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/fdtable.h>
  13 #include <linux/capability.h>
  14 #include <linux/dnotify.h>
  15 #include <linux/slab.h>
  16 #include <linux/module.h>
  17 #include <linux/security.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/signal.h>
  20 #include <linux/rcupdate.h>
  21 #include <linux/pid_namespace.h>
  22 #include <linux/smp_lock.h>
  23
  24 #include <asm/poll.h>
  25 #include <asm/siginfo.h>
  26 #include <asm/uaccess.h>
  27
  28 void set_close_on_exec(unsigned int fd, int flag)
  29 {
  30         struct files_struct *files = current->files;
  31         struct fdtable *fdt;
  32         spin_lock(&files->file_lock);
  33         fdt = files_fdtable(files);
  34         if (flag)
  35                 FD_SET(fd, fdt->close_on_exec);
  36         else
  37                 FD_CLR(fd, fdt->close_on_exec);
  38         spin_unlock(&files->file_lock);
  39 }
  40
  41 static int get_close_on_exec(unsigned int fd)
  42 {
  43         struct files_struct *files = current->files;
  44         struct fdtable *fdt;
  45         int res;
  46         rcu_read_lock();
  47         fdt = files_fdtable(files);
  48         res = FD_ISSET(fd, fdt->close_on_exec);
  49         rcu_read_unlock();
  50         return res;
  51 }
  52
  53 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
  54 {
  55         int err = -EBADF;
  56         struct file * file, *tofree;
  57         struct files_struct * files = current->files;
  58         struct fdtable *fdt;
  59
  60         if ((flags & ~O_CLOEXEC) != 0)
  61                 return -EINVAL;
  62
  63         if (unlikely(oldfd == newfd))
  64                 return -EINVAL;
  65
  66         spin_lock(&files->file_lock);
  67         err = expand_files(files, newfd);
  68         file = fcheck(oldfd);
  69         if (unlikely(!file))
  70                 goto Ebadf;
  71         if (unlikely(err < 0)) {
  72                 if (err == -EMFILE)
  73                         goto Ebadf;
  74                 goto out_unlock;
  75         }
  76         /*
  77          * We need to detect attempts to do dup2() over allocated but still
  78          * not finished descriptor.  NB: OpenBSD avoids that at the price of
  79          * extra work in their equivalent of fget() - they insert struct
  80          * file immediately after grabbing descriptor, mark it larval if
  81          * more work (e.g. actual opening) is needed and make sure that
  82          * fget() treats larval files as absent.  Potentially interesting,
  83          * but while extra work in fget() is trivial, locking implications
  84          * and amount of surgery on open()-related paths in VFS are not.
  85          * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
  86          * deadlocks in rather amusing ways, AFAICS.  All of that is out of
  87          * scope of POSIX or SUS, since neither considers shared descriptor
  88          * tables and this condition does not arise without those.
  89          */
  90         err = -EBUSY;
  91         fdt = files_fdtable(files);
  92         tofree = fdt->fd[newfd];
  93         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
  94                 goto out_unlock;
  95         get_file(file);
  96         rcu_assign_pointer(fdt->fd[newfd], file);
  97         FD_SET(newfd, fdt->open_fds);
  98         if (flags & O_CLOEXEC)
  99                 FD_SET(newfd, fdt->close_on_exec);
 100         else
 101                 FD_CLR(newfd, fdt->close_on_exec);
 102         spin_unlock(&files->file_lock);
 103
 104         if (tofree)
 105                 filp_close(tofree, files);
 106
 107         return newfd;
 108
 109 Ebadf:
 110         err = -EBADF;
 111 out_unlock:
 112         spin_unlock(&files->file_lock);
 113         return err;
 114 }
 115
 116 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
 117 {
 118         if (unlikely(newfd == oldfd)) { /* corner case */
 119                 struct files_struct *files = current->files;
 120                 rcu_read_lock();
 121                 if (!fcheck_files(files, oldfd))
 122                         oldfd = -EBADF;
 123                 rcu_read_unlock();
 124                 return oldfd;
 125         }
 126         return sys_dup3(oldfd, newfd, 0);
 127 }
 128
 129 SYSCALL_DEFINE1(dup, unsigned int, fildes)
 130 {
 131         int ret = -EBADF;
 132         struct file *file = fget(fildes);
 133
 134         if (file) {
 135                 ret = get_unused_fd();
 136                 if (ret >= 0)
 137                         fd_install(ret, file);
 138                 else
 139                         fput(file);
 140         }
 141         return ret;
 142 }
 143
 144 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 145
 146 static int setfl(int fd, struct file * filp, unsigned long arg)
 147 {
 148         struct inode * inode = filp->f_path.dentry->d_inode;
 149         int error = 0;
 150
 151         /*
 152          * O_APPEND cannot be cleared if the file is marked as append-only
 153          * and the file is open for write.
 154          */
 155         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 156                 return -EPERM;
 157
 158         /* O_NOATIME can only be set by the owner or superuser */
 159         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 160                 if (!is_owner_or_cap(inode))
 161                         return -EPERM;
 162
 163         /* required for strict SunOS emulation */
 164         if (O_NONBLOCK != O_NDELAY)
 165                if (arg & O_NDELAY)
 166                    arg |= O_NONBLOCK;
 167
 168         if (arg & O_DIRECT) {
 169                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 170                         !filp->f_mapping->a_ops->direct_IO)
 171                                 return -EINVAL;
 172         }
 173
 174         if (filp->f_op && filp->f_op->check_flags)
 175                 error = filp->f_op->check_flags(arg);
 176         if (error)
 177                 return error;
 178
 179         /*
 180          * We still need a lock here for now to keep multiple FASYNC calls
 181          * from racing with each other.
 182          */
 183         lock_kernel();
 184         if ((arg ^ filp->f_flags) & FASYNC) {
 185                 if (filp->f_op && filp->f_op->fasync) {
 186                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 187                         if (error < 0)
 188                                 goto out;
 189                 }
 190         }
 191
 192         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 193  out:
 194         unlock_kernel();
 195         return error;
 196 }
 197
 198 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 199                      uid_t uid, uid_t euid, int force)
 200 {
 201         write_lock_irq(&filp->f_owner.lock);
 202         if (force || !filp->f_owner.pid) {
 203                 put_pid(filp->f_owner.pid);
 204                 filp->f_owner.pid = get_pid(pid);
 205                 filp->f_owner.pid_type = type;
 206                 filp->f_owner.uid = uid;
 207                 filp->f_owner.euid = euid;
 208         }
 209         write_unlock_irq(&filp->f_owner.lock);
 210 }
 211
 212 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 213                 int force)
 214 {
 215         int err;
 216
 217         err = security_file_set_fowner(filp);
 218         if (err)
 219                 return err;
 220
 221         f_modown(filp, pid, type, current->uid, current->euid, force);
 222         return 0;
 223 }
 224 EXPORT_SYMBOL(__f_setown);
 225
 226 int f_setown(struct file *filp, unsigned long arg, int force)
 227 {
 228         enum pid_type type;
 229         struct pid *pid;
 230         int who = arg;
 231         int result;
 232         type = PIDTYPE_PID;
 233         if (who < 0) {
 234                 type = PIDTYPE_PGID;
 235                 who = -who;
 236         }
 237         rcu_read_lock();
 238         pid = find_vpid(who);
 239         result = __f_setown(filp, pid, type, force);
 240         rcu_read_unlock();
 241         return result;
 242 }
 243 EXPORT_SYMBOL(f_setown);
 244
 245 void f_delown(struct file *filp)
 246 {
 247         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 248 }
 249
 250 pid_t f_getown(struct file *filp)
 251 {
 252         pid_t pid;
 253         read_lock(&filp->f_owner.lock);
 254         pid = pid_vnr(filp->f_owner.pid);
 255         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 256                 pid = -pid;
 257         read_unlock(&filp->f_owner.lock);
 258         return pid;
 259 }
 260
 261 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 262                 struct file *filp)
 263 {
 264         long err = -EINVAL;
 265
 266         switch (cmd) {
 267         case F_DUPFD:
 268         case F_DUPFD_CLOEXEC:
 269                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 270                         break;
 271                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 272                 if (err >= 0) {
 273                         get_file(filp);
 274                         fd_install(err, filp);
 275                 }
 276                 break;
 277         case F_GETFD:
 278                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 279                 break;
 280         case F_SETFD:
 281                 err = 0;
 282                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 283                 break;
 284         case F_GETFL:
 285                 err = filp->f_flags;
 286                 break;
 287         case F_SETFL:
 288                 err = setfl(fd, filp, arg);
 289                 break;
 290         case F_GETLK:
 291                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 292                 break;
 293         case F_SETLK:
 294         case F_SETLKW:
 295                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 296                 break;
 297         case F_GETOWN:
 298                 /*
 299                  * XXX If f_owner is a process group, the
 300                  * negative return value will get converted
 301                  * into an error.  Oops.  If we keep the
 302                  * current syscall conventions, the only way
 303                  * to fix this will be in libc.
 304                  */
 305                 err = f_getown(filp);
 306                 force_successful_syscall_return();
 307                 break;
 308         case F_SETOWN:
 309                 err = f_setown(filp, arg, 1);
 310                 break;
 311         case F_GETSIG:
 312                 err = filp->f_owner.signum;
 313                 break;
 314         case F_SETSIG:
 315                 /* arg == 0 restores default behaviour. */
 316                 if (!valid_signal(arg)) {
 317                         break;
 318                 }
 319                 err = 0;
 320                 filp->f_owner.signum = arg;
 321                 break;
 322         case F_GETLEASE:
 323                 err = fcntl_getlease(filp);
 324                 break;
 325         case F_SETLEASE:
 326                 err = fcntl_setlease(fd, filp, arg);
 327                 break;
 328         case F_NOTIFY:
 329                 err = fcntl_dirnotify(fd, filp, arg);
 330                 break;
 331         default:
 332                 break;
 333         }
 334         return err;
 335 }
 336
 337 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
 338 {
 339         struct file *filp;
 340         long err = -EBADF;
 341
 342         filp = fget(fd);
 343         if (!filp)
 344                 goto out;
 345
 346         err = security_file_fcntl(filp, cmd, arg);
 347         if (err) {
 348                 fput(filp);
 349                 return err;
 350         }
 351
 352         err = do_fcntl(fd, cmd, arg, filp);
 353
 354         fput(filp);
 355 out:
 356         return err;
 357 }
 358
 359 #if BITS_PER_LONG == 32
 360 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
 361                 unsigned long, arg)
 362 {
 363         struct file * filp;
 364         long err;
 365
 366         err = -EBADF;
 367         filp = fget(fd);
 368         if (!filp)
 369                 goto out;
 370
 371         err = security_file_fcntl(filp, cmd, arg);
 372         if (err) {
 373                 fput(filp);
 374                 return err;
 375         }
 376         err = -EBADF;
 377
 378         switch (cmd) {
 379                 case F_GETLK64:
 380                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 381                         break;
 382                 case F_SETLK64:
 383                 case F_SETLKW64:
 384                         err = fcntl_setlk64(fd, filp, cmd,
 385                                         (struct flock64 __user *) arg);
 386                         break;
 387                 default:
 388                         err = do_fcntl(fd, cmd, arg, filp);
 389                         break;
 390         }
 391         fput(filp);
 392 out:
 393         return err;
 394 }
 395 #endif
 396
 397 /* Table to convert sigio signal codes into poll band bitmaps */
 398
 399 static const long band_table[NSIGPOLL] = {
 400         POLLIN | POLLRDNORM,                    /* POLL_IN */
 401         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 402         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 403         POLLERR,                                /* POLL_ERR */
 404         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 405         POLLHUP | POLLERR                       /* POLL_HUP */
 406 };
 407
 408 static inline int sigio_perm(struct task_struct *p,
 409                              struct fown_struct *fown, int sig)
 410 {
 411         return (((fown->euid == 0) ||
 412                  (fown->euid == p->suid) || (fown->euid == p->uid) ||
 413                  (fown->uid == p->suid) || (fown->uid == p->uid)) &&
 414                 !security_file_send_sigiotask(p, fown, sig));
 415 }
 416
 417 static void send_sigio_to_task(struct task_struct *p,
 418                                struct fown_struct *fown,
 419                                int fd,
 420                                int reason)
 421 {
 422         if (!sigio_perm(p, fown, fown->signum))
 423                 return;
 424
 425         switch (fown->signum) {
 426                 siginfo_t si;
 427                 default:
 428                         /* Queue a rt signal with the appropriate fd as its
 429                            value.  We use SI_SIGIO as the source, not
 430                            SI_KERNEL, since kernel signals always get
 431                            delivered even if we can't queue.  Failure to
 432                            queue in this case _should_ be reported; we fall
 433                            back to SIGIO in that case. --sct */
 434                         si.si_signo = fown->signum;
 435                         si.si_errno = 0;
 436                         si.si_code  = reason;
 437                         /* Make sure we are called with one of the POLL_*
 438                            reasons, otherwise we could leak kernel stack into
 439                            userspace.  */
 440                         BUG_ON((reason & __SI_MASK) != __SI_POLL);
 441                         if (reason - POLL_IN >= NSIGPOLL)
 442                                 si.si_band  = ~0L;
 443                         else
 444                                 si.si_band = band_table[reason - POLL_IN];
 445                         si.si_fd    = fd;
 446                         if (!group_send_sig_info(fown->signum, &si, p))
 447                                 break;
 448                 /* fall-through: fall back on the old plain SIGIO signal */
 449                 case 0:
 450                         group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
 451         }
 452 }
 453
 454 void send_sigio(struct fown_struct *fown, int fd, int band)
 455 {
 456         struct task_struct *p;
 457         enum pid_type type;
 458         struct pid *pid;
 459
 460         read_lock(&fown->lock);
 461         type = fown->pid_type;
 462         pid = fown->pid;
 463         if (!pid)
 464                 goto out_unlock_fown;
 465
 466         read_lock(&tasklist_lock);
 467         do_each_pid_task(pid, type, p) {
 468                 send_sigio_to_task(p, fown, fd, band);
 469         } while_each_pid_task(pid, type, p);
 470         read_unlock(&tasklist_lock);
 471  out_unlock_fown:
 472         read_unlock(&fown->lock);
 473 }
 474
 475 static void send_sigurg_to_task(struct task_struct *p,
 476                                 struct fown_struct *fown)
 477 {
 478         if (sigio_perm(p, fown, SIGURG))
 479                 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
 480 }
 481
 482 int send_sigurg(struct fown_struct *fown)
 483 {
 484         struct task_struct *p;
 485         enum pid_type type;
 486         struct pid *pid;
 487         int ret = 0;
 488
 489         read_lock(&fown->lock);
 490         type = fown->pid_type;
 491         pid = fown->pid;
 492         if (!pid)
 493                 goto out_unlock_fown;
 494
 495         ret = 1;
 496
 497         read_lock(&tasklist_lock);
 498         do_each_pid_task(pid, type, p) {
 499                 send_sigurg_to_task(p, fown);
 500         } while_each_pid_task(pid, type, p);
 501         read_unlock(&tasklist_lock);
 502  out_unlock_fown:
 503         read_unlock(&fown->lock);
 504         return ret;
 505 }
 506
 507 static DEFINE_RWLOCK(fasync_lock);
 508 static struct kmem_cache *fasync_cache __read_mostly;
 509
 510 /*
 511  * fasync_helper() is used by some character device drivers (mainly mice)
 512  * to set up the fasync queue. It returns negative on error, 0 if it did
 513  * no changes and positive if it added/deleted the entry.
 514  */
 515 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
 516 {
 517         struct fasync_struct *fa, **fp;
 518         struct fasync_struct *new = NULL;
 519         int result = 0;
 520
 521         if (on) {
 522                 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
 523                 if (!new)
 524                         return -ENOMEM;
 525         }
 526         write_lock_irq(&fasync_lock);
 527         for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
 528                 if (fa->fa_file == filp) {
 529                         if(on) {
 530                                 fa->fa_fd = fd;
 531                                 kmem_cache_free(fasync_cache, new);
 532                         } else {
 533                                 *fp = fa->fa_next;
 534                                 kmem_cache_free(fasync_cache, fa);
 535                                 result = 1;
 536                         }
 537                         goto out;
 538                 }
 539         }
 540
 541         if (on) {
 542                 new->magic = FASYNC_MAGIC;
 543                 new->fa_file = filp;
 544                 new->fa_fd = fd;
 545                 new->fa_next = *fapp;
 546                 *fapp = new;
 547                 result = 1;
 548         }
 549 out:
 550         write_unlock_irq(&fasync_lock);
 551         return result;
 552 }
 553
 554 EXPORT_SYMBOL(fasync_helper);
 555
 556 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
 557 {
 558         while (fa) {
 559                 struct fown_struct * fown;
 560                 if (fa->magic != FASYNC_MAGIC) {
 561                         printk(KERN_ERR "kill_fasync: bad magic number in "
 562                                "fasync_struct!\n");
 563                         return;
 564                 }
 565                 fown = &fa->fa_file->f_owner;
 566                 /* Don't send SIGURG to processes which have not set a
 567                    queued signum: SIGURG has its own default signalling
 568                    mechanism. */
 569                 if (!(sig == SIGURG && fown->signum == 0))
 570                         send_sigio(fown, fa->fa_fd, band);
 571                 fa = fa->fa_next;
 572         }
 573 }
 574
 575 EXPORT_SYMBOL(__kill_fasync);
 576
 577 void kill_fasync(struct fasync_struct **fp, int sig, int band)
 578 {
 579         /* First a quick test without locking: usually
 580          * the list is empty.
 581          */
 582         if (*fp) {
 583                 read_lock(&fasync_lock);
 584                 /* reread *fp after obtaining the lock */
 585                 __kill_fasync(*fp, sig, band);
 586                 read_unlock(&fasync_lock);
 587         }
 588 }
 589 EXPORT_SYMBOL(kill_fasync);
 590
 591 static int __init fasync_init(void)
 592 {
 593         fasync_cache = kmem_cache_create("fasync_cache",
 594                 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
 595         return 0;
 596 }
 597
 598 module_init(fasync_init)