fs/nfsd/nfsctl.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Syscall interface to knfsd.
   4  *
   5  * Copyright (C) 1995, 1996 Olaf Kirch <[email protected]>
   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/sunrpc/svc.h>
  19 #include <linux/module.h>
  20 #include <linux/fsnotify.h>
  21 #include <linux/nfslocalio.h>
  22
  23 #include "idmap.h"
  24 #include "nfsd.h"
  25 #include "cache.h"
  26 #include "state.h"
  27 #include "netns.h"
  28 #include "pnfs.h"
  29 #include "filecache.h"
  30 #include "trace.h"
  31 #include "netlink.h"
  32
  33 /*
  34  *      We have a single directory with several nodes in it.
  35  */
  36 enum {
  37         NFSD_Root = 1,
  38         NFSD_List,
  39         NFSD_Export_Stats,
  40         NFSD_Export_features,
  41         NFSD_Fh,
  42         NFSD_FO_UnlockIP,
  43         NFSD_FO_UnlockFS,
  44         NFSD_Threads,
  45         NFSD_Pool_Threads,
  46         NFSD_Pool_Stats,
  47         NFSD_Reply_Cache_Stats,
  48         NFSD_Versions,
  49         NFSD_Ports,
  50         NFSD_MaxBlkSize,
  51         NFSD_MaxConnections,
  52         NFSD_Filecache,
  53         NFSD_Leasetime,
  54         NFSD_Gracetime,
  55         NFSD_RecoveryDir,
  56         NFSD_V4EndGrace,
  57         NFSD_MaxReserved
  58 };
  59
  60 /*
  61  * write() for these nodes.
  62  */
  63 static ssize_t write_filehandle(struct file *file, char *buf, size_t size);
  64 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size);
  65 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size);
  66 static ssize_t write_threads(struct file *file, char *buf, size_t size);
  67 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size);
  68 static ssize_t write_versions(struct file *file, char *buf, size_t size);
  69 static ssize_t write_ports(struct file *file, char *buf, size_t size);
  70 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size);
  71 static ssize_t write_maxconn(struct file *file, char *buf, size_t size);
  72 #ifdef CONFIG_NFSD_V4
  73 static ssize_t write_leasetime(struct file *file, char *buf, size_t size);
  74 static ssize_t write_gracetime(struct file *file, char *buf, size_t size);
  75 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
  76 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size);
  77 #endif
  78 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size);
  79 #endif
  80
  81 static ssize_t (*const write_op[])(struct file *, char *, size_t) = {
  82         [NFSD_Fh] = write_filehandle,
  83         [NFSD_FO_UnlockIP] = write_unlock_ip,
  84         [NFSD_FO_UnlockFS] = write_unlock_fs,
  85         [NFSD_Threads] = write_threads,
  86         [NFSD_Pool_Threads] = write_pool_threads,
  87         [NFSD_Versions] = write_versions,
  88         [NFSD_Ports] = write_ports,
  89         [NFSD_MaxBlkSize] = write_maxblksize,
  90         [NFSD_MaxConnections] = write_maxconn,
  91 #ifdef CONFIG_NFSD_V4
  92         [NFSD_Leasetime] = write_leasetime,
  93         [NFSD_Gracetime] = write_gracetime,
  94 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
  95         [NFSD_RecoveryDir] = write_recoverydir,
  96 #endif
  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 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 179 {
 180         struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id);
 181
 182         return svc_pool_stats_open(&nn->nfsd_info, file);
 183 }
 184
 185 static const struct file_operations pool_stats_operations = {
 186         .open           = nfsd_pool_stats_open,
 187         .read           = seq_read,
 188         .llseek         = seq_lseek,
 189         .release        = seq_release,
 190 };
 191
 192 DEFINE_SHOW_ATTRIBUTE(nfsd_reply_cache_stats);
 193
 194 DEFINE_SHOW_ATTRIBUTE(nfsd_file_cache_stats);
 195
 196 /*----------------------------------------------------------------------------*/
 197 /*
 198  * payload - write methods
 199  */
 200
 201 static inline struct net *netns(struct file *file)
 202 {
 203         return file_inode(file)->i_sb->s_fs_info;
 204 }
 205
 206 /*
 207  * write_unlock_ip - Release all locks used by a client
 208  *
 209  * Experimental.
 210  *
 211  * Input:
 212  *                      buf:    '\n'-terminated C string containing a
 213  *                              presentation format IP address
 214  *                      size:   length of C string in @buf
 215  * Output:
 216  *      On success:     returns zero if all specified locks were released;
 217  *                      returns one if one or more locks were not released
 218  *      On error:       return code is negative errno value
 219  */
 220 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size)
 221 {
 222         struct sockaddr_storage address;
 223         struct sockaddr *sap = (struct sockaddr *)&address;
 224         size_t salen = sizeof(address);
 225         char *fo_path;
 226         struct net *net = netns(file);
 227
 228         /* sanity check */
 229         if (size == 0)
 230                 return -EINVAL;
 231
 232         if (buf[size-1] != '\n')
 233                 return -EINVAL;
 234
 235         fo_path = buf;
 236         if (qword_get(&buf, fo_path, size) < 0)
 237                 return -EINVAL;
 238
 239         if (rpc_pton(net, fo_path, size, sap, salen) == 0)
 240                 return -EINVAL;
 241
 242         trace_nfsd_ctl_unlock_ip(net, buf);
 243         return nlmsvc_unlock_all_by_ip(sap);
 244 }
 245
 246 /*
 247  * write_unlock_fs - Release all locks on a local file system
 248  *
 249  * Experimental.
 250  *
 251  * Input:
 252  *                      buf:    '\n'-terminated C string containing the
 253  *                              absolute pathname of a local file system
 254  *                      size:   length of C string in @buf
 255  * Output:
 256  *      On success:     returns zero if all specified locks were released;
 257  *                      returns one if one or more locks were not released
 258  *      On error:       return code is negative errno value
 259  */
 260 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size)
 261 {
 262         struct path path;
 263         char *fo_path;
 264         int error;
 265
 266         /* sanity check */
 267         if (size == 0)
 268                 return -EINVAL;
 269
 270         if (buf[size-1] != '\n')
 271                 return -EINVAL;
 272
 273         fo_path = buf;
 274         if (qword_get(&buf, fo_path, size) < 0)
 275                 return -EINVAL;
 276         trace_nfsd_ctl_unlock_fs(netns(file), fo_path);
 277         error = kern_path(fo_path, 0, &path);
 278         if (error)
 279                 return error;
 280
 281         /*
 282          * XXX: Needs better sanity checking.  Otherwise we could end up
 283          * releasing locks on the wrong file system.
 284          *
 285          * For example:
 286          * 1.  Does the path refer to a directory?
 287          * 2.  Is that directory a mount point, or
 288          * 3.  Is that directory the root of an exported file system?
 289          */
 290         error = nlmsvc_unlock_all_by_sb(path.dentry->d_sb);
 291         nfsd4_revoke_states(netns(file), path.dentry->d_sb);
 292
 293         path_put(&path);
 294         return error;
 295 }
 296
 297 /*
 298  * write_filehandle - Get a variable-length NFS file handle by path
 299  *
 300  * On input, the buffer contains a '\n'-terminated C string comprised of
 301  * three alphanumeric words separated by whitespace.  The string may
 302  * contain escape sequences.
 303  *
 304  * Input:
 305  *                      buf:
 306  *                              domain:         client domain name
 307  *                              path:           export pathname
 308  *                              maxsize:        numeric maximum size of
 309  *                                              @buf
 310  *                      size:   length of C string in @buf
 311  * Output:
 312  *      On success:     passed-in buffer filled with '\n'-terminated C
 313  *                      string containing a ASCII hex text version
 314  *                      of the NFS file handle;
 315  *                      return code is the size in bytes of the string
 316  *      On error:       return code is negative errno value
 317  */
 318 static ssize_t write_filehandle(struct file *file, char *buf, size_t size)
 319 {
 320         char *dname, *path;
 321         int maxsize;
 322         char *mesg = buf;
 323         int len;
 324         struct auth_domain *dom;
 325         struct knfsd_fh fh;
 326
 327         if (size == 0)
 328                 return -EINVAL;
 329
 330         if (buf[size-1] != '\n')
 331                 return -EINVAL;
 332         buf[size-1] = 0;
 333
 334         dname = mesg;
 335         len = qword_get(&mesg, dname, size);
 336         if (len <= 0)
 337                 return -EINVAL;
 338
 339         path = dname+len+1;
 340         len = qword_get(&mesg, path, size);
 341         if (len <= 0)
 342                 return -EINVAL;
 343
 344         len = get_int(&mesg, &maxsize);
 345         if (len)
 346                 return len;
 347
 348         if (maxsize < NFS_FHSIZE)
 349                 return -EINVAL;
 350         maxsize = min(maxsize, NFS3_FHSIZE);
 351
 352         if (qword_get(&mesg, mesg, size) > 0)
 353                 return -EINVAL;
 354
 355         trace_nfsd_ctl_filehandle(netns(file), dname, path, maxsize);
 356
 357         /* we have all the words, they are in buf.. */
 358         dom = unix_domain_find(dname);
 359         if (!dom)
 360                 return -ENOMEM;
 361
 362         len = exp_rootfh(netns(file), dom, path, &fh, maxsize);
 363         auth_domain_put(dom);
 364         if (len)
 365                 return len;
 366
 367         mesg = buf;
 368         len = SIMPLE_TRANSACTION_LIMIT;
 369         qword_addhex(&mesg, &len, fh.fh_raw, fh.fh_size);
 370         mesg[-1] = '\n';
 371         return mesg - buf;
 372 }
 373
 374 /*
 375  * write_threads - Start NFSD, or report the current number of running threads
 376  *
 377  * Input:
 378  *                      buf:            ignored
 379  *                      size:           zero
 380  * Output:
 381  *      On success:     passed-in buffer filled with '\n'-terminated C
 382  *                      string numeric value representing the number of
 383  *                      running NFSD threads;
 384  *                      return code is the size in bytes of the string
 385  *      On error:       return code is zero
 386  *
 387  * OR
 388  *
 389  * Input:
 390  *                      buf:            C string containing an unsigned
 391  *                                      integer value representing the
 392  *                                      number of NFSD threads to start
 393  *                      size:           non-zero length of C string in @buf
 394  * Output:
 395  *      On success:     NFS service is started;
 396  *                      passed-in buffer filled with '\n'-terminated C
 397  *                      string numeric value representing the number of
 398  *                      running NFSD threads;
 399  *                      return code is the size in bytes of the string
 400  *      On error:       return code is zero or a negative errno value
 401  */
 402 static ssize_t write_threads(struct file *file, char *buf, size_t size)
 403 {
 404         char *mesg = buf;
 405         int rv;
 406         struct net *net = netns(file);
 407
 408         if (size > 0) {
 409                 int newthreads;
 410                 rv = get_int(&mesg, &newthreads);
 411                 if (rv)
 412                         return rv;
 413                 if (newthreads < 0)
 414                         return -EINVAL;
 415                 trace_nfsd_ctl_threads(net, newthreads);
 416                 mutex_lock(&nfsd_mutex);
 417                 rv = nfsd_svc(1, &newthreads, net, file->f_cred, NULL);
 418                 mutex_unlock(&nfsd_mutex);
 419                 if (rv < 0)
 420                         return rv;
 421         } else
 422                 rv = nfsd_nrthreads(net);
 423
 424         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n", rv);
 425 }
 426
 427 /*
 428  * write_pool_threads - Set or report the current number of threads per pool
 429  *
 430  * Input:
 431  *                      buf:            ignored
 432  *                      size:           zero
 433  *
 434  * OR
 435  *
 436  * Input:
 437  *                      buf:            C string containing whitespace-
 438  *                                      separated unsigned integer values
 439  *                                      representing the number of NFSD
 440  *                                      threads to start in each pool
 441  *                      size:           non-zero length of C string in @buf
 442  * Output:
 443  *      On success:     passed-in buffer filled with '\n'-terminated C
 444  *                      string containing integer values representing the
 445  *                      number of NFSD threads in each pool;
 446  *                      return code is the size in bytes of the string
 447  *      On error:       return code is zero or a negative errno value
 448  */
 449 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size)
 450 {
 451         /* if size > 0, look for an array of number of threads per node
 452          * and apply them  then write out number of threads per node as reply
 453          */
 454         char *mesg = buf;
 455         int i;
 456         int rv;
 457         int len;
 458         int npools;
 459         int *nthreads;
 460         struct net *net = netns(file);
 461
 462         mutex_lock(&nfsd_mutex);
 463         npools = nfsd_nrpools(net);
 464         if (npools == 0) {
 465                 /*
 466                  * NFS is shut down.  The admin can start it by
 467                  * writing to the threads file but NOT the pool_threads
 468                  * file, sorry.  Report zero threads.
 469                  */
 470                 mutex_unlock(&nfsd_mutex);
 471                 strcpy(buf, "0\n");
 472                 return strlen(buf);
 473         }
 474
 475         nthreads = kcalloc(npools, sizeof(int), GFP_KERNEL);
 476         rv = -ENOMEM;
 477         if (nthreads == NULL)
 478                 goto out_free;
 479
 480         if (size > 0) {
 481                 for (i = 0; i < npools; i++) {
 482                         rv = get_int(&mesg, &nthreads[i]);
 483                         if (rv == -ENOENT)
 484                                 break;          /* fewer numbers than pools */
 485                         if (rv)
 486                                 goto out_free;  /* syntax error */
 487                         rv = -EINVAL;
 488                         if (nthreads[i] < 0)
 489                                 goto out_free;
 490                         trace_nfsd_ctl_pool_threads(net, i, nthreads[i]);
 491                 }
 492
 493                 /*
 494                  * There must always be a thread in pool 0; the admin
 495                  * can't shut down NFS completely using pool_threads.
 496                  */
 497                 if (nthreads[0] == 0)
 498                         nthreads[0] = 1;
 499
 500                 rv = nfsd_set_nrthreads(i, nthreads, net);
 501                 if (rv)
 502                         goto out_free;
 503         }
 504
 505         rv = nfsd_get_nrthreads(npools, nthreads, net);
 506         if (rv)
 507                 goto out_free;
 508
 509         mesg = buf;
 510         size = SIMPLE_TRANSACTION_LIMIT;
 511         for (i = 0; i < npools && size > 0; i++) {
 512                 snprintf(mesg, size, "%d%c", nthreads[i], (i == npools-1 ? '\n' : ' '));
 513                 len = strlen(mesg);
 514                 size -= len;
 515                 mesg += len;
 516         }
 517         rv = mesg - buf;
 518 out_free:
 519         kfree(nthreads);
 520         mutex_unlock(&nfsd_mutex);
 521         return rv;
 522 }
 523
 524 static ssize_t
 525 nfsd_print_version_support(struct nfsd_net *nn, char *buf, int remaining,
 526                 const char *sep, unsigned vers, int minor)
 527 {
 528         const char *format = minor < 0 ? "%s%c%u" : "%s%c%u.%u";
 529         bool supported = !!nfsd_vers(nn, vers, NFSD_TEST);
 530
 531         if (vers == 4 && minor >= 0 &&
 532             !nfsd_minorversion(nn, minor, NFSD_TEST))
 533                 supported = false;
 534         if (minor == 0 && supported)
 535                 /*
 536                  * special case for backward compatability.
 537                  * +4.0 is never reported, it is implied by
 538                  * +4, unless -4.0 is present.
 539                  */
 540                 return 0;
 541         return snprintf(buf, remaining, format, sep,
 542                         supported ? '+' : '-', vers, minor);
 543 }
 544
 545 static ssize_t __write_versions(struct file *file, char *buf, size_t size)
 546 {
 547         char *mesg = buf;
 548         char *vers, *minorp, sign;
 549         int len, num, remaining;
 550         ssize_t tlen = 0;
 551         char *sep;
 552         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 553
 554         if (size > 0) {
 555                 if (nn->nfsd_serv)
 556                         /* Cannot change versions without updating
 557                          * nn->nfsd_serv->sv_xdrsize, and reallocing
 558                          * rq_argp and rq_resp
 559                          */
 560                         return -EBUSY;
 561                 if (buf[size-1] != '\n')
 562                         return -EINVAL;
 563                 buf[size-1] = 0;
 564                 trace_nfsd_ctl_version(netns(file), buf);
 565
 566                 vers = mesg;
 567                 len = qword_get(&mesg, vers, size);
 568                 if (len <= 0) return -EINVAL;
 569                 do {
 570                         enum vers_op cmd;
 571                         unsigned minor;
 572                         sign = *vers;
 573                         if (sign == '+' || sign == '-')
 574                                 num = simple_strtol((vers+1), &minorp, 0);
 575                         else
 576                                 num = simple_strtol(vers, &minorp, 0);
 577                         if (*minorp == '.') {
 578                                 if (num != 4)
 579                                         return -EINVAL;
 580                                 if (kstrtouint(minorp+1, 0, &minor) < 0)
 581                                         return -EINVAL;
 582                         }
 583
 584                         cmd = sign == '-' ? NFSD_CLEAR : NFSD_SET;
 585                         switch(num) {
 586 #ifdef CONFIG_NFSD_V2
 587                         case 2:
 588 #endif
 589                         case 3:
 590                                 nfsd_vers(nn, num, cmd);
 591                                 break;
 592                         case 4:
 593                                 if (*minorp == '.') {
 594                                         if (nfsd_minorversion(nn, minor, cmd) < 0)
 595                                                 return -EINVAL;
 596                                 } else if ((cmd == NFSD_SET) != nfsd_vers(nn, num, NFSD_TEST)) {
 597                                         /*
 598                                          * Either we have +4 and no minors are enabled,
 599                                          * or we have -4 and at least one minor is enabled.
 600                                          * In either case, propagate 'cmd' to all minors.
 601                                          */
 602                                         minor = 0;
 603                                         while (nfsd_minorversion(nn, minor, cmd) >= 0)
 604                                                 minor++;
 605                                 }
 606                                 break;
 607                         default:
 608                                 /* Ignore requests to disable non-existent versions */
 609                                 if (cmd == NFSD_SET)
 610                                         return -EINVAL;
 611                         }
 612                         vers += len + 1;
 613                 } while ((len = qword_get(&mesg, vers, size)) > 0);
 614                 /* If all get turned off, turn them back on, as
 615                  * having no versions is BAD
 616                  */
 617                 nfsd_reset_versions(nn);
 618         }
 619
 620         /* Now write current state into reply buffer */
 621         sep = "";
 622         remaining = SIMPLE_TRANSACTION_LIMIT;
 623         for (num=2 ; num <= 4 ; num++) {
 624                 int minor;
 625                 if (!nfsd_vers(nn, num, NFSD_AVAIL))
 626                         continue;
 627
 628                 minor = -1;
 629                 do {
 630                         len = nfsd_print_version_support(nn, buf, remaining,
 631                                         sep, num, minor);
 632                         if (len >= remaining)
 633                                 goto out;
 634                         remaining -= len;
 635                         buf += len;
 636                         tlen += len;
 637                         minor++;
 638                         if (len)
 639                                 sep = " ";
 640                 } while (num == 4 && minor <= NFSD_SUPPORTED_MINOR_VERSION);
 641         }
 642 out:
 643         len = snprintf(buf, remaining, "\n");
 644         if (len >= remaining)
 645                 return -EINVAL;
 646         return tlen + len;
 647 }
 648
 649 /*
 650  * write_versions - Set or report the available NFS protocol versions
 651  *
 652  * Input:
 653  *                      buf:            ignored
 654  *                      size:           zero
 655  * Output:
 656  *      On success:     passed-in buffer filled with '\n'-terminated C
 657  *                      string containing positive or negative integer
 658  *                      values representing the current status of each
 659  *                      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  * OR
 664  *
 665  * Input:
 666  *                      buf:            C string containing whitespace-
 667  *                                      separated positive or negative
 668  *                                      integer values representing NFS
 669  *                                      protocol versions to enable ("+n")
 670  *                                      or disable ("-n")
 671  *                      size:           non-zero length of C string in @buf
 672  * Output:
 673  *      On success:     status of zero or more protocol versions has
 674  *                      been updated; passed-in buffer filled with
 675  *                      '\n'-terminated C string containing positive
 676  *                      or negative integer values representing the
 677  *                      current status of each protocol version;
 678  *                      return code is the size in bytes of the string
 679  *      On error:       return code is zero or a negative errno value
 680  */
 681 static ssize_t write_versions(struct file *file, char *buf, size_t size)
 682 {
 683         ssize_t rv;
 684
 685         mutex_lock(&nfsd_mutex);
 686         rv = __write_versions(file, buf, size);
 687         mutex_unlock(&nfsd_mutex);
 688         return rv;
 689 }
 690
 691 /*
 692  * Zero-length write.  Return a list of NFSD's current listener
 693  * transports.
 694  */
 695 static ssize_t __write_ports_names(char *buf, struct net *net)
 696 {
 697         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 698
 699         if (nn->nfsd_serv == NULL)
 700                 return 0;
 701         return svc_xprt_names(nn->nfsd_serv, buf, SIMPLE_TRANSACTION_LIMIT);
 702 }
 703
 704 /*
 705  * A single 'fd' number was written, in which case it must be for
 706  * a socket of a supported family/protocol, and we use it as an
 707  * nfsd listener.
 708  */
 709 static ssize_t __write_ports_addfd(char *buf, struct net *net, const struct cred *cred)
 710 {
 711         char *mesg = buf;
 712         int fd, err;
 713         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 714         struct svc_serv *serv;
 715
 716         err = get_int(&mesg, &fd);
 717         if (err != 0 || fd < 0)
 718                 return -EINVAL;
 719         trace_nfsd_ctl_ports_addfd(net, fd);
 720
 721         err = nfsd_create_serv(net);
 722         if (err != 0)
 723                 return err;
 724
 725         serv = nn->nfsd_serv;
 726         err = svc_addsock(serv, net, fd, buf, SIMPLE_TRANSACTION_LIMIT, cred);
 727
 728         if (!serv->sv_nrthreads && list_empty(&nn->nfsd_serv->sv_permsocks))
 729                 nfsd_destroy_serv(net);
 730
 731         return err;
 732 }
 733
 734 /*
 735  * A transport listener is added by writing its transport name and
 736  * a port number.
 737  */
 738 static ssize_t __write_ports_addxprt(char *buf, struct net *net, const struct cred *cred)
 739 {
 740         char transport[16];
 741         struct svc_xprt *xprt;
 742         int port, err;
 743         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 744         struct svc_serv *serv;
 745
 746         if (sscanf(buf, "%15s %5u", transport, &port) != 2)
 747                 return -EINVAL;
 748
 749         if (port < 1 || port > USHRT_MAX)
 750                 return -EINVAL;
 751         trace_nfsd_ctl_ports_addxprt(net, transport, port);
 752
 753         err = nfsd_create_serv(net);
 754         if (err != 0)
 755                 return err;
 756
 757         serv = nn->nfsd_serv;
 758         err = svc_xprt_create(serv, transport, net,
 759                               PF_INET, port, SVC_SOCK_ANONYMOUS, cred);
 760         if (err < 0)
 761                 goto out_err;
 762
 763         err = svc_xprt_create(serv, transport, net,
 764                               PF_INET6, port, SVC_SOCK_ANONYMOUS, cred);
 765         if (err < 0 && err != -EAFNOSUPPORT)
 766                 goto out_close;
 767
 768         return 0;
 769 out_close:
 770         xprt = svc_find_xprt(serv, transport, net, PF_INET, port);
 771         if (xprt != NULL) {
 772                 svc_xprt_close(xprt);
 773                 svc_xprt_put(xprt);
 774         }
 775 out_err:
 776         if (!serv->sv_nrthreads && list_empty(&nn->nfsd_serv->sv_permsocks))
 777                 nfsd_destroy_serv(net);
 778
 779         return err;
 780 }
 781
 782 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
 783                              struct net *net)
 784 {
 785         if (size == 0)
 786                 return __write_ports_names(buf, net);
 787
 788         if (isdigit(buf[0]))
 789                 return __write_ports_addfd(buf, net, file->f_cred);
 790
 791         if (isalpha(buf[0]))
 792                 return __write_ports_addxprt(buf, net, file->f_cred);
 793
 794         return -EINVAL;
 795 }
 796
 797 /*
 798  * write_ports - Pass a socket file descriptor or transport name to listen on
 799  *
 800  * Input:
 801  *                      buf:            ignored
 802  *                      size:           zero
 803  * Output:
 804  *      On success:     passed-in buffer filled with a '\n'-terminated C
 805  *                      string containing a whitespace-separated list of
 806  *                      named NFSD listeners;
 807  *                      return code is the size in bytes of the string
 808  *      On error:       return code is zero or a negative errno value
 809  *
 810  * OR
 811  *
 812  * Input:
 813  *                      buf:            C string containing an unsigned
 814  *                                      integer value representing a bound
 815  *                                      but unconnected socket that is to be
 816  *                                      used as an NFSD listener; listen(3)
 817  *                                      must be called for a SOCK_STREAM
 818  *                                      socket, otherwise it is ignored
 819  *                      size:           non-zero length of C string in @buf
 820  * Output:
 821  *      On success:     NFS service is started;
 822  *                      passed-in buffer filled with a '\n'-terminated C
 823  *                      string containing a unique alphanumeric name of
 824  *                      the listener;
 825  *                      return code is the size in bytes of the string
 826  *      On error:       return code is a negative errno value
 827  *
 828  * OR
 829  *
 830  * Input:
 831  *                      buf:            C string containing a transport
 832  *                                      name and an unsigned integer value
 833  *                                      representing the port to listen on,
 834  *                                      separated by whitespace
 835  *                      size:           non-zero length of C string in @buf
 836  * Output:
 837  *      On success:     returns zero; NFS service is started
 838  *      On error:       return code is a negative errno value
 839  */
 840 static ssize_t write_ports(struct file *file, char *buf, size_t size)
 841 {
 842         ssize_t rv;
 843
 844         mutex_lock(&nfsd_mutex);
 845         rv = __write_ports(file, buf, size, netns(file));
 846         mutex_unlock(&nfsd_mutex);
 847         return rv;
 848 }
 849
 850
 851 int nfsd_max_blksize;
 852
 853 /*
 854  * write_maxblksize - Set or report the current NFS blksize
 855  *
 856  * Input:
 857  *                      buf:            ignored
 858  *                      size:           zero
 859  *
 860  * OR
 861  *
 862  * Input:
 863  *                      buf:            C string containing an unsigned
 864  *                                      integer value representing the new
 865  *                                      NFS blksize
 866  *                      size:           non-zero length of C string in @buf
 867  * Output:
 868  *      On success:     passed-in buffer filled with '\n'-terminated C string
 869  *                      containing numeric value of the current NFS blksize
 870  *                      setting;
 871  *                      return code is the size in bytes of the string
 872  *      On error:       return code is zero or a negative errno value
 873  */
 874 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
 875 {
 876         char *mesg = buf;
 877         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 878
 879         if (size > 0) {
 880                 int bsize;
 881                 int rv = get_int(&mesg, &bsize);
 882                 if (rv)
 883                         return rv;
 884                 trace_nfsd_ctl_maxblksize(netns(file), bsize);
 885
 886                 /* force bsize into allowed range and
 887                  * required alignment.
 888                  */
 889                 bsize = max_t(int, bsize, 1024);
 890                 bsize = min_t(int, bsize, NFSSVC_MAXBLKSIZE);
 891                 bsize &= ~(1024-1);
 892                 mutex_lock(&nfsd_mutex);
 893                 if (nn->nfsd_serv) {
 894                         mutex_unlock(&nfsd_mutex);
 895                         return -EBUSY;
 896                 }
 897                 nfsd_max_blksize = bsize;
 898                 mutex_unlock(&nfsd_mutex);
 899         }
 900
 901         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
 902                                                         nfsd_max_blksize);
 903 }
 904
 905 /*
 906  * write_maxconn - Set or report the current max number of connections
 907  *
 908  * Input:
 909  *                      buf:            ignored
 910  *                      size:           zero
 911  * OR
 912  *
 913  * Input:
 914  *                      buf:            C string containing an unsigned
 915  *                                      integer value representing the new
 916  *                                      number of max connections
 917  *                      size:           non-zero length of C string in @buf
 918  * Output:
 919  *      On success:     passed-in buffer filled with '\n'-terminated C string
 920  *                      containing numeric value of max_connections setting
 921  *                      for this net namespace;
 922  *                      return code is the size in bytes of the string
 923  *      On error:       return code is zero or a negative errno value
 924  */
 925 static ssize_t write_maxconn(struct file *file, char *buf, size_t size)
 926 {
 927         char *mesg = buf;
 928         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 929         unsigned int maxconn = nn->max_connections;
 930
 931         if (size > 0) {
 932                 int rv = get_uint(&mesg, &maxconn);
 933
 934                 if (rv)
 935                         return rv;
 936                 trace_nfsd_ctl_maxconn(netns(file), maxconn);
 937                 nn->max_connections = maxconn;
 938         }
 939
 940         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%u\n", maxconn);
 941 }
 942
 943 #ifdef CONFIG_NFSD_V4
 944 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
 945                                   time64_t *time, struct nfsd_net *nn)
 946 {
 947         struct dentry *dentry = file_dentry(file);
 948         char *mesg = buf;
 949         int rv, i;
 950
 951         if (size > 0) {
 952                 if (nn->nfsd_serv)
 953                         return -EBUSY;
 954                 rv = get_int(&mesg, &i);
 955                 if (rv)
 956                         return rv;
 957                 trace_nfsd_ctl_time(netns(file), dentry->d_name.name,
 958                                     dentry->d_name.len, i);
 959
 960                 /*
 961                  * Some sanity checking.  We don't have a reason for
 962                  * these particular numbers, but problems with the
 963                  * extremes are:
 964                  *      - Too short: the briefest network outage may
 965                  *        cause clients to lose all their locks.  Also,
 966                  *        the frequent polling may be wasteful.
 967                  *      - Too long: do you really want reboot recovery
 968                  *        to take more than an hour?  Or to make other
 969                  *        clients wait an hour before being able to
 970                  *        revoke a dead client's locks?
 971                  */
 972                 if (i < 10 || i > 3600)
 973                         return -EINVAL;
 974                 *time = i;
 975         }
 976
 977         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%lld\n", *time);
 978 }
 979
 980 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
 981                                 time64_t *time, struct nfsd_net *nn)
 982 {
 983         ssize_t rv;
 984
 985         mutex_lock(&nfsd_mutex);
 986         rv = __nfsd4_write_time(file, buf, size, time, nn);
 987         mutex_unlock(&nfsd_mutex);
 988         return rv;
 989 }
 990
 991 /*
 992  * write_leasetime - Set or report the current NFSv4 lease time
 993  *
 994  * Input:
 995  *                      buf:            ignored
 996  *                      size:           zero
 997  *
 998  * OR
 999  *
1000  * Input:
1001  *                      buf:            C string containing an unsigned
1002  *                                      integer value representing the new
1003  *                                      NFSv4 lease expiry time
1004  *                      size:           non-zero length of C string in @buf
1005  * Output:
1006  *      On success:     passed-in buffer filled with '\n'-terminated C
1007  *                      string containing unsigned integer value of the
1008  *                      current lease expiry time;
1009  *                      return code is the size in bytes of the string
1010  *      On error:       return code is zero or a negative errno value
1011  */
1012 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
1013 {
1014         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1015         return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
1016 }
1017
1018 /*
1019  * write_gracetime - Set or report current NFSv4 grace period time
1020  *
1021  * As above, but sets the time of the NFSv4 grace period.
1022  *
1023  * Note this should never be set to less than the *previous*
1024  * lease-period time, but we don't try to enforce this.  (In the common
1025  * case (a new boot), we don't know what the previous lease time was
1026  * anyway.)
1027  */
1028 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
1029 {
1030         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1031         return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
1032 }
1033
1034 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1035 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
1036                                    struct nfsd_net *nn)
1037 {
1038         char *mesg = buf;
1039         char *recdir;
1040         int len, status;
1041
1042         if (size > 0) {
1043                 if (nn->nfsd_serv)
1044                         return -EBUSY;
1045                 if (size > PATH_MAX || buf[size-1] != '\n')
1046                         return -EINVAL;
1047                 buf[size-1] = 0;
1048
1049                 recdir = mesg;
1050                 len = qword_get(&mesg, recdir, size);
1051                 if (len <= 0)
1052                         return -EINVAL;
1053                 trace_nfsd_ctl_recoverydir(netns(file), recdir);
1054
1055                 status = nfs4_reset_recoverydir(recdir);
1056                 if (status)
1057                         return status;
1058         }
1059
1060         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1061                                                         nfs4_recoverydir());
1062 }
1063
1064 /*
1065  * write_recoverydir - Set or report the pathname of the recovery directory
1066  *
1067  * Input:
1068  *                      buf:            ignored
1069  *                      size:           zero
1070  *
1071  * OR
1072  *
1073  * Input:
1074  *                      buf:            C string containing the pathname
1075  *                                      of the directory on a local file
1076  *                                      system containing permanent NFSv4
1077  *                                      recovery data
1078  *                      size:           non-zero length of C string in @buf
1079  * Output:
1080  *      On success:     passed-in buffer filled with '\n'-terminated C string
1081  *                      containing the current recovery pathname setting;
1082  *                      return code is the size in bytes of the string
1083  *      On error:       return code is zero or a negative errno value
1084  */
1085 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1086 {
1087         ssize_t rv;
1088         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1089
1090         mutex_lock(&nfsd_mutex);
1091         rv = __write_recoverydir(file, buf, size, nn);
1092         mutex_unlock(&nfsd_mutex);
1093         return rv;
1094 }
1095 #endif
1096
1097 /*
1098  * write_v4_end_grace - release grace period for nfsd's v4.x lock manager
1099  *
1100  * Input:
1101  *                      buf:            ignored
1102  *                      size:           zero
1103  * OR
1104  *
1105  * Input:
1106  *                      buf:            any value
1107  *                      size:           non-zero length of C string in @buf
1108  * Output:
1109  *                      passed-in buffer filled with "Y" or "N" with a newline
1110  *                      and NULL-terminated C string. This indicates whether
1111  *                      the grace period has ended in the current net
1112  *                      namespace. Return code is the size in bytes of the
1113  *                      string. Writing a string that starts with 'Y', 'y', or
1114  *                      '1' to the file will end the grace period for nfsd's v4
1115  *                      lock manager.
1116  */
1117 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size)
1118 {
1119         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1120
1121         if (size > 0) {
1122                 switch(buf[0]) {
1123                 case 'Y':
1124                 case 'y':
1125                 case '1':
1126                         if (!nn->nfsd_serv)
1127                                 return -EBUSY;
1128                         trace_nfsd_end_grace(netns(file));
1129                         nfsd4_end_grace(nn);
1130                         break;
1131                 default:
1132                         return -EINVAL;
1133                 }
1134         }
1135
1136         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%c\n",
1137                          nn->grace_ended ? 'Y' : 'N');
1138 }
1139
1140 #endif
1141
1142 /*----------------------------------------------------------------------------*/
1143 /*
1144  *      populating the filesystem.
1145  */
1146
1147 /* Basically copying rpc_get_inode. */
1148 static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
1149 {
1150         struct inode *inode = new_inode(sb);
1151         if (!inode)
1152                 return NULL;
1153         /* Following advice from simple_fill_super documentation: */
1154         inode->i_ino = iunique(sb, NFSD_MaxReserved);
1155         inode->i_mode = mode;
1156         simple_inode_init_ts(inode);
1157         switch (mode & S_IFMT) {
1158         case S_IFDIR:
1159                 inode->i_fop = &simple_dir_operations;
1160                 inode->i_op = &simple_dir_inode_operations;
1161                 inc_nlink(inode);
1162                 break;
1163         case S_IFLNK:
1164                 inode->i_op = &simple_symlink_inode_operations;
1165                 break;
1166         default:
1167                 break;
1168         }
1169         return inode;
1170 }
1171
1172 static int __nfsd_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode, struct nfsdfs_client *ncl)
1173 {
1174         struct inode *inode;
1175
1176         inode = nfsd_get_inode(dir->i_sb, mode);
1177         if (!inode)
1178                 return -ENOMEM;
1179         if (ncl) {
1180                 inode->i_private = ncl;
1181                 kref_get(&ncl->cl_ref);
1182         }
1183         d_add(dentry, inode);
1184         inc_nlink(dir);
1185         fsnotify_mkdir(dir, dentry);
1186         return 0;
1187 }
1188
1189 static struct dentry *nfsd_mkdir(struct dentry *parent, struct nfsdfs_client *ncl, char *name)
1190 {
1191         struct inode *dir = parent->d_inode;
1192         struct dentry *dentry;
1193         int ret = -ENOMEM;
1194
1195         inode_lock(dir);
1196         dentry = d_alloc_name(parent, name);
1197         if (!dentry)
1198                 goto out_err;
1199         ret = __nfsd_mkdir(d_inode(parent), dentry, S_IFDIR | 0600, ncl);
1200         if (ret)
1201                 goto out_err;
1202 out:
1203         inode_unlock(dir);
1204         return dentry;
1205 out_err:
1206         dput(dentry);
1207         dentry = ERR_PTR(ret);
1208         goto out;
1209 }
1210
1211 #if IS_ENABLED(CONFIG_SUNRPC_GSS)
1212 static int __nfsd_symlink(struct inode *dir, struct dentry *dentry,
1213                           umode_t mode, const char *content)
1214 {
1215         struct inode *inode;
1216
1217         inode = nfsd_get_inode(dir->i_sb, mode);
1218         if (!inode)
1219                 return -ENOMEM;
1220
1221         inode->i_link = (char *)content;
1222         inode->i_size = strlen(content);
1223
1224         d_add(dentry, inode);
1225         inc_nlink(dir);
1226         fsnotify_create(dir, dentry);
1227         return 0;
1228 }
1229
1230 /*
1231  * @content is assumed to be a NUL-terminated string that lives
1232  * longer than the symlink itself.
1233  */
1234 static void _nfsd_symlink(struct dentry *parent, const char *name,
1235                           const char *content)
1236 {
1237         struct inode *dir = parent->d_inode;
1238         struct dentry *dentry;
1239         int ret;
1240
1241         inode_lock(dir);
1242         dentry = d_alloc_name(parent, name);
1243         if (!dentry)
1244                 goto out;
1245         ret = __nfsd_symlink(d_inode(parent), dentry, S_IFLNK | 0777, content);
1246         if (ret)
1247                 dput(dentry);
1248 out:
1249         inode_unlock(dir);
1250 }
1251 #else
1252 static inline void _nfsd_symlink(struct dentry *parent, const char *name,
1253                                  const char *content)
1254 {
1255 }
1256
1257 #endif
1258
1259 static void clear_ncl(struct dentry *dentry)
1260 {
1261         struct inode *inode = d_inode(dentry);
1262         struct nfsdfs_client *ncl = inode->i_private;
1263
1264         spin_lock(&inode->i_lock);
1265         inode->i_private = NULL;
1266         spin_unlock(&inode->i_lock);
1267         kref_put(&ncl->cl_ref, ncl->cl_release);
1268 }
1269
1270 struct nfsdfs_client *get_nfsdfs_client(struct inode *inode)
1271 {
1272         struct nfsdfs_client *nc;
1273
1274         spin_lock(&inode->i_lock);
1275         nc = inode->i_private;
1276         if (nc)
1277                 kref_get(&nc->cl_ref);
1278         spin_unlock(&inode->i_lock);
1279         return nc;
1280 }
1281
1282 /* XXX: cut'n'paste from simple_fill_super; figure out if we could share
1283  * code instead. */
1284 static  int nfsdfs_create_files(struct dentry *root,
1285                                 const struct tree_descr *files,
1286                                 struct nfsdfs_client *ncl,
1287                                 struct dentry **fdentries)
1288 {
1289         struct inode *dir = d_inode(root);
1290         struct inode *inode;
1291         struct dentry *dentry;
1292         int i;
1293
1294         inode_lock(dir);
1295         for (i = 0; files->name && files->name[0]; i++, files++) {
1296                 dentry = d_alloc_name(root, files->name);
1297                 if (!dentry)
1298                         goto out;
1299                 inode = nfsd_get_inode(d_inode(root)->i_sb,
1300                                         S_IFREG | files->mode);
1301                 if (!inode) {
1302                         dput(dentry);
1303                         goto out;
1304                 }
1305                 kref_get(&ncl->cl_ref);
1306                 inode->i_fop = files->ops;
1307                 inode->i_private = ncl;
1308                 d_add(dentry, inode);
1309                 fsnotify_create(dir, dentry);
1310                 if (fdentries)
1311                         fdentries[i] = dentry;
1312         }
1313         inode_unlock(dir);
1314         return 0;
1315 out:
1316         inode_unlock(dir);
1317         return -ENOMEM;
1318 }
1319
1320 /* on success, returns positive number unique to that client. */
1321 struct dentry *nfsd_client_mkdir(struct nfsd_net *nn,
1322                                  struct nfsdfs_client *ncl, u32 id,
1323                                  const struct tree_descr *files,
1324                                  struct dentry **fdentries)
1325 {
1326         struct dentry *dentry;
1327         char name[11];
1328         int ret;
1329
1330         sprintf(name, "%u", id);
1331
1332         dentry = nfsd_mkdir(nn->nfsd_client_dir, ncl, name);
1333         if (IS_ERR(dentry)) /* XXX: tossing errors? */
1334                 return NULL;
1335         ret = nfsdfs_create_files(dentry, files, ncl, fdentries);
1336         if (ret) {
1337                 nfsd_client_rmdir(dentry);
1338                 return NULL;
1339         }
1340         return dentry;
1341 }
1342
1343 /* Taken from __rpc_rmdir: */
1344 void nfsd_client_rmdir(struct dentry *dentry)
1345 {
1346         simple_recursive_removal(dentry, clear_ncl);
1347 }
1348
1349 static int nfsd_fill_super(struct super_block *sb, struct fs_context *fc)
1350 {
1351         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
1352                                                         nfsd_net_id);
1353         struct dentry *dentry;
1354         int ret;
1355
1356         static const struct tree_descr nfsd_files[] = {
1357                 [NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1358                 /* Per-export io stats use same ops as exports file */
1359                 [NFSD_Export_Stats] = {"export_stats", &exports_nfsd_operations, S_IRUGO},
1360                 [NFSD_Export_features] = {"export_features",
1361                                         &export_features_fops, S_IRUGO},
1362                 [NFSD_FO_UnlockIP] = {"unlock_ip",
1363                                         &transaction_ops, S_IWUSR|S_IRUSR},
1364                 [NFSD_FO_UnlockFS] = {"unlock_filesystem",
1365                                         &transaction_ops, S_IWUSR|S_IRUSR},
1366                 [NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1367                 [NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1368                 [NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1369                 [NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1370                 [NFSD_Reply_Cache_Stats] = {"reply_cache_stats",
1371                                         &nfsd_reply_cache_stats_fops, S_IRUGO},
1372                 [NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1373                 [NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1374                 [NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1375                 [NFSD_MaxConnections] = {"max_connections", &transaction_ops, S_IWUSR|S_IRUGO},
1376                 [NFSD_Filecache] = {"filecache", &nfsd_file_cache_stats_fops, S_IRUGO},
1377 #ifdef CONFIG_NFSD_V4
1378                 [NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1379                 [NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1380 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1381                 [NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1382 #endif
1383                 [NFSD_V4EndGrace] = {"v4_end_grace", &transaction_ops, S_IWUSR|S_IRUGO},
1384 #endif
1385                 /* last one */ {""}
1386         };
1387
1388         ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1389         if (ret)
1390                 return ret;
1391         _nfsd_symlink(sb->s_root, "supported_krb5_enctypes",
1392                       "/proc/net/rpc/gss_krb5_enctypes");
1393         dentry = nfsd_mkdir(sb->s_root, NULL, "clients");
1394         if (IS_ERR(dentry))
1395                 return PTR_ERR(dentry);
1396         nn->nfsd_client_dir = dentry;
1397         return 0;
1398 }
1399
1400 static int nfsd_fs_get_tree(struct fs_context *fc)
1401 {
1402         return get_tree_keyed(fc, nfsd_fill_super, get_net(fc->net_ns));
1403 }
1404
1405 static void nfsd_fs_free_fc(struct fs_context *fc)
1406 {
1407         if (fc->s_fs_info)
1408                 put_net(fc->s_fs_info);
1409 }
1410
1411 static const struct fs_context_operations nfsd_fs_context_ops = {
1412         .free           = nfsd_fs_free_fc,
1413         .get_tree       = nfsd_fs_get_tree,
1414 };
1415
1416 static int nfsd_init_fs_context(struct fs_context *fc)
1417 {
1418         put_user_ns(fc->user_ns);
1419         fc->user_ns = get_user_ns(fc->net_ns->user_ns);
1420         fc->ops = &nfsd_fs_context_ops;
1421         return 0;
1422 }
1423
1424 static void nfsd_umount(struct super_block *sb)
1425 {
1426         struct net *net = sb->s_fs_info;
1427
1428         nfsd_shutdown_threads(net);
1429
1430         kill_litter_super(sb);
1431         put_net(net);
1432 }
1433
1434 static struct file_system_type nfsd_fs_type = {
1435         .owner          = THIS_MODULE,
1436         .name           = "nfsd",
1437         .init_fs_context = nfsd_init_fs_context,
1438         .kill_sb        = nfsd_umount,
1439 };
1440 MODULE_ALIAS_FS("nfsd");
1441
1442 #ifdef CONFIG_PROC_FS
1443
1444 static int exports_proc_open(struct inode *inode, struct file *file)
1445 {
1446         return exports_net_open(current->nsproxy->net_ns, file);
1447 }
1448
1449 static const struct proc_ops exports_proc_ops = {
1450         .proc_open      = exports_proc_open,
1451         .proc_read      = seq_read,
1452         .proc_lseek     = seq_lseek,
1453         .proc_release   = seq_release,
1454 };
1455
1456 static int create_proc_exports_entry(void)
1457 {
1458         struct proc_dir_entry *entry;
1459
1460         entry = proc_mkdir("fs/nfs", NULL);
1461         if (!entry)
1462                 return -ENOMEM;
1463         entry = proc_create("exports", 0, entry, &exports_proc_ops);
1464         if (!entry) {
1465                 remove_proc_entry("fs/nfs", NULL);
1466                 return -ENOMEM;
1467         }
1468         return 0;
1469 }
1470 #else /* CONFIG_PROC_FS */
1471 static int create_proc_exports_entry(void)
1472 {
1473         return 0;
1474 }
1475 #endif
1476
1477 unsigned int nfsd_net_id;
1478
1479 static int nfsd_genl_rpc_status_compose_msg(struct sk_buff *skb,
1480                                             struct netlink_callback *cb,
1481                                             struct nfsd_genl_rqstp *rqstp)
1482 {
1483         void *hdr;
1484         u32 i;
1485
1486         hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1487                           &nfsd_nl_family, 0, NFSD_CMD_RPC_STATUS_GET);
1488         if (!hdr)
1489                 return -ENOBUFS;
1490
1491         if (nla_put_be32(skb, NFSD_A_RPC_STATUS_XID, rqstp->rq_xid) ||
1492             nla_put_u32(skb, NFSD_A_RPC_STATUS_FLAGS, rqstp->rq_flags) ||
1493             nla_put_u32(skb, NFSD_A_RPC_STATUS_PROG, rqstp->rq_prog) ||
1494             nla_put_u32(skb, NFSD_A_RPC_STATUS_PROC, rqstp->rq_proc) ||
1495             nla_put_u8(skb, NFSD_A_RPC_STATUS_VERSION, rqstp->rq_vers) ||
1496             nla_put_s64(skb, NFSD_A_RPC_STATUS_SERVICE_TIME,
1497                         ktime_to_us(rqstp->rq_stime),
1498                         NFSD_A_RPC_STATUS_PAD))
1499                 return -ENOBUFS;
1500
1501         switch (rqstp->rq_saddr.sa_family) {
1502         case AF_INET: {
1503                 const struct sockaddr_in *s_in, *d_in;
1504
1505                 s_in = (const struct sockaddr_in *)&rqstp->rq_saddr;
1506                 d_in = (const struct sockaddr_in *)&rqstp->rq_daddr;
1507                 if (nla_put_in_addr(skb, NFSD_A_RPC_STATUS_SADDR4,
1508                                     s_in->sin_addr.s_addr) ||
1509                     nla_put_in_addr(skb, NFSD_A_RPC_STATUS_DADDR4,
1510                                     d_in->sin_addr.s_addr) ||
1511                     nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1512                                  s_in->sin_port) ||
1513                     nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1514                                  d_in->sin_port))
1515                         return -ENOBUFS;
1516                 break;
1517         }
1518         case AF_INET6: {
1519                 const struct sockaddr_in6 *s_in, *d_in;
1520
1521                 s_in = (const struct sockaddr_in6 *)&rqstp->rq_saddr;
1522                 d_in = (const struct sockaddr_in6 *)&rqstp->rq_daddr;
1523                 if (nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_SADDR6,
1524                                      &s_in->sin6_addr) ||
1525                     nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_DADDR6,
1526                                      &d_in->sin6_addr) ||
1527                     nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1528                                  s_in->sin6_port) ||
1529                     nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1530                                  d_in->sin6_port))
1531                         return -ENOBUFS;
1532                 break;
1533         }
1534         }
1535
1536         for (i = 0; i < rqstp->rq_opcnt; i++)
1537                 if (nla_put_u32(skb, NFSD_A_RPC_STATUS_COMPOUND_OPS,
1538                                 rqstp->rq_opnum[i]))
1539                         return -ENOBUFS;
1540
1541         genlmsg_end(skb, hdr);
1542         return 0;
1543 }
1544
1545 /**
1546  * nfsd_nl_rpc_status_get_dumpit - Handle rpc_status_get dumpit
1547  * @skb: reply buffer
1548  * @cb: netlink metadata and command arguments
1549  *
1550  * Returns the size of the reply or a negative errno.
1551  */
1552 int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
1553                                   struct netlink_callback *cb)
1554 {
1555         int i, ret, rqstp_index = 0;
1556         struct nfsd_net *nn;
1557
1558         mutex_lock(&nfsd_mutex);
1559
1560         nn = net_generic(sock_net(skb->sk), nfsd_net_id);
1561         if (!nn->nfsd_serv) {
1562                 ret = -ENODEV;
1563                 goto out_unlock;
1564         }
1565
1566         rcu_read_lock();
1567
1568         for (i = 0; i < nn->nfsd_serv->sv_nrpools; i++) {
1569                 struct svc_rqst *rqstp;
1570
1571                 if (i < cb->args[0]) /* already consumed */
1572                         continue;
1573
1574                 rqstp_index = 0;
1575                 list_for_each_entry_rcu(rqstp,
1576                                 &nn->nfsd_serv->sv_pools[i].sp_all_threads,
1577                                 rq_all) {
1578                         struct nfsd_genl_rqstp genl_rqstp;
1579                         unsigned int status_counter;
1580
1581                         if (rqstp_index++ < cb->args[1]) /* already consumed */
1582                                 continue;
1583                         /*
1584                          * Acquire rq_status_counter before parsing the rqst
1585                          * fields. rq_status_counter is set to an odd value in
1586                          * order to notify the consumers the rqstp fields are
1587                          * meaningful.
1588                          */
1589                         status_counter =
1590                                 smp_load_acquire(&rqstp->rq_status_counter);
1591                         if (!(status_counter & 1))
1592                                 continue;
1593
1594                         genl_rqstp.rq_xid = rqstp->rq_xid;
1595                         genl_rqstp.rq_flags = rqstp->rq_flags;
1596                         genl_rqstp.rq_vers = rqstp->rq_vers;
1597                         genl_rqstp.rq_prog = rqstp->rq_prog;
1598                         genl_rqstp.rq_proc = rqstp->rq_proc;
1599                         genl_rqstp.rq_stime = rqstp->rq_stime;
1600                         genl_rqstp.rq_opcnt = 0;
1601                         memcpy(&genl_rqstp.rq_daddr, svc_daddr(rqstp),
1602                                sizeof(struct sockaddr));
1603                         memcpy(&genl_rqstp.rq_saddr, svc_addr(rqstp),
1604                                sizeof(struct sockaddr));
1605
1606 #ifdef CONFIG_NFSD_V4
1607                         if (rqstp->rq_vers == NFS4_VERSION &&
1608                             rqstp->rq_proc == NFSPROC4_COMPOUND) {
1609                                 /* NFSv4 compound */
1610                                 struct nfsd4_compoundargs *args;
1611                                 int j;
1612
1613                                 args = rqstp->rq_argp;
1614                                 genl_rqstp.rq_opcnt = args->opcnt;
1615                                 for (j = 0; j < genl_rqstp.rq_opcnt; j++)
1616                                         genl_rqstp.rq_opnum[j] =
1617                                                 args->ops[j].opnum;
1618                         }
1619 #endif /* CONFIG_NFSD_V4 */
1620
1621                         /*
1622                          * Acquire rq_status_counter before reporting the rqst
1623                          * fields to the user.
1624                          */
1625                         if (smp_load_acquire(&rqstp->rq_status_counter) !=
1626                             status_counter)
1627                                 continue;
1628
1629                         ret = nfsd_genl_rpc_status_compose_msg(skb, cb,
1630                                                                &genl_rqstp);
1631                         if (ret)
1632                                 goto out;
1633                 }
1634         }
1635
1636         cb->args[0] = i;
1637         cb->args[1] = rqstp_index;
1638         ret = skb->len;
1639 out:
1640         rcu_read_unlock();
1641 out_unlock:
1642         mutex_unlock(&nfsd_mutex);
1643
1644         return ret;
1645 }
1646
1647 /**
1648  * nfsd_nl_threads_set_doit - set the number of running threads
1649  * @skb: reply buffer
1650  * @info: netlink metadata and command arguments
1651  *
1652  * Return 0 on success or a negative errno.
1653  */
1654 int nfsd_nl_threads_set_doit(struct sk_buff *skb, struct genl_info *info)
1655 {
1656         int *nthreads, count = 0, nrpools, i, ret = -EOPNOTSUPP, rem;
1657         struct net *net = genl_info_net(info);
1658         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1659         const struct nlattr *attr;
1660         const char *scope = NULL;
1661
1662         if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_THREADS))
1663                 return -EINVAL;
1664
1665         /* count number of SERVER_THREADS values */
1666         nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
1667                 if (nla_type(attr) == NFSD_A_SERVER_THREADS)
1668                         count++;
1669         }
1670
1671         mutex_lock(&nfsd_mutex);
1672
1673         nrpools = max(count, nfsd_nrpools(net));
1674         nthreads = kcalloc(nrpools, sizeof(int), GFP_KERNEL);
1675         if (!nthreads) {
1676                 ret = -ENOMEM;
1677                 goto out_unlock;
1678         }
1679
1680         i = 0;
1681         nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
1682                 if (nla_type(attr) == NFSD_A_SERVER_THREADS) {
1683                         nthreads[i++] = nla_get_u32(attr);
1684                         if (i >= nrpools)
1685                                 break;
1686                 }
1687         }
1688
1689         if (info->attrs[NFSD_A_SERVER_GRACETIME] ||
1690             info->attrs[NFSD_A_SERVER_LEASETIME] ||
1691             info->attrs[NFSD_A_SERVER_SCOPE]) {
1692                 ret = -EBUSY;
1693                 if (nn->nfsd_serv && nn->nfsd_serv->sv_nrthreads)
1694                         goto out_unlock;
1695
1696                 ret = -EINVAL;
1697                 attr = info->attrs[NFSD_A_SERVER_GRACETIME];
1698                 if (attr) {
1699                         u32 gracetime = nla_get_u32(attr);
1700
1701                         if (gracetime < 10 || gracetime > 3600)
1702                                 goto out_unlock;
1703
1704                         nn->nfsd4_grace = gracetime;
1705                 }
1706
1707                 attr = info->attrs[NFSD_A_SERVER_LEASETIME];
1708                 if (attr) {
1709                         u32 leasetime = nla_get_u32(attr);
1710
1711                         if (leasetime < 10 || leasetime > 3600)
1712                                 goto out_unlock;
1713
1714                         nn->nfsd4_lease = leasetime;
1715                 }
1716
1717                 attr = info->attrs[NFSD_A_SERVER_SCOPE];
1718                 if (attr)
1719                         scope = nla_data(attr);
1720         }
1721
1722         ret = nfsd_svc(nrpools, nthreads, net, get_current_cred(), scope);
1723         if (ret > 0)
1724                 ret = 0;
1725 out_unlock:
1726         mutex_unlock(&nfsd_mutex);
1727         kfree(nthreads);
1728         return ret;
1729 }
1730
1731 /**
1732  * nfsd_nl_threads_get_doit - get the number of running threads
1733  * @skb: reply buffer
1734  * @info: netlink metadata and command arguments
1735  *
1736  * Return 0 on success or a negative errno.
1737  */
1738 int nfsd_nl_threads_get_doit(struct sk_buff *skb, struct genl_info *info)
1739 {
1740         struct net *net = genl_info_net(info);
1741         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1742         void *hdr;
1743         int err;
1744
1745         skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1746         if (!skb)
1747                 return -ENOMEM;
1748
1749         hdr = genlmsg_iput(skb, info);
1750         if (!hdr) {
1751                 err = -EMSGSIZE;
1752                 goto err_free_msg;
1753         }
1754
1755         mutex_lock(&nfsd_mutex);
1756
1757         err = nla_put_u32(skb, NFSD_A_SERVER_GRACETIME,
1758                           nn->nfsd4_grace) ||
1759               nla_put_u32(skb, NFSD_A_SERVER_LEASETIME,
1760                           nn->nfsd4_lease) ||
1761               nla_put_string(skb, NFSD_A_SERVER_SCOPE,
1762                           nn->nfsd_name);
1763         if (err)
1764                 goto err_unlock;
1765
1766         if (nn->nfsd_serv) {
1767                 int i;
1768
1769                 for (i = 0; i < nfsd_nrpools(net); ++i) {
1770                         struct svc_pool *sp = &nn->nfsd_serv->sv_pools[i];
1771
1772                         err = nla_put_u32(skb, NFSD_A_SERVER_THREADS,
1773                                           sp->sp_nrthreads);
1774                         if (err)
1775                                 goto err_unlock;
1776                 }
1777         } else {
1778                 err = nla_put_u32(skb, NFSD_A_SERVER_THREADS, 0);
1779                 if (err)
1780                         goto err_unlock;
1781         }
1782
1783         mutex_unlock(&nfsd_mutex);
1784
1785         genlmsg_end(skb, hdr);
1786
1787         return genlmsg_reply(skb, info);
1788
1789 err_unlock:
1790         mutex_unlock(&nfsd_mutex);
1791 err_free_msg:
1792         nlmsg_free(skb);
1793
1794         return err;
1795 }
1796
1797 /**
1798  * nfsd_nl_version_set_doit - set the nfs enabled versions
1799  * @skb: reply buffer
1800  * @info: netlink metadata and command arguments
1801  *
1802  * Return 0 on success or a negative errno.
1803  */
1804 int nfsd_nl_version_set_doit(struct sk_buff *skb, struct genl_info *info)
1805 {
1806         const struct nlattr *attr;
1807         struct nfsd_net *nn;
1808         int i, rem;
1809
1810         if (GENL_REQ_ATTR_CHECK(info, NFSD_A_SERVER_PROTO_VERSION))
1811                 return -EINVAL;
1812
1813         mutex_lock(&nfsd_mutex);
1814
1815         nn = net_generic(genl_info_net(info), nfsd_net_id);
1816         if (nn->nfsd_serv) {
1817                 mutex_unlock(&nfsd_mutex);
1818                 return -EBUSY;
1819         }
1820
1821         /* clear current supported versions. */
1822         nfsd_vers(nn, 2, NFSD_CLEAR);
1823         nfsd_vers(nn, 3, NFSD_CLEAR);
1824         for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++)
1825                 nfsd_minorversion(nn, i, NFSD_CLEAR);
1826
1827         nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
1828                 struct nlattr *tb[NFSD_A_VERSION_MAX + 1];
1829                 u32 major, minor = 0;
1830                 bool enabled;
1831
1832                 if (nla_type(attr) != NFSD_A_SERVER_PROTO_VERSION)
1833                         continue;
1834
1835                 if (nla_parse_nested(tb, NFSD_A_VERSION_MAX, attr,
1836                                      nfsd_version_nl_policy, info->extack) < 0)
1837                         continue;
1838
1839                 if (!tb[NFSD_A_VERSION_MAJOR])
1840                         continue;
1841
1842                 major = nla_get_u32(tb[NFSD_A_VERSION_MAJOR]);
1843                 if (tb[NFSD_A_VERSION_MINOR])
1844                         minor = nla_get_u32(tb[NFSD_A_VERSION_MINOR]);
1845
1846                 enabled = nla_get_flag(tb[NFSD_A_VERSION_ENABLED]);
1847
1848                 switch (major) {
1849                 case 4:
1850                         nfsd_minorversion(nn, minor, enabled ? NFSD_SET : NFSD_CLEAR);
1851                         break;
1852                 case 3:
1853                 case 2:
1854                         if (!minor)
1855                                 nfsd_vers(nn, major, enabled ? NFSD_SET : NFSD_CLEAR);
1856                         break;
1857                 default:
1858                         break;
1859                 }
1860         }
1861
1862         mutex_unlock(&nfsd_mutex);
1863
1864         return 0;
1865 }
1866
1867 /**
1868  * nfsd_nl_version_get_doit - get the enabled status for all supported nfs versions
1869  * @skb: reply buffer
1870  * @info: netlink metadata and command arguments
1871  *
1872  * Return 0 on success or a negative errno.
1873  */
1874 int nfsd_nl_version_get_doit(struct sk_buff *skb, struct genl_info *info)
1875 {
1876         struct nfsd_net *nn;
1877         int i, err;
1878         void *hdr;
1879
1880         skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
1881         if (!skb)
1882                 return -ENOMEM;
1883
1884         hdr = genlmsg_iput(skb, info);
1885         if (!hdr) {
1886                 err = -EMSGSIZE;
1887                 goto err_free_msg;
1888         }
1889
1890         mutex_lock(&nfsd_mutex);
1891         nn = net_generic(genl_info_net(info), nfsd_net_id);
1892
1893         for (i = 2; i <= 4; i++) {
1894                 int j;
1895
1896                 for (j = 0; j <= NFSD_SUPPORTED_MINOR_VERSION; j++) {
1897                         struct nlattr *attr;
1898
1899                         /* Don't record any versions the kernel doesn't have
1900                          * compiled in
1901                          */
1902                         if (!nfsd_support_version(i))
1903                                 continue;
1904
1905                         /* NFSv{2,3} does not support minor numbers */
1906                         if (i < 4 && j)
1907                                 continue;
1908
1909                         attr = nla_nest_start(skb,
1910                                               NFSD_A_SERVER_PROTO_VERSION);
1911                         if (!attr) {
1912                                 err = -EINVAL;
1913                                 goto err_nfsd_unlock;
1914                         }
1915
1916                         if (nla_put_u32(skb, NFSD_A_VERSION_MAJOR, i) ||
1917                             nla_put_u32(skb, NFSD_A_VERSION_MINOR, j)) {
1918                                 err = -EINVAL;
1919                                 goto err_nfsd_unlock;
1920                         }
1921
1922                         /* Set the enabled flag if the version is enabled */
1923                         if (nfsd_vers(nn, i, NFSD_TEST) &&
1924                             (i < 4 || nfsd_minorversion(nn, j, NFSD_TEST)) &&
1925                             nla_put_flag(skb, NFSD_A_VERSION_ENABLED)) {
1926                                 err = -EINVAL;
1927                                 goto err_nfsd_unlock;
1928                         }
1929
1930                         nla_nest_end(skb, attr);
1931                 }
1932         }
1933
1934         mutex_unlock(&nfsd_mutex);
1935         genlmsg_end(skb, hdr);
1936
1937         return genlmsg_reply(skb, info);
1938
1939 err_nfsd_unlock:
1940         mutex_unlock(&nfsd_mutex);
1941 err_free_msg:
1942         nlmsg_free(skb);
1943
1944         return err;
1945 }
1946
1947 /**
1948  * nfsd_nl_listener_set_doit - set the nfs running sockets
1949  * @skb: reply buffer
1950  * @info: netlink metadata and command arguments
1951  *
1952  * Return 0 on success or a negative errno.
1953  */
1954 int nfsd_nl_listener_set_doit(struct sk_buff *skb, struct genl_info *info)
1955 {
1956         struct net *net = genl_info_net(info);
1957         struct svc_xprt *xprt, *tmp;
1958         const struct nlattr *attr;
1959         struct svc_serv *serv;
1960         LIST_HEAD(permsocks);
1961         struct nfsd_net *nn;
1962         int err, rem;
1963
1964         mutex_lock(&nfsd_mutex);
1965
1966         err = nfsd_create_serv(net);
1967         if (err) {
1968                 mutex_unlock(&nfsd_mutex);
1969                 return err;
1970         }
1971
1972         nn = net_generic(net, nfsd_net_id);
1973         serv = nn->nfsd_serv;
1974
1975         spin_lock_bh(&serv->sv_lock);
1976
1977         /* Move all of the old listener sockets to a temp list */
1978         list_splice_init(&serv->sv_permsocks, &permsocks);
1979
1980         /*
1981          * Walk the list of server_socks from userland and move any that match
1982          * back to sv_permsocks
1983          */
1984         nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
1985                 struct nlattr *tb[NFSD_A_SOCK_MAX + 1];
1986                 const char *xcl_name;
1987                 struct sockaddr *sa;
1988
1989                 if (nla_type(attr) != NFSD_A_SERVER_SOCK_ADDR)
1990                         continue;
1991
1992                 if (nla_parse_nested(tb, NFSD_A_SOCK_MAX, attr,
1993                                      nfsd_sock_nl_policy, info->extack) < 0)
1994                         continue;
1995
1996                 if (!tb[NFSD_A_SOCK_ADDR] || !tb[NFSD_A_SOCK_TRANSPORT_NAME])
1997                         continue;
1998
1999                 if (nla_len(tb[NFSD_A_SOCK_ADDR]) < sizeof(*sa))
2000                         continue;
2001
2002                 xcl_name = nla_data(tb[NFSD_A_SOCK_TRANSPORT_NAME]);
2003                 sa = nla_data(tb[NFSD_A_SOCK_ADDR]);
2004
2005                 /* Put back any matching sockets */
2006                 list_for_each_entry_safe(xprt, tmp, &permsocks, xpt_list) {
2007                         /* This shouldn't be possible */
2008                         if (WARN_ON_ONCE(xprt->xpt_net != net)) {
2009                                 list_move(&xprt->xpt_list, &serv->sv_permsocks);
2010                                 continue;
2011                         }
2012
2013                         /* If everything matches, put it back */
2014                         if (!strcmp(xprt->xpt_class->xcl_name, xcl_name) &&
2015                             rpc_cmp_addr_port(sa, (struct sockaddr *)&xprt->xpt_local)) {
2016                                 list_move(&xprt->xpt_list, &serv->sv_permsocks);
2017                                 break;
2018                         }
2019                 }
2020         }
2021
2022         /* For now, no removing old sockets while server is running */
2023         if (serv->sv_nrthreads && !list_empty(&permsocks)) {
2024                 list_splice_init(&permsocks, &serv->sv_permsocks);
2025                 spin_unlock_bh(&serv->sv_lock);
2026                 err = -EBUSY;
2027                 goto out_unlock_mtx;
2028         }
2029
2030         /* Close the remaining sockets on the permsocks list */
2031         while (!list_empty(&permsocks)) {
2032                 xprt = list_first_entry(&permsocks, struct svc_xprt, xpt_list);
2033                 list_move(&xprt->xpt_list, &serv->sv_permsocks);
2034
2035                 /*
2036                  * Newly-created sockets are born with the BUSY bit set. Clear
2037                  * it if there are no threads, since nothing can pick it up
2038                  * in that case.
2039                  */
2040                 if (!serv->sv_nrthreads)
2041                         clear_bit(XPT_BUSY, &xprt->xpt_flags);
2042
2043                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
2044                 spin_unlock_bh(&serv->sv_lock);
2045                 svc_xprt_close(xprt);
2046                 spin_lock_bh(&serv->sv_lock);
2047         }
2048
2049         spin_unlock_bh(&serv->sv_lock);
2050
2051         /* walk list of addrs again, open any that still don't exist */
2052         nlmsg_for_each_attr(attr, info->nlhdr, GENL_HDRLEN, rem) {
2053                 struct nlattr *tb[NFSD_A_SOCK_MAX + 1];
2054                 const char *xcl_name;
2055                 struct sockaddr *sa;
2056                 int ret;
2057
2058                 if (nla_type(attr) != NFSD_A_SERVER_SOCK_ADDR)
2059                         continue;
2060
2061                 if (nla_parse_nested(tb, NFSD_A_SOCK_MAX, attr,
2062                                      nfsd_sock_nl_policy, info->extack) < 0)
2063                         continue;
2064
2065                 if (!tb[NFSD_A_SOCK_ADDR] || !tb[NFSD_A_SOCK_TRANSPORT_NAME])
2066                         continue;
2067
2068                 if (nla_len(tb[NFSD_A_SOCK_ADDR]) < sizeof(*sa))
2069                         continue;
2070
2071                 xcl_name = nla_data(tb[NFSD_A_SOCK_TRANSPORT_NAME]);
2072                 sa = nla_data(tb[NFSD_A_SOCK_ADDR]);
2073
2074                 xprt = svc_find_listener(serv, xcl_name, net, sa);
2075                 if (xprt) {
2076                         svc_xprt_put(xprt);
2077                         continue;
2078                 }
2079
2080                 ret = svc_xprt_create_from_sa(serv, xcl_name, net, sa, 0,
2081                                               get_current_cred());
2082                 /* always save the latest error */
2083                 if (ret < 0)
2084                         err = ret;
2085         }
2086
2087         if (!serv->sv_nrthreads && list_empty(&nn->nfsd_serv->sv_permsocks))
2088                 nfsd_destroy_serv(net);
2089
2090 out_unlock_mtx:
2091         mutex_unlock(&nfsd_mutex);
2092
2093         return err;
2094 }
2095
2096 /**
2097  * nfsd_nl_listener_get_doit - get the nfs running listeners
2098  * @skb: reply buffer
2099  * @info: netlink metadata and command arguments
2100  *
2101  * Return 0 on success or a negative errno.
2102  */
2103 int nfsd_nl_listener_get_doit(struct sk_buff *skb, struct genl_info *info)
2104 {
2105         struct svc_xprt *xprt;
2106         struct svc_serv *serv;
2107         struct nfsd_net *nn;
2108         void *hdr;
2109         int err;
2110
2111         skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
2112         if (!skb)
2113                 return -ENOMEM;
2114
2115         hdr = genlmsg_iput(skb, info);
2116         if (!hdr) {
2117                 err = -EMSGSIZE;
2118                 goto err_free_msg;
2119         }
2120
2121         mutex_lock(&nfsd_mutex);
2122         nn = net_generic(genl_info_net(info), nfsd_net_id);
2123
2124         /* no nfs server? Just send empty socket list */
2125         if (!nn->nfsd_serv)
2126                 goto out_unlock_mtx;
2127
2128         serv = nn->nfsd_serv;
2129         spin_lock_bh(&serv->sv_lock);
2130         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) {
2131                 struct nlattr *attr;
2132
2133                 attr = nla_nest_start(skb, NFSD_A_SERVER_SOCK_ADDR);
2134                 if (!attr) {
2135                         err = -EINVAL;
2136                         goto err_serv_unlock;
2137                 }
2138
2139                 if (nla_put_string(skb, NFSD_A_SOCK_TRANSPORT_NAME,
2140                                    xprt->xpt_class->xcl_name) ||
2141                     nla_put(skb, NFSD_A_SOCK_ADDR,
2142                             sizeof(struct sockaddr_storage),
2143                             &xprt->xpt_local)) {
2144                         err = -EINVAL;
2145                         goto err_serv_unlock;
2146                 }
2147
2148                 nla_nest_end(skb, attr);
2149         }
2150         spin_unlock_bh(&serv->sv_lock);
2151 out_unlock_mtx:
2152         mutex_unlock(&nfsd_mutex);
2153         genlmsg_end(skb, hdr);
2154
2155         return genlmsg_reply(skb, info);
2156
2157 err_serv_unlock:
2158         spin_unlock_bh(&serv->sv_lock);
2159         mutex_unlock(&nfsd_mutex);
2160 err_free_msg:
2161         nlmsg_free(skb);
2162
2163         return err;
2164 }
2165
2166 /**
2167  * nfsd_nl_pool_mode_set_doit - set the number of running threads
2168  * @skb: reply buffer
2169  * @info: netlink metadata and command arguments
2170  *
2171  * Return 0 on success or a negative errno.
2172  */
2173 int nfsd_nl_pool_mode_set_doit(struct sk_buff *skb, struct genl_info *info)
2174 {
2175         const struct nlattr *attr;
2176
2177         if (GENL_REQ_ATTR_CHECK(info, NFSD_A_POOL_MODE_MODE))
2178                 return -EINVAL;
2179
2180         attr = info->attrs[NFSD_A_POOL_MODE_MODE];
2181         return sunrpc_set_pool_mode(nla_data(attr));
2182 }
2183
2184 /**
2185  * nfsd_nl_pool_mode_get_doit - get info about pool_mode
2186  * @skb: reply buffer
2187  * @info: netlink metadata and command arguments
2188  *
2189  * Return 0 on success or a negative errno.
2190  */
2191 int nfsd_nl_pool_mode_get_doit(struct sk_buff *skb, struct genl_info *info)
2192 {
2193         struct net *net = genl_info_net(info);
2194         char buf[16];
2195         void *hdr;
2196         int err;
2197
2198         if (sunrpc_get_pool_mode(buf, ARRAY_SIZE(buf)) >= ARRAY_SIZE(buf))
2199                 return -ERANGE;
2200
2201         skb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
2202         if (!skb)
2203                 return -ENOMEM;
2204
2205         err = -EMSGSIZE;
2206         hdr = genlmsg_iput(skb, info);
2207         if (!hdr)
2208                 goto err_free_msg;
2209
2210         err = nla_put_string(skb, NFSD_A_POOL_MODE_MODE, buf) |
2211               nla_put_u32(skb, NFSD_A_POOL_MODE_NPOOLS, nfsd_nrpools(net));
2212         if (err)
2213                 goto err_free_msg;
2214
2215         genlmsg_end(skb, hdr);
2216         return genlmsg_reply(skb, info);
2217
2218 err_free_msg:
2219         nlmsg_free(skb);
2220         return err;
2221 }
2222
2223 /**
2224  * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
2225  * @net: a freshly-created network namespace
2226  *
2227  * This information stays around as long as the network namespace is
2228  * alive whether or not there is an NFSD instance running in the
2229  * namespace.
2230  *
2231  * Returns zero on success, or a negative errno otherwise.
2232  */
2233 static __net_init int nfsd_net_init(struct net *net)
2234 {
2235         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2236         int retval;
2237         int i;
2238
2239         retval = nfsd_export_init(net);
2240         if (retval)
2241                 goto out_export_error;
2242         retval = nfsd_idmap_init(net);
2243         if (retval)
2244                 goto out_idmap_error;
2245         retval = percpu_counter_init_many(nn->counter, 0, GFP_KERNEL,
2246                                           NFSD_STATS_COUNTERS_NUM);
2247         if (retval)
2248                 goto out_repcache_error;
2249         memset(&nn->nfsd_svcstats, 0, sizeof(nn->nfsd_svcstats));
2250         nn->nfsd_svcstats.program = &nfsd_programs[0];
2251         for (i = 0; i < sizeof(nn->nfsd_versions); i++)
2252                 nn->nfsd_versions[i] = nfsd_support_version(i);
2253         for (i = 0; i < sizeof(nn->nfsd4_minorversions); i++)
2254                 nn->nfsd4_minorversions[i] = nfsd_support_version(4);
2255         nn->nfsd_info.mutex = &nfsd_mutex;
2256         nn->nfsd_serv = NULL;
2257         nfsd4_init_leases_net(nn);
2258         get_random_bytes(&nn->siphash_key, sizeof(nn->siphash_key));
2259         seqlock_init(&nn->writeverf_lock);
2260         nfsd_proc_stat_init(net);
2261 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2262         INIT_LIST_HEAD(&nn->local_clients);
2263 #endif
2264         return 0;
2265
2266 out_repcache_error:
2267         nfsd_idmap_shutdown(net);
2268 out_idmap_error:
2269         nfsd_export_shutdown(net);
2270 out_export_error:
2271         return retval;
2272 }
2273
2274 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2275 /**
2276  * nfsd_net_pre_exit - Disconnect localio clients from net namespace
2277  * @net: a network namespace that is about to be destroyed
2278  *
2279  * This invalidated ->net pointers held by localio clients
2280  * while they can still safely access nn->counter.
2281  */
2282 static __net_exit void nfsd_net_pre_exit(struct net *net)
2283 {
2284         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2285
2286         nfs_uuid_invalidate_clients(&nn->local_clients);
2287 }
2288 #endif
2289
2290 /**
2291  * nfsd_net_exit - Release the nfsd_net portion of a net namespace
2292  * @net: a network namespace that is about to be destroyed
2293  *
2294  */
2295 static __net_exit void nfsd_net_exit(struct net *net)
2296 {
2297         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2298
2299         nfsd_proc_stat_shutdown(net);
2300         percpu_counter_destroy_many(nn->counter, NFSD_STATS_COUNTERS_NUM);
2301         nfsd_idmap_shutdown(net);
2302         nfsd_export_shutdown(net);
2303 }
2304
2305 static struct pernet_operations nfsd_net_ops = {
2306         .init = nfsd_net_init,
2307 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2308         .pre_exit = nfsd_net_pre_exit,
2309 #endif
2310         .exit = nfsd_net_exit,
2311         .id   = &nfsd_net_id,
2312         .size = sizeof(struct nfsd_net),
2313 };
2314
2315 static int __init init_nfsd(void)
2316 {
2317         int retval;
2318
2319         retval = nfsd4_init_slabs();
2320         if (retval)
2321                 return retval;
2322         retval = nfsd4_init_pnfs();
2323         if (retval)
2324                 goto out_free_slabs;
2325         retval = nfsd_drc_slab_create();
2326         if (retval)
2327                 goto out_free_pnfs;
2328         nfsd_lockd_init();      /* lockd->nfsd callbacks */
2329         retval = create_proc_exports_entry();
2330         if (retval)
2331                 goto out_free_lockd;
2332         retval = register_pernet_subsys(&nfsd_net_ops);
2333         if (retval < 0)
2334                 goto out_free_exports;
2335         retval = register_cld_notifier();
2336         if (retval)
2337                 goto out_free_subsys;
2338         retval = nfsd4_create_laundry_wq();
2339         if (retval)
2340                 goto out_free_cld;
2341         retval = register_filesystem(&nfsd_fs_type);
2342         if (retval)
2343                 goto out_free_all;
2344         retval = genl_register_family(&nfsd_nl_family);
2345         if (retval)
2346                 goto out_free_all;
2347         nfsd_localio_ops_init();
2348
2349         return 0;
2350 out_free_all:
2351         nfsd4_destroy_laundry_wq();
2352 out_free_cld:
2353         unregister_cld_notifier();
2354 out_free_subsys:
2355         unregister_pernet_subsys(&nfsd_net_ops);
2356 out_free_exports:
2357         remove_proc_entry("fs/nfs/exports", NULL);
2358         remove_proc_entry("fs/nfs", NULL);
2359 out_free_lockd:
2360         nfsd_lockd_shutdown();
2361         nfsd_drc_slab_free();
2362 out_free_pnfs:
2363         nfsd4_exit_pnfs();
2364 out_free_slabs:
2365         nfsd4_free_slabs();
2366         return retval;
2367 }
2368
2369 static void __exit exit_nfsd(void)
2370 {
2371         genl_unregister_family(&nfsd_nl_family);
2372         unregister_filesystem(&nfsd_fs_type);
2373         nfsd4_destroy_laundry_wq();
2374         unregister_cld_notifier();
2375         unregister_pernet_subsys(&nfsd_net_ops);
2376         nfsd_drc_slab_free();
2377         remove_proc_entry("fs/nfs/exports", NULL);
2378         remove_proc_entry("fs/nfs", NULL);
2379         nfsd_lockd_shutdown();
2380         nfsd4_free_slabs();
2381         nfsd4_exit_pnfs();
2382 }
2383
2384 MODULE_AUTHOR("Olaf Kirch <[email protected]>");
2385 MODULE_DESCRIPTION("In-kernel NFS server");
2386 MODULE_LICENSE("GPL");
2387 module_init(init_nfsd)
2388 module_exit(exit_nfsd)