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
  23 #include <asm/poll.h>
  24 #include <asm/siginfo.h>
  25 #include <asm/uaccess.h>
  26
  27 void set_close_on_exec(unsigned int fd, int flag)
  28 {
  29         struct files_struct *files = current->files;
  30         struct fdtable *fdt;
  31         spin_lock(&files->file_lock);
  32         fdt = files_fdtable(files);
  33         if (flag)
  34                 FD_SET(fd, fdt->close_on_exec);
  35         else
  36                 FD_CLR(fd, fdt->close_on_exec);
  37         spin_unlock(&files->file_lock);
  38 }
  39
  40 static int get_close_on_exec(unsigned int fd)
  41 {
  42         struct files_struct *files = current->files;
  43         struct fdtable *fdt;
  44         int res;
  45         rcu_read_lock();
  46         fdt = files_fdtable(files);
  47         res = FD_ISSET(fd, fdt->close_on_exec);
  48         rcu_read_unlock();
  49         return res;
  50 }
  51
  52 asmlinkage long sys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
  53 {
  54         int err = -EBADF;
  55         struct file * file, *tofree;
  56         struct files_struct * files = current->files;
  57         struct fdtable *fdt;
  58
  59         if ((flags & ~O_CLOEXEC) != 0)
  60                 return -EINVAL;
  61
  62         if (unlikely(oldfd == newfd))
  63                 return -EINVAL;
  64
  65         spin_lock(&files->file_lock);
  66         err = expand_files(files, newfd);
  67         file = fcheck(oldfd);
  68         if (unlikely(!file))
  69                 goto Ebadf;
  70         if (unlikely(err < 0)) {
  71                 if (err == -EMFILE)
  72                         goto Ebadf;
  73                 goto out_unlock;
  74         }
  75         /*
  76          * We need to detect attempts to do dup2() over allocated but still
  77          * not finished descriptor.  NB: OpenBSD avoids that at the price of
  78          * extra work in their equivalent of fget() - they insert struct
  79          * file immediately after grabbing descriptor, mark it larval if
  80          * more work (e.g. actual opening) is needed and make sure that
  81          * fget() treats larval files as absent.  Potentially interesting,
  82          * but while extra work in fget() is trivial, locking implications
  83          * and amount of surgery on open()-related paths in VFS are not.
  84          * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
  85          * deadlocks in rather amusing ways, AFAICS.  All of that is out of
  86          * scope of POSIX or SUS, since neither considers shared descriptor
  87          * tables and this condition does not arise without those.
  88          */
  89         err = -EBUSY;
  90         fdt = files_fdtable(files);
  91         tofree = fdt->fd[newfd];
  92         if (!tofree && FD_ISSET(newfd, fdt->open_fds))
  93                 goto out_unlock;
  94         get_file(file);
  95         rcu_assign_pointer(fdt->fd[newfd], file);
  96         FD_SET(newfd, fdt->open_fds);
  97         if (flags & O_CLOEXEC)
  98                 FD_SET(newfd, fdt->close_on_exec);
  99         else
 100                 FD_CLR(newfd, fdt->close_on_exec);
 101         spin_unlock(&files->file_lock);
 102
 103         if (tofree)
 104                 filp_close(tofree, files);
 105
 106         return newfd;
 107
 108 Ebadf:
 109         err = -EBADF;
 110 out_unlock:
 111         spin_unlock(&files->file_lock);
 112         return err;
 113 }
 114
 115 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
 116 {
 117         if (unlikely(newfd == oldfd)) { /* corner case */
 118                 struct files_struct *files = current->files;
 119                 rcu_read_lock();
 120                 if (!fcheck_files(files, oldfd))
 121                         oldfd = -EBADF;
 122                 rcu_read_unlock();
 123                 return oldfd;
 124         }
 125         return sys_dup3(oldfd, newfd, 0);
 126 }
 127
 128 asmlinkage long sys_dup(unsigned int fildes)
 129 {
 130         int ret = -EBADF;
 131         struct file *file = fget(fildes);
 132
 133         if (file) {
 134                 ret = get_unused_fd();
 135                 if (ret >= 0)
 136                         fd_install(ret, file);
 137                 else
 138                         fput(file);
 139         }
 140         return ret;
 141 }
 142
 143 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
 144
 145 static int setfl(int fd, struct file * filp, unsigned long arg)
 146 {
 147         struct inode * inode = filp->f_path.dentry->d_inode;
 148         int error = 0;
 149
 150         /*
 151          * O_APPEND cannot be cleared if the file is marked as append-only
 152          * and the file is open for write.
 153          */
 154         if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
 155                 return -EPERM;
 156
 157         /* O_NOATIME can only be set by the owner or superuser */
 158         if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
 159                 if (!is_owner_or_cap(inode))
 160                         return -EPERM;
 161
 162         /* required for strict SunOS emulation */
 163         if (O_NONBLOCK != O_NDELAY)
 164                if (arg & O_NDELAY)
 165                    arg |= O_NONBLOCK;
 166
 167         if (arg & O_DIRECT) {
 168                 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
 169                         !filp->f_mapping->a_ops->direct_IO)
 170                                 return -EINVAL;
 171         }
 172
 173         if (filp->f_op && filp->f_op->check_flags)
 174                 error = filp->f_op->check_flags(arg);
 175         if (error)
 176                 return error;
 177
 178         if ((arg ^ filp->f_flags) & FASYNC) {
 179                 if (filp->f_op && filp->f_op->fasync) {
 180                         error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
 181                         if (error < 0)
 182                                 goto out;
 183                 }
 184         }
 185
 186         filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
 187  out:
 188         return error;
 189 }
 190
 191 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 192                      uid_t uid, uid_t euid, int force)
 193 {
 194         write_lock_irq(&filp->f_owner.lock);
 195         if (force || !filp->f_owner.pid) {
 196                 put_pid(filp->f_owner.pid);
 197                 filp->f_owner.pid = get_pid(pid);
 198                 filp->f_owner.pid_type = type;
 199                 filp->f_owner.uid = uid;
 200                 filp->f_owner.euid = euid;
 201         }
 202         write_unlock_irq(&filp->f_owner.lock);
 203 }
 204
 205 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
 206                 int force)
 207 {
 208         const struct cred *cred = current_cred();
 209         int err;
 210
 211         err = security_file_set_fowner(filp);
 212         if (err)
 213                 return err;
 214
 215         f_modown(filp, pid, type, cred->uid, cred->euid, force);
 216         return 0;
 217 }
 218 EXPORT_SYMBOL(__f_setown);
 219
 220 int f_setown(struct file *filp, unsigned long arg, int force)
 221 {
 222         enum pid_type type;
 223         struct pid *pid;
 224         int who = arg;
 225         int result;
 226         type = PIDTYPE_PID;
 227         if (who < 0) {
 228                 type = PIDTYPE_PGID;
 229                 who = -who;
 230         }
 231         rcu_read_lock();
 232         pid = find_vpid(who);
 233         result = __f_setown(filp, pid, type, force);
 234         rcu_read_unlock();
 235         return result;
 236 }
 237 EXPORT_SYMBOL(f_setown);
 238
 239 void f_delown(struct file *filp)
 240 {
 241         f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
 242 }
 243
 244 pid_t f_getown(struct file *filp)
 245 {
 246         pid_t pid;
 247         read_lock(&filp->f_owner.lock);
 248         pid = pid_vnr(filp->f_owner.pid);
 249         if (filp->f_owner.pid_type == PIDTYPE_PGID)
 250                 pid = -pid;
 251         read_unlock(&filp->f_owner.lock);
 252         return pid;
 253 }
 254
 255 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 256                 struct file *filp)
 257 {
 258         long err = -EINVAL;
 259
 260         switch (cmd) {
 261         case F_DUPFD:
 262         case F_DUPFD_CLOEXEC:
 263                 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 264                         break;
 265                 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
 266                 if (err >= 0) {
 267                         get_file(filp);
 268                         fd_install(err, filp);
 269                 }
 270                 break;
 271         case F_GETFD:
 272                 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
 273                 break;
 274         case F_SETFD:
 275                 err = 0;
 276                 set_close_on_exec(fd, arg & FD_CLOEXEC);
 277                 break;
 278         case F_GETFL:
 279                 err = filp->f_flags;
 280                 break;
 281         case F_SETFL:
 282                 err = setfl(fd, filp, arg);
 283                 break;
 284         case F_GETLK:
 285                 err = fcntl_getlk(filp, (struct flock __user *) arg);
 286                 break;
 287         case F_SETLK:
 288         case F_SETLKW:
 289                 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
 290                 break;
 291         case F_GETOWN:
 292                 /*
 293                  * XXX If f_owner is a process group, the
 294                  * negative return value will get converted
 295                  * into an error.  Oops.  If we keep the
 296                  * current syscall conventions, the only way
 297                  * to fix this will be in libc.
 298                  */
 299                 err = f_getown(filp);
 300                 force_successful_syscall_return();
 301                 break;
 302         case F_SETOWN:
 303                 err = f_setown(filp, arg, 1);
 304                 break;
 305         case F_GETSIG:
 306                 err = filp->f_owner.signum;
 307                 break;
 308         case F_SETSIG:
 309                 /* arg == 0 restores default behaviour. */
 310                 if (!valid_signal(arg)) {
 311                         break;
 312                 }
 313                 err = 0;
 314                 filp->f_owner.signum = arg;
 315                 break;
 316         case F_GETLEASE:
 317                 err = fcntl_getlease(filp);
 318                 break;
 319         case F_SETLEASE:
 320                 err = fcntl_setlease(fd, filp, arg);
 321                 break;
 322         case F_NOTIFY:
 323                 err = fcntl_dirnotify(fd, filp, arg);
 324                 break;
 325         default:
 326                 break;
 327         }
 328         return err;
 329 }
 330
 331 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
 332 {
 333         struct file *filp;
 334         long err = -EBADF;
 335
 336         filp = fget(fd);
 337         if (!filp)
 338                 goto out;
 339
 340         err = security_file_fcntl(filp, cmd, arg);
 341         if (err) {
 342                 fput(filp);
 343                 return err;
 344         }
 345
 346         err = do_fcntl(fd, cmd, arg, filp);
 347
 348         fput(filp);
 349 out:
 350         return err;
 351 }
 352
 353 #if BITS_PER_LONG == 32
 354 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
 355 {
 356         struct file * filp;
 357         long err;
 358
 359         err = -EBADF;
 360         filp = fget(fd);
 361         if (!filp)
 362                 goto out;
 363
 364         err = security_file_fcntl(filp, cmd, arg);
 365         if (err) {
 366                 fput(filp);
 367                 return err;
 368         }
 369         err = -EBADF;
 370
 371         switch (cmd) {
 372                 case F_GETLK64:
 373                         err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
 374                         break;
 375                 case F_SETLK64:
 376                 case F_SETLKW64:
 377                         err = fcntl_setlk64(fd, filp, cmd,
 378                                         (struct flock64 __user *) arg);
 379                         break;
 380                 default:
 381                         err = do_fcntl(fd, cmd, arg, filp);
 382                         break;
 383         }
 384         fput(filp);
 385 out:
 386         return err;
 387 }
 388 #endif
 389
 390 /* Table to convert sigio signal codes into poll band bitmaps */
 391
 392 static const long band_table[NSIGPOLL] = {
 393         POLLIN | POLLRDNORM,                    /* POLL_IN */
 394         POLLOUT | POLLWRNORM | POLLWRBAND,      /* POLL_OUT */
 395         POLLIN | POLLRDNORM | POLLMSG,          /* POLL_MSG */
 396         POLLERR,                                /* POLL_ERR */
 397         POLLPRI | POLLRDBAND,                   /* POLL_PRI */
 398         POLLHUP | POLLERR                       /* POLL_HUP */
 399 };
 400
 401 static inline int sigio_perm(struct task_struct *p,
 402                              struct fown_struct *fown, int sig)
 403 {
 404         const struct cred *cred;
 405         int ret;
 406
 407         rcu_read_lock();
 408         cred = __task_cred(p);
 409         ret = ((fown->euid == 0 ||
 410                 fown->euid == cred->suid || fown->euid == cred->uid ||
 411                 fown->uid  == cred->suid || fown->uid  == cred->uid) &&
 412                !security_file_send_sigiotask(p, fown, sig));
 413         rcu_read_unlock();
 414         return ret;
 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)