autofs4: fix pending mount race
[linux-2.6-block.git] / fs / autofs4 / waitq.c
CommitLineData
1da177e4
LT
1/* -*- c -*- --------------------------------------------------------------- *
2 *
3 * linux/fs/autofs/waitq.c
4 *
5 * Copyright 1997-1998 Transmeta Corporation -- All Rights Reserved
5c0a32fc 6 * Copyright 2001-2006 Ian Kent <raven@themaw.net>
1da177e4
LT
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * ------------------------------------------------------------------------- */
13
14#include <linux/slab.h>
15#include <linux/time.h>
16#include <linux/signal.h>
17#include <linux/file.h>
18#include "autofs_i.h"
19
20/* We make this a static variable rather than a part of the superblock; it
21 is better if we don't reassign numbers easily even across filesystems */
22static autofs_wqt_t autofs4_next_wait_queue = 1;
23
24/* These are the signals we allow interrupting a pending mount */
25#define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGINT) | sigmask(SIGQUIT))
26
27void autofs4_catatonic_mode(struct autofs_sb_info *sbi)
28{
29 struct autofs_wait_queue *wq, *nwq;
30
5a11d4d0
IK
31 mutex_lock(&sbi->wq_mutex);
32 if (sbi->catatonic) {
33 mutex_unlock(&sbi->wq_mutex);
34 return;
35 }
36
1da177e4
LT
37 DPRINTK("entering catatonic mode");
38
39 sbi->catatonic = 1;
40 wq = sbi->queues;
41 sbi->queues = NULL; /* Erase all wait queues */
e77fbddf 42 while (wq) {
1da177e4
LT
43 nwq = wq->next;
44 wq->status = -ENOENT; /* Magic is gone - report failure */
70b52a0a
JM
45 if (wq->name.name) {
46 kfree(wq->name.name);
47 wq->name.name = NULL;
48 }
1da177e4
LT
49 wake_up_interruptible(&wq->queue);
50 wq = nwq;
51 }
ba8df43c
IK
52 fput(sbi->pipe); /* Close the pipe */
53 sbi->pipe = NULL;
5a11d4d0
IK
54 sbi->pipefd = -1;
55 mutex_unlock(&sbi->wq_mutex);
1da177e4
LT
56}
57
58static int autofs4_write(struct file *file, const void *addr, int bytes)
59{
60 unsigned long sigpipe, flags;
61 mm_segment_t fs;
62 const char *data = (const char *)addr;
63 ssize_t wr = 0;
64
65 /** WARNING: this is not safe for writing more than PIPE_BUF bytes! **/
66
67 sigpipe = sigismember(&current->pending.signal, SIGPIPE);
68
69 /* Save pointer to user space and point back to kernel space */
70 fs = get_fs();
71 set_fs(KERNEL_DS);
72
73 while (bytes &&
74 (wr = file->f_op->write(file,data,bytes,&file->f_pos)) > 0) {
75 data += wr;
76 bytes -= wr;
77 }
78
79 set_fs(fs);
80
81 /* Keep the currently executing process from receiving a
82 SIGPIPE unless it was already supposed to get one */
83 if (wr == -EPIPE && !sigpipe) {
84 spin_lock_irqsave(&current->sighand->siglock, flags);
85 sigdelset(&current->pending.signal, SIGPIPE);
86 recalc_sigpending();
87 spin_unlock_irqrestore(&current->sighand->siglock, flags);
88 }
89
90 return (bytes > 0);
91}
92
93static void autofs4_notify_daemon(struct autofs_sb_info *sbi,
94 struct autofs_wait_queue *wq,
95 int type)
96{
e8514478
IK
97 union {
98 struct autofs_packet_hdr hdr;
99 union autofs_packet_union v4_pkt;
100 union autofs_v5_packet_union v5_pkt;
101 } pkt;
1da177e4
LT
102 size_t pktsz;
103
104 DPRINTK("wait id = 0x%08lx, name = %.*s, type=%d",
70b52a0a 105 wq->wait_queue_token, wq->name.len, wq->name.name, type);
1da177e4
LT
106
107 memset(&pkt,0,sizeof pkt); /* For security reasons */
108
109 pkt.hdr.proto_version = sbi->version;
110 pkt.hdr.type = type;
5c0a32fc
IK
111 switch (type) {
112 /* Kernel protocol v4 missing and expire packets */
113 case autofs_ptype_missing:
114 {
e8514478 115 struct autofs_packet_missing *mp = &pkt.v4_pkt.missing;
1da177e4
LT
116
117 pktsz = sizeof(*mp);
118
119 mp->wait_queue_token = wq->wait_queue_token;
70b52a0a
JM
120 mp->len = wq->name.len;
121 memcpy(mp->name, wq->name.name, wq->name.len);
122 mp->name[wq->name.len] = '\0';
5c0a32fc
IK
123 break;
124 }
125 case autofs_ptype_expire_multi:
126 {
e8514478 127 struct autofs_packet_expire_multi *ep = &pkt.v4_pkt.expire_multi;
1da177e4
LT
128
129 pktsz = sizeof(*ep);
130
131 ep->wait_queue_token = wq->wait_queue_token;
70b52a0a
JM
132 ep->len = wq->name.len;
133 memcpy(ep->name, wq->name.name, wq->name.len);
134 ep->name[wq->name.len] = '\0';
5c0a32fc
IK
135 break;
136 }
137 /*
138 * Kernel protocol v5 packet for handling indirect and direct
139 * mount missing and expire requests
140 */
141 case autofs_ptype_missing_indirect:
142 case autofs_ptype_expire_indirect:
143 case autofs_ptype_missing_direct:
144 case autofs_ptype_expire_direct:
145 {
e8514478 146 struct autofs_v5_packet *packet = &pkt.v5_pkt.v5_packet;
5c0a32fc
IK
147
148 pktsz = sizeof(*packet);
149
150 packet->wait_queue_token = wq->wait_queue_token;
70b52a0a
JM
151 packet->len = wq->name.len;
152 memcpy(packet->name, wq->name.name, wq->name.len);
153 packet->name[wq->name.len] = '\0';
5c0a32fc
IK
154 packet->dev = wq->dev;
155 packet->ino = wq->ino;
156 packet->uid = wq->uid;
157 packet->gid = wq->gid;
158 packet->pid = wq->pid;
159 packet->tgid = wq->tgid;
160 break;
161 }
162 default:
1da177e4
LT
163 printk("autofs4_notify_daemon: bad type %d!\n", type);
164 return;
165 }
166
167 if (autofs4_write(sbi->pipe, &pkt, pktsz))
168 autofs4_catatonic_mode(sbi);
169}
170
171static int autofs4_getpath(struct autofs_sb_info *sbi,
172 struct dentry *dentry, char **name)
173{
174 struct dentry *root = sbi->sb->s_root;
175 struct dentry *tmp;
176 char *buf = *name;
177 char *p;
178 int len = 0;
179
180 spin_lock(&dcache_lock);
181 for (tmp = dentry ; tmp != root ; tmp = tmp->d_parent)
182 len += tmp->d_name.len + 1;
183
cab0936a 184 if (!len || --len > NAME_MAX) {
1da177e4
LT
185 spin_unlock(&dcache_lock);
186 return 0;
187 }
188
189 *(buf + len) = '\0';
190 p = buf + len - dentry->d_name.len;
191 strncpy(p, dentry->d_name.name, dentry->d_name.len);
192
193 for (tmp = dentry->d_parent; tmp != root ; tmp = tmp->d_parent) {
194 *(--p) = '/';
195 p -= tmp->d_name.len;
196 strncpy(p, tmp->d_name.name, tmp->d_name.len);
197 }
198 spin_unlock(&dcache_lock);
199
200 return len;
201}
202
a5370553 203static struct autofs_wait_queue *
70b52a0a 204autofs4_find_wait(struct autofs_sb_info *sbi, struct qstr *qstr)
a5370553
IK
205{
206 struct autofs_wait_queue *wq;
207
208 for (wq = sbi->queues; wq; wq = wq->next) {
70b52a0a
JM
209 if (wq->name.hash == qstr->hash &&
210 wq->name.len == qstr->len &&
211 wq->name.name &&
212 !memcmp(wq->name.name, qstr->name, qstr->len))
a5370553
IK
213 break;
214 }
215 return wq;
216}
217
a1362fe9
IK
218/*
219 * Check if we have a valid request.
220 * Returns
221 * 1 if the request should continue.
222 * In this case we can return an autofs_wait_queue entry if one is
223 * found or NULL to idicate a new wait needs to be created.
224 * 0 or a negative errno if the request shouldn't continue.
225 */
226static int validate_request(struct autofs_wait_queue **wait,
227 struct autofs_sb_info *sbi,
228 struct qstr *qstr,
229 struct dentry*dentry, enum autofs_notify notify)
230{
231 struct autofs_wait_queue *wq;
232 struct autofs_info *ino;
233
234 /* Wait in progress, continue; */
235 wq = autofs4_find_wait(sbi, qstr);
236 if (wq) {
237 *wait = wq;
238 return 1;
239 }
240
241 *wait = NULL;
242
243 /* If we don't yet have any info this is a new request */
244 ino = autofs4_dentry_ino(dentry);
245 if (!ino)
246 return 1;
247
248 /*
249 * If we've been asked to wait on an existing expire (NFY_NONE)
250 * but there is no wait in the queue ...
251 */
252 if (notify == NFY_NONE) {
253 /*
254 * Either we've betean the pending expire to post it's
255 * wait or it finished while we waited on the mutex.
256 * So we need to wait till either, the wait appears
257 * or the expire finishes.
258 */
259
260 while (ino->flags & AUTOFS_INF_EXPIRING) {
261 mutex_unlock(&sbi->wq_mutex);
262 schedule_timeout_interruptible(HZ/10);
263 if (mutex_lock_interruptible(&sbi->wq_mutex))
264 return -EINTR;
265
266 wq = autofs4_find_wait(sbi, qstr);
267 if (wq) {
268 *wait = wq;
269 return 1;
270 }
271 }
272
273 /*
274 * Not ideal but the status has already gone. Of the two
275 * cases where we wait on NFY_NONE neither depend on the
276 * return status of the wait.
277 */
278 return 0;
279 }
280
281 /*
282 * If we've been asked to trigger a mount and the request
283 * completed while we waited on the mutex ...
284 */
285 if (notify == NFY_MOUNT) {
286 /*
287 * If the dentry isn't hashed just go ahead and try the
288 * mount again with a new wait (not much else we can do).
289 */
290 if (!d_unhashed(dentry)) {
291 /*
292 * But if the dentry is hashed, that means that we
293 * got here through the revalidate path. Thus, we
294 * need to check if the dentry has been mounted
295 * while we waited on the wq_mutex. If it has,
296 * simply return success.
297 */
298 if (d_mountpoint(dentry))
299 return 0;
300 }
301 }
302
303 return 1;
304}
305
1da177e4
LT
306int autofs4_wait(struct autofs_sb_info *sbi, struct dentry *dentry,
307 enum autofs_notify notify)
308{
309 struct autofs_wait_queue *wq;
70b52a0a 310 struct qstr qstr;
1da177e4 311 char *name;
a1362fe9 312 int status, ret, type;
1da177e4
LT
313
314 /* In catatonic mode, we don't wait for nobody */
e77fbddf 315 if (sbi->catatonic)
1da177e4 316 return -ENOENT;
a1362fe9 317
1da177e4
LT
318 name = kmalloc(NAME_MAX + 1, GFP_KERNEL);
319 if (!name)
320 return -ENOMEM;
321
5c0a32fc 322 /* If this is a direct mount request create a dummy name */
44d53eb0 323 if (IS_ROOT(dentry) && (sbi->type & AUTOFS_TYPE_DIRECT))
70b52a0a 324 qstr.len = sprintf(name, "%p", dentry);
5c0a32fc 325 else {
70b52a0a
JM
326 qstr.len = autofs4_getpath(sbi, dentry, &name);
327 if (!qstr.len) {
5c0a32fc
IK
328 kfree(name);
329 return -ENOENT;
330 }
1da177e4 331 }
70b52a0a
JM
332 qstr.name = name;
333 qstr.hash = full_name_hash(name, qstr.len);
1da177e4 334
a1362fe9 335 if (mutex_lock_interruptible(&sbi->wq_mutex))
1da177e4 336 return -EINTR;
a5370553 337
a1362fe9
IK
338 ret = validate_request(&wq, sbi, &qstr, dentry, notify);
339 if (ret <= 0) {
340 if (ret == 0)
1d5599e3 341 mutex_unlock(&sbi->wq_mutex);
a1362fe9
IK
342 kfree(qstr.name);
343 return ret;
a5370553 344 }
cc9acc88 345
a5370553 346 if (!wq) {
1da177e4
LT
347 /* Create a new wait queue */
348 wq = kmalloc(sizeof(struct autofs_wait_queue),GFP_KERNEL);
e77fbddf 349 if (!wq) {
70b52a0a 350 kfree(qstr.name);
1d5599e3 351 mutex_unlock(&sbi->wq_mutex);
1da177e4
LT
352 return -ENOMEM;
353 }
354
355 wq->wait_queue_token = autofs4_next_wait_queue;
356 if (++autofs4_next_wait_queue == 0)
357 autofs4_next_wait_queue = 1;
358 wq->next = sbi->queues;
359 sbi->queues = wq;
360 init_waitqueue_head(&wq->queue);
70b52a0a 361 memcpy(&wq->name, &qstr, sizeof(struct qstr));
5c0a32fc
IK
362 wq->dev = autofs4_get_dev(sbi);
363 wq->ino = autofs4_get_ino(sbi);
364 wq->uid = current->uid;
365 wq->gid = current->gid;
366 wq->pid = current->pid;
367 wq->tgid = current->tgid;
1da177e4
LT
368 wq->status = -EINTR; /* Status return if interrupted */
369 atomic_set(&wq->wait_ctr, 2);
1d5599e3 370 mutex_unlock(&sbi->wq_mutex);
3e7b1919 371
5c0a32fc
IK
372 if (sbi->version < 5) {
373 if (notify == NFY_MOUNT)
374 type = autofs_ptype_missing;
375 else
376 type = autofs_ptype_expire_multi;
377 } else {
378 if (notify == NFY_MOUNT)
44d53eb0 379 type = (sbi->type & AUTOFS_TYPE_DIRECT) ?
5c0a32fc
IK
380 autofs_ptype_missing_direct :
381 autofs_ptype_missing_indirect;
382 else
44d53eb0 383 type = (sbi->type & AUTOFS_TYPE_DIRECT) ?
5c0a32fc
IK
384 autofs_ptype_expire_direct :
385 autofs_ptype_expire_indirect;
386 }
4dcd00b1 387
682d4fc9 388 DPRINTK("new wait id = 0x%08lx, name = %.*s, nfy=%d\n",
70b52a0a
JM
389 (unsigned long) wq->wait_queue_token, wq->name.len,
390 wq->name.name, notify);
4dcd00b1
IK
391
392 /* autofs4_notify_daemon() may block */
393 autofs4_notify_daemon(sbi, wq, type);
a5370553
IK
394 } else {
395 atomic_inc(&wq->wait_ctr);
396 mutex_unlock(&sbi->wq_mutex);
70b52a0a 397 kfree(qstr.name);
a5370553 398 DPRINTK("existing wait id = 0x%08lx, name = %.*s, nfy=%d",
70b52a0a
JM
399 (unsigned long) wq->wait_queue_token, wq->name.len,
400 wq->name.name, notify);
4dcd00b1
IK
401 }
402
5a11d4d0
IK
403 /*
404 * wq->name.name is NULL iff the lock is already released
405 * or the mount has been made catatonic.
406 */
70b52a0a 407 if (wq->name.name) {
1da177e4
LT
408 /* Block all but "shutdown" signals while waiting */
409 sigset_t oldset;
410 unsigned long irqflags;
411
412 spin_lock_irqsave(&current->sighand->siglock, irqflags);
413 oldset = current->blocked;
414 siginitsetinv(&current->blocked, SHUTDOWN_SIGS & ~oldset.sig[0]);
415 recalc_sigpending();
416 spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
417
70b52a0a 418 wait_event_interruptible(wq->queue, wq->name.name == NULL);
1da177e4
LT
419
420 spin_lock_irqsave(&current->sighand->siglock, irqflags);
421 current->blocked = oldset;
422 recalc_sigpending();
423 spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
424 } else {
425 DPRINTK("skipped sleeping");
426 }
427
428 status = wq->status;
429
430 /* Are we the last process to need status? */
431 if (atomic_dec_and_test(&wq->wait_ctr))
432 kfree(wq);
433
434 return status;
435}
436
437
438int autofs4_wait_release(struct autofs_sb_info *sbi, autofs_wqt_t wait_queue_token, int status)
439{
440 struct autofs_wait_queue *wq, **wql;
441
1d5599e3 442 mutex_lock(&sbi->wq_mutex);
c80544dc 443 for (wql = &sbi->queues; (wq = *wql) != NULL; wql = &wq->next) {
e77fbddf 444 if (wq->wait_queue_token == wait_queue_token)
1da177e4
LT
445 break;
446 }
447
e77fbddf 448 if (!wq) {
1d5599e3 449 mutex_unlock(&sbi->wq_mutex);
1da177e4
LT
450 return -EINVAL;
451 }
452
453 *wql = wq->next; /* Unlink from chain */
70b52a0a
JM
454 kfree(wq->name.name);
455 wq->name.name = NULL; /* Do not wait on this queue */
a1362fe9 456 mutex_unlock(&sbi->wq_mutex);
1da177e4
LT
457
458 wq->status = status;
459
460 if (atomic_dec_and_test(&wq->wait_ctr)) /* Is anyone still waiting for this guy? */
461 kfree(wq);
462 else
463 wake_up_interruptible(&wq->queue);
464
465 return 0;
466}
467