xref: /linux/fs/nfsd/nfsctl.c (revision 993498e537af9260e697219ce41b41b22b6199cc)
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 && !nn->nfsd_serv->sv_nrthreads && !nn->keep_active)
709 		nfsd_last_thread(net);
710 	else if (err >= 0 &&
711 		 !nn->nfsd_serv->sv_nrthreads && !xchg(&nn->keep_active, 1))
712 		svc_get(nn->nfsd_serv);
713 
714 	nfsd_put(net);
715 	return err;
716 }
717 
718 /*
719  * A transport listener is added by writing its transport name and
720  * a port number.
721  */
722 static ssize_t __write_ports_addxprt(char *buf, struct net *net, const struct cred *cred)
723 {
724 	char transport[16];
725 	struct svc_xprt *xprt;
726 	int port, err;
727 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
728 
729 	if (sscanf(buf, "%15s %5u", transport, &port) != 2)
730 		return -EINVAL;
731 
732 	if (port < 1 || port > USHRT_MAX)
733 		return -EINVAL;
734 	trace_nfsd_ctl_ports_addxprt(net, transport, port);
735 
736 	err = nfsd_create_serv(net);
737 	if (err != 0)
738 		return err;
739 
740 	err = svc_xprt_create(nn->nfsd_serv, transport, net,
741 			      PF_INET, port, SVC_SOCK_ANONYMOUS, cred);
742 	if (err < 0)
743 		goto out_err;
744 
745 	err = svc_xprt_create(nn->nfsd_serv, transport, net,
746 			      PF_INET6, port, SVC_SOCK_ANONYMOUS, cred);
747 	if (err < 0 && err != -EAFNOSUPPORT)
748 		goto out_close;
749 
750 	if (!nn->nfsd_serv->sv_nrthreads && !xchg(&nn->keep_active, 1))
751 		svc_get(nn->nfsd_serv);
752 
753 	nfsd_put(net);
754 	return 0;
755 out_close:
756 	xprt = svc_find_xprt(nn->nfsd_serv, transport, net, PF_INET, port);
757 	if (xprt != NULL) {
758 		svc_xprt_close(xprt);
759 		svc_xprt_put(xprt);
760 	}
761 out_err:
762 	if (!nn->nfsd_serv->sv_nrthreads && !nn->keep_active)
763 		nfsd_last_thread(net);
764 
765 	nfsd_put(net);
766 	return err;
767 }
768 
769 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
770 			     struct net *net)
771 {
772 	if (size == 0)
773 		return __write_ports_names(buf, net);
774 
775 	if (isdigit(buf[0]))
776 		return __write_ports_addfd(buf, net, file->f_cred);
777 
778 	if (isalpha(buf[0]))
779 		return __write_ports_addxprt(buf, net, file->f_cred);
780 
781 	return -EINVAL;
782 }
783 
784 /*
785  * write_ports - Pass a socket file descriptor or transport name to listen on
786  *
787  * Input:
788  *			buf:		ignored
789  *			size:		zero
790  * Output:
791  *	On success:	passed-in buffer filled with a '\n'-terminated C
792  *			string containing a whitespace-separated list of
793  *			named NFSD listeners;
794  *			return code is the size in bytes of the string
795  *	On error:	return code is zero or a negative errno value
796  *
797  * OR
798  *
799  * Input:
800  *			buf:		C string containing an unsigned
801  *					integer value representing a bound
802  *					but unconnected socket that is to be
803  *					used as an NFSD listener; listen(3)
804  *					must be called for a SOCK_STREAM
805  *					socket, otherwise it is ignored
806  *			size:		non-zero length of C string in @buf
807  * Output:
808  *	On success:	NFS service is started;
809  *			passed-in buffer filled with a '\n'-terminated C
810  *			string containing a unique alphanumeric name of
811  *			the listener;
812  *			return code is the size in bytes of the string
813  *	On error:	return code is a negative errno value
814  *
815  * OR
816  *
817  * Input:
818  *			buf:		C string containing a transport
819  *					name and an unsigned integer value
820  *					representing the port to listen on,
821  *					separated by whitespace
822  *			size:		non-zero length of C string in @buf
823  * Output:
824  *	On success:	returns zero; NFS service is started
825  *	On error:	return code is a negative errno value
826  */
827 static ssize_t write_ports(struct file *file, char *buf, size_t size)
828 {
829 	ssize_t rv;
830 
831 	mutex_lock(&nfsd_mutex);
832 	rv = __write_ports(file, buf, size, netns(file));
833 	mutex_unlock(&nfsd_mutex);
834 	return rv;
835 }
836 
837 
838 int nfsd_max_blksize;
839 
840 /*
841  * write_maxblksize - Set or report the current NFS blksize
842  *
843  * Input:
844  *			buf:		ignored
845  *			size:		zero
846  *
847  * OR
848  *
849  * Input:
850  *			buf:		C string containing an unsigned
851  *					integer value representing the new
852  *					NFS blksize
853  *			size:		non-zero length of C string in @buf
854  * Output:
855  *	On success:	passed-in buffer filled with '\n'-terminated C string
856  *			containing numeric value of the current NFS blksize
857  *			setting;
858  *			return code is the size in bytes of the string
859  *	On error:	return code is zero or a negative errno value
860  */
861 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
862 {
863 	char *mesg = buf;
864 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
865 
866 	if (size > 0) {
867 		int bsize;
868 		int rv = get_int(&mesg, &bsize);
869 		if (rv)
870 			return rv;
871 		trace_nfsd_ctl_maxblksize(netns(file), bsize);
872 
873 		/* force bsize into allowed range and
874 		 * required alignment.
875 		 */
876 		bsize = max_t(int, bsize, 1024);
877 		bsize = min_t(int, bsize, NFSSVC_MAXBLKSIZE);
878 		bsize &= ~(1024-1);
879 		mutex_lock(&nfsd_mutex);
880 		if (nn->nfsd_serv) {
881 			mutex_unlock(&nfsd_mutex);
882 			return -EBUSY;
883 		}
884 		nfsd_max_blksize = bsize;
885 		mutex_unlock(&nfsd_mutex);
886 	}
887 
888 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
889 							nfsd_max_blksize);
890 }
891 
892 /*
893  * write_maxconn - Set or report the current max number of connections
894  *
895  * Input:
896  *			buf:		ignored
897  *			size:		zero
898  * OR
899  *
900  * Input:
901  *			buf:		C string containing an unsigned
902  *					integer value representing the new
903  *					number of max connections
904  *			size:		non-zero length of C string in @buf
905  * Output:
906  *	On success:	passed-in buffer filled with '\n'-terminated C string
907  *			containing numeric value of max_connections setting
908  *			for this net namespace;
909  *			return code is the size in bytes of the string
910  *	On error:	return code is zero or a negative errno value
911  */
912 static ssize_t write_maxconn(struct file *file, char *buf, size_t size)
913 {
914 	char *mesg = buf;
915 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
916 	unsigned int maxconn = nn->max_connections;
917 
918 	if (size > 0) {
919 		int rv = get_uint(&mesg, &maxconn);
920 
921 		if (rv)
922 			return rv;
923 		trace_nfsd_ctl_maxconn(netns(file), maxconn);
924 		nn->max_connections = maxconn;
925 	}
926 
927 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%u\n", maxconn);
928 }
929 
930 #ifdef CONFIG_NFSD_V4
931 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
932 				  time64_t *time, struct nfsd_net *nn)
933 {
934 	struct dentry *dentry = file_dentry(file);
935 	char *mesg = buf;
936 	int rv, i;
937 
938 	if (size > 0) {
939 		if (nn->nfsd_serv)
940 			return -EBUSY;
941 		rv = get_int(&mesg, &i);
942 		if (rv)
943 			return rv;
944 		trace_nfsd_ctl_time(netns(file), dentry->d_name.name,
945 				    dentry->d_name.len, i);
946 
947 		/*
948 		 * Some sanity checking.  We don't have a reason for
949 		 * these particular numbers, but problems with the
950 		 * extremes are:
951 		 *	- Too short: the briefest network outage may
952 		 *	  cause clients to lose all their locks.  Also,
953 		 *	  the frequent polling may be wasteful.
954 		 *	- Too long: do you really want reboot recovery
955 		 *	  to take more than an hour?  Or to make other
956 		 *	  clients wait an hour before being able to
957 		 *	  revoke a dead client's locks?
958 		 */
959 		if (i < 10 || i > 3600)
960 			return -EINVAL;
961 		*time = i;
962 	}
963 
964 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%lld\n", *time);
965 }
966 
967 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
968 				time64_t *time, struct nfsd_net *nn)
969 {
970 	ssize_t rv;
971 
972 	mutex_lock(&nfsd_mutex);
973 	rv = __nfsd4_write_time(file, buf, size, time, nn);
974 	mutex_unlock(&nfsd_mutex);
975 	return rv;
976 }
977 
978 /*
979  * write_leasetime - Set or report the current NFSv4 lease time
980  *
981  * Input:
982  *			buf:		ignored
983  *			size:		zero
984  *
985  * OR
986  *
987  * Input:
988  *			buf:		C string containing an unsigned
989  *					integer value representing the new
990  *					NFSv4 lease expiry time
991  *			size:		non-zero length of C string in @buf
992  * Output:
993  *	On success:	passed-in buffer filled with '\n'-terminated C
994  *			string containing unsigned integer value of the
995  *			current lease expiry time;
996  *			return code is the size in bytes of the string
997  *	On error:	return code is zero or a negative errno value
998  */
999 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
1000 {
1001 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1002 	return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
1003 }
1004 
1005 /*
1006  * write_gracetime - Set or report current NFSv4 grace period time
1007  *
1008  * As above, but sets the time of the NFSv4 grace period.
1009  *
1010  * Note this should never be set to less than the *previous*
1011  * lease-period time, but we don't try to enforce this.  (In the common
1012  * case (a new boot), we don't know what the previous lease time was
1013  * anyway.)
1014  */
1015 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
1016 {
1017 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1018 	return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
1019 }
1020 
1021 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
1022 				   struct nfsd_net *nn)
1023 {
1024 	char *mesg = buf;
1025 	char *recdir;
1026 	int len, status;
1027 
1028 	if (size > 0) {
1029 		if (nn->nfsd_serv)
1030 			return -EBUSY;
1031 		if (size > PATH_MAX || buf[size-1] != '\n')
1032 			return -EINVAL;
1033 		buf[size-1] = 0;
1034 
1035 		recdir = mesg;
1036 		len = qword_get(&mesg, recdir, size);
1037 		if (len <= 0)
1038 			return -EINVAL;
1039 		trace_nfsd_ctl_recoverydir(netns(file), recdir);
1040 
1041 		status = nfs4_reset_recoverydir(recdir);
1042 		if (status)
1043 			return status;
1044 	}
1045 
1046 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1047 							nfs4_recoverydir());
1048 }
1049 
1050 /*
1051  * write_recoverydir - Set or report the pathname of the recovery directory
1052  *
1053  * Input:
1054  *			buf:		ignored
1055  *			size:		zero
1056  *
1057  * OR
1058  *
1059  * Input:
1060  *			buf:		C string containing the pathname
1061  *					of the directory on a local file
1062  *					system containing permanent NFSv4
1063  *					recovery data
1064  *			size:		non-zero length of C string in @buf
1065  * Output:
1066  *	On success:	passed-in buffer filled with '\n'-terminated C string
1067  *			containing the current recovery pathname setting;
1068  *			return code is the size in bytes of the string
1069  *	On error:	return code is zero or a negative errno value
1070  */
1071 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1072 {
1073 	ssize_t rv;
1074 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1075 
1076 	mutex_lock(&nfsd_mutex);
1077 	rv = __write_recoverydir(file, buf, size, nn);
1078 	mutex_unlock(&nfsd_mutex);
1079 	return rv;
1080 }
1081 
1082 /*
1083  * write_v4_end_grace - release grace period for nfsd's v4.x lock manager
1084  *
1085  * Input:
1086  *			buf:		ignored
1087  *			size:		zero
1088  * OR
1089  *
1090  * Input:
1091  *			buf:		any value
1092  *			size:		non-zero length of C string in @buf
1093  * Output:
1094  *			passed-in buffer filled with "Y" or "N" with a newline
1095  *			and NULL-terminated C string. This indicates whether
1096  *			the grace period has ended in the current net
1097  *			namespace. Return code is the size in bytes of the
1098  *			string. Writing a string that starts with 'Y', 'y', or
1099  *			'1' to the file will end the grace period for nfsd's v4
1100  *			lock manager.
1101  */
1102 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size)
1103 {
1104 	struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1105 
1106 	if (size > 0) {
1107 		switch(buf[0]) {
1108 		case 'Y':
1109 		case 'y':
1110 		case '1':
1111 			if (!nn->nfsd_serv)
1112 				return -EBUSY;
1113 			trace_nfsd_end_grace(netns(file));
1114 			nfsd4_end_grace(nn);
1115 			break;
1116 		default:
1117 			return -EINVAL;
1118 		}
1119 	}
1120 
1121 	return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%c\n",
1122 			 nn->grace_ended ? 'Y' : 'N');
1123 }
1124 
1125 #endif
1126 
1127 /*----------------------------------------------------------------------------*/
1128 /*
1129  *	populating the filesystem.
1130  */
1131 
1132 /* Basically copying rpc_get_inode. */
1133 static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
1134 {
1135 	struct inode *inode = new_inode(sb);
1136 	if (!inode)
1137 		return NULL;
1138 	/* Following advice from simple_fill_super documentation: */
1139 	inode->i_ino = iunique(sb, NFSD_MaxReserved);
1140 	inode->i_mode = mode;
1141 	simple_inode_init_ts(inode);
1142 	switch (mode & S_IFMT) {
1143 	case S_IFDIR:
1144 		inode->i_fop = &simple_dir_operations;
1145 		inode->i_op = &simple_dir_inode_operations;
1146 		inc_nlink(inode);
1147 		break;
1148 	case S_IFLNK:
1149 		inode->i_op = &simple_symlink_inode_operations;
1150 		break;
1151 	default:
1152 		break;
1153 	}
1154 	return inode;
1155 }
1156 
1157 static int __nfsd_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode, struct nfsdfs_client *ncl)
1158 {
1159 	struct inode *inode;
1160 
1161 	inode = nfsd_get_inode(dir->i_sb, mode);
1162 	if (!inode)
1163 		return -ENOMEM;
1164 	if (ncl) {
1165 		inode->i_private = ncl;
1166 		kref_get(&ncl->cl_ref);
1167 	}
1168 	d_add(dentry, inode);
1169 	inc_nlink(dir);
1170 	fsnotify_mkdir(dir, dentry);
1171 	return 0;
1172 }
1173 
1174 static struct dentry *nfsd_mkdir(struct dentry *parent, struct nfsdfs_client *ncl, char *name)
1175 {
1176 	struct inode *dir = parent->d_inode;
1177 	struct dentry *dentry;
1178 	int ret = -ENOMEM;
1179 
1180 	inode_lock(dir);
1181 	dentry = d_alloc_name(parent, name);
1182 	if (!dentry)
1183 		goto out_err;
1184 	ret = __nfsd_mkdir(d_inode(parent), dentry, S_IFDIR | 0600, ncl);
1185 	if (ret)
1186 		goto out_err;
1187 out:
1188 	inode_unlock(dir);
1189 	return dentry;
1190 out_err:
1191 	dput(dentry);
1192 	dentry = ERR_PTR(ret);
1193 	goto out;
1194 }
1195 
1196 #if IS_ENABLED(CONFIG_SUNRPC_GSS)
1197 static int __nfsd_symlink(struct inode *dir, struct dentry *dentry,
1198 			  umode_t mode, const char *content)
1199 {
1200 	struct inode *inode;
1201 
1202 	inode = nfsd_get_inode(dir->i_sb, mode);
1203 	if (!inode)
1204 		return -ENOMEM;
1205 
1206 	inode->i_link = (char *)content;
1207 	inode->i_size = strlen(content);
1208 
1209 	d_add(dentry, inode);
1210 	inc_nlink(dir);
1211 	fsnotify_create(dir, dentry);
1212 	return 0;
1213 }
1214 
1215 /*
1216  * @content is assumed to be a NUL-terminated string that lives
1217  * longer than the symlink itself.
1218  */
1219 static void _nfsd_symlink(struct dentry *parent, const char *name,
1220 			  const char *content)
1221 {
1222 	struct inode *dir = parent->d_inode;
1223 	struct dentry *dentry;
1224 	int ret;
1225 
1226 	inode_lock(dir);
1227 	dentry = d_alloc_name(parent, name);
1228 	if (!dentry)
1229 		goto out;
1230 	ret = __nfsd_symlink(d_inode(parent), dentry, S_IFLNK | 0777, content);
1231 	if (ret)
1232 		dput(dentry);
1233 out:
1234 	inode_unlock(dir);
1235 }
1236 #else
1237 static inline void _nfsd_symlink(struct dentry *parent, const char *name,
1238 				 const char *content)
1239 {
1240 }
1241 
1242 #endif
1243 
1244 static void clear_ncl(struct inode *inode)
1245 {
1246 	struct nfsdfs_client *ncl = inode->i_private;
1247 
1248 	inode->i_private = NULL;
1249 	kref_put(&ncl->cl_ref, ncl->cl_release);
1250 }
1251 
1252 static struct nfsdfs_client *__get_nfsdfs_client(struct inode *inode)
1253 {
1254 	struct nfsdfs_client *nc = inode->i_private;
1255 
1256 	if (nc)
1257 		kref_get(&nc->cl_ref);
1258 	return nc;
1259 }
1260 
1261 struct nfsdfs_client *get_nfsdfs_client(struct inode *inode)
1262 {
1263 	struct nfsdfs_client *nc;
1264 
1265 	inode_lock_shared(inode);
1266 	nc = __get_nfsdfs_client(inode);
1267 	inode_unlock_shared(inode);
1268 	return nc;
1269 }
1270 /* from __rpc_unlink */
1271 static void nfsdfs_remove_file(struct inode *dir, struct dentry *dentry)
1272 {
1273 	int ret;
1274 
1275 	clear_ncl(d_inode(dentry));
1276 	dget(dentry);
1277 	ret = simple_unlink(dir, dentry);
1278 	d_drop(dentry);
1279 	fsnotify_unlink(dir, dentry);
1280 	dput(dentry);
1281 	WARN_ON_ONCE(ret);
1282 }
1283 
1284 static void nfsdfs_remove_files(struct dentry *root)
1285 {
1286 	struct dentry *dentry, *tmp;
1287 
1288 	list_for_each_entry_safe(dentry, tmp, &root->d_subdirs, d_child) {
1289 		if (!simple_positive(dentry)) {
1290 			WARN_ON_ONCE(1); /* I think this can't happen? */
1291 			continue;
1292 		}
1293 		nfsdfs_remove_file(d_inode(root), dentry);
1294 	}
1295 }
1296 
1297 /* XXX: cut'n'paste from simple_fill_super; figure out if we could share
1298  * code instead. */
1299 static  int nfsdfs_create_files(struct dentry *root,
1300 				const struct tree_descr *files,
1301 				struct dentry **fdentries)
1302 {
1303 	struct inode *dir = d_inode(root);
1304 	struct inode *inode;
1305 	struct dentry *dentry;
1306 	int i;
1307 
1308 	inode_lock(dir);
1309 	for (i = 0; files->name && files->name[0]; i++, files++) {
1310 		dentry = d_alloc_name(root, files->name);
1311 		if (!dentry)
1312 			goto out;
1313 		inode = nfsd_get_inode(d_inode(root)->i_sb,
1314 					S_IFREG | files->mode);
1315 		if (!inode) {
1316 			dput(dentry);
1317 			goto out;
1318 		}
1319 		inode->i_fop = files->ops;
1320 		inode->i_private = __get_nfsdfs_client(dir);
1321 		d_add(dentry, inode);
1322 		fsnotify_create(dir, dentry);
1323 		if (fdentries)
1324 			fdentries[i] = dentry;
1325 	}
1326 	inode_unlock(dir);
1327 	return 0;
1328 out:
1329 	nfsdfs_remove_files(root);
1330 	inode_unlock(dir);
1331 	return -ENOMEM;
1332 }
1333 
1334 /* on success, returns positive number unique to that client. */
1335 struct dentry *nfsd_client_mkdir(struct nfsd_net *nn,
1336 				 struct nfsdfs_client *ncl, u32 id,
1337 				 const struct tree_descr *files,
1338 				 struct dentry **fdentries)
1339 {
1340 	struct dentry *dentry;
1341 	char name[11];
1342 	int ret;
1343 
1344 	sprintf(name, "%u", id);
1345 
1346 	dentry = nfsd_mkdir(nn->nfsd_client_dir, ncl, name);
1347 	if (IS_ERR(dentry)) /* XXX: tossing errors? */
1348 		return NULL;
1349 	ret = nfsdfs_create_files(dentry, files, fdentries);
1350 	if (ret) {
1351 		nfsd_client_rmdir(dentry);
1352 		return NULL;
1353 	}
1354 	return dentry;
1355 }
1356 
1357 /* Taken from __rpc_rmdir: */
1358 void nfsd_client_rmdir(struct dentry *dentry)
1359 {
1360 	struct inode *dir = d_inode(dentry->d_parent);
1361 	struct inode *inode = d_inode(dentry);
1362 	int ret;
1363 
1364 	inode_lock(dir);
1365 	nfsdfs_remove_files(dentry);
1366 	clear_ncl(inode);
1367 	dget(dentry);
1368 	ret = simple_rmdir(dir, dentry);
1369 	WARN_ON_ONCE(ret);
1370 	d_drop(dentry);
1371 	fsnotify_rmdir(dir, dentry);
1372 	dput(dentry);
1373 	inode_unlock(dir);
1374 }
1375 
1376 static int nfsd_fill_super(struct super_block *sb, struct fs_context *fc)
1377 {
1378 	struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
1379 							nfsd_net_id);
1380 	struct dentry *dentry;
1381 	int ret;
1382 
1383 	static const struct tree_descr nfsd_files[] = {
1384 		[NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1385 		/* Per-export io stats use same ops as exports file */
1386 		[NFSD_Export_Stats] = {"export_stats", &exports_nfsd_operations, S_IRUGO},
1387 		[NFSD_Export_features] = {"export_features",
1388 					&export_features_fops, S_IRUGO},
1389 		[NFSD_FO_UnlockIP] = {"unlock_ip",
1390 					&transaction_ops, S_IWUSR|S_IRUSR},
1391 		[NFSD_FO_UnlockFS] = {"unlock_filesystem",
1392 					&transaction_ops, S_IWUSR|S_IRUSR},
1393 		[NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1394 		[NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1395 		[NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1396 		[NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1397 		[NFSD_Reply_Cache_Stats] = {"reply_cache_stats",
1398 					&nfsd_reply_cache_stats_fops, S_IRUGO},
1399 		[NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1400 		[NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1401 		[NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1402 		[NFSD_MaxConnections] = {"max_connections", &transaction_ops, S_IWUSR|S_IRUGO},
1403 		[NFSD_Filecache] = {"filecache", &nfsd_file_cache_stats_fops, S_IRUGO},
1404 #ifdef CONFIG_NFSD_V4
1405 		[NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1406 		[NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1407 		[NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1408 		[NFSD_V4EndGrace] = {"v4_end_grace", &transaction_ops, S_IWUSR|S_IRUGO},
1409 #endif
1410 		/* last one */ {""}
1411 	};
1412 
1413 	ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1414 	if (ret)
1415 		return ret;
1416 	_nfsd_symlink(sb->s_root, "supported_krb5_enctypes",
1417 		      "/proc/net/rpc/gss_krb5_enctypes");
1418 	dentry = nfsd_mkdir(sb->s_root, NULL, "clients");
1419 	if (IS_ERR(dentry))
1420 		return PTR_ERR(dentry);
1421 	nn->nfsd_client_dir = dentry;
1422 	return 0;
1423 }
1424 
1425 static int nfsd_fs_get_tree(struct fs_context *fc)
1426 {
1427 	return get_tree_keyed(fc, nfsd_fill_super, get_net(fc->net_ns));
1428 }
1429 
1430 static void nfsd_fs_free_fc(struct fs_context *fc)
1431 {
1432 	if (fc->s_fs_info)
1433 		put_net(fc->s_fs_info);
1434 }
1435 
1436 static const struct fs_context_operations nfsd_fs_context_ops = {
1437 	.free		= nfsd_fs_free_fc,
1438 	.get_tree	= nfsd_fs_get_tree,
1439 };
1440 
1441 static int nfsd_init_fs_context(struct fs_context *fc)
1442 {
1443 	put_user_ns(fc->user_ns);
1444 	fc->user_ns = get_user_ns(fc->net_ns->user_ns);
1445 	fc->ops = &nfsd_fs_context_ops;
1446 	return 0;
1447 }
1448 
1449 static void nfsd_umount(struct super_block *sb)
1450 {
1451 	struct net *net = sb->s_fs_info;
1452 
1453 	nfsd_shutdown_threads(net);
1454 
1455 	kill_litter_super(sb);
1456 	put_net(net);
1457 }
1458 
1459 static struct file_system_type nfsd_fs_type = {
1460 	.owner		= THIS_MODULE,
1461 	.name		= "nfsd",
1462 	.init_fs_context = nfsd_init_fs_context,
1463 	.kill_sb	= nfsd_umount,
1464 };
1465 MODULE_ALIAS_FS("nfsd");
1466 
1467 #ifdef CONFIG_PROC_FS
1468 
1469 static int exports_proc_open(struct inode *inode, struct file *file)
1470 {
1471 	return exports_net_open(current->nsproxy->net_ns, file);
1472 }
1473 
1474 static const struct proc_ops exports_proc_ops = {
1475 	.proc_open	= exports_proc_open,
1476 	.proc_read	= seq_read,
1477 	.proc_lseek	= seq_lseek,
1478 	.proc_release	= seq_release,
1479 };
1480 
1481 static int create_proc_exports_entry(void)
1482 {
1483 	struct proc_dir_entry *entry;
1484 
1485 	entry = proc_mkdir("fs/nfs", NULL);
1486 	if (!entry)
1487 		return -ENOMEM;
1488 	entry = proc_create("exports", 0, entry, &exports_proc_ops);
1489 	if (!entry) {
1490 		remove_proc_entry("fs/nfs", NULL);
1491 		return -ENOMEM;
1492 	}
1493 	return 0;
1494 }
1495 #else /* CONFIG_PROC_FS */
1496 static int create_proc_exports_entry(void)
1497 {
1498 	return 0;
1499 }
1500 #endif
1501 
1502 unsigned int nfsd_net_id;
1503 
1504 /**
1505  * nfsd_nl_rpc_status_get_start - Prepare rpc_status_get dumpit
1506  * @cb: netlink metadata and command arguments
1507  *
1508  * Return values:
1509  *   %0: The rpc_status_get command may proceed
1510  *   %-ENODEV: There is no NFSD running in this namespace
1511  */
1512 int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb)
1513 {
1514 	struct nfsd_net *nn = net_generic(sock_net(cb->skb->sk), nfsd_net_id);
1515 	int ret = -ENODEV;
1516 
1517 	mutex_lock(&nfsd_mutex);
1518 	if (nn->nfsd_serv)
1519 		ret = 0;
1520 	else
1521 		mutex_unlock(&nfsd_mutex);
1522 
1523 	return ret;
1524 }
1525 
1526 static int nfsd_genl_rpc_status_compose_msg(struct sk_buff *skb,
1527 					    struct netlink_callback *cb,
1528 					    struct nfsd_genl_rqstp *rqstp)
1529 {
1530 	void *hdr;
1531 	u32 i;
1532 
1533 	hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1534 			  &nfsd_nl_family, 0, NFSD_CMD_RPC_STATUS_GET);
1535 	if (!hdr)
1536 		return -ENOBUFS;
1537 
1538 	if (nla_put_be32(skb, NFSD_A_RPC_STATUS_XID, rqstp->rq_xid) ||
1539 	    nla_put_u32(skb, NFSD_A_RPC_STATUS_FLAGS, rqstp->rq_flags) ||
1540 	    nla_put_u32(skb, NFSD_A_RPC_STATUS_PROG, rqstp->rq_prog) ||
1541 	    nla_put_u32(skb, NFSD_A_RPC_STATUS_PROC, rqstp->rq_proc) ||
1542 	    nla_put_u8(skb, NFSD_A_RPC_STATUS_VERSION, rqstp->rq_vers) ||
1543 	    nla_put_s64(skb, NFSD_A_RPC_STATUS_SERVICE_TIME,
1544 			ktime_to_us(rqstp->rq_stime),
1545 			NFSD_A_RPC_STATUS_PAD))
1546 		return -ENOBUFS;
1547 
1548 	switch (rqstp->rq_saddr.sa_family) {
1549 	case AF_INET: {
1550 		const struct sockaddr_in *s_in, *d_in;
1551 
1552 		s_in = (const struct sockaddr_in *)&rqstp->rq_saddr;
1553 		d_in = (const struct sockaddr_in *)&rqstp->rq_daddr;
1554 		if (nla_put_in_addr(skb, NFSD_A_RPC_STATUS_SADDR4,
1555 				    s_in->sin_addr.s_addr) ||
1556 		    nla_put_in_addr(skb, NFSD_A_RPC_STATUS_DADDR4,
1557 				    d_in->sin_addr.s_addr) ||
1558 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1559 				 s_in->sin_port) ||
1560 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1561 				 d_in->sin_port))
1562 			return -ENOBUFS;
1563 		break;
1564 	}
1565 	case AF_INET6: {
1566 		const struct sockaddr_in6 *s_in, *d_in;
1567 
1568 		s_in = (const struct sockaddr_in6 *)&rqstp->rq_saddr;
1569 		d_in = (const struct sockaddr_in6 *)&rqstp->rq_daddr;
1570 		if (nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_SADDR6,
1571 				     &s_in->sin6_addr) ||
1572 		    nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_DADDR6,
1573 				     &d_in->sin6_addr) ||
1574 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1575 				 s_in->sin6_port) ||
1576 		    nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1577 				 d_in->sin6_port))
1578 			return -ENOBUFS;
1579 		break;
1580 	}
1581 	}
1582 
1583 	for (i = 0; i < rqstp->rq_opcnt; i++)
1584 		if (nla_put_u32(skb, NFSD_A_RPC_STATUS_COMPOUND_OPS,
1585 				rqstp->rq_opnum[i]))
1586 			return -ENOBUFS;
1587 
1588 	genlmsg_end(skb, hdr);
1589 	return 0;
1590 }
1591 
1592 /**
1593  * nfsd_nl_rpc_status_get_dumpit - Handle rpc_status_get dumpit
1594  * @skb: reply buffer
1595  * @cb: netlink metadata and command arguments
1596  *
1597  * Returns the size of the reply or a negative errno.
1598  */
1599 int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
1600 				  struct netlink_callback *cb)
1601 {
1602 	struct nfsd_net *nn = net_generic(sock_net(skb->sk), nfsd_net_id);
1603 	int i, ret, rqstp_index = 0;
1604 
1605 	rcu_read_lock();
1606 
1607 	for (i = 0; i < nn->nfsd_serv->sv_nrpools; i++) {
1608 		struct svc_rqst *rqstp;
1609 
1610 		if (i < cb->args[0]) /* already consumed */
1611 			continue;
1612 
1613 		rqstp_index = 0;
1614 		list_for_each_entry_rcu(rqstp,
1615 				&nn->nfsd_serv->sv_pools[i].sp_all_threads,
1616 				rq_all) {
1617 			struct nfsd_genl_rqstp genl_rqstp;
1618 			unsigned int status_counter;
1619 
1620 			if (rqstp_index++ < cb->args[1]) /* already consumed */
1621 				continue;
1622 			/*
1623 			 * Acquire rq_status_counter before parsing the rqst
1624 			 * fields. rq_status_counter is set to an odd value in
1625 			 * order to notify the consumers the rqstp fields are
1626 			 * meaningful.
1627 			 */
1628 			status_counter =
1629 				smp_load_acquire(&rqstp->rq_status_counter);
1630 			if (!(status_counter & 1))
1631 				continue;
1632 
1633 			genl_rqstp.rq_xid = rqstp->rq_xid;
1634 			genl_rqstp.rq_flags = rqstp->rq_flags;
1635 			genl_rqstp.rq_vers = rqstp->rq_vers;
1636 			genl_rqstp.rq_prog = rqstp->rq_prog;
1637 			genl_rqstp.rq_proc = rqstp->rq_proc;
1638 			genl_rqstp.rq_stime = rqstp->rq_stime;
1639 			genl_rqstp.rq_opcnt = 0;
1640 			memcpy(&genl_rqstp.rq_daddr, svc_daddr(rqstp),
1641 			       sizeof(struct sockaddr));
1642 			memcpy(&genl_rqstp.rq_saddr, svc_addr(rqstp),
1643 			       sizeof(struct sockaddr));
1644 
1645 #ifdef CONFIG_NFSD_V4
1646 			if (rqstp->rq_vers == NFS4_VERSION &&
1647 			    rqstp->rq_proc == NFSPROC4_COMPOUND) {
1648 				/* NFSv4 compound */
1649 				struct nfsd4_compoundargs *args;
1650 				int j;
1651 
1652 				args = rqstp->rq_argp;
1653 				genl_rqstp.rq_opcnt = args->opcnt;
1654 				for (j = 0; j < genl_rqstp.rq_opcnt; j++)
1655 					genl_rqstp.rq_opnum[j] =
1656 						args->ops[j].opnum;
1657 			}
1658 #endif /* CONFIG_NFSD_V4 */
1659 
1660 			/*
1661 			 * Acquire rq_status_counter before reporting the rqst
1662 			 * fields to the user.
1663 			 */
1664 			if (smp_load_acquire(&rqstp->rq_status_counter) !=
1665 			    status_counter)
1666 				continue;
1667 
1668 			ret = nfsd_genl_rpc_status_compose_msg(skb, cb,
1669 							       &genl_rqstp);
1670 			if (ret)
1671 				goto out;
1672 		}
1673 	}
1674 
1675 	cb->args[0] = i;
1676 	cb->args[1] = rqstp_index;
1677 	ret = skb->len;
1678 out:
1679 	rcu_read_unlock();
1680 
1681 	return ret;
1682 }
1683 
1684 /**
1685  * nfsd_nl_rpc_status_get_done - rpc_status_get dumpit post-processing
1686  * @cb: netlink metadata and command arguments
1687  *
1688  * Return values:
1689  *   %0: Success
1690  */
1691 int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb)
1692 {
1693 	mutex_unlock(&nfsd_mutex);
1694 
1695 	return 0;
1696 }
1697 
1698 /**
1699  * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
1700  * @net: a freshly-created network namespace
1701  *
1702  * This information stays around as long as the network namespace is
1703  * alive whether or not there is an NFSD instance running in the
1704  * namespace.
1705  *
1706  * Returns zero on success, or a negative errno otherwise.
1707  */
1708 static __net_init int nfsd_net_init(struct net *net)
1709 {
1710 	int retval;
1711 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1712 
1713 	retval = nfsd_export_init(net);
1714 	if (retval)
1715 		goto out_export_error;
1716 	retval = nfsd_idmap_init(net);
1717 	if (retval)
1718 		goto out_idmap_error;
1719 	retval = nfsd_net_reply_cache_init(nn);
1720 	if (retval)
1721 		goto out_repcache_error;
1722 	nn->nfsd_versions = NULL;
1723 	nn->nfsd4_minorversions = NULL;
1724 	nfsd4_init_leases_net(nn);
1725 	get_random_bytes(&nn->siphash_key, sizeof(nn->siphash_key));
1726 	seqlock_init(&nn->writeverf_lock);
1727 
1728 	return 0;
1729 
1730 out_repcache_error:
1731 	nfsd_idmap_shutdown(net);
1732 out_idmap_error:
1733 	nfsd_export_shutdown(net);
1734 out_export_error:
1735 	return retval;
1736 }
1737 
1738 /**
1739  * nfsd_net_exit - Release the nfsd_net portion of a net namespace
1740  * @net: a network namespace that is about to be destroyed
1741  *
1742  */
1743 static __net_exit void nfsd_net_exit(struct net *net)
1744 {
1745 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1746 
1747 	nfsd_net_reply_cache_destroy(nn);
1748 	nfsd_idmap_shutdown(net);
1749 	nfsd_export_shutdown(net);
1750 	nfsd_netns_free_versions(nn);
1751 }
1752 
1753 static struct pernet_operations nfsd_net_ops = {
1754 	.init = nfsd_net_init,
1755 	.exit = nfsd_net_exit,
1756 	.id   = &nfsd_net_id,
1757 	.size = sizeof(struct nfsd_net),
1758 };
1759 
1760 static int __init init_nfsd(void)
1761 {
1762 	int retval;
1763 
1764 	retval = nfsd4_init_slabs();
1765 	if (retval)
1766 		return retval;
1767 	retval = nfsd4_init_pnfs();
1768 	if (retval)
1769 		goto out_free_slabs;
1770 	retval = nfsd_stat_init();	/* Statistics */
1771 	if (retval)
1772 		goto out_free_pnfs;
1773 	retval = nfsd_drc_slab_create();
1774 	if (retval)
1775 		goto out_free_stat;
1776 	nfsd_lockd_init();	/* lockd->nfsd callbacks */
1777 	retval = create_proc_exports_entry();
1778 	if (retval)
1779 		goto out_free_lockd;
1780 	retval = register_pernet_subsys(&nfsd_net_ops);
1781 	if (retval < 0)
1782 		goto out_free_exports;
1783 	retval = register_cld_notifier();
1784 	if (retval)
1785 		goto out_free_subsys;
1786 	retval = nfsd4_create_laundry_wq();
1787 	if (retval)
1788 		goto out_free_cld;
1789 	retval = register_filesystem(&nfsd_fs_type);
1790 	if (retval)
1791 		goto out_free_all;
1792 	retval = genl_register_family(&nfsd_nl_family);
1793 	if (retval)
1794 		goto out_free_all;
1795 
1796 	return 0;
1797 out_free_all:
1798 	nfsd4_destroy_laundry_wq();
1799 out_free_cld:
1800 	unregister_cld_notifier();
1801 out_free_subsys:
1802 	unregister_pernet_subsys(&nfsd_net_ops);
1803 out_free_exports:
1804 	remove_proc_entry("fs/nfs/exports", NULL);
1805 	remove_proc_entry("fs/nfs", NULL);
1806 out_free_lockd:
1807 	nfsd_lockd_shutdown();
1808 	nfsd_drc_slab_free();
1809 out_free_stat:
1810 	nfsd_stat_shutdown();
1811 out_free_pnfs:
1812 	nfsd4_exit_pnfs();
1813 out_free_slabs:
1814 	nfsd4_free_slabs();
1815 	return retval;
1816 }
1817 
1818 static void __exit exit_nfsd(void)
1819 {
1820 	genl_unregister_family(&nfsd_nl_family);
1821 	unregister_filesystem(&nfsd_fs_type);
1822 	nfsd4_destroy_laundry_wq();
1823 	unregister_cld_notifier();
1824 	unregister_pernet_subsys(&nfsd_net_ops);
1825 	nfsd_drc_slab_free();
1826 	remove_proc_entry("fs/nfs/exports", NULL);
1827 	remove_proc_entry("fs/nfs", NULL);
1828 	nfsd_stat_shutdown();
1829 	nfsd_lockd_shutdown();
1830 	nfsd4_free_slabs();
1831 	nfsd4_exit_pnfs();
1832 }
1833 
1834 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1835 MODULE_DESCRIPTION("In-kernel NFS server");
1836 MODULE_LICENSE("GPL");
1837 module_init(init_nfsd)
1838 module_exit(exit_nfsd)
1839