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/rpc_pipe_fs.h>
  18 #include <linux/module.h>
  19 #include <linux/fsnotify.h>
  20
  21 #include "idmap.h"
  22 #include "nfsd.h"
  23 #include "cache.h"
  24 #include "state.h"
  25 #include "netns.h"
  26 #include "pnfs.h"
  27 #include "filecache.h"
  28 #include "trace.h"
  29 #include "netlink.h"
  30
  31 /*
  32  *      We have a single directory with several nodes in it.
  33  */
  34 enum {
  35         NFSD_Root = 1,
  36         NFSD_List,
  37         NFSD_Export_Stats,
  38         NFSD_Export_features,
  39         NFSD_Fh,
  40         NFSD_FO_UnlockIP,
  41         NFSD_FO_UnlockFS,
  42         NFSD_Threads,
  43         NFSD_Pool_Threads,
  44         NFSD_Pool_Stats,
  45         NFSD_Reply_Cache_Stats,
  46         NFSD_Versions,
  47         NFSD_Ports,
  48         NFSD_MaxBlkSize,
  49         NFSD_MaxConnections,
  50         NFSD_Filecache,
  51 #ifdef CONFIG_NFSD_V4
  52         NFSD_Leasetime,
  53         NFSD_Gracetime,
  54         NFSD_RecoveryDir,
  55         NFSD_V4EndGrace,
  56 #endif
  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 const struct file_operations pool_stats_operations = {
 179         .open           = nfsd_pool_stats_open,
 180         .read           = seq_read,
 181         .llseek         = seq_lseek,
 182         .release        = seq_release,
 183 };
 184
 185 DEFINE_SHOW_ATTRIBUTE(nfsd_reply_cache_stats);
 186
 187 DEFINE_SHOW_ATTRIBUTE(nfsd_file_cache_stats);
 188
 189 /*----------------------------------------------------------------------------*/
 190 /*
 191  * payload - write methods
 192  */
 193
 194 static inline struct net *netns(struct file *file)
 195 {
 196         return file_inode(file)->i_sb->s_fs_info;
 197 }
 198
 199 /*
 200  * write_unlock_ip - Release all locks used by a client
 201  *
 202  * Experimental.
 203  *
 204  * Input:
 205  *                      buf:    '\n'-terminated C string containing a
 206  *                              presentation format IP address
 207  *                      size:   length of C string in @buf
 208  * Output:
 209  *      On success:     returns zero if all specified locks were released;
 210  *                      returns one if one or more locks were not released
 211  *      On error:       return code is negative errno value
 212  */
 213 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size)
 214 {
 215         struct sockaddr_storage address;
 216         struct sockaddr *sap = (struct sockaddr *)&address;
 217         size_t salen = sizeof(address);
 218         char *fo_path;
 219         struct net *net = netns(file);
 220
 221         /* sanity check */
 222         if (size == 0)
 223                 return -EINVAL;
 224
 225         if (buf[size-1] != '\n')
 226                 return -EINVAL;
 227
 228         fo_path = buf;
 229         if (qword_get(&buf, fo_path, size) < 0)
 230                 return -EINVAL;
 231
 232         if (rpc_pton(net, fo_path, size, sap, salen) == 0)
 233                 return -EINVAL;
 234
 235         trace_nfsd_ctl_unlock_ip(net, buf);
 236         return nlmsvc_unlock_all_by_ip(sap);
 237 }
 238
 239 /*
 240  * write_unlock_fs - Release all locks on a local file system
 241  *
 242  * Experimental.
 243  *
 244  * Input:
 245  *                      buf:    '\n'-terminated C string containing the
 246  *                              absolute pathname of a local file system
 247  *                      size:   length of C string in @buf
 248  * Output:
 249  *      On success:     returns zero if all specified locks were released;
 250  *                      returns one if one or more locks were not released
 251  *      On error:       return code is negative errno value
 252  */
 253 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size)
 254 {
 255         struct path path;
 256         char *fo_path;
 257         int error;
 258
 259         /* sanity check */
 260         if (size == 0)
 261                 return -EINVAL;
 262
 263         if (buf[size-1] != '\n')
 264                 return -EINVAL;
 265
 266         fo_path = buf;
 267         if (qword_get(&buf, fo_path, size) < 0)
 268                 return -EINVAL;
 269         trace_nfsd_ctl_unlock_fs(netns(file), fo_path);
 270         error = kern_path(fo_path, 0, &path);
 271         if (error)
 272                 return error;
 273
 274         /*
 275          * XXX: Needs better sanity checking.  Otherwise we could end up
 276          * releasing locks on the wrong file system.
 277          *
 278          * For example:
 279          * 1.  Does the path refer to a directory?
 280          * 2.  Is that directory a mount point, or
 281          * 3.  Is that directory the root of an exported file system?
 282          */
 283         error = nlmsvc_unlock_all_by_sb(path.dentry->d_sb);
 284         nfsd4_revoke_states(netns(file), path.dentry->d_sb);
 285
 286         path_put(&path);
 287         return error;
 288 }
 289
 290 /*
 291  * write_filehandle - Get a variable-length NFS file handle by path
 292  *
 293  * On input, the buffer contains a '\n'-terminated C string comprised of
 294  * three alphanumeric words separated by whitespace.  The string may
 295  * contain escape sequences.
 296  *
 297  * Input:
 298  *                      buf:
 299  *                              domain:         client domain name
 300  *                              path:           export pathname
 301  *                              maxsize:        numeric maximum size of
 302  *                                              @buf
 303  *                      size:   length of C string in @buf
 304  * Output:
 305  *      On success:     passed-in buffer filled with '\n'-terminated C
 306  *                      string containing a ASCII hex text version
 307  *                      of the NFS file handle;
 308  *                      return code is the size in bytes of the string
 309  *      On error:       return code is negative errno value
 310  */
 311 static ssize_t write_filehandle(struct file *file, char *buf, size_t size)
 312 {
 313         char *dname, *path;
 314         int maxsize;
 315         char *mesg = buf;
 316         int len;
 317         struct auth_domain *dom;
 318         struct knfsd_fh fh;
 319
 320         if (size == 0)
 321                 return -EINVAL;
 322
 323         if (buf[size-1] != '\n')
 324                 return -EINVAL;
 325         buf[size-1] = 0;
 326
 327         dname = mesg;
 328         len = qword_get(&mesg, dname, size);
 329         if (len <= 0)
 330                 return -EINVAL;
 331
 332         path = dname+len+1;
 333         len = qword_get(&mesg, path, size);
 334         if (len <= 0)
 335                 return -EINVAL;
 336
 337         len = get_int(&mesg, &maxsize);
 338         if (len)
 339                 return len;
 340
 341         if (maxsize < NFS_FHSIZE)
 342                 return -EINVAL;
 343         maxsize = min(maxsize, NFS3_FHSIZE);
 344
 345         if (qword_get(&mesg, mesg, size) > 0)
 346                 return -EINVAL;
 347
 348         trace_nfsd_ctl_filehandle(netns(file), dname, path, maxsize);
 349
 350         /* we have all the words, they are in buf.. */
 351         dom = unix_domain_find(dname);
 352         if (!dom)
 353                 return -ENOMEM;
 354
 355         len = exp_rootfh(netns(file), dom, path, &fh, maxsize);
 356         auth_domain_put(dom);
 357         if (len)
 358                 return len;
 359
 360         mesg = buf;
 361         len = SIMPLE_TRANSACTION_LIMIT;
 362         qword_addhex(&mesg, &len, fh.fh_raw, fh.fh_size);
 363         mesg[-1] = '\n';
 364         return mesg - buf;
 365 }
 366
 367 /*
 368  * write_threads - Start NFSD, or report the current number of running threads
 369  *
 370  * Input:
 371  *                      buf:            ignored
 372  *                      size:           zero
 373  * Output:
 374  *      On success:     passed-in buffer filled with '\n'-terminated C
 375  *                      string numeric value representing the number of
 376  *                      running NFSD threads;
 377  *                      return code is the size in bytes of the string
 378  *      On error:       return code is zero
 379  *
 380  * OR
 381  *
 382  * Input:
 383  *                      buf:            C string containing an unsigned
 384  *                                      integer value representing the
 385  *                                      number of NFSD threads to start
 386  *                      size:           non-zero length of C string in @buf
 387  * Output:
 388  *      On success:     NFS service is started;
 389  *                      passed-in buffer filled with '\n'-terminated C
 390  *                      string numeric value representing the number of
 391  *                      running NFSD threads;
 392  *                      return code is the size in bytes of the string
 393  *      On error:       return code is zero or a negative errno value
 394  */
 395 static ssize_t write_threads(struct file *file, char *buf, size_t size)
 396 {
 397         char *mesg = buf;
 398         int rv;
 399         struct net *net = netns(file);
 400
 401         if (size > 0) {
 402                 int newthreads;
 403                 rv = get_int(&mesg, &newthreads);
 404                 if (rv)
 405                         return rv;
 406                 if (newthreads < 0)
 407                         return -EINVAL;
 408                 trace_nfsd_ctl_threads(net, newthreads);
 409                 rv = nfsd_svc(newthreads, net, file->f_cred);
 410                 if (rv < 0)
 411                         return rv;
 412         } else
 413                 rv = nfsd_nrthreads(net);
 414
 415         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n", rv);
 416 }
 417
 418 /*
 419  * write_pool_threads - Set or report the current number of threads per pool
 420  *
 421  * Input:
 422  *                      buf:            ignored
 423  *                      size:           zero
 424  *
 425  * OR
 426  *
 427  * Input:
 428  *                      buf:            C string containing whitespace-
 429  *                                      separated unsigned integer values
 430  *                                      representing the number of NFSD
 431  *                                      threads to start in each pool
 432  *                      size:           non-zero length of C string in @buf
 433  * Output:
 434  *      On success:     passed-in buffer filled with '\n'-terminated C
 435  *                      string containing integer values representing the
 436  *                      number of NFSD threads in each pool;
 437  *                      return code is the size in bytes of the string
 438  *      On error:       return code is zero or a negative errno value
 439  */
 440 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size)
 441 {
 442         /* if size > 0, look for an array of number of threads per node
 443          * and apply them  then write out number of threads per node as reply
 444          */
 445         char *mesg = buf;
 446         int i;
 447         int rv;
 448         int len;
 449         int npools;
 450         int *nthreads;
 451         struct net *net = netns(file);
 452
 453         mutex_lock(&nfsd_mutex);
 454         npools = nfsd_nrpools(net);
 455         if (npools == 0) {
 456                 /*
 457                  * NFS is shut down.  The admin can start it by
 458                  * writing to the threads file but NOT the pool_threads
 459                  * file, sorry.  Report zero threads.
 460                  */
 461                 mutex_unlock(&nfsd_mutex);
 462                 strcpy(buf, "0\n");
 463                 return strlen(buf);
 464         }
 465
 466         nthreads = kcalloc(npools, sizeof(int), GFP_KERNEL);
 467         rv = -ENOMEM;
 468         if (nthreads == NULL)
 469                 goto out_free;
 470
 471         if (size > 0) {
 472                 for (i = 0; i < npools; i++) {
 473                         rv = get_int(&mesg, &nthreads[i]);
 474                         if (rv == -ENOENT)
 475                                 break;          /* fewer numbers than pools */
 476                         if (rv)
 477                                 goto out_free;  /* syntax error */
 478                         rv = -EINVAL;
 479                         if (nthreads[i] < 0)
 480                                 goto out_free;
 481                         trace_nfsd_ctl_pool_threads(net, i, nthreads[i]);
 482                 }
 483                 rv = nfsd_set_nrthreads(i, nthreads, net);
 484                 if (rv)
 485                         goto out_free;
 486         }
 487
 488         rv = nfsd_get_nrthreads(npools, nthreads, net);
 489         if (rv)
 490                 goto out_free;
 491
 492         mesg = buf;
 493         size = SIMPLE_TRANSACTION_LIMIT;
 494         for (i = 0; i < npools && size > 0; i++) {
 495                 snprintf(mesg, size, "%d%c", nthreads[i], (i == npools-1 ? '\n' : ' '));
 496                 len = strlen(mesg);
 497                 size -= len;
 498                 mesg += len;
 499         }
 500         rv = mesg - buf;
 501 out_free:
 502         kfree(nthreads);
 503         mutex_unlock(&nfsd_mutex);
 504         return rv;
 505 }
 506
 507 static ssize_t
 508 nfsd_print_version_support(struct nfsd_net *nn, char *buf, int remaining,
 509                 const char *sep, unsigned vers, int minor)
 510 {
 511         const char *format = minor < 0 ? "%s%c%u" : "%s%c%u.%u";
 512         bool supported = !!nfsd_vers(nn, vers, NFSD_TEST);
 513
 514         if (vers == 4 && minor >= 0 &&
 515             !nfsd_minorversion(nn, minor, NFSD_TEST))
 516                 supported = false;
 517         if (minor == 0 && supported)
 518                 /*
 519                  * special case for backward compatability.
 520                  * +4.0 is never reported, it is implied by
 521                  * +4, unless -4.0 is present.
 522                  */
 523                 return 0;
 524         return snprintf(buf, remaining, format, sep,
 525                         supported ? '+' : '-', vers, minor);
 526 }
 527
 528 static ssize_t __write_versions(struct file *file, char *buf, size_t size)
 529 {
 530         char *mesg = buf;
 531         char *vers, *minorp, sign;
 532         int len, num, remaining;
 533         ssize_t tlen = 0;
 534         char *sep;
 535         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 536
 537         if (size > 0) {
 538                 if (nn->nfsd_serv)
 539                         /* Cannot change versions without updating
 540                          * nn->nfsd_serv->sv_xdrsize, and reallocing
 541                          * rq_argp and rq_resp
 542                          */
 543                         return -EBUSY;
 544                 if (buf[size-1] != '\n')
 545                         return -EINVAL;
 546                 buf[size-1] = 0;
 547                 trace_nfsd_ctl_version(netns(file), buf);
 548
 549                 vers = mesg;
 550                 len = qword_get(&mesg, vers, size);
 551                 if (len <= 0) return -EINVAL;
 552                 do {
 553                         enum vers_op cmd;
 554                         unsigned minor;
 555                         sign = *vers;
 556                         if (sign == '+' || sign == '-')
 557                                 num = simple_strtol((vers+1), &minorp, 0);
 558                         else
 559                                 num = simple_strtol(vers, &minorp, 0);
 560                         if (*minorp == '.') {
 561                                 if (num != 4)
 562                                         return -EINVAL;
 563                                 if (kstrtouint(minorp+1, 0, &minor) < 0)
 564                                         return -EINVAL;
 565                         }
 566
 567                         cmd = sign == '-' ? NFSD_CLEAR : NFSD_SET;
 568                         switch(num) {
 569 #ifdef CONFIG_NFSD_V2
 570                         case 2:
 571 #endif
 572                         case 3:
 573                                 nfsd_vers(nn, num, cmd);
 574                                 break;
 575                         case 4:
 576                                 if (*minorp == '.') {
 577                                         if (nfsd_minorversion(nn, minor, cmd) < 0)
 578                                                 return -EINVAL;
 579                                 } else if ((cmd == NFSD_SET) != nfsd_vers(nn, num, NFSD_TEST)) {
 580                                         /*
 581                                          * Either we have +4 and no minors are enabled,
 582                                          * or we have -4 and at least one minor is enabled.
 583                                          * In either case, propagate 'cmd' to all minors.
 584                                          */
 585                                         minor = 0;
 586                                         while (nfsd_minorversion(nn, minor, cmd) >= 0)
 587                                                 minor++;
 588                                 }
 589                                 break;
 590                         default:
 591                                 /* Ignore requests to disable non-existent versions */
 592                                 if (cmd == NFSD_SET)
 593                                         return -EINVAL;
 594                         }
 595                         vers += len + 1;
 596                 } while ((len = qword_get(&mesg, vers, size)) > 0);
 597                 /* If all get turned off, turn them back on, as
 598                  * having no versions is BAD
 599                  */
 600                 nfsd_reset_versions(nn);
 601         }
 602
 603         /* Now write current state into reply buffer */
 604         sep = "";
 605         remaining = SIMPLE_TRANSACTION_LIMIT;
 606         for (num=2 ; num <= 4 ; num++) {
 607                 int minor;
 608                 if (!nfsd_vers(nn, num, NFSD_AVAIL))
 609                         continue;
 610
 611                 minor = -1;
 612                 do {
 613                         len = nfsd_print_version_support(nn, buf, remaining,
 614                                         sep, num, minor);
 615                         if (len >= remaining)
 616                                 goto out;
 617                         remaining -= len;
 618                         buf += len;
 619                         tlen += len;
 620                         minor++;
 621                         if (len)
 622                                 sep = " ";
 623                 } while (num == 4 && minor <= NFSD_SUPPORTED_MINOR_VERSION);
 624         }
 625 out:
 626         len = snprintf(buf, remaining, "\n");
 627         if (len >= remaining)
 628                 return -EINVAL;
 629         return tlen + len;
 630 }
 631
 632 /*
 633  * write_versions - Set or report the available NFS protocol versions
 634  *
 635  * Input:
 636  *                      buf:            ignored
 637  *                      size:           zero
 638  * Output:
 639  *      On success:     passed-in buffer filled with '\n'-terminated C
 640  *                      string containing positive or negative integer
 641  *                      values representing the current status of each
 642  *                      protocol version;
 643  *                      return code is the size in bytes of the string
 644  *      On error:       return code is zero or a negative errno value
 645  *
 646  * OR
 647  *
 648  * Input:
 649  *                      buf:            C string containing whitespace-
 650  *                                      separated positive or negative
 651  *                                      integer values representing NFS
 652  *                                      protocol versions to enable ("+n")
 653  *                                      or disable ("-n")
 654  *                      size:           non-zero length of C string in @buf
 655  * Output:
 656  *      On success:     status of zero or more protocol versions has
 657  *                      been updated; passed-in buffer filled with
 658  *                      '\n'-terminated C string containing positive
 659  *                      or negative integer values representing the
 660  *                      current status of each protocol version;
 661  *                      return code is the size in bytes of the string
 662  *      On error:       return code is zero or a negative errno value
 663  */
 664 static ssize_t write_versions(struct file *file, char *buf, size_t size)
 665 {
 666         ssize_t rv;
 667
 668         mutex_lock(&nfsd_mutex);
 669         rv = __write_versions(file, buf, size);
 670         mutex_unlock(&nfsd_mutex);
 671         return rv;
 672 }
 673
 674 /*
 675  * Zero-length write.  Return a list of NFSD's current listener
 676  * transports.
 677  */
 678 static ssize_t __write_ports_names(char *buf, struct net *net)
 679 {
 680         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 681
 682         if (nn->nfsd_serv == NULL)
 683                 return 0;
 684         return svc_xprt_names(nn->nfsd_serv, buf, SIMPLE_TRANSACTION_LIMIT);
 685 }
 686
 687 /*
 688  * A single 'fd' number was written, in which case it must be for
 689  * a socket of a supported family/protocol, and we use it as an
 690  * nfsd listener.
 691  */
 692 static ssize_t __write_ports_addfd(char *buf, struct net *net, const struct cred *cred)
 693 {
 694         char *mesg = buf;
 695         int fd, err;
 696         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 697         struct svc_serv *serv;
 698
 699         err = get_int(&mesg, &fd);
 700         if (err != 0 || fd < 0)
 701                 return -EINVAL;
 702         trace_nfsd_ctl_ports_addfd(net, fd);
 703
 704         err = nfsd_create_serv(net);
 705         if (err != 0)
 706                 return err;
 707
 708         serv = nn->nfsd_serv;
 709         err = svc_addsock(serv, net, fd, buf, SIMPLE_TRANSACTION_LIMIT, cred);
 710
 711         if (!serv->sv_nrthreads && list_empty(&nn->nfsd_serv->sv_permsocks))
 712                 nfsd_destroy_serv(net);
 713
 714         return err;
 715 }
 716
 717 /*
 718  * A transport listener is added by writing its transport name and
 719  * a port number.
 720  */
 721 static ssize_t __write_ports_addxprt(char *buf, struct net *net, const struct cred *cred)
 722 {
 723         char transport[16];
 724         struct svc_xprt *xprt;
 725         int port, err;
 726         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 727         struct svc_serv *serv;
 728
 729         if (sscanf(buf, "%15s %5u", transport, &port) != 2)
 730                 return -EINVAL;
 731
 732         if (port < 1 || port > USHRT_MAX)
 733                 return -EINVAL;
 734         trace_nfsd_ctl_ports_addxprt(net, transport, port);
 735
 736         err = nfsd_create_serv(net);
 737         if (err != 0)
 738                 return err;
 739
 740         serv = nn->nfsd_serv;
 741         err = svc_xprt_create(serv, transport, net,
 742                               PF_INET, port, SVC_SOCK_ANONYMOUS, cred);
 743         if (err < 0)
 744                 goto out_err;
 745
 746         err = svc_xprt_create(serv, transport, net,
 747                               PF_INET6, port, SVC_SOCK_ANONYMOUS, cred);
 748         if (err < 0 && err != -EAFNOSUPPORT)
 749                 goto out_close;
 750
 751         return 0;
 752 out_close:
 753         xprt = svc_find_xprt(serv, transport, net, PF_INET, port);
 754         if (xprt != NULL) {
 755                 svc_xprt_close(xprt);
 756                 svc_xprt_put(xprt);
 757         }
 758 out_err:
 759         if (!serv->sv_nrthreads && list_empty(&nn->nfsd_serv->sv_permsocks))
 760                 nfsd_destroy_serv(net);
 761
 762         return err;
 763 }
 764
 765 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
 766                              struct net *net)
 767 {
 768         if (size == 0)
 769                 return __write_ports_names(buf, net);
 770
 771         if (isdigit(buf[0]))
 772                 return __write_ports_addfd(buf, net, file->f_cred);
 773
 774         if (isalpha(buf[0]))
 775                 return __write_ports_addxprt(buf, net, file->f_cred);
 776
 777         return -EINVAL;
 778 }
 779
 780 /*
 781  * write_ports - Pass a socket file descriptor or transport name to listen on
 782  *
 783  * Input:
 784  *                      buf:            ignored
 785  *                      size:           zero
 786  * Output:
 787  *      On success:     passed-in buffer filled with a '\n'-terminated C
 788  *                      string containing a whitespace-separated list of
 789  *                      named NFSD listeners;
 790  *                      return code is the size in bytes of the string
 791  *      On error:       return code is zero or a negative errno value
 792  *
 793  * OR
 794  *
 795  * Input:
 796  *                      buf:            C string containing an unsigned
 797  *                                      integer value representing a bound
 798  *                                      but unconnected socket that is to be
 799  *                                      used as an NFSD listener; listen(3)
 800  *                                      must be called for a SOCK_STREAM
 801  *                                      socket, otherwise it is ignored
 802  *                      size:           non-zero length of C string in @buf
 803  * Output:
 804  *      On success:     NFS service is started;
 805  *                      passed-in buffer filled with a '\n'-terminated C
 806  *                      string containing a unique alphanumeric name of
 807  *                      the listener;
 808  *                      return code is the size in bytes of the string
 809  *      On error:       return code is a negative errno value
 810  *
 811  * OR
 812  *
 813  * Input:
 814  *                      buf:            C string containing a transport
 815  *                                      name and an unsigned integer value
 816  *                                      representing the port to listen on,
 817  *                                      separated by whitespace
 818  *                      size:           non-zero length of C string in @buf
 819  * Output:
 820  *      On success:     returns zero; NFS service is started
 821  *      On error:       return code is a negative errno value
 822  */
 823 static ssize_t write_ports(struct file *file, char *buf, size_t size)
 824 {
 825         ssize_t rv;
 826
 827         mutex_lock(&nfsd_mutex);
 828         rv = __write_ports(file, buf, size, netns(file));
 829         mutex_unlock(&nfsd_mutex);
 830         return rv;
 831 }
 832
 833
 834 int nfsd_max_blksize;
 835
 836 /*
 837  * write_maxblksize - Set or report the current NFS blksize
 838  *
 839  * Input:
 840  *                      buf:            ignored
 841  *                      size:           zero
 842  *
 843  * OR
 844  *
 845  * Input:
 846  *                      buf:            C string containing an unsigned
 847  *                                      integer value representing the new
 848  *                                      NFS blksize
 849  *                      size:           non-zero length of C string in @buf
 850  * Output:
 851  *      On success:     passed-in buffer filled with '\n'-terminated C string
 852  *                      containing numeric value of the current NFS blksize
 853  *                      setting;
 854  *                      return code is the size in bytes of the string
 855  *      On error:       return code is zero or a negative errno value
 856  */
 857 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
 858 {
 859         char *mesg = buf;
 860         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 861
 862         if (size > 0) {
 863                 int bsize;
 864                 int rv = get_int(&mesg, &bsize);
 865                 if (rv)
 866                         return rv;
 867                 trace_nfsd_ctl_maxblksize(netns(file), bsize);
 868
 869                 /* force bsize into allowed range and
 870                  * required alignment.
 871                  */
 872                 bsize = max_t(int, bsize, 1024);
 873                 bsize = min_t(int, bsize, NFSSVC_MAXBLKSIZE);
 874                 bsize &= ~(1024-1);
 875                 mutex_lock(&nfsd_mutex);
 876                 if (nn->nfsd_serv) {
 877                         mutex_unlock(&nfsd_mutex);
 878                         return -EBUSY;
 879                 }
 880                 nfsd_max_blksize = bsize;
 881                 mutex_unlock(&nfsd_mutex);
 882         }
 883
 884         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
 885                                                         nfsd_max_blksize);
 886 }
 887
 888 /*
 889  * write_maxconn - Set or report the current max number of connections
 890  *
 891  * Input:
 892  *                      buf:            ignored
 893  *                      size:           zero
 894  * OR
 895  *
 896  * Input:
 897  *                      buf:            C string containing an unsigned
 898  *                                      integer value representing the new
 899  *                                      number of max connections
 900  *                      size:           non-zero length of C string in @buf
 901  * Output:
 902  *      On success:     passed-in buffer filled with '\n'-terminated C string
 903  *                      containing numeric value of max_connections setting
 904  *                      for this net namespace;
 905  *                      return code is the size in bytes of the string
 906  *      On error:       return code is zero or a negative errno value
 907  */
 908 static ssize_t write_maxconn(struct file *file, char *buf, size_t size)
 909 {
 910         char *mesg = buf;
 911         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 912         unsigned int maxconn = nn->max_connections;
 913
 914         if (size > 0) {
 915                 int rv = get_uint(&mesg, &maxconn);
 916
 917                 if (rv)
 918                         return rv;
 919                 trace_nfsd_ctl_maxconn(netns(file), maxconn);
 920                 nn->max_connections = maxconn;
 921         }
 922
 923         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%u\n", maxconn);
 924 }
 925
 926 #ifdef CONFIG_NFSD_V4
 927 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
 928                                   time64_t *time, struct nfsd_net *nn)
 929 {
 930         struct dentry *dentry = file_dentry(file);
 931         char *mesg = buf;
 932         int rv, i;
 933
 934         if (size > 0) {
 935                 if (nn->nfsd_serv)
 936                         return -EBUSY;
 937                 rv = get_int(&mesg, &i);
 938                 if (rv)
 939                         return rv;
 940                 trace_nfsd_ctl_time(netns(file), dentry->d_name.name,
 941                                     dentry->d_name.len, i);
 942
 943                 /*
 944                  * Some sanity checking.  We don't have a reason for
 945                  * these particular numbers, but problems with the
 946                  * extremes are:
 947                  *      - Too short: the briefest network outage may
 948                  *        cause clients to lose all their locks.  Also,
 949                  *        the frequent polling may be wasteful.
 950                  *      - Too long: do you really want reboot recovery
 951                  *        to take more than an hour?  Or to make other
 952                  *        clients wait an hour before being able to
 953                  *        revoke a dead client's locks?
 954                  */
 955                 if (i < 10 || i > 3600)
 956                         return -EINVAL;
 957                 *time = i;
 958         }
 959
 960         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%lld\n", *time);
 961 }
 962
 963 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
 964                                 time64_t *time, struct nfsd_net *nn)
 965 {
 966         ssize_t rv;
 967
 968         mutex_lock(&nfsd_mutex);
 969         rv = __nfsd4_write_time(file, buf, size, time, nn);
 970         mutex_unlock(&nfsd_mutex);
 971         return rv;
 972 }
 973
 974 /*
 975  * write_leasetime - Set or report the current NFSv4 lease time
 976  *
 977  * Input:
 978  *                      buf:            ignored
 979  *                      size:           zero
 980  *
 981  * OR
 982  *
 983  * Input:
 984  *                      buf:            C string containing an unsigned
 985  *                                      integer value representing the new
 986  *                                      NFSv4 lease expiry time
 987  *                      size:           non-zero length of C string in @buf
 988  * Output:
 989  *      On success:     passed-in buffer filled with '\n'-terminated C
 990  *                      string containing unsigned integer value of the
 991  *                      current lease expiry time;
 992  *                      return code is the size in bytes of the string
 993  *      On error:       return code is zero or a negative errno value
 994  */
 995 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
 996 {
 997         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 998         return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
 999 }
1000
1001 /*
1002  * write_gracetime - Set or report current NFSv4 grace period time
1003  *
1004  * As above, but sets the time of the NFSv4 grace period.
1005  *
1006  * Note this should never be set to less than the *previous*
1007  * lease-period time, but we don't try to enforce this.  (In the common
1008  * case (a new boot), we don't know what the previous lease time was
1009  * anyway.)
1010  */
1011 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
1012 {
1013         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1014         return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
1015 }
1016
1017 #ifdef CONFIG_NFSD_LEGACY_CLIENT_TRACKING
1018 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
1019                                    struct nfsd_net *nn)
1020 {
1021         char *mesg = buf;
1022         char *recdir;
1023         int len, status;
1024
1025         if (size > 0) {
1026                 if (nn->nfsd_serv)
1027                         return -EBUSY;
1028                 if (size > PATH_MAX || buf[size-1] != '\n')
1029                         return -EINVAL;
1030                 buf[size-1] = 0;
1031
1032                 recdir = mesg;
1033                 len = qword_get(&mesg, recdir, size);
1034                 if (len <= 0)
1035                         return -EINVAL;
1036                 trace_nfsd_ctl_recoverydir(netns(file), recdir);
1037
1038                 status = nfs4_reset_recoverydir(recdir);
1039                 if (status)
1040                         return status;
1041         }
1042
1043         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1044                                                         nfs4_recoverydir());
1045 }
1046
1047 /*
1048  * write_recoverydir - Set or report the pathname of the recovery directory
1049  *
1050  * Input:
1051  *                      buf:            ignored
1052  *                      size:           zero
1053  *
1054  * OR
1055  *
1056  * Input:
1057  *                      buf:            C string containing the pathname
1058  *                                      of the directory on a local file
1059  *                                      system containing permanent NFSv4
1060  *                                      recovery data
1061  *                      size:           non-zero length of C string in @buf
1062  * Output:
1063  *      On success:     passed-in buffer filled with '\n'-terminated C string
1064  *                      containing the current recovery pathname setting;
1065  *                      return code is the size in bytes of the string
1066  *      On error:       return code is zero or a negative errno value
1067  */
1068 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1069 {
1070         ssize_t rv;
1071         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1072
1073         mutex_lock(&nfsd_mutex);
1074         rv = __write_recoverydir(file, buf, size, nn);
1075         mutex_unlock(&nfsd_mutex);
1076         return rv;
1077 }
1078 #endif
1079
1080 /*
1081  * write_v4_end_grace - release grace period for nfsd's v4.x lock manager
1082  *
1083  * Input:
1084  *                      buf:            ignored
1085  *                      size:           zero
1086  * OR
1087  *
1088  * Input:
1089  *                      buf:            any value
1090  *                      size:           non-zero length of C string in @buf
1091  * Output:
1092  *                      passed-in buffer filled with "Y" or "N" with a newline
1093  *                      and NULL-terminated C string. This indicates whether
1094  *                      the grace period has ended in the current net
1095  *                      namespace. Return code is the size in bytes of the
1096  *                      string. Writing a string that starts with 'Y', 'y', or
1097  *                      '1' to the file will end the grace period for nfsd's v4
1098  *                      lock manager.
1099  */
1100 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size)
1101 {
1102         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1103
1104         if (size > 0) {
1105                 switch(buf[0]) {
1106                 case 'Y':
1107                 case 'y':
1108                 case '1':
1109                         if (!nn->nfsd_serv)
1110                                 return -EBUSY;
1111                         trace_nfsd_end_grace(netns(file));
1112                         nfsd4_end_grace(nn);
1113                         break;
1114                 default:
1115                         return -EINVAL;
1116                 }
1117         }
1118
1119         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%c\n",
1120                          nn->grace_ended ? 'Y' : 'N');
1121 }
1122
1123 #endif
1124
1125 /*----------------------------------------------------------------------------*/
1126 /*
1127  *      populating the filesystem.
1128  */
1129
1130 /* Basically copying rpc_get_inode. */
1131 static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
1132 {
1133         struct inode *inode = new_inode(sb);
1134         if (!inode)
1135                 return NULL;
1136         /* Following advice from simple_fill_super documentation: */
1137         inode->i_ino = iunique(sb, NFSD_MaxReserved);
1138         inode->i_mode = mode;
1139         simple_inode_init_ts(inode);
1140         switch (mode & S_IFMT) {
1141         case S_IFDIR:
1142                 inode->i_fop = &simple_dir_operations;
1143                 inode->i_op = &simple_dir_inode_operations;
1144                 inc_nlink(inode);
1145                 break;
1146         case S_IFLNK:
1147                 inode->i_op = &simple_symlink_inode_operations;
1148                 break;
1149         default:
1150                 break;
1151         }
1152         return inode;
1153 }
1154
1155 static int __nfsd_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode, struct nfsdfs_client *ncl)
1156 {
1157         struct inode *inode;
1158
1159         inode = nfsd_get_inode(dir->i_sb, mode);
1160         if (!inode)
1161                 return -ENOMEM;
1162         if (ncl) {
1163                 inode->i_private = ncl;
1164                 kref_get(&ncl->cl_ref);
1165         }
1166         d_add(dentry, inode);
1167         inc_nlink(dir);
1168         fsnotify_mkdir(dir, dentry);
1169         return 0;
1170 }
1171
1172 static struct dentry *nfsd_mkdir(struct dentry *parent, struct nfsdfs_client *ncl, char *name)
1173 {
1174         struct inode *dir = parent->d_inode;
1175         struct dentry *dentry;
1176         int ret = -ENOMEM;
1177
1178         inode_lock(dir);
1179         dentry = d_alloc_name(parent, name);
1180         if (!dentry)
1181                 goto out_err;
1182         ret = __nfsd_mkdir(d_inode(parent), dentry, S_IFDIR | 0600, ncl);
1183         if (ret)
1184                 goto out_err;
1185 out:
1186         inode_unlock(dir);
1187         return dentry;
1188 out_err:
1189         dput(dentry);
1190         dentry = ERR_PTR(ret);
1191         goto out;
1192 }
1193
1194 #if IS_ENABLED(CONFIG_SUNRPC_GSS)
1195 static int __nfsd_symlink(struct inode *dir, struct dentry *dentry,
1196                           umode_t mode, const char *content)
1197 {
1198         struct inode *inode;
1199
1200         inode = nfsd_get_inode(dir->i_sb, mode);
1201         if (!inode)
1202                 return -ENOMEM;
1203
1204         inode->i_link = (char *)content;
1205         inode->i_size = strlen(content);
1206
1207         d_add(dentry, inode);
1208         inc_nlink(dir);
1209         fsnotify_create(dir, dentry);
1210         return 0;
1211 }
1212
1213 /*
1214  * @content is assumed to be a NUL-terminated string that lives
1215  * longer than the symlink itself.
1216  */
1217 static void _nfsd_symlink(struct dentry *parent, const char *name,
1218                           const char *content)
1219 {
1220         struct inode *dir = parent->d_inode;
1221         struct dentry *dentry;
1222         int ret;
1223
1224         inode_lock(dir);
1225         dentry = d_alloc_name(parent, name);
1226         if (!dentry)
1227                 goto out;
1228         ret = __nfsd_symlink(d_inode(parent), dentry, S_IFLNK | 0777, content);
1229         if (ret)
1230                 dput(dentry);
1231 out:
1232         inode_unlock(dir);
1233 }
1234 #else
1235 static inline void _nfsd_symlink(struct dentry *parent, const char *name,
1236                                  const char *content)
1237 {
1238 }
1239
1240 #endif
1241
1242 static void clear_ncl(struct dentry *dentry)
1243 {
1244         struct inode *inode = d_inode(dentry);
1245         struct nfsdfs_client *ncl = inode->i_private;
1246
1247         spin_lock(&inode->i_lock);
1248         inode->i_private = NULL;
1249         spin_unlock(&inode->i_lock);
1250         kref_put(&ncl->cl_ref, ncl->cl_release);
1251 }
1252
1253 struct nfsdfs_client *get_nfsdfs_client(struct inode *inode)
1254 {
1255         struct nfsdfs_client *nc;
1256
1257         spin_lock(&inode->i_lock);
1258         nc = inode->i_private;
1259         if (nc)
1260                 kref_get(&nc->cl_ref);
1261         spin_unlock(&inode->i_lock);
1262         return nc;
1263 }
1264
1265 /* XXX: cut'n'paste from simple_fill_super; figure out if we could share
1266  * code instead. */
1267 static  int nfsdfs_create_files(struct dentry *root,
1268                                 const struct tree_descr *files,
1269                                 struct nfsdfs_client *ncl,
1270                                 struct dentry **fdentries)
1271 {
1272         struct inode *dir = d_inode(root);
1273         struct inode *inode;
1274         struct dentry *dentry;
1275         int i;
1276
1277         inode_lock(dir);
1278         for (i = 0; files->name && files->name[0]; i++, files++) {
1279                 dentry = d_alloc_name(root, files->name);
1280                 if (!dentry)
1281                         goto out;
1282                 inode = nfsd_get_inode(d_inode(root)->i_sb,
1283                                         S_IFREG | files->mode);
1284                 if (!inode) {
1285                         dput(dentry);
1286                         goto out;
1287                 }
1288                 kref_get(&ncl->cl_ref);
1289                 inode->i_fop = files->ops;
1290                 inode->i_private = ncl;
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         inode_unlock(dir);
1300         return -ENOMEM;
1301 }
1302
1303 /* on success, returns positive number unique to that client. */
1304 struct dentry *nfsd_client_mkdir(struct nfsd_net *nn,
1305                                  struct nfsdfs_client *ncl, u32 id,
1306                                  const struct tree_descr *files,
1307                                  struct dentry **fdentries)
1308 {
1309         struct dentry *dentry;
1310         char name[11];
1311         int ret;
1312
1313         sprintf(name, "%u", id);
1314
1315         dentry = nfsd_mkdir(nn->nfsd_client_dir, ncl, name);
1316         if (IS_ERR(dentry)) /* XXX: tossing errors? */
1317                 return NULL;
1318         ret = nfsdfs_create_files(dentry, files, ncl, fdentries);
1319         if (ret) {
1320                 nfsd_client_rmdir(dentry);
1321                 return NULL;
1322         }
1323         return dentry;
1324 }
1325
1326 /* Taken from __rpc_rmdir: */
1327 void nfsd_client_rmdir(struct dentry *dentry)
1328 {
1329         simple_recursive_removal(dentry, clear_ncl);
1330 }
1331
1332 static int nfsd_fill_super(struct super_block *sb, struct fs_context *fc)
1333 {
1334         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
1335                                                         nfsd_net_id);
1336         struct dentry *dentry;
1337         int ret;
1338
1339         static const struct tree_descr nfsd_files[] = {
1340                 [NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1341                 /* Per-export io stats use same ops as exports file */
1342                 [NFSD_Export_Stats] = {"export_stats", &exports_nfsd_operations, S_IRUGO},
1343                 [NFSD_Export_features] = {"export_features",
1344                                         &export_features_fops, S_IRUGO},
1345                 [NFSD_FO_UnlockIP] = {"unlock_ip",
1346                                         &transaction_ops, S_IWUSR|S_IRUSR},
1347                 [NFSD_FO_UnlockFS] = {"unlock_filesystem",
1348                                         &transaction_ops, S_IWUSR|S_IRUSR},
1349                 [NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1350                 [NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1351                 [NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1352                 [NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1353                 [NFSD_Reply_Cache_Stats] = {"reply_cache_stats",
1354                                         &nfsd_reply_cache_stats_fops, S_IRUGO},
1355                 [NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1356                 [NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1357                 [NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1358                 [NFSD_MaxConnections] = {"max_connections", &transaction_ops, S_IWUSR|S_IRUGO},
1359                 [NFSD_Filecache] = {"filecache", &nfsd_file_cache_stats_fops, S_IRUGO},
1360 #ifdef CONFIG_NFSD_V4
1361                 [NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1362                 [NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1363                 [NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1364                 [NFSD_V4EndGrace] = {"v4_end_grace", &transaction_ops, S_IWUSR|S_IRUGO},
1365 #endif
1366                 /* last one */ {""}
1367         };
1368
1369         ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1370         if (ret)
1371                 return ret;
1372         _nfsd_symlink(sb->s_root, "supported_krb5_enctypes",
1373                       "/proc/net/rpc/gss_krb5_enctypes");
1374         dentry = nfsd_mkdir(sb->s_root, NULL, "clients");
1375         if (IS_ERR(dentry))
1376                 return PTR_ERR(dentry);
1377         nn->nfsd_client_dir = dentry;
1378         return 0;
1379 }
1380
1381 static int nfsd_fs_get_tree(struct fs_context *fc)
1382 {
1383         return get_tree_keyed(fc, nfsd_fill_super, get_net(fc->net_ns));
1384 }
1385
1386 static void nfsd_fs_free_fc(struct fs_context *fc)
1387 {
1388         if (fc->s_fs_info)
1389                 put_net(fc->s_fs_info);
1390 }
1391
1392 static const struct fs_context_operations nfsd_fs_context_ops = {
1393         .free           = nfsd_fs_free_fc,
1394         .get_tree       = nfsd_fs_get_tree,
1395 };
1396
1397 static int nfsd_init_fs_context(struct fs_context *fc)
1398 {
1399         put_user_ns(fc->user_ns);
1400         fc->user_ns = get_user_ns(fc->net_ns->user_ns);
1401         fc->ops = &nfsd_fs_context_ops;
1402         return 0;
1403 }
1404
1405 static void nfsd_umount(struct super_block *sb)
1406 {
1407         struct net *net = sb->s_fs_info;
1408
1409         nfsd_shutdown_threads(net);
1410
1411         kill_litter_super(sb);
1412         put_net(net);
1413 }
1414
1415 static struct file_system_type nfsd_fs_type = {
1416         .owner          = THIS_MODULE,
1417         .name           = "nfsd",
1418         .init_fs_context = nfsd_init_fs_context,
1419         .kill_sb        = nfsd_umount,
1420 };
1421 MODULE_ALIAS_FS("nfsd");
1422
1423 #ifdef CONFIG_PROC_FS
1424
1425 static int exports_proc_open(struct inode *inode, struct file *file)
1426 {
1427         return exports_net_open(current->nsproxy->net_ns, file);
1428 }
1429
1430 static const struct proc_ops exports_proc_ops = {
1431         .proc_open      = exports_proc_open,
1432         .proc_read      = seq_read,
1433         .proc_lseek     = seq_lseek,
1434         .proc_release   = seq_release,
1435 };
1436
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 /**
1461  * nfsd_nl_rpc_status_get_start - Prepare rpc_status_get dumpit
1462  * @cb: netlink metadata and command arguments
1463  *
1464  * Return values:
1465  *   %0: The rpc_status_get command may proceed
1466  *   %-ENODEV: There is no NFSD running in this namespace
1467  */
1468 int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb)
1469 {
1470         struct nfsd_net *nn = net_generic(sock_net(cb->skb->sk), nfsd_net_id);
1471         int ret = -ENODEV;
1472
1473         mutex_lock(&nfsd_mutex);
1474         if (nn->nfsd_serv)
1475                 ret = 0;
1476         else
1477                 mutex_unlock(&nfsd_mutex);
1478
1479         return ret;
1480 }
1481
1482 static int nfsd_genl_rpc_status_compose_msg(struct sk_buff *skb,
1483                                             struct netlink_callback *cb,
1484                                             struct nfsd_genl_rqstp *rqstp)
1485 {
1486         void *hdr;
1487         u32 i;
1488
1489         hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1490                           &nfsd_nl_family, 0, NFSD_CMD_RPC_STATUS_GET);
1491         if (!hdr)
1492                 return -ENOBUFS;
1493
1494         if (nla_put_be32(skb, NFSD_A_RPC_STATUS_XID, rqstp->rq_xid) ||
1495             nla_put_u32(skb, NFSD_A_RPC_STATUS_FLAGS, rqstp->rq_flags) ||
1496             nla_put_u32(skb, NFSD_A_RPC_STATUS_PROG, rqstp->rq_prog) ||
1497             nla_put_u32(skb, NFSD_A_RPC_STATUS_PROC, rqstp->rq_proc) ||
1498             nla_put_u8(skb, NFSD_A_RPC_STATUS_VERSION, rqstp->rq_vers) ||
1499             nla_put_s64(skb, NFSD_A_RPC_STATUS_SERVICE_TIME,
1500                         ktime_to_us(rqstp->rq_stime),
1501                         NFSD_A_RPC_STATUS_PAD))
1502                 return -ENOBUFS;
1503
1504         switch (rqstp->rq_saddr.sa_family) {
1505         case AF_INET: {
1506                 const struct sockaddr_in *s_in, *d_in;
1507
1508                 s_in = (const struct sockaddr_in *)&rqstp->rq_saddr;
1509                 d_in = (const struct sockaddr_in *)&rqstp->rq_daddr;
1510                 if (nla_put_in_addr(skb, NFSD_A_RPC_STATUS_SADDR4,
1511                                     s_in->sin_addr.s_addr) ||
1512                     nla_put_in_addr(skb, NFSD_A_RPC_STATUS_DADDR4,
1513                                     d_in->sin_addr.s_addr) ||
1514                     nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1515                                  s_in->sin_port) ||
1516                     nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1517                                  d_in->sin_port))
1518                         return -ENOBUFS;
1519                 break;
1520         }
1521         case AF_INET6: {
1522                 const struct sockaddr_in6 *s_in, *d_in;
1523
1524                 s_in = (const struct sockaddr_in6 *)&rqstp->rq_saddr;
1525                 d_in = (const struct sockaddr_in6 *)&rqstp->rq_daddr;
1526                 if (nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_SADDR6,
1527                                      &s_in->sin6_addr) ||
1528                     nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_DADDR6,
1529                                      &d_in->sin6_addr) ||
1530                     nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1531                                  s_in->sin6_port) ||
1532                     nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1533                                  d_in->sin6_port))
1534                         return -ENOBUFS;
1535                 break;
1536         }
1537         }
1538
1539         for (i = 0; i < rqstp->rq_opcnt; i++)
1540                 if (nla_put_u32(skb, NFSD_A_RPC_STATUS_COMPOUND_OPS,
1541                                 rqstp->rq_opnum[i]))
1542                         return -ENOBUFS;
1543
1544         genlmsg_end(skb, hdr);
1545         return 0;
1546 }
1547
1548 /**
1549  * nfsd_nl_rpc_status_get_dumpit - Handle rpc_status_get dumpit
1550  * @skb: reply buffer
1551  * @cb: netlink metadata and command arguments
1552  *
1553  * Returns the size of the reply or a negative errno.
1554  */
1555 int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
1556                                   struct netlink_callback *cb)
1557 {
1558         struct nfsd_net *nn = net_generic(sock_net(skb->sk), nfsd_net_id);
1559         int i, ret, rqstp_index = 0;
1560
1561         rcu_read_lock();
1562
1563         for (i = 0; i < nn->nfsd_serv->sv_nrpools; i++) {
1564                 struct svc_rqst *rqstp;
1565
1566                 if (i < cb->args[0]) /* already consumed */
1567                         continue;
1568
1569                 rqstp_index = 0;
1570                 list_for_each_entry_rcu(rqstp,
1571                                 &nn->nfsd_serv->sv_pools[i].sp_all_threads,
1572                                 rq_all) {
1573                         struct nfsd_genl_rqstp genl_rqstp;
1574                         unsigned int status_counter;
1575
1576                         if (rqstp_index++ < cb->args[1]) /* already consumed */
1577                                 continue;
1578                         /*
1579                          * Acquire rq_status_counter before parsing the rqst
1580                          * fields. rq_status_counter is set to an odd value in
1581                          * order to notify the consumers the rqstp fields are
1582                          * meaningful.
1583                          */
1584                         status_counter =
1585                                 smp_load_acquire(&rqstp->rq_status_counter);
1586                         if (!(status_counter & 1))
1587                                 continue;
1588
1589                         genl_rqstp.rq_xid = rqstp->rq_xid;
1590                         genl_rqstp.rq_flags = rqstp->rq_flags;
1591                         genl_rqstp.rq_vers = rqstp->rq_vers;
1592                         genl_rqstp.rq_prog = rqstp->rq_prog;
1593                         genl_rqstp.rq_proc = rqstp->rq_proc;
1594                         genl_rqstp.rq_stime = rqstp->rq_stime;
1595                         genl_rqstp.rq_opcnt = 0;
1596                         memcpy(&genl_rqstp.rq_daddr, svc_daddr(rqstp),
1597                                sizeof(struct sockaddr));
1598                         memcpy(&genl_rqstp.rq_saddr, svc_addr(rqstp),
1599                                sizeof(struct sockaddr));
1600
1601 #ifdef CONFIG_NFSD_V4
1602                         if (rqstp->rq_vers == NFS4_VERSION &&
1603                             rqstp->rq_proc == NFSPROC4_COMPOUND) {
1604                                 /* NFSv4 compound */
1605                                 struct nfsd4_compoundargs *args;
1606                                 int j;
1607
1608                                 args = rqstp->rq_argp;
1609                                 genl_rqstp.rq_opcnt = args->opcnt;
1610                                 for (j = 0; j < genl_rqstp.rq_opcnt; j++)
1611                                         genl_rqstp.rq_opnum[j] =
1612                                                 args->ops[j].opnum;
1613                         }
1614 #endif /* CONFIG_NFSD_V4 */
1615
1616                         /*
1617                          * Acquire rq_status_counter before reporting the rqst
1618                          * fields to the user.
1619                          */
1620                         if (smp_load_acquire(&rqstp->rq_status_counter) !=
1621                             status_counter)
1622                                 continue;
1623
1624                         ret = nfsd_genl_rpc_status_compose_msg(skb, cb,
1625                                                                &genl_rqstp);
1626                         if (ret)
1627                                 goto out;
1628                 }
1629         }
1630
1631         cb->args[0] = i;
1632         cb->args[1] = rqstp_index;
1633         ret = skb->len;
1634 out:
1635         rcu_read_unlock();
1636
1637         return ret;
1638 }
1639
1640 /**
1641  * nfsd_nl_rpc_status_get_done - rpc_status_get dumpit post-processing
1642  * @cb: netlink metadata and command arguments
1643  *
1644  * Return values:
1645  *   %0: Success
1646  */
1647 int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb)
1648 {
1649         mutex_unlock(&nfsd_mutex);
1650
1651         return 0;
1652 }
1653
1654 /**
1655  * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
1656  * @net: a freshly-created network namespace
1657  *
1658  * This information stays around as long as the network namespace is
1659  * alive whether or not there is an NFSD instance running in the
1660  * namespace.
1661  *
1662  * Returns zero on success, or a negative errno otherwise.
1663  */
1664 static __net_init int nfsd_net_init(struct net *net)
1665 {
1666         int retval;
1667         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1668
1669         retval = nfsd_export_init(net);
1670         if (retval)
1671                 goto out_export_error;
1672         retval = nfsd_idmap_init(net);
1673         if (retval)
1674                 goto out_idmap_error;
1675         retval = nfsd_stat_counters_init(nn);
1676         if (retval)
1677                 goto out_repcache_error;
1678         memset(&nn->nfsd_svcstats, 0, sizeof(nn->nfsd_svcstats));
1679         nn->nfsd_svcstats.program = &nfsd_program;
1680         nn->nfsd_versions = NULL;
1681         nn->nfsd4_minorversions = NULL;
1682         nfsd4_init_leases_net(nn);
1683         get_random_bytes(&nn->siphash_key, sizeof(nn->siphash_key));
1684         seqlock_init(&nn->writeverf_lock);
1685         nfsd_proc_stat_init(net);
1686
1687         return 0;
1688
1689 out_repcache_error:
1690         nfsd_idmap_shutdown(net);
1691 out_idmap_error:
1692         nfsd_export_shutdown(net);
1693 out_export_error:
1694         return retval;
1695 }
1696
1697 /**
1698  * nfsd_net_exit - Release the nfsd_net portion of a net namespace
1699  * @net: a network namespace that is about to be destroyed
1700  *
1701  */
1702 static __net_exit void nfsd_net_exit(struct net *net)
1703 {
1704         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1705
1706         nfsd_proc_stat_shutdown(net);
1707         nfsd_stat_counters_destroy(nn);
1708         nfsd_idmap_shutdown(net);
1709         nfsd_export_shutdown(net);
1710         nfsd_netns_free_versions(nn);
1711 }
1712
1713 static struct pernet_operations nfsd_net_ops = {
1714         .init = nfsd_net_init,
1715         .exit = nfsd_net_exit,
1716         .id   = &nfsd_net_id,
1717         .size = sizeof(struct nfsd_net),
1718 };
1719
1720 static int __init init_nfsd(void)
1721 {
1722         int retval;
1723
1724         retval = nfsd4_init_slabs();
1725         if (retval)
1726                 return retval;
1727         retval = nfsd4_init_pnfs();
1728         if (retval)
1729                 goto out_free_slabs;
1730         retval = nfsd_drc_slab_create();
1731         if (retval)
1732                 goto out_free_pnfs;
1733         nfsd_lockd_init();      /* lockd->nfsd callbacks */
1734         retval = create_proc_exports_entry();
1735         if (retval)
1736                 goto out_free_lockd;
1737         retval = register_pernet_subsys(&nfsd_net_ops);
1738         if (retval < 0)
1739                 goto out_free_exports;
1740         retval = register_cld_notifier();
1741         if (retval)
1742                 goto out_free_subsys;
1743         retval = nfsd4_create_laundry_wq();
1744         if (retval)
1745                 goto out_free_cld;
1746         retval = register_filesystem(&nfsd_fs_type);
1747         if (retval)
1748                 goto out_free_all;
1749         retval = genl_register_family(&nfsd_nl_family);
1750         if (retval)
1751                 goto out_free_all;
1752
1753         return 0;
1754 out_free_all:
1755         nfsd4_destroy_laundry_wq();
1756 out_free_cld:
1757         unregister_cld_notifier();
1758 out_free_subsys:
1759         unregister_pernet_subsys(&nfsd_net_ops);
1760 out_free_exports:
1761         remove_proc_entry("fs/nfs/exports", NULL);
1762         remove_proc_entry("fs/nfs", NULL);
1763 out_free_lockd:
1764         nfsd_lockd_shutdown();
1765         nfsd_drc_slab_free();
1766 out_free_pnfs:
1767         nfsd4_exit_pnfs();
1768 out_free_slabs:
1769         nfsd4_free_slabs();
1770         return retval;
1771 }
1772
1773 static void __exit exit_nfsd(void)
1774 {
1775         genl_unregister_family(&nfsd_nl_family);
1776         unregister_filesystem(&nfsd_fs_type);
1777         nfsd4_destroy_laundry_wq();
1778         unregister_cld_notifier();
1779         unregister_pernet_subsys(&nfsd_net_ops);
1780         nfsd_drc_slab_free();
1781         remove_proc_entry("fs/nfs/exports", NULL);
1782         remove_proc_entry("fs/nfs", NULL);
1783         nfsd_lockd_shutdown();
1784         nfsd4_free_slabs();
1785         nfsd4_exit_pnfs();
1786 }
1787
1788 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1789 MODULE_DESCRIPTION("In-kernel NFS server");
1790 MODULE_LICENSE("GPL");
1791 module_init(init_nfsd)
1792 module_exit(exit_nfsd)