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 /* we could possibly check if the cancel of an orphan has resulted in the lkb
142 being removed and then remove that lkb from the orphans list and free it */
144 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
147 struct dlm_user_args *ua;
148 struct dlm_user_proc *proc;
149 int eol = 0, ast_type;
151 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
154 ls = lkb->lkb_resource->res_ls;
155 mutex_lock(&ls->ls_clear_proc_locks);
157 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
158 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
159 lkb->ua so we can't try to use it. This second check is necessary
160 for cases where a completion ast is received for an operation that
161 began before clear_proc_locks did its cancel/unlock. */
163 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
166 DLM_ASSERT(lkb->lkb_astparam, dlm_print_lkb(lkb););
167 ua = (struct dlm_user_args *)lkb->lkb_astparam;
170 if (type == AST_BAST && ua->bastaddr == NULL)
173 spin_lock(&proc->asts_spin);
175 ast_type = lkb->lkb_ast_type;
176 lkb->lkb_ast_type |= type;
179 kref_get(&lkb->lkb_ref);
180 list_add_tail(&lkb->lkb_astqueue, &proc->asts);
181 wake_up_interruptible(&proc->wait);
183 if (type == AST_COMP && (ast_type & AST_COMP))
184 log_debug(ls, "ast overlap %x status %x %x",
185 lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
187 /* Figure out if this lock is at the end of its life and no longer
188 available for the application to use. The lkb still exists until
189 the final ast is read. A lock becomes EOL in three situations:
190 1. a noqueue request fails with EAGAIN
191 2. an unlock completes with EUNLOCK
192 3. a cancel of a waiting request completes with ECANCEL
193 An EOL lock needs to be removed from the process's list of locks.
194 And we can't allow any new operation on an EOL lock. This is
195 not related to the lifetime of the lkb struct which is managed
196 entirely by refcount. */
198 if (type == AST_COMP &&
199 lkb->lkb_grmode == DLM_LOCK_IV &&
200 ua->lksb.sb_status == -EAGAIN)
202 else if (ua->lksb.sb_status == -DLM_EUNLOCK ||
203 (ua->lksb.sb_status == -DLM_ECANCEL &&
204 lkb->lkb_grmode == DLM_LOCK_IV))
207 lkb->lkb_ast_type &= ~AST_BAST;
208 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
211 /* We want to copy the lvb to userspace when the completion
212 ast is read if the status is 0, the lock has an lvb and
213 lvb_ops says we should. We could probably have set_lvb_lock()
214 set update_user_lvb instead and not need old_mode */
216 if ((lkb->lkb_ast_type & AST_COMP) &&
217 (lkb->lkb_lksb->sb_status == 0) &&
218 lkb->lkb_lksb->sb_lvbptr &&
219 dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
220 ua->update_user_lvb = 1;
222 ua->update_user_lvb = 0;
224 spin_unlock(&proc->asts_spin);
227 spin_lock(&ua->proc->locks_spin);
228 if (!list_empty(&lkb->lkb_ownqueue)) {
229 list_del_init(&lkb->lkb_ownqueue);
232 spin_unlock(&ua->proc->locks_spin);
235 mutex_unlock(&ls->ls_clear_proc_locks);
238 static int device_user_lock(struct dlm_user_proc *proc,
239 struct dlm_lock_params *params)
242 struct dlm_user_args *ua;
245 ls = dlm_find_lockspace_local(proc->lockspace);
249 if (!params->castaddr || !params->lksb) {
254 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
258 ua->user_lksb = params->lksb;
259 ua->castparam = params->castparam;
260 ua->castaddr = params->castaddr;
261 ua->bastparam = params->bastparam;
262 ua->bastaddr = params->bastaddr;
263 ua->xid = params->xid;
265 if (params->flags & DLM_LKF_CONVERT)
266 error = dlm_user_convert(ls, ua,
267 params->mode, params->flags,
268 params->lkid, params->lvb,
269 (unsigned long) params->timeout);
271 error = dlm_user_request(ls, ua,
272 params->mode, params->flags,
273 params->name, params->namelen,
274 (unsigned long) params->timeout);
276 error = ua->lksb.sb_lkid;
279 dlm_put_lockspace(ls);
283 static int device_user_unlock(struct dlm_user_proc *proc,
284 struct dlm_lock_params *params)
287 struct dlm_user_args *ua;
290 ls = dlm_find_lockspace_local(proc->lockspace);
294 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
298 ua->user_lksb = params->lksb;
299 ua->castparam = params->castparam;
300 ua->castaddr = params->castaddr;
302 if (params->flags & DLM_LKF_CANCEL)
303 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
305 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
308 dlm_put_lockspace(ls);
312 static int create_misc_device(struct dlm_ls *ls, char *name)
317 len = strlen(name) + strlen(name_prefix) + 2;
318 ls->ls_device.name = kzalloc(len, GFP_KERNEL);
319 if (!ls->ls_device.name)
322 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
324 ls->ls_device.fops = &device_fops;
325 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
327 error = misc_register(&ls->ls_device);
329 kfree(ls->ls_device.name);
335 static int device_user_purge(struct dlm_user_proc *proc,
336 struct dlm_purge_params *params)
341 ls = dlm_find_lockspace_local(proc->lockspace);
345 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
347 dlm_put_lockspace(ls);
351 static int device_create_lockspace(struct dlm_lspace_params *params)
353 dlm_lockspace_t *lockspace;
357 if (!capable(CAP_SYS_ADMIN))
360 error = dlm_new_lockspace(params->name, strlen(params->name),
361 &lockspace, params->flags, DLM_USER_LVB_LEN);
365 ls = dlm_find_lockspace_local(lockspace);
369 error = create_misc_device(ls, params->name);
370 dlm_put_lockspace(ls);
373 dlm_release_lockspace(lockspace, 0);
375 error = ls->ls_device.minor;
380 static int device_remove_lockspace(struct dlm_lspace_params *params)
382 dlm_lockspace_t *lockspace;
384 int error, force = 0;
386 if (!capable(CAP_SYS_ADMIN))
389 ls = dlm_find_lockspace_device(params->minor);
393 /* Deregister the misc device first, so we don't have
394 * a device that's not attached to a lockspace. If
395 * dlm_release_lockspace fails then we can recreate it
397 error = misc_deregister(&ls->ls_device);
399 dlm_put_lockspace(ls);
402 kfree(ls->ls_device.name);
404 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
407 lockspace = ls->ls_local_handle;
409 /* dlm_release_lockspace waits for references to go to zero,
410 so all processes will need to close their device for the ls
411 before the release will procede */
413 dlm_put_lockspace(ls);
414 error = dlm_release_lockspace(lockspace, force);
416 create_misc_device(ls, ls->ls_name);
421 /* Check the user's version matches ours */
422 static int check_version(struct dlm_write_request *req)
424 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
425 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
426 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
428 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
429 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
435 DLM_DEVICE_VERSION_MAJOR,
436 DLM_DEVICE_VERSION_MINOR,
437 DLM_DEVICE_VERSION_PATCH);
447 * dlm_user_request -> request_lock
448 * dlm_user_convert -> convert_lock
451 * dlm_user_unlock -> unlock_lock
452 * dlm_user_cancel -> cancel_lock
454 * device_create_lockspace
457 * device_remove_lockspace
458 * dlm_release_lockspace
461 /* a write to a lockspace device is a lock or unlock request, a write
462 to the control device is to create/remove a lockspace */
464 static ssize_t device_write(struct file *file, const char __user *buf,
465 size_t count, loff_t *ppos)
467 struct dlm_user_proc *proc = file->private_data;
468 struct dlm_write_request *kbuf;
469 sigset_t tmpsig, allsigs;
473 if (count < sizeof(struct dlm_write_request32))
475 if (count < sizeof(struct dlm_write_request))
479 kbuf = kmalloc(count, GFP_KERNEL);
483 if (copy_from_user(kbuf, buf, count)) {
488 if (check_version(kbuf)) {
494 if (!kbuf->is64bit) {
495 struct dlm_write_request32 *k32buf;
496 k32buf = (struct dlm_write_request32 *)kbuf;
497 kbuf = kmalloc(count + (sizeof(struct dlm_write_request) -
498 sizeof(struct dlm_write_request32)), GFP_KERNEL);
503 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
504 compat_input(kbuf, k32buf);
509 /* do we really need this? can a write happen after a close? */
510 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
511 test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
514 sigfillset(&allsigs);
515 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
523 log_print("no locking on control device");
526 error = device_user_lock(proc, &kbuf->i.lock);
529 case DLM_USER_UNLOCK:
531 log_print("no locking on control device");
534 error = device_user_unlock(proc, &kbuf->i.lock);
537 case DLM_USER_CREATE_LOCKSPACE:
539 log_print("create/remove only on control device");
542 error = device_create_lockspace(&kbuf->i.lspace);
545 case DLM_USER_REMOVE_LOCKSPACE:
547 log_print("create/remove only on control device");
550 error = device_remove_lockspace(&kbuf->i.lspace);
555 log_print("no locking on control device");
558 error = device_user_purge(proc, &kbuf->i.purge);
562 log_print("Unknown command passed to DLM device : %d\n",
567 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
574 /* Every process that opens the lockspace device has its own "proc" structure
575 hanging off the open file that's used to keep track of locks owned by the
576 process and asts that need to be delivered to the process. */
578 static int device_open(struct inode *inode, struct file *file)
580 struct dlm_user_proc *proc;
583 ls = dlm_find_lockspace_device(iminor(inode));
587 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
589 dlm_put_lockspace(ls);
593 proc->lockspace = ls->ls_local_handle;
594 INIT_LIST_HEAD(&proc->asts);
595 INIT_LIST_HEAD(&proc->locks);
596 INIT_LIST_HEAD(&proc->unlocking);
597 spin_lock_init(&proc->asts_spin);
598 spin_lock_init(&proc->locks_spin);
599 init_waitqueue_head(&proc->wait);
600 file->private_data = proc;
605 static int device_close(struct inode *inode, struct file *file)
607 struct dlm_user_proc *proc = file->private_data;
609 sigset_t tmpsig, allsigs;
611 ls = dlm_find_lockspace_local(proc->lockspace);
615 sigfillset(&allsigs);
616 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
618 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
620 dlm_clear_proc_locks(ls, proc);
622 /* at this point no more lkb's should exist for this lockspace,
623 so there's no chance of dlm_user_add_ast() being called and
624 looking for lkb->ua->proc */
627 file->private_data = NULL;
629 dlm_put_lockspace(ls);
630 dlm_put_lockspace(ls); /* for the find in device_open() */
632 /* FIXME: AUTOFREE: if this ls is no longer used do
633 device_remove_lockspace() */
635 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
641 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
642 int bmode, char __user *buf, size_t count)
645 struct dlm_lock_result32 result32;
647 struct dlm_lock_result result;
653 memset(&result, 0, sizeof(struct dlm_lock_result));
654 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
655 result.version[1] = DLM_DEVICE_VERSION_MINOR;
656 result.version[2] = DLM_DEVICE_VERSION_PATCH;
657 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
658 result.user_lksb = ua->user_lksb;
660 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
661 in a conversion unless the conversion is successful. See code
662 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
663 notes that a new blocking AST address and parameter are set even if
664 the conversion fails, so maybe we should just do that. */
666 if (type == AST_BAST) {
667 result.user_astaddr = ua->bastaddr;
668 result.user_astparam = ua->bastparam;
669 result.bast_mode = bmode;
671 result.user_astaddr = ua->castaddr;
672 result.user_astparam = ua->castparam;
677 len = sizeof(struct dlm_lock_result32);
680 len = sizeof(struct dlm_lock_result);
683 /* copy lvb to userspace if there is one, it's been updated, and
684 the user buffer has space for it */
686 if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
687 count >= len + DLM_USER_LVB_LEN) {
688 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
694 result.lvb_offset = len;
695 len += DLM_USER_LVB_LEN;
702 compat_output(&result, &result32);
703 resultptr = &result32;
707 if (copy_to_user(buf, resultptr, struct_len))
715 static int copy_version_to_user(char __user *buf, size_t count)
717 struct dlm_device_version ver;
719 memset(&ver, 0, sizeof(struct dlm_device_version));
720 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
721 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
722 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
724 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
726 return sizeof(struct dlm_device_version);
729 /* a read returns a single ast described in a struct dlm_lock_result */
731 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
734 struct dlm_user_proc *proc = file->private_data;
736 struct dlm_user_args *ua;
737 DECLARE_WAITQUEUE(wait, current);
738 int error, type=0, bmode=0, removed = 0;
740 if (count == sizeof(struct dlm_device_version)) {
741 error = copy_version_to_user(buf, count);
746 log_print("non-version read from control device %zu", count);
751 if (count < sizeof(struct dlm_lock_result32))
753 if (count < sizeof(struct dlm_lock_result))
757 /* do we really need this? can a read happen after a close? */
758 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
761 spin_lock(&proc->asts_spin);
762 if (list_empty(&proc->asts)) {
763 if (file->f_flags & O_NONBLOCK) {
764 spin_unlock(&proc->asts_spin);
768 add_wait_queue(&proc->wait, &wait);
771 set_current_state(TASK_INTERRUPTIBLE);
772 if (list_empty(&proc->asts) && !signal_pending(current)) {
773 spin_unlock(&proc->asts_spin);
775 spin_lock(&proc->asts_spin);
778 set_current_state(TASK_RUNNING);
779 remove_wait_queue(&proc->wait, &wait);
781 if (signal_pending(current)) {
782 spin_unlock(&proc->asts_spin);
787 /* there may be both completion and blocking asts to return for
788 the lkb, don't remove lkb from asts list unless no asts remain */
790 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
792 if (lkb->lkb_ast_type & AST_COMP) {
793 lkb->lkb_ast_type &= ~AST_COMP;
795 } else if (lkb->lkb_ast_type & AST_BAST) {
796 lkb->lkb_ast_type &= ~AST_BAST;
798 bmode = lkb->lkb_bastmode;
801 if (!lkb->lkb_ast_type) {
802 list_del(&lkb->lkb_astqueue);
805 spin_unlock(&proc->asts_spin);
807 ua = (struct dlm_user_args *)lkb->lkb_astparam;
808 error = copy_result_to_user(ua,
809 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
810 type, bmode, buf, count);
812 /* removes reference for the proc->asts lists added by
813 dlm_user_add_ast() and may result in the lkb being freed */
820 static unsigned int device_poll(struct file *file, poll_table *wait)
822 struct dlm_user_proc *proc = file->private_data;
824 poll_wait(file, &proc->wait, wait);
826 spin_lock(&proc->asts_spin);
827 if (!list_empty(&proc->asts)) {
828 spin_unlock(&proc->asts_spin);
829 return POLLIN | POLLRDNORM;
831 spin_unlock(&proc->asts_spin);
835 static int ctl_device_open(struct inode *inode, struct file *file)
837 file->private_data = NULL;
841 static int ctl_device_close(struct inode *inode, struct file *file)
846 static const struct file_operations device_fops = {
848 .release = device_close,
850 .write = device_write,
852 .owner = THIS_MODULE,
855 static const struct file_operations ctl_device_fops = {
856 .open = ctl_device_open,
857 .release = ctl_device_close,
859 .write = device_write,
860 .owner = THIS_MODULE,
863 int dlm_user_init(void)
867 ctl_device.name = "dlm-control";
868 ctl_device.fops = &ctl_device_fops;
869 ctl_device.minor = MISC_DYNAMIC_MINOR;
871 error = misc_register(&ctl_device);
873 log_print("misc_register failed for control device");
878 void dlm_user_exit(void)
880 misc_deregister(&ctl_device);