The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/sys_generic.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)sys_generic.c       8.5 (Berkeley) 1/21/94
   35  */
   36 
   37 #include <sys/cdefs.h>
   38 __FBSDID("$FreeBSD: stable/9/sys/kern/sys_generic.c 275231 2014-11-29 05:08:29Z hselasky $");
   39 
   40 #include "opt_capsicum.h"
   41 #include "opt_compat.h"
   42 #include "opt_ktrace.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/systm.h>
   46 #include <sys/sysproto.h>
   47 #include <sys/capability.h>
   48 #include <sys/filedesc.h>
   49 #include <sys/filio.h>
   50 #include <sys/fcntl.h>
   51 #include <sys/file.h>
   52 #include <sys/proc.h>
   53 #include <sys/signalvar.h>
   54 #include <sys/socketvar.h>
   55 #include <sys/uio.h>
   56 #include <sys/kernel.h>
   57 #include <sys/ktr.h>
   58 #include <sys/limits.h>
   59 #include <sys/malloc.h>
   60 #include <sys/poll.h>
   61 #include <sys/resourcevar.h>
   62 #include <sys/selinfo.h>
   63 #include <sys/sleepqueue.h>
   64 #include <sys/syscallsubr.h>
   65 #include <sys/sysctl.h>
   66 #include <sys/sysent.h>
   67 #include <sys/vnode.h>
   68 #include <sys/bio.h>
   69 #include <sys/buf.h>
   70 #include <sys/condvar.h>
   71 #ifdef KTRACE
   72 #include <sys/ktrace.h>
   73 #endif
   74 
   75 #include <security/audit/audit.h>
   76 
   77 /*
   78  * The following macro defines how many bytes will be allocated from
   79  * the stack instead of memory allocated when passing the IOCTL data
   80  * structures from userspace and to the kernel. Some IOCTLs having
   81  * small data structures are used very frequently and this small
   82  * buffer on the stack gives a significant speedup improvement for
   83  * those requests. The value of this define should be greater or equal
   84  * to 64 bytes and should also be power of two. The data structure is
   85  * currently hard-aligned to a 8-byte boundary on the stack. This
   86  * should currently be sufficient for all supported platforms.
   87  */
   88 #define SYS_IOCTL_SMALL_SIZE    128     /* bytes */
   89 #define SYS_IOCTL_SMALL_ALIGN   8       /* bytes */
   90 
   91 int iosize_max_clamp = 1;
   92 SYSCTL_INT(_debug, OID_AUTO, iosize_max_clamp, CTLFLAG_RW,
   93     &iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX");
   94 int devfs_iosize_max_clamp = 1;
   95 SYSCTL_INT(_debug, OID_AUTO, devfs_iosize_max_clamp, CTLFLAG_RW,
   96     &devfs_iosize_max_clamp, 0, "Clamp max i/o size to INT_MAX for devices");
   97 
   98 /*
   99  * Assert that the return value of read(2) and write(2) syscalls fits
  100  * into a register.  If not, an architecture will need to provide the
  101  * usermode wrappers to reconstruct the result.
  102  */
  103 CTASSERT(sizeof(register_t) >= sizeof(size_t));
  104 
  105 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
  106 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
  107 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
  108 
  109 static int      pollout(struct thread *, struct pollfd *, struct pollfd *,
  110                     u_int);
  111 static int      pollscan(struct thread *, struct pollfd *, u_int);
  112 static int      pollrescan(struct thread *);
  113 static int      selscan(struct thread *, fd_mask **, fd_mask **, int);
  114 static int      selrescan(struct thread *, fd_mask **, fd_mask **);
  115 static void     selfdalloc(struct thread *, void *);
  116 static void     selfdfree(struct seltd *, struct selfd *);
  117 static int      dofileread(struct thread *, int, struct file *, struct uio *,
  118                     off_t, int);
  119 static int      dofilewrite(struct thread *, int, struct file *, struct uio *,
  120                     off_t, int);
  121 static void     doselwakeup(struct selinfo *, int);
  122 static void     seltdinit(struct thread *);
  123 static int      seltdwait(struct thread *, int);
  124 static void     seltdclear(struct thread *);
  125 
  126 /*
  127  * One seltd per-thread allocated on demand as needed.
  128  *
  129  *      t - protected by st_mtx
  130  *      k - Only accessed by curthread or read-only
  131  */
  132 struct seltd {
  133         STAILQ_HEAD(, selfd)    st_selq;        /* (k) List of selfds. */
  134         struct selfd            *st_free1;      /* (k) free fd for read set. */
  135         struct selfd            *st_free2;      /* (k) free fd for write set. */
  136         struct mtx              st_mtx;         /* Protects struct seltd */
  137         struct cv               st_wait;        /* (t) Wait channel. */
  138         int                     st_flags;       /* (t) SELTD_ flags. */
  139 };
  140 
  141 #define SELTD_PENDING   0x0001                  /* We have pending events. */
  142 #define SELTD_RESCAN    0x0002                  /* Doing a rescan. */
  143 
  144 /*
  145  * One selfd allocated per-thread per-file-descriptor.
  146  *      f - protected by sf_mtx
  147  */
  148 struct selfd {
  149         STAILQ_ENTRY(selfd)     sf_link;        /* (k) fds owned by this td. */
  150         TAILQ_ENTRY(selfd)      sf_threads;     /* (f) fds on this selinfo. */
  151         struct selinfo          *sf_si;         /* (f) selinfo when linked. */
  152         struct mtx              *sf_mtx;        /* Pointer to selinfo mtx. */
  153         struct seltd            *sf_td;         /* (k) owning seltd. */
  154         void                    *sf_cookie;     /* (k) fd or pollfd. */
  155 };
  156 
  157 static uma_zone_t selfd_zone;
  158 static struct mtx_pool *mtxpool_select;
  159 
  160 #ifndef _SYS_SYSPROTO_H_
  161 struct read_args {
  162         int     fd;
  163         void    *buf;
  164         size_t  nbyte;
  165 };
  166 #endif
  167 int
  168 sys_read(td, uap)
  169         struct thread *td;
  170         struct read_args *uap;
  171 {
  172         struct uio auio;
  173         struct iovec aiov;
  174         int error;
  175 
  176         if (uap->nbyte > IOSIZE_MAX)
  177                 return (EINVAL);
  178         aiov.iov_base = uap->buf;
  179         aiov.iov_len = uap->nbyte;
  180         auio.uio_iov = &aiov;
  181         auio.uio_iovcnt = 1;
  182         auio.uio_resid = uap->nbyte;
  183         auio.uio_segflg = UIO_USERSPACE;
  184         error = kern_readv(td, uap->fd, &auio);
  185         return(error);
  186 }
  187 
  188 /*
  189  * Positioned read system call
  190  */
  191 #ifndef _SYS_SYSPROTO_H_
  192 struct pread_args {
  193         int     fd;
  194         void    *buf;
  195         size_t  nbyte;
  196         int     pad;
  197         off_t   offset;
  198 };
  199 #endif
  200 int
  201 sys_pread(td, uap)
  202         struct thread *td;
  203         struct pread_args *uap;
  204 {
  205         struct uio auio;
  206         struct iovec aiov;
  207         int error;
  208 
  209         if (uap->nbyte > IOSIZE_MAX)
  210                 return (EINVAL);
  211         aiov.iov_base = uap->buf;
  212         aiov.iov_len = uap->nbyte;
  213         auio.uio_iov = &aiov;
  214         auio.uio_iovcnt = 1;
  215         auio.uio_resid = uap->nbyte;
  216         auio.uio_segflg = UIO_USERSPACE;
  217         error = kern_preadv(td, uap->fd, &auio, uap->offset);
  218         return(error);
  219 }
  220 
  221 int
  222 freebsd6_pread(td, uap)
  223         struct thread *td;
  224         struct freebsd6_pread_args *uap;
  225 {
  226         struct pread_args oargs;
  227 
  228         oargs.fd = uap->fd;
  229         oargs.buf = uap->buf;
  230         oargs.nbyte = uap->nbyte;
  231         oargs.offset = uap->offset;
  232         return (sys_pread(td, &oargs));
  233 }
  234 
  235 /*
  236  * Scatter read system call.
  237  */
  238 #ifndef _SYS_SYSPROTO_H_
  239 struct readv_args {
  240         int     fd;
  241         struct  iovec *iovp;
  242         u_int   iovcnt;
  243 };
  244 #endif
  245 int
  246 sys_readv(struct thread *td, struct readv_args *uap)
  247 {
  248         struct uio *auio;
  249         int error;
  250 
  251         error = copyinuio(uap->iovp, uap->iovcnt, &auio);
  252         if (error)
  253                 return (error);
  254         error = kern_readv(td, uap->fd, auio);
  255         free(auio, M_IOV);
  256         return (error);
  257 }
  258 
  259 int
  260 kern_readv(struct thread *td, int fd, struct uio *auio)
  261 {
  262         struct file *fp;
  263         int error;
  264 
  265         error = fget_read(td, fd, CAP_READ | CAP_SEEK, &fp);
  266         if (error)
  267                 return (error);
  268         error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
  269         fdrop(fp, td);
  270         return (error);
  271 }
  272 
  273 /*
  274  * Scatter positioned read system call.
  275  */
  276 #ifndef _SYS_SYSPROTO_H_
  277 struct preadv_args {
  278         int     fd;
  279         struct  iovec *iovp;
  280         u_int   iovcnt;
  281         off_t   offset;
  282 };
  283 #endif
  284 int
  285 sys_preadv(struct thread *td, struct preadv_args *uap)
  286 {
  287         struct uio *auio;
  288         int error;
  289 
  290         error = copyinuio(uap->iovp, uap->iovcnt, &auio);
  291         if (error)
  292                 return (error);
  293         error = kern_preadv(td, uap->fd, auio, uap->offset);
  294         free(auio, M_IOV);
  295         return (error);
  296 }
  297 
  298 int
  299 kern_preadv(td, fd, auio, offset)
  300         struct thread *td;
  301         int fd;
  302         struct uio *auio;
  303         off_t offset;
  304 {
  305         struct file *fp;
  306         int error;
  307 
  308         error = fget_read(td, fd, CAP_READ, &fp);
  309         if (error)
  310                 return (error);
  311         if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
  312                 error = ESPIPE;
  313         else if (offset < 0 && fp->f_vnode->v_type != VCHR)
  314                 error = EINVAL;
  315         else
  316                 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
  317         fdrop(fp, td);
  318         return (error);
  319 }
  320 
  321 /*
  322  * Common code for readv and preadv that reads data in
  323  * from a file using the passed in uio, offset, and flags.
  324  */
  325 static int
  326 dofileread(td, fd, fp, auio, offset, flags)
  327         struct thread *td;
  328         int fd;
  329         struct file *fp;
  330         struct uio *auio;
  331         off_t offset;
  332         int flags;
  333 {
  334         ssize_t cnt;
  335         int error;
  336 #ifdef KTRACE
  337         struct uio *ktruio = NULL;
  338 #endif
  339 
  340         /* Finish zero length reads right here */
  341         if (auio->uio_resid == 0) {
  342                 td->td_retval[0] = 0;
  343                 return(0);
  344         }
  345         auio->uio_rw = UIO_READ;
  346         auio->uio_offset = offset;
  347         auio->uio_td = td;
  348 #ifdef KTRACE
  349         if (KTRPOINT(td, KTR_GENIO)) 
  350                 ktruio = cloneuio(auio);
  351 #endif
  352         cnt = auio->uio_resid;
  353         if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
  354                 if (auio->uio_resid != cnt && (error == ERESTART ||
  355                     error == EINTR || error == EWOULDBLOCK))
  356                         error = 0;
  357         }
  358         cnt -= auio->uio_resid;
  359 #ifdef KTRACE
  360         if (ktruio != NULL) {
  361                 ktruio->uio_resid = cnt;
  362                 ktrgenio(fd, UIO_READ, ktruio, error);
  363         }
  364 #endif
  365         td->td_retval[0] = cnt;
  366         return (error);
  367 }
  368 
  369 #ifndef _SYS_SYSPROTO_H_
  370 struct write_args {
  371         int     fd;
  372         const void *buf;
  373         size_t  nbyte;
  374 };
  375 #endif
  376 int
  377 sys_write(td, uap)
  378         struct thread *td;
  379         struct write_args *uap;
  380 {
  381         struct uio auio;
  382         struct iovec aiov;
  383         int error;
  384 
  385         if (uap->nbyte > IOSIZE_MAX)
  386                 return (EINVAL);
  387         aiov.iov_base = (void *)(uintptr_t)uap->buf;
  388         aiov.iov_len = uap->nbyte;
  389         auio.uio_iov = &aiov;
  390         auio.uio_iovcnt = 1;
  391         auio.uio_resid = uap->nbyte;
  392         auio.uio_segflg = UIO_USERSPACE;
  393         error = kern_writev(td, uap->fd, &auio);
  394         return(error);
  395 }
  396 
  397 /*
  398  * Positioned write system call.
  399  */
  400 #ifndef _SYS_SYSPROTO_H_
  401 struct pwrite_args {
  402         int     fd;
  403         const void *buf;
  404         size_t  nbyte;
  405         int     pad;
  406         off_t   offset;
  407 };
  408 #endif
  409 int
  410 sys_pwrite(td, uap)
  411         struct thread *td;
  412         struct pwrite_args *uap;
  413 {
  414         struct uio auio;
  415         struct iovec aiov;
  416         int error;
  417 
  418         if (uap->nbyte > IOSIZE_MAX)
  419                 return (EINVAL);
  420         aiov.iov_base = (void *)(uintptr_t)uap->buf;
  421         aiov.iov_len = uap->nbyte;
  422         auio.uio_iov = &aiov;
  423         auio.uio_iovcnt = 1;
  424         auio.uio_resid = uap->nbyte;
  425         auio.uio_segflg = UIO_USERSPACE;
  426         error = kern_pwritev(td, uap->fd, &auio, uap->offset);
  427         return(error);
  428 }
  429 
  430 int
  431 freebsd6_pwrite(td, uap)
  432         struct thread *td;
  433         struct freebsd6_pwrite_args *uap;
  434 {
  435         struct pwrite_args oargs;
  436 
  437         oargs.fd = uap->fd;
  438         oargs.buf = uap->buf;
  439         oargs.nbyte = uap->nbyte;
  440         oargs.offset = uap->offset;
  441         return (sys_pwrite(td, &oargs));
  442 }
  443 
  444 /*
  445  * Gather write system call.
  446  */
  447 #ifndef _SYS_SYSPROTO_H_
  448 struct writev_args {
  449         int     fd;
  450         struct  iovec *iovp;
  451         u_int   iovcnt;
  452 };
  453 #endif
  454 int
  455 sys_writev(struct thread *td, struct writev_args *uap)
  456 {
  457         struct uio *auio;
  458         int error;
  459 
  460         error = copyinuio(uap->iovp, uap->iovcnt, &auio);
  461         if (error)
  462                 return (error);
  463         error = kern_writev(td, uap->fd, auio);
  464         free(auio, M_IOV);
  465         return (error);
  466 }
  467 
  468 int
  469 kern_writev(struct thread *td, int fd, struct uio *auio)
  470 {
  471         struct file *fp;
  472         int error;
  473 
  474         error = fget_write(td, fd, CAP_WRITE | CAP_SEEK, &fp);
  475         if (error)
  476                 return (error);
  477         error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
  478         fdrop(fp, td);
  479         return (error);
  480 }
  481 
  482 /*
  483  * Gather positioned write system call.
  484  */
  485 #ifndef _SYS_SYSPROTO_H_
  486 struct pwritev_args {
  487         int     fd;
  488         struct  iovec *iovp;
  489         u_int   iovcnt;
  490         off_t   offset;
  491 };
  492 #endif
  493 int
  494 sys_pwritev(struct thread *td, struct pwritev_args *uap)
  495 {
  496         struct uio *auio;
  497         int error;
  498 
  499         error = copyinuio(uap->iovp, uap->iovcnt, &auio);
  500         if (error)
  501                 return (error);
  502         error = kern_pwritev(td, uap->fd, auio, uap->offset);
  503         free(auio, M_IOV);
  504         return (error);
  505 }
  506 
  507 int
  508 kern_pwritev(td, fd, auio, offset)
  509         struct thread *td;
  510         struct uio *auio;
  511         int fd;
  512         off_t offset;
  513 {
  514         struct file *fp;
  515         int error;
  516 
  517         error = fget_write(td, fd, CAP_WRITE, &fp);
  518         if (error)
  519                 return (error);
  520         if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
  521                 error = ESPIPE;
  522         else if (offset < 0 && fp->f_vnode->v_type != VCHR)
  523                 error = EINVAL;
  524         else
  525                 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
  526         fdrop(fp, td);
  527         return (error);
  528 }
  529 
  530 /*
  531  * Common code for writev and pwritev that writes data to
  532  * a file using the passed in uio, offset, and flags.
  533  */
  534 static int
  535 dofilewrite(td, fd, fp, auio, offset, flags)
  536         struct thread *td;
  537         int fd;
  538         struct file *fp;
  539         struct uio *auio;
  540         off_t offset;
  541         int flags;
  542 {
  543         ssize_t cnt;
  544         int error;
  545 #ifdef KTRACE
  546         struct uio *ktruio = NULL;
  547 #endif
  548 
  549         auio->uio_rw = UIO_WRITE;
  550         auio->uio_td = td;
  551         auio->uio_offset = offset;
  552 #ifdef KTRACE
  553         if (KTRPOINT(td, KTR_GENIO))
  554                 ktruio = cloneuio(auio);
  555 #endif
  556         cnt = auio->uio_resid;
  557         if (fp->f_type == DTYPE_VNODE &&
  558             (fp->f_vnread_flags & FDEVFS_VNODE) == 0)
  559                 bwillwrite();
  560         if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
  561                 if (auio->uio_resid != cnt && (error == ERESTART ||
  562                     error == EINTR || error == EWOULDBLOCK))
  563                         error = 0;
  564                 /* Socket layer is responsible for issuing SIGPIPE. */
  565                 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
  566                         PROC_LOCK(td->td_proc);
  567                         tdsignal(td, SIGPIPE);
  568                         PROC_UNLOCK(td->td_proc);
  569                 }
  570         }
  571         cnt -= auio->uio_resid;
  572 #ifdef KTRACE
  573         if (ktruio != NULL) {
  574                 ktruio->uio_resid = cnt;
  575                 ktrgenio(fd, UIO_WRITE, ktruio, error);
  576         }
  577 #endif
  578         td->td_retval[0] = cnt;
  579         return (error);
  580 }
  581 
  582 /*
  583  * Truncate a file given a file descriptor.
  584  *
  585  * Can't use fget_write() here, since must return EINVAL and not EBADF if the
  586  * descriptor isn't writable.
  587  */
  588 int
  589 kern_ftruncate(td, fd, length)
  590         struct thread *td;
  591         int fd;
  592         off_t length;
  593 {
  594         struct file *fp;
  595         int error;
  596 
  597         AUDIT_ARG_FD(fd);
  598         if (length < 0)
  599                 return (EINVAL);
  600         error = fget(td, fd, CAP_FTRUNCATE, &fp);
  601         if (error)
  602                 return (error);
  603         AUDIT_ARG_FILE(td->td_proc, fp);
  604         if (!(fp->f_flag & FWRITE)) {
  605                 fdrop(fp, td);
  606                 return (EINVAL);
  607         }
  608         error = fo_truncate(fp, length, td->td_ucred, td);
  609         fdrop(fp, td);
  610         return (error);
  611 }
  612 
  613 #ifndef _SYS_SYSPROTO_H_
  614 struct ftruncate_args {
  615         int     fd;
  616         int     pad;
  617         off_t   length;
  618 };
  619 #endif
  620 int
  621 sys_ftruncate(td, uap)
  622         struct thread *td;
  623         struct ftruncate_args *uap;
  624 {
  625 
  626         return (kern_ftruncate(td, uap->fd, uap->length));
  627 }
  628 
  629 #if defined(COMPAT_43)
  630 #ifndef _SYS_SYSPROTO_H_
  631 struct oftruncate_args {
  632         int     fd;
  633         long    length;
  634 };
  635 #endif
  636 int
  637 oftruncate(td, uap)
  638         struct thread *td;
  639         struct oftruncate_args *uap;
  640 {
  641 
  642         return (kern_ftruncate(td, uap->fd, uap->length));
  643 }
  644 #endif /* COMPAT_43 */
  645 
  646 #ifndef _SYS_SYSPROTO_H_
  647 struct ioctl_args {
  648         int     fd;
  649         u_long  com;
  650         caddr_t data;
  651 };
  652 #endif
  653 /* ARGSUSED */
  654 int
  655 sys_ioctl(struct thread *td, struct ioctl_args *uap)
  656 {
  657         u_char smalldata[SYS_IOCTL_SMALL_SIZE] __aligned(SYS_IOCTL_SMALL_ALIGN);
  658         u_long com;
  659         int arg, error;
  660         u_int size;
  661         caddr_t data;
  662 
  663         if (uap->com > 0xffffffff) {
  664                 printf(
  665                     "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
  666                     td->td_proc->p_pid, td->td_name, uap->com);
  667                 uap->com &= 0xffffffff;
  668         }
  669         com = uap->com;
  670 
  671         /*
  672          * Interpret high order word to find amount of data to be
  673          * copied to/from the user's address space.
  674          */
  675         size = IOCPARM_LEN(com);
  676         if ((size > IOCPARM_MAX) ||
  677             ((com & (IOC_VOID  | IOC_IN | IOC_OUT)) == 0) ||
  678 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
  679             ((com & IOC_OUT) && size == 0) ||
  680 #else
  681             ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
  682 #endif
  683             ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
  684                 return (ENOTTY);
  685 
  686         if (size > 0) {
  687                 if (com & IOC_VOID) {
  688                         /* Integer argument. */
  689                         arg = (intptr_t)uap->data;
  690                         data = (void *)&arg;
  691                         size = 0;
  692                 } else {
  693                         if (size > SYS_IOCTL_SMALL_SIZE)
  694                                 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
  695                         else
  696                                 data = smalldata;
  697                 }
  698         } else
  699                 data = (void *)&uap->data;
  700         if (com & IOC_IN) {
  701                 error = copyin(uap->data, data, (u_int)size);
  702                 if (error != 0)
  703                         goto out;
  704         } else if (com & IOC_OUT) {
  705                 /*
  706                  * Zero the buffer so the user always
  707                  * gets back something deterministic.
  708                  */
  709                 bzero(data, size);
  710         }
  711 
  712         error = kern_ioctl(td, uap->fd, com, data);
  713 
  714         if (error == 0 && (com & IOC_OUT))
  715                 error = copyout(data, uap->data, (u_int)size);
  716 
  717 out:
  718         if (size > SYS_IOCTL_SMALL_SIZE)
  719                 free(data, M_IOCTLOPS);
  720         return (error);
  721 }
  722 
  723 int
  724 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
  725 {
  726         struct file *fp;
  727         struct filedesc *fdp;
  728         int error;
  729         int tmp;
  730 
  731         AUDIT_ARG_FD(fd);
  732         AUDIT_ARG_CMD(com);
  733         if ((error = fget(td, fd, CAP_IOCTL, &fp)) != 0)
  734                 return (error);
  735         if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
  736                 fdrop(fp, td);
  737                 return (EBADF);
  738         }
  739         fdp = td->td_proc->p_fd;
  740         switch (com) {
  741         case FIONCLEX:
  742                 FILEDESC_XLOCK(fdp);
  743                 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
  744                 FILEDESC_XUNLOCK(fdp);
  745                 goto out;
  746         case FIOCLEX:
  747                 FILEDESC_XLOCK(fdp);
  748                 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
  749                 FILEDESC_XUNLOCK(fdp);
  750                 goto out;
  751         case FIONBIO:
  752                 if ((tmp = *(int *)data))
  753                         atomic_set_int(&fp->f_flag, FNONBLOCK);
  754                 else
  755                         atomic_clear_int(&fp->f_flag, FNONBLOCK);
  756                 data = (void *)&tmp;
  757                 break;
  758         case FIOASYNC:
  759                 if ((tmp = *(int *)data))
  760                         atomic_set_int(&fp->f_flag, FASYNC);
  761                 else
  762                         atomic_clear_int(&fp->f_flag, FASYNC);
  763                 data = (void *)&tmp;
  764                 break;
  765         }
  766 
  767         error = fo_ioctl(fp, com, data, td->td_ucred, td);
  768 out:
  769         fdrop(fp, td);
  770         return (error);
  771 }
  772 
  773 int
  774 poll_no_poll(int events)
  775 {
  776         /*
  777          * Return true for read/write.  If the user asked for something
  778          * special, return POLLNVAL, so that clients have a way of
  779          * determining reliably whether or not the extended
  780          * functionality is present without hard-coding knowledge
  781          * of specific filesystem implementations.
  782          */
  783         if (events & ~POLLSTANDARD)
  784                 return (POLLNVAL);
  785 
  786         return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
  787 }
  788 
  789 int
  790 sys_pselect(struct thread *td, struct pselect_args *uap)
  791 {
  792         struct timespec ts;
  793         struct timeval tv, *tvp;
  794         sigset_t set, *uset;
  795         int error;
  796 
  797         if (uap->ts != NULL) {
  798                 error = copyin(uap->ts, &ts, sizeof(ts));
  799                 if (error != 0)
  800                     return (error);
  801                 TIMESPEC_TO_TIMEVAL(&tv, &ts);
  802                 tvp = &tv;
  803         } else
  804                 tvp = NULL;
  805         if (uap->sm != NULL) {
  806                 error = copyin(uap->sm, &set, sizeof(set));
  807                 if (error != 0)
  808                         return (error);
  809                 uset = &set;
  810         } else
  811                 uset = NULL;
  812         return (kern_pselect(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
  813             uset, NFDBITS));
  814 }
  815 
  816 int
  817 kern_pselect(struct thread *td, int nd, fd_set *in, fd_set *ou, fd_set *ex,
  818     struct timeval *tvp, sigset_t *uset, int abi_nfdbits)
  819 {
  820         int error;
  821 
  822         if (uset != NULL) {
  823                 error = kern_sigprocmask(td, SIG_SETMASK, uset,
  824                     &td->td_oldsigmask, 0);
  825                 if (error != 0)
  826                         return (error);
  827                 td->td_pflags |= TDP_OLDMASK;
  828                 /*
  829                  * Make sure that ast() is called on return to
  830                  * usermode and TDP_OLDMASK is cleared, restoring old
  831                  * sigmask.
  832                  */
  833                 thread_lock(td);
  834                 td->td_flags |= TDF_ASTPENDING;
  835                 thread_unlock(td);
  836         }
  837         error = kern_select(td, nd, in, ou, ex, tvp, abi_nfdbits);
  838         return (error);
  839 }
  840 
  841 #ifndef _SYS_SYSPROTO_H_
  842 struct select_args {
  843         int     nd;
  844         fd_set  *in, *ou, *ex;
  845         struct  timeval *tv;
  846 };
  847 #endif
  848 int
  849 sys_select(struct thread *td, struct select_args *uap)
  850 {
  851         struct timeval tv, *tvp;
  852         int error;
  853 
  854         if (uap->tv != NULL) {
  855                 error = copyin(uap->tv, &tv, sizeof(tv));
  856                 if (error)
  857                         return (error);
  858                 tvp = &tv;
  859         } else
  860                 tvp = NULL;
  861 
  862         return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp,
  863             NFDBITS));
  864 }
  865 
  866 /*
  867  * In the unlikely case when user specified n greater then the last
  868  * open file descriptor, check that no bits are set after the last
  869  * valid fd.  We must return EBADF if any is set.
  870  *
  871  * There are applications that rely on the behaviour.
  872  *
  873  * nd is fd_lastfile + 1.
  874  */
  875 static int
  876 select_check_badfd(fd_set *fd_in, int nd, int ndu, int abi_nfdbits)
  877 {
  878         char *addr, *oaddr;
  879         int b, i, res;
  880         uint8_t bits;
  881 
  882         if (nd >= ndu || fd_in == NULL)
  883                 return (0);
  884 
  885         oaddr = NULL;
  886         bits = 0; /* silence gcc */
  887         for (i = nd; i < ndu; i++) {
  888                 b = i / NBBY;
  889 #if BYTE_ORDER == LITTLE_ENDIAN
  890                 addr = (char *)fd_in + b;
  891 #else
  892                 addr = (char *)fd_in;
  893                 if (abi_nfdbits == NFDBITS) {
  894                         addr += rounddown(b, sizeof(fd_mask)) +
  895                             sizeof(fd_mask) - 1 - b % sizeof(fd_mask);
  896                 } else {
  897                         addr += rounddown(b, sizeof(uint32_t)) +
  898                             sizeof(uint32_t) - 1 - b % sizeof(uint32_t);
  899                 }
  900 #endif
  901                 if (addr != oaddr) {
  902                         res = fubyte(addr);
  903                         if (res == -1)
  904                                 return (EFAULT);
  905                         oaddr = addr;
  906                         bits = res;
  907                 }
  908                 if ((bits & (1 << (i % NBBY))) != 0)
  909                         return (EBADF);
  910         }
  911         return (0);
  912 }
  913 
  914 int
  915 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
  916     fd_set *fd_ex, struct timeval *tvp, int abi_nfdbits)
  917 {
  918         struct filedesc *fdp;
  919         /*
  920          * The magic 2048 here is chosen to be just enough for FD_SETSIZE
  921          * infds with the new FD_SETSIZE of 1024, and more than enough for
  922          * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
  923          * of 256.
  924          */
  925         fd_mask s_selbits[howmany(2048, NFDBITS)];
  926         fd_mask *ibits[3], *obits[3], *selbits, *sbp;
  927         struct timeval atv, rtv, ttv;
  928         int error, lf, ndu, timo;
  929         u_int nbufbytes, ncpbytes, ncpubytes, nfdbits;
  930 
  931         if (nd < 0)
  932                 return (EINVAL);
  933         fdp = td->td_proc->p_fd;
  934         ndu = nd;
  935         lf = fdp->fd_lastfile;
  936         if (nd > lf + 1)
  937                 nd = lf + 1;
  938 
  939         error = select_check_badfd(fd_in, nd, ndu, abi_nfdbits);
  940         if (error != 0)
  941                 return (error);
  942         error = select_check_badfd(fd_ou, nd, ndu, abi_nfdbits);
  943         if (error != 0)
  944                 return (error);
  945         error = select_check_badfd(fd_ex, nd, ndu, abi_nfdbits);
  946         if (error != 0)
  947                 return (error);
  948 
  949         /*
  950          * Allocate just enough bits for the non-null fd_sets.  Use the
  951          * preallocated auto buffer if possible.
  952          */
  953         nfdbits = roundup(nd, NFDBITS);
  954         ncpbytes = nfdbits / NBBY;
  955         ncpubytes = roundup(nd, abi_nfdbits) / NBBY;
  956         nbufbytes = 0;
  957         if (fd_in != NULL)
  958                 nbufbytes += 2 * ncpbytes;
  959         if (fd_ou != NULL)
  960                 nbufbytes += 2 * ncpbytes;
  961         if (fd_ex != NULL)
  962                 nbufbytes += 2 * ncpbytes;
  963         if (nbufbytes <= sizeof s_selbits)
  964                 selbits = &s_selbits[0];
  965         else
  966                 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
  967 
  968         /*
  969          * Assign pointers into the bit buffers and fetch the input bits.
  970          * Put the output buffers together so that they can be bzeroed
  971          * together.
  972          */
  973         sbp = selbits;
  974 #define getbits(name, x) \
  975         do {                                                            \
  976                 if (name == NULL) {                                     \
  977                         ibits[x] = NULL;                                \
  978                         obits[x] = NULL;                                \
  979                 } else {                                                \
  980                         ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp;   \
  981                         obits[x] = sbp;                                 \
  982                         sbp += ncpbytes / sizeof *sbp;                  \
  983                         error = copyin(name, ibits[x], ncpubytes);      \
  984                         if (error != 0)                                 \
  985                                 goto done;                              \
  986                         bzero((char *)ibits[x] + ncpubytes,             \
  987                             ncpbytes - ncpubytes);                      \
  988                 }                                                       \
  989         } while (0)
  990         getbits(fd_in, 0);
  991         getbits(fd_ou, 1);
  992         getbits(fd_ex, 2);
  993 #undef  getbits
  994 
  995 #if BYTE_ORDER == BIG_ENDIAN && defined(__LP64__)
  996         /*
  997          * XXX: swizzle_fdset assumes that if abi_nfdbits != NFDBITS,
  998          * we are running under 32-bit emulation. This should be more
  999          * generic.
 1000          */
 1001 #define swizzle_fdset(bits)                                             \
 1002         if (abi_nfdbits != NFDBITS && bits != NULL) {                   \
 1003                 int i;                                                  \
 1004                 for (i = 0; i < ncpbytes / sizeof *sbp; i++)            \
 1005                         bits[i] = (bits[i] >> 32) | (bits[i] << 32);    \
 1006         }
 1007 #else
 1008 #define swizzle_fdset(bits)
 1009 #endif
 1010 
 1011         /* Make sure the bit order makes it through an ABI transition */
 1012         swizzle_fdset(ibits[0]);
 1013         swizzle_fdset(ibits[1]);
 1014         swizzle_fdset(ibits[2]);
 1015         
 1016         if (nbufbytes != 0)
 1017                 bzero(selbits, nbufbytes / 2);
 1018 
 1019         if (tvp != NULL) {
 1020                 atv = *tvp;
 1021                 if (itimerfix(&atv)) {
 1022                         error = EINVAL;
 1023                         goto done;
 1024                 }
 1025                 getmicrouptime(&rtv);
 1026                 timevaladd(&atv, &rtv);
 1027         } else {
 1028                 atv.tv_sec = 0;
 1029                 atv.tv_usec = 0;
 1030         }
 1031         timo = 0;
 1032         seltdinit(td);
 1033         /* Iterate until the timeout expires or descriptors become ready. */
 1034         for (;;) {
 1035                 error = selscan(td, ibits, obits, nd);
 1036                 if (error || td->td_retval[0] != 0)
 1037                         break;
 1038                 if (atv.tv_sec || atv.tv_usec) {
 1039                         getmicrouptime(&rtv);
 1040                         if (timevalcmp(&rtv, &atv, >=))
 1041                                 break;
 1042                         ttv = atv;
 1043                         timevalsub(&ttv, &rtv);
 1044                         timo = ttv.tv_sec > 24 * 60 * 60 ?
 1045                             24 * 60 * 60 * hz : tvtohz(&ttv);
 1046                 }
 1047                 error = seltdwait(td, timo);
 1048                 if (error)
 1049                         break;
 1050                 error = selrescan(td, ibits, obits);
 1051                 if (error || td->td_retval[0] != 0)
 1052                         break;
 1053         }
 1054         seltdclear(td);
 1055 
 1056 done:
 1057         /* select is not restarted after signals... */
 1058         if (error == ERESTART)
 1059                 error = EINTR;
 1060         if (error == EWOULDBLOCK)
 1061                 error = 0;
 1062 
 1063         /* swizzle bit order back, if necessary */
 1064         swizzle_fdset(obits[0]);
 1065         swizzle_fdset(obits[1]);
 1066         swizzle_fdset(obits[2]);
 1067 #undef swizzle_fdset
 1068 
 1069 #define putbits(name, x) \
 1070         if (name && (error2 = copyout(obits[x], name, ncpubytes))) \
 1071                 error = error2;
 1072         if (error == 0) {
 1073                 int error2;
 1074 
 1075                 putbits(fd_in, 0);
 1076                 putbits(fd_ou, 1);
 1077                 putbits(fd_ex, 2);
 1078 #undef putbits
 1079         }
 1080         if (selbits != &s_selbits[0])
 1081                 free(selbits, M_SELECT);
 1082 
 1083         return (error);
 1084 }
 1085 /* 
 1086  * Convert a select bit set to poll flags.
 1087  *
 1088  * The backend always returns POLLHUP/POLLERR if appropriate and we
 1089  * return this as a set bit in any set.
 1090  */
 1091 static int select_flags[3] = {
 1092     POLLRDNORM | POLLHUP | POLLERR,
 1093     POLLWRNORM | POLLHUP | POLLERR,
 1094     POLLRDBAND | POLLERR
 1095 };
 1096 
 1097 /*
 1098  * Compute the fo_poll flags required for a fd given by the index and
 1099  * bit position in the fd_mask array.
 1100  */
 1101 static __inline int
 1102 selflags(fd_mask **ibits, int idx, fd_mask bit)
 1103 {
 1104         int flags;
 1105         int msk;
 1106 
 1107         flags = 0;
 1108         for (msk = 0; msk < 3; msk++) {
 1109                 if (ibits[msk] == NULL)
 1110                         continue;
 1111                 if ((ibits[msk][idx] & bit) == 0)
 1112                         continue;
 1113                 flags |= select_flags[msk];
 1114         }
 1115         return (flags);
 1116 }
 1117 
 1118 /*
 1119  * Set the appropriate output bits given a mask of fired events and the
 1120  * input bits originally requested.
 1121  */
 1122 static __inline int
 1123 selsetbits(fd_mask **ibits, fd_mask **obits, int idx, fd_mask bit, int events)
 1124 {
 1125         int msk;
 1126         int n;
 1127 
 1128         n = 0;
 1129         for (msk = 0; msk < 3; msk++) {
 1130                 if ((events & select_flags[msk]) == 0)
 1131                         continue;
 1132                 if (ibits[msk] == NULL)
 1133                         continue;
 1134                 if ((ibits[msk][idx] & bit) == 0)
 1135                         continue;
 1136                 /*
 1137                  * XXX Check for a duplicate set.  This can occur because a
 1138                  * socket calls selrecord() twice for each poll() call
 1139                  * resulting in two selfds per real fd.  selrescan() will
 1140                  * call selsetbits twice as a result.
 1141                  */
 1142                 if ((obits[msk][idx] & bit) != 0)
 1143                         continue;
 1144                 obits[msk][idx] |= bit;
 1145                 n++;
 1146         }
 1147 
 1148         return (n);
 1149 }
 1150 
 1151 static __inline int
 1152 getselfd_cap(struct filedesc *fdp, int fd, struct file **fpp)
 1153 {
 1154         struct file *fp;
 1155 #ifdef CAPABILITIES
 1156         struct file *fp_fromcap;
 1157         int error;
 1158 #endif
 1159 
 1160         if ((fp = fget_unlocked(fdp, fd)) == NULL)
 1161                 return (EBADF);
 1162 #ifdef CAPABILITIES
 1163         /*
 1164          * If the file descriptor is for a capability, test rights and use
 1165          * the file descriptor references by the capability.
 1166          */
 1167         error = cap_funwrap(fp, CAP_POLL_EVENT, &fp_fromcap);
 1168         if (error) {
 1169                 fdrop(fp, curthread);
 1170                 return (error);
 1171         }
 1172         if (fp != fp_fromcap) {
 1173                 fhold(fp_fromcap);
 1174                 fdrop(fp, curthread);
 1175                 fp = fp_fromcap;
 1176         }
 1177 #endif /* CAPABILITIES */
 1178         *fpp = fp;
 1179         return (0);
 1180 }
 1181 
 1182 /*
 1183  * Traverse the list of fds attached to this thread's seltd and check for
 1184  * completion.
 1185  */
 1186 static int
 1187 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
 1188 {
 1189         struct filedesc *fdp;
 1190         struct selinfo *si;
 1191         struct seltd *stp;
 1192         struct selfd *sfp;
 1193         struct selfd *sfn;
 1194         struct file *fp;
 1195         fd_mask bit;
 1196         int fd, ev, n, idx;
 1197         int error;
 1198 
 1199         fdp = td->td_proc->p_fd;
 1200         stp = td->td_sel;
 1201         n = 0;
 1202         STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
 1203                 fd = (int)(uintptr_t)sfp->sf_cookie;
 1204                 si = sfp->sf_si;
 1205                 selfdfree(stp, sfp);
 1206                 /* If the selinfo wasn't cleared the event didn't fire. */
 1207                 if (si != NULL)
 1208                         continue;
 1209                 error = getselfd_cap(fdp, fd, &fp);
 1210                 if (error)
 1211                         return (error);
 1212                 idx = fd / NFDBITS;
 1213                 bit = (fd_mask)1 << (fd % NFDBITS);
 1214                 ev = fo_poll(fp, selflags(ibits, idx, bit), td->td_ucred, td);
 1215                 fdrop(fp, td);
 1216                 if (ev != 0)
 1217                         n += selsetbits(ibits, obits, idx, bit, ev);
 1218         }
 1219         stp->st_flags = 0;
 1220         td->td_retval[0] = n;
 1221         return (0);
 1222 }
 1223 
 1224 /*
 1225  * Perform the initial filedescriptor scan and register ourselves with
 1226  * each selinfo.
 1227  */
 1228 static int
 1229 selscan(td, ibits, obits, nfd)
 1230         struct thread *td;
 1231         fd_mask **ibits, **obits;
 1232         int nfd;
 1233 {
 1234         struct filedesc *fdp;
 1235         struct file *fp;
 1236         fd_mask bit;
 1237         int ev, flags, end, fd;
 1238         int n, idx;
 1239         int error;
 1240 
 1241         fdp = td->td_proc->p_fd;
 1242         n = 0;
 1243         for (idx = 0, fd = 0; fd < nfd; idx++) {
 1244                 end = imin(fd + NFDBITS, nfd);
 1245                 for (bit = 1; fd < end; bit <<= 1, fd++) {
 1246                         /* Compute the list of events we're interested in. */
 1247                         flags = selflags(ibits, idx, bit);
 1248                         if (flags == 0)
 1249                                 continue;
 1250                         error = getselfd_cap(fdp, fd, &fp);
 1251                         if (error)
 1252                                 return (error);
 1253                         selfdalloc(td, (void *)(uintptr_t)fd);
 1254                         ev = fo_poll(fp, flags, td->td_ucred, td);
 1255                         fdrop(fp, td);
 1256                         if (ev != 0)
 1257                                 n += selsetbits(ibits, obits, idx, bit, ev);
 1258                 }
 1259         }
 1260 
 1261         td->td_retval[0] = n;
 1262         return (0);
 1263 }
 1264 
 1265 #ifndef _SYS_SYSPROTO_H_
 1266 struct poll_args {
 1267         struct pollfd *fds;
 1268         u_int   nfds;
 1269         int     timeout;
 1270 };
 1271 #endif
 1272 int
 1273 sys_poll(td, uap)
 1274         struct thread *td;
 1275         struct poll_args *uap;
 1276 {
 1277         struct pollfd *bits;
 1278         struct pollfd smallbits[32];
 1279         struct timeval atv, rtv, ttv;
 1280         int error, timo;
 1281         u_int nfds;
 1282         size_t ni;
 1283 
 1284         nfds = uap->nfds;
 1285         if (nfds > maxfilesperproc && nfds > FD_SETSIZE) 
 1286                 return (EINVAL);
 1287         ni = nfds * sizeof(struct pollfd);
 1288         if (ni > sizeof(smallbits))
 1289                 bits = malloc(ni, M_TEMP, M_WAITOK);
 1290         else
 1291                 bits = smallbits;
 1292         error = copyin(uap->fds, bits, ni);
 1293         if (error)
 1294                 goto done;
 1295         if (uap->timeout != INFTIM) {
 1296                 atv.tv_sec = uap->timeout / 1000;
 1297                 atv.tv_usec = (uap->timeout % 1000) * 1000;
 1298                 if (itimerfix(&atv)) {
 1299                         error = EINVAL;
 1300                         goto done;
 1301                 }
 1302                 getmicrouptime(&rtv);
 1303                 timevaladd(&atv, &rtv);
 1304         } else {
 1305                 atv.tv_sec = 0;
 1306                 atv.tv_usec = 0;
 1307         }
 1308         timo = 0;
 1309         seltdinit(td);
 1310         /* Iterate until the timeout expires or descriptors become ready. */
 1311         for (;;) {
 1312                 error = pollscan(td, bits, nfds);
 1313                 if (error || td->td_retval[0] != 0)
 1314                         break;
 1315                 if (atv.tv_sec || atv.tv_usec) {
 1316                         getmicrouptime(&rtv);
 1317                         if (timevalcmp(&rtv, &atv, >=))
 1318                                 break;
 1319                         ttv = atv;
 1320                         timevalsub(&ttv, &rtv);
 1321                         timo = ttv.tv_sec > 24 * 60 * 60 ?
 1322                             24 * 60 * 60 * hz : tvtohz(&ttv);
 1323                 }
 1324                 error = seltdwait(td, timo);
 1325                 if (error)
 1326                         break;
 1327                 error = pollrescan(td);
 1328                 if (error || td->td_retval[0] != 0)
 1329                         break;
 1330         }
 1331         seltdclear(td);
 1332 
 1333 done:
 1334         /* poll is not restarted after signals... */
 1335         if (error == ERESTART)
 1336                 error = EINTR;
 1337         if (error == EWOULDBLOCK)
 1338                 error = 0;
 1339         if (error == 0) {
 1340                 error = pollout(td, bits, uap->fds, nfds);
 1341                 if (error)
 1342                         goto out;
 1343         }
 1344 out:
 1345         if (ni > sizeof(smallbits))
 1346                 free(bits, M_TEMP);
 1347         return (error);
 1348 }
 1349 
 1350 static int
 1351 pollrescan(struct thread *td)
 1352 {
 1353         struct seltd *stp;
 1354         struct selfd *sfp;
 1355         struct selfd *sfn;
 1356         struct selinfo *si;
 1357         struct filedesc *fdp;
 1358         struct file *fp;
 1359         struct pollfd *fd;
 1360         int n;
 1361 
 1362         n = 0;
 1363         fdp = td->td_proc->p_fd;
 1364         stp = td->td_sel;
 1365         FILEDESC_SLOCK(fdp);
 1366         STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
 1367                 fd = (struct pollfd *)sfp->sf_cookie;
 1368                 si = sfp->sf_si;
 1369                 selfdfree(stp, sfp);
 1370                 /* If the selinfo wasn't cleared the event didn't fire. */
 1371                 if (si != NULL)
 1372                         continue;
 1373                 fp = fdp->fd_ofiles[fd->fd];
 1374 #ifdef CAPABILITIES
 1375                 if ((fp == NULL)
 1376                     || (cap_funwrap(fp, CAP_POLL_EVENT, &fp) != 0)) {
 1377 #else
 1378                 if (fp == NULL) {
 1379 #endif
 1380                         fd->revents = POLLNVAL;
 1381                         n++;
 1382                         continue;
 1383                 }
 1384 
 1385                 /*
 1386                  * Note: backend also returns POLLHUP and
 1387                  * POLLERR if appropriate.
 1388                  */
 1389                 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
 1390                 if (fd->revents != 0)
 1391                         n++;
 1392         }
 1393         FILEDESC_SUNLOCK(fdp);
 1394         stp->st_flags = 0;
 1395         td->td_retval[0] = n;
 1396         return (0);
 1397 }
 1398 
 1399 
 1400 static int
 1401 pollout(td, fds, ufds, nfd)
 1402         struct thread *td;
 1403         struct pollfd *fds;
 1404         struct pollfd *ufds;
 1405         u_int nfd;
 1406 {
 1407         int error = 0;
 1408         u_int i = 0;
 1409         u_int n = 0;
 1410 
 1411         for (i = 0; i < nfd; i++) {
 1412                 error = copyout(&fds->revents, &ufds->revents,
 1413                     sizeof(ufds->revents));
 1414                 if (error)
 1415                         return (error);
 1416                 if (fds->revents != 0)
 1417                         n++;
 1418                 fds++;
 1419                 ufds++;
 1420         }
 1421         td->td_retval[0] = n;
 1422         return (0);
 1423 }
 1424 
 1425 static int
 1426 pollscan(td, fds, nfd)
 1427         struct thread *td;
 1428         struct pollfd *fds;
 1429         u_int nfd;
 1430 {
 1431         struct filedesc *fdp = td->td_proc->p_fd;
 1432         int i;
 1433         struct file *fp;
 1434         int n = 0;
 1435 
 1436         FILEDESC_SLOCK(fdp);
 1437         for (i = 0; i < nfd; i++, fds++) {
 1438                 if (fds->fd >= fdp->fd_nfiles) {
 1439                         fds->revents = POLLNVAL;
 1440                         n++;
 1441                 } else if (fds->fd < 0) {
 1442                         fds->revents = 0;
 1443                 } else {
 1444                         fp = fdp->fd_ofiles[fds->fd];
 1445 #ifdef CAPABILITIES
 1446                         if ((fp == NULL)
 1447                             || (cap_funwrap(fp, CAP_POLL_EVENT, &fp) != 0)) {
 1448 #else
 1449                         if (fp == NULL) {
 1450 #endif
 1451                                 fds->revents = POLLNVAL;
 1452                                 n++;
 1453                         } else {
 1454                                 /*
 1455                                  * Note: backend also returns POLLHUP and
 1456                                  * POLLERR if appropriate.
 1457                                  */
 1458                                 selfdalloc(td, fds);
 1459                                 fds->revents = fo_poll(fp, fds->events,
 1460                                     td->td_ucred, td);
 1461                                 /*
 1462                                  * POSIX requires POLLOUT to be never
 1463                                  * set simultaneously with POLLHUP.
 1464                                  */
 1465                                 if ((fds->revents & POLLHUP) != 0)
 1466                                         fds->revents &= ~POLLOUT;
 1467 
 1468                                 if (fds->revents != 0)
 1469                                         n++;
 1470                         }
 1471                 }
 1472         }
 1473         FILEDESC_SUNLOCK(fdp);
 1474         td->td_retval[0] = n;
 1475         return (0);
 1476 }
 1477 
 1478 /*
 1479  * OpenBSD poll system call.
 1480  *
 1481  * XXX this isn't quite a true representation..  OpenBSD uses select ops.
 1482  */
 1483 #ifndef _SYS_SYSPROTO_H_
 1484 struct openbsd_poll_args {
 1485         struct pollfd *fds;
 1486         u_int   nfds;
 1487         int     timeout;
 1488 };
 1489 #endif
 1490 int
 1491 sys_openbsd_poll(td, uap)
 1492         register struct thread *td;
 1493         register struct openbsd_poll_args *uap;
 1494 {
 1495         return (sys_poll(td, (struct poll_args *)uap));
 1496 }
 1497 
 1498 /*
 1499  * XXX This was created specifically to support netncp and netsmb.  This
 1500  * allows the caller to specify a socket to wait for events on.  It returns
 1501  * 0 if any events matched and an error otherwise.  There is no way to
 1502  * determine which events fired.
 1503  */
 1504 int
 1505 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
 1506 {
 1507         struct timeval atv, rtv, ttv;
 1508         int error, timo;
 1509 
 1510         if (tvp != NULL) {
 1511                 atv = *tvp;
 1512                 if (itimerfix(&atv))
 1513                         return (EINVAL);
 1514                 getmicrouptime(&rtv);
 1515                 timevaladd(&atv, &rtv);
 1516         } else {
 1517                 atv.tv_sec = 0;
 1518                 atv.tv_usec = 0;
 1519         }
 1520 
 1521         timo = 0;
 1522         seltdinit(td);
 1523         /*
 1524          * Iterate until the timeout expires or the socket becomes ready.
 1525          */
 1526         for (;;) {
 1527                 selfdalloc(td, NULL);
 1528                 error = sopoll(so, events, NULL, td);
 1529                 /* error here is actually the ready events. */
 1530                 if (error)
 1531                         return (0);
 1532                 if (atv.tv_sec || atv.tv_usec) {
 1533                         getmicrouptime(&rtv);
 1534                         if (timevalcmp(&rtv, &atv, >=)) {
 1535                                 seltdclear(td);
 1536                                 return (EWOULDBLOCK);
 1537                         }
 1538                         ttv = atv;
 1539                         timevalsub(&ttv, &rtv);
 1540                         timo = ttv.tv_sec > 24 * 60 * 60 ?
 1541                             24 * 60 * 60 * hz : tvtohz(&ttv);
 1542                 }
 1543                 error = seltdwait(td, timo);
 1544                 seltdclear(td);
 1545                 if (error)
 1546                         break;
 1547         }
 1548         /* XXX Duplicates ncp/smb behavior. */
 1549         if (error == ERESTART)
 1550                 error = 0;
 1551         return (error);
 1552 }
 1553 
 1554 /*
 1555  * Preallocate two selfds associated with 'cookie'.  Some fo_poll routines
 1556  * have two select sets, one for read and another for write.
 1557  */
 1558 static void
 1559 selfdalloc(struct thread *td, void *cookie)
 1560 {
 1561         struct seltd *stp;
 1562 
 1563         stp = td->td_sel;
 1564         if (stp->st_free1 == NULL)
 1565                 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
 1566         stp->st_free1->sf_td = stp;
 1567         stp->st_free1->sf_cookie = cookie;
 1568         if (stp->st_free2 == NULL)
 1569                 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
 1570         stp->st_free2->sf_td = stp;
 1571         stp->st_free2->sf_cookie = cookie;
 1572 }
 1573 
 1574 static void
 1575 selfdfree(struct seltd *stp, struct selfd *sfp)
 1576 {
 1577         STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
 1578         mtx_lock(sfp->sf_mtx);
 1579         if (sfp->sf_si)
 1580                 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
 1581         mtx_unlock(sfp->sf_mtx);
 1582         uma_zfree(selfd_zone, sfp);
 1583 }
 1584 
 1585 /* Drain the waiters tied to all the selfd belonging the specified selinfo. */
 1586 void
 1587 seldrain(sip)
 1588         struct selinfo *sip;
 1589 {
 1590 
 1591         /*
 1592          * This feature is already provided by doselwakeup(), thus it is
 1593          * enough to go for it.
 1594          * Eventually, the context, should take care to avoid races
 1595          * between thread calling select()/poll() and file descriptor
 1596          * detaching, but, again, the races are just the same as
 1597          * selwakeup().
 1598          */
 1599         doselwakeup(sip, -1);
 1600 }
 1601 
 1602 /*
 1603  * Record a select request.
 1604  */
 1605 void
 1606 selrecord(selector, sip)
 1607         struct thread *selector;
 1608         struct selinfo *sip;
 1609 {
 1610         struct selfd *sfp;
 1611         struct seltd *stp;
 1612         struct mtx *mtxp;
 1613 
 1614         stp = selector->td_sel;
 1615         /*
 1616          * Don't record when doing a rescan.
 1617          */
 1618         if (stp->st_flags & SELTD_RESCAN)
 1619                 return;
 1620         /*
 1621          * Grab one of the preallocated descriptors.
 1622          */
 1623         sfp = NULL;
 1624         if ((sfp = stp->st_free1) != NULL)
 1625                 stp->st_free1 = NULL;
 1626         else if ((sfp = stp->st_free2) != NULL)
 1627                 stp->st_free2 = NULL;
 1628         else
 1629                 panic("selrecord: No free selfd on selq");
 1630         mtxp = sip->si_mtx;
 1631         if (mtxp == NULL)
 1632                 mtxp = mtx_pool_find(mtxpool_select, sip);
 1633         /*
 1634          * Initialize the sfp and queue it in the thread.
 1635          */
 1636         sfp->sf_si = sip;
 1637         sfp->sf_mtx = mtxp;
 1638         STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
 1639         /*
 1640          * Now that we've locked the sip, check for initialization.
 1641          */
 1642         mtx_lock(mtxp);
 1643         if (sip->si_mtx == NULL) {
 1644                 sip->si_mtx = mtxp;
 1645                 TAILQ_INIT(&sip->si_tdlist);
 1646         }
 1647         /*
 1648          * Add this thread to the list of selfds listening on this selinfo.
 1649          */
 1650         TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
 1651         mtx_unlock(sip->si_mtx);
 1652 }
 1653 
 1654 /* Wake up a selecting thread. */
 1655 void
 1656 selwakeup(sip)
 1657         struct selinfo *sip;
 1658 {
 1659         doselwakeup(sip, -1);
 1660 }
 1661 
 1662 /* Wake up a selecting thread, and set its priority. */
 1663 void
 1664 selwakeuppri(sip, pri)
 1665         struct selinfo *sip;
 1666         int pri;
 1667 {
 1668         doselwakeup(sip, pri);
 1669 }
 1670 
 1671 /*
 1672  * Do a wakeup when a selectable event occurs.
 1673  */
 1674 static void
 1675 doselwakeup(sip, pri)
 1676         struct selinfo *sip;
 1677         int pri;
 1678 {
 1679         struct selfd *sfp;
 1680         struct selfd *sfn;
 1681         struct seltd *stp;
 1682 
 1683         /* If it's not initialized there can't be any waiters. */
 1684         if (sip->si_mtx == NULL)
 1685                 return;
 1686         /*
 1687          * Locking the selinfo locks all selfds associated with it.
 1688          */
 1689         mtx_lock(sip->si_mtx);
 1690         TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
 1691                 /*
 1692                  * Once we remove this sfp from the list and clear the
 1693                  * sf_si seltdclear will know to ignore this si.
 1694                  */
 1695                 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
 1696                 sfp->sf_si = NULL;
 1697                 stp = sfp->sf_td;
 1698                 mtx_lock(&stp->st_mtx);
 1699                 stp->st_flags |= SELTD_PENDING;
 1700                 cv_broadcastpri(&stp->st_wait, pri);
 1701                 mtx_unlock(&stp->st_mtx);
 1702         }
 1703         mtx_unlock(sip->si_mtx);
 1704 }
 1705 
 1706 static void
 1707 seltdinit(struct thread *td)
 1708 {
 1709         struct seltd *stp;
 1710 
 1711         if ((stp = td->td_sel) != NULL)
 1712                 goto out;
 1713         td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
 1714         mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
 1715         cv_init(&stp->st_wait, "select");
 1716 out:
 1717         stp->st_flags = 0;
 1718         STAILQ_INIT(&stp->st_selq);
 1719 }
 1720 
 1721 static int
 1722 seltdwait(struct thread *td, int timo)
 1723 {
 1724         struct seltd *stp;
 1725         int error;
 1726 
 1727         stp = td->td_sel;
 1728         /*
 1729          * An event of interest may occur while we do not hold the seltd
 1730          * locked so check the pending flag before we sleep.
 1731          */
 1732         mtx_lock(&stp->st_mtx);
 1733         /*
 1734          * Any further calls to selrecord will be a rescan.
 1735          */
 1736         stp->st_flags |= SELTD_RESCAN;
 1737         if (stp->st_flags & SELTD_PENDING) {
 1738                 mtx_unlock(&stp->st_mtx);
 1739                 return (0);
 1740         }
 1741         if (timo > 0)
 1742                 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
 1743         else
 1744                 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
 1745         mtx_unlock(&stp->st_mtx);
 1746 
 1747         return (error);
 1748 }
 1749 
 1750 void
 1751 seltdfini(struct thread *td)
 1752 {
 1753         struct seltd *stp;
 1754 
 1755         stp = td->td_sel;
 1756         if (stp == NULL)
 1757                 return;
 1758         if (stp->st_free1)
 1759                 uma_zfree(selfd_zone, stp->st_free1);
 1760         if (stp->st_free2)
 1761                 uma_zfree(selfd_zone, stp->st_free2);
 1762         td->td_sel = NULL;
 1763         free(stp, M_SELECT);
 1764 }
 1765 
 1766 /*
 1767  * Remove the references to the thread from all of the objects we were
 1768  * polling.
 1769  */
 1770 static void
 1771 seltdclear(struct thread *td)
 1772 {
 1773         struct seltd *stp;
 1774         struct selfd *sfp;
 1775         struct selfd *sfn;
 1776 
 1777         stp = td->td_sel;
 1778         STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
 1779                 selfdfree(stp, sfp);
 1780         stp->st_flags = 0;
 1781 }
 1782 
 1783 static void selectinit(void *);
 1784 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
 1785 static void
 1786 selectinit(void *dummy __unused)
 1787 {
 1788 
 1789         selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
 1790             NULL, NULL, UMA_ALIGN_PTR, 0);
 1791         mtxpool_select = mtx_pool_create("select mtxpool", 128, MTX_DEF);
 1792 }

Cache object: bc4ce937c4ddb0bccc79418e95ddf129


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.