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