xref: /openbmc/linux/fs/dlm/plock.c (revision 63dc02bd)
1 /*
2  * Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
3  *
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 version 2.
7  */
8 
9 #include <linux/fs.h>
10 #include <linux/miscdevice.h>
11 #include <linux/poll.h>
12 #include <linux/dlm.h>
13 #include <linux/dlm_plock.h>
14 #include <linux/slab.h>
15 
16 #include "dlm_internal.h"
17 #include "lockspace.h"
18 
19 static spinlock_t ops_lock;
20 static struct list_head send_list;
21 static struct list_head recv_list;
22 static wait_queue_head_t send_wq;
23 static wait_queue_head_t recv_wq;
24 
25 struct plock_op {
26 	struct list_head list;
27 	int done;
28 	struct dlm_plock_info info;
29 };
30 
31 struct plock_xop {
32 	struct plock_op xop;
33 	void *callback;
34 	void *fl;
35 	void *file;
36 	struct file_lock flc;
37 };
38 
39 
40 static inline void set_version(struct dlm_plock_info *info)
41 {
42 	info->version[0] = DLM_PLOCK_VERSION_MAJOR;
43 	info->version[1] = DLM_PLOCK_VERSION_MINOR;
44 	info->version[2] = DLM_PLOCK_VERSION_PATCH;
45 }
46 
47 static int check_version(struct dlm_plock_info *info)
48 {
49 	if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
50 	    (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
51 		log_print("plock device version mismatch: "
52 			  "kernel (%u.%u.%u), user (%u.%u.%u)",
53 			  DLM_PLOCK_VERSION_MAJOR,
54 			  DLM_PLOCK_VERSION_MINOR,
55 			  DLM_PLOCK_VERSION_PATCH,
56 			  info->version[0],
57 			  info->version[1],
58 			  info->version[2]);
59 		return -EINVAL;
60 	}
61 	return 0;
62 }
63 
64 static void send_op(struct plock_op *op)
65 {
66 	set_version(&op->info);
67 	INIT_LIST_HEAD(&op->list);
68 	spin_lock(&ops_lock);
69 	list_add_tail(&op->list, &send_list);
70 	spin_unlock(&ops_lock);
71 	wake_up(&send_wq);
72 }
73 
74 /* If a process was killed while waiting for the only plock on a file,
75    locks_remove_posix will not see any lock on the file so it won't
76    send an unlock-close to us to pass on to userspace to clean up the
77    abandoned waiter.  So, we have to insert the unlock-close when the
78    lock call is interrupted. */
79 
80 static void do_unlock_close(struct dlm_ls *ls, u64 number,
81 			    struct file *file, struct file_lock *fl)
82 {
83 	struct plock_op *op;
84 
85 	op = kzalloc(sizeof(*op), GFP_NOFS);
86 	if (!op)
87 		return;
88 
89 	op->info.optype		= DLM_PLOCK_OP_UNLOCK;
90 	op->info.pid		= fl->fl_pid;
91 	op->info.fsid		= ls->ls_global_id;
92 	op->info.number		= number;
93 	op->info.start		= 0;
94 	op->info.end		= OFFSET_MAX;
95 	if (fl->fl_lmops && fl->fl_lmops->lm_grant)
96 		op->info.owner	= (__u64) fl->fl_pid;
97 	else
98 		op->info.owner	= (__u64)(long) fl->fl_owner;
99 
100 	op->info.flags |= DLM_PLOCK_FL_CLOSE;
101 	send_op(op);
102 }
103 
104 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
105 		   int cmd, struct file_lock *fl)
106 {
107 	struct dlm_ls *ls;
108 	struct plock_op *op;
109 	struct plock_xop *xop;
110 	int rv;
111 
112 	ls = dlm_find_lockspace_local(lockspace);
113 	if (!ls)
114 		return -EINVAL;
115 
116 	xop = kzalloc(sizeof(*xop), GFP_NOFS);
117 	if (!xop) {
118 		rv = -ENOMEM;
119 		goto out;
120 	}
121 
122 	op = &xop->xop;
123 	op->info.optype		= DLM_PLOCK_OP_LOCK;
124 	op->info.pid		= fl->fl_pid;
125 	op->info.ex		= (fl->fl_type == F_WRLCK);
126 	op->info.wait		= IS_SETLKW(cmd);
127 	op->info.fsid		= ls->ls_global_id;
128 	op->info.number		= number;
129 	op->info.start		= fl->fl_start;
130 	op->info.end		= fl->fl_end;
131 	if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
132 		/* fl_owner is lockd which doesn't distinguish
133 		   processes on the nfs client */
134 		op->info.owner	= (__u64) fl->fl_pid;
135 		xop->callback	= fl->fl_lmops->lm_grant;
136 		locks_init_lock(&xop->flc);
137 		locks_copy_lock(&xop->flc, fl);
138 		xop->fl		= fl;
139 		xop->file	= file;
140 	} else {
141 		op->info.owner	= (__u64)(long) fl->fl_owner;
142 		xop->callback	= NULL;
143 	}
144 
145 	send_op(op);
146 
147 	if (xop->callback == NULL) {
148 		rv = wait_event_killable(recv_wq, (op->done != 0));
149 		if (rv == -ERESTARTSYS) {
150 			log_debug(ls, "dlm_posix_lock: wait killed %llx",
151 				  (unsigned long long)number);
152 			spin_lock(&ops_lock);
153 			list_del(&op->list);
154 			spin_unlock(&ops_lock);
155 			kfree(xop);
156 			do_unlock_close(ls, number, file, fl);
157 			goto out;
158 		}
159 	} else {
160 		rv = FILE_LOCK_DEFERRED;
161 		goto out;
162 	}
163 
164 	spin_lock(&ops_lock);
165 	if (!list_empty(&op->list)) {
166 		log_error(ls, "dlm_posix_lock: op on list %llx",
167 			  (unsigned long long)number);
168 		list_del(&op->list);
169 	}
170 	spin_unlock(&ops_lock);
171 
172 	rv = op->info.rv;
173 
174 	if (!rv) {
175 		if (posix_lock_file_wait(file, fl) < 0)
176 			log_error(ls, "dlm_posix_lock: vfs lock error %llx",
177 				  (unsigned long long)number);
178 	}
179 
180 	kfree(xop);
181 out:
182 	dlm_put_lockspace(ls);
183 	return rv;
184 }
185 EXPORT_SYMBOL_GPL(dlm_posix_lock);
186 
187 /* Returns failure iff a successful lock operation should be canceled */
188 static int dlm_plock_callback(struct plock_op *op)
189 {
190 	struct file *file;
191 	struct file_lock *fl;
192 	struct file_lock *flc;
193 	int (*notify)(void *, void *, int) = NULL;
194 	struct plock_xop *xop = (struct plock_xop *)op;
195 	int rv = 0;
196 
197 	spin_lock(&ops_lock);
198 	if (!list_empty(&op->list)) {
199 		log_print("dlm_plock_callback: op on list %llx",
200 			  (unsigned long long)op->info.number);
201 		list_del(&op->list);
202 	}
203 	spin_unlock(&ops_lock);
204 
205 	/* check if the following 2 are still valid or make a copy */
206 	file = xop->file;
207 	flc = &xop->flc;
208 	fl = xop->fl;
209 	notify = xop->callback;
210 
211 	if (op->info.rv) {
212 		notify(fl, NULL, op->info.rv);
213 		goto out;
214 	}
215 
216 	/* got fs lock; bookkeep locally as well: */
217 	flc->fl_flags &= ~FL_SLEEP;
218 	if (posix_lock_file(file, flc, NULL)) {
219 		/*
220 		 * This can only happen in the case of kmalloc() failure.
221 		 * The filesystem's own lock is the authoritative lock,
222 		 * so a failure to get the lock locally is not a disaster.
223 		 * As long as the fs cannot reliably cancel locks (especially
224 		 * in a low-memory situation), we're better off ignoring
225 		 * this failure than trying to recover.
226 		 */
227 		log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
228 			  (unsigned long long)op->info.number, file, fl);
229 	}
230 
231 	rv = notify(fl, NULL, 0);
232 	if (rv) {
233 		/* XXX: We need to cancel the fs lock here: */
234 		log_print("dlm_plock_callback: lock granted after lock request "
235 			  "failed; dangling lock!\n");
236 		goto out;
237 	}
238 
239 out:
240 	kfree(xop);
241 	return rv;
242 }
243 
244 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
245 		     struct file_lock *fl)
246 {
247 	struct dlm_ls *ls;
248 	struct plock_op *op;
249 	int rv;
250 
251 	ls = dlm_find_lockspace_local(lockspace);
252 	if (!ls)
253 		return -EINVAL;
254 
255 	op = kzalloc(sizeof(*op), GFP_NOFS);
256 	if (!op) {
257 		rv = -ENOMEM;
258 		goto out;
259 	}
260 
261 	if (posix_lock_file_wait(file, fl) < 0)
262 		log_error(ls, "dlm_posix_unlock: vfs unlock error %llx",
263 			  (unsigned long long)number);
264 
265 	op->info.optype		= DLM_PLOCK_OP_UNLOCK;
266 	op->info.pid		= fl->fl_pid;
267 	op->info.fsid		= ls->ls_global_id;
268 	op->info.number		= number;
269 	op->info.start		= fl->fl_start;
270 	op->info.end		= fl->fl_end;
271 	if (fl->fl_lmops && fl->fl_lmops->lm_grant)
272 		op->info.owner	= (__u64) fl->fl_pid;
273 	else
274 		op->info.owner	= (__u64)(long) fl->fl_owner;
275 
276 	if (fl->fl_flags & FL_CLOSE) {
277 		op->info.flags |= DLM_PLOCK_FL_CLOSE;
278 		send_op(op);
279 		rv = 0;
280 		goto out;
281 	}
282 
283 	send_op(op);
284 	wait_event(recv_wq, (op->done != 0));
285 
286 	spin_lock(&ops_lock);
287 	if (!list_empty(&op->list)) {
288 		log_error(ls, "dlm_posix_unlock: op on list %llx",
289 			  (unsigned long long)number);
290 		list_del(&op->list);
291 	}
292 	spin_unlock(&ops_lock);
293 
294 	rv = op->info.rv;
295 
296 	if (rv == -ENOENT)
297 		rv = 0;
298 
299 	kfree(op);
300 out:
301 	dlm_put_lockspace(ls);
302 	return rv;
303 }
304 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
305 
306 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
307 		  struct file_lock *fl)
308 {
309 	struct dlm_ls *ls;
310 	struct plock_op *op;
311 	int rv;
312 
313 	ls = dlm_find_lockspace_local(lockspace);
314 	if (!ls)
315 		return -EINVAL;
316 
317 	op = kzalloc(sizeof(*op), GFP_NOFS);
318 	if (!op) {
319 		rv = -ENOMEM;
320 		goto out;
321 	}
322 
323 	op->info.optype		= DLM_PLOCK_OP_GET;
324 	op->info.pid		= fl->fl_pid;
325 	op->info.ex		= (fl->fl_type == F_WRLCK);
326 	op->info.fsid		= ls->ls_global_id;
327 	op->info.number		= number;
328 	op->info.start		= fl->fl_start;
329 	op->info.end		= fl->fl_end;
330 	if (fl->fl_lmops && fl->fl_lmops->lm_grant)
331 		op->info.owner	= (__u64) fl->fl_pid;
332 	else
333 		op->info.owner	= (__u64)(long) fl->fl_owner;
334 
335 	send_op(op);
336 	wait_event(recv_wq, (op->done != 0));
337 
338 	spin_lock(&ops_lock);
339 	if (!list_empty(&op->list)) {
340 		log_error(ls, "dlm_posix_get: op on list %llx",
341 			  (unsigned long long)number);
342 		list_del(&op->list);
343 	}
344 	spin_unlock(&ops_lock);
345 
346 	/* info.rv from userspace is 1 for conflict, 0 for no-conflict,
347 	   -ENOENT if there are no locks on the file */
348 
349 	rv = op->info.rv;
350 
351 	fl->fl_type = F_UNLCK;
352 	if (rv == -ENOENT)
353 		rv = 0;
354 	else if (rv > 0) {
355 		locks_init_lock(fl);
356 		fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
357 		fl->fl_flags = FL_POSIX;
358 		fl->fl_pid = op->info.pid;
359 		fl->fl_start = op->info.start;
360 		fl->fl_end = op->info.end;
361 		rv = 0;
362 	}
363 
364 	kfree(op);
365 out:
366 	dlm_put_lockspace(ls);
367 	return rv;
368 }
369 EXPORT_SYMBOL_GPL(dlm_posix_get);
370 
371 /* a read copies out one plock request from the send list */
372 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
373 			loff_t *ppos)
374 {
375 	struct dlm_plock_info info;
376 	struct plock_op *op = NULL;
377 
378 	if (count < sizeof(info))
379 		return -EINVAL;
380 
381 	spin_lock(&ops_lock);
382 	if (!list_empty(&send_list)) {
383 		op = list_entry(send_list.next, struct plock_op, list);
384 		if (op->info.flags & DLM_PLOCK_FL_CLOSE)
385 			list_del(&op->list);
386 		else
387 			list_move(&op->list, &recv_list);
388 		memcpy(&info, &op->info, sizeof(info));
389 	}
390 	spin_unlock(&ops_lock);
391 
392 	if (!op)
393 		return -EAGAIN;
394 
395 	/* there is no need to get a reply from userspace for unlocks
396 	   that were generated by the vfs cleaning up for a close
397 	   (the process did not make an unlock call). */
398 
399 	if (op->info.flags & DLM_PLOCK_FL_CLOSE)
400 		kfree(op);
401 
402 	if (copy_to_user(u, &info, sizeof(info)))
403 		return -EFAULT;
404 	return sizeof(info);
405 }
406 
407 /* a write copies in one plock result that should match a plock_op
408    on the recv list */
409 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
410 			 loff_t *ppos)
411 {
412 	struct dlm_plock_info info;
413 	struct plock_op *op;
414 	int found = 0, do_callback = 0;
415 
416 	if (count != sizeof(info))
417 		return -EINVAL;
418 
419 	if (copy_from_user(&info, u, sizeof(info)))
420 		return -EFAULT;
421 
422 	if (check_version(&info))
423 		return -EINVAL;
424 
425 	spin_lock(&ops_lock);
426 	list_for_each_entry(op, &recv_list, list) {
427 		if (op->info.fsid == info.fsid &&
428 		    op->info.number == info.number &&
429 		    op->info.owner == info.owner) {
430 			struct plock_xop *xop = (struct plock_xop *)op;
431 			list_del_init(&op->list);
432 			memcpy(&op->info, &info, sizeof(info));
433 			if (xop->callback)
434 				do_callback = 1;
435 			else
436 				op->done = 1;
437 			found = 1;
438 			break;
439 		}
440 	}
441 	spin_unlock(&ops_lock);
442 
443 	if (found) {
444 		if (do_callback)
445 			dlm_plock_callback(op);
446 		else
447 			wake_up(&recv_wq);
448 	} else
449 		log_print("dev_write no op %x %llx", info.fsid,
450 			  (unsigned long long)info.number);
451 	return count;
452 }
453 
454 static unsigned int dev_poll(struct file *file, poll_table *wait)
455 {
456 	unsigned int mask = 0;
457 
458 	poll_wait(file, &send_wq, wait);
459 
460 	spin_lock(&ops_lock);
461 	if (!list_empty(&send_list))
462 		mask = POLLIN | POLLRDNORM;
463 	spin_unlock(&ops_lock);
464 
465 	return mask;
466 }
467 
468 static const struct file_operations dev_fops = {
469 	.read    = dev_read,
470 	.write   = dev_write,
471 	.poll    = dev_poll,
472 	.owner   = THIS_MODULE,
473 	.llseek  = noop_llseek,
474 };
475 
476 static struct miscdevice plock_dev_misc = {
477 	.minor = MISC_DYNAMIC_MINOR,
478 	.name = DLM_PLOCK_MISC_NAME,
479 	.fops = &dev_fops
480 };
481 
482 int dlm_plock_init(void)
483 {
484 	int rv;
485 
486 	spin_lock_init(&ops_lock);
487 	INIT_LIST_HEAD(&send_list);
488 	INIT_LIST_HEAD(&recv_list);
489 	init_waitqueue_head(&send_wq);
490 	init_waitqueue_head(&recv_wq);
491 
492 	rv = misc_register(&plock_dev_misc);
493 	if (rv)
494 		log_print("dlm_plock_init: misc_register failed %d", rv);
495 	return rv;
496 }
497 
498 void dlm_plock_exit(void)
499 {
500 	if (misc_deregister(&plock_dev_misc) < 0)
501 		log_print("dlm_plock_exit: misc_deregister failed");
502 }
503 
504