xref: /openbmc/linux/fs/nfs/nfs42proc.c (revision fadbafc1)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2014 Anna Schumaker <Anna.Schumaker@Netapp.com>
4  */
5 #include <linux/fs.h>
6 #include <linux/sunrpc/addr.h>
7 #include <linux/sunrpc/sched.h>
8 #include <linux/nfs.h>
9 #include <linux/nfs3.h>
10 #include <linux/nfs4.h>
11 #include <linux/nfs_xdr.h>
12 #include <linux/nfs_fs.h>
13 #include "nfs4_fs.h"
14 #include "nfs42.h"
15 #include "iostat.h"
16 #include "pnfs.h"
17 #include "nfs4session.h"
18 #include "internal.h"
19 #include "delegation.h"
20 #include "nfs4trace.h"
21 
22 #define NFSDBG_FACILITY NFSDBG_PROC
23 static int nfs42_do_offload_cancel_async(struct file *dst, nfs4_stateid *std);
24 
25 static void nfs42_set_netaddr(struct file *filep, struct nfs42_netaddr *naddr)
26 {
27 	struct nfs_client *clp = (NFS_SERVER(file_inode(filep)))->nfs_client;
28 	unsigned short port = 2049;
29 
30 	rcu_read_lock();
31 	naddr->netid_len = scnprintf(naddr->netid,
32 					sizeof(naddr->netid), "%s",
33 					rpc_peeraddr2str(clp->cl_rpcclient,
34 					RPC_DISPLAY_NETID));
35 	naddr->addr_len = scnprintf(naddr->addr,
36 					sizeof(naddr->addr),
37 					"%s.%u.%u",
38 					rpc_peeraddr2str(clp->cl_rpcclient,
39 					RPC_DISPLAY_ADDR),
40 					port >> 8, port & 255);
41 	rcu_read_unlock();
42 }
43 
44 static int _nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,
45 		struct nfs_lock_context *lock, loff_t offset, loff_t len)
46 {
47 	struct inode *inode = file_inode(filep);
48 	struct nfs_server *server = NFS_SERVER(inode);
49 	u32 bitmask[NFS_BITMASK_SZ];
50 	struct nfs42_falloc_args args = {
51 		.falloc_fh	= NFS_FH(inode),
52 		.falloc_offset	= offset,
53 		.falloc_length	= len,
54 		.falloc_bitmask	= bitmask,
55 	};
56 	struct nfs42_falloc_res res = {
57 		.falloc_server	= server,
58 	};
59 	int status;
60 
61 	msg->rpc_argp = &args;
62 	msg->rpc_resp = &res;
63 
64 	status = nfs4_set_rw_stateid(&args.falloc_stateid, lock->open_context,
65 			lock, FMODE_WRITE);
66 	if (status) {
67 		if (status == -EAGAIN)
68 			status = -NFS4ERR_BAD_STATEID;
69 		return status;
70 	}
71 
72 	nfs4_bitmask_set(bitmask, server->cache_consistency_bitmask, inode,
73 			 NFS_INO_INVALID_BLOCKS);
74 
75 	res.falloc_fattr = nfs_alloc_fattr();
76 	if (!res.falloc_fattr)
77 		return -ENOMEM;
78 
79 	status = nfs4_call_sync(server->client, server, msg,
80 				&args.seq_args, &res.seq_res, 0);
81 	if (status == 0) {
82 		if (nfs_should_remove_suid(inode)) {
83 			spin_lock(&inode->i_lock);
84 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
85 			spin_unlock(&inode->i_lock);
86 		}
87 		status = nfs_post_op_update_inode_force_wcc(inode,
88 							    res.falloc_fattr);
89 	}
90 	if (msg->rpc_proc == &nfs4_procedures[NFSPROC4_CLNT_ALLOCATE])
91 		trace_nfs4_fallocate(inode, &args, status);
92 	else
93 		trace_nfs4_deallocate(inode, &args, status);
94 	kfree(res.falloc_fattr);
95 	return status;
96 }
97 
98 static int nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,
99 				loff_t offset, loff_t len)
100 {
101 	struct inode *inode = file_inode(filep);
102 	struct nfs_server *server = NFS_SERVER(inode);
103 	struct nfs4_exception exception = { };
104 	struct nfs_lock_context *lock;
105 	int err;
106 
107 	lock = nfs_get_lock_context(nfs_file_open_context(filep));
108 	if (IS_ERR(lock))
109 		return PTR_ERR(lock);
110 
111 	exception.inode = inode;
112 	exception.state = lock->open_context->state;
113 
114 	err = nfs_sync_inode(inode);
115 	if (err)
116 		goto out;
117 
118 	do {
119 		err = _nfs42_proc_fallocate(msg, filep, lock, offset, len);
120 		if (err == -ENOTSUPP) {
121 			err = -EOPNOTSUPP;
122 			break;
123 		}
124 		err = nfs4_handle_exception(server, err, &exception);
125 	} while (exception.retry);
126 out:
127 	nfs_put_lock_context(lock);
128 	return err;
129 }
130 
131 int nfs42_proc_allocate(struct file *filep, loff_t offset, loff_t len)
132 {
133 	struct rpc_message msg = {
134 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_ALLOCATE],
135 	};
136 	struct inode *inode = file_inode(filep);
137 	int err;
138 
139 	if (!nfs_server_capable(inode, NFS_CAP_ALLOCATE))
140 		return -EOPNOTSUPP;
141 
142 	inode_lock(inode);
143 
144 	err = nfs42_proc_fallocate(&msg, filep, offset, len);
145 	if (err == -EOPNOTSUPP)
146 		NFS_SERVER(inode)->caps &= ~NFS_CAP_ALLOCATE;
147 
148 	inode_unlock(inode);
149 	return err;
150 }
151 
152 int nfs42_proc_deallocate(struct file *filep, loff_t offset, loff_t len)
153 {
154 	struct rpc_message msg = {
155 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_DEALLOCATE],
156 	};
157 	struct inode *inode = file_inode(filep);
158 	int err;
159 
160 	if (!nfs_server_capable(inode, NFS_CAP_DEALLOCATE))
161 		return -EOPNOTSUPP;
162 
163 	inode_lock(inode);
164 
165 	err = nfs42_proc_fallocate(&msg, filep, offset, len);
166 	if (err == 0)
167 		truncate_pagecache_range(inode, offset, (offset + len) -1);
168 	if (err == -EOPNOTSUPP)
169 		NFS_SERVER(inode)->caps &= ~NFS_CAP_DEALLOCATE;
170 
171 	inode_unlock(inode);
172 	return err;
173 }
174 
175 static int handle_async_copy(struct nfs42_copy_res *res,
176 			     struct nfs_server *dst_server,
177 			     struct nfs_server *src_server,
178 			     struct file *src,
179 			     struct file *dst,
180 			     nfs4_stateid *src_stateid,
181 			     bool *restart)
182 {
183 	struct nfs4_copy_state *copy, *tmp_copy = NULL, *iter;
184 	int status = NFS4_OK;
185 	struct nfs_open_context *dst_ctx = nfs_file_open_context(dst);
186 	struct nfs_open_context *src_ctx = nfs_file_open_context(src);
187 
188 	copy = kzalloc(sizeof(struct nfs4_copy_state), GFP_KERNEL);
189 	if (!copy)
190 		return -ENOMEM;
191 
192 	spin_lock(&dst_server->nfs_client->cl_lock);
193 	list_for_each_entry(iter,
194 				&dst_server->nfs_client->pending_cb_stateids,
195 				copies) {
196 		if (memcmp(&res->write_res.stateid, &iter->stateid,
197 				NFS4_STATEID_SIZE))
198 			continue;
199 		tmp_copy = iter;
200 		list_del(&iter->copies);
201 		break;
202 	}
203 	if (tmp_copy) {
204 		spin_unlock(&dst_server->nfs_client->cl_lock);
205 		kfree(copy);
206 		copy = tmp_copy;
207 		goto out;
208 	}
209 
210 	memcpy(&copy->stateid, &res->write_res.stateid, NFS4_STATEID_SIZE);
211 	init_completion(&copy->completion);
212 	copy->parent_dst_state = dst_ctx->state;
213 	copy->parent_src_state = src_ctx->state;
214 
215 	list_add_tail(&copy->copies, &dst_server->ss_copies);
216 	spin_unlock(&dst_server->nfs_client->cl_lock);
217 
218 	if (dst_server != src_server) {
219 		spin_lock(&src_server->nfs_client->cl_lock);
220 		list_add_tail(&copy->src_copies, &src_server->ss_copies);
221 		spin_unlock(&src_server->nfs_client->cl_lock);
222 	}
223 
224 	status = wait_for_completion_interruptible(&copy->completion);
225 	spin_lock(&dst_server->nfs_client->cl_lock);
226 	list_del_init(&copy->copies);
227 	spin_unlock(&dst_server->nfs_client->cl_lock);
228 	if (dst_server != src_server) {
229 		spin_lock(&src_server->nfs_client->cl_lock);
230 		list_del_init(&copy->src_copies);
231 		spin_unlock(&src_server->nfs_client->cl_lock);
232 	}
233 	if (status == -ERESTARTSYS) {
234 		goto out_cancel;
235 	} else if (copy->flags || copy->error == NFS4ERR_PARTNER_NO_AUTH) {
236 		status = -EAGAIN;
237 		*restart = true;
238 		goto out_cancel;
239 	}
240 out:
241 	res->write_res.count = copy->count;
242 	memcpy(&res->write_res.verifier, &copy->verf, sizeof(copy->verf));
243 	status = -copy->error;
244 
245 out_free:
246 	kfree(copy);
247 	return status;
248 out_cancel:
249 	nfs42_do_offload_cancel_async(dst, &copy->stateid);
250 	if (!nfs42_files_from_same_server(src, dst))
251 		nfs42_do_offload_cancel_async(src, src_stateid);
252 	goto out_free;
253 }
254 
255 static int process_copy_commit(struct file *dst, loff_t pos_dst,
256 			       struct nfs42_copy_res *res)
257 {
258 	struct nfs_commitres cres;
259 	int status = -ENOMEM;
260 
261 	cres.verf = kzalloc(sizeof(struct nfs_writeverf), GFP_KERNEL);
262 	if (!cres.verf)
263 		goto out;
264 
265 	status = nfs4_proc_commit(dst, pos_dst, res->write_res.count, &cres);
266 	if (status)
267 		goto out_free;
268 	if (nfs_write_verifier_cmp(&res->write_res.verifier.verifier,
269 				    &cres.verf->verifier)) {
270 		dprintk("commit verf differs from copy verf\n");
271 		status = -EAGAIN;
272 	}
273 out_free:
274 	kfree(cres.verf);
275 out:
276 	return status;
277 }
278 
279 /**
280  * nfs42_copy_dest_done - perform inode cache updates after clone/copy offload
281  * @inode: pointer to destination inode
282  * @pos: destination offset
283  * @len: copy length
284  *
285  * Punch a hole in the inode page cache, so that the NFS client will
286  * know to retrieve new data.
287  * Update the file size if necessary, and then mark the inode as having
288  * invalid cached values for change attribute, ctime, mtime and space used.
289  */
290 static void nfs42_copy_dest_done(struct inode *inode, loff_t pos, loff_t len)
291 {
292 	loff_t newsize = pos + len;
293 	loff_t end = newsize - 1;
294 
295 	WARN_ON_ONCE(invalidate_inode_pages2_range(inode->i_mapping,
296 				pos >> PAGE_SHIFT, end >> PAGE_SHIFT));
297 
298 	spin_lock(&inode->i_lock);
299 	if (newsize > i_size_read(inode))
300 		i_size_write(inode, newsize);
301 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE |
302 					     NFS_INO_INVALID_CTIME |
303 					     NFS_INO_INVALID_MTIME |
304 					     NFS_INO_INVALID_BLOCKS);
305 	spin_unlock(&inode->i_lock);
306 }
307 
308 static ssize_t _nfs42_proc_copy(struct file *src,
309 				struct nfs_lock_context *src_lock,
310 				struct file *dst,
311 				struct nfs_lock_context *dst_lock,
312 				struct nfs42_copy_args *args,
313 				struct nfs42_copy_res *res,
314 				struct nl4_server *nss,
315 				nfs4_stateid *cnr_stateid,
316 				bool *restart)
317 {
318 	struct rpc_message msg = {
319 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_COPY],
320 		.rpc_argp = args,
321 		.rpc_resp = res,
322 	};
323 	struct inode *dst_inode = file_inode(dst);
324 	struct inode *src_inode = file_inode(src);
325 	struct nfs_server *dst_server = NFS_SERVER(dst_inode);
326 	struct nfs_server *src_server = NFS_SERVER(src_inode);
327 	loff_t pos_src = args->src_pos;
328 	loff_t pos_dst = args->dst_pos;
329 	size_t count = args->count;
330 	ssize_t status;
331 
332 	if (nss) {
333 		args->cp_src = nss;
334 		nfs4_stateid_copy(&args->src_stateid, cnr_stateid);
335 	} else {
336 		status = nfs4_set_rw_stateid(&args->src_stateid,
337 				src_lock->open_context, src_lock, FMODE_READ);
338 		if (status) {
339 			if (status == -EAGAIN)
340 				status = -NFS4ERR_BAD_STATEID;
341 			return status;
342 		}
343 	}
344 	status = nfs_filemap_write_and_wait_range(src->f_mapping,
345 			pos_src, pos_src + (loff_t)count - 1);
346 	if (status)
347 		return status;
348 
349 	status = nfs4_set_rw_stateid(&args->dst_stateid, dst_lock->open_context,
350 				     dst_lock, FMODE_WRITE);
351 	if (status) {
352 		if (status == -EAGAIN)
353 			status = -NFS4ERR_BAD_STATEID;
354 		return status;
355 	}
356 
357 	status = nfs_sync_inode(dst_inode);
358 	if (status)
359 		return status;
360 
361 	res->commit_res.verf = NULL;
362 	if (args->sync) {
363 		res->commit_res.verf =
364 			kzalloc(sizeof(struct nfs_writeverf), GFP_KERNEL);
365 		if (!res->commit_res.verf)
366 			return -ENOMEM;
367 	}
368 	set_bit(NFS_CLNT_SRC_SSC_COPY_STATE,
369 		&src_lock->open_context->state->flags);
370 	set_bit(NFS_CLNT_DST_SSC_COPY_STATE,
371 		&dst_lock->open_context->state->flags);
372 
373 	status = nfs4_call_sync(dst_server->client, dst_server, &msg,
374 				&args->seq_args, &res->seq_res, 0);
375 	trace_nfs4_copy(src_inode, dst_inode, args, res, nss, status);
376 	if (status == -ENOTSUPP)
377 		dst_server->caps &= ~NFS_CAP_COPY;
378 	if (status)
379 		goto out;
380 
381 	if (args->sync &&
382 		nfs_write_verifier_cmp(&res->write_res.verifier.verifier,
383 				    &res->commit_res.verf->verifier)) {
384 		status = -EAGAIN;
385 		goto out;
386 	}
387 
388 	if (!res->synchronous) {
389 		status = handle_async_copy(res, dst_server, src_server, src,
390 				dst, &args->src_stateid, restart);
391 		if (status)
392 			goto out;
393 	}
394 
395 	if ((!res->synchronous || !args->sync) &&
396 			res->write_res.verifier.committed != NFS_FILE_SYNC) {
397 		status = process_copy_commit(dst, pos_dst, res);
398 		if (status)
399 			goto out;
400 	}
401 
402 	nfs42_copy_dest_done(dst_inode, pos_dst, res->write_res.count);
403 	nfs_invalidate_atime(src_inode);
404 	status = res->write_res.count;
405 out:
406 	if (args->sync)
407 		kfree(res->commit_res.verf);
408 	return status;
409 }
410 
411 ssize_t nfs42_proc_copy(struct file *src, loff_t pos_src,
412 			struct file *dst, loff_t pos_dst, size_t count,
413 			struct nl4_server *nss,
414 			nfs4_stateid *cnr_stateid, bool sync)
415 {
416 	struct nfs_server *server = NFS_SERVER(file_inode(dst));
417 	struct nfs_lock_context *src_lock;
418 	struct nfs_lock_context *dst_lock;
419 	struct nfs42_copy_args args = {
420 		.src_fh		= NFS_FH(file_inode(src)),
421 		.src_pos	= pos_src,
422 		.dst_fh		= NFS_FH(file_inode(dst)),
423 		.dst_pos	= pos_dst,
424 		.count		= count,
425 		.sync		= sync,
426 	};
427 	struct nfs42_copy_res res;
428 	struct nfs4_exception src_exception = {
429 		.inode		= file_inode(src),
430 		.stateid	= &args.src_stateid,
431 	};
432 	struct nfs4_exception dst_exception = {
433 		.inode		= file_inode(dst),
434 		.stateid	= &args.dst_stateid,
435 	};
436 	ssize_t err, err2;
437 	bool restart = false;
438 
439 	src_lock = nfs_get_lock_context(nfs_file_open_context(src));
440 	if (IS_ERR(src_lock))
441 		return PTR_ERR(src_lock);
442 
443 	src_exception.state = src_lock->open_context->state;
444 
445 	dst_lock = nfs_get_lock_context(nfs_file_open_context(dst));
446 	if (IS_ERR(dst_lock)) {
447 		err = PTR_ERR(dst_lock);
448 		goto out_put_src_lock;
449 	}
450 
451 	dst_exception.state = dst_lock->open_context->state;
452 
453 	do {
454 		inode_lock(file_inode(dst));
455 		err = _nfs42_proc_copy(src, src_lock,
456 				dst, dst_lock,
457 				&args, &res,
458 				nss, cnr_stateid, &restart);
459 		inode_unlock(file_inode(dst));
460 
461 		if (err >= 0)
462 			break;
463 		if (err == -ENOTSUPP &&
464 				nfs42_files_from_same_server(src, dst)) {
465 			err = -EOPNOTSUPP;
466 			break;
467 		} else if (err == -EAGAIN) {
468 			if (!restart) {
469 				dst_exception.retry = 1;
470 				continue;
471 			}
472 			break;
473 		} else if (err == -NFS4ERR_OFFLOAD_NO_REQS && !args.sync) {
474 			args.sync = true;
475 			dst_exception.retry = 1;
476 			continue;
477 		} else if ((err == -ESTALE ||
478 				err == -NFS4ERR_OFFLOAD_DENIED ||
479 				err == -ENOTSUPP) &&
480 				!nfs42_files_from_same_server(src, dst)) {
481 			nfs42_do_offload_cancel_async(src, &args.src_stateid);
482 			err = -EOPNOTSUPP;
483 			break;
484 		}
485 
486 		err2 = nfs4_handle_exception(server, err, &src_exception);
487 		err  = nfs4_handle_exception(server, err, &dst_exception);
488 		if (!err)
489 			err = err2;
490 	} while (src_exception.retry || dst_exception.retry);
491 
492 	nfs_put_lock_context(dst_lock);
493 out_put_src_lock:
494 	nfs_put_lock_context(src_lock);
495 	return err;
496 }
497 
498 struct nfs42_offloadcancel_data {
499 	struct nfs_server *seq_server;
500 	struct nfs42_offload_status_args args;
501 	struct nfs42_offload_status_res res;
502 };
503 
504 static void nfs42_offload_cancel_prepare(struct rpc_task *task, void *calldata)
505 {
506 	struct nfs42_offloadcancel_data *data = calldata;
507 
508 	nfs4_setup_sequence(data->seq_server->nfs_client,
509 				&data->args.osa_seq_args,
510 				&data->res.osr_seq_res, task);
511 }
512 
513 static void nfs42_offload_cancel_done(struct rpc_task *task, void *calldata)
514 {
515 	struct nfs42_offloadcancel_data *data = calldata;
516 
517 	trace_nfs4_offload_cancel(&data->args, task->tk_status);
518 	nfs41_sequence_done(task, &data->res.osr_seq_res);
519 	if (task->tk_status &&
520 		nfs4_async_handle_error(task, data->seq_server, NULL,
521 			NULL) == -EAGAIN)
522 		rpc_restart_call_prepare(task);
523 }
524 
525 static void nfs42_free_offloadcancel_data(void *data)
526 {
527 	kfree(data);
528 }
529 
530 static const struct rpc_call_ops nfs42_offload_cancel_ops = {
531 	.rpc_call_prepare = nfs42_offload_cancel_prepare,
532 	.rpc_call_done = nfs42_offload_cancel_done,
533 	.rpc_release = nfs42_free_offloadcancel_data,
534 };
535 
536 static int nfs42_do_offload_cancel_async(struct file *dst,
537 					 nfs4_stateid *stateid)
538 {
539 	struct nfs_server *dst_server = NFS_SERVER(file_inode(dst));
540 	struct nfs42_offloadcancel_data *data = NULL;
541 	struct nfs_open_context *ctx = nfs_file_open_context(dst);
542 	struct rpc_task *task;
543 	struct rpc_message msg = {
544 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OFFLOAD_CANCEL],
545 		.rpc_cred = ctx->cred,
546 	};
547 	struct rpc_task_setup task_setup_data = {
548 		.rpc_client = dst_server->client,
549 		.rpc_message = &msg,
550 		.callback_ops = &nfs42_offload_cancel_ops,
551 		.workqueue = nfsiod_workqueue,
552 		.flags = RPC_TASK_ASYNC,
553 	};
554 	int status;
555 
556 	if (!(dst_server->caps & NFS_CAP_OFFLOAD_CANCEL))
557 		return -EOPNOTSUPP;
558 
559 	data = kzalloc(sizeof(struct nfs42_offloadcancel_data), GFP_KERNEL);
560 	if (data == NULL)
561 		return -ENOMEM;
562 
563 	data->seq_server = dst_server;
564 	data->args.osa_src_fh = NFS_FH(file_inode(dst));
565 	memcpy(&data->args.osa_stateid, stateid,
566 		sizeof(data->args.osa_stateid));
567 	msg.rpc_argp = &data->args;
568 	msg.rpc_resp = &data->res;
569 	task_setup_data.callback_data = data;
570 	nfs4_init_sequence(&data->args.osa_seq_args, &data->res.osr_seq_res,
571 			   1, 0);
572 	task = rpc_run_task(&task_setup_data);
573 	if (IS_ERR(task))
574 		return PTR_ERR(task);
575 	status = rpc_wait_for_completion_task(task);
576 	if (status == -ENOTSUPP)
577 		dst_server->caps &= ~NFS_CAP_OFFLOAD_CANCEL;
578 	rpc_put_task(task);
579 	return status;
580 }
581 
582 static int _nfs42_proc_copy_notify(struct file *src, struct file *dst,
583 				   struct nfs42_copy_notify_args *args,
584 				   struct nfs42_copy_notify_res *res)
585 {
586 	struct nfs_server *src_server = NFS_SERVER(file_inode(src));
587 	struct rpc_message msg = {
588 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_COPY_NOTIFY],
589 		.rpc_argp = args,
590 		.rpc_resp = res,
591 	};
592 	int status;
593 	struct nfs_open_context *ctx;
594 	struct nfs_lock_context *l_ctx;
595 
596 	ctx = get_nfs_open_context(nfs_file_open_context(src));
597 	l_ctx = nfs_get_lock_context(ctx);
598 	if (IS_ERR(l_ctx)) {
599 		status = PTR_ERR(l_ctx);
600 		goto out;
601 	}
602 
603 	status = nfs4_set_rw_stateid(&args->cna_src_stateid, ctx, l_ctx,
604 				     FMODE_READ);
605 	nfs_put_lock_context(l_ctx);
606 	if (status) {
607 		if (status == -EAGAIN)
608 			status = -NFS4ERR_BAD_STATEID;
609 		goto out;
610 	}
611 
612 	status = nfs4_call_sync(src_server->client, src_server, &msg,
613 				&args->cna_seq_args, &res->cnr_seq_res, 0);
614 	trace_nfs4_copy_notify(file_inode(src), args, res, status);
615 	if (status == -ENOTSUPP)
616 		src_server->caps &= ~NFS_CAP_COPY_NOTIFY;
617 
618 out:
619 	put_nfs_open_context(nfs_file_open_context(src));
620 	return status;
621 }
622 
623 int nfs42_proc_copy_notify(struct file *src, struct file *dst,
624 				struct nfs42_copy_notify_res *res)
625 {
626 	struct nfs_server *src_server = NFS_SERVER(file_inode(src));
627 	struct nfs42_copy_notify_args *args;
628 	struct nfs4_exception exception = {
629 		.inode = file_inode(src),
630 	};
631 	int status;
632 
633 	if (!(src_server->caps & NFS_CAP_COPY_NOTIFY))
634 		return -EOPNOTSUPP;
635 
636 	args = kzalloc(sizeof(struct nfs42_copy_notify_args), GFP_KERNEL);
637 	if (args == NULL)
638 		return -ENOMEM;
639 
640 	args->cna_src_fh  = NFS_FH(file_inode(src)),
641 	args->cna_dst.nl4_type = NL4_NETADDR;
642 	nfs42_set_netaddr(dst, &args->cna_dst.u.nl4_addr);
643 	exception.stateid = &args->cna_src_stateid;
644 
645 	do {
646 		status = _nfs42_proc_copy_notify(src, dst, args, res);
647 		if (status == -ENOTSUPP) {
648 			status = -EOPNOTSUPP;
649 			goto out;
650 		}
651 		status = nfs4_handle_exception(src_server, status, &exception);
652 	} while (exception.retry);
653 
654 out:
655 	kfree(args);
656 	return status;
657 }
658 
659 static loff_t _nfs42_proc_llseek(struct file *filep,
660 		struct nfs_lock_context *lock, loff_t offset, int whence)
661 {
662 	struct inode *inode = file_inode(filep);
663 	struct nfs42_seek_args args = {
664 		.sa_fh		= NFS_FH(inode),
665 		.sa_offset	= offset,
666 		.sa_what	= (whence == SEEK_HOLE) ?
667 					NFS4_CONTENT_HOLE : NFS4_CONTENT_DATA,
668 	};
669 	struct nfs42_seek_res res;
670 	struct rpc_message msg = {
671 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_SEEK],
672 		.rpc_argp = &args,
673 		.rpc_resp = &res,
674 	};
675 	struct nfs_server *server = NFS_SERVER(inode);
676 	int status;
677 
678 	if (!nfs_server_capable(inode, NFS_CAP_SEEK))
679 		return -ENOTSUPP;
680 
681 	status = nfs4_set_rw_stateid(&args.sa_stateid, lock->open_context,
682 			lock, FMODE_READ);
683 	if (status) {
684 		if (status == -EAGAIN)
685 			status = -NFS4ERR_BAD_STATEID;
686 		return status;
687 	}
688 
689 	status = nfs_filemap_write_and_wait_range(inode->i_mapping,
690 			offset, LLONG_MAX);
691 	if (status)
692 		return status;
693 
694 	status = nfs4_call_sync(server->client, server, &msg,
695 				&args.seq_args, &res.seq_res, 0);
696 	trace_nfs4_llseek(inode, &args, &res, status);
697 	if (status == -ENOTSUPP)
698 		server->caps &= ~NFS_CAP_SEEK;
699 	if (status)
700 		return status;
701 
702 	if (whence == SEEK_DATA && res.sr_eof)
703 		return -NFS4ERR_NXIO;
704 	else
705 		return vfs_setpos(filep, res.sr_offset, inode->i_sb->s_maxbytes);
706 }
707 
708 loff_t nfs42_proc_llseek(struct file *filep, loff_t offset, int whence)
709 {
710 	struct nfs_server *server = NFS_SERVER(file_inode(filep));
711 	struct nfs4_exception exception = { };
712 	struct nfs_lock_context *lock;
713 	loff_t err;
714 
715 	lock = nfs_get_lock_context(nfs_file_open_context(filep));
716 	if (IS_ERR(lock))
717 		return PTR_ERR(lock);
718 
719 	exception.inode = file_inode(filep);
720 	exception.state = lock->open_context->state;
721 
722 	do {
723 		err = _nfs42_proc_llseek(filep, lock, offset, whence);
724 		if (err >= 0)
725 			break;
726 		if (err == -ENOTSUPP) {
727 			err = -EOPNOTSUPP;
728 			break;
729 		}
730 		err = nfs4_handle_exception(server, err, &exception);
731 	} while (exception.retry);
732 
733 	nfs_put_lock_context(lock);
734 	return err;
735 }
736 
737 
738 static void
739 nfs42_layoutstat_prepare(struct rpc_task *task, void *calldata)
740 {
741 	struct nfs42_layoutstat_data *data = calldata;
742 	struct inode *inode = data->inode;
743 	struct nfs_server *server = NFS_SERVER(inode);
744 	struct pnfs_layout_hdr *lo;
745 
746 	spin_lock(&inode->i_lock);
747 	lo = NFS_I(inode)->layout;
748 	if (!pnfs_layout_is_valid(lo)) {
749 		spin_unlock(&inode->i_lock);
750 		rpc_exit(task, 0);
751 		return;
752 	}
753 	nfs4_stateid_copy(&data->args.stateid, &lo->plh_stateid);
754 	spin_unlock(&inode->i_lock);
755 	nfs4_setup_sequence(server->nfs_client, &data->args.seq_args,
756 			    &data->res.seq_res, task);
757 }
758 
759 static void
760 nfs42_layoutstat_done(struct rpc_task *task, void *calldata)
761 {
762 	struct nfs42_layoutstat_data *data = calldata;
763 	struct inode *inode = data->inode;
764 	struct pnfs_layout_hdr *lo;
765 
766 	if (!nfs4_sequence_done(task, &data->res.seq_res))
767 		return;
768 
769 	switch (task->tk_status) {
770 	case 0:
771 		return;
772 	case -NFS4ERR_BADHANDLE:
773 	case -ESTALE:
774 		pnfs_destroy_layout(NFS_I(inode));
775 		break;
776 	case -NFS4ERR_EXPIRED:
777 	case -NFS4ERR_ADMIN_REVOKED:
778 	case -NFS4ERR_DELEG_REVOKED:
779 	case -NFS4ERR_STALE_STATEID:
780 	case -NFS4ERR_BAD_STATEID:
781 		spin_lock(&inode->i_lock);
782 		lo = NFS_I(inode)->layout;
783 		if (pnfs_layout_is_valid(lo) &&
784 		    nfs4_stateid_match(&data->args.stateid,
785 					     &lo->plh_stateid)) {
786 			LIST_HEAD(head);
787 
788 			/*
789 			 * Mark the bad layout state as invalid, then retry
790 			 * with the current stateid.
791 			 */
792 			pnfs_mark_layout_stateid_invalid(lo, &head);
793 			spin_unlock(&inode->i_lock);
794 			pnfs_free_lseg_list(&head);
795 			nfs_commit_inode(inode, 0);
796 		} else
797 			spin_unlock(&inode->i_lock);
798 		break;
799 	case -NFS4ERR_OLD_STATEID:
800 		spin_lock(&inode->i_lock);
801 		lo = NFS_I(inode)->layout;
802 		if (pnfs_layout_is_valid(lo) &&
803 		    nfs4_stateid_match_other(&data->args.stateid,
804 					&lo->plh_stateid)) {
805 			/* Do we need to delay before resending? */
806 			if (!nfs4_stateid_is_newer(&lo->plh_stateid,
807 						&data->args.stateid))
808 				rpc_delay(task, HZ);
809 			rpc_restart_call_prepare(task);
810 		}
811 		spin_unlock(&inode->i_lock);
812 		break;
813 	case -ENOTSUPP:
814 	case -EOPNOTSUPP:
815 		NFS_SERVER(inode)->caps &= ~NFS_CAP_LAYOUTSTATS;
816 	}
817 
818 	trace_nfs4_layoutstats(inode, &data->args.stateid, task->tk_status);
819 }
820 
821 static void
822 nfs42_layoutstat_release(void *calldata)
823 {
824 	struct nfs42_layoutstat_data *data = calldata;
825 	struct nfs42_layoutstat_devinfo *devinfo = data->args.devinfo;
826 	int i;
827 
828 	for (i = 0; i < data->args.num_dev; i++) {
829 		if (devinfo[i].ld_private.ops && devinfo[i].ld_private.ops->free)
830 			devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
831 	}
832 
833 	pnfs_put_layout_hdr(NFS_I(data->args.inode)->layout);
834 	smp_mb__before_atomic();
835 	clear_bit(NFS_INO_LAYOUTSTATS, &NFS_I(data->args.inode)->flags);
836 	smp_mb__after_atomic();
837 	nfs_iput_and_deactive(data->inode);
838 	kfree(data->args.devinfo);
839 	kfree(data);
840 }
841 
842 static const struct rpc_call_ops nfs42_layoutstat_ops = {
843 	.rpc_call_prepare = nfs42_layoutstat_prepare,
844 	.rpc_call_done = nfs42_layoutstat_done,
845 	.rpc_release = nfs42_layoutstat_release,
846 };
847 
848 int nfs42_proc_layoutstats_generic(struct nfs_server *server,
849 				   struct nfs42_layoutstat_data *data)
850 {
851 	struct rpc_message msg = {
852 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LAYOUTSTATS],
853 		.rpc_argp = &data->args,
854 		.rpc_resp = &data->res,
855 	};
856 	struct rpc_task_setup task_setup = {
857 		.rpc_client = server->client,
858 		.rpc_message = &msg,
859 		.callback_ops = &nfs42_layoutstat_ops,
860 		.callback_data = data,
861 		.flags = RPC_TASK_ASYNC,
862 	};
863 	struct rpc_task *task;
864 
865 	data->inode = nfs_igrab_and_active(data->args.inode);
866 	if (!data->inode) {
867 		nfs42_layoutstat_release(data);
868 		return -EAGAIN;
869 	}
870 	nfs4_init_sequence(&data->args.seq_args, &data->res.seq_res, 0, 0);
871 	task = rpc_run_task(&task_setup);
872 	if (IS_ERR(task))
873 		return PTR_ERR(task);
874 	rpc_put_task(task);
875 	return 0;
876 }
877 
878 static struct nfs42_layouterror_data *
879 nfs42_alloc_layouterror_data(struct pnfs_layout_segment *lseg, gfp_t gfp_flags)
880 {
881 	struct nfs42_layouterror_data *data;
882 	struct inode *inode = lseg->pls_layout->plh_inode;
883 
884 	data = kzalloc(sizeof(*data), gfp_flags);
885 	if (data) {
886 		data->args.inode = data->inode = nfs_igrab_and_active(inode);
887 		if (data->inode) {
888 			data->lseg = pnfs_get_lseg(lseg);
889 			if (data->lseg)
890 				return data;
891 			nfs_iput_and_deactive(data->inode);
892 		}
893 		kfree(data);
894 	}
895 	return NULL;
896 }
897 
898 static void
899 nfs42_free_layouterror_data(struct nfs42_layouterror_data *data)
900 {
901 	pnfs_put_lseg(data->lseg);
902 	nfs_iput_and_deactive(data->inode);
903 	kfree(data);
904 }
905 
906 static void
907 nfs42_layouterror_prepare(struct rpc_task *task, void *calldata)
908 {
909 	struct nfs42_layouterror_data *data = calldata;
910 	struct inode *inode = data->inode;
911 	struct nfs_server *server = NFS_SERVER(inode);
912 	struct pnfs_layout_hdr *lo = data->lseg->pls_layout;
913 	unsigned i;
914 
915 	spin_lock(&inode->i_lock);
916 	if (!pnfs_layout_is_valid(lo)) {
917 		spin_unlock(&inode->i_lock);
918 		rpc_exit(task, 0);
919 		return;
920 	}
921 	for (i = 0; i < data->args.num_errors; i++)
922 		nfs4_stateid_copy(&data->args.errors[i].stateid,
923 				&lo->plh_stateid);
924 	spin_unlock(&inode->i_lock);
925 	nfs4_setup_sequence(server->nfs_client, &data->args.seq_args,
926 			    &data->res.seq_res, task);
927 }
928 
929 static void
930 nfs42_layouterror_done(struct rpc_task *task, void *calldata)
931 {
932 	struct nfs42_layouterror_data *data = calldata;
933 	struct inode *inode = data->inode;
934 	struct pnfs_layout_hdr *lo = data->lseg->pls_layout;
935 
936 	if (!nfs4_sequence_done(task, &data->res.seq_res))
937 		return;
938 
939 	switch (task->tk_status) {
940 	case 0:
941 		return;
942 	case -NFS4ERR_BADHANDLE:
943 	case -ESTALE:
944 		pnfs_destroy_layout(NFS_I(inode));
945 		break;
946 	case -NFS4ERR_EXPIRED:
947 	case -NFS4ERR_ADMIN_REVOKED:
948 	case -NFS4ERR_DELEG_REVOKED:
949 	case -NFS4ERR_STALE_STATEID:
950 	case -NFS4ERR_BAD_STATEID:
951 		spin_lock(&inode->i_lock);
952 		if (pnfs_layout_is_valid(lo) &&
953 		    nfs4_stateid_match(&data->args.errors[0].stateid,
954 					     &lo->plh_stateid)) {
955 			LIST_HEAD(head);
956 
957 			/*
958 			 * Mark the bad layout state as invalid, then retry
959 			 * with the current stateid.
960 			 */
961 			pnfs_mark_layout_stateid_invalid(lo, &head);
962 			spin_unlock(&inode->i_lock);
963 			pnfs_free_lseg_list(&head);
964 			nfs_commit_inode(inode, 0);
965 		} else
966 			spin_unlock(&inode->i_lock);
967 		break;
968 	case -NFS4ERR_OLD_STATEID:
969 		spin_lock(&inode->i_lock);
970 		if (pnfs_layout_is_valid(lo) &&
971 		    nfs4_stateid_match_other(&data->args.errors[0].stateid,
972 					&lo->plh_stateid)) {
973 			/* Do we need to delay before resending? */
974 			if (!nfs4_stateid_is_newer(&lo->plh_stateid,
975 						&data->args.errors[0].stateid))
976 				rpc_delay(task, HZ);
977 			rpc_restart_call_prepare(task);
978 		}
979 		spin_unlock(&inode->i_lock);
980 		break;
981 	case -ENOTSUPP:
982 	case -EOPNOTSUPP:
983 		NFS_SERVER(inode)->caps &= ~NFS_CAP_LAYOUTERROR;
984 	}
985 
986 	trace_nfs4_layouterror(inode, &data->args.errors[0].stateid,
987 			       task->tk_status);
988 }
989 
990 static void
991 nfs42_layouterror_release(void *calldata)
992 {
993 	struct nfs42_layouterror_data *data = calldata;
994 
995 	nfs42_free_layouterror_data(data);
996 }
997 
998 static const struct rpc_call_ops nfs42_layouterror_ops = {
999 	.rpc_call_prepare = nfs42_layouterror_prepare,
1000 	.rpc_call_done = nfs42_layouterror_done,
1001 	.rpc_release = nfs42_layouterror_release,
1002 };
1003 
1004 int nfs42_proc_layouterror(struct pnfs_layout_segment *lseg,
1005 		const struct nfs42_layout_error *errors, size_t n)
1006 {
1007 	struct inode *inode = lseg->pls_layout->plh_inode;
1008 	struct nfs42_layouterror_data *data;
1009 	struct rpc_task *task;
1010 	struct rpc_message msg = {
1011 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LAYOUTERROR],
1012 	};
1013 	struct rpc_task_setup task_setup = {
1014 		.rpc_message = &msg,
1015 		.callback_ops = &nfs42_layouterror_ops,
1016 		.flags = RPC_TASK_ASYNC,
1017 	};
1018 	unsigned int i;
1019 
1020 	if (!nfs_server_capable(inode, NFS_CAP_LAYOUTERROR))
1021 		return -EOPNOTSUPP;
1022 	if (n > NFS42_LAYOUTERROR_MAX)
1023 		return -EINVAL;
1024 	data = nfs42_alloc_layouterror_data(lseg, nfs_io_gfp_mask());
1025 	if (!data)
1026 		return -ENOMEM;
1027 	for (i = 0; i < n; i++) {
1028 		data->args.errors[i] = errors[i];
1029 		data->args.num_errors++;
1030 		data->res.num_errors++;
1031 	}
1032 	msg.rpc_argp = &data->args;
1033 	msg.rpc_resp = &data->res;
1034 	task_setup.callback_data = data;
1035 	task_setup.rpc_client = NFS_SERVER(inode)->client;
1036 	nfs4_init_sequence(&data->args.seq_args, &data->res.seq_res, 0, 0);
1037 	task = rpc_run_task(&task_setup);
1038 	if (IS_ERR(task))
1039 		return PTR_ERR(task);
1040 	rpc_put_task(task);
1041 	return 0;
1042 }
1043 EXPORT_SYMBOL_GPL(nfs42_proc_layouterror);
1044 
1045 static int _nfs42_proc_clone(struct rpc_message *msg, struct file *src_f,
1046 		struct file *dst_f, struct nfs_lock_context *src_lock,
1047 		struct nfs_lock_context *dst_lock, loff_t src_offset,
1048 		loff_t dst_offset, loff_t count)
1049 {
1050 	struct inode *src_inode = file_inode(src_f);
1051 	struct inode *dst_inode = file_inode(dst_f);
1052 	struct nfs_server *server = NFS_SERVER(dst_inode);
1053 	__u32 dst_bitmask[NFS_BITMASK_SZ];
1054 	struct nfs42_clone_args args = {
1055 		.src_fh = NFS_FH(src_inode),
1056 		.dst_fh = NFS_FH(dst_inode),
1057 		.src_offset = src_offset,
1058 		.dst_offset = dst_offset,
1059 		.count = count,
1060 		.dst_bitmask = dst_bitmask,
1061 	};
1062 	struct nfs42_clone_res res = {
1063 		.server	= server,
1064 	};
1065 	int status;
1066 
1067 	msg->rpc_argp = &args;
1068 	msg->rpc_resp = &res;
1069 
1070 	status = nfs4_set_rw_stateid(&args.src_stateid, src_lock->open_context,
1071 			src_lock, FMODE_READ);
1072 	if (status) {
1073 		if (status == -EAGAIN)
1074 			status = -NFS4ERR_BAD_STATEID;
1075 		return status;
1076 	}
1077 	status = nfs4_set_rw_stateid(&args.dst_stateid, dst_lock->open_context,
1078 			dst_lock, FMODE_WRITE);
1079 	if (status) {
1080 		if (status == -EAGAIN)
1081 			status = -NFS4ERR_BAD_STATEID;
1082 		return status;
1083 	}
1084 
1085 	res.dst_fattr = nfs_alloc_fattr();
1086 	if (!res.dst_fattr)
1087 		return -ENOMEM;
1088 
1089 	nfs4_bitmask_set(dst_bitmask, server->cache_consistency_bitmask,
1090 			 dst_inode, NFS_INO_INVALID_BLOCKS);
1091 
1092 	status = nfs4_call_sync(server->client, server, msg,
1093 				&args.seq_args, &res.seq_res, 0);
1094 	trace_nfs4_clone(src_inode, dst_inode, &args, status);
1095 	if (status == 0) {
1096 		nfs42_copy_dest_done(dst_inode, dst_offset, count);
1097 		status = nfs_post_op_update_inode(dst_inode, res.dst_fattr);
1098 	}
1099 
1100 	kfree(res.dst_fattr);
1101 	return status;
1102 }
1103 
1104 int nfs42_proc_clone(struct file *src_f, struct file *dst_f,
1105 		     loff_t src_offset, loff_t dst_offset, loff_t count)
1106 {
1107 	struct rpc_message msg = {
1108 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLONE],
1109 	};
1110 	struct inode *inode = file_inode(src_f);
1111 	struct nfs_server *server = NFS_SERVER(file_inode(src_f));
1112 	struct nfs_lock_context *src_lock;
1113 	struct nfs_lock_context *dst_lock;
1114 	struct nfs4_exception src_exception = { };
1115 	struct nfs4_exception dst_exception = { };
1116 	int err, err2;
1117 
1118 	if (!nfs_server_capable(inode, NFS_CAP_CLONE))
1119 		return -EOPNOTSUPP;
1120 
1121 	src_lock = nfs_get_lock_context(nfs_file_open_context(src_f));
1122 	if (IS_ERR(src_lock))
1123 		return PTR_ERR(src_lock);
1124 
1125 	src_exception.inode = file_inode(src_f);
1126 	src_exception.state = src_lock->open_context->state;
1127 
1128 	dst_lock = nfs_get_lock_context(nfs_file_open_context(dst_f));
1129 	if (IS_ERR(dst_lock)) {
1130 		err = PTR_ERR(dst_lock);
1131 		goto out_put_src_lock;
1132 	}
1133 
1134 	dst_exception.inode = file_inode(dst_f);
1135 	dst_exception.state = dst_lock->open_context->state;
1136 
1137 	do {
1138 		err = _nfs42_proc_clone(&msg, src_f, dst_f, src_lock, dst_lock,
1139 					src_offset, dst_offset, count);
1140 		if (err == -ENOTSUPP || err == -EOPNOTSUPP) {
1141 			NFS_SERVER(inode)->caps &= ~NFS_CAP_CLONE;
1142 			err = -EOPNOTSUPP;
1143 			break;
1144 		}
1145 
1146 		err2 = nfs4_handle_exception(server, err, &src_exception);
1147 		err = nfs4_handle_exception(server, err, &dst_exception);
1148 		if (!err)
1149 			err = err2;
1150 	} while (src_exception.retry || dst_exception.retry);
1151 
1152 	nfs_put_lock_context(dst_lock);
1153 out_put_src_lock:
1154 	nfs_put_lock_context(src_lock);
1155 	return err;
1156 }
1157 
1158 #define NFS4XATTR_MAXPAGES DIV_ROUND_UP(XATTR_SIZE_MAX, PAGE_SIZE)
1159 
1160 static int _nfs42_proc_removexattr(struct inode *inode, const char *name)
1161 {
1162 	struct nfs_server *server = NFS_SERVER(inode);
1163 	struct nfs42_removexattrargs args = {
1164 		.fh = NFS_FH(inode),
1165 		.xattr_name = name,
1166 	};
1167 	struct nfs42_removexattrres res;
1168 	struct rpc_message msg = {
1169 		.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_REMOVEXATTR],
1170 		.rpc_argp = &args,
1171 		.rpc_resp = &res,
1172 	};
1173 	int ret;
1174 	unsigned long timestamp = jiffies;
1175 
1176 	ret = nfs4_call_sync(server->client, server, &msg, &args.seq_args,
1177 	    &res.seq_res, 1);
1178 	trace_nfs4_removexattr(inode, name, ret);
1179 	if (!ret)
1180 		nfs4_update_changeattr(inode, &res.cinfo, timestamp, 0);
1181 
1182 	return ret;
1183 }
1184 
1185 static int _nfs42_proc_setxattr(struct inode *inode, const char *name,
1186 				const void *buf, size_t buflen, int flags)
1187 {
1188 	struct nfs_server *server = NFS_SERVER(inode);
1189 	struct page *pages[NFS4XATTR_MAXPAGES];
1190 	struct nfs42_setxattrargs arg = {
1191 		.fh		= NFS_FH(inode),
1192 		.xattr_pages	= pages,
1193 		.xattr_len	= buflen,
1194 		.xattr_name	= name,
1195 		.xattr_flags	= flags,
1196 	};
1197 	struct nfs42_setxattrres res;
1198 	struct rpc_message msg = {
1199 		.rpc_proc	= &nfs4_procedures[NFSPROC4_CLNT_SETXATTR],
1200 		.rpc_argp	= &arg,
1201 		.rpc_resp	= &res,
1202 	};
1203 	int ret, np;
1204 	unsigned long timestamp = jiffies;
1205 
1206 	if (buflen > server->sxasize)
1207 		return -ERANGE;
1208 
1209 	if (buflen > 0) {
1210 		np = nfs4_buf_to_pages_noslab(buf, buflen, arg.xattr_pages);
1211 		if (np < 0)
1212 			return np;
1213 	} else
1214 		np = 0;
1215 
1216 	ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args,
1217 	    &res.seq_res, 1);
1218 	trace_nfs4_setxattr(inode, name, ret);
1219 
1220 	for (; np > 0; np--)
1221 		put_page(pages[np - 1]);
1222 
1223 	if (!ret)
1224 		nfs4_update_changeattr(inode, &res.cinfo, timestamp, 0);
1225 
1226 	return ret;
1227 }
1228 
1229 static ssize_t _nfs42_proc_getxattr(struct inode *inode, const char *name,
1230 				void *buf, size_t buflen, struct page **pages,
1231 				size_t plen)
1232 {
1233 	struct nfs_server *server = NFS_SERVER(inode);
1234 	struct nfs42_getxattrargs arg = {
1235 		.fh		= NFS_FH(inode),
1236 		.xattr_name	= name,
1237 	};
1238 	struct nfs42_getxattrres res;
1239 	struct rpc_message msg = {
1240 		.rpc_proc	= &nfs4_procedures[NFSPROC4_CLNT_GETXATTR],
1241 		.rpc_argp	= &arg,
1242 		.rpc_resp	= &res,
1243 	};
1244 	ssize_t ret;
1245 
1246 	arg.xattr_len = plen;
1247 	arg.xattr_pages = pages;
1248 
1249 	ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args,
1250 	    &res.seq_res, 0);
1251 	trace_nfs4_getxattr(inode, name, ret);
1252 	if (ret < 0)
1253 		return ret;
1254 
1255 	/*
1256 	 * Normally, the caching is done one layer up, but for successful
1257 	 * RPCS, always cache the result here, even if the caller was
1258 	 * just querying the length, or if the reply was too big for
1259 	 * the caller. This avoids a second RPC in the case of the
1260 	 * common query-alloc-retrieve cycle for xattrs.
1261 	 *
1262 	 * Note that xattr_len is always capped to XATTR_SIZE_MAX.
1263 	 */
1264 
1265 	nfs4_xattr_cache_add(inode, name, NULL, pages, res.xattr_len);
1266 
1267 	if (buflen) {
1268 		if (res.xattr_len > buflen)
1269 			return -ERANGE;
1270 		_copy_from_pages(buf, pages, 0, res.xattr_len);
1271 	}
1272 
1273 	return res.xattr_len;
1274 }
1275 
1276 static ssize_t _nfs42_proc_listxattrs(struct inode *inode, void *buf,
1277 				 size_t buflen, u64 *cookiep, bool *eofp)
1278 {
1279 	struct nfs_server *server = NFS_SERVER(inode);
1280 	struct page **pages;
1281 	struct nfs42_listxattrsargs arg = {
1282 		.fh		= NFS_FH(inode),
1283 		.cookie		= *cookiep,
1284 	};
1285 	struct nfs42_listxattrsres res = {
1286 		.eof = false,
1287 		.xattr_buf = buf,
1288 		.xattr_len = buflen,
1289 	};
1290 	struct rpc_message msg = {
1291 		.rpc_proc	= &nfs4_procedures[NFSPROC4_CLNT_LISTXATTRS],
1292 		.rpc_argp	= &arg,
1293 		.rpc_resp	= &res,
1294 	};
1295 	u32 xdrlen;
1296 	int ret, np, i;
1297 
1298 
1299 	ret = -ENOMEM;
1300 	res.scratch = alloc_page(GFP_KERNEL);
1301 	if (!res.scratch)
1302 		goto out;
1303 
1304 	xdrlen = nfs42_listxattr_xdrsize(buflen);
1305 	if (xdrlen > server->lxasize)
1306 		xdrlen = server->lxasize;
1307 	np = xdrlen / PAGE_SIZE + 1;
1308 
1309 	pages = kcalloc(np, sizeof(struct page *), GFP_KERNEL);
1310 	if (!pages)
1311 		goto out_free_scratch;
1312 	for (i = 0; i < np; i++) {
1313 		pages[i] = alloc_page(GFP_KERNEL);
1314 		if (!pages[i])
1315 			goto out_free_pages;
1316 	}
1317 
1318 	arg.xattr_pages = pages;
1319 	arg.count = xdrlen;
1320 
1321 	ret = nfs4_call_sync(server->client, server, &msg, &arg.seq_args,
1322 	    &res.seq_res, 0);
1323 	trace_nfs4_listxattr(inode, ret);
1324 
1325 	if (ret >= 0) {
1326 		ret = res.copied;
1327 		*cookiep = res.cookie;
1328 		*eofp = res.eof;
1329 	}
1330 
1331 out_free_pages:
1332 	while (--np >= 0) {
1333 		if (pages[np])
1334 			__free_page(pages[np]);
1335 	}
1336 	kfree(pages);
1337 out_free_scratch:
1338 	__free_page(res.scratch);
1339 out:
1340 	return ret;
1341 
1342 }
1343 
1344 ssize_t nfs42_proc_getxattr(struct inode *inode, const char *name,
1345 			      void *buf, size_t buflen)
1346 {
1347 	struct nfs4_exception exception = { };
1348 	ssize_t err, np, i;
1349 	struct page **pages;
1350 
1351 	np = nfs_page_array_len(0, buflen ?: XATTR_SIZE_MAX);
1352 	pages = kmalloc_array(np, sizeof(*pages), GFP_KERNEL);
1353 	if (!pages)
1354 		return -ENOMEM;
1355 
1356 	for (i = 0; i < np; i++) {
1357 		pages[i] = alloc_page(GFP_KERNEL);
1358 		if (!pages[i]) {
1359 			np = i + 1;
1360 			err = -ENOMEM;
1361 			goto out;
1362 		}
1363 	}
1364 
1365 	/*
1366 	 * The GETXATTR op has no length field in the call, and the
1367 	 * xattr data is at the end of the reply.
1368 	 *
1369 	 * There is no downside in using the page-aligned length. It will
1370 	 * allow receiving and caching xattrs that are too large for the
1371 	 * caller but still fit in the page-rounded value.
1372 	 */
1373 	do {
1374 		err = _nfs42_proc_getxattr(inode, name, buf, buflen,
1375 			pages, np * PAGE_SIZE);
1376 		if (err >= 0)
1377 			break;
1378 		err = nfs4_handle_exception(NFS_SERVER(inode), err,
1379 				&exception);
1380 	} while (exception.retry);
1381 
1382 out:
1383 	while (--np >= 0)
1384 		__free_page(pages[np]);
1385 	kfree(pages);
1386 
1387 	return err;
1388 }
1389 
1390 int nfs42_proc_setxattr(struct inode *inode, const char *name,
1391 			      const void *buf, size_t buflen, int flags)
1392 {
1393 	struct nfs4_exception exception = { };
1394 	int err;
1395 
1396 	do {
1397 		err = _nfs42_proc_setxattr(inode, name, buf, buflen, flags);
1398 		if (!err)
1399 			break;
1400 		err = nfs4_handle_exception(NFS_SERVER(inode), err,
1401 				&exception);
1402 	} while (exception.retry);
1403 
1404 	return err;
1405 }
1406 
1407 ssize_t nfs42_proc_listxattrs(struct inode *inode, void *buf,
1408 			      size_t buflen, u64 *cookiep, bool *eofp)
1409 {
1410 	struct nfs4_exception exception = { };
1411 	ssize_t err;
1412 
1413 	do {
1414 		err = _nfs42_proc_listxattrs(inode, buf, buflen,
1415 		    cookiep, eofp);
1416 		if (err >= 0)
1417 			break;
1418 		err = nfs4_handle_exception(NFS_SERVER(inode), err,
1419 				&exception);
1420 	} while (exception.retry);
1421 
1422 	return err;
1423 }
1424 
1425 int nfs42_proc_removexattr(struct inode *inode, const char *name)
1426 {
1427 	struct nfs4_exception exception = { };
1428 	int err;
1429 
1430 	do {
1431 		err = _nfs42_proc_removexattr(inode, name);
1432 		if (!err)
1433 			break;
1434 		err = nfs4_handle_exception(NFS_SERVER(inode), err,
1435 				&exception);
1436 	} while (exception.retry);
1437 
1438 	return err;
1439 }
1440