fs/fcntl.c

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