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         /*
  52          * The below MUST come last.  Otherwise we leave a hole in nfsd_files[]
  53          * with !CONFIG_NFSD_V4 and simple_fill_super() goes oops
  54          */
  55 #ifdef CONFIG_NFSD_V4
  56         NFSD_Leasetime,
  57         NFSD_Gracetime,
  58         NFSD_RecoveryDir,
  59         NFSD_V4EndGrace,
  60 #endif
  61         NFSD_MaxReserved
  62 };
  63
  64 /*
  65  * write() for these nodes.
  66  */
  67 static ssize_t write_filehandle(struct file *file, char *buf, size_t size);
  68 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size);
  69 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size);
  70 static ssize_t write_threads(struct file *file, char *buf, size_t size);
  71 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size);
  72 static ssize_t write_versions(struct file *file, char *buf, size_t size);
  73 static ssize_t write_ports(struct file *file, char *buf, size_t size);
  74 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size);
  75 static ssize_t write_maxconn(struct file *file, char *buf, size_t size);
  76 #ifdef CONFIG_NFSD_V4
  77 static ssize_t write_leasetime(struct file *file, char *buf, size_t size);
  78 static ssize_t write_gracetime(struct file *file, char *buf, size_t size);
  79 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size);
  80 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size);
  81 #endif
  82
  83 static ssize_t (*const write_op[])(struct file *, char *, size_t) = {
  84         [NFSD_Fh] = write_filehandle,
  85         [NFSD_FO_UnlockIP] = write_unlock_ip,
  86         [NFSD_FO_UnlockFS] = write_unlock_fs,
  87         [NFSD_Threads] = write_threads,
  88         [NFSD_Pool_Threads] = write_pool_threads,
  89         [NFSD_Versions] = write_versions,
  90         [NFSD_Ports] = write_ports,
  91         [NFSD_MaxBlkSize] = write_maxblksize,
  92         [NFSD_MaxConnections] = write_maxconn,
  93 #ifdef CONFIG_NFSD_V4
  94         [NFSD_Leasetime] = write_leasetime,
  95         [NFSD_Gracetime] = write_gracetime,
  96         [NFSD_RecoveryDir] = write_recoverydir,
  97         [NFSD_V4EndGrace] = write_v4_end_grace,
  98 #endif
  99 };
 100
 101 static ssize_t nfsctl_transaction_write(struct file *file, const char __user *buf, size_t size, loff_t *pos)
 102 {
 103         ino_t ino =  file_inode(file)->i_ino;
 104         char *data;
 105         ssize_t rv;
 106
 107         if (ino >= ARRAY_SIZE(write_op) || !write_op[ino])
 108                 return -EINVAL;
 109
 110         data = simple_transaction_get(file, buf, size);
 111         if (IS_ERR(data))
 112                 return PTR_ERR(data);
 113
 114         rv = write_op[ino](file, data, size);
 115         if (rv < 0)
 116                 return rv;
 117
 118         simple_transaction_set(file, rv);
 119         return size;
 120 }
 121
 122 static ssize_t nfsctl_transaction_read(struct file *file, char __user *buf, size_t size, loff_t *pos)
 123 {
 124         if (! file->private_data) {
 125                 /* An attempt to read a transaction file without writing
 126                  * causes a 0-byte write so that the file can return
 127                  * state information
 128                  */
 129                 ssize_t rv = nfsctl_transaction_write(file, buf, 0, pos);
 130                 if (rv < 0)
 131                         return rv;
 132         }
 133         return simple_transaction_read(file, buf, size, pos);
 134 }
 135
 136 static const struct file_operations transaction_ops = {
 137         .write          = nfsctl_transaction_write,
 138         .read           = nfsctl_transaction_read,
 139         .release        = simple_transaction_release,
 140         .llseek         = default_llseek,
 141 };
 142
 143 static int exports_net_open(struct net *net, struct file *file)
 144 {
 145         int err;
 146         struct seq_file *seq;
 147         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 148
 149         err = seq_open(file, &nfs_exports_op);
 150         if (err)
 151                 return err;
 152
 153         seq = file->private_data;
 154         seq->private = nn->svc_export_cache;
 155         return 0;
 156 }
 157
 158 static int exports_nfsd_open(struct inode *inode, struct file *file)
 159 {
 160         return exports_net_open(inode->i_sb->s_fs_info, file);
 161 }
 162
 163 static const struct file_operations exports_nfsd_operations = {
 164         .open           = exports_nfsd_open,
 165         .read           = seq_read,
 166         .llseek         = seq_lseek,
 167         .release        = seq_release,
 168 };
 169
 170 static int export_features_show(struct seq_file *m, void *v)
 171 {
 172         seq_printf(m, "0x%x 0x%x\n", NFSEXP_ALLFLAGS, NFSEXP_SECINFO_FLAGS);
 173         return 0;
 174 }
 175
 176 DEFINE_SHOW_ATTRIBUTE(export_features);
 177
 178 static const struct file_operations pool_stats_operations = {
 179         .open           = nfsd_pool_stats_open,
 180         .read           = seq_read,
 181         .llseek         = seq_lseek,
 182         .release        = nfsd_pool_stats_release,
 183 };
 184
 185 DEFINE_SHOW_ATTRIBUTE(nfsd_reply_cache_stats);
 186
 187 DEFINE_SHOW_ATTRIBUTE(nfsd_file_cache_stats);
 188
 189 /*----------------------------------------------------------------------------*/
 190 /*
 191  * payload - write methods
 192  */
 193
 194 static inline struct net *netns(struct file *file)
 195 {
 196         return file_inode(file)->i_sb->s_fs_info;
 197 }
 198
 199 /*
 200  * write_unlock_ip - Release all locks used by a client
 201  *
 202  * Experimental.
 203  *
 204  * Input:
 205  *                      buf:    '\n'-terminated C string containing a
 206  *                              presentation format IP address
 207  *                      size:   length of C string in @buf
 208  * Output:
 209  *      On success:     returns zero if all specified locks were released;
 210  *                      returns one if one or more locks were not released
 211  *      On error:       return code is negative errno value
 212  */
 213 static ssize_t write_unlock_ip(struct file *file, char *buf, size_t size)
 214 {
 215         struct sockaddr_storage address;
 216         struct sockaddr *sap = (struct sockaddr *)&address;
 217         size_t salen = sizeof(address);
 218         char *fo_path;
 219         struct net *net = netns(file);
 220
 221         /* sanity check */
 222         if (size == 0)
 223                 return -EINVAL;
 224
 225         if (buf[size-1] != '\n')
 226                 return -EINVAL;
 227
 228         fo_path = buf;
 229         if (qword_get(&buf, fo_path, size) < 0)
 230                 return -EINVAL;
 231
 232         if (rpc_pton(net, fo_path, size, sap, salen) == 0)
 233                 return -EINVAL;
 234
 235         trace_nfsd_ctl_unlock_ip(net, buf);
 236         return nlmsvc_unlock_all_by_ip(sap);
 237 }
 238
 239 /*
 240  * write_unlock_fs - Release all locks on a local file system
 241  *
 242  * Experimental.
 243  *
 244  * Input:
 245  *                      buf:    '\n'-terminated C string containing the
 246  *                              absolute pathname of a local file system
 247  *                      size:   length of C string in @buf
 248  * Output:
 249  *      On success:     returns zero if all specified locks were released;
 250  *                      returns one if one or more locks were not released
 251  *      On error:       return code is negative errno value
 252  */
 253 static ssize_t write_unlock_fs(struct file *file, char *buf, size_t size)
 254 {
 255         struct path path;
 256         char *fo_path;
 257         int error;
 258
 259         /* sanity check */
 260         if (size == 0)
 261                 return -EINVAL;
 262
 263         if (buf[size-1] != '\n')
 264                 return -EINVAL;
 265
 266         fo_path = buf;
 267         if (qword_get(&buf, fo_path, size) < 0)
 268                 return -EINVAL;
 269         trace_nfsd_ctl_unlock_fs(netns(file), fo_path);
 270         error = kern_path(fo_path, 0, &path);
 271         if (error)
 272                 return error;
 273
 274         /*
 275          * XXX: Needs better sanity checking.  Otherwise we could end up
 276          * releasing locks on the wrong file system.
 277          *
 278          * For example:
 279          * 1.  Does the path refer to a directory?
 280          * 2.  Is that directory a mount point, or
 281          * 3.  Is that directory the root of an exported file system?
 282          */
 283         error = nlmsvc_unlock_all_by_sb(path.dentry->d_sb);
 284
 285         path_put(&path);
 286         return error;
 287 }
 288
 289 /*
 290  * write_filehandle - Get a variable-length NFS file handle by path
 291  *
 292  * On input, the buffer contains a '\n'-terminated C string comprised of
 293  * three alphanumeric words separated by whitespace.  The string may
 294  * contain escape sequences.
 295  *
 296  * Input:
 297  *                      buf:
 298  *                              domain:         client domain name
 299  *                              path:           export pathname
 300  *                              maxsize:        numeric maximum size of
 301  *                                              @buf
 302  *                      size:   length of C string in @buf
 303  * Output:
 304  *      On success:     passed-in buffer filled with '\n'-terminated C
 305  *                      string containing a ASCII hex text version
 306  *                      of the NFS file handle;
 307  *                      return code is the size in bytes of the string
 308  *      On error:       return code is negative errno value
 309  */
 310 static ssize_t write_filehandle(struct file *file, char *buf, size_t size)
 311 {
 312         char *dname, *path;
 313         int maxsize;
 314         char *mesg = buf;
 315         int len;
 316         struct auth_domain *dom;
 317         struct knfsd_fh fh;
 318
 319         if (size == 0)
 320                 return -EINVAL;
 321
 322         if (buf[size-1] != '\n')
 323                 return -EINVAL;
 324         buf[size-1] = 0;
 325
 326         dname = mesg;
 327         len = qword_get(&mesg, dname, size);
 328         if (len <= 0)
 329                 return -EINVAL;
 330
 331         path = dname+len+1;
 332         len = qword_get(&mesg, path, size);
 333         if (len <= 0)
 334                 return -EINVAL;
 335
 336         len = get_int(&mesg, &maxsize);
 337         if (len)
 338                 return len;
 339
 340         if (maxsize < NFS_FHSIZE)
 341                 return -EINVAL;
 342         maxsize = min(maxsize, NFS3_FHSIZE);
 343
 344         if (qword_get(&mesg, mesg, size) > 0)
 345                 return -EINVAL;
 346
 347         trace_nfsd_ctl_filehandle(netns(file), dname, path, maxsize);
 348
 349         /* we have all the words, they are in buf.. */
 350         dom = unix_domain_find(dname);
 351         if (!dom)
 352                 return -ENOMEM;
 353
 354         len = exp_rootfh(netns(file), dom, path, &fh, maxsize);
 355         auth_domain_put(dom);
 356         if (len)
 357                 return len;
 358
 359         mesg = buf;
 360         len = SIMPLE_TRANSACTION_LIMIT;
 361         qword_addhex(&mesg, &len, fh.fh_raw, fh.fh_size);
 362         mesg[-1] = '\n';
 363         return mesg - buf;
 364 }
 365
 366 /*
 367  * write_threads - Start NFSD, or report the current number of running threads
 368  *
 369  * Input:
 370  *                      buf:            ignored
 371  *                      size:           zero
 372  * Output:
 373  *      On success:     passed-in buffer filled with '\n'-terminated C
 374  *                      string numeric value representing the number of
 375  *                      running NFSD threads;
 376  *                      return code is the size in bytes of the string
 377  *      On error:       return code is zero
 378  *
 379  * OR
 380  *
 381  * Input:
 382  *                      buf:            C string containing an unsigned
 383  *                                      integer value representing the
 384  *                                      number of NFSD threads to start
 385  *                      size:           non-zero length of C string in @buf
 386  * Output:
 387  *      On success:     NFS service is started;
 388  *                      passed-in buffer filled with '\n'-terminated C
 389  *                      string numeric value representing the number of
 390  *                      running NFSD threads;
 391  *                      return code is the size in bytes of the string
 392  *      On error:       return code is zero or a negative errno value
 393  */
 394 static ssize_t write_threads(struct file *file, char *buf, size_t size)
 395 {
 396         char *mesg = buf;
 397         int rv;
 398         struct net *net = netns(file);
 399
 400         if (size > 0) {
 401                 int newthreads;
 402                 rv = get_int(&mesg, &newthreads);
 403                 if (rv)
 404                         return rv;
 405                 if (newthreads < 0)
 406                         return -EINVAL;
 407                 trace_nfsd_ctl_threads(net, newthreads);
 408                 rv = nfsd_svc(newthreads, net, file->f_cred);
 409                 if (rv < 0)
 410                         return rv;
 411         } else
 412                 rv = nfsd_nrthreads(net);
 413
 414         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n", rv);
 415 }
 416
 417 /*
 418  * write_pool_threads - Set or report the current number of threads per pool
 419  *
 420  * Input:
 421  *                      buf:            ignored
 422  *                      size:           zero
 423  *
 424  * OR
 425  *
 426  * Input:
 427  *                      buf:            C string containing whitespace-
 428  *                                      separated unsigned integer values
 429  *                                      representing the number of NFSD
 430  *                                      threads to start in each pool
 431  *                      size:           non-zero length of C string in @buf
 432  * Output:
 433  *      On success:     passed-in buffer filled with '\n'-terminated C
 434  *                      string containing integer values representing the
 435  *                      number of NFSD threads in each pool;
 436  *                      return code is the size in bytes of the string
 437  *      On error:       return code is zero or a negative errno value
 438  */
 439 static ssize_t write_pool_threads(struct file *file, char *buf, size_t size)
 440 {
 441         /* if size > 0, look for an array of number of threads per node
 442          * and apply them  then write out number of threads per node as reply
 443          */
 444         char *mesg = buf;
 445         int i;
 446         int rv;
 447         int len;
 448         int npools;
 449         int *nthreads;
 450         struct net *net = netns(file);
 451
 452         mutex_lock(&nfsd_mutex);
 453         npools = nfsd_nrpools(net);
 454         if (npools == 0) {
 455                 /*
 456                  * NFS is shut down.  The admin can start it by
 457                  * writing to the threads file but NOT the pool_threads
 458                  * file, sorry.  Report zero threads.
 459                  */
 460                 mutex_unlock(&nfsd_mutex);
 461                 strcpy(buf, "0\n");
 462                 return strlen(buf);
 463         }
 464
 465         nthreads = kcalloc(npools, sizeof(int), GFP_KERNEL);
 466         rv = -ENOMEM;
 467         if (nthreads == NULL)
 468                 goto out_free;
 469
 470         if (size > 0) {
 471                 for (i = 0; i < npools; i++) {
 472                         rv = get_int(&mesg, &nthreads[i]);
 473                         if (rv == -ENOENT)
 474                                 break;          /* fewer numbers than pools */
 475                         if (rv)
 476                                 goto out_free;  /* syntax error */
 477                         rv = -EINVAL;
 478                         if (nthreads[i] < 0)
 479                                 goto out_free;
 480                         trace_nfsd_ctl_pool_threads(net, i, nthreads[i]);
 481                 }
 482                 rv = nfsd_set_nrthreads(i, nthreads, net);
 483                 if (rv)
 484                         goto out_free;
 485         }
 486
 487         rv = nfsd_get_nrthreads(npools, nthreads, net);
 488         if (rv)
 489                 goto out_free;
 490
 491         mesg = buf;
 492         size = SIMPLE_TRANSACTION_LIMIT;
 493         for (i = 0; i < npools && size > 0; i++) {
 494                 snprintf(mesg, size, "%d%c", nthreads[i], (i == npools-1 ? '\n' : ' '));
 495                 len = strlen(mesg);
 496                 size -= len;
 497                 mesg += len;
 498         }
 499         rv = mesg - buf;
 500 out_free:
 501         kfree(nthreads);
 502         mutex_unlock(&nfsd_mutex);
 503         return rv;
 504 }
 505
 506 static ssize_t
 507 nfsd_print_version_support(struct nfsd_net *nn, char *buf, int remaining,
 508                 const char *sep, unsigned vers, int minor)
 509 {
 510         const char *format = minor < 0 ? "%s%c%u" : "%s%c%u.%u";
 511         bool supported = !!nfsd_vers(nn, vers, NFSD_TEST);
 512
 513         if (vers == 4 && minor >= 0 &&
 514             !nfsd_minorversion(nn, minor, NFSD_TEST))
 515                 supported = false;
 516         if (minor == 0 && supported)
 517                 /*
 518                  * special case for backward compatability.
 519                  * +4.0 is never reported, it is implied by
 520                  * +4, unless -4.0 is present.
 521                  */
 522                 return 0;
 523         return snprintf(buf, remaining, format, sep,
 524                         supported ? '+' : '-', vers, minor);
 525 }
 526
 527 static ssize_t __write_versions(struct file *file, char *buf, size_t size)
 528 {
 529         char *mesg = buf;
 530         char *vers, *minorp, sign;
 531         int len, num, remaining;
 532         ssize_t tlen = 0;
 533         char *sep;
 534         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 535
 536         if (size > 0) {
 537                 if (nn->nfsd_serv)
 538                         /* Cannot change versions without updating
 539                          * nn->nfsd_serv->sv_xdrsize, and reallocing
 540                          * rq_argp and rq_resp
 541                          */
 542                         return -EBUSY;
 543                 if (buf[size-1] != '\n')
 544                         return -EINVAL;
 545                 buf[size-1] = 0;
 546                 trace_nfsd_ctl_version(netns(file), buf);
 547
 548                 vers = mesg;
 549                 len = qword_get(&mesg, vers, size);
 550                 if (len <= 0) return -EINVAL;
 551                 do {
 552                         enum vers_op cmd;
 553                         unsigned minor;
 554                         sign = *vers;
 555                         if (sign == '+' || sign == '-')
 556                                 num = simple_strtol((vers+1), &minorp, 0);
 557                         else
 558                                 num = simple_strtol(vers, &minorp, 0);
 559                         if (*minorp == '.') {
 560                                 if (num != 4)
 561                                         return -EINVAL;
 562                                 if (kstrtouint(minorp+1, 0, &minor) < 0)
 563                                         return -EINVAL;
 564                         }
 565
 566                         cmd = sign == '-' ? NFSD_CLEAR : NFSD_SET;
 567                         switch(num) {
 568 #ifdef CONFIG_NFSD_V2
 569                         case 2:
 570 #endif
 571                         case 3:
 572                                 nfsd_vers(nn, num, cmd);
 573                                 break;
 574                         case 4:
 575                                 if (*minorp == '.') {
 576                                         if (nfsd_minorversion(nn, minor, cmd) < 0)
 577                                                 return -EINVAL;
 578                                 } else if ((cmd == NFSD_SET) != nfsd_vers(nn, num, NFSD_TEST)) {
 579                                         /*
 580                                          * Either we have +4 and no minors are enabled,
 581                                          * or we have -4 and at least one minor is enabled.
 582                                          * In either case, propagate 'cmd' to all minors.
 583                                          */
 584                                         minor = 0;
 585                                         while (nfsd_minorversion(nn, minor, cmd) >= 0)
 586                                                 minor++;
 587                                 }
 588                                 break;
 589                         default:
 590                                 /* Ignore requests to disable non-existent versions */
 591                                 if (cmd == NFSD_SET)
 592                                         return -EINVAL;
 593                         }
 594                         vers += len + 1;
 595                 } while ((len = qword_get(&mesg, vers, size)) > 0);
 596                 /* If all get turned off, turn them back on, as
 597                  * having no versions is BAD
 598                  */
 599                 nfsd_reset_versions(nn);
 600         }
 601
 602         /* Now write current state into reply buffer */
 603         sep = "";
 604         remaining = SIMPLE_TRANSACTION_LIMIT;
 605         for (num=2 ; num <= 4 ; num++) {
 606                 int minor;
 607                 if (!nfsd_vers(nn, num, NFSD_AVAIL))
 608                         continue;
 609
 610                 minor = -1;
 611                 do {
 612                         len = nfsd_print_version_support(nn, buf, remaining,
 613                                         sep, num, minor);
 614                         if (len >= remaining)
 615                                 goto out;
 616                         remaining -= len;
 617                         buf += len;
 618                         tlen += len;
 619                         minor++;
 620                         if (len)
 621                                 sep = " ";
 622                 } while (num == 4 && minor <= NFSD_SUPPORTED_MINOR_VERSION);
 623         }
 624 out:
 625         len = snprintf(buf, remaining, "\n");
 626         if (len >= remaining)
 627                 return -EINVAL;
 628         return tlen + len;
 629 }
 630
 631 /*
 632  * write_versions - Set or report the available NFS protocol versions
 633  *
 634  * Input:
 635  *                      buf:            ignored
 636  *                      size:           zero
 637  * Output:
 638  *      On success:     passed-in buffer filled with '\n'-terminated C
 639  *                      string containing positive or negative integer
 640  *                      values representing the current status of each
 641  *                      protocol version;
 642  *                      return code is the size in bytes of the string
 643  *      On error:       return code is zero or a negative errno value
 644  *
 645  * OR
 646  *
 647  * Input:
 648  *                      buf:            C string containing whitespace-
 649  *                                      separated positive or negative
 650  *                                      integer values representing NFS
 651  *                                      protocol versions to enable ("+n")
 652  *                                      or disable ("-n")
 653  *                      size:           non-zero length of C string in @buf
 654  * Output:
 655  *      On success:     status of zero or more protocol versions has
 656  *                      been updated; passed-in buffer filled with
 657  *                      '\n'-terminated C string containing positive
 658  *                      or negative integer values representing the
 659  *                      current status of each protocol version;
 660  *                      return code is the size in bytes of the string
 661  *      On error:       return code is zero or a negative errno value
 662  */
 663 static ssize_t write_versions(struct file *file, char *buf, size_t size)
 664 {
 665         ssize_t rv;
 666
 667         mutex_lock(&nfsd_mutex);
 668         rv = __write_versions(file, buf, size);
 669         mutex_unlock(&nfsd_mutex);
 670         return rv;
 671 }
 672
 673 /*
 674  * Zero-length write.  Return a list of NFSD's current listener
 675  * transports.
 676  */
 677 static ssize_t __write_ports_names(char *buf, struct net *net)
 678 {
 679         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 680
 681         if (nn->nfsd_serv == NULL)
 682                 return 0;
 683         return svc_xprt_names(nn->nfsd_serv, buf, SIMPLE_TRANSACTION_LIMIT);
 684 }
 685
 686 /*
 687  * A single 'fd' number was written, in which case it must be for
 688  * a socket of a supported family/protocol, and we use it as an
 689  * nfsd listener.
 690  */
 691 static ssize_t __write_ports_addfd(char *buf, struct net *net, const struct cred *cred)
 692 {
 693         char *mesg = buf;
 694         int fd, err;
 695         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 696         struct svc_serv *serv;
 697
 698         err = get_int(&mesg, &fd);
 699         if (err != 0 || fd < 0)
 700                 return -EINVAL;
 701         trace_nfsd_ctl_ports_addfd(net, fd);
 702
 703         err = nfsd_create_serv(net);
 704         if (err != 0)
 705                 return err;
 706
 707         serv = nn->nfsd_serv;
 708         err = svc_addsock(serv, net, fd, buf, SIMPLE_TRANSACTION_LIMIT, cred);
 709
 710         if (err < 0 && !serv->sv_nrthreads && !nn->keep_active)
 711                 nfsd_last_thread(net);
 712         else if (err >= 0 && !serv->sv_nrthreads && !xchg(&nn->keep_active, 1))
 713                 svc_get(serv);
 714
 715         svc_put(serv);
 716         return err;
 717 }
 718
 719 /*
 720  * A transport listener is added by writing its transport name and
 721  * a port number.
 722  */
 723 static ssize_t __write_ports_addxprt(char *buf, struct net *net, const struct cred *cred)
 724 {
 725         char transport[16];
 726         struct svc_xprt *xprt;
 727         int port, err;
 728         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 729         struct svc_serv *serv;
 730
 731         if (sscanf(buf, "%15s %5u", transport, &port) != 2)
 732                 return -EINVAL;
 733
 734         if (port < 1 || port > USHRT_MAX)
 735                 return -EINVAL;
 736         trace_nfsd_ctl_ports_addxprt(net, transport, port);
 737
 738         err = nfsd_create_serv(net);
 739         if (err != 0)
 740                 return err;
 741
 742         serv = nn->nfsd_serv;
 743         err = svc_xprt_create(serv, transport, net,
 744                               PF_INET, port, SVC_SOCK_ANONYMOUS, cred);
 745         if (err < 0)
 746                 goto out_err;
 747
 748         err = svc_xprt_create(serv, transport, net,
 749                               PF_INET6, port, SVC_SOCK_ANONYMOUS, cred);
 750         if (err < 0 && err != -EAFNOSUPPORT)
 751                 goto out_close;
 752
 753         if (!serv->sv_nrthreads && !xchg(&nn->keep_active, 1))
 754                 svc_get(serv);
 755
 756         svc_put(serv);
 757         return 0;
 758 out_close:
 759         xprt = svc_find_xprt(serv, transport, net, PF_INET, port);
 760         if (xprt != NULL) {
 761                 svc_xprt_close(xprt);
 762                 svc_xprt_put(xprt);
 763         }
 764 out_err:
 765         if (!serv->sv_nrthreads && !nn->keep_active)
 766                 nfsd_last_thread(net);
 767
 768         svc_put(serv);
 769         return err;
 770 }
 771
 772 static ssize_t __write_ports(struct file *file, char *buf, size_t size,
 773                              struct net *net)
 774 {
 775         if (size == 0)
 776                 return __write_ports_names(buf, net);
 777
 778         if (isdigit(buf[0]))
 779                 return __write_ports_addfd(buf, net, file->f_cred);
 780
 781         if (isalpha(buf[0]))
 782                 return __write_ports_addxprt(buf, net, file->f_cred);
 783
 784         return -EINVAL;
 785 }
 786
 787 /*
 788  * write_ports - Pass a socket file descriptor or transport name to listen on
 789  *
 790  * Input:
 791  *                      buf:            ignored
 792  *                      size:           zero
 793  * Output:
 794  *      On success:     passed-in buffer filled with a '\n'-terminated C
 795  *                      string containing a whitespace-separated list of
 796  *                      named NFSD listeners;
 797  *                      return code is the size in bytes of the string
 798  *      On error:       return code is zero or a negative errno value
 799  *
 800  * OR
 801  *
 802  * Input:
 803  *                      buf:            C string containing an unsigned
 804  *                                      integer value representing a bound
 805  *                                      but unconnected socket that is to be
 806  *                                      used as an NFSD listener; listen(3)
 807  *                                      must be called for a SOCK_STREAM
 808  *                                      socket, otherwise it is ignored
 809  *                      size:           non-zero length of C string in @buf
 810  * Output:
 811  *      On success:     NFS service is started;
 812  *                      passed-in buffer filled with a '\n'-terminated C
 813  *                      string containing a unique alphanumeric name of
 814  *                      the listener;
 815  *                      return code is the size in bytes of the string
 816  *      On error:       return code is a negative errno value
 817  *
 818  * OR
 819  *
 820  * Input:
 821  *                      buf:            C string containing a transport
 822  *                                      name and an unsigned integer value
 823  *                                      representing the port to listen on,
 824  *                                      separated by whitespace
 825  *                      size:           non-zero length of C string in @buf
 826  * Output:
 827  *      On success:     returns zero; NFS service is started
 828  *      On error:       return code is a negative errno value
 829  */
 830 static ssize_t write_ports(struct file *file, char *buf, size_t size)
 831 {
 832         ssize_t rv;
 833
 834         mutex_lock(&nfsd_mutex);
 835         rv = __write_ports(file, buf, size, netns(file));
 836         mutex_unlock(&nfsd_mutex);
 837         return rv;
 838 }
 839
 840
 841 int nfsd_max_blksize;
 842
 843 /*
 844  * write_maxblksize - Set or report the current NFS blksize
 845  *
 846  * Input:
 847  *                      buf:            ignored
 848  *                      size:           zero
 849  *
 850  * OR
 851  *
 852  * Input:
 853  *                      buf:            C string containing an unsigned
 854  *                                      integer value representing the new
 855  *                                      NFS blksize
 856  *                      size:           non-zero length of C string in @buf
 857  * Output:
 858  *      On success:     passed-in buffer filled with '\n'-terminated C string
 859  *                      containing numeric value of the current NFS blksize
 860  *                      setting;
 861  *                      return code is the size in bytes of the string
 862  *      On error:       return code is zero or a negative errno value
 863  */
 864 static ssize_t write_maxblksize(struct file *file, char *buf, size_t size)
 865 {
 866         char *mesg = buf;
 867         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 868
 869         if (size > 0) {
 870                 int bsize;
 871                 int rv = get_int(&mesg, &bsize);
 872                 if (rv)
 873                         return rv;
 874                 trace_nfsd_ctl_maxblksize(netns(file), bsize);
 875
 876                 /* force bsize into allowed range and
 877                  * required alignment.
 878                  */
 879                 bsize = max_t(int, bsize, 1024);
 880                 bsize = min_t(int, bsize, NFSSVC_MAXBLKSIZE);
 881                 bsize &= ~(1024-1);
 882                 mutex_lock(&nfsd_mutex);
 883                 if (nn->nfsd_serv) {
 884                         mutex_unlock(&nfsd_mutex);
 885                         return -EBUSY;
 886                 }
 887                 nfsd_max_blksize = bsize;
 888                 mutex_unlock(&nfsd_mutex);
 889         }
 890
 891         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%d\n",
 892                                                         nfsd_max_blksize);
 893 }
 894
 895 /*
 896  * write_maxconn - Set or report the current max number of connections
 897  *
 898  * Input:
 899  *                      buf:            ignored
 900  *                      size:           zero
 901  * OR
 902  *
 903  * Input:
 904  *                      buf:            C string containing an unsigned
 905  *                                      integer value representing the new
 906  *                                      number of max connections
 907  *                      size:           non-zero length of C string in @buf
 908  * Output:
 909  *      On success:     passed-in buffer filled with '\n'-terminated C string
 910  *                      containing numeric value of max_connections setting
 911  *                      for this net namespace;
 912  *                      return code is the size in bytes of the string
 913  *      On error:       return code is zero or a negative errno value
 914  */
 915 static ssize_t write_maxconn(struct file *file, char *buf, size_t size)
 916 {
 917         char *mesg = buf;
 918         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
 919         unsigned int maxconn = nn->max_connections;
 920
 921         if (size > 0) {
 922                 int rv = get_uint(&mesg, &maxconn);
 923
 924                 if (rv)
 925                         return rv;
 926                 trace_nfsd_ctl_maxconn(netns(file), maxconn);
 927                 nn->max_connections = maxconn;
 928         }
 929
 930         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%u\n", maxconn);
 931 }
 932
 933 #ifdef CONFIG_NFSD_V4
 934 static ssize_t __nfsd4_write_time(struct file *file, char *buf, size_t size,
 935                                   time64_t *time, struct nfsd_net *nn)
 936 {
 937         struct dentry *dentry = file_dentry(file);
 938         char *mesg = buf;
 939         int rv, i;
 940
 941         if (size > 0) {
 942                 if (nn->nfsd_serv)
 943                         return -EBUSY;
 944                 rv = get_int(&mesg, &i);
 945                 if (rv)
 946                         return rv;
 947                 trace_nfsd_ctl_time(netns(file), dentry->d_name.name,
 948                                     dentry->d_name.len, i);
 949
 950                 /*
 951                  * Some sanity checking.  We don't have a reason for
 952                  * these particular numbers, but problems with the
 953                  * extremes are:
 954                  *      - Too short: the briefest network outage may
 955                  *        cause clients to lose all their locks.  Also,
 956                  *        the frequent polling may be wasteful.
 957                  *      - Too long: do you really want reboot recovery
 958                  *        to take more than an hour?  Or to make other
 959                  *        clients wait an hour before being able to
 960                  *        revoke a dead client's locks?
 961                  */
 962                 if (i < 10 || i > 3600)
 963                         return -EINVAL;
 964                 *time = i;
 965         }
 966
 967         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%lld\n", *time);
 968 }
 969
 970 static ssize_t nfsd4_write_time(struct file *file, char *buf, size_t size,
 971                                 time64_t *time, struct nfsd_net *nn)
 972 {
 973         ssize_t rv;
 974
 975         mutex_lock(&nfsd_mutex);
 976         rv = __nfsd4_write_time(file, buf, size, time, nn);
 977         mutex_unlock(&nfsd_mutex);
 978         return rv;
 979 }
 980
 981 /*
 982  * write_leasetime - Set or report the current NFSv4 lease time
 983  *
 984  * Input:
 985  *                      buf:            ignored
 986  *                      size:           zero
 987  *
 988  * OR
 989  *
 990  * Input:
 991  *                      buf:            C string containing an unsigned
 992  *                                      integer value representing the new
 993  *                                      NFSv4 lease expiry time
 994  *                      size:           non-zero length of C string in @buf
 995  * Output:
 996  *      On success:     passed-in buffer filled with '\n'-terminated C
 997  *                      string containing unsigned integer value of the
 998  *                      current lease expiry time;
 999  *                      return code is the size in bytes of the string
1000  *      On error:       return code is zero or a negative errno value
1001  */
1002 static ssize_t write_leasetime(struct file *file, char *buf, size_t size)
1003 {
1004         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1005         return nfsd4_write_time(file, buf, size, &nn->nfsd4_lease, nn);
1006 }
1007
1008 /*
1009  * write_gracetime - Set or report current NFSv4 grace period time
1010  *
1011  * As above, but sets the time of the NFSv4 grace period.
1012  *
1013  * Note this should never be set to less than the *previous*
1014  * lease-period time, but we don't try to enforce this.  (In the common
1015  * case (a new boot), we don't know what the previous lease time was
1016  * anyway.)
1017  */
1018 static ssize_t write_gracetime(struct file *file, char *buf, size_t size)
1019 {
1020         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1021         return nfsd4_write_time(file, buf, size, &nn->nfsd4_grace, nn);
1022 }
1023
1024 static ssize_t __write_recoverydir(struct file *file, char *buf, size_t size,
1025                                    struct nfsd_net *nn)
1026 {
1027         char *mesg = buf;
1028         char *recdir;
1029         int len, status;
1030
1031         if (size > 0) {
1032                 if (nn->nfsd_serv)
1033                         return -EBUSY;
1034                 if (size > PATH_MAX || buf[size-1] != '\n')
1035                         return -EINVAL;
1036                 buf[size-1] = 0;
1037
1038                 recdir = mesg;
1039                 len = qword_get(&mesg, recdir, size);
1040                 if (len <= 0)
1041                         return -EINVAL;
1042                 trace_nfsd_ctl_recoverydir(netns(file), recdir);
1043
1044                 status = nfs4_reset_recoverydir(recdir);
1045                 if (status)
1046                         return status;
1047         }
1048
1049         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%s\n",
1050                                                         nfs4_recoverydir());
1051 }
1052
1053 /*
1054  * write_recoverydir - Set or report the pathname of the recovery directory
1055  *
1056  * Input:
1057  *                      buf:            ignored
1058  *                      size:           zero
1059  *
1060  * OR
1061  *
1062  * Input:
1063  *                      buf:            C string containing the pathname
1064  *                                      of the directory on a local file
1065  *                                      system containing permanent NFSv4
1066  *                                      recovery data
1067  *                      size:           non-zero length of C string in @buf
1068  * Output:
1069  *      On success:     passed-in buffer filled with '\n'-terminated C string
1070  *                      containing the current recovery pathname setting;
1071  *                      return code is the size in bytes of the string
1072  *      On error:       return code is zero or a negative errno value
1073  */
1074 static ssize_t write_recoverydir(struct file *file, char *buf, size_t size)
1075 {
1076         ssize_t rv;
1077         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1078
1079         mutex_lock(&nfsd_mutex);
1080         rv = __write_recoverydir(file, buf, size, nn);
1081         mutex_unlock(&nfsd_mutex);
1082         return rv;
1083 }
1084
1085 /*
1086  * write_v4_end_grace - release grace period for nfsd's v4.x lock manager
1087  *
1088  * Input:
1089  *                      buf:            ignored
1090  *                      size:           zero
1091  * OR
1092  *
1093  * Input:
1094  *                      buf:            any value
1095  *                      size:           non-zero length of C string in @buf
1096  * Output:
1097  *                      passed-in buffer filled with "Y" or "N" with a newline
1098  *                      and NULL-terminated C string. This indicates whether
1099  *                      the grace period has ended in the current net
1100  *                      namespace. Return code is the size in bytes of the
1101  *                      string. Writing a string that starts with 'Y', 'y', or
1102  *                      '1' to the file will end the grace period for nfsd's v4
1103  *                      lock manager.
1104  */
1105 static ssize_t write_v4_end_grace(struct file *file, char *buf, size_t size)
1106 {
1107         struct nfsd_net *nn = net_generic(netns(file), nfsd_net_id);
1108
1109         if (size > 0) {
1110                 switch(buf[0]) {
1111                 case 'Y':
1112                 case 'y':
1113                 case '1':
1114                         if (!nn->nfsd_serv)
1115                                 return -EBUSY;
1116                         trace_nfsd_end_grace(netns(file));
1117                         nfsd4_end_grace(nn);
1118                         break;
1119                 default:
1120                         return -EINVAL;
1121                 }
1122         }
1123
1124         return scnprintf(buf, SIMPLE_TRANSACTION_LIMIT, "%c\n",
1125                          nn->grace_ended ? 'Y' : 'N');
1126 }
1127
1128 #endif
1129
1130 /*----------------------------------------------------------------------------*/
1131 /*
1132  *      populating the filesystem.
1133  */
1134
1135 /* Basically copying rpc_get_inode. */
1136 static struct inode *nfsd_get_inode(struct super_block *sb, umode_t mode)
1137 {
1138         struct inode *inode = new_inode(sb);
1139         if (!inode)
1140                 return NULL;
1141         /* Following advice from simple_fill_super documentation: */
1142         inode->i_ino = iunique(sb, NFSD_MaxReserved);
1143         inode->i_mode = mode;
1144         simple_inode_init_ts(inode);
1145         switch (mode & S_IFMT) {
1146         case S_IFDIR:
1147                 inode->i_fop = &simple_dir_operations;
1148                 inode->i_op = &simple_dir_inode_operations;
1149                 inc_nlink(inode);
1150                 break;
1151         case S_IFLNK:
1152                 inode->i_op = &simple_symlink_inode_operations;
1153                 break;
1154         default:
1155                 break;
1156         }
1157         return inode;
1158 }
1159
1160 static int __nfsd_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode, struct nfsdfs_client *ncl)
1161 {
1162         struct inode *inode;
1163
1164         inode = nfsd_get_inode(dir->i_sb, mode);
1165         if (!inode)
1166                 return -ENOMEM;
1167         if (ncl) {
1168                 inode->i_private = ncl;
1169                 kref_get(&ncl->cl_ref);
1170         }
1171         d_add(dentry, inode);
1172         inc_nlink(dir);
1173         fsnotify_mkdir(dir, dentry);
1174         return 0;
1175 }
1176
1177 static struct dentry *nfsd_mkdir(struct dentry *parent, struct nfsdfs_client *ncl, char *name)
1178 {
1179         struct inode *dir = parent->d_inode;
1180         struct dentry *dentry;
1181         int ret = -ENOMEM;
1182
1183         inode_lock(dir);
1184         dentry = d_alloc_name(parent, name);
1185         if (!dentry)
1186                 goto out_err;
1187         ret = __nfsd_mkdir(d_inode(parent), dentry, S_IFDIR | 0600, ncl);
1188         if (ret)
1189                 goto out_err;
1190 out:
1191         inode_unlock(dir);
1192         return dentry;
1193 out_err:
1194         dput(dentry);
1195         dentry = ERR_PTR(ret);
1196         goto out;
1197 }
1198
1199 #if IS_ENABLED(CONFIG_SUNRPC_GSS)
1200 static int __nfsd_symlink(struct inode *dir, struct dentry *dentry,
1201                           umode_t mode, const char *content)
1202 {
1203         struct inode *inode;
1204
1205         inode = nfsd_get_inode(dir->i_sb, mode);
1206         if (!inode)
1207                 return -ENOMEM;
1208
1209         inode->i_link = (char *)content;
1210         inode->i_size = strlen(content);
1211
1212         d_add(dentry, inode);
1213         inc_nlink(dir);
1214         fsnotify_create(dir, dentry);
1215         return 0;
1216 }
1217
1218 /*
1219  * @content is assumed to be a NUL-terminated string that lives
1220  * longer than the symlink itself.
1221  */
1222 static void _nfsd_symlink(struct dentry *parent, const char *name,
1223                           const char *content)
1224 {
1225         struct inode *dir = parent->d_inode;
1226         struct dentry *dentry;
1227         int ret;
1228
1229         inode_lock(dir);
1230         dentry = d_alloc_name(parent, name);
1231         if (!dentry)
1232                 goto out;
1233         ret = __nfsd_symlink(d_inode(parent), dentry, S_IFLNK | 0777, content);
1234         if (ret)
1235                 dput(dentry);
1236 out:
1237         inode_unlock(dir);
1238 }
1239 #else
1240 static inline void _nfsd_symlink(struct dentry *parent, const char *name,
1241                                  const char *content)
1242 {
1243 }
1244
1245 #endif
1246
1247 static void clear_ncl(struct inode *inode)
1248 {
1249         struct nfsdfs_client *ncl = inode->i_private;
1250
1251         inode->i_private = NULL;
1252         kref_put(&ncl->cl_ref, ncl->cl_release);
1253 }
1254
1255 static struct nfsdfs_client *__get_nfsdfs_client(struct inode *inode)
1256 {
1257         struct nfsdfs_client *nc = inode->i_private;
1258
1259         if (nc)
1260                 kref_get(&nc->cl_ref);
1261         return nc;
1262 }
1263
1264 struct nfsdfs_client *get_nfsdfs_client(struct inode *inode)
1265 {
1266         struct nfsdfs_client *nc;
1267
1268         inode_lock_shared(inode);
1269         nc = __get_nfsdfs_client(inode);
1270         inode_unlock_shared(inode);
1271         return nc;
1272 }
1273 /* from __rpc_unlink */
1274 static void nfsdfs_remove_file(struct inode *dir, struct dentry *dentry)
1275 {
1276         int ret;
1277
1278         clear_ncl(d_inode(dentry));
1279         dget(dentry);
1280         ret = simple_unlink(dir, dentry);
1281         d_drop(dentry);
1282         fsnotify_unlink(dir, dentry);
1283         dput(dentry);
1284         WARN_ON_ONCE(ret);
1285 }
1286
1287 static void nfsdfs_remove_files(struct dentry *root)
1288 {
1289         struct dentry *dentry, *tmp;
1290
1291         list_for_each_entry_safe(dentry, tmp, &root->d_subdirs, d_child) {
1292                 if (!simple_positive(dentry)) {
1293                         WARN_ON_ONCE(1); /* I think this can't happen? */
1294                         continue;
1295                 }
1296                 nfsdfs_remove_file(d_inode(root), dentry);
1297         }
1298 }
1299
1300 /* XXX: cut'n'paste from simple_fill_super; figure out if we could share
1301  * code instead. */
1302 static  int nfsdfs_create_files(struct dentry *root,
1303                                 const struct tree_descr *files,
1304                                 struct dentry **fdentries)
1305 {
1306         struct inode *dir = d_inode(root);
1307         struct inode *inode;
1308         struct dentry *dentry;
1309         int i;
1310
1311         inode_lock(dir);
1312         for (i = 0; files->name && files->name[0]; i++, files++) {
1313                 dentry = d_alloc_name(root, files->name);
1314                 if (!dentry)
1315                         goto out;
1316                 inode = nfsd_get_inode(d_inode(root)->i_sb,
1317                                         S_IFREG | files->mode);
1318                 if (!inode) {
1319                         dput(dentry);
1320                         goto out;
1321                 }
1322                 inode->i_fop = files->ops;
1323                 inode->i_private = __get_nfsdfs_client(dir);
1324                 d_add(dentry, inode);
1325                 fsnotify_create(dir, dentry);
1326                 if (fdentries)
1327                         fdentries[i] = dentry;
1328         }
1329         inode_unlock(dir);
1330         return 0;
1331 out:
1332         nfsdfs_remove_files(root);
1333         inode_unlock(dir);
1334         return -ENOMEM;
1335 }
1336
1337 /* on success, returns positive number unique to that client. */
1338 struct dentry *nfsd_client_mkdir(struct nfsd_net *nn,
1339                                  struct nfsdfs_client *ncl, u32 id,
1340                                  const struct tree_descr *files,
1341                                  struct dentry **fdentries)
1342 {
1343         struct dentry *dentry;
1344         char name[11];
1345         int ret;
1346
1347         sprintf(name, "%u", id);
1348
1349         dentry = nfsd_mkdir(nn->nfsd_client_dir, ncl, name);
1350         if (IS_ERR(dentry)) /* XXX: tossing errors? */
1351                 return NULL;
1352         ret = nfsdfs_create_files(dentry, files, fdentries);
1353         if (ret) {
1354                 nfsd_client_rmdir(dentry);
1355                 return NULL;
1356         }
1357         return dentry;
1358 }
1359
1360 /* Taken from __rpc_rmdir: */
1361 void nfsd_client_rmdir(struct dentry *dentry)
1362 {
1363         struct inode *dir = d_inode(dentry->d_parent);
1364         struct inode *inode = d_inode(dentry);
1365         int ret;
1366
1367         inode_lock(dir);
1368         nfsdfs_remove_files(dentry);
1369         clear_ncl(inode);
1370         dget(dentry);
1371         ret = simple_rmdir(dir, dentry);
1372         WARN_ON_ONCE(ret);
1373         d_drop(dentry);
1374         fsnotify_rmdir(dir, dentry);
1375         dput(dentry);
1376         inode_unlock(dir);
1377 }
1378
1379 static int nfsd_fill_super(struct super_block *sb, struct fs_context *fc)
1380 {
1381         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
1382                                                         nfsd_net_id);
1383         struct dentry *dentry;
1384         int ret;
1385
1386         static const struct tree_descr nfsd_files[] = {
1387                 [NFSD_List] = {"exports", &exports_nfsd_operations, S_IRUGO},
1388                 /* Per-export io stats use same ops as exports file */
1389                 [NFSD_Export_Stats] = {"export_stats", &exports_nfsd_operations, S_IRUGO},
1390                 [NFSD_Export_features] = {"export_features",
1391                                         &export_features_fops, S_IRUGO},
1392                 [NFSD_FO_UnlockIP] = {"unlock_ip",
1393                                         &transaction_ops, S_IWUSR|S_IRUSR},
1394                 [NFSD_FO_UnlockFS] = {"unlock_filesystem",
1395                                         &transaction_ops, S_IWUSR|S_IRUSR},
1396                 [NFSD_Fh] = {"filehandle", &transaction_ops, S_IWUSR|S_IRUSR},
1397                 [NFSD_Threads] = {"threads", &transaction_ops, S_IWUSR|S_IRUSR},
1398                 [NFSD_Pool_Threads] = {"pool_threads", &transaction_ops, S_IWUSR|S_IRUSR},
1399                 [NFSD_Pool_Stats] = {"pool_stats", &pool_stats_operations, S_IRUGO},
1400                 [NFSD_Reply_Cache_Stats] = {"reply_cache_stats",
1401                                         &nfsd_reply_cache_stats_fops, S_IRUGO},
1402                 [NFSD_Versions] = {"versions", &transaction_ops, S_IWUSR|S_IRUSR},
1403                 [NFSD_Ports] = {"portlist", &transaction_ops, S_IWUSR|S_IRUGO},
1404                 [NFSD_MaxBlkSize] = {"max_block_size", &transaction_ops, S_IWUSR|S_IRUGO},
1405                 [NFSD_MaxConnections] = {"max_connections", &transaction_ops, S_IWUSR|S_IRUGO},
1406                 [NFSD_Filecache] = {"filecache", &nfsd_file_cache_stats_fops, S_IRUGO},
1407 #ifdef CONFIG_NFSD_V4
1408                 [NFSD_Leasetime] = {"nfsv4leasetime", &transaction_ops, S_IWUSR|S_IRUSR},
1409                 [NFSD_Gracetime] = {"nfsv4gracetime", &transaction_ops, S_IWUSR|S_IRUSR},
1410                 [NFSD_RecoveryDir] = {"nfsv4recoverydir", &transaction_ops, S_IWUSR|S_IRUSR},
1411                 [NFSD_V4EndGrace] = {"v4_end_grace", &transaction_ops, S_IWUSR|S_IRUGO},
1412 #endif
1413                 /* last one */ {""}
1414         };
1415
1416         ret = simple_fill_super(sb, 0x6e667364, nfsd_files);
1417         if (ret)
1418                 return ret;
1419         _nfsd_symlink(sb->s_root, "supported_krb5_enctypes",
1420                       "/proc/net/rpc/gss_krb5_enctypes");
1421         dentry = nfsd_mkdir(sb->s_root, NULL, "clients");
1422         if (IS_ERR(dentry))
1423                 return PTR_ERR(dentry);
1424         nn->nfsd_client_dir = dentry;
1425         return 0;
1426 }
1427
1428 static int nfsd_fs_get_tree(struct fs_context *fc)
1429 {
1430         return get_tree_keyed(fc, nfsd_fill_super, get_net(fc->net_ns));
1431 }
1432
1433 static void nfsd_fs_free_fc(struct fs_context *fc)
1434 {
1435         if (fc->s_fs_info)
1436                 put_net(fc->s_fs_info);
1437 }
1438
1439 static const struct fs_context_operations nfsd_fs_context_ops = {
1440         .free           = nfsd_fs_free_fc,
1441         .get_tree       = nfsd_fs_get_tree,
1442 };
1443
1444 static int nfsd_init_fs_context(struct fs_context *fc)
1445 {
1446         put_user_ns(fc->user_ns);
1447         fc->user_ns = get_user_ns(fc->net_ns->user_ns);
1448         fc->ops = &nfsd_fs_context_ops;
1449         return 0;
1450 }
1451
1452 static void nfsd_umount(struct super_block *sb)
1453 {
1454         struct net *net = sb->s_fs_info;
1455
1456         nfsd_shutdown_threads(net);
1457
1458         kill_litter_super(sb);
1459         put_net(net);
1460 }
1461
1462 static struct file_system_type nfsd_fs_type = {
1463         .owner          = THIS_MODULE,
1464         .name           = "nfsd",
1465         .init_fs_context = nfsd_init_fs_context,
1466         .kill_sb        = nfsd_umount,
1467 };
1468 MODULE_ALIAS_FS("nfsd");
1469
1470 #ifdef CONFIG_PROC_FS
1471
1472 static int exports_proc_open(struct inode *inode, struct file *file)
1473 {
1474         return exports_net_open(current->nsproxy->net_ns, file);
1475 }
1476
1477 static const struct proc_ops exports_proc_ops = {
1478         .proc_open      = exports_proc_open,
1479         .proc_read      = seq_read,
1480         .proc_lseek     = seq_lseek,
1481         .proc_release   = seq_release,
1482 };
1483
1484 static int create_proc_exports_entry(void)
1485 {
1486         struct proc_dir_entry *entry;
1487
1488         entry = proc_mkdir("fs/nfs", NULL);
1489         if (!entry)
1490                 return -ENOMEM;
1491         entry = proc_create("exports", 0, entry, &exports_proc_ops);
1492         if (!entry) {
1493                 remove_proc_entry("fs/nfs", NULL);
1494                 return -ENOMEM;
1495         }
1496         return 0;
1497 }
1498 #else /* CONFIG_PROC_FS */
1499 static int create_proc_exports_entry(void)
1500 {
1501         return 0;
1502 }
1503 #endif
1504
1505 unsigned int nfsd_net_id;
1506
1507 /**
1508  * nfsd_nl_rpc_status_get_start - Prepare rpc_status_get dumpit
1509  * @cb: netlink metadata and command arguments
1510  *
1511  * Return values:
1512  *   %0: The rpc_status_get command may proceed
1513  *   %-ENODEV: There is no NFSD running in this namespace
1514  */
1515 int nfsd_nl_rpc_status_get_start(struct netlink_callback *cb)
1516 {
1517         struct nfsd_net *nn = net_generic(sock_net(cb->skb->sk), nfsd_net_id);
1518         int ret = -ENODEV;
1519
1520         mutex_lock(&nfsd_mutex);
1521         if (nn->nfsd_serv)
1522                 ret = 0;
1523         else
1524                 mutex_unlock(&nfsd_mutex);
1525
1526         return ret;
1527 }
1528
1529 static int nfsd_genl_rpc_status_compose_msg(struct sk_buff *skb,
1530                                             struct netlink_callback *cb,
1531                                             struct nfsd_genl_rqstp *rqstp)
1532 {
1533         void *hdr;
1534         u32 i;
1535
1536         hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1537                           &nfsd_nl_family, 0, NFSD_CMD_RPC_STATUS_GET);
1538         if (!hdr)
1539                 return -ENOBUFS;
1540
1541         if (nla_put_be32(skb, NFSD_A_RPC_STATUS_XID, rqstp->rq_xid) ||
1542             nla_put_u32(skb, NFSD_A_RPC_STATUS_FLAGS, rqstp->rq_flags) ||
1543             nla_put_u32(skb, NFSD_A_RPC_STATUS_PROG, rqstp->rq_prog) ||
1544             nla_put_u32(skb, NFSD_A_RPC_STATUS_PROC, rqstp->rq_proc) ||
1545             nla_put_u8(skb, NFSD_A_RPC_STATUS_VERSION, rqstp->rq_vers) ||
1546             nla_put_s64(skb, NFSD_A_RPC_STATUS_SERVICE_TIME,
1547                         ktime_to_us(rqstp->rq_stime),
1548                         NFSD_A_RPC_STATUS_PAD))
1549                 return -ENOBUFS;
1550
1551         switch (rqstp->rq_saddr.sa_family) {
1552         case AF_INET: {
1553                 const struct sockaddr_in *s_in, *d_in;
1554
1555                 s_in = (const struct sockaddr_in *)&rqstp->rq_saddr;
1556                 d_in = (const struct sockaddr_in *)&rqstp->rq_daddr;
1557                 if (nla_put_in_addr(skb, NFSD_A_RPC_STATUS_SADDR4,
1558                                     s_in->sin_addr.s_addr) ||
1559                     nla_put_in_addr(skb, NFSD_A_RPC_STATUS_DADDR4,
1560                                     d_in->sin_addr.s_addr) ||
1561                     nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1562                                  s_in->sin_port) ||
1563                     nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1564                                  d_in->sin_port))
1565                         return -ENOBUFS;
1566                 break;
1567         }
1568         case AF_INET6: {
1569                 const struct sockaddr_in6 *s_in, *d_in;
1570
1571                 s_in = (const struct sockaddr_in6 *)&rqstp->rq_saddr;
1572                 d_in = (const struct sockaddr_in6 *)&rqstp->rq_daddr;
1573                 if (nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_SADDR6,
1574                                      &s_in->sin6_addr) ||
1575                     nla_put_in6_addr(skb, NFSD_A_RPC_STATUS_DADDR6,
1576                                      &d_in->sin6_addr) ||
1577                     nla_put_be16(skb, NFSD_A_RPC_STATUS_SPORT,
1578                                  s_in->sin6_port) ||
1579                     nla_put_be16(skb, NFSD_A_RPC_STATUS_DPORT,
1580                                  d_in->sin6_port))
1581                         return -ENOBUFS;
1582                 break;
1583         }
1584         }
1585
1586         for (i = 0; i < rqstp->rq_opcnt; i++)
1587                 if (nla_put_u32(skb, NFSD_A_RPC_STATUS_COMPOUND_OPS,
1588                                 rqstp->rq_opnum[i]))
1589                         return -ENOBUFS;
1590
1591         genlmsg_end(skb, hdr);
1592         return 0;
1593 }
1594
1595 /**
1596  * nfsd_nl_rpc_status_get_dumpit - Handle rpc_status_get dumpit
1597  * @skb: reply buffer
1598  * @cb: netlink metadata and command arguments
1599  *
1600  * Returns the size of the reply or a negative errno.
1601  */
1602 int nfsd_nl_rpc_status_get_dumpit(struct sk_buff *skb,
1603                                   struct netlink_callback *cb)
1604 {
1605         struct nfsd_net *nn = net_generic(sock_net(skb->sk), nfsd_net_id);
1606         int i, ret, rqstp_index = 0;
1607
1608         rcu_read_lock();
1609
1610         for (i = 0; i < nn->nfsd_serv->sv_nrpools; i++) {
1611                 struct svc_rqst *rqstp;
1612
1613                 if (i < cb->args[0]) /* already consumed */
1614                         continue;
1615
1616                 rqstp_index = 0;
1617                 list_for_each_entry_rcu(rqstp,
1618                                 &nn->nfsd_serv->sv_pools[i].sp_all_threads,
1619                                 rq_all) {
1620                         struct nfsd_genl_rqstp genl_rqstp;
1621                         unsigned int status_counter;
1622
1623                         if (rqstp_index++ < cb->args[1]) /* already consumed */
1624                                 continue;
1625                         /*
1626                          * Acquire rq_status_counter before parsing the rqst
1627                          * fields. rq_status_counter is set to an odd value in
1628                          * order to notify the consumers the rqstp fields are
1629                          * meaningful.
1630                          */
1631                         status_counter =
1632                                 smp_load_acquire(&rqstp->rq_status_counter);
1633                         if (!(status_counter & 1))
1634                                 continue;
1635
1636                         genl_rqstp.rq_xid = rqstp->rq_xid;
1637                         genl_rqstp.rq_flags = rqstp->rq_flags;
1638                         genl_rqstp.rq_vers = rqstp->rq_vers;
1639                         genl_rqstp.rq_prog = rqstp->rq_prog;
1640                         genl_rqstp.rq_proc = rqstp->rq_proc;
1641                         genl_rqstp.rq_stime = rqstp->rq_stime;
1642                         genl_rqstp.rq_opcnt = 0;
1643                         memcpy(&genl_rqstp.rq_daddr, svc_daddr(rqstp),
1644                                sizeof(struct sockaddr));
1645                         memcpy(&genl_rqstp.rq_saddr, svc_addr(rqstp),
1646                                sizeof(struct sockaddr));
1647
1648 #ifdef CONFIG_NFSD_V4
1649                         if (rqstp->rq_vers == NFS4_VERSION &&
1650                             rqstp->rq_proc == NFSPROC4_COMPOUND) {
1651                                 /* NFSv4 compound */
1652                                 struct nfsd4_compoundargs *args;
1653                                 int j;
1654
1655                                 args = rqstp->rq_argp;
1656                                 genl_rqstp.rq_opcnt = args->opcnt;
1657                                 for (j = 0; j < genl_rqstp.rq_opcnt; j++)
1658                                         genl_rqstp.rq_opnum[j] =
1659                                                 args->ops[j].opnum;
1660                         }
1661 #endif /* CONFIG_NFSD_V4 */
1662
1663                         /*
1664                          * Acquire rq_status_counter before reporting the rqst
1665                          * fields to the user.
1666                          */
1667                         if (smp_load_acquire(&rqstp->rq_status_counter) !=
1668                             status_counter)
1669                                 continue;
1670
1671                         ret = nfsd_genl_rpc_status_compose_msg(skb, cb,
1672                                                                &genl_rqstp);
1673                         if (ret)
1674                                 goto out;
1675                 }
1676         }
1677
1678         cb->args[0] = i;
1679         cb->args[1] = rqstp_index;
1680         ret = skb->len;
1681 out:
1682         rcu_read_unlock();
1683
1684         return ret;
1685 }
1686
1687 /**
1688  * nfsd_nl_rpc_status_get_done - rpc_status_get dumpit post-processing
1689  * @cb: netlink metadata and command arguments
1690  *
1691  * Return values:
1692  *   %0: Success
1693  */
1694 int nfsd_nl_rpc_status_get_done(struct netlink_callback *cb)
1695 {
1696         mutex_unlock(&nfsd_mutex);
1697
1698         return 0;
1699 }
1700
1701 /**
1702  * nfsd_net_init - Prepare the nfsd_net portion of a new net namespace
1703  * @net: a freshly-created network namespace
1704  *
1705  * This information stays around as long as the network namespace is
1706  * alive whether or not there is an NFSD instance running in the
1707  * namespace.
1708  *
1709  * Returns zero on success, or a negative errno otherwise.
1710  */
1711 static __net_init int nfsd_net_init(struct net *net)
1712 {
1713         int retval;
1714         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1715
1716         retval = nfsd_export_init(net);
1717         if (retval)
1718                 goto out_export_error;
1719         retval = nfsd_idmap_init(net);
1720         if (retval)
1721                 goto out_idmap_error;
1722         retval = nfsd_net_reply_cache_init(nn);
1723         if (retval)
1724                 goto out_repcache_error;
1725         nn->nfsd_versions = NULL;
1726         nn->nfsd4_minorversions = NULL;
1727         nfsd4_init_leases_net(nn);
1728         get_random_bytes(&nn->siphash_key, sizeof(nn->siphash_key));
1729         seqlock_init(&nn->writeverf_lock);
1730
1731         return 0;
1732
1733 out_repcache_error:
1734         nfsd_idmap_shutdown(net);
1735 out_idmap_error:
1736         nfsd_export_shutdown(net);
1737 out_export_error:
1738         return retval;
1739 }
1740
1741 /**
1742  * nfsd_net_exit - Release the nfsd_net portion of a net namespace
1743  * @net: a network namespace that is about to be destroyed
1744  *
1745  */
1746 static __net_exit void nfsd_net_exit(struct net *net)
1747 {
1748         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1749
1750         nfsd_net_reply_cache_destroy(nn);
1751         nfsd_idmap_shutdown(net);
1752         nfsd_export_shutdown(net);
1753         nfsd_netns_free_versions(nn);
1754 }
1755
1756 static struct pernet_operations nfsd_net_ops = {
1757         .init = nfsd_net_init,
1758         .exit = nfsd_net_exit,
1759         .id   = &nfsd_net_id,
1760         .size = sizeof(struct nfsd_net),
1761 };
1762
1763 static int __init init_nfsd(void)
1764 {
1765         int retval;
1766
1767         retval = nfsd4_init_slabs();
1768         if (retval)
1769                 return retval;
1770         retval = nfsd4_init_pnfs();
1771         if (retval)
1772                 goto out_free_slabs;
1773         retval = nfsd_stat_init();      /* Statistics */
1774         if (retval)
1775                 goto out_free_pnfs;
1776         retval = nfsd_drc_slab_create();
1777         if (retval)
1778                 goto out_free_stat;
1779         nfsd_lockd_init();      /* lockd->nfsd callbacks */
1780         retval = create_proc_exports_entry();
1781         if (retval)
1782                 goto out_free_lockd;
1783         retval = register_pernet_subsys(&nfsd_net_ops);
1784         if (retval < 0)
1785                 goto out_free_exports;
1786         retval = register_cld_notifier();
1787         if (retval)
1788                 goto out_free_subsys;
1789         retval = nfsd4_create_laundry_wq();
1790         if (retval)
1791                 goto out_free_cld;
1792         retval = register_filesystem(&nfsd_fs_type);
1793         if (retval)
1794                 goto out_free_all;
1795         retval = genl_register_family(&nfsd_nl_family);
1796         if (retval)
1797                 goto out_free_all;
1798
1799         return 0;
1800 out_free_all:
1801         nfsd4_destroy_laundry_wq();
1802 out_free_cld:
1803         unregister_cld_notifier();
1804 out_free_subsys:
1805         unregister_pernet_subsys(&nfsd_net_ops);
1806 out_free_exports:
1807         remove_proc_entry("fs/nfs/exports", NULL);
1808         remove_proc_entry("fs/nfs", NULL);
1809 out_free_lockd:
1810         nfsd_lockd_shutdown();
1811         nfsd_drc_slab_free();
1812 out_free_stat:
1813         nfsd_stat_shutdown();
1814 out_free_pnfs:
1815         nfsd4_exit_pnfs();
1816 out_free_slabs:
1817         nfsd4_free_slabs();
1818         return retval;
1819 }
1820
1821 static void __exit exit_nfsd(void)
1822 {
1823         genl_unregister_family(&nfsd_nl_family);
1824         unregister_filesystem(&nfsd_fs_type);
1825         nfsd4_destroy_laundry_wq();
1826         unregister_cld_notifier();
1827         unregister_pernet_subsys(&nfsd_net_ops);
1828         nfsd_drc_slab_free();
1829         remove_proc_entry("fs/nfs/exports", NULL);
1830         remove_proc_entry("fs/nfs", NULL);
1831         nfsd_stat_shutdown();
1832         nfsd_lockd_shutdown();
1833         nfsd4_free_slabs();
1834         nfsd4_exit_pnfs();
1835 }
1836
1837 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1838 MODULE_DESCRIPTION("In-kernel NFS server");
1839 MODULE_LICENSE("GPL");
1840 module_init(init_nfsd)
1841 module_exit(exit_nfsd)