xref: /openbmc/linux/fs/nfs/delegation.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
1 /*
2  * linux/fs/nfs/delegation.c
3  *
4  * Copyright (C) 2004 Trond Myklebust
5  *
6  * NFS file delegation management
7  *
8  */
9 #include <linux/completion.h>
10 #include <linux/kthread.h>
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/spinlock.h>
14 
15 #include <linux/nfs4.h>
16 #include <linux/nfs_fs.h>
17 #include <linux/nfs_xdr.h>
18 
19 #include "nfs4_fs.h"
20 #include "delegation.h"
21 #include "internal.h"
22 
23 static void nfs_do_free_delegation(struct nfs_delegation *delegation)
24 {
25 	kfree(delegation);
26 }
27 
28 static void nfs_free_delegation_callback(struct rcu_head *head)
29 {
30 	struct nfs_delegation *delegation = container_of(head, struct nfs_delegation, rcu);
31 
32 	nfs_do_free_delegation(delegation);
33 }
34 
35 static void nfs_free_delegation(struct nfs_delegation *delegation)
36 {
37 	struct rpc_cred *cred;
38 
39 	cred = rcu_dereference(delegation->cred);
40 	rcu_assign_pointer(delegation->cred, NULL);
41 	call_rcu(&delegation->rcu, nfs_free_delegation_callback);
42 	if (cred)
43 		put_rpccred(cred);
44 }
45 
46 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
47 {
48 	struct inode *inode = state->inode;
49 	struct file_lock *fl;
50 	int status;
51 
52 	for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
53 		if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
54 			continue;
55 		if (nfs_file_open_context(fl->fl_file) != ctx)
56 			continue;
57 		status = nfs4_lock_delegation_recall(state, fl);
58 		if (status >= 0)
59 			continue;
60 		switch (status) {
61 			default:
62 				printk(KERN_ERR "%s: unhandled error %d.\n",
63 						__FUNCTION__, status);
64 			case -NFS4ERR_EXPIRED:
65 				/* kill_proc(fl->fl_pid, SIGLOST, 1); */
66 			case -NFS4ERR_STALE_CLIENTID:
67 				nfs4_schedule_state_recovery(NFS_SERVER(inode)->nfs_client);
68 				goto out_err;
69 		}
70 	}
71 	return 0;
72 out_err:
73 	return status;
74 }
75 
76 static void nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
77 {
78 	struct nfs_inode *nfsi = NFS_I(inode);
79 	struct nfs_open_context *ctx;
80 	struct nfs4_state *state;
81 	int err;
82 
83 again:
84 	spin_lock(&inode->i_lock);
85 	list_for_each_entry(ctx, &nfsi->open_files, list) {
86 		state = ctx->state;
87 		if (state == NULL)
88 			continue;
89 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
90 			continue;
91 		if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
92 			continue;
93 		get_nfs_open_context(ctx);
94 		spin_unlock(&inode->i_lock);
95 		err = nfs4_open_delegation_recall(ctx, state, stateid);
96 		if (err >= 0)
97 			err = nfs_delegation_claim_locks(ctx, state);
98 		put_nfs_open_context(ctx);
99 		if (err != 0)
100 			return;
101 		goto again;
102 	}
103 	spin_unlock(&inode->i_lock);
104 }
105 
106 /*
107  * Set up a delegation on an inode
108  */
109 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
110 {
111 	struct nfs_delegation *delegation = NFS_I(inode)->delegation;
112 	struct rpc_cred *oldcred;
113 
114 	if (delegation == NULL)
115 		return;
116 	memcpy(delegation->stateid.data, res->delegation.data,
117 			sizeof(delegation->stateid.data));
118 	delegation->type = res->delegation_type;
119 	delegation->maxsize = res->maxsize;
120 	oldcred = delegation->cred;
121 	delegation->cred = get_rpccred(cred);
122 	delegation->flags &= ~NFS_DELEGATION_NEED_RECLAIM;
123 	NFS_I(inode)->delegation_state = delegation->type;
124 	smp_wmb();
125 	put_rpccred(oldcred);
126 }
127 
128 /*
129  * Set up a delegation on an inode
130  */
131 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
132 {
133 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
134 	struct nfs_inode *nfsi = NFS_I(inode);
135 	struct nfs_delegation *delegation;
136 	int status = 0;
137 
138 	delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
139 	if (delegation == NULL)
140 		return -ENOMEM;
141 	memcpy(delegation->stateid.data, res->delegation.data,
142 			sizeof(delegation->stateid.data));
143 	delegation->type = res->delegation_type;
144 	delegation->maxsize = res->maxsize;
145 	delegation->change_attr = nfsi->change_attr;
146 	delegation->cred = get_rpccred(cred);
147 	delegation->inode = inode;
148 
149 	spin_lock(&clp->cl_lock);
150 	if (rcu_dereference(nfsi->delegation) == NULL) {
151 		list_add_rcu(&delegation->super_list, &clp->cl_delegations);
152 		nfsi->delegation_state = delegation->type;
153 		rcu_assign_pointer(nfsi->delegation, delegation);
154 		delegation = NULL;
155 	} else {
156 		if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
157 					sizeof(delegation->stateid)) != 0 ||
158 				delegation->type != nfsi->delegation->type) {
159 			printk("%s: server %u.%u.%u.%u, handed out a duplicate delegation!\n",
160 					__FUNCTION__, NIPQUAD(clp->cl_addr.sin_addr));
161 			status = -EIO;
162 		}
163 	}
164 
165 	/* Ensure we revalidate the attributes and page cache! */
166 	spin_lock(&inode->i_lock);
167 	nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
168 	spin_unlock(&inode->i_lock);
169 
170 	spin_unlock(&clp->cl_lock);
171 	kfree(delegation);
172 	return status;
173 }
174 
175 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
176 {
177 	int res = 0;
178 
179 	res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid);
180 	nfs_free_delegation(delegation);
181 	return res;
182 }
183 
184 /* Sync all data to disk upon delegation return */
185 static void nfs_msync_inode(struct inode *inode)
186 {
187 	filemap_fdatawrite(inode->i_mapping);
188 	nfs_wb_all(inode);
189 	filemap_fdatawait(inode->i_mapping);
190 }
191 
192 /*
193  * Basic procedure for returning a delegation to the server
194  */
195 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
196 {
197 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
198 	struct nfs_inode *nfsi = NFS_I(inode);
199 
200 	nfs_msync_inode(inode);
201 	down_read(&clp->cl_sem);
202 	/* Guard against new delegated open calls */
203 	down_write(&nfsi->rwsem);
204 	nfs_delegation_claim_opens(inode, &delegation->stateid);
205 	up_write(&nfsi->rwsem);
206 	up_read(&clp->cl_sem);
207 	nfs_msync_inode(inode);
208 
209 	return nfs_do_return_delegation(inode, delegation);
210 }
211 
212 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
213 {
214 	struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
215 
216 	if (delegation == NULL)
217 		goto nomatch;
218 	if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
219 				sizeof(delegation->stateid.data)) != 0)
220 		goto nomatch;
221 	list_del_rcu(&delegation->super_list);
222 	nfsi->delegation_state = 0;
223 	rcu_assign_pointer(nfsi->delegation, NULL);
224 	return delegation;
225 nomatch:
226 	return NULL;
227 }
228 
229 int nfs_inode_return_delegation(struct inode *inode)
230 {
231 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
232 	struct nfs_inode *nfsi = NFS_I(inode);
233 	struct nfs_delegation *delegation;
234 	int err = 0;
235 
236 	if (rcu_dereference(nfsi->delegation) != NULL) {
237 		spin_lock(&clp->cl_lock);
238 		delegation = nfs_detach_delegation_locked(nfsi, NULL);
239 		spin_unlock(&clp->cl_lock);
240 		if (delegation != NULL)
241 			err = __nfs_inode_return_delegation(inode, delegation);
242 	}
243 	return err;
244 }
245 
246 /*
247  * Return all delegations associated to a super block
248  */
249 void nfs_return_all_delegations(struct super_block *sb)
250 {
251 	struct nfs_client *clp = NFS_SB(sb)->nfs_client;
252 	struct nfs_delegation *delegation;
253 	struct inode *inode;
254 
255 	if (clp == NULL)
256 		return;
257 restart:
258 	rcu_read_lock();
259 	list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
260 		if (delegation->inode->i_sb != sb)
261 			continue;
262 		inode = igrab(delegation->inode);
263 		if (inode == NULL)
264 			continue;
265 		spin_lock(&clp->cl_lock);
266 		delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
267 		spin_unlock(&clp->cl_lock);
268 		rcu_read_unlock();
269 		if (delegation != NULL)
270 			__nfs_inode_return_delegation(inode, delegation);
271 		iput(inode);
272 		goto restart;
273 	}
274 	rcu_read_unlock();
275 }
276 
277 static int nfs_do_expire_all_delegations(void *ptr)
278 {
279 	struct nfs_client *clp = ptr;
280 	struct nfs_delegation *delegation;
281 	struct inode *inode;
282 
283 	allow_signal(SIGKILL);
284 restart:
285 	if (test_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) != 0)
286 		goto out;
287 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0)
288 		goto out;
289 	rcu_read_lock();
290 	list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
291 		inode = igrab(delegation->inode);
292 		if (inode == NULL)
293 			continue;
294 		spin_lock(&clp->cl_lock);
295 		delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
296 		spin_unlock(&clp->cl_lock);
297 		rcu_read_unlock();
298 		if (delegation)
299 			__nfs_inode_return_delegation(inode, delegation);
300 		iput(inode);
301 		goto restart;
302 	}
303 	rcu_read_unlock();
304 out:
305 	nfs_put_client(clp);
306 	module_put_and_exit(0);
307 }
308 
309 void nfs_expire_all_delegations(struct nfs_client *clp)
310 {
311 	struct task_struct *task;
312 
313 	__module_get(THIS_MODULE);
314 	atomic_inc(&clp->cl_count);
315 	task = kthread_run(nfs_do_expire_all_delegations, clp,
316 			"%u.%u.%u.%u-delegreturn",
317 			NIPQUAD(clp->cl_addr.sin_addr));
318 	if (!IS_ERR(task))
319 		return;
320 	nfs_put_client(clp);
321 	module_put(THIS_MODULE);
322 }
323 
324 /*
325  * Return all delegations following an NFS4ERR_CB_PATH_DOWN error.
326  */
327 void nfs_handle_cb_pathdown(struct nfs_client *clp)
328 {
329 	struct nfs_delegation *delegation;
330 	struct inode *inode;
331 
332 	if (clp == NULL)
333 		return;
334 restart:
335 	rcu_read_lock();
336 	list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
337 		inode = igrab(delegation->inode);
338 		if (inode == NULL)
339 			continue;
340 		spin_lock(&clp->cl_lock);
341 		delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
342 		spin_unlock(&clp->cl_lock);
343 		rcu_read_unlock();
344 		if (delegation != NULL)
345 			__nfs_inode_return_delegation(inode, delegation);
346 		iput(inode);
347 		goto restart;
348 	}
349 	rcu_read_unlock();
350 }
351 
352 struct recall_threadargs {
353 	struct inode *inode;
354 	struct nfs_client *clp;
355 	const nfs4_stateid *stateid;
356 
357 	struct completion started;
358 	int result;
359 };
360 
361 static int recall_thread(void *data)
362 {
363 	struct recall_threadargs *args = (struct recall_threadargs *)data;
364 	struct inode *inode = igrab(args->inode);
365 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
366 	struct nfs_inode *nfsi = NFS_I(inode);
367 	struct nfs_delegation *delegation;
368 
369 	daemonize("nfsv4-delegreturn");
370 
371 	nfs_msync_inode(inode);
372 	down_read(&clp->cl_sem);
373 	down_write(&nfsi->rwsem);
374 	spin_lock(&clp->cl_lock);
375 	delegation = nfs_detach_delegation_locked(nfsi, args->stateid);
376 	if (delegation != NULL)
377 		args->result = 0;
378 	else
379 		args->result = -ENOENT;
380 	spin_unlock(&clp->cl_lock);
381 	complete(&args->started);
382 	nfs_delegation_claim_opens(inode, args->stateid);
383 	up_write(&nfsi->rwsem);
384 	up_read(&clp->cl_sem);
385 	nfs_msync_inode(inode);
386 
387 	if (delegation != NULL)
388 		nfs_do_return_delegation(inode, delegation);
389 	iput(inode);
390 	module_put_and_exit(0);
391 }
392 
393 /*
394  * Asynchronous delegation recall!
395  */
396 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid)
397 {
398 	struct recall_threadargs data = {
399 		.inode = inode,
400 		.stateid = stateid,
401 	};
402 	int status;
403 
404 	init_completion(&data.started);
405 	__module_get(THIS_MODULE);
406 	status = kernel_thread(recall_thread, &data, CLONE_KERNEL);
407 	if (status < 0)
408 		goto out_module_put;
409 	wait_for_completion(&data.started);
410 	return data.result;
411 out_module_put:
412 	module_put(THIS_MODULE);
413 	return status;
414 }
415 
416 /*
417  * Retrieve the inode associated with a delegation
418  */
419 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle)
420 {
421 	struct nfs_delegation *delegation;
422 	struct inode *res = NULL;
423 	rcu_read_lock();
424 	list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
425 		if (nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
426 			res = igrab(delegation->inode);
427 			break;
428 		}
429 	}
430 	rcu_read_unlock();
431 	return res;
432 }
433 
434 /*
435  * Mark all delegations as needing to be reclaimed
436  */
437 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
438 {
439 	struct nfs_delegation *delegation;
440 	rcu_read_lock();
441 	list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list)
442 		delegation->flags |= NFS_DELEGATION_NEED_RECLAIM;
443 	rcu_read_unlock();
444 }
445 
446 /*
447  * Reap all unclaimed delegations after reboot recovery is done
448  */
449 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
450 {
451 	struct nfs_delegation *delegation;
452 restart:
453 	rcu_read_lock();
454 	list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
455 		if ((delegation->flags & NFS_DELEGATION_NEED_RECLAIM) == 0)
456 			continue;
457 		spin_lock(&clp->cl_lock);
458 		delegation = nfs_detach_delegation_locked(NFS_I(delegation->inode), NULL);
459 		spin_unlock(&clp->cl_lock);
460 		rcu_read_unlock();
461 		if (delegation != NULL)
462 			nfs_free_delegation(delegation);
463 		goto restart;
464 	}
465 	rcu_read_unlock();
466 }
467 
468 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
469 {
470 	struct nfs_inode *nfsi = NFS_I(inode);
471 	struct nfs_delegation *delegation;
472 	int ret = 0;
473 
474 	rcu_read_lock();
475 	delegation = rcu_dereference(nfsi->delegation);
476 	if (delegation != NULL) {
477 		memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
478 		ret = 1;
479 	}
480 	rcu_read_unlock();
481 	return ret;
482 }
483