xref: /linux/fs/nfsd/nfsctl.c (revision 70ab9ec9166db90ab8980aff4f7083512ecddd1f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Syscall interface to knfsd.
4  *
5  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
6  */
7 
8 #include <linux/slab.h>
9 #include <linux/namei.h>
10 #include <linux/ctype.h>
11 #include <linux/fs_context.h>
12 
13 #include <linux/sunrpc/svcsock.h>
14 #include <linux/lockd/lockd.h>
15 #include <linux/sunrpc/addr.h>
16 #include <linux/sunrpc/gss_api.h>
17 #include <linux/sunrpc/rpc_pipe_fs.h>
18 #include <linux/module.h>
19 #include <linux/fsnotify.h>
20 
21 #include "idmap.h"
22 #include "nfsd.h"
23 #include "cache.h"
24 #include "state.h"
25 #include "netns.h"
26 #include "pnfs.h"
27 #include "filecache.h"
28 #include "trace.h"
29 #include "netlink.h"
30 
31 /*
32  *	We have a single directory with several nodes in it.
33  */
34 enum {
35 	NFSD_Root = 1,
36 	NFSD_List,
37 	NFSD_Export_Stats,
38 	NFSD_Export_features,
39 	NFSD_Fh,
40 	NFSD_FO_UnlockIP,
41 	NFSD_FO_UnlockFS,
42 	NFSD_Threads,
43 	NFSD_Pool_Threads,
44 	NFSD_Pool_Stats,
45 	NFSD_Reply_Cache_Stats,
46 	NFSD_Versions,
47 	NFSD_Ports,
48 	NFSD_MaxBlkSize,
49 	NFSD_MaxConnections,
50 	NFSD_Filecache,
51 	/*
52 	 * The below MUST come last.  Otherwise we leave a hole in nfsd_files[]
53 	 * with !CONFIG_NFSD_V4 and simple_fill_super() goes oops
54 	 */
55 #ifdef CONFIG_NFSD_V4
56 	NFSD_Leasetime,
57 	NFSD_Gracetime,
58 	NFSD_RecoveryDir,
59 	NFSD_V4EndGrace,
60 #endif
61 	NFSD_MaxReserved
62 };
63 
64 /*
65  * write() for these nodes.
66  */
67 static ssize_t write_filehandle(struct file *file, char *buf, size_t size);
68 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size);
69 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size);
70 static ssize_t write_threads(struct file *file, char *buf, size_t size);
71 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size);
72 static ssize_t write_versions(struct file *file, char *buf, size_t size);
73 static ssize_t write_ports(struct file *file, char *buf, size_t size);
74 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size);
75 static ssize_t write_maxconn(struct file *file, char *buf, size_t size);
76 #ifdef CONFIG_NFSD_V4
77 static ssize_t write_leasetime(struct file *file, char *buf, size_t size);
78 static ssize_t write_gracetime(struct file *file, char *buf, size_t size);
79 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size);
80 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size);
81 #endif
82 
83 static ssize_t (*const write_op[])(struct file *, char *, size_t) = {
84 	[NFSD_Fh] = write_filehandle,
85 	[NFSD_FO_UnlockIP] = write_unlock_ip,
86 	[NFSD_FO_UnlockFS] = write_unlock_fs,
87 	[NFSD_Threads] = write_threads,
88 	[NFSD_Pool_Threads] = write_pool_threads,
89 	[NFSD_Versions] = write_versions,
90 	[NFSD_Ports] = write_ports,
91 	[NFSD_MaxBlkSize] = write_maxblksize,
92 	[NFSD_MaxConnections] = write_maxconn,
93 #ifdef CONFIG_NFSD_V4
94 	[NFSD_Leasetime] = write_leasetime,
95 	[NFSD_Gracetime] = write_gracetime,
96 	[NFSD_RecoveryDir] = write_recoverydir,
97 	[NFSD_V4EndGrace] = write_v4_end_grace,
98 #endif
99 };
100 
101 static ssize_t nfsctl_transaction_write(struct file *file, const char __user *buf, size_t size, loff_t *pos)
102 {
103 	ino_t ino =  file_inode(file)->i_ino;
104 	char *data;
105 	ssize_t rv;
106 
107 	if (ino >= ARRAY_SIZE(write_op) || !write_op[ino])
108 		return -EINVAL;
109 
110 	data = simple_transaction_get(file, buf, size);
111 	if (IS_ERR(data))
112 		return PTR_ERR(data);
113 
114 	rv = write_op[ino](file, data, size);
115 	if (rv < 0)
116 		return rv;
117 
118 	simple_transaction_set(file, rv);
119 	return size;
120 }
121 
122 static ssize_t nfsctl_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
123 {
124 	if (! file->private_data) {
125 		/* An attempt to read a transaction file without writing
126 		 * causes a 0-byte write so that the file can return
127 		 * state information
128 		 */
129 		ssize_t rv = nfsctl_transaction_write(file, buf, 0, pos);
130 		if (rv < 0)
131 			return rv;
132 	}
133 	return simple_transaction_read(file, buf, size, pos);
134 }
135 
136 static const struct file_operations transaction_ops = {
137 	.write		= nfsctl_transaction_write,
138 	.read		= nfsctl_transaction_read,
139 	.release	= simple_transaction_release,
140 	.llseek		= default_llseek,
141 };
142 
143 static int exports_net_open(struct net *net, struct file *file)
144 {
145 	int err;
146 	struct seq_file *seq;
147 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
148 
149 	err = seq_open(file, &nfs_exports_op);
150 	if (err)
151 		return err;
152 
153 	seq = file->private_data;
154 	seq->private = nn->svc_export_cache;
155 	return 0;
156 }
157 
158 static int exports_nfsd_open(struct inode *inode, struct file *file)
159 {
160 	return exports_net_open(inode->i_sb->s_fs_info, file);
161 }
162 
163 static const struct file_operations exports_nfsd_operations = {
164 	.open		= exports_nfsd_open,
165 	.read		= seq_read,
166 	.llseek		= seq_lseek,
167 	.release	= seq_release,
168 };
169 
170 static int export_features_show(struct seq_file *m, void *v)
171 {
172 	seq_printf(m, "0x%x 0x%x\n", NFSEXP_ALLFLAGS, NFSEXP_SECINFO_FLAGS);
173 	return 0;
174 }
175 
176 DEFINE_SHOW_ATTRIBUTE(export_features);
177 
178 static const struct file_operations pool_stats_operations = {
179 	.open		= nfsd_pool_stats_open,
180 	.read		= seq_read,
181 	.llseek		= seq_lseek,
182 	.release	= nfsd_pool_stats_release,
183 };
184 
185 DEFINE_SHOW_ATTRIBUTE(nfsd_reply_cache_stats);
186 
187 DEFINE_SHOW_ATTRIBUTE(nfsd_file_cache_stats);
188 
189 /*----------------------------------------------------------------------------*/
190 /*
191  * payload - write methods
192  */
193 
194 static inline struct net *netns(struct file *file)
195 {
196 	return file_inode(file)->i_sb->s_fs_info;
197 }
198 
199 /*
200  * write_unlock_ip - Release all locks used by a client
201  *
202  * Experimental.
203  *
204  * Input:
205  *			buf:	'\n'-terminated C string containing a
206  *				presentation format IP address
207  *			size:	length of C string in @buf
208  * Output:
209  *	On success:	returns zero if all specified locks were released;
210  *			returns one if one or more locks were not released
211  *	On error:	return code is negative errno value
212  */
213 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size)
214 {
215 	struct sockaddr_storage address;
216 	struct sockaddr *sap = (struct sockaddr *)&address;
217 	size_t salen = sizeof(address);
218 	char *fo_path;
219 	struct net *net = netns(file);
220 
221 	/* sanity check */
222 	if (size == 0)
223 		return -EINVAL;
224 
225 	if (buf[size-1] != '\n')
226 		return -EINVAL;
227 
228 	fo_path = buf;
229 	if (qword_get(&buf, fo_path, size) < 0)
230 		return -EINVAL;
231 
232 	if (rpc_pton(net, fo_path, size, sap, salen) == 0)
233 		return -EINVAL;
234 
235 	trace_nfsd_ctl_unlock_ip(net, buf);
236 	return nlmsvc_unlock_all_by_ip(sap);
237 }
238 
239 /*
240  * write_unlock_fs - Release all locks on a local file system
241  *
242  * Experimental.
243  *
244  * Input:
245  *			buf:	'\n'-terminated C string containing the
246  *				absolute pathname of a local file system
247  *			size:	length of C string in @buf
248  * Output:
249  *	On success:	returns zero if all specified locks were released;
250  *			returns one if one or more locks were not released
251  *	On error:	return code is negative errno value
252  */
253 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size)
254 {
255 	struct path path;
256 	char *fo_path;
257 	int error;
258 
259 	/* sanity check */
260 	if (size == 0)
261 		return -EINVAL;
262 
263 	if (buf[size-1] != '\n')
264 		return -EINVAL;
265 
266 	fo_path = buf;
267 	if (qword_get(&buf, fo_path, size) < 0)
268 		return -EINVAL;
269 	trace_nfsd_ctl_unlock_fs(netns(file), fo_path);
270 	error = kern_path(fo_path, 0, &path);
271 	if (error)
272 		return error;
273 
274 	/*
275 	 * XXX: Needs better sanity checking.  Otherwise we could end up
276 	 * releasing locks on the wrong file system.
277 	 *
278 	 * For example:
279 	 * 1.  Does the path refer to a directory?
280 	 * 2.  Is that directory a mount point, or
281 	 * 3.  Is that directory the root of an exported file system?
282 	 */
283 	error = nlmsvc_unlock_all_by_sb(path.dentry->d_sb);
284 
285 	path_put(&path);
286 	return error;
287 }
288 
289 /*
290  * write_filehandle - Get a variable-length NFS file handle by path
291  *
292  * On input, the buffer contains a '\n'-terminated C string comprised of
293  * three alphanumeric words separated by whitespace.  The string may
294  * contain escape sequences.
295  *
296  * Input:
297  *			buf:
298  *				domain:		client domain name
299  *				path:		export pathname
300  *				maxsize:	numeric maximum size of
301  *						@buf
302  *			size:	length of C string in @buf
303  * Output:
304  *	On success:	passed-in buffer filled with '\n'-terminated C
305  *			string containing a ASCII hex text version
306  *			of the NFS file handle;
307  *			return code is the size in bytes of the string
308  *	On error:	return code is negative errno value
309  */
310 static ssize_t write_filehandle(struct file *file, char *buf, size_t size)
311 {
312 	char *dname, *path;
313 	int maxsize;
314 	char *mesg = buf;
315 	int len;
316 	struct auth_domain *dom;
317 	struct knfsd_fh fh;
318 
319 	if (size == 0)
320 		return -EINVAL;
321 
322 	if (buf[size-1] != '\n')
323 		return -EINVAL;
324 	buf[size-1] = 0;
325 
326 	dname = mesg;
327 	len = qword_get(&mesg, dname, size);
328 	if (len <= 0)
329 		return -EINVAL;
330 
331 	path = dname+len+1;
332 	len = qword_get(&mesg, path, size);
333 	if (len <= 0)
334 		return -EINVAL;
335 
336 	len = get_int(&mesg, &maxsize);
337 	if (len)
338 		return len;
339 
340 	if (maxsize < NFS_FHSIZE)
341 		return -EINVAL;
342 	maxsize = min(maxsize, NFS3_FHSIZE);
343 
344 	if (qword_get(&mesg, mesg, size) > 0)
345 		return -EINVAL;
346 
347 	trace_nfsd_ctl_filehandle(netns(file), dname, path, maxsize);
348 
349 	/* we have all the words, they are in buf.. */
350 	dom = unix_domain_find(dname);
351 	if (!dom)
352 		return -ENOMEM;
353 
354 	len = exp_rootfh(netns(file), dom, path, &fh, maxsize);
355 	auth_domain_put(dom);
356 	if (len)
357 		return len;
358 
359 	mesg = buf;
360 	len = SIMPLE_TRANSACTION_LIMIT;
361 	qword_addhex(&mesg, &len, fh.fh_raw, fh.fh_size);
362 	mesg[-1] = '\n';
363 	return mesg - buf;
364 }
365 
366 /*
367  * write_threads - Start NFSD, or report the current number of running threads
368  *
369  * Input:
370  *			buf:		ignored
371  *			size:		zero
372  * Output:
373  *	On success:	passed-in buffer filled with '\n'-terminated C
374  *			string numeric value representing the number of
375  *			running NFSD threads;
376  *			return code is the size in bytes of the string
377  *	On error:	return code is zero
378  *
379  * OR
380  *
381  * Input:
382  *			buf:		C string containing an unsigned
383  *					integer value representing the
384  *					number of NFSD threads to start
385  *			size:		non-zero length of C string in @buf
386  * Output:
387  *	On success:	NFS service is started;
388  *			passed-in buffer filled with '\n'-terminated C
389  *			string numeric value representing the number of
390  *			running NFSD threads;
391  *			return code is the size in bytes of the string
392  *	On error:	return code is zero or a negative errno value
393  */
394 static ssize_t write_threads(struct file *file, char *buf, size_t size)
395 {
396 	char *mesg = buf;
397 	int rv;
398 	struct net *net = netns(file);
399 
400 	if (size > 0) {
401 		int newthreads;
402 		rv = get_int(&mesg, &newthreads);
403 		if (rv)
404 			return rv;
405 		if (newthreads < 0)
406 			return -EINVAL;
407 		trace_nfsd_ctl_threads(net, newthreads);
408 		rv = nfsd_svc(newthreads, net, file->f_cred);
409 		if (rv < 0)
410 			return rv;
411 	} else
412 		rv = nfsd_nrthreads(net);
413 
414 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n", rv);
415 }
416 
417 /*
418  * write_pool_threads - Set or report the current number of threads per pool
419  *
420  * Input:
421  *			buf:		ignored
422  *			size:		zero
423  *
424  * OR
425  *
426  * Input:
427  *			buf:		C string containing whitespace-
428  *					separated unsigned integer values
429  *					representing the number of NFSD
430  *					threads to start in each pool
431  *			size:		non-zero length of C string in @buf
432  * Output:
433  *	On success:	passed-in buffer filled with '\n'-terminated C
434  *			string containing integer values representing the
435  *			number of NFSD threads in each pool;
436  *			return code is the size in bytes of the string
437  *	On error:	return code is zero or a negative errno value
438  */
439 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size)
440 {
441 	/* if size > 0, look for an array of number of threads per node
442 	 * and apply them  then write out number of threads per node as reply
443 	 */
444 	char *mesg = buf;
445 	int i;
446 	int rv;
447 	int len;
448 	int npools;
449 	int *nthreads;
450 	struct net *net = netns(file);
451 
452 	mutex_lock(&nfsd_mutex);
453 	npools = nfsd_nrpools(net);
454 	if (npools == 0) {
455 		/*
456 		 * NFS is shut down.  The admin can start it by
457 		 * writing to the threads file but NOT the pool_threads
458 		 * file, sorry.  Report zero threads.
459 		 */
460 		mutex_unlock(&nfsd_mutex);
461 		strcpy(buf, "0\n");
462 		return strlen(buf);
463 	}
464 
465 	nthreads = kcalloc(npools, sizeof(int), GFP_KERNEL);
466 	rv = -ENOMEM;
467 	if (nthreads == NULL)
468 		goto out_free;
469 
470 	if (size > 0) {
471 		for (i = 0; i < npools; i++) {
472 			rv = get_int(&mesg, &nthreads[i]);
473 			if (rv == -ENOENT)
474 				break;		/* fewer numbers than pools */
475 			if (rv)
476 				goto out_free;	/* syntax error */
477 			rv = -EINVAL;
478 			if (nthreads[i] < 0)
479 				goto out_free;
480 			trace_nfsd_ctl_pool_threads(net, i, nthreads[i]);
481 		}
482 		rv = nfsd_set_nrthreads(i, nthreads, net);
483 		if (rv)
484 			goto out_free;
485 	}
486 
487 	rv = nfsd_get_nrthreads(npools, nthreads, net);
488 	if (rv)
489 		goto out_free;
490 
491 	mesg = buf;
492 	size = SIMPLE_TRANSACTION_LIMIT;
493 	for (i = 0; i < npools && size > 0; i++) {
494 		snprintf(mesg, size, "%d%c", nthreads[i], (i == npools-1 ? '\n' : ' '));
495 		len = strlen(mesg);
496 		size -= len;
497 		mesg += len;
498 	}
499 	rv = mesg - buf;
500 out_free:
501 	kfree(nthreads);
502 	mutex_unlock(&nfsd_mutex);
503 	return rv;
504 }
505 
506 static ssize_t
507 nfsd_print_version_support(struct nfsd_net *nn, char *buf, int remaining,
508 		const char *sep, unsigned vers, int minor)
509 {
510 	const char *format = minor < 0 ? "%s%c%u" : "%s%c%u.%u";
511 	bool supported = !!nfsd_vers(nn, vers, NFSD_TEST);
512 
513 	if (vers == 4 && minor >= 0 &&
514 	    !nfsd_minorversion(nn, minor, NFSD_TEST))
515 		supported = false;
516 	if (minor == 0 && supported)
517 		/*
518 		 * special case for backward compatability.
519 		 * +4.0 is never reported, it is implied by
520 		 * +4, unless -4.0 is present.
521 		 */
522 		return 0;
523 	return snprintf(buf, remaining, format, sep,
524 			supported ? '+' : '-', vers, minor);
525 }
526 
527 static ssize_t __write_versions(struct file *file, char *buf, size_t size)
528 {
529 	char *mesg = buf;
530 	char *vers, *minorp, sign;
531 	int len, num, remaining;
532 	ssize_t tlen = 0;
533 	char *sep;
534 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
535 
536 	if (size > 0) {
537 		if (nn->nfsd_serv)
538 			/* Cannot change versions without updating
539 			 * nn->nfsd_serv->sv_xdrsize, and reallocing
540 			 * rq_argp and rq_resp
541 			 */
542 			return -EBUSY;
543 		if (buf[size-1] != '\n')
544 			return -EINVAL;
545 		buf[size-1] = 0;
546 		trace_nfsd_ctl_version(netns(file), buf);
547 
548 		vers = mesg;
549 		len = qword_get(&mesg, vers, size);
550 		if (len <= 0) return -EINVAL;
551 		do {
552 			enum vers_op cmd;
553 			unsigned minor;
554 			sign = *vers;
555 			if (sign == '+' || sign == '-')
556 				num = simple_strtol((vers+1), &minorp, 0);
557 			else
558 				num = simple_strtol(vers, &minorp, 0);
559 			if (*minorp == '.') {
560 				if (num != 4)
561 					return -EINVAL;
562 				if (kstrtouint(minorp+1, 0, &minor) < 0)
563 					return -EINVAL;
564 			}
565 
566 			cmd = sign == '-' ? NFSD_CLEAR : NFSD_SET;
567 			switch(num) {
568 #ifdef CONFIG_NFSD_V2
569 			case 2:
570 #endif
571 			case 3:
572 				nfsd_vers(nn, num, cmd);
573 				break;
574 			case 4:
575 				if (*minorp == '.') {
576 					if (nfsd_minorversion(nn, minor, cmd) < 0)
577 						return -EINVAL;
578 				} else if ((cmd == NFSD_SET) != nfsd_vers(nn, num, NFSD_TEST)) {
579 					/*
580 					 * Either we have +4 and no minors are enabled,
581 					 * or we have -4 and at least one minor is enabled.
582 					 * In either case, propagate 'cmd' to all minors.
583 					 */
584 					minor = 0;
585 					while (nfsd_minorversion(nn, minor, cmd) >= 0)
586 						minor++;
587 				}
588 				break;
589 			default:
590 				/* Ignore requests to disable non-existent versions */
591 				if (cmd == NFSD_SET)
592 					return -EINVAL;
593 			}
594 			vers += len + 1;
595 		} while ((len = qword_get(&mesg, vers, size)) > 0);
596 		/* If all get turned off, turn them back on, as
597 		 * having no versions is BAD
598 		 */
599 		nfsd_reset_versions(nn);
600 	}
601 
602 	/* Now write current state into reply buffer */
603 	sep = "";
604 	remaining = SIMPLE_TRANSACTION_LIMIT;
605 	for (num=2 ; num <= 4 ; num++) {
606 		int minor;
607 		if (!nfsd_vers(nn, num, NFSD_AVAIL))
608 			continue;
609 
610 		minor = -1;
611 		do {
612 			len = nfsd_print_version_support(nn, buf, remaining,
613 					sep, num, minor);
614 			if (len >= remaining)
615 				goto out;
616 			remaining -= len;
617 			buf += len;
618 			tlen += len;
619 			minor++;
620 			if (len)
621 				sep = " ";
622 		} while (num == 4 && minor <= NFSD_SUPPORTED_MINOR_VERSION);
623 	}
624 out:
625 	len = snprintf(buf, remaining, "\n");
626 	if (len >= remaining)
627 		return -EINVAL;
628 	return tlen + len;
629 }
630 
631 /*
632  * write_versions - Set or report the available NFS protocol versions
633  *
634  * Input:
635  *			buf:		ignored
636  *			size:		zero
637  * Output:
638  *	On success:	passed-in buffer filled with '\n'-terminated C
639  *			string containing positive or negative integer
640  *			values representing the current status of each
641  *			protocol version;
642  *			return code is the size in bytes of the string
643  *	On error:	return code is zero or a negative errno value
644  *
645  * OR
646  *
647  * Input:
648  *			buf:		C string containing whitespace-
649  *					separated positive or negative
650  *					integer values representing NFS
651  *					protocol versions to enable ("+n")
652  *					or disable ("-n")
653  *			size:		non-zero length of C string in @buf
654  * Output:
655  *	On success:	status of zero or more protocol versions has
656  *			been updated; passed-in buffer filled with
657  *			'\n'-terminated C string containing positive
658  *			or negative integer values representing the
659  *			current status of each protocol version;
660  *			return code is the size in bytes of the string
661  *	On error:	return code is zero or a negative errno value
662  */
663 static ssize_t write_versions(struct file *file, char *buf, size_t size)
664 {
665 	ssize_t rv;
666 
667 	mutex_lock(&nfsd_mutex);
668 	rv = __write_versions(file, buf, size);
669 	mutex_unlock(&nfsd_mutex);
670 	return rv;
671 }
672 
673 /*
674  * Zero-length write.  Return a list of NFSD's current listener
675  * transports.
676  */
677 static ssize_t __write_ports_names(char *buf, struct net *net)
678 {
679 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
680 
681 	if (nn->nfsd_serv == NULL)
682 		return 0;
683 	return svc_xprt_names(nn->nfsd_serv, buf, SIMPLE_TRANSACTION_LIMIT);
684 }
685 
686 /*
687  * A single 'fd' number was written, in which case it must be for
688  * a socket of a supported family/protocol, and we use it as an
689  * nfsd listener.
690  */
691 static ssize_t __write_ports_addfd(char *buf, struct net *net, const struct cred *cred)
692 {
693 	char *mesg = buf;
694 	int fd, err;
695 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
696 
697 	err = get_int(&mesg, &fd);
698 	if (err != 0 || fd < 0)
699 		return -EINVAL;
700 	trace_nfsd_ctl_ports_addfd(net, fd);
701 
702 	err = nfsd_create_serv(net);
703 	if (err != 0)
704 		return err;
705 
706 	err = svc_addsock(nn->nfsd_serv, net, fd, buf, SIMPLE_TRANSACTION_LIMIT, cred);
707 
708 	if (err >= 0 &&
709 	    !nn->nfsd_serv->sv_nrthreads && !xchg(&nn->keep_active, 1))
710 		svc_get(nn->nfsd_serv);
711 
712 	nfsd_put(net);
713 	return err;
714 }
715 
716 /*
717  * A transport listener is added by writing its transport name and
718  * a port number.
719  */
720 static ssize_t __write_ports_addxprt(char *buf, struct net *net, const struct cred *cred)
721 {
722 	char transport[16];
723 	struct svc_xprt *xprt;
724 	int port, err;
725 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
726 
727 	if (sscanf(buf, "%15s %5u", transport, &port) != 2)
728 		return -EINVAL;
729 
730 	if (port < 1 || port > USHRT_MAX)
731 		return -EINVAL;
732 	trace_nfsd_ctl_ports_addxprt(net, transport, port);
733 
734 	err = nfsd_create_serv(net);
735 	if (err != 0)
736 		return err;
737 
738 	err = svc_xprt_create(nn->nfsd_serv, transport, net,
739 			      PF_INET, port, SVC_SOCK_ANONYMOUS, cred);
740 	if (err < 0)
741 		goto out_err;
742 
743 	err = svc_xprt_create(nn->nfsd_serv, transport, net,
744 			      PF_INET6, port, SVC_SOCK_ANONYMOUS, cred);
745 	if (err < 0 && err != -EAFNOSUPPORT)
746 		goto out_close;
747 
748 	if (!nn->nfsd_serv->sv_nrthreads && !xchg(&nn->keep_active, 1))
749 		svc_get(nn->nfsd_serv);
750 
751 	nfsd_put(net);
752 	return 0;
753 out_close:
754 	xprt = svc_find_xprt(nn->nfsd_serv, transport, net, PF_INET, port);
755 	if (xprt != NULL) {
756 		svc_xprt_close(xprt);
757 		svc_xprt_put(xprt);
758 	}
759 out_err:
760 	nfsd_put(net);
761 	return err;
762 }
763 
764 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
765 			     struct net *net)
766 {
767 	if (size == 0)
768 		return __write_ports_names(buf, net);
769 
770 	if (isdigit(buf[0]))
771 		return __write_ports_addfd(buf, net, file->f_cred);
772 
773 	if (isalpha(buf[0]))
774 		return __write_ports_addxprt(buf, net, file->f_cred);
775 
776 	return -EINVAL;
777 }
778 
779 /*
780  * write_ports - Pass a socket file descriptor or transport name to listen on
781  *
782  * Input:
783  *			buf:		ignored
784  *			size:		zero
785  * Output:
786  *	On success:	passed-in buffer filled with a '\n'-terminated C
787  *			string containing a whitespace-separated list of
788  *			named NFSD listeners;
789  *			return code is the size in bytes of the string
790  *	On error:	return code is zero or a negative errno value
791  *
792  * OR
793  *
794  * Input:
795  *			buf:		C string containing an unsigned
796  *					integer value representing a bound
797  *					but unconnected socket that is to be
798  *					used as an NFSD listener; listen(3)
799  *					must be called for a SOCK_STREAM
800  *					socket, otherwise it is ignored
801  *			size:		non-zero length of C string in @buf
802  * Output:
803  *	On success:	NFS service is started;
804  *			passed-in buffer filled with a '\n'-terminated C
805  *			string containing a unique alphanumeric name of
806  *			the listener;
807  *			return code is the size in bytes of the string
808  *	On error:	return code is a negative errno value
809  *
810  * OR
811  *
812  * Input:
813  *			buf:		C string containing a transport
814  *					name and an unsigned integer value
815  *					representing the port to listen on,
816  *					separated by whitespace
817  *			size:		non-zero length of C string in @buf
818  * Output:
819  *	On success:	returns zero; NFS service is started
820  *	On error:	return code is a negative errno value
821  */
822 static ssize_t write_ports(struct file *file, char *buf, size_t size)
823 {
824 	ssize_t rv;
825 
826 	mutex_lock(&nfsd_mutex);
827 	rv = __write_ports(file, buf, size, netns(file));
828 	mutex_unlock(&nfsd_mutex);
829 	return rv;
830 }
831 
832 
833 int nfsd_max_blksize;
834 
835 /*
836  * write_maxblksize - Set or report the current NFS blksize
837  *
838  * Input:
839  *			buf:		ignored
840  *			size:		zero
841  *
842  * OR
843  *
844  * Input:
845  *			buf:		C string containing an unsigned
846  *					integer value representing the new
847  *					NFS blksize
848  *			size:		non-zero length of C string in @buf
849  * Output:
850  *	On success:	passed-in buffer filled with '\n'-terminated C string
851  *			containing numeric value of the current NFS blksize
852  *			setting;
853  *			return code is the size in bytes of the string
854  *	On error:	return code is zero or a negative errno value
855  */
856 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
857 {
858 	char *mesg = buf;
859 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
860 
861 	if (size > 0) {
862 		int bsize;
863 		int rv = get_int(&mesg, &bsize);
864 		if (rv)
865 			return rv;
866 		trace_nfsd_ctl_maxblksize(netns(file), bsize);
867 
868 		/* force bsize into allowed range and
869 		 * required alignment.
870 		 */
871 		bsize = max_t(int, bsize, 1024);
872 		bsize = min_t(int, bsize, NFSSVC_MAXBLKSIZE);
873 		bsize &= ~(1024-1);
874 		mutex_lock(&nfsd_mutex);
875 		if (nn->nfsd_serv) {
876 			mutex_unlock(&nfsd_mutex);
877 			return -EBUSY;
878 		}
879 		nfsd_max_blksize = bsize;
880 		mutex_unlock(&nfsd_mutex);
881 	}
882 
883 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
884 							nfsd_max_blksize);
885 }
886 
887 /*
888  * write_maxconn - Set or report the current max number of connections
889  *
890  * Input:
891  *			buf:		ignored
892  *			size:		zero
893  * OR
894  *
895  * Input:
896  *			buf:		C string containing an unsigned
897  *					integer value representing the new
898  *					number of max connections
899  *			size:		non-zero length of C string in @buf
900  * Output:
901  *	On success:	passed-in buffer filled with '\n'-terminated C string
902  *			containing numeric value of max_connections setting
903  *			for this net namespace;
904  *			return code is the size in bytes of the string
905  *	On error:	return code is zero or a negative errno value
906  */
907 static ssize_t write_maxconn(struct file *file, char *buf, size_t size)
908 {
909 	char *mesg = buf;
910 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
911 	unsigned int maxconn = nn->max_connections;
912 
913 	if (size > 0) {
914 		int rv = get_uint(&mesg, &maxconn);
915 
916 		if (rv)
917 			return rv;
918 		trace_nfsd_ctl_maxconn(netns(file), maxconn);
919 		nn->max_connections = maxconn;
920 	}
921 
922 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%u\n", maxconn);
923 }
924 
925 #ifdef CONFIG_NFSD_V4
926 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
927 				  time64_t *time, struct nfsd_net *nn)
928 {
929 	struct dentry *dentry = file_dentry(file);
930 	char *mesg = buf;
931 	int rv, i;
932 
933 	if (size > 0) {
934 		if (nn->nfsd_serv)
935 			return -EBUSY;
936 		rv = get_int(&mesg, &i);
937 		if (rv)
938 			return rv;
939 		trace_nfsd_ctl_time(netns(file), dentry->d_name.name,
940 				    dentry->d_name.len, i);
941 
942 		/*
943 		 * Some sanity checking.  We don't have a reason for
944 		 * these particular numbers, but problems with the
945 		 * extremes are:
946 		 *	- Too short: the briefest network outage may
947 		 *	  cause clients to lose all their locks.  Also,
948 		 *	  the frequent polling may be wasteful.
949 		 *	- Too long: do you really want reboot recovery
950 		 *	  to take more than an hour?  Or to make other
951 		 *	  clients wait an hour before being able to
952 		 *	  revoke a dead client's locks?
953 		 */
954 		if (i < 10 || i > 3600)
955 			return -EINVAL;
956 		*time = i;
957 	}
958 
959 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%lld\n", *time);
960 }
961 
962 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
963 				time64_t *time, struct nfsd_net *nn)
964 {
965 	ssize_t rv;
966 
967 	mutex_lock(&nfsd_mutex);
968 	rv = __nfsd4_write_time(file, buf, size, time, nn);
969 	mutex_unlock(&nfsd_mutex);
970 	return rv;
971 }
972 
973 /*
974  * write_leasetime - Set or report the current NFSv4 lease time
975  *
976  * Input:
977  *			buf:		ignored
978  *			size:		zero
979  *
980  * OR
981  *
982  * Input:
983  *			buf:		C string containing an unsigned
984  *					integer value representing the new
985  *					NFSv4 lease expiry time
986  *			size:		non-zero length of C string in @buf
987  * Output:
988  *	On success:	passed-in buffer filled with '\n'-terminated C
989  *			string containing unsigned integer value of the
990  *			current lease expiry time;
991  *			return code is the size in bytes of the string
992  *	On error:	return code is zero or a negative errno value
993  */
994 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
995 {
996 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
997 	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
998 }
999 
1000 /*
1001  * write_gracetime - Set or report current NFSv4 grace period time
1002  *
1003  * As above, but sets the time of the NFSv4 grace period.
1004  *
1005  * Note this should never be set to less than the *previous*
1006  * lease-period time, but we don't try to enforce this.  (In the common
1007  * case (a new boot), we don't know what the previous lease time was
1008  * anyway.)
1009  */
1010 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
1011 {
1012 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1013 	return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
1014 }
1015 
1016 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
1017 				   struct nfsd_net *nn)
1018 {
1019 	char *mesg = buf;
1020 	char *recdir;
1021 	int len, status;
1022 
1023 	if (size > 0) {
1024 		if (nn->nfsd_serv)
1025 			return -EBUSY;
1026 		if (size > PATH_MAX || buf[size-1] != '\n')
1027 			return -EINVAL;
1028 		buf[size-1] = 0;
1029 
1030 		recdir = mesg;
1031 		len = qword_get(&mesg, recdir, size);
1032 		if (len <= 0)
1033 			return -EINVAL;
1034 		trace_nfsd_ctl_recoverydir(netns(file), recdir);
1035 
1036 		status = nfs4_reset_recoverydir(recdir);
1037 		if (status)
1038 			return status;
1039 	}
1040 
1041 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1042 							nfs4_recoverydir());
1043 }
1044 
1045 /*
1046  * write_recoverydir - Set or report the pathname of the recovery directory
1047  *
1048  * Input:
1049  *			buf:		ignored
1050  *			size:		zero
1051  *
1052  * OR
1053  *
1054  * Input:
1055  *			buf:		C string containing the pathname
1056  *					of the directory on a local file
1057  *					system containing permanent NFSv4
1058  *					recovery data
1059  *			size:		non-zero length of C string in @buf
1060  * Output:
1061  *	On success:	passed-in buffer filled with '\n'-terminated C string
1062  *			containing the current recovery pathname setting;
1063  *			return code is the size in bytes of the string
1064  *	On error:	return code is zero or a negative errno value
1065  */
1066 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1067 {
1068 	ssize_t rv;
1069 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1070 
1071 	mutex_lock(&nfsd_mutex);
1072 	rv = __write_recoverydir(file, buf, size, nn);
1073 	mutex_unlock(&nfsd_mutex);
1074 	return rv;
1075 }
1076 
1077 /*
1078  * write_v4_end_grace - release grace period for nfsd's v4.x lock manager
1079  *
1080  * Input:
1081  *			buf:		ignored
1082  *			size:		zero
1083  * OR
1084  *
1085  * Input:
1086  *			buf:		any value
1087  *			size:		non-zero length of C string in @buf
1088  * Output:
1089  *			passed-in buffer filled with "Y" or "N" with a newline
1090  *			and NULL-terminated C string. This indicates whether
1091  *			the grace period has ended in the current net
1092  *			namespace. Return code is the size in bytes of the
1093  *			string. Writing a string that starts with 'Y', 'y', or
1094  *			'1' to the file will end the grace period for nfsd's v4
1095  *			lock manager.
1096  */
1097 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size)
1098 {
1099 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1100 
1101 	if (size > 0) {
1102 		switch(buf[0]) {
1103 		case 'Y':
1104 		case 'y':
1105 		case '1':
1106 			if (!nn->nfsd_serv)
1107 				return -EBUSY;
1108 			trace_nfsd_end_grace(netns(file));
1109 			nfsd4_end_grace(nn);
1110 			break;
1111 		default:
1112 			return -EINVAL;
1113 		}
1114 	}
1115 
1116 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%c\n",
1117 			 nn->grace_ended ? 'Y' : 'N');
1118 }
1119 
1120 #endif
1121 
1122 /*----------------------------------------------------------------------------*/
1123 /*
1124  *	populating the filesystem.
1125  */
1126 
1127 /* Basically copying rpc_get_inode. */
1128 static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
1129 {
1130 	struct inode *inode = new_inode(sb);
1131 	if (!inode)
1132 		return NULL;
1133 	/* Following advice from simple_fill_super documentation: */
1134 	inode->i_ino = iunique(sb, NFSD_MaxReserved);
1135 	inode->i_mode = mode;
1136 	simple_inode_init_ts(inode);
1137 	switch (mode & S_IFMT) {
1138 	case S_IFDIR:
1139 		inode->i_fop = &simple_dir_operations;
1140 		inode->i_op = &simple_dir_inode_operations;
1141 		inc_nlink(inode);
1142 		break;
1143 	case S_IFLNK:
1144 		inode->i_op = &simple_symlink_inode_operations;
1145 		break;
1146 	default:
1147 		break;
1148 	}
1149 	return inode;
1150 }
1151 
1152 static int __nfsd_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode, struct nfsdfs_client *ncl)
1153 {
1154 	struct inode *inode;
1155 
1156 	inode = nfsd_get_inode(dir->i_sb, mode);
1157 	if (!inode)
1158 		return -ENOMEM;
1159 	if (ncl) {
1160 		inode->i_private = ncl;
1161 		kref_get(&ncl->cl_ref);
1162 	}
1163 	d_add(dentry, inode);
1164 	inc_nlink(dir);
1165 	fsnotify_mkdir(dir, dentry);
1166 	return 0;
1167 }
1168 
1169 static struct dentry *nfsd_mkdir(struct dentry *parent, struct nfsdfs_client *ncl, char *name)
1170 {
1171 	struct inode *dir = parent->d_inode;
1172 	struct dentry *dentry;
1173 	int ret = -ENOMEM;
1174 
1175 	inode_lock(dir);
1176 	dentry = d_alloc_name(parent, name);
1177 	if (!dentry)
1178 		goto out_err;
1179 	ret = __nfsd_mkdir(d_inode(parent), dentry, S_IFDIR | 0600, ncl);
1180 	if (ret)
1181 		goto out_err;
1182 out:
1183 	inode_unlock(dir);
1184 	return dentry;
1185 out_err:
1186 	dput(dentry);
1187 	dentry = ERR_PTR(ret);
1188 	goto out;
1189 }
1190 
1191 #if IS_ENABLED(CONFIG_SUNRPC_GSS)
1192 static int __nfsd_symlink(struct inode *dir, struct dentry *dentry,
1193 			  umode_t mode, const char *content)
1194 {
1195 	struct inode *inode;
1196 
1197 	inode = nfsd_get_inode(dir->i_sb, mode);
1198 	if (!inode)
1199 		return -ENOMEM;
1200 
1201 	inode->i_link = (char *)content;
1202 	inode->i_size = strlen(content);
1203 
1204 	d_add(dentry, inode);
1205 	inc_nlink(dir);
1206 	fsnotify_create(dir, dentry);
1207 	return 0;
1208 }
1209 
1210 /*
1211  * @content is assumed to be a NUL-terminated string that lives
1212  * longer than the symlink itself.
1213  */
1214 static void _nfsd_symlink(struct dentry *parent, const char *name,
1215 			  const char *content)
1216 {
1217 	struct inode *dir = parent->d_inode;
1218 	struct dentry *dentry;
1219 	int ret;
1220 
1221 	inode_lock(dir);
1222 	dentry = d_alloc_name(parent, name);
1223 	if (!dentry)
1224 		goto out;
1225 	ret = __nfsd_symlink(d_inode(parent), dentry, S_IFLNK | 0777, content);
1226 	if (ret)
1227 		dput(dentry);
1228 out:
1229 	inode_unlock(dir);
1230 }
1231 #else
1232 static inline void _nfsd_symlink(struct dentry *parent, const char *name,
1233 				 const char *content)
1234 {
1235 }
1236 
1237 #endif
1238 
1239 static void clear_ncl(struct inode *inode)
1240 {
1241 	struct nfsdfs_client *ncl = inode->i_private;
1242 
1243 	inode->i_private = NULL;
1244 	kref_put(&ncl->cl_ref, ncl->cl_release);
1245 }
1246 
1247 static struct nfsdfs_client *__get_nfsdfs_client(struct inode *inode)
1248 {
1249 	struct nfsdfs_client *nc = inode->i_private;
1250 
1251 	if (nc)
1252 		kref_get(&nc->cl_ref);
1253 	return nc;
1254 }
1255 
1256 struct nfsdfs_client *get_nfsdfs_client(struct inode *inode)
1257 {
1258 	struct nfsdfs_client *nc;
1259 
1260 	inode_lock_shared(inode);
1261 	nc = __get_nfsdfs_client(inode);
1262 	inode_unlock_shared(inode);
1263 	return nc;
1264 }
1265 /* from __rpc_unlink */
1266 static void nfsdfs_remove_file(struct inode *dir, struct dentry *dentry)
1267 {
1268 	int ret;
1269 
1270 	clear_ncl(d_inode(dentry));
1271 	dget(dentry);
1272 	ret = simple_unlink(dir, dentry);
1273 	d_drop(dentry);
1274 	fsnotify_unlink(dir, dentry);
1275 	dput(dentry);
1276 	WARN_ON_ONCE(ret);
1277 }
1278 
1279 static void nfsdfs_remove_files(struct dentry *root)
1280 {
1281 	struct dentry *dentry, *tmp;
1282 
1283 	list_for_each_entry_safe(dentry, tmp, &root->d_subdirs, d_child) {
1284 		if (!simple_positive(dentry)) {
1285 			WARN_ON_ONCE(1); /* I think this can't happen? */
1286 			continue;
1287 		}
1288 		nfsdfs_remove_file(d_inode(root), dentry);
1289 	}
1290 }
1291 
1292 /* XXX: cut'n'paste from simple_fill_super; figure out if we could share
1293  * code instead. */
1294 static  int nfsdfs_create_files(struct dentry *root,
1295 				const struct tree_descr *files,
1296 				struct dentry **fdentries)
1297 {
1298 	struct inode *dir = d_inode(root);
1299 	struct inode *inode;
1300 	struct dentry *dentry;
1301 	int i;
1302 
1303 	inode_lock(dir);
1304 	for (i = 0; files->name && files->name[0]; i++, files++) {
1305 		dentry = d_alloc_name(root, files->name);
1306 		if (!dentry)
1307 			goto out;
1308 		inode = nfsd_get_inode(d_inode(root)->i_sb,
1309 					S_IFREG | files->mode);
1310 		if (!inode) {
1311 			dput(dentry);
1312 			goto out;
1313 		}
1314 		inode->i_fop = files->ops;
1315 		inode->i_private = __get_nfsdfs_client(dir);
1316 		d_add(dentry, inode);
1317 		fsnotify_create(dir, dentry);
1318 		if (fdentries)
1319 			fdentries[i] = dentry;
1320 	}
1321 	inode_unlock(dir);
1322 	return 0;
1323 out:
1324 	nfsdfs_remove_files(root);
1325 	inode_unlock(dir);
1326 	return -ENOMEM;
1327 }
1328 
1329 /* on success, returns positive number unique to that client. */
1330 struct dentry *nfsd_client_mkdir(struct nfsd_net *nn,
1331 				 struct nfsdfs_client *ncl, u32 id,
1332 				 const struct tree_descr *files,
1333 				 struct dentry **fdentries)
1334 {
1335 	struct dentry *dentry;
1336 	char name[11];
1337 	int ret;
1338 
1339 	sprintf(name, "%u", id);
1340 
1341 	dentry = nfsd_mkdir(nn->nfsd_client_dir, ncl, name);
1342 	if (IS_ERR(dentry)) /* XXX: tossing errors? */
1343 		return NULL;
1344 	ret = nfsdfs_create_files(dentry, files, fdentries);
1345 	if (ret) {
1346 		nfsd_client_rmdir(dentry);
1347 		return NULL;
1348 	}
1349 	return dentry;
1350 }
1351 
1352 /* Taken from __rpc_rmdir: */
1353 void nfsd_client_rmdir(struct dentry *dentry)
1354 {
1355 	struct inode *dir = d_inode(dentry->d_parent);
1356 	struct inode *inode = d_inode(dentry);
1357 	int ret;
1358 
1359 	inode_lock(dir);
1360 	nfsdfs_remove_files(dentry);
1361 	clear_ncl(inode);
1362 	dget(dentry);
1363 	ret = simple_rmdir(dir, dentry);
1364 	WARN_ON_ONCE(ret);
1365 	d_drop(dentry);
1366 	fsnotify_rmdir(dir, dentry);
1367 	dput(dentry);
1368 	inode_unlock(dir);
1369 }
1370 
1371 static int nfsd_fill_super(struct super_block *sb, struct fs_context *fc)
1372 {
1373 	struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
1374 							nfsd_net_id);
1375 	struct dentry *dentry;
1376 	int ret;
1377 
1378 	static const struct tree_descr nfsd_files[] = {
1379 		[NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1380 		/* Per-export io stats use same ops as exports file */
1381 		[NFSD_Export_Stats] = {"export_stats", &exports_nfsd_operations, S_IRUGO},
1382 		[NFSD_Export_features] = {"export_features",
1383 					&export_features_fops, S_IRUGO},
1384 		[NFSD_FO_UnlockIP] = {"unlock_ip",
1385 					&transaction_ops, S_IWUSR|S_IRUSR},
1386 		[NFSD_FO_UnlockFS] = {"unlock_filesystem",
1387 					&transaction_ops, S_IWUSR|S_IRUSR},
1388 		[NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1389 		[NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1390 		[NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1391 		[NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1392 		[NFSD_Reply_Cache_Stats] = {"reply_cache_stats",
1393 					&nfsd_reply_cache_stats_fops, S_IRUGO},
1394 		[NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1395 		[NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1396 		[NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1397 		[NFSD_MaxConnections] = {"max_connections", &transaction_ops, S_IWUSR|S_IRUGO},
1398 		[NFSD_Filecache] = {"filecache", &nfsd_file_cache_stats_fops, S_IRUGO},
1399 #ifdef CONFIG_NFSD_V4
1400 		[NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1401 		[NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1402 		[NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1403 		[NFSD_V4EndGrace] = {"v4_end_grace", &transaction_ops, S_IWUSR|S_IRUGO},
1404 #endif
1405 		/* last one */ {""}
1406 	};
1407 
1408 	ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1409 	if (ret)
1410 		return ret;
1411 	_nfsd_symlink(sb->s_root, "supported_krb5_enctypes",
1412 		      "/proc/net/rpc/gss_krb5_enctypes");
1413 	dentry = nfsd_mkdir(sb->s_root, NULL, "clients");
1414 	if (IS_ERR(dentry))
1415 		return PTR_ERR(dentry);
1416 	nn->nfsd_client_dir = dentry;
1417 	return 0;
1418 }
1419 
1420 static int nfsd_fs_get_tree(struct fs_context *fc)
1421 {
1422 	return get_tree_keyed(fc, nfsd_fill_super, get_net(fc->net_ns));
1423 }
1424 
1425 static void nfsd_fs_free_fc(struct fs_context *fc)
1426 {
1427 	if (fc->s_fs_info)
1428 		put_net(fc->s_fs_info);
1429 }
1430 
1431 static const struct fs_context_operations nfsd_fs_context_ops = {
1432 	.free		= nfsd_fs_free_fc,
1433 	.get_tree	= nfsd_fs_get_tree,
1434 };
1435 
1436 static int nfsd_init_fs_context(struct fs_context *fc)
1437 {
1438 	put_user_ns(fc->user_ns);
1439 	fc->user_ns = get_user_ns(fc->net_ns->user_ns);
1440 	fc->ops = &nfsd_fs_context_ops;
1441 	return 0;
1442 }
1443 
1444 static void nfsd_umount(struct super_block *sb)
1445 {
1446 	struct net *net = sb->s_fs_info;
1447 
1448 	nfsd_shutdown_threads(net);
1449 
1450 	kill_litter_super(sb);
1451 	put_net(net);
1452 }
1453 
1454 static struct file_system_type nfsd_fs_type = {
1455 	.owner		= THIS_MODULE,
1456 	.name		= "nfsd",
1457 	.init_fs_context = nfsd_init_fs_context,
1458 	.kill_sb	= nfsd_umount,
1459 };
1460 MODULE_ALIAS_FS("nfsd");
1461 
1462 #ifdef CONFIG_PROC_FS
1463 
1464 static int exports_proc_open(struct inode *inode, struct file *file)
1465 {
1466 	return exports_net_open(current->nsproxy->net_ns, file);
1467 }
1468 
1469 static const struct proc_ops exports_proc_ops = {
1470 	.proc_open	= exports_proc_open,
1471 	.proc_read	= seq_read,
1472 	.proc_lseek	= seq_lseek,
1473 	.proc_release	= seq_release,
1474 };
1475 
1476 static int create_proc_exports_entry(void)
1477 {
1478 	struct proc_dir_entry *entry;
1479 
1480 	entry = proc_mkdir("fs/nfs", NULL);
1481 	if (!entry)
1482 		return -ENOMEM;
1483 	entry = proc_create("exports", 0, entry, &exports_proc_ops);
1484 	if (!entry) {
1485 		remove_proc_entry("fs/nfs", NULL);
1486 		return -ENOMEM;
1487 	}
1488 	return 0;
1489 }
1490 #else /* CONFIG_PROC_FS */
1491 static int create_proc_exports_entry(void)
1492 {
1493 	return 0;
1494 }
1495 #endif
1496 
1497 unsigned int nfsd_net_id;
1498 
1499 /**
1500  * nfsd_nl_rpc_status_get_start - Prepare rpc_status_get dumpit
1501  * @cb: netlink metadata and command arguments
1502  *
1503  * Return values:
1504  *   %0: The rpc_status_get command may proceed
1505  *   %-ENODEV: There is no NFSD running in this namespace
1506  */
1507 int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb)
1508 {
1509 	struct nfsd_net *nn = net_generic(sock_net(cb->skb->sk), nfsd_net_id);
1510 	int ret = -ENODEV;
1511 
1512 	mutex_lock(&nfsd_mutex);
1513 	if (nn->nfsd_serv) {
1514 		svc_get(nn->nfsd_serv);
1515 		ret = 0;
1516 	}
1517 	mutex_unlock(&nfsd_mutex);
1518 
1519 	return ret;
1520 }
1521 
1522 static int nfsd_genl_rpc_status_compose_msg(struct sk_buff *skb,
1523 					    struct netlink_callback *cb,
1524 					    struct nfsd_genl_rqstp *rqstp)
1525 {
1526 	void *hdr;
1527 	u32 i;
1528 
1529 	hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1530 			  &nfsd_nl_family, 0, NFSD_CMD_RPC_STATUS_GET);
1531 	if (!hdr)
1532 		return -ENOBUFS;
1533 
1534 	if (nla_put_be32(skb, NFSD_A_RPC_STATUS_XID, rqstp->rq_xid) ||
1535 	    nla_put_u32(skb, NFSD_A_RPC_STATUS_FLAGS, rqstp->rq_flags) ||
1536 	    nla_put_u32(skb, NFSD_A_RPC_STATUS_PROG, rqstp->rq_prog) ||
1537 	    nla_put_u32(skb, NFSD_A_RPC_STATUS_PROC, rqstp->rq_proc) ||
1538 	    nla_put_u8(skb, NFSD_A_RPC_STATUS_VERSION, rqstp->rq_vers) ||
1539 	    nla_put_s64(skb, NFSD_A_RPC_STATUS_SERVICE_TIME,
1540 			ktime_to_us(rqstp->rq_stime),
1541 			NFSD_A_RPC_STATUS_PAD))
1542 		return -ENOBUFS;
1543 
1544 	switch (rqstp->rq_saddr.sa_family) {
1545 	case AF_INET: {
1546 		const struct sockaddr_in *s_in, *d_in;
1547 
1548 		s_in = (const struct sockaddr_in *)&rqstp->rq_saddr;
1549 		d_in = (const struct sockaddr_in *)&rqstp->rq_daddr;
1550 		if (nla_put_in_addr(skb, NFSD_A_RPC_STATUS_SADDR4,
1551 				    s_in->sin_addr.s_addr) ||
1552 		    nla_put_in_addr(skb, NFSD_A_RPC_STATUS_DADDR4,
1553 				    d_in->sin_addr.s_addr) ||
1554 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1555 				 s_in->sin_port) ||
1556 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1557 				 d_in->sin_port))
1558 			return -ENOBUFS;
1559 		break;
1560 	}
1561 	case AF_INET6: {
1562 		const struct sockaddr_in6 *s_in, *d_in;
1563 
1564 		s_in = (const struct sockaddr_in6 *)&rqstp->rq_saddr;
1565 		d_in = (const struct sockaddr_in6 *)&rqstp->rq_daddr;
1566 		if (nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_SADDR6,
1567 				     &s_in->sin6_addr) ||
1568 		    nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_DADDR6,
1569 				     &d_in->sin6_addr) ||
1570 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1571 				 s_in->sin6_port) ||
1572 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1573 				 d_in->sin6_port))
1574 			return -ENOBUFS;
1575 		break;
1576 	}
1577 	}
1578 
1579 	for (i = 0; i < rqstp->rq_opcnt; i++)
1580 		if (nla_put_u32(skb, NFSD_A_RPC_STATUS_COMPOUND_OPS,
1581 				rqstp->rq_opnum[i]))
1582 			return -ENOBUFS;
1583 
1584 	genlmsg_end(skb, hdr);
1585 	return 0;
1586 }
1587 
1588 /**
1589  * nfsd_nl_rpc_status_get_dumpit - Handle rpc_status_get dumpit
1590  * @skb: reply buffer
1591  * @cb: netlink metadata and command arguments
1592  *
1593  * Returns the size of the reply or a negative errno.
1594  */
1595 int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
1596 				  struct netlink_callback *cb)
1597 {
1598 	struct nfsd_net *nn = net_generic(sock_net(skb->sk), nfsd_net_id);
1599 	int i, ret, rqstp_index = 0;
1600 
1601 	rcu_read_lock();
1602 
1603 	for (i = 0; i < nn->nfsd_serv->sv_nrpools; i++) {
1604 		struct svc_rqst *rqstp;
1605 
1606 		if (i < cb->args[0]) /* already consumed */
1607 			continue;
1608 
1609 		rqstp_index = 0;
1610 		list_for_each_entry_rcu(rqstp,
1611 				&nn->nfsd_serv->sv_pools[i].sp_all_threads,
1612 				rq_all) {
1613 			struct nfsd_genl_rqstp genl_rqstp;
1614 			unsigned int status_counter;
1615 
1616 			if (rqstp_index++ < cb->args[1]) /* already consumed */
1617 				continue;
1618 			/*
1619 			 * Acquire rq_status_counter before parsing the rqst
1620 			 * fields. rq_status_counter is set to an odd value in
1621 			 * order to notify the consumers the rqstp fields are
1622 			 * meaningful.
1623 			 */
1624 			status_counter =
1625 				smp_load_acquire(&rqstp->rq_status_counter);
1626 			if (!(status_counter & 1))
1627 				continue;
1628 
1629 			genl_rqstp.rq_xid = rqstp->rq_xid;
1630 			genl_rqstp.rq_flags = rqstp->rq_flags;
1631 			genl_rqstp.rq_vers = rqstp->rq_vers;
1632 			genl_rqstp.rq_prog = rqstp->rq_prog;
1633 			genl_rqstp.rq_proc = rqstp->rq_proc;
1634 			genl_rqstp.rq_stime = rqstp->rq_stime;
1635 			genl_rqstp.rq_opcnt = 0;
1636 			memcpy(&genl_rqstp.rq_daddr, svc_daddr(rqstp),
1637 			       sizeof(struct sockaddr));
1638 			memcpy(&genl_rqstp.rq_saddr, svc_addr(rqstp),
1639 			       sizeof(struct sockaddr));
1640 
1641 #ifdef CONFIG_NFSD_V4
1642 			if (rqstp->rq_vers == NFS4_VERSION &&
1643 			    rqstp->rq_proc == NFSPROC4_COMPOUND) {
1644 				/* NFSv4 compound */
1645 				struct nfsd4_compoundargs *args;
1646 				int j;
1647 
1648 				args = rqstp->rq_argp;
1649 				genl_rqstp.rq_opcnt = args->opcnt;
1650 				for (j = 0; j < genl_rqstp.rq_opcnt; j++)
1651 					genl_rqstp.rq_opnum[j] =
1652 						args->ops[j].opnum;
1653 			}
1654 #endif /* CONFIG_NFSD_V4 */
1655 
1656 			/*
1657 			 * Acquire rq_status_counter before reporting the rqst
1658 			 * fields to the user.
1659 			 */
1660 			if (smp_load_acquire(&rqstp->rq_status_counter) !=
1661 			    status_counter)
1662 				continue;
1663 
1664 			ret = nfsd_genl_rpc_status_compose_msg(skb, cb,
1665 							       &genl_rqstp);
1666 			if (ret)
1667 				goto out;
1668 		}
1669 	}
1670 
1671 	cb->args[0] = i;
1672 	cb->args[1] = rqstp_index;
1673 	ret = skb->len;
1674 out:
1675 	rcu_read_unlock();
1676 
1677 	return ret;
1678 }
1679 
1680 /**
1681  * nfsd_nl_rpc_status_get_done - rpc_status_get dumpit post-processing
1682  * @cb: netlink metadata and command arguments
1683  *
1684  * Return values:
1685  *   %0: Success
1686  */
1687 int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb)
1688 {
1689 	mutex_lock(&nfsd_mutex);
1690 	nfsd_put(sock_net(cb->skb->sk));
1691 	mutex_unlock(&nfsd_mutex);
1692 
1693 	return 0;
1694 }
1695 
1696 /**
1697  * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
1698  * @net: a freshly-created network namespace
1699  *
1700  * This information stays around as long as the network namespace is
1701  * alive whether or not there is an NFSD instance running in the
1702  * namespace.
1703  *
1704  * Returns zero on success, or a negative errno otherwise.
1705  */
1706 static __net_init int nfsd_net_init(struct net *net)
1707 {
1708 	int retval;
1709 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1710 
1711 	retval = nfsd_export_init(net);
1712 	if (retval)
1713 		goto out_export_error;
1714 	retval = nfsd_idmap_init(net);
1715 	if (retval)
1716 		goto out_idmap_error;
1717 	retval = nfsd_net_reply_cache_init(nn);
1718 	if (retval)
1719 		goto out_repcache_error;
1720 	nn->nfsd_versions = NULL;
1721 	nn->nfsd4_minorversions = NULL;
1722 	nfsd4_init_leases_net(nn);
1723 	get_random_bytes(&nn->siphash_key, sizeof(nn->siphash_key));
1724 	seqlock_init(&nn->writeverf_lock);
1725 
1726 	return 0;
1727 
1728 out_repcache_error:
1729 	nfsd_idmap_shutdown(net);
1730 out_idmap_error:
1731 	nfsd_export_shutdown(net);
1732 out_export_error:
1733 	return retval;
1734 }
1735 
1736 /**
1737  * nfsd_net_exit - Release the nfsd_net portion of a net namespace
1738  * @net: a network namespace that is about to be destroyed
1739  *
1740  */
1741 static __net_exit void nfsd_net_exit(struct net *net)
1742 {
1743 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1744 
1745 	nfsd_net_reply_cache_destroy(nn);
1746 	nfsd_idmap_shutdown(net);
1747 	nfsd_export_shutdown(net);
1748 	nfsd_netns_free_versions(nn);
1749 }
1750 
1751 static struct pernet_operations nfsd_net_ops = {
1752 	.init = nfsd_net_init,
1753 	.exit = nfsd_net_exit,
1754 	.id   = &nfsd_net_id,
1755 	.size = sizeof(struct nfsd_net),
1756 };
1757 
1758 static int __init init_nfsd(void)
1759 {
1760 	int retval;
1761 
1762 	retval = nfsd4_init_slabs();
1763 	if (retval)
1764 		return retval;
1765 	retval = nfsd4_init_pnfs();
1766 	if (retval)
1767 		goto out_free_slabs;
1768 	retval = nfsd_stat_init();	/* Statistics */
1769 	if (retval)
1770 		goto out_free_pnfs;
1771 	retval = nfsd_drc_slab_create();
1772 	if (retval)
1773 		goto out_free_stat;
1774 	nfsd_lockd_init();	/* lockd->nfsd callbacks */
1775 	retval = create_proc_exports_entry();
1776 	if (retval)
1777 		goto out_free_lockd;
1778 	retval = register_pernet_subsys(&nfsd_net_ops);
1779 	if (retval < 0)
1780 		goto out_free_exports;
1781 	retval = register_cld_notifier();
1782 	if (retval)
1783 		goto out_free_subsys;
1784 	retval = nfsd4_create_laundry_wq();
1785 	if (retval)
1786 		goto out_free_cld;
1787 	retval = register_filesystem(&nfsd_fs_type);
1788 	if (retval)
1789 		goto out_free_all;
1790 	retval = genl_register_family(&nfsd_nl_family);
1791 	if (retval)
1792 		goto out_free_all;
1793 
1794 	return 0;
1795 out_free_all:
1796 	nfsd4_destroy_laundry_wq();
1797 out_free_cld:
1798 	unregister_cld_notifier();
1799 out_free_subsys:
1800 	unregister_pernet_subsys(&nfsd_net_ops);
1801 out_free_exports:
1802 	remove_proc_entry("fs/nfs/exports", NULL);
1803 	remove_proc_entry("fs/nfs", NULL);
1804 out_free_lockd:
1805 	nfsd_lockd_shutdown();
1806 	nfsd_drc_slab_free();
1807 out_free_stat:
1808 	nfsd_stat_shutdown();
1809 out_free_pnfs:
1810 	nfsd4_exit_pnfs();
1811 out_free_slabs:
1812 	nfsd4_free_slabs();
1813 	return retval;
1814 }
1815 
1816 static void __exit exit_nfsd(void)
1817 {
1818 	genl_unregister_family(&nfsd_nl_family);
1819 	unregister_filesystem(&nfsd_fs_type);
1820 	nfsd4_destroy_laundry_wq();
1821 	unregister_cld_notifier();
1822 	unregister_pernet_subsys(&nfsd_net_ops);
1823 	nfsd_drc_slab_free();
1824 	remove_proc_entry("fs/nfs/exports", NULL);
1825 	remove_proc_entry("fs/nfs", NULL);
1826 	nfsd_stat_shutdown();
1827 	nfsd_lockd_shutdown();
1828 	nfsd4_free_slabs();
1829 	nfsd4_exit_pnfs();
1830 }
1831 
1832 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1833 MODULE_DESCRIPTION("In-kernel NFS server");
1834 MODULE_LICENSE("GPL");
1835 module_init(init_nfsd)
1836 module_exit(exit_nfsd)
1837