
213 if (type == AST_COMP && (ast_type & AST_COMP))
214 log_debug(ls, "ast overlap %x status %x %x",
215 lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
217 eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
219 lkb->lkb_ast_type &= ~AST_BAST;
220 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
223 /* We want to copy the lvb to userspace when the completion
224 ast is read if the status is 0, the lock has an lvb and
225 lvb_ops says we should. We could probably have set_lvb_lock()
226 set update_user_lvb instead and not need old_mode */
228 if ((lkb->lkb_ast_type & AST_COMP) &&
229 (lkb->lkb_lksb->sb_status == 0) &&
230 lkb->lkb_lksb->sb_lvbptr &&
231 dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
232 ua->update_user_lvb = 1;
234 ua->update_user_lvb = 0;
236 spin_unlock(&proc->asts_spin);
239 spin_lock(&ua->proc->locks_spin);
240 if (!list_empty(&lkb->lkb_ownqueue)) {
241 list_del_init(&lkb->lkb_ownqueue);
244 spin_unlock(&ua->proc->locks_spin);
247 mutex_unlock(&ls->ls_clear_proc_locks);
250 static int device_user_lock(struct dlm_user_proc *proc,
251 struct dlm_lock_params *params)
254 struct dlm_user_args *ua;
257 ls = dlm_find_lockspace_local(proc->lockspace);
261 if (!params->castaddr || !params->lksb) {
266 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
270 ua->user_lksb = params->lksb;
271 ua->castparam = params->castparam;
272 ua->castaddr = params->castaddr;
273 ua->bastparam = params->bastparam;
274 ua->bastaddr = params->bastaddr;
275 ua->xid = params->xid;
277 if (params->flags & DLM_LKF_CONVERT)
278 error = dlm_user_convert(ls, ua,
279 params->mode, params->flags,
280 params->lkid, params->lvb,
281 (unsigned long) params->timeout);
283 error = dlm_user_request(ls, ua,
284 params->mode, params->flags,
285 params->name, params->namelen,
286 (unsigned long) params->timeout);
288 error = ua->lksb.sb_lkid;
291 dlm_put_lockspace(ls);
295 static int device_user_unlock(struct dlm_user_proc *proc,
296 struct dlm_lock_params *params)
299 struct dlm_user_args *ua;
302 ls = dlm_find_lockspace_local(proc->lockspace);
306 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
310 ua->user_lksb = params->lksb;
311 ua->castparam = params->castparam;
312 ua->castaddr = params->castaddr;
314 if (params->flags & DLM_LKF_CANCEL)
315 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
317 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
320 dlm_put_lockspace(ls);
324 static int device_user_deadlock(struct dlm_user_proc *proc,
325 struct dlm_lock_params *params)
330 ls = dlm_find_lockspace_local(proc->lockspace);
334 error = dlm_user_deadlock(ls, params->flags, params->lkid);
336 dlm_put_lockspace(ls);
340 static int create_misc_device(struct dlm_ls *ls, char *name)
345 len = strlen(name) + strlen(name_prefix) + 2;
346 ls->ls_device.name = kzalloc(len, GFP_KERNEL);
347 if (!ls->ls_device.name)
350 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
352 ls->ls_device.fops = &device_fops;
353 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
355 error = misc_register(&ls->ls_device);
357 kfree(ls->ls_device.name);
363 static int device_user_purge(struct dlm_user_proc *proc,
364 struct dlm_purge_params *params)
369 ls = dlm_find_lockspace_local(proc->lockspace);
373 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
375 dlm_put_lockspace(ls);
379 static int device_create_lockspace(struct dlm_lspace_params *params)
381 dlm_lockspace_t *lockspace;
385 if (!capable(CAP_SYS_ADMIN))
388 error = dlm_new_lockspace(params->name, strlen(params->name),
389 &lockspace, params->flags, DLM_USER_LVB_LEN);
393 ls = dlm_find_lockspace_local(lockspace);
397 error = create_misc_device(ls, params->name);
398 dlm_put_lockspace(ls);
401 dlm_release_lockspace(lockspace, 0);
403 error = ls->ls_device.minor;
408 static int device_remove_lockspace(struct dlm_lspace_params *params)
410 dlm_lockspace_t *lockspace;
412 int error, force = 0;
414 if (!capable(CAP_SYS_ADMIN))
417 ls = dlm_find_lockspace_device(params->minor);
421 /* Deregister the misc device first, so we don't have
422 * a device that's not attached to a lockspace. If
423 * dlm_release_lockspace fails then we can recreate it
425 error = misc_deregister(&ls->ls_device);
427 dlm_put_lockspace(ls);
430 kfree(ls->ls_device.name);
432 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
435 lockspace = ls->ls_local_handle;
437 /* dlm_release_lockspace waits for references to go to zero,
438 so all processes will need to close their device for the ls
439 before the release will procede */
441 dlm_put_lockspace(ls);
442 error = dlm_release_lockspace(lockspace, force);
444 create_misc_device(ls, ls->ls_name);
449 /* Check the user's version matches ours */
450 static int check_version(struct dlm_write_request *req)
452 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
453 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
454 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
456 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
457 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
459 task_pid_nr(current),
463 DLM_DEVICE_VERSION_MAJOR,
464 DLM_DEVICE_VERSION_MINOR,
465 DLM_DEVICE_VERSION_PATCH);
475 * dlm_user_request -> request_lock
476 * dlm_user_convert -> convert_lock
479 * dlm_user_unlock -> unlock_lock
480 * dlm_user_cancel -> cancel_lock
482 * device_create_lockspace
485 * device_remove_lockspace
486 * dlm_release_lockspace
489 /* a write to a lockspace device is a lock or unlock request, a write
490 to the control device is to create/remove a lockspace */
492 static ssize_t device_write(struct file *file, const char __user *buf,
493 size_t count, loff_t *ppos)
495 struct dlm_user_proc *proc = file->private_data;
496 struct dlm_write_request *kbuf;
497 sigset_t tmpsig, allsigs;
501 if (count < sizeof(struct dlm_write_request32))
503 if (count < sizeof(struct dlm_write_request))
507 kbuf = kmalloc(count, GFP_KERNEL);
511 if (copy_from_user(kbuf, buf, count)) {
516 if (check_version(kbuf)) {
522 if (!kbuf->is64bit) {
523 struct dlm_write_request32 *k32buf;
524 k32buf = (struct dlm_write_request32 *)kbuf;
525 kbuf = kmalloc(count + (sizeof(struct dlm_write_request) -
526 sizeof(struct dlm_write_request32)), GFP_KERNEL);
531 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
532 compat_input(kbuf, k32buf);
537 /* do we really need this? can a write happen after a close? */
538 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
539 test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
542 sigfillset(&allsigs);
543 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
551 log_print("no locking on control device");
554 error = device_user_lock(proc, &kbuf->i.lock);
557 case DLM_USER_UNLOCK:
559 log_print("no locking on control device");
562 error = device_user_unlock(proc, &kbuf->i.lock);
565 case DLM_USER_DEADLOCK:
567 log_print("no locking on control device");
570 error = device_user_deadlock(proc, &kbuf->i.lock);
573 case DLM_USER_CREATE_LOCKSPACE:
575 log_print("create/remove only on control device");
578 error = device_create_lockspace(&kbuf->i.lspace);
581 case DLM_USER_REMOVE_LOCKSPACE:
583 log_print("create/remove only on control device");
586 error = device_remove_lockspace(&kbuf->i.lspace);
591 log_print("no locking on control device");
594 error = device_user_purge(proc, &kbuf->i.purge);
598 log_print("Unknown command passed to DLM device : %d\n",
603 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
610 /* Every process that opens the lockspace device has its own "proc" structure
611 hanging off the open file that's used to keep track of locks owned by the
612 process and asts that need to be delivered to the process. */
614 static int device_open(struct inode *inode, struct file *file)
616 struct dlm_user_proc *proc;
619 ls = dlm_find_lockspace_device(iminor(inode));
623 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
625 dlm_put_lockspace(ls);
629 proc->lockspace = ls->ls_local_handle;
630 INIT_LIST_HEAD(&proc->asts);
631 INIT_LIST_HEAD(&proc->locks);
632 INIT_LIST_HEAD(&proc->unlocking);
633 spin_lock_init(&proc->asts_spin);
634 spin_lock_init(&proc->locks_spin);
635 init_waitqueue_head(&proc->wait);
636 file->private_data = proc;
641 static int device_close(struct inode *inode, struct file *file)
643 struct dlm_user_proc *proc = file->private_data;
645 sigset_t tmpsig, allsigs;
647 ls = dlm_find_lockspace_local(proc->lockspace);
651 sigfillset(&allsigs);
652 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
654 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
656 dlm_clear_proc_locks(ls, proc);
658 /* at this point no more lkb's should exist for this lockspace,
659 so there's no chance of dlm_user_add_ast() being called and
660 looking for lkb->ua->proc */
663 file->private_data = NULL;
665 dlm_put_lockspace(ls);
666 dlm_put_lockspace(ls); /* for the find in device_open() */
668 /* FIXME: AUTOFREE: if this ls is no longer used do
669 device_remove_lockspace() */
671 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
677 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
678 int bmode, char __user *buf, size_t count)
681 struct dlm_lock_result32 result32;
683 struct dlm_lock_result result;
689 memset(&result, 0, sizeof(struct dlm_lock_result));
690 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
691 result.version[1] = DLM_DEVICE_VERSION_MINOR;
692 result.version[2] = DLM_DEVICE_VERSION_PATCH;
693 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
694 result.user_lksb = ua->user_lksb;
696 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
697 in a conversion unless the conversion is successful. See code
698 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
699 notes that a new blocking AST address and parameter are set even if
700 the conversion fails, so maybe we should just do that. */
702 if (type == AST_BAST) {
703 result.user_astaddr = ua->bastaddr;
704 result.user_astparam = ua->bastparam;
705 result.bast_mode = bmode;
707 result.user_astaddr = ua->castaddr;
708 result.user_astparam = ua->castparam;
713 len = sizeof(struct dlm_lock_result32);
716 len = sizeof(struct dlm_lock_result);
719 /* copy lvb to userspace if there is one, it's been updated, and
720 the user buffer has space for it */
722 if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
723 count >= len + DLM_USER_LVB_LEN) {
724 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
730 result.lvb_offset = len;
731 len += DLM_USER_LVB_LEN;
738 compat_output(&result, &result32);
739 resultptr = &result32;
743 if (copy_to_user(buf, resultptr, struct_len))
751 static int copy_version_to_user(char __user *buf, size_t count)
753 struct dlm_device_version ver;
755 memset(&ver, 0, sizeof(struct dlm_device_version));
756 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
757 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
758 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
760 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
762 return sizeof(struct dlm_device_version);
765 /* a read returns a single ast described in a struct dlm_lock_result */
767 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
770 struct dlm_user_proc *proc = file->private_data;
772 struct dlm_user_args *ua;
773 DECLARE_WAITQUEUE(wait, current);
774 int error, type=0, bmode=0, removed = 0;
776 if (count == sizeof(struct dlm_device_version)) {
777 error = copy_version_to_user(buf, count);
782 log_print("non-version read from control device %zu", count);
787 if (count < sizeof(struct dlm_lock_result32))
789 if (count < sizeof(struct dlm_lock_result))
793 /* do we really need this? can a read happen after a close? */
794 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
797 spin_lock(&proc->asts_spin);
798 if (list_empty(&proc->asts)) {
799 if (file->f_flags & O_NONBLOCK) {
800 spin_unlock(&proc->asts_spin);
804 add_wait_queue(&proc->wait, &wait);
807 set_current_state(TASK_INTERRUPTIBLE);
808 if (list_empty(&proc->asts) && !signal_pending(current)) {
809 spin_unlock(&proc->asts_spin);
811 spin_lock(&proc->asts_spin);
814 set_current_state(TASK_RUNNING);
815 remove_wait_queue(&proc->wait, &wait);
817 if (signal_pending(current)) {
818 spin_unlock(&proc->asts_spin);
823 /* there may be both completion and blocking asts to return for
824 the lkb, don't remove lkb from asts list unless no asts remain */
826 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
828 if (lkb->lkb_ast_type & AST_COMP) {
829 lkb->lkb_ast_type &= ~AST_COMP;
831 } else if (lkb->lkb_ast_type & AST_BAST) {
832 lkb->lkb_ast_type &= ~AST_BAST;
834 bmode = lkb->lkb_bastmode;
837 if (!lkb->lkb_ast_type) {
838 list_del(&lkb->lkb_astqueue);
841 spin_unlock(&proc->asts_spin);
843 ua = (struct dlm_user_args *)lkb->lkb_astparam;
844 error = copy_result_to_user(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 int dlm_user_init(void)
903 ctl_device.name = "dlm-control";
904 ctl_device.fops = &ctl_device_fops;
905 ctl_device.minor = MISC_DYNAMIC_MINOR;
907 error = misc_register(&ctl_device);
909 log_print("misc_register failed for control device");
914 void dlm_user_exit(void)
916 misc_deregister(&ctl_device);