fs/autofs4/waitq.c

   1 /*
   2  * Copyright 1997-1998 Transmeta Corporation -- All Rights Reserved
   3  * Copyright 2001-2006 Ian Kent <raven@themaw.net>
   4  *
   5  * This file is part of the Linux kernel and is made available under
   6  * the terms of the GNU General Public License, version 2, or at your
   7  * option, any later version, incorporated herein by reference.
   8  */
   9
  10 #include <linux/slab.h>
  11 #include <linux/time.h>
  12 #include <linux/signal.h>
  13 #include <linux/file.h>
  14 #include "autofs_i.h"
  15
  16 /* We make this a static variable rather than a part of the superblock; it
  17  * is better if we don't reassign numbers easily even across filesystems
  18  */
  19 static autofs_wqt_t autofs4_next_wait_queue = 1;
  20
  21 /* These are the signals we allow interrupting a pending mount */
  22 #define SHUTDOWN_SIGS   (sigmask(SIGKILL) | sigmask(SIGINT) | sigmask(SIGQUIT))
  23
  24 void autofs4_catatonic_mode(struct autofs_sb_info *sbi)
  25 {
  26         struct autofs_wait_queue *wq, *nwq;
  27
  28         mutex_lock(&sbi->wq_mutex);
  29         if (sbi->catatonic) {
  30                 mutex_unlock(&sbi->wq_mutex);
  31                 return;
  32         }
  33
  34         pr_debug("entering catatonic mode\n");
  35
  36         sbi->catatonic = 1;
  37         wq = sbi->queues;
  38         sbi->queues = NULL;     /* Erase all wait queues */
  39         while (wq) {
  40                 nwq = wq->next;
  41                 wq->status = -ENOENT; /* Magic is gone - report failure */
  42                 kfree(wq->name.name);
  43                 wq->name.name = NULL;
  44                 wq->wait_ctr--;
  45                 wake_up_interruptible(&wq->queue);
  46                 wq = nwq;
  47         }
  48         fput(sbi->pipe);        /* Close the pipe */
  49         sbi->pipe = NULL;
  50         sbi->pipefd = -1;
  51         mutex_unlock(&sbi->wq_mutex);
  52 }
  53
  54 static int autofs4_write(struct autofs_sb_info *sbi,
  55                          struct file *file, const void *addr, int bytes)
  56 {
  57         unsigned long sigpipe, flags;
  58         mm_segment_t fs;
  59         const char *data = (const char *)addr;
  60         ssize_t wr = 0;
  61
  62         sigpipe = sigismember(&current->pending.signal, SIGPIPE);
  63
  64         /* Save pointer to user space and point back to kernel space */
  65         fs = get_fs();
  66         set_fs(KERNEL_DS);
  67
  68         mutex_lock(&sbi->pipe_mutex);
  69         while (bytes) {
  70                 wr = __vfs_write(file, data, bytes, &file->f_pos);
  71                 if (wr <= 0)
  72                         break;
  73                 data += wr;
  74                 bytes -= wr;
  75         }
  76         mutex_unlock(&sbi->pipe_mutex);
  77
  78         set_fs(fs);
  79
  80         /* Keep the currently executing process from receiving a
  81          * SIGPIPE unless it was already supposed to get one
  82          */
  83         if (wr == -EPIPE && !sigpipe) {
  84                 spin_lock_irqsave(&current->sighand->siglock, flags);
  85                 sigdelset(&current->pending.signal, SIGPIPE);
  86                 recalc_sigpending();
  87                 spin_unlock_irqrestore(&current->sighand->siglock, flags);
  88         }
  89
  90         /* if 'wr' returned 0 (impossible) we assume -EIO (safe) */
  91         return bytes == 0 ? 0 : wr < 0 ? wr : -EIO;
  92 }
  93
  94 static void autofs4_notify_daemon(struct autofs_sb_info *sbi,
  95                                  struct autofs_wait_queue *wq,
  96                                  int type)
  97 {
  98         union {
  99                 struct autofs_packet_hdr hdr;
 100                 union autofs_packet_union v4_pkt;
 101                 union autofs_v5_packet_union v5_pkt;
 102         } pkt;
 103         struct file *pipe = NULL;
 104         size_t pktsz;
 105         int ret;
 106
 107         pr_debug("wait id = 0x%08lx, name = %.*s, type=%d\n",
 108                  (unsigned long) wq->wait_queue_token,
 109                  wq->name.len, wq->name.name, type);
 110
 111         memset(&pkt, 0, sizeof(pkt)); /* For security reasons */
 112
 113         pkt.hdr.proto_version = sbi->version;
 114         pkt.hdr.type = type;
 115
 116         switch (type) {
 117         /* Kernel protocol v4 missing and expire packets */
 118         case autofs_ptype_missing:
 119         {
 120                 struct autofs_packet_missing *mp = &pkt.v4_pkt.missing;
 121
 122                 pktsz = sizeof(*mp);
 123
 124                 mp->wait_queue_token = wq->wait_queue_token;
 125                 mp->len = wq->name.len;
 126                 memcpy(mp->name, wq->name.name, wq->name.len);
 127                 mp->name[wq->name.len] = '\0';
 128                 break;
 129         }
 130         case autofs_ptype_expire_multi:
 131         {
 132                 struct autofs_packet_expire_multi *ep =
 133                                         &pkt.v4_pkt.expire_multi;
 134
 135                 pktsz = sizeof(*ep);
 136
 137                 ep->wait_queue_token = wq->wait_queue_token;
 138                 ep->len = wq->name.len;
 139                 memcpy(ep->name, wq->name.name, wq->name.len);
 140                 ep->name[wq->name.len] = '\0';
 141                 break;
 142         }
 143         /*
 144          * Kernel protocol v5 packet for handling indirect and direct
 145          * mount missing and expire requests
 146          */
 147         case autofs_ptype_missing_indirect:
 148         case autofs_ptype_expire_indirect:
 149         case autofs_ptype_missing_direct:
 150         case autofs_ptype_expire_direct:
 151         {
 152                 struct autofs_v5_packet *packet = &pkt.v5_pkt.v5_packet;
 153                 struct user_namespace *user_ns = sbi->pipe->f_cred->user_ns;
 154
 155                 pktsz = sizeof(*packet);
 156
 157                 packet->wait_queue_token = wq->wait_queue_token;
 158                 packet->len = wq->name.len;
 159                 memcpy(packet->name, wq->name.name, wq->name.len);
 160                 packet->name[wq->name.len] = '\0';
 161                 packet->dev = wq->dev;
 162                 packet->ino = wq->ino;
 163                 packet->uid = from_kuid_munged(user_ns, wq->uid);
 164                 packet->gid = from_kgid_munged(user_ns, wq->gid);
 165                 packet->pid = wq->pid;
 166                 packet->tgid = wq->tgid;
 167                 break;
 168         }
 169         default:
 170                 pr_warn("bad type %d!\n", type);
 171                 mutex_unlock(&sbi->wq_mutex);
 172                 return;
 173         }
 174
 175         pipe = get_file(sbi->pipe);
 176
 177         mutex_unlock(&sbi->wq_mutex);
 178
 179         switch (ret = autofs4_write(sbi, pipe, &pkt, pktsz)) {
 180         case 0:
 181                 break;
 182         case -ENOMEM:
 183         case -ERESTARTSYS:
 184                 /* Just fail this one */
 185                 autofs4_wait_release(sbi, wq->wait_queue_token, ret);
 186                 break;
 187         default:
 188                 autofs4_catatonic_mode(sbi);
 189                 break;
 190         }
 191         fput(pipe);
 192 }
 193
 194 static int autofs4_getpath(struct autofs_sb_info *sbi,
 195                            struct dentry *dentry, char **name)
 196 {
 197         struct dentry *root = sbi->sb->s_root;
 198         struct dentry *tmp;
 199         char *buf;
 200         char *p;
 201         int len;
 202         unsigned seq;
 203
 204 rename_retry:
 205         buf = *name;
 206         len = 0;
 207
 208         seq = read_seqbegin(&rename_lock);
 209         rcu_read_lock();
 210         spin_lock(&sbi->fs_lock);
 211         for (tmp = dentry ; tmp != root ; tmp = tmp->d_parent)
 212                 len += tmp->d_name.len + 1;
 213
 214         if (!len || --len > NAME_MAX) {
 215                 spin_unlock(&sbi->fs_lock);
 216                 rcu_read_unlock();
 217                 if (read_seqretry(&rename_lock, seq))
 218                         goto rename_retry;
 219                 return 0;
 220         }
 221
 222         *(buf + len) = '\0';
 223         p = buf + len - dentry->d_name.len;
 224         strncpy(p, dentry->d_name.name, dentry->d_name.len);
 225
 226         for (tmp = dentry->d_parent; tmp != root ; tmp = tmp->d_parent) {
 227                 *(--p) = '/';
 228                 p -= tmp->d_name.len;
 229                 strncpy(p, tmp->d_name.name, tmp->d_name.len);
 230         }
 231         spin_unlock(&sbi->fs_lock);
 232         rcu_read_unlock();
 233         if (read_seqretry(&rename_lock, seq))
 234                 goto rename_retry;
 235
 236         return len;
 237 }
 238
 239 static struct autofs_wait_queue *
 240 autofs4_find_wait(struct autofs_sb_info *sbi, const struct qstr *qstr)
 241 {
 242         struct autofs_wait_queue *wq;
 243
 244         for (wq = sbi->queues; wq; wq = wq->next) {
 245                 if (wq->name.hash == qstr->hash &&
 246                     wq->name.len == qstr->len &&
 247                     wq->name.name &&
 248                     !memcmp(wq->name.name, qstr->name, qstr->len))
 249                         break;
 250         }
 251         return wq;
 252 }
 253
 254 /*
 255  * Check if we have a valid request.
 256  * Returns
 257  * 1 if the request should continue.
 258  *   In this case we can return an autofs_wait_queue entry if one is
 259  *   found or NULL to idicate a new wait needs to be created.
 260  * 0 or a negative errno if the request shouldn't continue.
 261  */
 262 static int validate_request(struct autofs_wait_queue **wait,
 263                             struct autofs_sb_info *sbi,
 264                             const struct qstr *qstr,
 265                             struct dentry *dentry, enum autofs_notify notify)
 266 {
 267         struct autofs_wait_queue *wq;
 268         struct autofs_info *ino;
 269
 270         if (sbi->catatonic)
 271                 return -ENOENT;
 272
 273         /* Wait in progress, continue; */
 274         wq = autofs4_find_wait(sbi, qstr);
 275         if (wq) {
 276                 *wait = wq;
 277                 return 1;
 278         }
 279
 280         *wait = NULL;
 281
 282         /* If we don't yet have any info this is a new request */
 283         ino = autofs4_dentry_ino(dentry);
 284         if (!ino)
 285                 return 1;
 286
 287         /*
 288          * If we've been asked to wait on an existing expire (NFY_NONE)
 289          * but there is no wait in the queue ...
 290          */
 291         if (notify == NFY_NONE) {
 292                 /*
 293                  * Either we've betean the pending expire to post it's
 294                  * wait or it finished while we waited on the mutex.
 295                  * So we need to wait till either, the wait appears
 296                  * or the expire finishes.
 297                  */
 298
 299                 while (ino->flags & AUTOFS_INF_EXPIRING) {
 300                         mutex_unlock(&sbi->wq_mutex);
 301                         schedule_timeout_interruptible(HZ/10);
 302                         if (mutex_lock_interruptible(&sbi->wq_mutex))
 303                                 return -EINTR;
 304
 305                         if (sbi->catatonic)
 306                                 return -ENOENT;
 307
 308                         wq = autofs4_find_wait(sbi, qstr);
 309                         if (wq) {
 310                                 *wait = wq;
 311                                 return 1;
 312                         }
 313                 }
 314
 315                 /*
 316                  * Not ideal but the status has already gone. Of the two
 317                  * cases where we wait on NFY_NONE neither depend on the
 318                  * return status of the wait.
 319                  */
 320                 return 0;
 321         }
 322
 323         /*
 324          * If we've been asked to trigger a mount and the request
 325          * completed while we waited on the mutex ...
 326          */
 327         if (notify == NFY_MOUNT) {
 328                 struct dentry *new = NULL;
 329                 int valid = 1;
 330
 331                 /*
 332                  * If the dentry was successfully mounted while we slept
 333                  * on the wait queue mutex we can return success. If it
 334                  * isn't mounted (doesn't have submounts for the case of
 335                  * a multi-mount with no mount at it's base) we can
 336                  * continue on and create a new request.
 337                  */
 338                 if (!IS_ROOT(dentry)) {
 339                         if (d_unhashed(dentry) &&
 340                             d_really_is_positive(dentry)) {
 341                                 struct dentry *parent = dentry->d_parent;
 342
 343                                 new = d_lookup(parent, &dentry->d_name);
 344                                 if (new)
 345                                         dentry = new;
 346                         }
 347                 }
 348                 if (have_submounts(dentry))
 349                         valid = 0;
 350
 351                 if (new)
 352                         dput(new);
 353                 return valid;
 354         }
 355
 356         return 1;
 357 }
 358
 359 int autofs4_wait(struct autofs_sb_info *sbi,
 360                  struct dentry *dentry, enum autofs_notify notify)
 361 {
 362         struct autofs_wait_queue *wq;
 363         struct qstr qstr;
 364         char *name;
 365         int status, ret, type;
 366         pid_t pid;
 367         pid_t tgid;
 368
 369         /* In catatonic mode, we don't wait for nobody */
 370         if (sbi->catatonic)
 371                 return -ENOENT;
 372
 373         /*
 374          * Try translating pids to the namespace of the daemon.
 375          *
 376          * Zero means failure: we are in an unrelated pid namespace.
 377          */
 378         pid = task_pid_nr_ns(current, ns_of_pid(sbi->oz_pgrp));
 379         tgid = task_tgid_nr_ns(current, ns_of_pid(sbi->oz_pgrp));
 380         if (pid == 0 || tgid == 0)
 381                 return -ENOENT;
 382
 383         if (d_really_is_negative(dentry)) {
 384                 /*
 385                  * A wait for a negative dentry is invalid for certain
 386                  * cases. A direct or offset mount "always" has its mount
 387                  * point directory created and so the request dentry must
 388                  * be positive or the map key doesn't exist. The situation
 389                  * is very similar for indirect mounts except only dentrys
 390                  * in the root of the autofs file system may be negative.
 391                  */
 392                 if (autofs_type_trigger(sbi->type))
 393                         return -ENOENT;
 394                 else if (!IS_ROOT(dentry->d_parent))
 395                         return -ENOENT;
 396         }
 397
 398         name = kmalloc(NAME_MAX + 1, GFP_KERNEL);
 399         if (!name)
 400                 return -ENOMEM;
 401
 402         /* If this is a direct mount request create a dummy name */
 403         if (IS_ROOT(dentry) && autofs_type_trigger(sbi->type))
 404                 qstr.len = sprintf(name, "%p", dentry);
 405         else {
 406                 qstr.len = autofs4_getpath(sbi, dentry, &name);
 407                 if (!qstr.len) {
 408                         kfree(name);
 409                         return -ENOENT;
 410                 }
 411         }
 412         qstr.name = name;
 413         qstr.hash = full_name_hash(dentry, name, qstr.len);
 414
 415         if (mutex_lock_interruptible(&sbi->wq_mutex)) {
 416                 kfree(qstr.name);
 417                 return -EINTR;
 418         }
 419
 420         ret = validate_request(&wq, sbi, &qstr, dentry, notify);
 421         if (ret <= 0) {
 422                 if (ret != -EINTR)
 423                         mutex_unlock(&sbi->wq_mutex);
 424                 kfree(qstr.name);
 425                 return ret;
 426         }
 427
 428         if (!wq) {
 429                 /* Create a new wait queue */
 430                 wq = kmalloc(sizeof(struct autofs_wait_queue), GFP_KERNEL);
 431                 if (!wq) {
 432                         kfree(qstr.name);
 433                         mutex_unlock(&sbi->wq_mutex);
 434                         return -ENOMEM;
 435                 }
 436
 437                 wq->wait_queue_token = autofs4_next_wait_queue;
 438                 if (++autofs4_next_wait_queue == 0)
 439                         autofs4_next_wait_queue = 1;
 440                 wq->next = sbi->queues;
 441                 sbi->queues = wq;
 442                 init_waitqueue_head(&wq->queue);
 443                 memcpy(&wq->name, &qstr, sizeof(struct qstr));
 444                 wq->dev = autofs4_get_dev(sbi);
 445                 wq->ino = autofs4_get_ino(sbi);
 446                 wq->uid = current_cred()->uid;
 447                 wq->gid = current_cred()->gid;
 448                 wq->pid = pid;
 449                 wq->tgid = tgid;
 450                 wq->status = -EINTR; /* Status return if interrupted */
 451                 wq->wait_ctr = 2;
 452
 453                 if (sbi->version < 5) {
 454                         if (notify == NFY_MOUNT)
 455                                 type = autofs_ptype_missing;
 456                         else
 457                                 type = autofs_ptype_expire_multi;
 458                 } else {
 459                         if (notify == NFY_MOUNT)
 460                                 type = autofs_type_trigger(sbi->type) ?
 461                                         autofs_ptype_missing_direct :
 462                                          autofs_ptype_missing_indirect;
 463                         else
 464                                 type = autofs_type_trigger(sbi->type) ?
 465                                         autofs_ptype_expire_direct :
 466                                         autofs_ptype_expire_indirect;
 467                 }
 468
 469                 pr_debug("new wait id = 0x%08lx, name = %.*s, nfy=%d\n",
 470                          (unsigned long) wq->wait_queue_token, wq->name.len,
 471                          wq->name.name, notify);
 472
 473                 /*
 474                  * autofs4_notify_daemon() may block; it will unlock ->wq_mutex
 475                  */
 476                 autofs4_notify_daemon(sbi, wq, type);
 477         } else {
 478                 wq->wait_ctr++;
 479                 pr_debug("existing wait id = 0x%08lx, name = %.*s, nfy=%d\n",
 480                          (unsigned long) wq->wait_queue_token, wq->name.len,
 481                          wq->name.name, notify);
 482                 mutex_unlock(&sbi->wq_mutex);
 483                 kfree(qstr.name);
 484         }
 485
 486         /*
 487          * wq->name.name is NULL iff the lock is already released
 488          * or the mount has been made catatonic.
 489          */
 490         if (wq->name.name) {
 491                 /* Block all but "shutdown" signals while waiting */
 492                 unsigned long shutdown_sigs_mask;
 493                 unsigned long irqflags;
 494                 sigset_t oldset;
 495
 496                 spin_lock_irqsave(&current->sighand->siglock, irqflags);
 497                 oldset = current->blocked;
 498                 shutdown_sigs_mask = SHUTDOWN_SIGS & ~oldset.sig[0];
 499                 siginitsetinv(&current->blocked, shutdown_sigs_mask);
 500                 recalc_sigpending();
 501                 spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
 502
 503                 wait_event_interruptible(wq->queue, wq->name.name == NULL);
 504
 505                 spin_lock_irqsave(&current->sighand->siglock, irqflags);
 506                 current->blocked = oldset;
 507                 recalc_sigpending();
 508                 spin_unlock_irqrestore(&current->sighand->siglock, irqflags);
 509         } else {
 510                 pr_debug("skipped sleeping\n");
 511         }
 512
 513         status = wq->status;
 514
 515         /*
 516          * For direct and offset mounts we need to track the requester's
 517          * uid and gid in the dentry info struct. This is so it can be
 518          * supplied, on request, by the misc device ioctl interface.
 519          * This is needed during daemon resatart when reconnecting
 520          * to existing, active, autofs mounts. The uid and gid (and
 521          * related string values) may be used for macro substitution
 522          * in autofs mount maps.
 523          */
 524         if (!status) {
 525                 struct autofs_info *ino;
 526                 struct dentry *de = NULL;
 527
 528                 /* direct mount or browsable map */
 529                 ino = autofs4_dentry_ino(dentry);
 530                 if (!ino) {
 531                         /* If not lookup actual dentry used */
 532                         de = d_lookup(dentry->d_parent, &dentry->d_name);
 533                         if (de)
 534                                 ino = autofs4_dentry_ino(de);
 535                 }
 536
 537                 /* Set mount requester */
 538                 if (ino) {
 539                         spin_lock(&sbi->fs_lock);
 540                         ino->uid = wq->uid;
 541                         ino->gid = wq->gid;
 542                         spin_unlock(&sbi->fs_lock);
 543                 }
 544
 545                 if (de)
 546                         dput(de);
 547         }
 548
 549         /* Are we the last process to need status? */
 550         mutex_lock(&sbi->wq_mutex);
 551         if (!--wq->wait_ctr)
 552                 kfree(wq);
 553         mutex_unlock(&sbi->wq_mutex);
 554
 555         return status;
 556 }
 557
 558
 559 int autofs4_wait_release(struct autofs_sb_info *sbi, autofs_wqt_t wait_queue_token, int status)
 560 {
 561         struct autofs_wait_queue *wq, **wql;
 562
 563         mutex_lock(&sbi->wq_mutex);
 564         for (wql = &sbi->queues; (wq = *wql) != NULL; wql = &wq->next) {
 565                 if (wq->wait_queue_token == wait_queue_token)
 566                         break;
 567         }
 568
 569         if (!wq) {
 570                 mutex_unlock(&sbi->wq_mutex);
 571                 return -EINVAL;
 572         }
 573
 574         *wql = wq->next;        /* Unlink from chain */
 575         kfree(wq->name.name);
 576         wq->name.name = NULL;   /* Do not wait on this queue */
 577         wq->status = status;
 578         wake_up_interruptible(&wq->queue);
 579         if (!--wq->wait_ctr)
 580                 kfree(wq);
 581         mutex_unlock(&sbi->wq_mutex);
 582
 583         return 0;
 584 }