blob: d2b66bad7d50458a9725d9aa5eabb47b017932b0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/lockd/svclock.c
3 *
4 * Handling of server-side locks, mostly of the blocked variety.
5 * This is the ugliest part of lockd because we tread on very thin ice.
6 * GRANT and CANCEL calls may get stuck, meet in mid-flight, etc.
7 * IMNSHO introducing the grant callback into the NLM protocol was one
8 * of the worst ideas Sun ever had. Except maybe for the idea of doing
9 * NFS file locking at all.
10 *
11 * I'm trying hard to avoid race conditions by protecting most accesses
12 * to a file's list of blocked locks through a semaphore. The global
13 * list of blocked locks is not protected in this fashion however.
14 * Therefore, some functions (such as the RPC callback for the async grant
15 * call) move blocked locks towards the head of the list *while some other
16 * process might be traversing it*. This should not be a problem in
17 * practice, because this will only cause functions traversing the list
18 * to visit some blocks twice.
19 *
20 * Copyright (C) 1996, Olaf Kirch <okir@monad.swb.de>
21 */
22
23#include <linux/config.h>
24#include <linux/types.h>
25#include <linux/errno.h>
26#include <linux/kernel.h>
27#include <linux/sched.h>
28#include <linux/smp_lock.h>
29#include <linux/sunrpc/clnt.h>
30#include <linux/sunrpc/svc.h>
31#include <linux/lockd/nlm.h>
32#include <linux/lockd/lockd.h>
33
34#define NLMDBG_FACILITY NLMDBG_SVCLOCK
35
36#ifdef CONFIG_LOCKD_V4
37#define nlm_deadlock nlm4_deadlock
38#else
39#define nlm_deadlock nlm_lck_denied
40#endif
41
Trond Myklebust6849c0c2006-03-20 13:44:39 -050042static void nlmsvc_release_block(struct nlm_block *block);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static void nlmsvc_insert_block(struct nlm_block *block, unsigned long);
44static int nlmsvc_remove_block(struct nlm_block *block);
Trond Myklebust963d8fe2006-01-03 09:55:04 +010045
Trond Myklebust5e1abf82006-03-20 13:44:39 -050046static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock);
47static void nlmsvc_freegrantargs(struct nlm_rqst *call);
Trond Myklebust963d8fe2006-01-03 09:55:04 +010048static const struct rpc_call_ops nlmsvc_grant_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
51 * The list of blocked locks to retry
52 */
53static struct nlm_block * nlm_blocked;
54
55/*
56 * Insert a blocked lock into the global list
57 */
58static void
59nlmsvc_insert_block(struct nlm_block *block, unsigned long when)
60{
61 struct nlm_block **bp, *b;
62
63 dprintk("lockd: nlmsvc_insert_block(%p, %ld)\n", block, when);
Trond Myklebust6849c0c2006-03-20 13:44:39 -050064 kref_get(&block->b_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 if (block->b_queued)
66 nlmsvc_remove_block(block);
67 bp = &nlm_blocked;
68 if (when != NLM_NEVER) {
69 if ((when += jiffies) == NLM_NEVER)
70 when ++;
71 while ((b = *bp) && time_before_eq(b->b_when,when) && b->b_when != NLM_NEVER)
72 bp = &b->b_next;
73 } else
74 while ((b = *bp) != 0)
75 bp = &b->b_next;
76
77 block->b_queued = 1;
78 block->b_when = when;
79 block->b_next = b;
80 *bp = block;
81}
82
83/*
84 * Remove a block from the global list
85 */
86static int
87nlmsvc_remove_block(struct nlm_block *block)
88{
89 struct nlm_block **bp, *b;
90
91 if (!block->b_queued)
92 return 1;
93 for (bp = &nlm_blocked; (b = *bp) != 0; bp = &b->b_next) {
94 if (b == block) {
95 *bp = block->b_next;
96 block->b_queued = 0;
Trond Myklebust6849c0c2006-03-20 13:44:39 -050097 nlmsvc_release_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 return 1;
99 }
100 }
101
102 return 0;
103}
104
105/*
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500106 * Find a block for a given lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 */
108static struct nlm_block *
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500109nlmsvc_lookup_block(struct nlm_file *file, struct nlm_lock *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110{
111 struct nlm_block **head, *block;
112 struct file_lock *fl;
113
114 dprintk("lockd: nlmsvc_lookup_block f=%p pd=%d %Ld-%Ld ty=%d\n",
115 file, lock->fl.fl_pid,
116 (long long)lock->fl.fl_start,
117 (long long)lock->fl.fl_end, lock->fl.fl_type);
118 for (head = &nlm_blocked; (block = *head) != 0; head = &block->b_next) {
Trond Myklebust92737232006-03-20 13:44:45 -0500119 fl = &block->b_call->a_args.lock.fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 dprintk("lockd: check f=%p pd=%d %Ld-%Ld ty=%d cookie=%s\n",
121 block->b_file, fl->fl_pid,
122 (long long)fl->fl_start,
123 (long long)fl->fl_end, fl->fl_type,
Trond Myklebust92737232006-03-20 13:44:45 -0500124 nlmdbg_cookie2a(&block->b_call->a_args.cookie));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (block->b_file == file && nlm_compare_locks(fl, &lock->fl)) {
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500126 kref_get(&block->b_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 return block;
128 }
129 }
130
131 return NULL;
132}
133
134static inline int nlm_cookie_match(struct nlm_cookie *a, struct nlm_cookie *b)
135{
136 if(a->len != b->len)
137 return 0;
138 if(memcmp(a->data,b->data,a->len))
139 return 0;
140 return 1;
141}
142
143/*
144 * Find a block with a given NLM cookie.
145 */
146static inline struct nlm_block *
147nlmsvc_find_block(struct nlm_cookie *cookie, struct sockaddr_in *sin)
148{
149 struct nlm_block *block;
150
151 for (block = nlm_blocked; block; block = block->b_next) {
152 dprintk("cookie: head of blocked queue %p, block %p\n",
153 nlm_blocked, block);
Trond Myklebust92737232006-03-20 13:44:45 -0500154 if (nlm_cookie_match(&block->b_call->a_args.cookie,cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 && nlm_cmp_addr(sin, &block->b_host->h_addr))
156 break;
157 }
158
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500159 if (block != NULL)
160 kref_get(&block->b_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 return block;
162}
163
164/*
165 * Create a block and initialize it.
166 *
167 * Note: we explicitly set the cookie of the grant reply to that of
168 * the blocked lock request. The spec explicitly mentions that the client
169 * should _not_ rely on the callback containing the same cookie as the
170 * request, but (as I found out later) that's because some implementations
171 * do just this. Never mind the standards comittees, they support our
172 * logging industries.
173 */
174static inline struct nlm_block *
175nlmsvc_create_block(struct svc_rqst *rqstp, struct nlm_file *file,
176 struct nlm_lock *lock, struct nlm_cookie *cookie)
177{
178 struct nlm_block *block;
179 struct nlm_host *host;
Trond Myklebust92737232006-03-20 13:44:45 -0500180 struct nlm_rqst *call = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
182 /* Create host handle for callback */
Trond Myklebust686517f2006-03-20 13:44:39 -0500183 host = nlmsvc_lookup_host(rqstp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 if (host == NULL)
185 return NULL;
186
Trond Myklebust92737232006-03-20 13:44:45 -0500187 call = nlm_alloc_call(host);
188 if (call == NULL)
189 return NULL;
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 /* Allocate memory for block, and initialize arguments */
Trond Myklebust92737232006-03-20 13:44:45 -0500192 block = kzalloc(sizeof(*block), GFP_KERNEL);
193 if (block == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 goto failed;
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500195 kref_init(&block->b_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Trond Myklebust92737232006-03-20 13:44:45 -0500197 if (!nlmsvc_setgrantargs(call, lock))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 goto failed_free;
199
200 /* Set notifier function for VFS, and init args */
Trond Myklebust92737232006-03-20 13:44:45 -0500201 call->a_args.lock.fl.fl_flags |= FL_SLEEP;
202 call->a_args.lock.fl.fl_lmops = &nlmsvc_lock_operations;
203 call->a_args.cookie = *cookie; /* see above */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205 dprintk("lockd: created block %p...\n", block);
206
207 /* Create and initialize the block */
208 block->b_daemon = rqstp->rq_server;
209 block->b_host = host;
210 block->b_file = file;
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500211 file->f_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213 /* Add to file's list of blocks */
214 block->b_fnext = file->f_blocks;
215 file->f_blocks = block;
216
217 /* Set up RPC arguments for callback */
Trond Myklebust92737232006-03-20 13:44:45 -0500218 block->b_call = call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 call->a_flags = RPC_TASK_ASYNC;
Trond Myklebust92737232006-03-20 13:44:45 -0500220 call->a_block = block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
222 return block;
223
224failed_free:
225 kfree(block);
226failed:
Trond Myklebust92737232006-03-20 13:44:45 -0500227 nlm_release_call(call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return NULL;
229}
230
231/*
232 * Delete a block. If the lock was cancelled or the grant callback
233 * failed, unlock is set to 1.
234 * It is the caller's responsibility to check whether the file
235 * can be closed hereafter.
236 */
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500237static int nlmsvc_unlink_block(struct nlm_block *block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Trond Myklebust09c79382006-03-20 13:44:38 -0500239 int status;
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500240 dprintk("lockd: unlinking block %p...\n", block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242 /* Remove block from list */
Trond Myklebust92737232006-03-20 13:44:45 -0500243 status = posix_unblock_lock(block->b_file->f_file, &block->b_call->a_args.lock.fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 nlmsvc_remove_block(block);
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500245 return status;
246}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500248static void nlmsvc_free_block(struct kref *kref)
249{
250 struct nlm_block *block = container_of(kref, struct nlm_block, b_count);
251 struct nlm_file *file = block->b_file;
252 struct nlm_block **bp;
253
254 dprintk("lockd: freeing block %p...\n", block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500256 down(&file->f_sema);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 /* Remove block from file's list of blocks */
258 for (bp = &file->f_blocks; *bp; bp = &(*bp)->b_fnext) {
259 if (*bp == block) {
260 *bp = block->b_fnext;
261 break;
262 }
263 }
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500264 up(&file->f_sema);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Trond Myklebust92737232006-03-20 13:44:45 -0500266 nlmsvc_freegrantargs(block->b_call);
267 nlm_release_call(block->b_call);
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500268 nlm_release_file(block->b_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 kfree(block);
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500270}
271
272static void nlmsvc_release_block(struct nlm_block *block)
273{
274 if (block != NULL)
275 kref_put(&block->b_count, nlmsvc_free_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500278static void nlmsvc_act_mark(struct nlm_host *host, struct nlm_file *file)
279{
280 struct nlm_block *block;
281
282 down(&file->f_sema);
283 for (block = file->f_blocks; block != NULL; block = block->b_fnext)
284 block->b_host->h_inuse = 1;
285 up(&file->f_sema);
286}
287
288static void nlmsvc_act_unlock(struct nlm_host *host, struct nlm_file *file)
289{
290 struct nlm_block *block;
291
292restart:
293 down(&file->f_sema);
294 for (block = file->f_blocks; block != NULL; block = block->b_fnext) {
295 if (host != NULL && host != block->b_host)
296 continue;
297 if (!block->b_queued)
298 continue;
299 kref_get(&block->b_count);
300 up(&file->f_sema);
301 nlmsvc_unlink_block(block);
302 nlmsvc_release_block(block);
303 goto restart;
304 }
305 up(&file->f_sema);
306}
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/*
309 * Loop over all blocks and perform the action specified.
310 * (NLM_ACT_CHECK handled by nlmsvc_inspect_file).
311 */
J. Bruce Fieldsf3ee4392006-03-20 23:24:13 -0500312void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313nlmsvc_traverse_blocks(struct nlm_host *host, struct nlm_file *file, int action)
314{
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500315 if (action == NLM_ACT_MARK)
316 nlmsvc_act_mark(host, file);
317 else
318 nlmsvc_act_unlock(host, file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319}
320
321/*
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500322 * Initialize arguments for GRANTED call. The nlm_rqst structure
323 * has been cleared already.
324 */
325static int nlmsvc_setgrantargs(struct nlm_rqst *call, struct nlm_lock *lock)
326{
327 locks_copy_lock(&call->a_args.lock.fl, &lock->fl);
328 memcpy(&call->a_args.lock.fh, &lock->fh, sizeof(call->a_args.lock.fh));
329 call->a_args.lock.caller = system_utsname.nodename;
330 call->a_args.lock.oh.len = lock->oh.len;
331
332 /* set default data area */
333 call->a_args.lock.oh.data = call->a_owner;
334 call->a_args.lock.svid = lock->fl.fl_pid;
335
336 if (lock->oh.len > NLMCLNT_OHSIZE) {
337 void *data = kmalloc(lock->oh.len, GFP_KERNEL);
Trond Myklebust92737232006-03-20 13:44:45 -0500338 if (!data)
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500339 return 0;
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500340 call->a_args.lock.oh.data = (u8 *) data;
341 }
342
343 memcpy(call->a_args.lock.oh.data, lock->oh.data, lock->oh.len);
344 return 1;
345}
346
347static void nlmsvc_freegrantargs(struct nlm_rqst *call)
348{
Trond Myklebust92737232006-03-20 13:44:45 -0500349 if (call->a_args.lock.oh.data != call->a_owner)
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500350 kfree(call->a_args.lock.oh.data);
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500351}
352
353/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * Attempt to establish a lock, and if it can't be granted, block it
355 * if required.
356 */
357u32
358nlmsvc_lock(struct svc_rqst *rqstp, struct nlm_file *file,
359 struct nlm_lock *lock, int wait, struct nlm_cookie *cookie)
360{
Trond Myklebust09c79382006-03-20 13:44:38 -0500361 struct nlm_block *block, *newblock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 int error;
Andy Adamson15dadef2006-03-20 13:44:24 -0500363 u32 ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 dprintk("lockd: nlmsvc_lock(%s/%ld, ty=%d, pi=%d, %Ld-%Ld, bl=%d)\n",
366 file->f_file->f_dentry->d_inode->i_sb->s_id,
367 file->f_file->f_dentry->d_inode->i_ino,
368 lock->fl.fl_type, lock->fl.fl_pid,
369 (long long)lock->fl.fl_start,
370 (long long)lock->fl.fl_end,
371 wait);
372
373
Trond Myklebust09c79382006-03-20 13:44:38 -0500374 lock->fl.fl_flags &= ~FL_SLEEP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375again:
376 /* Lock file against concurrent access */
377 down(&file->f_sema);
Trond Myklebust09c79382006-03-20 13:44:38 -0500378 /* Get existing block (in case client is busy-waiting) */
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500379 block = nlmsvc_lookup_block(file, lock);
Trond Myklebust09c79382006-03-20 13:44:38 -0500380 if (block == NULL) {
381 if (newblock != NULL)
Trond Myklebust92737232006-03-20 13:44:45 -0500382 lock = &newblock->b_call->a_args.lock;
Trond Myklebust09c79382006-03-20 13:44:38 -0500383 } else
Trond Myklebust92737232006-03-20 13:44:45 -0500384 lock = &block->b_call->a_args.lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Andy Adamsona85f1932006-03-20 13:44:25 -0500386 error = posix_lock_file(file->f_file, &lock->fl);
Trond Myklebust09c79382006-03-20 13:44:38 -0500387 lock->fl.fl_flags &= ~FL_SLEEP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Andy Adamsona85f1932006-03-20 13:44:25 -0500389 dprintk("lockd: posix_lock_file returned %d\n", error);
390
Trond Myklebust09c79382006-03-20 13:44:38 -0500391 switch(error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 case 0:
Andy Adamson15dadef2006-03-20 13:44:24 -0500393 ret = nlm_granted;
394 goto out;
Trond Myklebust09c79382006-03-20 13:44:38 -0500395 case -EAGAIN:
396 break;
397 case -EDEADLK:
Andy Adamson15dadef2006-03-20 13:44:24 -0500398 ret = nlm_deadlock;
399 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 default: /* includes ENOLCK */
Andy Adamson15dadef2006-03-20 13:44:24 -0500401 ret = nlm_lck_denied_nolocks;
402 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 }
404
Trond Myklebust09c79382006-03-20 13:44:38 -0500405 ret = nlm_lck_denied;
406 if (!wait)
407 goto out;
408
409 ret = nlm_lck_blocked;
410 if (block != NULL)
411 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 /* If we don't have a block, create and initialize it. Then
414 * retry because we may have slept in kmalloc. */
415 /* We have to release f_sema as nlmsvc_create_block may try to
416 * to claim it while doing host garbage collection */
Trond Myklebust09c79382006-03-20 13:44:38 -0500417 if (newblock == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 up(&file->f_sema);
419 dprintk("lockd: blocking on this lock (allocating).\n");
Trond Myklebust09c79382006-03-20 13:44:38 -0500420 if (!(newblock = nlmsvc_create_block(rqstp, file, lock, cookie)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 return nlm_lck_denied_nolocks;
422 goto again;
423 }
424
425 /* Append to list of blocked */
Trond Myklebust09c79382006-03-20 13:44:38 -0500426 nlmsvc_insert_block(newblock, NLM_NEVER);
Andy Adamson15dadef2006-03-20 13:44:24 -0500427out:
Trond Myklebust09c79382006-03-20 13:44:38 -0500428 up(&file->f_sema);
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500429 nlmsvc_release_block(newblock);
430 nlmsvc_release_block(block);
Andy Adamson15dadef2006-03-20 13:44:24 -0500431 dprintk("lockd: nlmsvc_lock returned %u\n", ret);
432 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
435/*
436 * Test for presence of a conflicting lock.
437 */
438u32
439nlmsvc_testlock(struct nlm_file *file, struct nlm_lock *lock,
440 struct nlm_lock *conflock)
441{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 dprintk("lockd: nlmsvc_testlock(%s/%ld, ty=%d, %Ld-%Ld)\n",
443 file->f_file->f_dentry->d_inode->i_sb->s_id,
444 file->f_file->f_dentry->d_inode->i_ino,
445 lock->fl.fl_type,
446 (long long)lock->fl.fl_start,
447 (long long)lock->fl.fl_end);
448
Andy Adamson8dc7c312006-03-20 13:44:26 -0500449 if (posix_test_lock(file->f_file, &lock->fl, &conflock->fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 dprintk("lockd: conflicting lock(ty=%d, %Ld-%Ld)\n",
Andy Adamson8dc7c312006-03-20 13:44:26 -0500451 conflock->fl.fl_type,
452 (long long)conflock->fl.fl_start,
453 (long long)conflock->fl.fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 conflock->caller = "somehost"; /* FIXME */
455 conflock->oh.len = 0; /* don't return OH info */
Andy Adamson8dc7c312006-03-20 13:44:26 -0500456 conflock->svid = conflock->fl.fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 return nlm_lck_denied;
458 }
459
460 return nlm_granted;
461}
462
463/*
464 * Remove a lock.
465 * This implies a CANCEL call: We send a GRANT_MSG, the client replies
466 * with a GRANT_RES call which gets lost, and calls UNLOCK immediately
467 * afterwards. In this case the block will still be there, and hence
468 * must be removed.
469 */
470u32
471nlmsvc_unlock(struct nlm_file *file, struct nlm_lock *lock)
472{
473 int error;
474
475 dprintk("lockd: nlmsvc_unlock(%s/%ld, pi=%d, %Ld-%Ld)\n",
476 file->f_file->f_dentry->d_inode->i_sb->s_id,
477 file->f_file->f_dentry->d_inode->i_ino,
478 lock->fl.fl_pid,
479 (long long)lock->fl.fl_start,
480 (long long)lock->fl.fl_end);
481
482 /* First, cancel any lock that might be there */
483 nlmsvc_cancel_blocked(file, lock);
484
485 lock->fl.fl_type = F_UNLCK;
486 error = posix_lock_file(file->f_file, &lock->fl);
487
488 return (error < 0)? nlm_lck_denied_nolocks : nlm_granted;
489}
490
491/*
492 * Cancel a previously blocked request.
493 *
494 * A cancel request always overrides any grant that may currently
495 * be in progress.
496 * The calling procedure must check whether the file can be closed.
497 */
498u32
499nlmsvc_cancel_blocked(struct nlm_file *file, struct nlm_lock *lock)
500{
501 struct nlm_block *block;
J. Bruce Fields64a318e2006-01-03 09:55:46 +0100502 int status = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 dprintk("lockd: nlmsvc_cancel(%s/%ld, pi=%d, %Ld-%Ld)\n",
505 file->f_file->f_dentry->d_inode->i_sb->s_id,
506 file->f_file->f_dentry->d_inode->i_ino,
507 lock->fl.fl_pid,
508 (long long)lock->fl.fl_start,
509 (long long)lock->fl.fl_end);
510
511 down(&file->f_sema);
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500512 block = nlmsvc_lookup_block(file, lock);
513 up(&file->f_sema);
514 if (block != NULL) {
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500515 status = nlmsvc_unlink_block(block);
516 nlmsvc_release_block(block);
517 }
J. Bruce Fields64a318e2006-01-03 09:55:46 +0100518 return status ? nlm_lck_denied : nlm_granted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519}
520
521/*
522 * Unblock a blocked lock request. This is a callback invoked from the
523 * VFS layer when a lock on which we blocked is removed.
524 *
525 * This function doesn't grant the blocked lock instantly, but rather moves
526 * the block to the head of nlm_blocked where it can be picked up by lockd.
527 */
528static void
529nlmsvc_notify_blocked(struct file_lock *fl)
530{
531 struct nlm_block **bp, *block;
532
533 dprintk("lockd: VFS unblock notification for block %p\n", fl);
534 for (bp = &nlm_blocked; (block = *bp) != 0; bp = &block->b_next) {
Trond Myklebust92737232006-03-20 13:44:45 -0500535 if (nlm_compare_locks(&block->b_call->a_args.lock.fl, fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 nlmsvc_insert_block(block, 0);
537 svc_wake_up(block->b_daemon);
538 return;
539 }
540 }
541
542 printk(KERN_WARNING "lockd: notification for unknown block!\n");
543}
544
545static int nlmsvc_same_owner(struct file_lock *fl1, struct file_lock *fl2)
546{
547 return fl1->fl_owner == fl2->fl_owner && fl1->fl_pid == fl2->fl_pid;
548}
549
550struct lock_manager_operations nlmsvc_lock_operations = {
551 .fl_compare_owner = nlmsvc_same_owner,
552 .fl_notify = nlmsvc_notify_blocked,
553};
554
555/*
556 * Try to claim a lock that was previously blocked.
557 *
558 * Note that we use both the RPC_GRANTED_MSG call _and_ an async
559 * RPC thread when notifying the client. This seems like overkill...
560 * Here's why:
561 * - we don't want to use a synchronous RPC thread, otherwise
562 * we might find ourselves hanging on a dead portmapper.
563 * - Some lockd implementations (e.g. HP) don't react to
564 * RPC_GRANTED calls; they seem to insist on RPC_GRANTED_MSG calls.
565 */
566static void
567nlmsvc_grant_blocked(struct nlm_block *block)
568{
569 struct nlm_file *file = block->b_file;
Trond Myklebust92737232006-03-20 13:44:45 -0500570 struct nlm_lock *lock = &block->b_call->a_args.lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 int error;
572
573 dprintk("lockd: grant blocked lock %p\n", block);
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 /* Unlink block request from list */
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500576 nlmsvc_unlink_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 /* If b_granted is true this means we've been here before.
579 * Just retry the grant callback, possibly refreshing the RPC
580 * binding */
581 if (block->b_granted) {
582 nlm_rebind_host(block->b_host);
583 goto callback;
584 }
585
586 /* Try the lock operation again */
Trond Myklebust09c79382006-03-20 13:44:38 -0500587 lock->fl.fl_flags |= FL_SLEEP;
Andy Adamson5de0e502006-03-20 13:44:25 -0500588 error = posix_lock_file(file->f_file, &lock->fl);
Trond Myklebust09c79382006-03-20 13:44:38 -0500589 lock->fl.fl_flags &= ~FL_SLEEP;
590
Andy Adamson5de0e502006-03-20 13:44:25 -0500591 switch (error) {
592 case 0:
593 break;
594 case -EAGAIN:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 dprintk("lockd: lock still blocked\n");
596 nlmsvc_insert_block(block, NLM_NEVER);
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500597 return;
Andy Adamson5de0e502006-03-20 13:44:25 -0500598 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 printk(KERN_WARNING "lockd: unexpected error %d in %s!\n",
600 -error, __FUNCTION__);
601 nlmsvc_insert_block(block, 10 * HZ);
Trond Myklebustd9f6eb72006-03-20 13:44:47 -0500602 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 }
604
605callback:
606 /* Lock was granted by VFS. */
607 dprintk("lockd: GRANTing blocked lock.\n");
608 block->b_granted = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 /* Schedule next grant callback in 30 seconds */
611 nlmsvc_insert_block(block, 30 * HZ);
612
613 /* Call the client */
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500614 kref_get(&block->b_count);
Trond Myklebust92737232006-03-20 13:44:45 -0500615 if (nlm_async_call(block->b_call, NLMPROC_GRANTED_MSG,
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100616 &nlmsvc_grant_ops) < 0)
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500617 nlmsvc_release_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
619
620/*
621 * This is the callback from the RPC layer when the NLM_GRANTED_MSG
622 * RPC call has succeeded or timed out.
623 * Like all RPC callbacks, it is invoked by the rpciod process, so it
624 * better not sleep. Therefore, we put the blocked lock on the nlm_blocked
625 * chain once more in order to have it removed by lockd itself (which can
626 * then sleep on the file semaphore without disrupting e.g. the nfs client).
627 */
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100628static void nlmsvc_grant_callback(struct rpc_task *task, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100630 struct nlm_rqst *call = data;
Trond Myklebust92737232006-03-20 13:44:45 -0500631 struct nlm_block *block = call->a_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 unsigned long timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 dprintk("lockd: GRANT_MSG RPC callback\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
636 /* Technically, we should down the file semaphore here. Since we
637 * move the block towards the head of the queue only, no harm
638 * can be done, though. */
639 if (task->tk_status < 0) {
640 /* RPC error: Re-insert for retransmission */
641 timeout = 10 * HZ;
642 } else if (block->b_done) {
643 /* Block already removed, kill it for real */
644 timeout = 0;
645 } else {
646 /* Call was successful, now wait for client callback */
647 timeout = 60 * HZ;
648 }
649 nlmsvc_insert_block(block, timeout);
650 svc_wake_up(block->b_daemon);
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500651}
652
653void nlmsvc_grant_release(void *data)
654{
Trond Myklebust6041b792006-03-20 13:44:45 -0500655 struct nlm_rqst *call = data;
656
657 nlmsvc_release_block(call->a_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100660static const struct rpc_call_ops nlmsvc_grant_ops = {
661 .rpc_call_done = nlmsvc_grant_callback,
Trond Myklebust5e1abf82006-03-20 13:44:39 -0500662 .rpc_release = nlmsvc_grant_release,
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100663};
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665/*
666 * We received a GRANT_RES callback. Try to find the corresponding
667 * block.
668 */
669void
670nlmsvc_grant_reply(struct svc_rqst *rqstp, struct nlm_cookie *cookie, u32 status)
671{
672 struct nlm_block *block;
673 struct nlm_file *file;
674
675 dprintk("grant_reply: looking for cookie %x, host (%08x), s=%d \n",
676 *(unsigned int *)(cookie->data),
677 ntohl(rqstp->rq_addr.sin_addr.s_addr), status);
678 if (!(block = nlmsvc_find_block(cookie, &rqstp->rq_addr)))
679 return;
680 file = block->b_file;
681
J. Bruce Fieldsf2321422006-01-03 09:55:42 +0100682 if (block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (status == NLM_LCK_DENIED_GRACE_PERIOD) {
684 /* Try again in a couple of seconds */
685 nlmsvc_insert_block(block, 10 * HZ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 } else {
687 /* Lock is now held by client, or has been rejected.
688 * In both cases, the block should be removed. */
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500689 nlmsvc_unlink_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 }
691 }
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500692 nlmsvc_release_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
695/*
696 * Retry all blocked locks that have been notified. This is where lockd
697 * picks up locks that can be granted, or grant notifications that must
698 * be retransmitted.
699 */
700unsigned long
701nlmsvc_retry_blocked(void)
702{
703 struct nlm_block *block;
704
705 dprintk("nlmsvc_retry_blocked(%p, when=%ld)\n",
706 nlm_blocked,
707 nlm_blocked? nlm_blocked->b_when : 0);
708 while ((block = nlm_blocked) != 0) {
709 if (block->b_when == NLM_NEVER)
710 break;
711 if (time_after(block->b_when,jiffies))
712 break;
713 dprintk("nlmsvc_retry_blocked(%p, when=%ld, done=%d)\n",
714 block, block->b_when, block->b_done);
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500715 kref_get(&block->b_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (block->b_done)
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500717 nlmsvc_unlink_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 else
719 nlmsvc_grant_blocked(block);
Trond Myklebust6849c0c2006-03-20 13:44:39 -0500720 nlmsvc_release_block(block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
722
723 if ((block = nlm_blocked) && block->b_when != NLM_NEVER)
724 return (block->b_when - jiffies);
725
726 return MAX_SCHEDULE_TIMEOUT;
727}