fs/fcntl.c

   1 /*
   2  *  linux/fs/fcntl.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/init.h>
   8 #include <linux/mm.h>
   9 #include <linux/fs.h>
  10 #include <linux/file.h>
  11 #include <linux/dnotify.h>
  12 #include <linux/smp_lock.h>
  13 #include <linux/slab.h>
  14 #include <linux/module.h>
  15 #include <linux/security.h>
  16 #include <linux/ptrace.h>
  17
  18 #include <asm/poll.h>
  19 #include <asm/siginfo.h>
  20 #include <asm/uaccess.h>
  21
  22 void fastcall set_close_on_exec(unsigned int fd, int flag)
  23 {
  24         struct files_struct *files = current->files;
  25         spin_lock(&files->file_lock);
  26         if (flag)
  27                 FD_SET(fd, files->close_on_exec);
  28         else
  29                 FD_CLR(fd, files->close_on_exec);
  30         spin_unlock(&files->file_lock);
  31 }
  32
  33 static inline int get_close_on_exec(unsigned int fd)
  34 {
  35         struct files_struct *files = current->files;
  36         int res;
  37         spin_lock(&files->file_lock);
  38         res = FD_ISSET(fd, files->close_on_exec);
  39         spin_unlock(&files->file_lock);
  40         return res;
  41 }
  42
  43
  44 /* Expand files.  Return <0 on error; 0 nothing done; 1 files expanded,
  45  * we may have blocked.
  46  *
  47  * Should be called with the files->file_lock spinlock held for write.
  48  */
  49 static int expand_files(struct files_struct *files, int nr)
  50 {
  51         int err, expand = 0;
  52 #ifdef FDSET_DEBUG
  53         printk (KERN_ERR "%s %d: nr = %d\n", __FUNCTION__, current->pid, nr);
  54 #endif
  55
  56         if (nr >= files->max_fdset) {
  57                 expand = 1;
  58                 if ((err = expand_fdset(files, nr)))
  59                         goto out;
  60         }
  61         if (nr >= files->max_fds) {
  62                 expand = 1;
  63                 if ((err = expand_fd_array(files, nr)))
  64                         goto out;
  65         }
  66         err = expand;
  67  out:
  68 #ifdef FDSET_DEBUG
  69         if (err)
  70                 printk (KERN_ERR "%s %d: return %d\n", __FUNCTION__, current->pid, err);
  71 #endif
  72         return err;
  73 }
  74
  75 /*
  76  * locate_fd finds a free file descriptor in the open_fds fdset,
  77  * expanding the fd arrays if necessary.  Must be called with the
  78  * file_lock held for write.
  79  */
  80
  81 static int locate_fd(struct files_struct *files,
  82                             struct file *file, unsigned int orig_start)
  83 {
  84         unsigned int newfd;
  85         unsigned int start;
  86         int error;
  87
  88         error = -EINVAL;
  89         if (orig_start >= current->rlim[RLIMIT_NOFILE].rlim_cur)
  90                 goto out;
  91
  92 repeat:
  93         /*
  94          * Someone might have closed fd's in the range
  95          * orig_start..files->next_fd
  96          */
  97         start = orig_start;
  98         if (start < files->next_fd)
  99                 start = files->next_fd;
 100
 101         newfd = start;
 102         if (start < files->max_fdset) {
 103                 newfd = find_next_zero_bit(files->open_fds->fds_bits,
 104                         files->max_fdset, start);
 105         }
 106
 107         error = -EMFILE;
 108         if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 109                 goto out;
 110
 111         error = expand_files(files, newfd);
 112         if (error < 0)
 113                 goto out;
 114
 115         /*
 116          * If we needed to expand the fs array we
 117          * might have blocked - try again.
 118          */
 119         if (error)
 120                 goto repeat;
 121
 122         if (start <= files->next_fd)
 123                 files->next_fd = newfd + 1;
 124
 125         error = newfd;
 126
 127 out:
 128         return error;
 129 }
 130
 131 static int dupfd(struct file *file, unsigned int start)
 132 {
 133         struct files_struct * files = current->files;
 134         int fd;
 135
 136         spin_lock(&files->file_lock);
 137         fd = locate_fd(files, file, start);
 138         if (fd >= 0) {
 139                 FD_SET(fd, files->open_fds);
 140                 FD_CLR(fd, files->close_on_exec);
 141                 spin_unlock(&files->file_lock);
 142                 fd_install(fd, file);
 143         } else {
 144                 spin_unlock(&files->file_lock);
 145                 fput(file);
 146         }
 147
 148         return fd;
 149 }
 150
 151 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 152 {
 153         int err = -EBADF;
 154         struct file * file, *tofree;
 155         struct files_struct * files = current->files;
 156
 157         spin_lock(&files->file_lock);
 158         if (!(file = fcheck(oldfd)))
 159                 goto out_unlock;
 160         err = newfd;
 161         if (newfd == oldfd)
 162                 goto out_unlock;
 163         err = -EBADF;
 164         if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
 165                 goto out_unlock;
 166         get_file(file);                 /* We are now finished with oldfd */
 167
 168         err = expand_files(files, newfd);
 169         if (err < 0)
 170                 goto out_fput;
 171
 172         /* To avoid races with open() and dup(), we will mark the fd as
 173          * in-use in the open-file bitmap throughout the entire dup2()
 174          * process.  This is quite safe: do_close() uses the fd array
 175          * entry, not the bitmap, to decide what work needs to be
 176          * done.  --sct */
 177         /* Doesn't work. open() might be there first. --AV */
 178
 179         /* Yes. It's a race. In user space. Nothing sane to do */
 180         err = -EBUSY;
 181         tofree = files->fd[newfd];
 182         if (!tofree && FD_ISSET(newfd, files->open_fds))
 183                 goto out_fput;
 184
 185         files->fd[newfd] = file;
 186         FD_SET(newfd, files->open_fds);
 187         FD_CLR(newfd, files->close_on_exec);
 188         spin_unlock(&files->file_lock);
 189
 190         if (tofree)
 191                 filp_close(tofree, files);
 192         err = newfd;
 193 out:
 194         return err;
 195 out_unlock:
 196         spin_unlock(&files->file_lock);
 197         goto out;
 198
 199 out_fput:
 200         spin_unlock(&files->file_lock);
 201         fput(file);
 202         goto out;
 203 }
 204
 205 asmlinkage long sys_dup(unsigned int fildes)
 206 {
 207         int ret = -EBADF;
 208         struct file * file = fget(fildes);
 209
 210         if (file)
 211                 ret = dupfd(file, 0);
 212         return ret;
 213 }
 214
 215 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 216
 217 static int setfl(int fd, struct file * filp, unsigned long arg)
 218 {
 219         struct inode * inode = filp->f_dentry->d_inode;
 220         int error = 0;
 221
 222         /* O_APPEND cannot be cleared if the file is marked as append-only */
 223         if (!(arg & O_APPEND) && IS_APPEND(inode))
 224                 return -EPERM;
 225
 226         /* O_NOATIME can only be set by the owner or superuser */
 227         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 228                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
 229                         return -EPERM;
 230
 231         /* required for strict SunOS emulation */
 232         if (O_NONBLOCK != O_NDELAY)
 233                if (arg & O_NDELAY)
 234                    arg |= O_NONBLOCK;
 235
 236         if (arg & O_DIRECT) {
 237                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 238                         !filp->f_mapping->a_ops->direct_IO)
 239                                 return -EINVAL;
 240         }
 241
 242         if (filp->f_op && filp->f_op->check_flags)
 243                 error = filp->f_op->check_flags(arg);
 244         if (error)
 245                 return error;
 246
 247         lock_kernel();
 248         if ((arg ^ filp->f_flags) & FASYNC) {
 249                 if (filp->f_op && filp->f_op->fasync) {
 250                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 251                         if (error < 0)
 252                                 goto out;
 253                 }
 254         }
 255
 256         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 257  out:
 258         unlock_kernel();
 259         return error;
 260 }
 261
 262 static void f_modown(struct file *filp, unsigned long pid,
 263                      uid_t uid, uid_t euid, int force)
 264 {
 265         write_lock_irq(&filp->f_owner.lock);
 266         if (force || !filp->f_owner.pid) {
 267                 filp->f_owner.pid = pid;
 268                 filp->f_owner.uid = uid;
 269                 filp->f_owner.euid = euid;
 270         }
 271         write_unlock_irq(&filp->f_owner.lock);
 272 }
 273
 274 int f_setown(struct file *filp, unsigned long arg, int force)
 275 {
 276         int err;
 277
 278         err = security_file_set_fowner(filp);
 279         if (err)
 280                 return err;
 281
 282         f_modown(filp, arg, current->uid, current->euid, force);
 283         return 0;
 284 }
 285
 286 EXPORT_SYMBOL(f_setown);
 287
 288 void f_delown(struct file *filp)
 289 {
 290         f_modown(filp, 0, 0, 0, 1);
 291 }
 292
 293 EXPORT_SYMBOL(f_delown);
 294
 295 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 296                 struct file *filp)
 297 {
 298         long err = -EINVAL;
 299
 300         switch (cmd) {
 301         case F_DUPFD:
 302                 get_file(filp);
 303                 err = dupfd(filp, arg);
 304                 break;
 305         case F_GETFD:
 306                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 307                 break;
 308         case F_SETFD:
 309                 err = 0;
 310                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 311                 break;
 312         case F_GETFL:
 313                 err = filp->f_flags;
 314                 break;
 315         case F_SETFL:
 316                 err = setfl(fd, filp, arg);
 317                 break;
 318         case F_GETLK:
 319                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 320                 break;
 321         case F_SETLK:
 322         case F_SETLKW:
 323                 err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
 324                 break;
 325         case F_GETOWN:
 326                 /*
 327                  * XXX If f_owner is a process group, the
 328                  * negative return value will get converted
 329                  * into an error.  Oops.  If we keep the
 330                  * current syscall conventions, the only way
 331                  * to fix this will be in libc.
 332                  */
 333                 err = filp->f_owner.pid;
 334                 force_successful_syscall_return();
 335                 break;
 336         case F_SETOWN:
 337                 err = f_setown(filp, arg, 1);
 338                 break;
 339         case F_GETSIG:
 340                 err = filp->f_owner.signum;
 341                 break;
 342         case F_SETSIG:
 343                 /* arg == 0 restores default behaviour. */
 344                 if (arg < 0 || arg > _NSIG) {
 345                         break;
 346                 }
 347                 err = 0;
 348                 filp->f_owner.signum = arg;
 349                 break;
 350         case F_GETLEASE:
 351                 err = fcntl_getlease(filp);
 352                 break;
 353         case F_SETLEASE:
 354                 err = fcntl_setlease(fd, filp, arg);
 355                 break;
 356         case F_NOTIFY:
 357                 err = fcntl_dirnotify(fd, filp, arg);
 358                 break;
 359         default:
 360                 break;
 361         }
 362         return err;
 363 }
 364
 365 asmlinkage long sys_fcntl(int fd, unsigned int cmd, unsigned long arg)
 366 {
 367         struct file *filp;
 368         long err = -EBADF;
 369
 370         filp = fget(fd);
 371         if (!filp)
 372                 goto out;
 373
 374         err = security_file_fcntl(filp, cmd, arg);
 375         if (err) {
 376                 fput(filp);
 377                 return err;
 378         }
 379
 380         err = do_fcntl(fd, cmd, arg, filp);
 381
 382         fput(filp);
 383 out:
 384         return err;
 385 }
 386
 387 #if BITS_PER_LONG == 32
 388 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 389 {
 390         struct file * filp;
 391         long err;
 392
 393         err = -EBADF;
 394         filp = fget(fd);
 395         if (!filp)
 396                 goto out;
 397
 398         err = security_file_fcntl(filp, cmd, arg);
 399         if (err) {
 400                 fput(filp);
 401                 return err;
 402         }
 403         err = -EBADF;
 404
 405         switch (cmd) {
 406                 case F_GETLK64:
 407                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 408                         break;
 409                 case F_SETLK64:
 410                 case F_SETLKW64:
 411                         err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
 412                         break;
 413                 default:
 414                         err = do_fcntl(fd, cmd, arg, filp);
 415                         break;
 416         }
 417         fput(filp);
 418 out:
 419         return err;
 420 }
 421 #endif
 422
 423 /* Table to convert sigio signal codes into poll band bitmaps */
 424
 425 static long band_table[NSIGPOLL] = {
 426         POLLIN | POLLRDNORM,                    /* POLL_IN */
 427         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 428         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 429         POLLERR,                                /* POLL_ERR */
 430         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 431         POLLHUP | POLLERR                       /* POLL_HUP */
 432 };
 433
 434 static inline int sigio_perm(struct task_struct *p,
 435                              struct fown_struct *fown)
 436 {
 437         return ((fown->euid == 0) ||
 438                 (fown->euid == p->suid) || (fown->euid == p->uid) ||
 439                 (fown->uid == p->suid) || (fown->uid == p->uid));
 440 }
 441
 442 static void send_sigio_to_task(struct task_struct *p,
 443                                struct fown_struct *fown,
 444                                int fd,
 445                                int reason)
 446 {
 447         if (!sigio_perm(p, fown))
 448                 return;
 449
 450         if (security_file_send_sigiotask(p, fown, fd, reason))
 451                 return;
 452
 453         switch (fown->signum) {
 454                 siginfo_t si;
 455                 default:
 456                         /* Queue a rt signal with the appropriate fd as its
 457                            value.  We use SI_SIGIO as the source, not
 458                            SI_KERNEL, since kernel signals always get
 459                            delivered even if we can't queue.  Failure to
 460                            queue in this case _should_ be reported; we fall
 461                            back to SIGIO in that case. --sct */
 462                         si.si_signo = fown->signum;
 463                         si.si_errno = 0;
 464                         si.si_code  = reason;
 465                         /* Make sure we are called with one of the POLL_*
 466                            reasons, otherwise we could leak kernel stack into
 467                            userspace.  */
 468                         if ((reason & __SI_MASK) != __SI_POLL)
 469                                 BUG();
 470                         if (reason - POLL_IN >= NSIGPOLL)
 471                                 si.si_band  = ~0L;
 472                         else
 473                                 si.si_band = band_table[reason - POLL_IN];
 474                         si.si_fd    = fd;
 475                         if (!send_sig_info(fown->signum, &si, p))
 476                                 break;
 477                 /* fall-through: fall back on the old plain SIGIO signal */
 478                 case 0:
 479                         send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
 480         }
 481 }
 482
 483 void send_sigio(struct fown_struct *fown, int fd, int band)
 484 {
 485         struct task_struct *p;
 486         int pid;
 487
 488         read_lock(&fown->lock);
 489         pid = fown->pid;
 490         if (!pid)
 491                 goto out_unlock_fown;
 492
 493         read_lock(&tasklist_lock);
 494         if (pid > 0) {
 495                 p = find_task_by_pid(pid);
 496                 if (p) {
 497                         send_sigio_to_task(p, fown, fd, band);
 498                 }
 499         } else {
 500                 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
 501                         send_sigio_to_task(p, fown, fd, band);
 502                 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
 503         }
 504         read_unlock(&tasklist_lock);
 505  out_unlock_fown:
 506         read_unlock(&fown->lock);
 507 }
 508
 509 static void send_sigurg_to_task(struct task_struct *p,
 510                                 struct fown_struct *fown)
 511 {
 512         if (sigio_perm(p, fown))
 513                 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
 514 }
 515
 516 int send_sigurg(struct fown_struct *fown)
 517 {
 518         struct task_struct *p;
 519         int pid, ret = 0;
 520
 521         read_lock(&fown->lock);
 522         pid = fown->pid;
 523         if (!pid)
 524                 goto out_unlock_fown;
 525
 526         ret = 1;
 527
 528         read_lock(&tasklist_lock);
 529         if (pid > 0) {
 530                 p = find_task_by_pid(pid);
 531                 if (p) {
 532                         send_sigurg_to_task(p, fown);
 533                 }
 534         } else {
 535                 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
 536                         send_sigurg_to_task(p, fown);
 537                 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
 538         }
 539         read_unlock(&tasklist_lock);
 540  out_unlock_fown:
 541         read_unlock(&fown->lock);
 542         return ret;
 543 }
 544
 545 static rwlock_t fasync_lock = RW_LOCK_UNLOCKED;
 546 static kmem_cache_t *fasync_cache;
 547
 548 /*
 549  * fasync_helper() is used by some character device drivers (mainly mice)
 550  * to set up the fasync queue. It returns negative on error, 0 if it did
 551  * no changes and positive if it added/deleted the entry.
 552  */
 553 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 554 {
 555         struct fasync_struct *fa, **fp;
 556         struct fasync_struct *new = NULL;
 557         int result = 0;
 558
 559         if (on) {
 560                 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
 561                 if (!new)
 562                         return -ENOMEM;
 563         }
 564         write_lock_irq(&fasync_lock);
 565         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 566                 if (fa->fa_file == filp) {
 567                         if(on) {
 568                                 fa->fa_fd = fd;
 569                                 kmem_cache_free(fasync_cache, new);
 570                         } else {
 571                                 *fp = fa->fa_next;
 572                                 kmem_cache_free(fasync_cache, fa);
 573                                 result = 1;
 574                         }
 575                         goto out;
 576                 }
 577         }
 578
 579         if (on) {
 580                 new->magic = FASYNC_MAGIC;
 581                 new->fa_file = filp;
 582                 new->fa_fd = fd;
 583                 new->fa_next = *fapp;
 584                 *fapp = new;
 585                 result = 1;
 586         }
 587 out:
 588         write_unlock_irq(&fasync_lock);
 589         return result;
 590 }
 591
 592 EXPORT_SYMBOL(fasync_helper);
 593
 594 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 595 {
 596         while (fa) {
 597                 struct fown_struct * fown;
 598                 if (fa->magic != FASYNC_MAGIC) {
 599                         printk(KERN_ERR "kill_fasync: bad magic number in "
 600                                "fasync_struct!\n");
 601                         return;
 602                 }
 603                 fown = &fa->fa_file->f_owner;
 604                 /* Don't send SIGURG to processes which have not set a
 605                    queued signum: SIGURG has its own default signalling
 606                    mechanism. */
 607                 if (!(sig == SIGURG && fown->signum == 0))
 608                         send_sigio(fown, fa->fa_fd, band);
 609                 fa = fa->fa_next;
 610         }
 611 }
 612
 613 EXPORT_SYMBOL(__kill_fasync);
 614
 615 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 616 {
 617         /* First a quick test without locking: usually
 618          * the list is empty.
 619          */
 620         if (*fp) {
 621                 read_lock(&fasync_lock);
 622                 /* reread *fp after obtaining the lock */
 623                 __kill_fasync(*fp, sig, band);
 624                 read_unlock(&fasync_lock);
 625         }
 626 }
 627 EXPORT_SYMBOL(kill_fasync);
 628
 629 static int __init fasync_init(void)
 630 {
 631         fasync_cache = kmem_cache_create("fasync_cache",
 632                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
 633         return 0;
 634 }
 635
 636 module_init(fasync_init)