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 <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/gss_krb5_enctypes.h>
  18 #include <linux/sunrpc/rpc_pipe_fs.h>
  19 #include <linux/module.h>
  20 #include <linux/fsnotify.h>
  21
  22 #include "idmap.h"
  23 #include "nfsd.h"
  24 #include "cache.h"
  25 #include "state.h"
  26 #include "netns.h"
  27 #include "pnfs.h"
  28
  29 /*
  30  *      We have a single directory with several nodes in it.
  31  */
  32 enum {
  33         NFSD_Root = 1,
  34         NFSD_List,
  35         NFSD_Export_Stats,
  36         NFSD_Export_features,
  37         NFSD_Fh,
  38         NFSD_FO_UnlockIP,
  39         NFSD_FO_UnlockFS,
  40         NFSD_Threads,
  41         NFSD_Pool_Threads,
  42         NFSD_Pool_Stats,
  43         NFSD_Reply_Cache_Stats,
  44         NFSD_Versions,
  45         NFSD_Ports,
  46         NFSD_MaxBlkSize,
  47         NFSD_MaxConnections,
  48         NFSD_SupportedEnctypes,
  49         /*
  50          * The below MUST come last.  Otherwise we leave a hole in nfsd_files[]
  51          * with !CONFIG_NFSD_V4 and simple_fill_super() goes oops
  52          */
  53 #ifdef CONFIG_NFSD_V4
  54         NFSD_Leasetime,
  55         NFSD_Gracetime,
  56         NFSD_RecoveryDir,
  57         NFSD_V4EndGrace,
  58 #endif
  59         NFSD_MaxReserved
  60 };
  61
  62 /*
  63  * write() for these nodes.
  64  */
  65 static ssize_t write_filehandle(struct file *file, char *buf, size_t size);
  66 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size);
  67 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size);
  68 static ssize_t write_threads(struct file *file, char *buf, size_t size);
  69 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size);
  70 static ssize_t write_versions(struct file *file, char *buf, size_t size);
  71 static ssize_t write_ports(struct file *file, char *buf, size_t size);
  72 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size);
  73 static ssize_t write_maxconn(struct file *file, char *buf, size_t size);
  74 #ifdef CONFIG_NFSD_V4
  75 static ssize_t write_leasetime(struct file *file, char *buf, size_t size);
  76 static ssize_t write_gracetime(struct file *file, char *buf, size_t size);
  77 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size);
  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         [NFSD_RecoveryDir] = write_recoverydir,
  95         [NFSD_V4EndGrace] = write_v4_end_grace,
  96 #endif
  97 };
  98
  99 static ssize_t nfsctl_transaction_write(struct file *file, const char __user *buf, size_t size, loff_t *pos)
 100 {
 101         ino_t ino =  file_inode(file)->i_ino;
 102         char *data;
 103         ssize_t rv;
 104
 105         if (ino >= ARRAY_SIZE(write_op) || !write_op[ino])
 106                 return -EINVAL;
 107
 108         data = simple_transaction_get(file, buf, size);
 109         if (IS_ERR(data))
 110                 return PTR_ERR(data);
 111
 112         rv =  write_op[ino](file, data, size);
 113         if (rv >= 0) {
 114                 simple_transaction_set(file, rv);
 115                 rv = size;
 116         }
 117         return rv;
 118 }
 119
 120 static ssize_t nfsctl_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
 121 {
 122         if (! file->private_data) {
 123                 /* An attempt to read a transaction file without writing
 124                  * causes a 0-byte write so that the file can return
 125                  * state information
 126                  */
 127                 ssize_t rv = nfsctl_transaction_write(file, buf, 0, pos);
 128                 if (rv < 0)
 129                         return rv;
 130         }
 131         return simple_transaction_read(file, buf, size, pos);
 132 }
 133
 134 static const struct file_operations transaction_ops = {
 135         .write          = nfsctl_transaction_write,
 136         .read           = nfsctl_transaction_read,
 137         .release        = simple_transaction_release,
 138         .llseek         = default_llseek,
 139 };
 140
 141 static int exports_net_open(struct net *net, struct file *file)
 142 {
 143         int err;
 144         struct seq_file *seq;
 145         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 146
 147         err = seq_open(file, &nfs_exports_op);
 148         if (err)
 149                 return err;
 150
 151         seq = file->private_data;
 152         seq->private = nn->svc_export_cache;
 153         return 0;
 154 }
 155
 156 static int exports_proc_open(struct inode *inode, struct file *file)
 157 {
 158         return exports_net_open(current->nsproxy->net_ns, file);
 159 }
 160
 161 static const struct proc_ops exports_proc_ops = {
 162         .proc_open      = exports_proc_open,
 163         .proc_read      = seq_read,
 164         .proc_lseek     = seq_lseek,
 165         .proc_release   = seq_release,
 166 };
 167
 168 static int exports_nfsd_open(struct inode *inode, struct file *file)
 169 {
 170         return exports_net_open(inode->i_sb->s_fs_info, file);
 171 }
 172
 173 static const struct file_operations exports_nfsd_operations = {
 174         .open           = exports_nfsd_open,
 175         .read           = seq_read,
 176         .llseek         = seq_lseek,
 177         .release        = seq_release,
 178 };
 179
 180 static int export_features_show(struct seq_file *m, void *v)
 181 {
 182         seq_printf(m, "0x%x 0x%x\n", NFSEXP_ALLFLAGS, NFSEXP_SECINFO_FLAGS);
 183         return 0;
 184 }
 185
 186 static int export_features_open(struct inode *inode, struct file *file)
 187 {
 188         return single_open(file, export_features_show, NULL);
 189 }
 190
 191 static const struct file_operations export_features_operations = {
 192         .open           = export_features_open,
 193         .read           = seq_read,
 194         .llseek         = seq_lseek,
 195         .release        = single_release,
 196 };
 197
 198 #if defined(CONFIG_SUNRPC_GSS) || defined(CONFIG_SUNRPC_GSS_MODULE)
 199 static int supported_enctypes_show(struct seq_file *m, void *v)
 200 {
 201         seq_printf(m, KRB5_SUPPORTED_ENCTYPES);
 202         return 0;
 203 }
 204
 205 static int supported_enctypes_open(struct inode *inode, struct file *file)
 206 {
 207         return single_open(file, supported_enctypes_show, NULL);
 208 }
 209
 210 static const struct file_operations supported_enctypes_ops = {
 211         .open           = supported_enctypes_open,
 212         .read           = seq_read,
 213         .llseek         = seq_lseek,
 214         .release        = single_release,
 215 };
 216 #endif /* CONFIG_SUNRPC_GSS or CONFIG_SUNRPC_GSS_MODULE */
 217
 218 static const struct file_operations pool_stats_operations = {
 219         .open           = nfsd_pool_stats_open,
 220         .read           = seq_read,
 221         .llseek         = seq_lseek,
 222         .release        = nfsd_pool_stats_release,
 223 };
 224
 225 static const struct file_operations reply_cache_stats_operations = {
 226         .open           = nfsd_reply_cache_stats_open,
 227         .read           = seq_read,
 228         .llseek         = seq_lseek,
 229         .release        = single_release,
 230 };
 231
 232 /*----------------------------------------------------------------------------*/
 233 /*
 234  * payload - write methods
 235  */
 236
 237 static inline struct net *netns(struct file *file)
 238 {
 239         return file_inode(file)->i_sb->s_fs_info;
 240 }
 241
 242 /*
 243  * write_unlock_ip - Release all locks used by a client
 244  *
 245  * Experimental.
 246  *
 247  * Input:
 248  *                      buf:    '\n'-terminated C string containing a
 249  *                              presentation format IP address
 250  *                      size:   length of C string in @buf
 251  * Output:
 252  *      On success:     returns zero if all specified locks were released;
 253  *                      returns one if one or more locks were not released
 254  *      On error:       return code is negative errno value
 255  */
 256 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size)
 257 {
 258         struct sockaddr_storage address;
 259         struct sockaddr *sap = (struct sockaddr *)&address;
 260         size_t salen = sizeof(address);
 261         char *fo_path;
 262         struct net *net = netns(file);
 263
 264         /* sanity check */
 265         if (size == 0)
 266                 return -EINVAL;
 267
 268         if (buf[size-1] != '\n')
 269                 return -EINVAL;
 270
 271         fo_path = buf;
 272         if (qword_get(&buf, fo_path, size) < 0)
 273                 return -EINVAL;
 274
 275         if (rpc_pton(net, fo_path, size, sap, salen) == 0)
 276                 return -EINVAL;
 277
 278         return nlmsvc_unlock_all_by_ip(sap);
 279 }
 280
 281 /*
 282  * write_unlock_fs - Release all locks on a local file system
 283  *
 284  * Experimental.
 285  *
 286  * Input:
 287  *                      buf:    '\n'-terminated C string containing the
 288  *                              absolute pathname of a local file system
 289  *                      size:   length of C string in @buf
 290  * Output:
 291  *      On success:     returns zero if all specified locks were released;
 292  *                      returns one if one or more locks were not released
 293  *      On error:       return code is negative errno value
 294  */
 295 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size)
 296 {
 297         struct path path;
 298         char *fo_path;
 299         int error;
 300
 301         /* sanity check */
 302         if (size == 0)
 303                 return -EINVAL;
 304
 305         if (buf[size-1] != '\n')
 306                 return -EINVAL;
 307
 308         fo_path = buf;
 309         if (qword_get(&buf, fo_path, size) < 0)
 310                 return -EINVAL;
 311
 312         error = kern_path(fo_path, 0, &path);
 313         if (error)
 314                 return error;
 315
 316         /*
 317          * XXX: Needs better sanity checking.  Otherwise we could end up
 318          * releasing locks on the wrong file system.
 319          *
 320          * For example:
 321          * 1.  Does the path refer to a directory?
 322          * 2.  Is that directory a mount point, or
 323          * 3.  Is that directory the root of an exported file system?
 324          */
 325         error = nlmsvc_unlock_all_by_sb(path.dentry->d_sb);
 326
 327         path_put(&path);
 328         return error;
 329 }
 330
 331 /*
 332  * write_filehandle - Get a variable-length NFS file handle by path
 333  *
 334  * On input, the buffer contains a '\n'-terminated C string comprised of
 335  * three alphanumeric words separated by whitespace.  The string may
 336  * contain escape sequences.
 337  *
 338  * Input:
 339  *                      buf:
 340  *                              domain:         client domain name
 341  *                              path:           export pathname
 342  *                              maxsize:        numeric maximum size of
 343  *                                              @buf
 344  *                      size:   length of C string in @buf
 345  * Output:
 346  *      On success:     passed-in buffer filled with '\n'-terminated C
 347  *                      string containing a ASCII hex text version
 348  *                      of the NFS file handle;
 349  *                      return code is the size in bytes of the string
 350  *      On error:       return code is negative errno value
 351  */
 352 static ssize_t write_filehandle(struct file *file, char *buf, size_t size)
 353 {
 354         char *dname, *path;
 355         int maxsize;
 356         char *mesg = buf;
 357         int len;
 358         struct auth_domain *dom;
 359         struct knfsd_fh fh;
 360
 361         if (size == 0)
 362                 return -EINVAL;
 363
 364         if (buf[size-1] != '\n')
 365                 return -EINVAL;
 366         buf[size-1] = 0;
 367
 368         dname = mesg;
 369         len = qword_get(&mesg, dname, size);
 370         if (len <= 0)
 371                 return -EINVAL;
 372
 373         path = dname+len+1;
 374         len = qword_get(&mesg, path, size);
 375         if (len <= 0)
 376                 return -EINVAL;
 377
 378         len = get_int(&mesg, &maxsize);
 379         if (len)
 380                 return len;
 381
 382         if (maxsize < NFS_FHSIZE)
 383                 return -EINVAL;
 384         maxsize = min(maxsize, NFS3_FHSIZE);
 385
 386         if (qword_get(&mesg, mesg, size)>0)
 387                 return -EINVAL;
 388
 389         /* we have all the words, they are in buf.. */
 390         dom = unix_domain_find(dname);
 391         if (!dom)
 392                 return -ENOMEM;
 393
 394         len = exp_rootfh(netns(file), dom, path, &fh,  maxsize);
 395         auth_domain_put(dom);
 396         if (len)
 397                 return len;
 398
 399         mesg = buf;
 400         len = SIMPLE_TRANSACTION_LIMIT;
 401         qword_addhex(&mesg, &len, (char*)&fh.fh_base, fh.fh_size);
 402         mesg[-1] = '\n';
 403         return mesg - buf;
 404 }
 405
 406 /*
 407  * write_threads - Start NFSD, or report the current number of running threads
 408  *
 409  * Input:
 410  *                      buf:            ignored
 411  *                      size:           zero
 412  * Output:
 413  *      On success:     passed-in buffer filled with '\n'-terminated C
 414  *                      string numeric value representing the number of
 415  *                      running NFSD threads;
 416  *                      return code is the size in bytes of the string
 417  *      On error:       return code is zero
 418  *
 419  * OR
 420  *
 421  * Input:
 422  *                      buf:            C string containing an unsigned
 423  *                                      integer value representing the
 424  *                                      number of NFSD threads to start
 425  *                      size:           non-zero length of C string in @buf
 426  * Output:
 427  *      On success:     NFS service is started;
 428  *                      passed-in buffer filled with '\n'-terminated C
 429  *                      string numeric value representing the number of
 430  *                      running NFSD threads;
 431  *                      return code is the size in bytes of the string
 432  *      On error:       return code is zero or a negative errno value
 433  */
 434 static ssize_t write_threads(struct file *file, char *buf, size_t size)
 435 {
 436         char *mesg = buf;
 437         int rv;
 438         struct net *net = netns(file);
 439
 440         if (size > 0) {
 441                 int newthreads;
 442                 rv = get_int(&mesg, &newthreads);
 443                 if (rv)
 444                         return rv;
 445                 if (newthreads < 0)
 446                         return -EINVAL;
 447                 rv = nfsd_svc(newthreads, net, file->f_cred);
 448                 if (rv < 0)
 449                         return rv;
 450         } else
 451                 rv = nfsd_nrthreads(net);
 452
 453         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n", rv);
 454 }
 455
 456 /*
 457  * write_pool_threads - Set or report the current number of threads per pool
 458  *
 459  * Input:
 460  *                      buf:            ignored
 461  *                      size:           zero
 462  *
 463  * OR
 464  *
 465  * Input:
 466  *                      buf:            C string containing whitespace-
 467  *                                      separated unsigned integer values
 468  *                                      representing the number of NFSD
 469  *                                      threads to start in each pool
 470  *                      size:           non-zero length of C string in @buf
 471  * Output:
 472  *      On success:     passed-in buffer filled with '\n'-terminated C
 473  *                      string containing integer values representing the
 474  *                      number of NFSD threads in each pool;
 475  *                      return code is the size in bytes of the string
 476  *      On error:       return code is zero or a negative errno value
 477  */
 478 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size)
 479 {
 480         /* if size > 0, look for an array of number of threads per node
 481          * and apply them  then write out number of threads per node as reply
 482          */
 483         char *mesg = buf;
 484         int i;
 485         int rv;
 486         int len;
 487         int npools;
 488         int *nthreads;
 489         struct net *net = netns(file);
 490
 491         mutex_lock(&nfsd_mutex);
 492         npools = nfsd_nrpools(net);
 493         if (npools == 0) {
 494                 /*
 495                  * NFS is shut down.  The admin can start it by
 496                  * writing to the threads file but NOT the pool_threads
 497                  * file, sorry.  Report zero threads.
 498                  */
 499                 mutex_unlock(&nfsd_mutex);
 500                 strcpy(buf, "0\n");
 501                 return strlen(buf);
 502         }
 503
 504         nthreads = kcalloc(npools, sizeof(int), GFP_KERNEL);
 505         rv = -ENOMEM;
 506         if (nthreads == NULL)
 507                 goto out_free;
 508
 509         if (size > 0) {
 510                 for (i = 0; i < npools; i++) {
 511                         rv = get_int(&mesg, &nthreads[i]);
 512                         if (rv == -ENOENT)
 513                                 break;          /* fewer numbers than pools */
 514                         if (rv)
 515                                 goto out_free;  /* syntax error */
 516                         rv = -EINVAL;
 517                         if (nthreads[i] < 0)
 518                                 goto out_free;
 519                 }
 520                 rv = nfsd_set_nrthreads(i, nthreads, net);
 521                 if (rv)
 522                         goto out_free;
 523         }
 524
 525         rv = nfsd_get_nrthreads(npools, nthreads, net);
 526         if (rv)
 527                 goto out_free;
 528
 529         mesg = buf;
 530         size = SIMPLE_TRANSACTION_LIMIT;
 531         for (i = 0; i < npools && size > 0; i++) {
 532                 snprintf(mesg, size, "%d%c", nthreads[i], (i == npools-1 ? '\n' : ' '));
 533                 len = strlen(mesg);
 534                 size -= len;
 535                 mesg += len;
 536         }
 537         rv = mesg - buf;
 538 out_free:
 539         kfree(nthreads);
 540         mutex_unlock(&nfsd_mutex);
 541         return rv;
 542 }
 543
 544 static ssize_t
 545 nfsd_print_version_support(struct nfsd_net *nn, char *buf, int remaining,
 546                 const char *sep, unsigned vers, int minor)
 547 {
 548         const char *format = minor < 0 ? "%s%c%u" : "%s%c%u.%u";
 549         bool supported = !!nfsd_vers(nn, vers, NFSD_TEST);
 550
 551         if (vers == 4 && minor >= 0 &&
 552             !nfsd_minorversion(nn, minor, NFSD_TEST))
 553                 supported = false;
 554         if (minor == 0 && supported)
 555                 /*
 556                  * special case for backward compatability.
 557                  * +4.0 is never reported, it is implied by
 558                  * +4, unless -4.0 is present.
 559                  */
 560                 return 0;
 561         return snprintf(buf, remaining, format, sep,
 562                         supported ? '+' : '-', vers, minor);
 563 }
 564
 565 static ssize_t __write_versions(struct file *file, char *buf, size_t size)
 566 {
 567         char *mesg = buf;
 568         char *vers, *minorp, sign;
 569         int len, num, remaining;
 570         ssize_t tlen = 0;
 571         char *sep;
 572         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 573
 574         if (size>0) {
 575                 if (nn->nfsd_serv)
 576                         /* Cannot change versions without updating
 577                          * nn->nfsd_serv->sv_xdrsize, and reallocing
 578                          * rq_argp and rq_resp
 579                          */
 580                         return -EBUSY;
 581                 if (buf[size-1] != '\n')
 582                         return -EINVAL;
 583                 buf[size-1] = 0;
 584
 585                 vers = mesg;
 586                 len = qword_get(&mesg, vers, size);
 587                 if (len <= 0) return -EINVAL;
 588                 do {
 589                         enum vers_op cmd;
 590                         unsigned minor;
 591                         sign = *vers;
 592                         if (sign == '+' || sign == '-')
 593                                 num = simple_strtol((vers+1), &minorp, 0);
 594                         else
 595                                 num = simple_strtol(vers, &minorp, 0);
 596                         if (*minorp == '.') {
 597                                 if (num != 4)
 598                                         return -EINVAL;
 599                                 if (kstrtouint(minorp+1, 0, &minor) < 0)
 600                                         return -EINVAL;
 601                         }
 602
 603                         cmd = sign == '-' ? NFSD_CLEAR : NFSD_SET;
 604                         switch(num) {
 605                         case 2:
 606                         case 3:
 607                                 nfsd_vers(nn, num, cmd);
 608                                 break;
 609                         case 4:
 610                                 if (*minorp == '.') {
 611                                         if (nfsd_minorversion(nn, minor, cmd) < 0)
 612                                                 return -EINVAL;
 613                                 } else if ((cmd == NFSD_SET) != nfsd_vers(nn, num, NFSD_TEST)) {
 614                                         /*
 615                                          * Either we have +4 and no minors are enabled,
 616                                          * or we have -4 and at least one minor is enabled.
 617                                          * In either case, propagate 'cmd' to all minors.
 618                                          */
 619                                         minor = 0;
 620                                         while (nfsd_minorversion(nn, minor, cmd) >= 0)
 621                                                 minor++;
 622                                 }
 623                                 break;
 624                         default:
 625                                 return -EINVAL;
 626                         }
 627                         vers += len + 1;
 628                 } while ((len = qword_get(&mesg, vers, size)) > 0);
 629                 /* If all get turned off, turn them back on, as
 630                  * having no versions is BAD
 631                  */
 632                 nfsd_reset_versions(nn);
 633         }
 634
 635         /* Now write current state into reply buffer */
 636         len = 0;
 637         sep = "";
 638         remaining = SIMPLE_TRANSACTION_LIMIT;
 639         for (num=2 ; num <= 4 ; num++) {
 640                 int minor;
 641                 if (!nfsd_vers(nn, num, NFSD_AVAIL))
 642                         continue;
 643
 644                 minor = -1;
 645                 do {
 646                         len = nfsd_print_version_support(nn, buf, remaining,
 647                                         sep, num, minor);
 648                         if (len >= remaining)
 649                                 goto out;
 650                         remaining -= len;
 651                         buf += len;
 652                         tlen += len;
 653                         minor++;
 654                         if (len)
 655                                 sep = " ";
 656                 } while (num == 4 && minor <= NFSD_SUPPORTED_MINOR_VERSION);
 657         }
 658 out:
 659         len = snprintf(buf, remaining, "\n");
 660         if (len >= remaining)
 661                 return -EINVAL;
 662         return tlen + len;
 663 }
 664
 665 /*
 666  * write_versions - Set or report the available NFS protocol versions
 667  *
 668  * Input:
 669  *                      buf:            ignored
 670  *                      size:           zero
 671  * Output:
 672  *      On success:     passed-in buffer filled with '\n'-terminated C
 673  *                      string containing positive or negative integer
 674  *                      values representing the current status of each
 675  *                      protocol version;
 676  *                      return code is the size in bytes of the string
 677  *      On error:       return code is zero or a negative errno value
 678  *
 679  * OR
 680  *
 681  * Input:
 682  *                      buf:            C string containing whitespace-
 683  *                                      separated positive or negative
 684  *                                      integer values representing NFS
 685  *                                      protocol versions to enable ("+n")
 686  *                                      or disable ("-n")
 687  *                      size:           non-zero length of C string in @buf
 688  * Output:
 689  *      On success:     status of zero or more protocol versions has
 690  *                      been updated; passed-in buffer filled with
 691  *                      '\n'-terminated C string containing positive
 692  *                      or negative integer values representing the
 693  *                      current status of each protocol version;
 694  *                      return code is the size in bytes of the string
 695  *      On error:       return code is zero or a negative errno value
 696  */
 697 static ssize_t write_versions(struct file *file, char *buf, size_t size)
 698 {
 699         ssize_t rv;
 700
 701         mutex_lock(&nfsd_mutex);
 702         rv = __write_versions(file, buf, size);
 703         mutex_unlock(&nfsd_mutex);
 704         return rv;
 705 }
 706
 707 /*
 708  * Zero-length write.  Return a list of NFSD's current listener
 709  * transports.
 710  */
 711 static ssize_t __write_ports_names(char *buf, struct net *net)
 712 {
 713         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 714
 715         if (nn->nfsd_serv == NULL)
 716                 return 0;
 717         return svc_xprt_names(nn->nfsd_serv, buf, SIMPLE_TRANSACTION_LIMIT);
 718 }
 719
 720 /*
 721  * A single 'fd' number was written, in which case it must be for
 722  * a socket of a supported family/protocol, and we use it as an
 723  * nfsd listener.
 724  */
 725 static ssize_t __write_ports_addfd(char *buf, struct net *net, const struct cred *cred)
 726 {
 727         char *mesg = buf;
 728         int fd, err;
 729         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 730
 731         err = get_int(&mesg, &fd);
 732         if (err != 0 || fd < 0)
 733                 return -EINVAL;
 734
 735         if (svc_alien_sock(net, fd)) {
 736                 printk(KERN_ERR "%s: socket net is different to NFSd's one\n", __func__);
 737                 return -EINVAL;
 738         }
 739
 740         err = nfsd_create_serv(net);
 741         if (err != 0)
 742                 return err;
 743
 744         err = svc_addsock(nn->nfsd_serv, fd, buf, SIMPLE_TRANSACTION_LIMIT, cred);
 745         if (err < 0) {
 746                 nfsd_destroy(net);
 747                 return err;
 748         }
 749
 750         /* Decrease the count, but don't shut down the service */
 751         nn->nfsd_serv->sv_nrthreads--;
 752         return err;
 753 }
 754
 755 /*
 756  * A transport listener is added by writing it's transport name and
 757  * a port number.
 758  */
 759 static ssize_t __write_ports_addxprt(char *buf, struct net *net, const struct cred *cred)
 760 {
 761         char transport[16];
 762         struct svc_xprt *xprt;
 763         int port, err;
 764         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 765
 766         if (sscanf(buf, "%15s %5u", transport, &port) != 2)
 767                 return -EINVAL;
 768
 769         if (port < 1 || port > USHRT_MAX)
 770                 return -EINVAL;
 771
 772         err = nfsd_create_serv(net);
 773         if (err != 0)
 774                 return err;
 775
 776         err = svc_create_xprt(nn->nfsd_serv, transport, net,
 777                                 PF_INET, port, SVC_SOCK_ANONYMOUS, cred);
 778         if (err < 0)
 779                 goto out_err;
 780
 781         err = svc_create_xprt(nn->nfsd_serv, transport, net,
 782                                 PF_INET6, port, SVC_SOCK_ANONYMOUS, cred);
 783         if (err < 0 && err != -EAFNOSUPPORT)
 784                 goto out_close;
 785
 786         /* Decrease the count, but don't shut down the service */
 787         nn->nfsd_serv->sv_nrthreads--;
 788         return 0;
 789 out_close:
 790         xprt = svc_find_xprt(nn->nfsd_serv, transport, net, PF_INET, port);
 791         if (xprt != NULL) {
 792                 svc_close_xprt(xprt);
 793                 svc_xprt_put(xprt);
 794         }
 795 out_err:
 796         nfsd_destroy(net);
 797         return err;
 798 }
 799
 800 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
 801                              struct net *net)
 802 {
 803         if (size == 0)
 804                 return __write_ports_names(buf, net);
 805
 806         if (isdigit(buf[0]))
 807                 return __write_ports_addfd(buf, net, file->f_cred);
 808
 809         if (isalpha(buf[0]))
 810                 return __write_ports_addxprt(buf, net, file->f_cred);
 811
 812         return -EINVAL;
 813 }
 814
 815 /*
 816  * write_ports - Pass a socket file descriptor or transport name to listen on
 817  *
 818  * Input:
 819  *                      buf:            ignored
 820  *                      size:           zero
 821  * Output:
 822  *      On success:     passed-in buffer filled with a '\n'-terminated C
 823  *                      string containing a whitespace-separated list of
 824  *                      named NFSD listeners;
 825  *                      return code is the size in bytes of the string
 826  *      On error:       return code is zero or a negative errno value
 827  *
 828  * OR
 829  *
 830  * Input:
 831  *                      buf:            C string containing an unsigned
 832  *                                      integer value representing a bound
 833  *                                      but unconnected socket that is to be
 834  *                                      used as an NFSD listener; listen(3)
 835  *                                      must be called for a SOCK_STREAM
 836  *                                      socket, otherwise it is ignored
 837  *                      size:           non-zero length of C string in @buf
 838  * Output:
 839  *      On success:     NFS service is started;
 840  *                      passed-in buffer filled with a '\n'-terminated C
 841  *                      string containing a unique alphanumeric name of
 842  *                      the listener;
 843  *                      return code is the size in bytes of the string
 844  *      On error:       return code is a negative errno value
 845  *
 846  * OR
 847  *
 848  * Input:
 849  *                      buf:            C string containing a transport
 850  *                                      name and an unsigned integer value
 851  *                                      representing the port to listen on,
 852  *                                      separated by whitespace
 853  *                      size:           non-zero length of C string in @buf
 854  * Output:
 855  *      On success:     returns zero; NFS service is started
 856  *      On error:       return code is a negative errno value
 857  */
 858 static ssize_t write_ports(struct file *file, char *buf, size_t size)
 859 {
 860         ssize_t rv;
 861
 862         mutex_lock(&nfsd_mutex);
 863         rv = __write_ports(file, buf, size, netns(file));
 864         mutex_unlock(&nfsd_mutex);
 865         return rv;
 866 }
 867
 868
 869 int nfsd_max_blksize;
 870
 871 /*
 872  * write_maxblksize - Set or report the current NFS blksize
 873  *
 874  * Input:
 875  *                      buf:            ignored
 876  *                      size:           zero
 877  *
 878  * OR
 879  *
 880  * Input:
 881  *                      buf:            C string containing an unsigned
 882  *                                      integer value representing the new
 883  *                                      NFS blksize
 884  *                      size:           non-zero length of C string in @buf
 885  * Output:
 886  *      On success:     passed-in buffer filled with '\n'-terminated C string
 887  *                      containing numeric value of the current NFS blksize
 888  *                      setting;
 889  *                      return code is the size in bytes of the string
 890  *      On error:       return code is zero or a negative errno value
 891  */
 892 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
 893 {
 894         char *mesg = buf;
 895         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 896
 897         if (size > 0) {
 898                 int bsize;
 899                 int rv = get_int(&mesg, &bsize);
 900                 if (rv)
 901                         return rv;
 902                 /* force bsize into allowed range and
 903                  * required alignment.
 904                  */
 905                 bsize = max_t(int, bsize, 1024);
 906                 bsize = min_t(int, bsize, NFSSVC_MAXBLKSIZE);
 907                 bsize &= ~(1024-1);
 908                 mutex_lock(&nfsd_mutex);
 909                 if (nn->nfsd_serv) {
 910                         mutex_unlock(&nfsd_mutex);
 911                         return -EBUSY;
 912                 }
 913                 nfsd_max_blksize = bsize;
 914                 mutex_unlock(&nfsd_mutex);
 915         }
 916
 917         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
 918                                                         nfsd_max_blksize);
 919 }
 920
 921 /*
 922  * write_maxconn - Set or report the current max number of connections
 923  *
 924  * Input:
 925  *                      buf:            ignored
 926  *                      size:           zero
 927  * OR
 928  *
 929  * Input:
 930  *                      buf:            C string containing an unsigned
 931  *                                      integer value representing the new
 932  *                                      number of max connections
 933  *                      size:           non-zero length of C string in @buf
 934  * Output:
 935  *      On success:     passed-in buffer filled with '\n'-terminated C string
 936  *                      containing numeric value of max_connections setting
 937  *                      for this net namespace;
 938  *                      return code is the size in bytes of the string
 939  *      On error:       return code is zero or a negative errno value
 940  */
 941 static ssize_t write_maxconn(struct file *file, char *buf, size_t size)
 942 {
 943         char *mesg = buf;
 944         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 945         unsigned int maxconn = nn->max_connections;
 946
 947         if (size > 0) {
 948                 int rv = get_uint(&mesg, &maxconn);
 949
 950                 if (rv)
 951                         return rv;
 952                 nn->max_connections = maxconn;
 953         }
 954
 955         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%u\n", maxconn);
 956 }
 957
 958 #ifdef CONFIG_NFSD_V4
 959 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
 960                                   time64_t *time, struct nfsd_net *nn)
 961 {
 962         char *mesg = buf;
 963         int rv, i;
 964
 965         if (size > 0) {
 966                 if (nn->nfsd_serv)
 967                         return -EBUSY;
 968                 rv = get_int(&mesg, &i);
 969                 if (rv)
 970                         return rv;
 971                 /*
 972                  * Some sanity checking.  We don't have a reason for
 973                  * these particular numbers, but problems with the
 974                  * extremes are:
 975                  *      - Too short: the briefest network outage may
 976                  *        cause clients to lose all their locks.  Also,
 977                  *        the frequent polling may be wasteful.
 978                  *      - Too long: do you really want reboot recovery
 979                  *        to take more than an hour?  Or to make other
 980                  *        clients wait an hour before being able to
 981                  *        revoke a dead client's locks?
 982                  */
 983                 if (i < 10 || i > 3600)
 984                         return -EINVAL;
 985                 *time = i;
 986         }
 987
 988         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%lld\n", *time);
 989 }
 990
 991 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
 992                                 time64_t *time, struct nfsd_net *nn)
 993 {
 994         ssize_t rv;
 995
 996         mutex_lock(&nfsd_mutex);
 997         rv = __nfsd4_write_time(file, buf, size, time, nn);
 998         mutex_unlock(&nfsd_mutex);
 999         return rv;
1000 }
1001
1002 /*
1003  * write_leasetime - Set or report the current NFSv4 lease time
1004  *
1005  * Input:
1006  *                      buf:            ignored
1007  *                      size:           zero
1008  *
1009  * OR
1010  *
1011  * Input:
1012  *                      buf:            C string containing an unsigned
1013  *                                      integer value representing the new
1014  *                                      NFSv4 lease expiry time
1015  *                      size:           non-zero length of C string in @buf
1016  * Output:
1017  *      On success:     passed-in buffer filled with '\n'-terminated C
1018  *                      string containing unsigned integer value of the
1019  *                      current lease expiry time;
1020  *                      return code is the size in bytes of the string
1021  *      On error:       return code is zero or a negative errno value
1022  */
1023 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
1024 {
1025         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1026         return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
1027 }
1028
1029 /*
1030  * write_gracetime - Set or report current NFSv4 grace period time
1031  *
1032  * As above, but sets the time of the NFSv4 grace period.
1033  *
1034  * Note this should never be set to less than the *previous*
1035  * lease-period time, but we don't try to enforce this.  (In the common
1036  * case (a new boot), we don't know what the previous lease time was
1037  * anyway.)
1038  */
1039 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
1040 {
1041         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1042         return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
1043 }
1044
1045 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
1046                                    struct nfsd_net *nn)
1047 {
1048         char *mesg = buf;
1049         char *recdir;
1050         int len, status;
1051
1052         if (size > 0) {
1053                 if (nn->nfsd_serv)
1054                         return -EBUSY;
1055                 if (size > PATH_MAX || buf[size-1] != '\n')
1056                         return -EINVAL;
1057                 buf[size-1] = 0;
1058
1059                 recdir = mesg;
1060                 len = qword_get(&mesg, recdir, size);
1061                 if (len <= 0)
1062                         return -EINVAL;
1063
1064                 status = nfs4_reset_recoverydir(recdir);
1065                 if (status)
1066                         return status;
1067         }
1068
1069         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1070                                                         nfs4_recoverydir());
1071 }
1072
1073 /*
1074  * write_recoverydir - Set or report the pathname of the recovery directory
1075  *
1076  * Input:
1077  *                      buf:            ignored
1078  *                      size:           zero
1079  *
1080  * OR
1081  *
1082  * Input:
1083  *                      buf:            C string containing the pathname
1084  *                                      of the directory on a local file
1085  *                                      system containing permanent NFSv4
1086  *                                      recovery data
1087  *                      size:           non-zero length of C string in @buf
1088  * Output:
1089  *      On success:     passed-in buffer filled with '\n'-terminated C string
1090  *                      containing the current recovery pathname setting;
1091  *                      return code is the size in bytes of the string
1092  *      On error:       return code is zero or a negative errno value
1093  */
1094 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1095 {
1096         ssize_t rv;
1097         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1098
1099         mutex_lock(&nfsd_mutex);
1100         rv = __write_recoverydir(file, buf, size, nn);
1101         mutex_unlock(&nfsd_mutex);
1102         return rv;
1103 }
1104
1105 /*
1106  * write_v4_end_grace - release grace period for nfsd's v4.x lock manager
1107  *
1108  * Input:
1109  *                      buf:            ignored
1110  *                      size:           zero
1111  * OR
1112  *
1113  * Input:
1114  *                      buf:            any value
1115  *                      size:           non-zero length of C string in @buf
1116  * Output:
1117  *                      passed-in buffer filled with "Y" or "N" with a newline
1118  *                      and NULL-terminated C string. This indicates whether
1119  *                      the grace period has ended in the current net
1120  *                      namespace. Return code is the size in bytes of the
1121  *                      string. Writing a string that starts with 'Y', 'y', or
1122  *                      '1' to the file will end the grace period for nfsd's v4
1123  *                      lock manager.
1124  */
1125 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size)
1126 {
1127         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1128
1129         if (size > 0) {
1130                 switch(buf[0]) {
1131                 case 'Y':
1132                 case 'y':
1133                 case '1':
1134                         if (!nn->nfsd_serv)
1135                                 return -EBUSY;
1136                         nfsd4_end_grace(nn);
1137                         break;
1138                 default:
1139                         return -EINVAL;
1140                 }
1141         }
1142
1143         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%c\n",
1144                          nn->grace_ended ? 'Y' : 'N');
1145 }
1146
1147 #endif
1148
1149 /*----------------------------------------------------------------------------*/
1150 /*
1151  *      populating the filesystem.
1152  */
1153
1154 /* Basically copying rpc_get_inode. */
1155 static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
1156 {
1157         struct inode *inode = new_inode(sb);
1158         if (!inode)
1159                 return NULL;
1160         /* Following advice from simple_fill_super documentation: */
1161         inode->i_ino = iunique(sb, NFSD_MaxReserved);
1162         inode->i_mode = mode;
1163         inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
1164         switch (mode & S_IFMT) {
1165         case S_IFDIR:
1166                 inode->i_fop = &simple_dir_operations;
1167                 inode->i_op = &simple_dir_inode_operations;
1168                 inc_nlink(inode);
1169                 break;
1170         default:
1171                 break;
1172         }
1173         return inode;
1174 }
1175
1176 static int __nfsd_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode, struct nfsdfs_client *ncl)
1177 {
1178         struct inode *inode;
1179
1180         inode = nfsd_get_inode(dir->i_sb, mode);
1181         if (!inode)
1182                 return -ENOMEM;
1183         if (ncl) {
1184                 inode->i_private = ncl;
1185                 kref_get(&ncl->cl_ref);
1186         }
1187         d_add(dentry, inode);
1188         inc_nlink(dir);
1189         fsnotify_mkdir(dir, dentry);
1190         return 0;
1191 }
1192
1193 static struct dentry *nfsd_mkdir(struct dentry *parent, struct nfsdfs_client *ncl, char *name)
1194 {
1195         struct inode *dir = parent->d_inode;
1196         struct dentry *dentry;
1197         int ret = -ENOMEM;
1198
1199         inode_lock(dir);
1200         dentry = d_alloc_name(parent, name);
1201         if (!dentry)
1202                 goto out_err;
1203         ret = __nfsd_mkdir(d_inode(parent), dentry, S_IFDIR | 0600, ncl);
1204         if (ret)
1205                 goto out_err;
1206 out:
1207         inode_unlock(dir);
1208         return dentry;
1209 out_err:
1210         dput(dentry);
1211         dentry = ERR_PTR(ret);
1212         goto out;
1213 }
1214
1215 static void clear_ncl(struct inode *inode)
1216 {
1217         struct nfsdfs_client *ncl = inode->i_private;
1218
1219         inode->i_private = NULL;
1220         kref_put(&ncl->cl_ref, ncl->cl_release);
1221 }
1222
1223 static struct nfsdfs_client *__get_nfsdfs_client(struct inode *inode)
1224 {
1225         struct nfsdfs_client *nc = inode->i_private;
1226
1227         if (nc)
1228                 kref_get(&nc->cl_ref);
1229         return nc;
1230 }
1231
1232 struct nfsdfs_client *get_nfsdfs_client(struct inode *inode)
1233 {
1234         struct nfsdfs_client *nc;
1235
1236         inode_lock_shared(inode);
1237         nc = __get_nfsdfs_client(inode);
1238         inode_unlock_shared(inode);
1239         return nc;
1240 }
1241 /* from __rpc_unlink */
1242 static void nfsdfs_remove_file(struct inode *dir, struct dentry *dentry)
1243 {
1244         int ret;
1245
1246         clear_ncl(d_inode(dentry));
1247         dget(dentry);
1248         ret = simple_unlink(dir, dentry);
1249         d_delete(dentry);
1250         dput(dentry);
1251         WARN_ON_ONCE(ret);
1252 }
1253
1254 static void nfsdfs_remove_files(struct dentry *root)
1255 {
1256         struct dentry *dentry, *tmp;
1257
1258         list_for_each_entry_safe(dentry, tmp, &root->d_subdirs, d_child) {
1259                 if (!simple_positive(dentry)) {
1260                         WARN_ON_ONCE(1); /* I think this can't happen? */
1261                         continue;
1262                 }
1263                 nfsdfs_remove_file(d_inode(root), dentry);
1264         }
1265 }
1266
1267 /* XXX: cut'n'paste from simple_fill_super; figure out if we could share
1268  * code instead. */
1269 static  int nfsdfs_create_files(struct dentry *root,
1270                                 const struct tree_descr *files,
1271                                 struct dentry **fdentries)
1272 {
1273         struct inode *dir = d_inode(root);
1274         struct inode *inode;
1275         struct dentry *dentry;
1276         int i;
1277
1278         inode_lock(dir);
1279         for (i = 0; files->name && files->name[0]; i++, files++) {
1280                 dentry = d_alloc_name(root, files->name);
1281                 if (!dentry)
1282                         goto out;
1283                 inode = nfsd_get_inode(d_inode(root)->i_sb,
1284                                         S_IFREG | files->mode);
1285                 if (!inode) {
1286                         dput(dentry);
1287                         goto out;
1288                 }
1289                 inode->i_fop = files->ops;
1290                 inode->i_private = __get_nfsdfs_client(dir);
1291                 d_add(dentry, inode);
1292                 fsnotify_create(dir, dentry);
1293                 if (fdentries)
1294                         fdentries[i] = dentry;
1295         }
1296         inode_unlock(dir);
1297         return 0;
1298 out:
1299         nfsdfs_remove_files(root);
1300         inode_unlock(dir);
1301         return -ENOMEM;
1302 }
1303
1304 /* on success, returns positive number unique to that client. */
1305 struct dentry *nfsd_client_mkdir(struct nfsd_net *nn,
1306                                  struct nfsdfs_client *ncl, u32 id,
1307                                  const struct tree_descr *files,
1308                                  struct dentry **fdentries)
1309 {
1310         struct dentry *dentry;
1311         char name[11];
1312         int ret;
1313
1314         sprintf(name, "%u", id);
1315
1316         dentry = nfsd_mkdir(nn->nfsd_client_dir, ncl, name);
1317         if (IS_ERR(dentry)) /* XXX: tossing errors? */
1318                 return NULL;
1319         ret = nfsdfs_create_files(dentry, files, fdentries);
1320         if (ret) {
1321                 nfsd_client_rmdir(dentry);
1322                 return NULL;
1323         }
1324         return dentry;
1325 }
1326
1327 /* Taken from __rpc_rmdir: */
1328 void nfsd_client_rmdir(struct dentry *dentry)
1329 {
1330         struct inode *dir = d_inode(dentry->d_parent);
1331         struct inode *inode = d_inode(dentry);
1332         int ret;
1333
1334         inode_lock(dir);
1335         nfsdfs_remove_files(dentry);
1336         clear_ncl(inode);
1337         dget(dentry);
1338         ret = simple_rmdir(dir, dentry);
1339         WARN_ON_ONCE(ret);
1340         fsnotify_rmdir(dir, dentry);
1341         d_delete(dentry);
1342         dput(dentry);
1343         inode_unlock(dir);
1344 }
1345
1346 static int nfsd_fill_super(struct super_block *sb, struct fs_context *fc)
1347 {
1348         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
1349                                                         nfsd_net_id);
1350         struct dentry *dentry;
1351         int ret;
1352
1353         static const struct tree_descr nfsd_files[] = {
1354                 [NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1355                 /* Per-export io stats use same ops as exports file */
1356                 [NFSD_Export_Stats] = {"export_stats", &exports_nfsd_operations, S_IRUGO},
1357                 [NFSD_Export_features] = {"export_features",
1358                                         &export_features_operations, S_IRUGO},
1359                 [NFSD_FO_UnlockIP] = {"unlock_ip",
1360                                         &transaction_ops, S_IWUSR|S_IRUSR},
1361                 [NFSD_FO_UnlockFS] = {"unlock_filesystem",
1362                                         &transaction_ops, S_IWUSR|S_IRUSR},
1363                 [NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1364                 [NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1365                 [NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1366                 [NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1367                 [NFSD_Reply_Cache_Stats] = {"reply_cache_stats", &reply_cache_stats_operations, S_IRUGO},
1368                 [NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1369                 [NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1370                 [NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1371                 [NFSD_MaxConnections] = {"max_connections", &transaction_ops, S_IWUSR|S_IRUGO},
1372 #if defined(CONFIG_SUNRPC_GSS) || defined(CONFIG_SUNRPC_GSS_MODULE)
1373                 [NFSD_SupportedEnctypes] = {"supported_krb5_enctypes", &supported_enctypes_ops, S_IRUGO},
1374 #endif /* CONFIG_SUNRPC_GSS or CONFIG_SUNRPC_GSS_MODULE */
1375 #ifdef CONFIG_NFSD_V4
1376                 [NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1377                 [NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1378                 [NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1379                 [NFSD_V4EndGrace] = {"v4_end_grace", &transaction_ops, S_IWUSR|S_IRUGO},
1380 #endif
1381                 /* last one */ {""}
1382         };
1383
1384         ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1385         if (ret)
1386                 return ret;
1387         dentry = nfsd_mkdir(sb->s_root, NULL, "clients");
1388         if (IS_ERR(dentry))
1389                 return PTR_ERR(dentry);
1390         nn->nfsd_client_dir = dentry;
1391         return 0;
1392 }
1393
1394 static int nfsd_fs_get_tree(struct fs_context *fc)
1395 {
1396         return get_tree_keyed(fc, nfsd_fill_super, get_net(fc->net_ns));
1397 }
1398
1399 static void nfsd_fs_free_fc(struct fs_context *fc)
1400 {
1401         if (fc->s_fs_info)
1402                 put_net(fc->s_fs_info);
1403 }
1404
1405 static const struct fs_context_operations nfsd_fs_context_ops = {
1406         .free           = nfsd_fs_free_fc,
1407         .get_tree       = nfsd_fs_get_tree,
1408 };
1409
1410 static int nfsd_init_fs_context(struct fs_context *fc)
1411 {
1412         put_user_ns(fc->user_ns);
1413         fc->user_ns = get_user_ns(fc->net_ns->user_ns);
1414         fc->ops = &nfsd_fs_context_ops;
1415         return 0;
1416 }
1417
1418 static void nfsd_umount(struct super_block *sb)
1419 {
1420         struct net *net = sb->s_fs_info;
1421
1422         nfsd_shutdown_threads(net);
1423
1424         kill_litter_super(sb);
1425         put_net(net);
1426 }
1427
1428 static struct file_system_type nfsd_fs_type = {
1429         .owner          = THIS_MODULE,
1430         .name           = "nfsd",
1431         .init_fs_context = nfsd_init_fs_context,
1432         .kill_sb        = nfsd_umount,
1433 };
1434 MODULE_ALIAS_FS("nfsd");
1435
1436 #ifdef CONFIG_PROC_FS
1437 static int create_proc_exports_entry(void)
1438 {
1439         struct proc_dir_entry *entry;
1440
1441         entry = proc_mkdir("fs/nfs", NULL);
1442         if (!entry)
1443                 return -ENOMEM;
1444         entry = proc_create("exports", 0, entry, &exports_proc_ops);
1445         if (!entry) {
1446                 remove_proc_entry("fs/nfs", NULL);
1447                 return -ENOMEM;
1448         }
1449         return 0;
1450 }
1451 #else /* CONFIG_PROC_FS */
1452 static int create_proc_exports_entry(void)
1453 {
1454         return 0;
1455 }
1456 #endif
1457
1458 unsigned int nfsd_net_id;
1459
1460 static __net_init int nfsd_init_net(struct net *net)
1461 {
1462         int retval;
1463         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1464
1465         retval = nfsd_export_init(net);
1466         if (retval)
1467                 goto out_export_error;
1468         retval = nfsd_idmap_init(net);
1469         if (retval)
1470                 goto out_idmap_error;
1471         nn->nfsd_versions = NULL;
1472         nn->nfsd4_minorversions = NULL;
1473         retval = nfsd_reply_cache_init(nn);
1474         if (retval)
1475                 goto out_drc_error;
1476         nn->nfsd4_lease = 90;   /* default lease time */
1477         nn->nfsd4_grace = 90;
1478         nn->somebody_reclaimed = false;
1479         nn->track_reclaim_completes = false;
1480         nn->clverifier_counter = prandom_u32();
1481         nn->clientid_base = prandom_u32();
1482         nn->clientid_counter = nn->clientid_base + 1;
1483         nn->s2s_cp_cl_id = nn->clientid_counter++;
1484
1485         atomic_set(&nn->ntf_refcnt, 0);
1486         init_waitqueue_head(&nn->ntf_wq);
1487         seqlock_init(&nn->boot_lock);
1488
1489         return 0;
1490
1491 out_drc_error:
1492         nfsd_idmap_shutdown(net);
1493 out_idmap_error:
1494         nfsd_export_shutdown(net);
1495 out_export_error:
1496         return retval;
1497 }
1498
1499 static __net_exit void nfsd_exit_net(struct net *net)
1500 {
1501         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1502
1503         nfsd_reply_cache_shutdown(nn);
1504         nfsd_idmap_shutdown(net);
1505         nfsd_export_shutdown(net);
1506         nfsd_netns_free_versions(net_generic(net, nfsd_net_id));
1507 }
1508
1509 static struct pernet_operations nfsd_net_ops = {
1510         .init = nfsd_init_net,
1511         .exit = nfsd_exit_net,
1512         .id   = &nfsd_net_id,
1513         .size = sizeof(struct nfsd_net),
1514 };
1515
1516 static int __init init_nfsd(void)
1517 {
1518         int retval;
1519         printk(KERN_INFO "Installing knfsd (copyright (C) 1996 okir@monad.swb.de).\n");
1520
1521         retval = register_cld_notifier();
1522         if (retval)
1523                 return retval;
1524         retval = nfsd4_init_slabs();
1525         if (retval)
1526                 goto out_unregister_notifier;
1527         retval = nfsd4_init_pnfs();
1528         if (retval)
1529                 goto out_free_slabs;
1530         retval = nfsd_stat_init();      /* Statistics */
1531         if (retval)
1532                 goto out_free_pnfs;
1533         retval = nfsd_drc_slab_create();
1534         if (retval)
1535                 goto out_free_stat;
1536         nfsd_lockd_init();      /* lockd->nfsd callbacks */
1537         retval = create_proc_exports_entry();
1538         if (retval)
1539                 goto out_free_lockd;
1540         retval = register_filesystem(&nfsd_fs_type);
1541         if (retval)
1542                 goto out_free_exports;
1543         retval = register_pernet_subsys(&nfsd_net_ops);
1544         if (retval < 0)
1545                 goto out_free_all;
1546         return 0;
1547 out_free_all:
1548         unregister_pernet_subsys(&nfsd_net_ops);
1549 out_free_exports:
1550         remove_proc_entry("fs/nfs/exports", NULL);
1551         remove_proc_entry("fs/nfs", NULL);
1552 out_free_lockd:
1553         nfsd_lockd_shutdown();
1554         nfsd_drc_slab_free();
1555 out_free_stat:
1556         nfsd_stat_shutdown();
1557 out_free_pnfs:
1558         nfsd4_exit_pnfs();
1559 out_free_slabs:
1560         nfsd4_free_slabs();
1561 out_unregister_notifier:
1562         unregister_cld_notifier();
1563         return retval;
1564 }
1565
1566 static void __exit exit_nfsd(void)
1567 {
1568         unregister_pernet_subsys(&nfsd_net_ops);
1569         nfsd_drc_slab_free();
1570         remove_proc_entry("fs/nfs/exports", NULL);
1571         remove_proc_entry("fs/nfs", NULL);
1572         nfsd_stat_shutdown();
1573         nfsd_lockd_shutdown();
1574         nfsd4_free_slabs();
1575         nfsd4_exit_pnfs();
1576         unregister_filesystem(&nfsd_fs_type);
1577         unregister_cld_notifier();
1578 }
1579
1580 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1581 MODULE_LICENSE("GPL");
1582 module_init(init_nfsd)
1583 module_exit(exit_nfsd)