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/vfs_vnops.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
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  * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
   11  * Copyright (c) 2013, 2014 The FreeBSD Foundation
   12  *
   13  * Portions of this software were developed by Konstantin Belousov
   14  * under sponsorship from the FreeBSD Foundation.
   15  *
   16  * Redistribution and use in source and binary forms, with or without
   17  * modification, are permitted provided that the following conditions
   18  * are met:
   19  * 1. Redistributions of source code must retain the above copyright
   20  *    notice, this list of conditions and the following disclaimer.
   21  * 2. Redistributions in binary form must reproduce the above copyright
   22  *    notice, this list of conditions and the following disclaimer in the
   23  *    documentation and/or other materials provided with the distribution.
   24  * 4. Neither the name of the University nor the names of its contributors
   25  *    may be used to endorse or promote products derived from this software
   26  *    without specific prior written permission.
   27  *
   28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   38  * SUCH DAMAGE.
   39  *
   40  *      @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
   41  */
   42 
   43 #include <sys/cdefs.h>
   44 __FBSDID("$FreeBSD: releng/10.3/sys/kern/vfs_vnops.c 295614 2016-02-14 18:17:58Z kib $");
   45 
   46 #include <sys/param.h>
   47 #include <sys/systm.h>
   48 #include <sys/disk.h>
   49 #include <sys/fcntl.h>
   50 #include <sys/file.h>
   51 #include <sys/kdb.h>
   52 #include <sys/stat.h>
   53 #include <sys/priv.h>
   54 #include <sys/proc.h>
   55 #include <sys/limits.h>
   56 #include <sys/lock.h>
   57 #include <sys/mount.h>
   58 #include <sys/mutex.h>
   59 #include <sys/namei.h>
   60 #include <sys/vnode.h>
   61 #include <sys/bio.h>
   62 #include <sys/buf.h>
   63 #include <sys/filio.h>
   64 #include <sys/resourcevar.h>
   65 #include <sys/rwlock.h>
   66 #include <sys/sx.h>
   67 #include <sys/sysctl.h>
   68 #include <sys/ttycom.h>
   69 #include <sys/conf.h>
   70 #include <sys/syslog.h>
   71 #include <sys/unistd.h>
   72 
   73 #include <security/audit/audit.h>
   74 #include <security/mac/mac_framework.h>
   75 
   76 #include <vm/vm.h>
   77 #include <vm/vm_extern.h>
   78 #include <vm/pmap.h>
   79 #include <vm/vm_map.h>
   80 #include <vm/vm_object.h>
   81 #include <vm/vm_page.h>
   82 
   83 static fo_rdwr_t        vn_read;
   84 static fo_rdwr_t        vn_write;
   85 static fo_rdwr_t        vn_io_fault;
   86 static fo_truncate_t    vn_truncate;
   87 static fo_ioctl_t       vn_ioctl;
   88 static fo_poll_t        vn_poll;
   89 static fo_kqfilter_t    vn_kqfilter;
   90 static fo_stat_t        vn_statfile;
   91 static fo_close_t       vn_closefile;
   92 
   93 struct  fileops vnops = {
   94         .fo_read = vn_io_fault,
   95         .fo_write = vn_io_fault,
   96         .fo_truncate = vn_truncate,
   97         .fo_ioctl = vn_ioctl,
   98         .fo_poll = vn_poll,
   99         .fo_kqfilter = vn_kqfilter,
  100         .fo_stat = vn_statfile,
  101         .fo_close = vn_closefile,
  102         .fo_chmod = vn_chmod,
  103         .fo_chown = vn_chown,
  104         .fo_sendfile = vn_sendfile,
  105         .fo_seek = vn_seek,
  106         .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
  107 };
  108 
  109 static const int io_hold_cnt = 16;
  110 static int vn_io_fault_enable = 1;
  111 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
  112     &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
  113 static int vn_io_fault_prefault = 0;
  114 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
  115     &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
  116 static u_long vn_io_faults_cnt;
  117 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
  118     &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
  119 
  120 /*
  121  * Returns true if vn_io_fault mode of handling the i/o request should
  122  * be used.
  123  */
  124 static bool
  125 do_vn_io_fault(struct vnode *vp, struct uio *uio)
  126 {
  127         struct mount *mp;
  128 
  129         return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
  130             (mp = vp->v_mount) != NULL &&
  131             (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
  132 }
  133 
  134 /*
  135  * Structure used to pass arguments to vn_io_fault1(), to do either
  136  * file- or vnode-based I/O calls.
  137  */
  138 struct vn_io_fault_args {
  139         enum {
  140                 VN_IO_FAULT_FOP,
  141                 VN_IO_FAULT_VOP
  142         } kind;
  143         struct ucred *cred;
  144         int flags;
  145         union {
  146                 struct fop_args_tag {
  147                         struct file *fp;
  148                         fo_rdwr_t *doio;
  149                 } fop_args;
  150                 struct vop_args_tag {
  151                         struct vnode *vp;
  152                 } vop_args;
  153         } args;
  154 };
  155 
  156 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
  157     struct vn_io_fault_args *args, struct thread *td);
  158 
  159 int
  160 vn_open(ndp, flagp, cmode, fp)
  161         struct nameidata *ndp;
  162         int *flagp, cmode;
  163         struct file *fp;
  164 {
  165         struct thread *td = ndp->ni_cnd.cn_thread;
  166 
  167         return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
  168 }
  169 
  170 /*
  171  * Common code for vnode open operations via a name lookup.
  172  * Lookup the vnode and invoke VOP_CREATE if needed.
  173  * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
  174  * 
  175  * Note that this does NOT free nameidata for the successful case,
  176  * due to the NDINIT being done elsewhere.
  177  */
  178 int
  179 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
  180     struct ucred *cred, struct file *fp)
  181 {
  182         struct vnode *vp;
  183         struct mount *mp;
  184         struct thread *td = ndp->ni_cnd.cn_thread;
  185         struct vattr vat;
  186         struct vattr *vap = &vat;
  187         int fmode, error;
  188 
  189 restart:
  190         fmode = *flagp;
  191         if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
  192             O_EXCL | O_DIRECTORY))
  193                 return (EINVAL);
  194         else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
  195                 ndp->ni_cnd.cn_nameiop = CREATE;
  196                 /*
  197                  * Set NOCACHE to avoid flushing the cache when
  198                  * rolling in many files at once.
  199                 */
  200                 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
  201                 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
  202                         ndp->ni_cnd.cn_flags |= FOLLOW;
  203                 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
  204                         ndp->ni_cnd.cn_flags |= AUDITVNODE1;
  205                 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
  206                         ndp->ni_cnd.cn_flags |= NOCAPCHECK;
  207                 bwillwrite();
  208                 if ((error = namei(ndp)) != 0)
  209                         return (error);
  210                 if (ndp->ni_vp == NULL) {
  211                         VATTR_NULL(vap);
  212                         vap->va_type = VREG;
  213                         vap->va_mode = cmode;
  214                         if (fmode & O_EXCL)
  215                                 vap->va_vaflags |= VA_EXCLUSIVE;
  216                         if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
  217                                 NDFREE(ndp, NDF_ONLY_PNBUF);
  218                                 vput(ndp->ni_dvp);
  219                                 if ((error = vn_start_write(NULL, &mp,
  220                                     V_XSLEEP | PCATCH)) != 0)
  221                                         return (error);
  222                                 goto restart;
  223                         }
  224                         if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
  225                                 ndp->ni_cnd.cn_flags |= MAKEENTRY;
  226 #ifdef MAC
  227                         error = mac_vnode_check_create(cred, ndp->ni_dvp,
  228                             &ndp->ni_cnd, vap);
  229                         if (error == 0)
  230 #endif
  231                                 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
  232                                                    &ndp->ni_cnd, vap);
  233                         vput(ndp->ni_dvp);
  234                         vn_finished_write(mp);
  235                         if (error) {
  236                                 NDFREE(ndp, NDF_ONLY_PNBUF);
  237                                 return (error);
  238                         }
  239                         fmode &= ~O_TRUNC;
  240                         vp = ndp->ni_vp;
  241                 } else {
  242                         if (ndp->ni_dvp == ndp->ni_vp)
  243                                 vrele(ndp->ni_dvp);
  244                         else
  245                                 vput(ndp->ni_dvp);
  246                         ndp->ni_dvp = NULL;
  247                         vp = ndp->ni_vp;
  248                         if (fmode & O_EXCL) {
  249                                 error = EEXIST;
  250                                 goto bad;
  251                         }
  252                         fmode &= ~O_CREAT;
  253                 }
  254         } else {
  255                 ndp->ni_cnd.cn_nameiop = LOOKUP;
  256                 ndp->ni_cnd.cn_flags = ISOPEN |
  257                     ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
  258                 if (!(fmode & FWRITE))
  259                         ndp->ni_cnd.cn_flags |= LOCKSHARED;
  260                 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
  261                         ndp->ni_cnd.cn_flags |= AUDITVNODE1;
  262                 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
  263                         ndp->ni_cnd.cn_flags |= NOCAPCHECK;
  264                 if ((error = namei(ndp)) != 0)
  265                         return (error);
  266                 vp = ndp->ni_vp;
  267         }
  268         error = vn_open_vnode(vp, fmode, cred, td, fp);
  269         if (error)
  270                 goto bad;
  271         *flagp = fmode;
  272         return (0);
  273 bad:
  274         NDFREE(ndp, NDF_ONLY_PNBUF);
  275         vput(vp);
  276         *flagp = fmode;
  277         ndp->ni_vp = NULL;
  278         return (error);
  279 }
  280 
  281 /*
  282  * Common code for vnode open operations once a vnode is located.
  283  * Check permissions, and call the VOP_OPEN routine.
  284  */
  285 int
  286 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
  287     struct thread *td, struct file *fp)
  288 {
  289         struct mount *mp;
  290         accmode_t accmode;
  291         struct flock lf;
  292         int error, have_flock, lock_flags, type;
  293 
  294         if (vp->v_type == VLNK)
  295                 return (EMLINK);
  296         if (vp->v_type == VSOCK)
  297                 return (EOPNOTSUPP);
  298         if (vp->v_type != VDIR && fmode & O_DIRECTORY)
  299                 return (ENOTDIR);
  300         accmode = 0;
  301         if (fmode & (FWRITE | O_TRUNC)) {
  302                 if (vp->v_type == VDIR)
  303                         return (EISDIR);
  304                 accmode |= VWRITE;
  305         }
  306         if (fmode & FREAD)
  307                 accmode |= VREAD;
  308         if (fmode & FEXEC)
  309                 accmode |= VEXEC;
  310         if ((fmode & O_APPEND) && (fmode & FWRITE))
  311                 accmode |= VAPPEND;
  312 #ifdef MAC
  313         error = mac_vnode_check_open(cred, vp, accmode);
  314         if (error)
  315                 return (error);
  316 #endif
  317         if ((fmode & O_CREAT) == 0) {
  318                 if (accmode & VWRITE) {
  319                         error = vn_writechk(vp);
  320                         if (error)
  321                                 return (error);
  322                 }
  323                 if (accmode) {
  324                         error = VOP_ACCESS(vp, accmode, cred, td);
  325                         if (error)
  326                                 return (error);
  327                 }
  328         }
  329         if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
  330                 vn_lock(vp, LK_UPGRADE | LK_RETRY);
  331         if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
  332                 return (error);
  333 
  334         if (fmode & (O_EXLOCK | O_SHLOCK)) {
  335                 KASSERT(fp != NULL, ("open with flock requires fp"));
  336                 lock_flags = VOP_ISLOCKED(vp);
  337                 VOP_UNLOCK(vp, 0);
  338                 lf.l_whence = SEEK_SET;
  339                 lf.l_start = 0;
  340                 lf.l_len = 0;
  341                 if (fmode & O_EXLOCK)
  342                         lf.l_type = F_WRLCK;
  343                 else
  344                         lf.l_type = F_RDLCK;
  345                 type = F_FLOCK;
  346                 if ((fmode & FNONBLOCK) == 0)
  347                         type |= F_WAIT;
  348                 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
  349                 have_flock = (error == 0);
  350                 vn_lock(vp, lock_flags | LK_RETRY);
  351                 if (error == 0 && vp->v_iflag & VI_DOOMED)
  352                         error = ENOENT;
  353                 /*
  354                  * Another thread might have used this vnode as an
  355                  * executable while the vnode lock was dropped.
  356                  * Ensure the vnode is still able to be opened for
  357                  * writing after the lock has been obtained.
  358                  */
  359                 if (error == 0 && accmode & VWRITE)
  360                         error = vn_writechk(vp);
  361                 if (error) {
  362                         VOP_UNLOCK(vp, 0);
  363                         if (have_flock) {
  364                                 lf.l_whence = SEEK_SET;
  365                                 lf.l_start = 0;
  366                                 lf.l_len = 0;
  367                                 lf.l_type = F_UNLCK;
  368                                 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
  369                                     F_FLOCK);
  370                         }
  371                         vn_start_write(vp, &mp, V_WAIT);
  372                         vn_lock(vp, lock_flags | LK_RETRY);
  373                         (void)VOP_CLOSE(vp, fmode, cred, td);
  374                         vn_finished_write(mp);
  375                         /* Prevent second close from fdrop()->vn_close(). */
  376                         if (fp != NULL)
  377                                 fp->f_ops= &badfileops;
  378                         return (error);
  379                 }
  380                 fp->f_flag |= FHASLOCK;
  381         }
  382         if (fmode & FWRITE) {
  383                 VOP_ADD_WRITECOUNT(vp, 1);
  384                 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
  385                     __func__, vp, vp->v_writecount);
  386         }
  387         ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
  388         return (0);
  389 }
  390 
  391 /*
  392  * Check for write permissions on the specified vnode.
  393  * Prototype text segments cannot be written.
  394  */
  395 int
  396 vn_writechk(vp)
  397         register struct vnode *vp;
  398 {
  399 
  400         ASSERT_VOP_LOCKED(vp, "vn_writechk");
  401         /*
  402          * If there's shared text associated with
  403          * the vnode, try to free it up once.  If
  404          * we fail, we can't allow writing.
  405          */
  406         if (VOP_IS_TEXT(vp))
  407                 return (ETXTBSY);
  408 
  409         return (0);
  410 }
  411 
  412 /*
  413  * Vnode close call
  414  */
  415 int
  416 vn_close(vp, flags, file_cred, td)
  417         register struct vnode *vp;
  418         int flags;
  419         struct ucred *file_cred;
  420         struct thread *td;
  421 {
  422         struct mount *mp;
  423         int error, lock_flags;
  424 
  425         if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
  426             MNT_EXTENDED_SHARED(vp->v_mount))
  427                 lock_flags = LK_SHARED;
  428         else
  429                 lock_flags = LK_EXCLUSIVE;
  430 
  431         vn_start_write(vp, &mp, V_WAIT);
  432         vn_lock(vp, lock_flags | LK_RETRY);
  433         if (flags & FWRITE) {
  434                 VNASSERT(vp->v_writecount > 0, vp, 
  435                     ("vn_close: negative writecount"));
  436                 VOP_ADD_WRITECOUNT(vp, -1);
  437                 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
  438                     __func__, vp, vp->v_writecount);
  439         }
  440         error = VOP_CLOSE(vp, flags, file_cred, td);
  441         vput(vp);
  442         vn_finished_write(mp);
  443         return (error);
  444 }
  445 
  446 /*
  447  * Heuristic to detect sequential operation.
  448  */
  449 static int
  450 sequential_heuristic(struct uio *uio, struct file *fp)
  451 {
  452 
  453         ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
  454         if (fp->f_flag & FRDAHEAD)
  455                 return (fp->f_seqcount << IO_SEQSHIFT);
  456 
  457         /*
  458          * Offset 0 is handled specially.  open() sets f_seqcount to 1 so
  459          * that the first I/O is normally considered to be slightly
  460          * sequential.  Seeking to offset 0 doesn't change sequentiality
  461          * unless previous seeks have reduced f_seqcount to 0, in which
  462          * case offset 0 is not special.
  463          */
  464         if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
  465             uio->uio_offset == fp->f_nextoff) {
  466                 /*
  467                  * f_seqcount is in units of fixed-size blocks so that it
  468                  * depends mainly on the amount of sequential I/O and not
  469                  * much on the number of sequential I/O's.  The fixed size
  470                  * of 16384 is hard-coded here since it is (not quite) just
  471                  * a magic size that works well here.  This size is more
  472                  * closely related to the best I/O size for real disks than
  473                  * to any block size used by software.
  474                  */
  475                 fp->f_seqcount += howmany(uio->uio_resid, 16384);
  476                 if (fp->f_seqcount > IO_SEQMAX)
  477                         fp->f_seqcount = IO_SEQMAX;
  478                 return (fp->f_seqcount << IO_SEQSHIFT);
  479         }
  480 
  481         /* Not sequential.  Quickly draw-down sequentiality. */
  482         if (fp->f_seqcount > 1)
  483                 fp->f_seqcount = 1;
  484         else
  485                 fp->f_seqcount = 0;
  486         return (0);
  487 }
  488 
  489 /*
  490  * Package up an I/O request on a vnode into a uio and do it.
  491  */
  492 int
  493 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
  494     enum uio_seg segflg, int ioflg, struct ucred *active_cred,
  495     struct ucred *file_cred, ssize_t *aresid, struct thread *td)
  496 {
  497         struct uio auio;
  498         struct iovec aiov;
  499         struct mount *mp;
  500         struct ucred *cred;
  501         void *rl_cookie;
  502         struct vn_io_fault_args args;
  503         int error, lock_flags;
  504 
  505         auio.uio_iov = &aiov;
  506         auio.uio_iovcnt = 1;
  507         aiov.iov_base = base;
  508         aiov.iov_len = len;
  509         auio.uio_resid = len;
  510         auio.uio_offset = offset;
  511         auio.uio_segflg = segflg;
  512         auio.uio_rw = rw;
  513         auio.uio_td = td;
  514         error = 0;
  515 
  516         if ((ioflg & IO_NODELOCKED) == 0) {
  517                 if ((ioflg & IO_RANGELOCKED) == 0) {
  518                         if (rw == UIO_READ) {
  519                                 rl_cookie = vn_rangelock_rlock(vp, offset,
  520                                     offset + len);
  521                         } else {
  522                                 rl_cookie = vn_rangelock_wlock(vp, offset,
  523                                     offset + len);
  524                         }
  525                 } else
  526                         rl_cookie = NULL;
  527                 mp = NULL;
  528                 if (rw == UIO_WRITE) { 
  529                         if (vp->v_type != VCHR &&
  530                             (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
  531                             != 0)
  532                                 goto out;
  533                         if (MNT_SHARED_WRITES(mp) ||
  534                             ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
  535                                 lock_flags = LK_SHARED;
  536                         else
  537                                 lock_flags = LK_EXCLUSIVE;
  538                 } else
  539                         lock_flags = LK_SHARED;
  540                 vn_lock(vp, lock_flags | LK_RETRY);
  541         } else
  542                 rl_cookie = NULL;
  543 
  544         ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
  545 #ifdef MAC
  546         if ((ioflg & IO_NOMACCHECK) == 0) {
  547                 if (rw == UIO_READ)
  548                         error = mac_vnode_check_read(active_cred, file_cred,
  549                             vp);
  550                 else
  551                         error = mac_vnode_check_write(active_cred, file_cred,
  552                             vp);
  553         }
  554 #endif
  555         if (error == 0) {
  556                 if (file_cred != NULL)
  557                         cred = file_cred;
  558                 else
  559                         cred = active_cred;
  560                 if (do_vn_io_fault(vp, &auio)) {
  561                         args.kind = VN_IO_FAULT_VOP;
  562                         args.cred = cred;
  563                         args.flags = ioflg;
  564                         args.args.vop_args.vp = vp;
  565                         error = vn_io_fault1(vp, &auio, &args, td);
  566                 } else if (rw == UIO_READ) {
  567                         error = VOP_READ(vp, &auio, ioflg, cred);
  568                 } else /* if (rw == UIO_WRITE) */ {
  569                         error = VOP_WRITE(vp, &auio, ioflg, cred);
  570                 }
  571         }
  572         if (aresid)
  573                 *aresid = auio.uio_resid;
  574         else
  575                 if (auio.uio_resid && error == 0)
  576                         error = EIO;
  577         if ((ioflg & IO_NODELOCKED) == 0) {
  578                 VOP_UNLOCK(vp, 0);
  579                 if (mp != NULL)
  580                         vn_finished_write(mp);
  581         }
  582  out:
  583         if (rl_cookie != NULL)
  584                 vn_rangelock_unlock(vp, rl_cookie);
  585         return (error);
  586 }
  587 
  588 /*
  589  * Package up an I/O request on a vnode into a uio and do it.  The I/O
  590  * request is split up into smaller chunks and we try to avoid saturating
  591  * the buffer cache while potentially holding a vnode locked, so we 
  592  * check bwillwrite() before calling vn_rdwr().  We also call kern_yield()
  593  * to give other processes a chance to lock the vnode (either other processes
  594  * core'ing the same binary, or unrelated processes scanning the directory).
  595  */
  596 int
  597 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
  598     file_cred, aresid, td)
  599         enum uio_rw rw;
  600         struct vnode *vp;
  601         void *base;
  602         size_t len;
  603         off_t offset;
  604         enum uio_seg segflg;
  605         int ioflg;
  606         struct ucred *active_cred;
  607         struct ucred *file_cred;
  608         size_t *aresid;
  609         struct thread *td;
  610 {
  611         int error = 0;
  612         ssize_t iaresid;
  613 
  614         do {
  615                 int chunk;
  616 
  617                 /*
  618                  * Force `offset' to a multiple of MAXBSIZE except possibly
  619                  * for the first chunk, so that filesystems only need to
  620                  * write full blocks except possibly for the first and last
  621                  * chunks.
  622                  */
  623                 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
  624 
  625                 if (chunk > len)
  626                         chunk = len;
  627                 if (rw != UIO_READ && vp->v_type == VREG)
  628                         bwillwrite();
  629                 iaresid = 0;
  630                 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
  631                     ioflg, active_cred, file_cred, &iaresid, td);
  632                 len -= chunk;   /* aresid calc already includes length */
  633                 if (error)
  634                         break;
  635                 offset += chunk;
  636                 base = (char *)base + chunk;
  637                 kern_yield(PRI_USER);
  638         } while (len);
  639         if (aresid)
  640                 *aresid = len + iaresid;
  641         return (error);
  642 }
  643 
  644 off_t
  645 foffset_lock(struct file *fp, int flags)
  646 {
  647         struct mtx *mtxp;
  648         off_t res;
  649 
  650         KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
  651 
  652 #if OFF_MAX <= LONG_MAX
  653         /*
  654          * Caller only wants the current f_offset value.  Assume that
  655          * the long and shorter integer types reads are atomic.
  656          */
  657         if ((flags & FOF_NOLOCK) != 0)
  658                 return (fp->f_offset);
  659 #endif
  660 
  661         /*
  662          * According to McKusick the vn lock was protecting f_offset here.
  663          * It is now protected by the FOFFSET_LOCKED flag.
  664          */
  665         mtxp = mtx_pool_find(mtxpool_sleep, fp);
  666         mtx_lock(mtxp);
  667         if ((flags & FOF_NOLOCK) == 0) {
  668                 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
  669                         fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
  670                         msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
  671                             "vofflock", 0);
  672                 }
  673                 fp->f_vnread_flags |= FOFFSET_LOCKED;
  674         }
  675         res = fp->f_offset;
  676         mtx_unlock(mtxp);
  677         return (res);
  678 }
  679 
  680 void
  681 foffset_unlock(struct file *fp, off_t val, int flags)
  682 {
  683         struct mtx *mtxp;
  684 
  685         KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
  686 
  687 #if OFF_MAX <= LONG_MAX
  688         if ((flags & FOF_NOLOCK) != 0) {
  689                 if ((flags & FOF_NOUPDATE) == 0)
  690                         fp->f_offset = val;
  691                 if ((flags & FOF_NEXTOFF) != 0)
  692                         fp->f_nextoff = val;
  693                 return;
  694         }
  695 #endif
  696 
  697         mtxp = mtx_pool_find(mtxpool_sleep, fp);
  698         mtx_lock(mtxp);
  699         if ((flags & FOF_NOUPDATE) == 0)
  700                 fp->f_offset = val;
  701         if ((flags & FOF_NEXTOFF) != 0)
  702                 fp->f_nextoff = val;
  703         if ((flags & FOF_NOLOCK) == 0) {
  704                 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
  705                     ("Lost FOFFSET_LOCKED"));
  706                 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
  707                         wakeup(&fp->f_vnread_flags);
  708                 fp->f_vnread_flags = 0;
  709         }
  710         mtx_unlock(mtxp);
  711 }
  712 
  713 void
  714 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
  715 {
  716 
  717         if ((flags & FOF_OFFSET) == 0)
  718                 uio->uio_offset = foffset_lock(fp, flags);
  719 }
  720 
  721 void
  722 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
  723 {
  724 
  725         if ((flags & FOF_OFFSET) == 0)
  726                 foffset_unlock(fp, uio->uio_offset, flags);
  727 }
  728 
  729 static int
  730 get_advice(struct file *fp, struct uio *uio)
  731 {
  732         struct mtx *mtxp;
  733         int ret;
  734 
  735         ret = POSIX_FADV_NORMAL;
  736         if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
  737                 return (ret);
  738 
  739         mtxp = mtx_pool_find(mtxpool_sleep, fp);
  740         mtx_lock(mtxp);
  741         if (fp->f_advice != NULL &&
  742             uio->uio_offset >= fp->f_advice->fa_start &&
  743             uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
  744                 ret = fp->f_advice->fa_advice;
  745         mtx_unlock(mtxp);
  746         return (ret);
  747 }
  748 
  749 /*
  750  * File table vnode read routine.
  751  */
  752 static int
  753 vn_read(fp, uio, active_cred, flags, td)
  754         struct file *fp;
  755         struct uio *uio;
  756         struct ucred *active_cred;
  757         int flags;
  758         struct thread *td;
  759 {
  760         struct vnode *vp;
  761         struct mtx *mtxp;
  762         int error, ioflag;
  763         int advice;
  764         off_t offset, start, end;
  765 
  766         KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
  767             uio->uio_td, td));
  768         KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
  769         vp = fp->f_vnode;
  770         ioflag = 0;
  771         if (fp->f_flag & FNONBLOCK)
  772                 ioflag |= IO_NDELAY;
  773         if (fp->f_flag & O_DIRECT)
  774                 ioflag |= IO_DIRECT;
  775         advice = get_advice(fp, uio);
  776         vn_lock(vp, LK_SHARED | LK_RETRY);
  777 
  778         switch (advice) {
  779         case POSIX_FADV_NORMAL:
  780         case POSIX_FADV_SEQUENTIAL:
  781         case POSIX_FADV_NOREUSE:
  782                 ioflag |= sequential_heuristic(uio, fp);
  783                 break;
  784         case POSIX_FADV_RANDOM:
  785                 /* Disable read-ahead for random I/O. */
  786                 break;
  787         }
  788         offset = uio->uio_offset;
  789 
  790 #ifdef MAC
  791         error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
  792         if (error == 0)
  793 #endif
  794                 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
  795         fp->f_nextoff = uio->uio_offset;
  796         VOP_UNLOCK(vp, 0);
  797         if (error == 0 && advice == POSIX_FADV_NOREUSE &&
  798             offset != uio->uio_offset) {
  799                 /*
  800                  * Use POSIX_FADV_DONTNEED to flush clean pages and
  801                  * buffers for the backing file after a
  802                  * POSIX_FADV_NOREUSE read(2).  To optimize the common
  803                  * case of using POSIX_FADV_NOREUSE with sequential
  804                  * access, track the previous implicit DONTNEED
  805                  * request and grow this request to include the
  806                  * current read(2) in addition to the previous
  807                  * DONTNEED.  With purely sequential access this will
  808                  * cause the DONTNEED requests to continously grow to
  809                  * cover all of the previously read regions of the
  810                  * file.  This allows filesystem blocks that are
  811                  * accessed by multiple calls to read(2) to be flushed
  812                  * once the last read(2) finishes.
  813                  */
  814                 start = offset;
  815                 end = uio->uio_offset - 1;
  816                 mtxp = mtx_pool_find(mtxpool_sleep, fp);
  817                 mtx_lock(mtxp);
  818                 if (fp->f_advice != NULL &&
  819                     fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
  820                         if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
  821                                 start = fp->f_advice->fa_prevstart;
  822                         else if (fp->f_advice->fa_prevstart != 0 &&
  823                             fp->f_advice->fa_prevstart == end + 1)
  824                                 end = fp->f_advice->fa_prevend;
  825                         fp->f_advice->fa_prevstart = start;
  826                         fp->f_advice->fa_prevend = end;
  827                 }
  828                 mtx_unlock(mtxp);
  829                 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
  830         }
  831         return (error);
  832 }
  833 
  834 /*
  835  * File table vnode write routine.
  836  */
  837 static int
  838 vn_write(fp, uio, active_cred, flags, td)
  839         struct file *fp;
  840         struct uio *uio;
  841         struct ucred *active_cred;
  842         int flags;
  843         struct thread *td;
  844 {
  845         struct vnode *vp;
  846         struct mount *mp;
  847         struct mtx *mtxp;
  848         int error, ioflag, lock_flags;
  849         int advice;
  850         off_t offset, start, end;
  851 
  852         KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
  853             uio->uio_td, td));
  854         KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
  855         vp = fp->f_vnode;
  856         if (vp->v_type == VREG)
  857                 bwillwrite();
  858         ioflag = IO_UNIT;
  859         if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
  860                 ioflag |= IO_APPEND;
  861         if (fp->f_flag & FNONBLOCK)
  862                 ioflag |= IO_NDELAY;
  863         if (fp->f_flag & O_DIRECT)
  864                 ioflag |= IO_DIRECT;
  865         if ((fp->f_flag & O_FSYNC) ||
  866             (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
  867                 ioflag |= IO_SYNC;
  868         mp = NULL;
  869         if (vp->v_type != VCHR &&
  870             (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
  871                 goto unlock;
  872 
  873         advice = get_advice(fp, uio);
  874 
  875         if (MNT_SHARED_WRITES(mp) ||
  876             (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
  877                 lock_flags = LK_SHARED;
  878         } else {
  879                 lock_flags = LK_EXCLUSIVE;
  880         }
  881 
  882         vn_lock(vp, lock_flags | LK_RETRY);
  883         switch (advice) {
  884         case POSIX_FADV_NORMAL:
  885         case POSIX_FADV_SEQUENTIAL:
  886         case POSIX_FADV_NOREUSE:
  887                 ioflag |= sequential_heuristic(uio, fp);
  888                 break;
  889         case POSIX_FADV_RANDOM:
  890                 /* XXX: Is this correct? */
  891                 break;
  892         }
  893         offset = uio->uio_offset;
  894 
  895 #ifdef MAC
  896         error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
  897         if (error == 0)
  898 #endif
  899                 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
  900         fp->f_nextoff = uio->uio_offset;
  901         VOP_UNLOCK(vp, 0);
  902         if (vp->v_type != VCHR)
  903                 vn_finished_write(mp);
  904         if (error == 0 && advice == POSIX_FADV_NOREUSE &&
  905             offset != uio->uio_offset) {
  906                 /*
  907                  * Use POSIX_FADV_DONTNEED to flush clean pages and
  908                  * buffers for the backing file after a
  909                  * POSIX_FADV_NOREUSE write(2).  To optimize the
  910                  * common case of using POSIX_FADV_NOREUSE with
  911                  * sequential access, track the previous implicit
  912                  * DONTNEED request and grow this request to include
  913                  * the current write(2) in addition to the previous
  914                  * DONTNEED.  With purely sequential access this will
  915                  * cause the DONTNEED requests to continously grow to
  916                  * cover all of the previously written regions of the
  917                  * file.
  918                  *
  919                  * Note that the blocks just written are almost
  920                  * certainly still dirty, so this only works when
  921                  * VOP_ADVISE() calls from subsequent writes push out
  922                  * the data written by this write(2) once the backing
  923                  * buffers are clean.  However, as compared to forcing
  924                  * IO_DIRECT, this gives much saner behavior.  Write
  925                  * clustering is still allowed, and clean pages are
  926                  * merely moved to the cache page queue rather than
  927                  * outright thrown away.  This means a subsequent
  928                  * read(2) can still avoid hitting the disk if the
  929                  * pages have not been reclaimed.
  930                  *
  931                  * This does make POSIX_FADV_NOREUSE largely useless
  932                  * with non-sequential access.  However, sequential
  933                  * access is the more common use case and the flag is
  934                  * merely advisory.
  935                  */
  936                 start = offset;
  937                 end = uio->uio_offset - 1;
  938                 mtxp = mtx_pool_find(mtxpool_sleep, fp);
  939                 mtx_lock(mtxp);
  940                 if (fp->f_advice != NULL &&
  941                     fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
  942                         if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
  943                                 start = fp->f_advice->fa_prevstart;
  944                         else if (fp->f_advice->fa_prevstart != 0 &&
  945                             fp->f_advice->fa_prevstart == end + 1)
  946                                 end = fp->f_advice->fa_prevend;
  947                         fp->f_advice->fa_prevstart = start;
  948                         fp->f_advice->fa_prevend = end;
  949                 }
  950                 mtx_unlock(mtxp);
  951                 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
  952         }
  953         
  954 unlock:
  955         return (error);
  956 }
  957 
  958 /*
  959  * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
  960  * prevent the following deadlock:
  961  *
  962  * Assume that the thread A reads from the vnode vp1 into userspace
  963  * buffer buf1 backed by the pages of vnode vp2.  If a page in buf1 is
  964  * currently not resident, then system ends up with the call chain
  965  *   vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
  966  *     vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
  967  * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
  968  * If, at the same time, thread B reads from vnode vp2 into buffer buf2
  969  * backed by the pages of vnode vp1, and some page in buf2 is not
  970  * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
  971  *
  972  * To prevent the lock order reversal and deadlock, vn_io_fault() does
  973  * not allow page faults to happen during VOP_READ() or VOP_WRITE().
  974  * Instead, it first tries to do the whole range i/o with pagefaults
  975  * disabled. If all pages in the i/o buffer are resident and mapped,
  976  * VOP will succeed (ignoring the genuine filesystem errors).
  977  * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
  978  * i/o in chunks, with all pages in the chunk prefaulted and held
  979  * using vm_fault_quick_hold_pages().
  980  *
  981  * Filesystems using this deadlock avoidance scheme should use the
  982  * array of the held pages from uio, saved in the curthread->td_ma,
  983  * instead of doing uiomove().  A helper function
  984  * vn_io_fault_uiomove() converts uiomove request into
  985  * uiomove_fromphys() over td_ma array.
  986  *
  987  * Since vnode locks do not cover the whole i/o anymore, rangelocks
  988  * make the current i/o request atomic with respect to other i/os and
  989  * truncations.
  990  */
  991 
  992 /*
  993  * Decode vn_io_fault_args and perform the corresponding i/o.
  994  */
  995 static int
  996 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
  997     struct thread *td)
  998 {
  999 
 1000         switch (args->kind) {
 1001         case VN_IO_FAULT_FOP:
 1002                 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
 1003                     uio, args->cred, args->flags, td));
 1004         case VN_IO_FAULT_VOP:
 1005                 if (uio->uio_rw == UIO_READ) {
 1006                         return (VOP_READ(args->args.vop_args.vp, uio,
 1007                             args->flags, args->cred));
 1008                 } else if (uio->uio_rw == UIO_WRITE) {
 1009                         return (VOP_WRITE(args->args.vop_args.vp, uio,
 1010                             args->flags, args->cred));
 1011                 }
 1012                 break;
 1013         }
 1014         panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
 1015             uio->uio_rw);
 1016 }
 1017 
 1018 static int
 1019 vn_io_fault_touch(char *base, const struct uio *uio)
 1020 {
 1021         int r;
 1022 
 1023         r = fubyte(base);
 1024         if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
 1025                 return (EFAULT);
 1026         return (0);
 1027 }
 1028 
 1029 static int
 1030 vn_io_fault_prefault_user(const struct uio *uio)
 1031 {
 1032         char *base;
 1033         const struct iovec *iov;
 1034         size_t len;
 1035         ssize_t resid;
 1036         int error, i;
 1037 
 1038         KASSERT(uio->uio_segflg == UIO_USERSPACE,
 1039             ("vn_io_fault_prefault userspace"));
 1040 
 1041         error = i = 0;
 1042         iov = uio->uio_iov;
 1043         resid = uio->uio_resid;
 1044         base = iov->iov_base;
 1045         len = iov->iov_len;
 1046         while (resid > 0) {
 1047                 error = vn_io_fault_touch(base, uio);
 1048                 if (error != 0)
 1049                         break;
 1050                 if (len < PAGE_SIZE) {
 1051                         if (len != 0) {
 1052                                 error = vn_io_fault_touch(base + len - 1, uio);
 1053                                 if (error != 0)
 1054                                         break;
 1055                                 resid -= len;
 1056                         }
 1057                         if (++i >= uio->uio_iovcnt)
 1058                                 break;
 1059                         iov = uio->uio_iov + i;
 1060                         base = iov->iov_base;
 1061                         len = iov->iov_len;
 1062                 } else {
 1063                         len -= PAGE_SIZE;
 1064                         base += PAGE_SIZE;
 1065                         resid -= PAGE_SIZE;
 1066                 }
 1067         }
 1068         return (error);
 1069 }
 1070 
 1071 /*
 1072  * Common code for vn_io_fault(), agnostic to the kind of i/o request.
 1073  * Uses vn_io_fault_doio() to make the call to an actual i/o function.
 1074  * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
 1075  * into args and call vn_io_fault1() to handle faults during the user
 1076  * mode buffer accesses.
 1077  */
 1078 static int
 1079 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
 1080     struct thread *td)
 1081 {
 1082         vm_page_t ma[io_hold_cnt + 2];
 1083         struct uio *uio_clone, short_uio;
 1084         struct iovec short_iovec[1];
 1085         vm_page_t *prev_td_ma;
 1086         vm_prot_t prot;
 1087         vm_offset_t addr, end;
 1088         size_t len, resid;
 1089         ssize_t adv;
 1090         int error, cnt, save, saveheld, prev_td_ma_cnt;
 1091 
 1092         if (vn_io_fault_prefault) {
 1093                 error = vn_io_fault_prefault_user(uio);
 1094                 if (error != 0)
 1095                         return (error); /* Or ignore ? */
 1096         }
 1097 
 1098         prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
 1099 
 1100         /*
 1101          * The UFS follows IO_UNIT directive and replays back both
 1102          * uio_offset and uio_resid if an error is encountered during the
 1103          * operation.  But, since the iovec may be already advanced,
 1104          * uio is still in an inconsistent state.
 1105          *
 1106          * Cache a copy of the original uio, which is advanced to the redo
 1107          * point using UIO_NOCOPY below.
 1108          */
 1109         uio_clone = cloneuio(uio);
 1110         resid = uio->uio_resid;
 1111 
 1112         short_uio.uio_segflg = UIO_USERSPACE;
 1113         short_uio.uio_rw = uio->uio_rw;
 1114         short_uio.uio_td = uio->uio_td;
 1115 
 1116         save = vm_fault_disable_pagefaults();
 1117         error = vn_io_fault_doio(args, uio, td);
 1118         if (error != EFAULT)
 1119                 goto out;
 1120 
 1121         atomic_add_long(&vn_io_faults_cnt, 1);
 1122         uio_clone->uio_segflg = UIO_NOCOPY;
 1123         uiomove(NULL, resid - uio->uio_resid, uio_clone);
 1124         uio_clone->uio_segflg = uio->uio_segflg;
 1125 
 1126         saveheld = curthread_pflags_set(TDP_UIOHELD);
 1127         prev_td_ma = td->td_ma;
 1128         prev_td_ma_cnt = td->td_ma_cnt;
 1129 
 1130         while (uio_clone->uio_resid != 0) {
 1131                 len = uio_clone->uio_iov->iov_len;
 1132                 if (len == 0) {
 1133                         KASSERT(uio_clone->uio_iovcnt >= 1,
 1134                             ("iovcnt underflow"));
 1135                         uio_clone->uio_iov++;
 1136                         uio_clone->uio_iovcnt--;
 1137                         continue;
 1138                 }
 1139                 if (len > io_hold_cnt * PAGE_SIZE)
 1140                         len = io_hold_cnt * PAGE_SIZE;
 1141                 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
 1142                 end = round_page(addr + len);
 1143                 if (end < addr) {
 1144                         error = EFAULT;
 1145                         break;
 1146                 }
 1147                 cnt = atop(end - trunc_page(addr));
 1148                 /*
 1149                  * A perfectly misaligned address and length could cause
 1150                  * both the start and the end of the chunk to use partial
 1151                  * page.  +2 accounts for such a situation.
 1152                  */
 1153                 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
 1154                     addr, len, prot, ma, io_hold_cnt + 2);
 1155                 if (cnt == -1) {
 1156                         error = EFAULT;
 1157                         break;
 1158                 }
 1159                 short_uio.uio_iov = &short_iovec[0];
 1160                 short_iovec[0].iov_base = (void *)addr;
 1161                 short_uio.uio_iovcnt = 1;
 1162                 short_uio.uio_resid = short_iovec[0].iov_len = len;
 1163                 short_uio.uio_offset = uio_clone->uio_offset;
 1164                 td->td_ma = ma;
 1165                 td->td_ma_cnt = cnt;
 1166 
 1167                 error = vn_io_fault_doio(args, &short_uio, td);
 1168                 vm_page_unhold_pages(ma, cnt);
 1169                 adv = len - short_uio.uio_resid;
 1170 
 1171                 uio_clone->uio_iov->iov_base =
 1172                     (char *)uio_clone->uio_iov->iov_base + adv;
 1173                 uio_clone->uio_iov->iov_len -= adv;
 1174                 uio_clone->uio_resid -= adv;
 1175                 uio_clone->uio_offset += adv;
 1176 
 1177                 uio->uio_resid -= adv;
 1178                 uio->uio_offset += adv;
 1179 
 1180                 if (error != 0 || adv == 0)
 1181                         break;
 1182         }
 1183         td->td_ma = prev_td_ma;
 1184         td->td_ma_cnt = prev_td_ma_cnt;
 1185         curthread_pflags_restore(saveheld);
 1186 out:
 1187         vm_fault_enable_pagefaults(save);
 1188         free(uio_clone, M_IOV);
 1189         return (error);
 1190 }
 1191 
 1192 static int
 1193 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
 1194     int flags, struct thread *td)
 1195 {
 1196         fo_rdwr_t *doio;
 1197         struct vnode *vp;
 1198         void *rl_cookie;
 1199         struct vn_io_fault_args args;
 1200         int error;
 1201 
 1202         doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
 1203         vp = fp->f_vnode;
 1204         foffset_lock_uio(fp, uio, flags);
 1205         if (do_vn_io_fault(vp, uio)) {
 1206                 args.kind = VN_IO_FAULT_FOP;
 1207                 args.args.fop_args.fp = fp;
 1208                 args.args.fop_args.doio = doio;
 1209                 args.cred = active_cred;
 1210                 args.flags = flags | FOF_OFFSET;
 1211                 if (uio->uio_rw == UIO_READ) {
 1212                         rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
 1213                             uio->uio_offset + uio->uio_resid);
 1214                 } else if ((fp->f_flag & O_APPEND) != 0 ||
 1215                     (flags & FOF_OFFSET) == 0) {
 1216                         /* For appenders, punt and lock the whole range. */
 1217                         rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
 1218                 } else {
 1219                         rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
 1220                             uio->uio_offset + uio->uio_resid);
 1221                 }
 1222                 error = vn_io_fault1(vp, uio, &args, td);
 1223                 vn_rangelock_unlock(vp, rl_cookie);
 1224         } else {
 1225                 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
 1226         }
 1227         foffset_unlock_uio(fp, uio, flags);
 1228         return (error);
 1229 }
 1230 
 1231 /*
 1232  * Helper function to perform the requested uiomove operation using
 1233  * the held pages for io->uio_iov[0].iov_base buffer instead of
 1234  * copyin/copyout.  Access to the pages with uiomove_fromphys()
 1235  * instead of iov_base prevents page faults that could occur due to
 1236  * pmap_collect() invalidating the mapping created by
 1237  * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
 1238  * object cleanup revoking the write access from page mappings.
 1239  *
 1240  * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
 1241  * instead of plain uiomove().
 1242  */
 1243 int
 1244 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
 1245 {
 1246         struct uio transp_uio;
 1247         struct iovec transp_iov[1];
 1248         struct thread *td;
 1249         size_t adv;
 1250         int error, pgadv;
 1251 
 1252         td = curthread;
 1253         if ((td->td_pflags & TDP_UIOHELD) == 0 ||
 1254             uio->uio_segflg != UIO_USERSPACE)
 1255                 return (uiomove(data, xfersize, uio));
 1256 
 1257         KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
 1258         transp_iov[0].iov_base = data;
 1259         transp_uio.uio_iov = &transp_iov[0];
 1260         transp_uio.uio_iovcnt = 1;
 1261         if (xfersize > uio->uio_resid)
 1262                 xfersize = uio->uio_resid;
 1263         transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
 1264         transp_uio.uio_offset = 0;
 1265         transp_uio.uio_segflg = UIO_SYSSPACE;
 1266         /*
 1267          * Since transp_iov points to data, and td_ma page array
 1268          * corresponds to original uio->uio_iov, we need to invert the
 1269          * direction of the i/o operation as passed to
 1270          * uiomove_fromphys().
 1271          */
 1272         switch (uio->uio_rw) {
 1273         case UIO_WRITE:
 1274                 transp_uio.uio_rw = UIO_READ;
 1275                 break;
 1276         case UIO_READ:
 1277                 transp_uio.uio_rw = UIO_WRITE;
 1278                 break;
 1279         }
 1280         transp_uio.uio_td = uio->uio_td;
 1281         error = uiomove_fromphys(td->td_ma,
 1282             ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
 1283             xfersize, &transp_uio);
 1284         adv = xfersize - transp_uio.uio_resid;
 1285         pgadv =
 1286             (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
 1287             (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
 1288         td->td_ma += pgadv;
 1289         KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
 1290             pgadv));
 1291         td->td_ma_cnt -= pgadv;
 1292         uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
 1293         uio->uio_iov->iov_len -= adv;
 1294         uio->uio_resid -= adv;
 1295         uio->uio_offset += adv;
 1296         return (error);
 1297 }
 1298 
 1299 int
 1300 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
 1301     struct uio *uio)
 1302 {
 1303         struct thread *td;
 1304         vm_offset_t iov_base;
 1305         int cnt, pgadv;
 1306 
 1307         td = curthread;
 1308         if ((td->td_pflags & TDP_UIOHELD) == 0 ||
 1309             uio->uio_segflg != UIO_USERSPACE)
 1310                 return (uiomove_fromphys(ma, offset, xfersize, uio));
 1311 
 1312         KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
 1313         cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
 1314         iov_base = (vm_offset_t)uio->uio_iov->iov_base;
 1315         switch (uio->uio_rw) {
 1316         case UIO_WRITE:
 1317                 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
 1318                     offset, cnt);
 1319                 break;
 1320         case UIO_READ:
 1321                 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
 1322                     cnt);
 1323                 break;
 1324         }
 1325         pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
 1326         td->td_ma += pgadv;
 1327         KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
 1328             pgadv));
 1329         td->td_ma_cnt -= pgadv;
 1330         uio->uio_iov->iov_base = (char *)(iov_base + cnt);
 1331         uio->uio_iov->iov_len -= cnt;
 1332         uio->uio_resid -= cnt;
 1333         uio->uio_offset += cnt;
 1334         return (0);
 1335 }
 1336 
 1337 
 1338 /*
 1339  * File table truncate routine.
 1340  */
 1341 static int
 1342 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
 1343     struct thread *td)
 1344 {
 1345         struct vattr vattr;
 1346         struct mount *mp;
 1347         struct vnode *vp;
 1348         void *rl_cookie;
 1349         int error;
 1350 
 1351         vp = fp->f_vnode;
 1352 
 1353         /*
 1354          * Lock the whole range for truncation.  Otherwise split i/o
 1355          * might happen partly before and partly after the truncation.
 1356          */
 1357         rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
 1358         error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
 1359         if (error)
 1360                 goto out1;
 1361         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1362         if (vp->v_type == VDIR) {
 1363                 error = EISDIR;
 1364                 goto out;
 1365         }
 1366 #ifdef MAC
 1367         error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
 1368         if (error)
 1369                 goto out;
 1370 #endif
 1371         error = vn_writechk(vp);
 1372         if (error == 0) {
 1373                 VATTR_NULL(&vattr);
 1374                 vattr.va_size = length;
 1375                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
 1376         }
 1377 out:
 1378         VOP_UNLOCK(vp, 0);
 1379         vn_finished_write(mp);
 1380 out1:
 1381         vn_rangelock_unlock(vp, rl_cookie);
 1382         return (error);
 1383 }
 1384 
 1385 /*
 1386  * File table vnode stat routine.
 1387  */
 1388 static int
 1389 vn_statfile(fp, sb, active_cred, td)
 1390         struct file *fp;
 1391         struct stat *sb;
 1392         struct ucred *active_cred;
 1393         struct thread *td;
 1394 {
 1395         struct vnode *vp = fp->f_vnode;
 1396         int error;
 1397 
 1398         vn_lock(vp, LK_SHARED | LK_RETRY);
 1399         error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
 1400         VOP_UNLOCK(vp, 0);
 1401 
 1402         return (error);
 1403 }
 1404 
 1405 /*
 1406  * Stat a vnode; implementation for the stat syscall
 1407  */
 1408 int
 1409 vn_stat(vp, sb, active_cred, file_cred, td)
 1410         struct vnode *vp;
 1411         register struct stat *sb;
 1412         struct ucred *active_cred;
 1413         struct ucred *file_cred;
 1414         struct thread *td;
 1415 {
 1416         struct vattr vattr;
 1417         register struct vattr *vap;
 1418         int error;
 1419         u_short mode;
 1420 
 1421 #ifdef MAC
 1422         error = mac_vnode_check_stat(active_cred, file_cred, vp);
 1423         if (error)
 1424                 return (error);
 1425 #endif
 1426 
 1427         vap = &vattr;
 1428 
 1429         /*
 1430          * Initialize defaults for new and unusual fields, so that file
 1431          * systems which don't support these fields don't need to know
 1432          * about them.
 1433          */
 1434         vap->va_birthtime.tv_sec = -1;
 1435         vap->va_birthtime.tv_nsec = 0;
 1436         vap->va_fsid = VNOVAL;
 1437         vap->va_rdev = NODEV;
 1438 
 1439         error = VOP_GETATTR(vp, vap, active_cred);
 1440         if (error)
 1441                 return (error);
 1442 
 1443         /*
 1444          * Zero the spare stat fields
 1445          */
 1446         bzero(sb, sizeof *sb);
 1447 
 1448         /*
 1449          * Copy from vattr table
 1450          */
 1451         if (vap->va_fsid != VNOVAL)
 1452                 sb->st_dev = vap->va_fsid;
 1453         else
 1454                 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
 1455         sb->st_ino = vap->va_fileid;
 1456         mode = vap->va_mode;
 1457         switch (vap->va_type) {
 1458         case VREG:
 1459                 mode |= S_IFREG;
 1460                 break;
 1461         case VDIR:
 1462                 mode |= S_IFDIR;
 1463                 break;
 1464         case VBLK:
 1465                 mode |= S_IFBLK;
 1466                 break;
 1467         case VCHR:
 1468                 mode |= S_IFCHR;
 1469                 break;
 1470         case VLNK:
 1471                 mode |= S_IFLNK;
 1472                 break;
 1473         case VSOCK:
 1474                 mode |= S_IFSOCK;
 1475                 break;
 1476         case VFIFO:
 1477                 mode |= S_IFIFO;
 1478                 break;
 1479         default:
 1480                 return (EBADF);
 1481         };
 1482         sb->st_mode = mode;
 1483         sb->st_nlink = vap->va_nlink;
 1484         sb->st_uid = vap->va_uid;
 1485         sb->st_gid = vap->va_gid;
 1486         sb->st_rdev = vap->va_rdev;
 1487         if (vap->va_size > OFF_MAX)
 1488                 return (EOVERFLOW);
 1489         sb->st_size = vap->va_size;
 1490         sb->st_atim = vap->va_atime;
 1491         sb->st_mtim = vap->va_mtime;
 1492         sb->st_ctim = vap->va_ctime;
 1493         sb->st_birthtim = vap->va_birthtime;
 1494 
 1495         /*
 1496          * According to www.opengroup.org, the meaning of st_blksize is 
 1497          *   "a filesystem-specific preferred I/O block size for this 
 1498          *    object.  In some filesystem types, this may vary from file
 1499          *    to file"
 1500          * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
 1501          */
 1502 
 1503         sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
 1504         
 1505         sb->st_flags = vap->va_flags;
 1506         if (priv_check(td, PRIV_VFS_GENERATION))
 1507                 sb->st_gen = 0;
 1508         else
 1509                 sb->st_gen = vap->va_gen;
 1510 
 1511         sb->st_blocks = vap->va_bytes / S_BLKSIZE;
 1512         return (0);
 1513 }
 1514 
 1515 /*
 1516  * File table vnode ioctl routine.
 1517  */
 1518 static int
 1519 vn_ioctl(fp, com, data, active_cred, td)
 1520         struct file *fp;
 1521         u_long com;
 1522         void *data;
 1523         struct ucred *active_cred;
 1524         struct thread *td;
 1525 {
 1526         struct vattr vattr;
 1527         struct vnode *vp;
 1528         int error;
 1529 
 1530         vp = fp->f_vnode;
 1531         switch (vp->v_type) {
 1532         case VDIR:
 1533         case VREG:
 1534                 switch (com) {
 1535                 case FIONREAD:
 1536                         vn_lock(vp, LK_SHARED | LK_RETRY);
 1537                         error = VOP_GETATTR(vp, &vattr, active_cred);
 1538                         VOP_UNLOCK(vp, 0);
 1539                         if (error == 0)
 1540                                 *(int *)data = vattr.va_size - fp->f_offset;
 1541                         return (error);
 1542                 case FIONBIO:
 1543                 case FIOASYNC:
 1544                         return (0);
 1545                 default:
 1546                         return (VOP_IOCTL(vp, com, data, fp->f_flag,
 1547                             active_cred, td));
 1548                 }
 1549         default:
 1550                 return (ENOTTY);
 1551         }
 1552 }
 1553 
 1554 /*
 1555  * File table vnode poll routine.
 1556  */
 1557 static int
 1558 vn_poll(fp, events, active_cred, td)
 1559         struct file *fp;
 1560         int events;
 1561         struct ucred *active_cred;
 1562         struct thread *td;
 1563 {
 1564         struct vnode *vp;
 1565         int error;
 1566 
 1567         vp = fp->f_vnode;
 1568 #ifdef MAC
 1569         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1570         error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
 1571         VOP_UNLOCK(vp, 0);
 1572         if (!error)
 1573 #endif
 1574 
 1575         error = VOP_POLL(vp, events, fp->f_cred, td);
 1576         return (error);
 1577 }
 1578 
 1579 /*
 1580  * Acquire the requested lock and then check for validity.  LK_RETRY
 1581  * permits vn_lock to return doomed vnodes.
 1582  */
 1583 int
 1584 _vn_lock(struct vnode *vp, int flags, char *file, int line)
 1585 {
 1586         int error;
 1587 
 1588         VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
 1589             ("vn_lock called with no locktype."));
 1590         do {
 1591 #ifdef DEBUG_VFS_LOCKS
 1592                 KASSERT(vp->v_holdcnt != 0,
 1593                     ("vn_lock %p: zero hold count", vp));
 1594 #endif
 1595                 error = VOP_LOCK1(vp, flags, file, line);
 1596                 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
 1597                 KASSERT((flags & LK_RETRY) == 0 || error == 0,
 1598                     ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
 1599                     flags, error));
 1600                 /*
 1601                  * Callers specify LK_RETRY if they wish to get dead vnodes.
 1602                  * If RETRY is not set, we return ENOENT instead.
 1603                  */
 1604                 if (error == 0 && vp->v_iflag & VI_DOOMED &&
 1605                     (flags & LK_RETRY) == 0) {
 1606                         VOP_UNLOCK(vp, 0);
 1607                         error = ENOENT;
 1608                         break;
 1609                 }
 1610         } while (flags & LK_RETRY && error != 0);
 1611         return (error);
 1612 }
 1613 
 1614 /*
 1615  * File table vnode close routine.
 1616  */
 1617 static int
 1618 vn_closefile(fp, td)
 1619         struct file *fp;
 1620         struct thread *td;
 1621 {
 1622         struct vnode *vp;
 1623         struct flock lf;
 1624         int error;
 1625 
 1626         vp = fp->f_vnode;
 1627         fp->f_ops = &badfileops;
 1628 
 1629         if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
 1630                 vref(vp);
 1631 
 1632         error = vn_close(vp, fp->f_flag, fp->f_cred, td);
 1633 
 1634         if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
 1635                 lf.l_whence = SEEK_SET;
 1636                 lf.l_start = 0;
 1637                 lf.l_len = 0;
 1638                 lf.l_type = F_UNLCK;
 1639                 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
 1640                 vrele(vp);
 1641         }
 1642         return (error);
 1643 }
 1644 
 1645 /*
 1646  * Preparing to start a filesystem write operation. If the operation is
 1647  * permitted, then we bump the count of operations in progress and
 1648  * proceed. If a suspend request is in progress, we wait until the
 1649  * suspension is over, and then proceed.
 1650  */
 1651 static int
 1652 vn_start_write_locked(struct mount *mp, int flags)
 1653 {
 1654         int error, mflags;
 1655 
 1656         mtx_assert(MNT_MTX(mp), MA_OWNED);
 1657         error = 0;
 1658 
 1659         /*
 1660          * Check on status of suspension.
 1661          */
 1662         if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
 1663             mp->mnt_susp_owner != curthread) {
 1664                 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
 1665                     (flags & PCATCH) : 0) | (PUSER - 1);
 1666                 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
 1667                         if (flags & V_NOWAIT) {
 1668                                 error = EWOULDBLOCK;
 1669                                 goto unlock;
 1670                         }
 1671                         error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
 1672                             "suspfs", 0);
 1673                         if (error)
 1674                                 goto unlock;
 1675                 }
 1676         }
 1677         if (flags & V_XSLEEP)
 1678                 goto unlock;
 1679         mp->mnt_writeopcount++;
 1680 unlock:
 1681         if (error != 0 || (flags & V_XSLEEP) != 0)
 1682                 MNT_REL(mp);
 1683         MNT_IUNLOCK(mp);
 1684         return (error);
 1685 }
 1686 
 1687 int
 1688 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
 1689 {
 1690         struct mount *mp;
 1691         int error;
 1692 
 1693         KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
 1694             ("V_MNTREF requires mp"));
 1695 
 1696         error = 0;
 1697         /*
 1698          * If a vnode is provided, get and return the mount point that
 1699          * to which it will write.
 1700          */
 1701         if (vp != NULL) {
 1702                 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
 1703                         *mpp = NULL;
 1704                         if (error != EOPNOTSUPP)
 1705                                 return (error);
 1706                         return (0);
 1707                 }
 1708         }
 1709         if ((mp = *mpp) == NULL)
 1710                 return (0);
 1711 
 1712         /*
 1713          * VOP_GETWRITEMOUNT() returns with the mp refcount held through
 1714          * a vfs_ref().
 1715          * As long as a vnode is not provided we need to acquire a
 1716          * refcount for the provided mountpoint too, in order to
 1717          * emulate a vfs_ref().
 1718          */
 1719         MNT_ILOCK(mp);
 1720         if (vp == NULL && (flags & V_MNTREF) == 0)
 1721                 MNT_REF(mp);
 1722 
 1723         return (vn_start_write_locked(mp, flags));
 1724 }
 1725 
 1726 /*
 1727  * Secondary suspension. Used by operations such as vop_inactive
 1728  * routines that are needed by the higher level functions. These
 1729  * are allowed to proceed until all the higher level functions have
 1730  * completed (indicated by mnt_writeopcount dropping to zero). At that
 1731  * time, these operations are halted until the suspension is over.
 1732  */
 1733 int
 1734 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
 1735 {
 1736         struct mount *mp;
 1737         int error;
 1738 
 1739         KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
 1740             ("V_MNTREF requires mp"));
 1741 
 1742  retry:
 1743         if (vp != NULL) {
 1744                 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
 1745                         *mpp = NULL;
 1746                         if (error != EOPNOTSUPP)
 1747                                 return (error);
 1748                         return (0);
 1749                 }
 1750         }
 1751         /*
 1752          * If we are not suspended or have not yet reached suspended
 1753          * mode, then let the operation proceed.
 1754          */
 1755         if ((mp = *mpp) == NULL)
 1756                 return (0);
 1757 
 1758         /*
 1759          * VOP_GETWRITEMOUNT() returns with the mp refcount held through
 1760          * a vfs_ref().
 1761          * As long as a vnode is not provided we need to acquire a
 1762          * refcount for the provided mountpoint too, in order to
 1763          * emulate a vfs_ref().
 1764          */
 1765         MNT_ILOCK(mp);
 1766         if (vp == NULL && (flags & V_MNTREF) == 0)
 1767                 MNT_REF(mp);
 1768         if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
 1769                 mp->mnt_secondary_writes++;
 1770                 mp->mnt_secondary_accwrites++;
 1771                 MNT_IUNLOCK(mp);
 1772                 return (0);
 1773         }
 1774         if (flags & V_NOWAIT) {
 1775                 MNT_REL(mp);
 1776                 MNT_IUNLOCK(mp);
 1777                 return (EWOULDBLOCK);
 1778         }
 1779         /*
 1780          * Wait for the suspension to finish.
 1781          */
 1782         error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
 1783             ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
 1784             "suspfs", 0);
 1785         vfs_rel(mp);
 1786         if (error == 0)
 1787                 goto retry;
 1788         return (error);
 1789 }
 1790 
 1791 /*
 1792  * Filesystem write operation has completed. If we are suspending and this
 1793  * operation is the last one, notify the suspender that the suspension is
 1794  * now in effect.
 1795  */
 1796 void
 1797 vn_finished_write(mp)
 1798         struct mount *mp;
 1799 {
 1800         if (mp == NULL)
 1801                 return;
 1802         MNT_ILOCK(mp);
 1803         MNT_REL(mp);
 1804         mp->mnt_writeopcount--;
 1805         if (mp->mnt_writeopcount < 0)
 1806                 panic("vn_finished_write: neg cnt");
 1807         if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
 1808             mp->mnt_writeopcount <= 0)
 1809                 wakeup(&mp->mnt_writeopcount);
 1810         MNT_IUNLOCK(mp);
 1811 }
 1812 
 1813 
 1814 /*
 1815  * Filesystem secondary write operation has completed. If we are
 1816  * suspending and this operation is the last one, notify the suspender
 1817  * that the suspension is now in effect.
 1818  */
 1819 void
 1820 vn_finished_secondary_write(mp)
 1821         struct mount *mp;
 1822 {
 1823         if (mp == NULL)
 1824                 return;
 1825         MNT_ILOCK(mp);
 1826         MNT_REL(mp);
 1827         mp->mnt_secondary_writes--;
 1828         if (mp->mnt_secondary_writes < 0)
 1829                 panic("vn_finished_secondary_write: neg cnt");
 1830         if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
 1831             mp->mnt_secondary_writes <= 0)
 1832                 wakeup(&mp->mnt_secondary_writes);
 1833         MNT_IUNLOCK(mp);
 1834 }
 1835 
 1836 
 1837 
 1838 /*
 1839  * Request a filesystem to suspend write operations.
 1840  */
 1841 int
 1842 vfs_write_suspend(struct mount *mp, int flags)
 1843 {
 1844         int error;
 1845 
 1846         MNT_ILOCK(mp);
 1847         if (mp->mnt_susp_owner == curthread) {
 1848                 MNT_IUNLOCK(mp);
 1849                 return (EALREADY);
 1850         }
 1851         while (mp->mnt_kern_flag & MNTK_SUSPEND)
 1852                 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
 1853 
 1854         /*
 1855          * Unmount holds a write reference on the mount point.  If we
 1856          * own busy reference and drain for writers, we deadlock with
 1857          * the reference draining in the unmount path.  Callers of
 1858          * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
 1859          * vfs_busy() reference is owned and caller is not in the
 1860          * unmount context.
 1861          */
 1862         if ((flags & VS_SKIP_UNMOUNT) != 0 &&
 1863             (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
 1864                 MNT_IUNLOCK(mp);
 1865                 return (EBUSY);
 1866         }
 1867 
 1868         mp->mnt_kern_flag |= MNTK_SUSPEND;
 1869         mp->mnt_susp_owner = curthread;
 1870         if (mp->mnt_writeopcount > 0)
 1871                 (void) msleep(&mp->mnt_writeopcount, 
 1872                     MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
 1873         else
 1874                 MNT_IUNLOCK(mp);
 1875         if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
 1876                 vfs_write_resume(mp, 0);
 1877         return (error);
 1878 }
 1879 
 1880 /*
 1881  * Request a filesystem to resume write operations.
 1882  */
 1883 void
 1884 vfs_write_resume(struct mount *mp, int flags)
 1885 {
 1886 
 1887         MNT_ILOCK(mp);
 1888         if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
 1889                 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
 1890                 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
 1891                                        MNTK_SUSPENDED);
 1892                 mp->mnt_susp_owner = NULL;
 1893                 wakeup(&mp->mnt_writeopcount);
 1894                 wakeup(&mp->mnt_flag);
 1895                 curthread->td_pflags &= ~TDP_IGNSUSP;
 1896                 if ((flags & VR_START_WRITE) != 0) {
 1897                         MNT_REF(mp);
 1898                         mp->mnt_writeopcount++;
 1899                 }
 1900                 MNT_IUNLOCK(mp);
 1901                 if ((flags & VR_NO_SUSPCLR) == 0)
 1902                         VFS_SUSP_CLEAN(mp);
 1903         } else if ((flags & VR_START_WRITE) != 0) {
 1904                 MNT_REF(mp);
 1905                 vn_start_write_locked(mp, 0);
 1906         } else {
 1907                 MNT_IUNLOCK(mp);
 1908         }
 1909 }
 1910 
 1911 /*
 1912  * Helper loop around vfs_write_suspend() for filesystem unmount VFS
 1913  * methods.
 1914  */
 1915 int
 1916 vfs_write_suspend_umnt(struct mount *mp)
 1917 {
 1918         int error;
 1919 
 1920         KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
 1921             ("vfs_write_suspend_umnt: recursed"));
 1922 
 1923         /* dounmount() already called vn_start_write(). */
 1924         for (;;) {
 1925                 vn_finished_write(mp);
 1926                 error = vfs_write_suspend(mp, 0);
 1927                 if (error != 0) {
 1928                         vn_start_write(NULL, &mp, V_WAIT);
 1929                         return (error);
 1930                 }
 1931                 MNT_ILOCK(mp);
 1932                 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
 1933                         break;
 1934                 MNT_IUNLOCK(mp);
 1935                 vn_start_write(NULL, &mp, V_WAIT);
 1936         }
 1937         mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
 1938         wakeup(&mp->mnt_flag);
 1939         MNT_IUNLOCK(mp);
 1940         curthread->td_pflags |= TDP_IGNSUSP;
 1941         return (0);
 1942 }
 1943 
 1944 /*
 1945  * Implement kqueues for files by translating it to vnode operation.
 1946  */
 1947 static int
 1948 vn_kqfilter(struct file *fp, struct knote *kn)
 1949 {
 1950 
 1951         return (VOP_KQFILTER(fp->f_vnode, kn));
 1952 }
 1953 
 1954 /*
 1955  * Simplified in-kernel wrapper calls for extended attribute access.
 1956  * Both calls pass in a NULL credential, authorizing as "kernel" access.
 1957  * Set IO_NODELOCKED in ioflg if the vnode is already locked.
 1958  */
 1959 int
 1960 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
 1961     const char *attrname, int *buflen, char *buf, struct thread *td)
 1962 {
 1963         struct uio      auio;
 1964         struct iovec    iov;
 1965         int     error;
 1966 
 1967         iov.iov_len = *buflen;
 1968         iov.iov_base = buf;
 1969 
 1970         auio.uio_iov = &iov;
 1971         auio.uio_iovcnt = 1;
 1972         auio.uio_rw = UIO_READ;
 1973         auio.uio_segflg = UIO_SYSSPACE;
 1974         auio.uio_td = td;
 1975         auio.uio_offset = 0;
 1976         auio.uio_resid = *buflen;
 1977 
 1978         if ((ioflg & IO_NODELOCKED) == 0)
 1979                 vn_lock(vp, LK_SHARED | LK_RETRY);
 1980 
 1981         ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
 1982 
 1983         /* authorize attribute retrieval as kernel */
 1984         error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
 1985             td);
 1986 
 1987         if ((ioflg & IO_NODELOCKED) == 0)
 1988                 VOP_UNLOCK(vp, 0);
 1989 
 1990         if (error == 0) {
 1991                 *buflen = *buflen - auio.uio_resid;
 1992         }
 1993 
 1994         return (error);
 1995 }
 1996 
 1997 /*
 1998  * XXX failure mode if partially written?
 1999  */
 2000 int
 2001 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
 2002     const char *attrname, int buflen, char *buf, struct thread *td)
 2003 {
 2004         struct uio      auio;
 2005         struct iovec    iov;
 2006         struct mount    *mp;
 2007         int     error;
 2008 
 2009         iov.iov_len = buflen;
 2010         iov.iov_base = buf;
 2011 
 2012         auio.uio_iov = &iov;
 2013         auio.uio_iovcnt = 1;
 2014         auio.uio_rw = UIO_WRITE;
 2015         auio.uio_segflg = UIO_SYSSPACE;
 2016         auio.uio_td = td;
 2017         auio.uio_offset = 0;
 2018         auio.uio_resid = buflen;
 2019 
 2020         if ((ioflg & IO_NODELOCKED) == 0) {
 2021                 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
 2022                         return (error);
 2023                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 2024         }
 2025 
 2026         ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
 2027 
 2028         /* authorize attribute setting as kernel */
 2029         error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
 2030 
 2031         if ((ioflg & IO_NODELOCKED) == 0) {
 2032                 vn_finished_write(mp);
 2033                 VOP_UNLOCK(vp, 0);
 2034         }
 2035 
 2036         return (error);
 2037 }
 2038 
 2039 int
 2040 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
 2041     const char *attrname, struct thread *td)
 2042 {
 2043         struct mount    *mp;
 2044         int     error;
 2045 
 2046         if ((ioflg & IO_NODELOCKED) == 0) {
 2047                 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
 2048                         return (error);
 2049                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 2050         }
 2051 
 2052         ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
 2053 
 2054         /* authorize attribute removal as kernel */
 2055         error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
 2056         if (error == EOPNOTSUPP)
 2057                 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
 2058                     NULL, td);
 2059 
 2060         if ((ioflg & IO_NODELOCKED) == 0) {
 2061                 vn_finished_write(mp);
 2062                 VOP_UNLOCK(vp, 0);
 2063         }
 2064 
 2065         return (error);
 2066 }
 2067 
 2068 static int
 2069 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
 2070     struct vnode **rvp)
 2071 {
 2072 
 2073         return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
 2074 }
 2075 
 2076 int
 2077 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
 2078 {
 2079 
 2080         return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
 2081             lkflags, rvp));
 2082 }
 2083 
 2084 int
 2085 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
 2086     int lkflags, struct vnode **rvp)
 2087 {
 2088         struct mount *mp;
 2089         int ltype, error;
 2090 
 2091         ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
 2092         mp = vp->v_mount;
 2093         ltype = VOP_ISLOCKED(vp);
 2094         KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
 2095             ("vn_vget_ino: vp not locked"));
 2096         error = vfs_busy(mp, MBF_NOWAIT);
 2097         if (error != 0) {
 2098                 vfs_ref(mp);
 2099                 VOP_UNLOCK(vp, 0);
 2100                 error = vfs_busy(mp, 0);
 2101                 vn_lock(vp, ltype | LK_RETRY);
 2102                 vfs_rel(mp);
 2103                 if (error != 0)
 2104                         return (ENOENT);
 2105                 if (vp->v_iflag & VI_DOOMED) {
 2106                         vfs_unbusy(mp);
 2107                         return (ENOENT);
 2108                 }
 2109         }
 2110         VOP_UNLOCK(vp, 0);
 2111         error = alloc(mp, alloc_arg, lkflags, rvp);
 2112         vfs_unbusy(mp);
 2113         if (*rvp != vp)
 2114                 vn_lock(vp, ltype | LK_RETRY);
 2115         if (vp->v_iflag & VI_DOOMED) {
 2116                 if (error == 0) {
 2117                         if (*rvp == vp)
 2118                                 vunref(vp);
 2119                         else
 2120                                 vput(*rvp);
 2121                 }
 2122                 error = ENOENT;
 2123         }
 2124         return (error);
 2125 }
 2126 
 2127 int
 2128 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
 2129     const struct thread *td)
 2130 {
 2131 
 2132         if (vp->v_type != VREG || td == NULL)
 2133                 return (0);
 2134         PROC_LOCK(td->td_proc);
 2135         if ((uoff_t)uio->uio_offset + uio->uio_resid >
 2136             lim_cur(td->td_proc, RLIMIT_FSIZE)) {
 2137                 kern_psignal(td->td_proc, SIGXFSZ);
 2138                 PROC_UNLOCK(td->td_proc);
 2139                 return (EFBIG);
 2140         }
 2141         PROC_UNLOCK(td->td_proc);
 2142         return (0);
 2143 }
 2144 
 2145 int
 2146 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
 2147     struct thread *td)
 2148 {
 2149         struct vnode *vp;
 2150 
 2151         vp = fp->f_vnode;
 2152 #ifdef AUDIT
 2153         vn_lock(vp, LK_SHARED | LK_RETRY);
 2154         AUDIT_ARG_VNODE1(vp);
 2155         VOP_UNLOCK(vp, 0);
 2156 #endif
 2157         return (setfmode(td, active_cred, vp, mode));
 2158 }
 2159 
 2160 int
 2161 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
 2162     struct thread *td)
 2163 {
 2164         struct vnode *vp;
 2165 
 2166         vp = fp->f_vnode;
 2167 #ifdef AUDIT
 2168         vn_lock(vp, LK_SHARED | LK_RETRY);
 2169         AUDIT_ARG_VNODE1(vp);
 2170         VOP_UNLOCK(vp, 0);
 2171 #endif
 2172         return (setfown(td, active_cred, vp, uid, gid));
 2173 }
 2174 
 2175 void
 2176 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
 2177 {
 2178         vm_object_t object;
 2179 
 2180         if ((object = vp->v_object) == NULL)
 2181                 return;
 2182         VM_OBJECT_WLOCK(object);
 2183         vm_object_page_remove(object, start, end, 0);
 2184         VM_OBJECT_WUNLOCK(object);
 2185 }
 2186 
 2187 int
 2188 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
 2189 {
 2190         struct vattr va;
 2191         daddr_t bn, bnp;
 2192         uint64_t bsize;
 2193         off_t noff;
 2194         int error;
 2195 
 2196         KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
 2197             ("Wrong command %lu", cmd));
 2198 
 2199         if (vn_lock(vp, LK_SHARED) != 0)
 2200                 return (EBADF);
 2201         if (vp->v_type != VREG) {
 2202                 error = ENOTTY;
 2203                 goto unlock;
 2204         }
 2205         error = VOP_GETATTR(vp, &va, cred);
 2206         if (error != 0)
 2207                 goto unlock;
 2208         noff = *off;
 2209         if (noff >= va.va_size) {
 2210                 error = ENXIO;
 2211                 goto unlock;
 2212         }
 2213         bsize = vp->v_mount->mnt_stat.f_iosize;
 2214         for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
 2215                 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
 2216                 if (error == EOPNOTSUPP) {
 2217                         error = ENOTTY;
 2218                         goto unlock;
 2219                 }
 2220                 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
 2221                     (bnp != -1 && cmd == FIOSEEKDATA)) {
 2222                         noff = bn * bsize;
 2223                         if (noff < *off)
 2224                                 noff = *off;
 2225                         goto unlock;
 2226                 }
 2227         }
 2228         if (noff > va.va_size)
 2229                 noff = va.va_size;
 2230         /* noff == va.va_size. There is an implicit hole at the end of file. */
 2231         if (cmd == FIOSEEKDATA)
 2232                 error = ENXIO;
 2233 unlock:
 2234         VOP_UNLOCK(vp, 0);
 2235         if (error == 0)
 2236                 *off = noff;
 2237         return (error);
 2238 }
 2239 
 2240 int
 2241 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
 2242 {
 2243         struct ucred *cred;
 2244         struct vnode *vp;
 2245         struct vattr vattr;
 2246         off_t foffset, size;
 2247         int error, noneg;
 2248 
 2249         cred = td->td_ucred;
 2250         vp = fp->f_vnode;
 2251         foffset = foffset_lock(fp, 0);
 2252         noneg = (vp->v_type != VCHR);
 2253         error = 0;
 2254         switch (whence) {
 2255         case L_INCR:
 2256                 if (noneg &&
 2257                     (foffset < 0 ||
 2258                     (offset > 0 && foffset > OFF_MAX - offset))) {
 2259                         error = EOVERFLOW;
 2260                         break;
 2261                 }
 2262                 offset += foffset;
 2263                 break;
 2264         case L_XTND:
 2265                 vn_lock(vp, LK_SHARED | LK_RETRY);
 2266                 error = VOP_GETATTR(vp, &vattr, cred);
 2267                 VOP_UNLOCK(vp, 0);
 2268                 if (error)
 2269                         break;
 2270 
 2271                 /*
 2272                  * If the file references a disk device, then fetch
 2273                  * the media size and use that to determine the ending
 2274                  * offset.
 2275                  */
 2276                 if (vattr.va_size == 0 && vp->v_type == VCHR &&
 2277                     fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
 2278                         vattr.va_size = size;
 2279                 if (noneg &&
 2280                     (vattr.va_size > OFF_MAX ||
 2281                     (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
 2282                         error = EOVERFLOW;
 2283                         break;
 2284                 }
 2285                 offset += vattr.va_size;
 2286                 break;
 2287         case L_SET:
 2288                 break;
 2289         case SEEK_DATA:
 2290                 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
 2291                 break;
 2292         case SEEK_HOLE:
 2293                 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
 2294                 break;
 2295         default:
 2296                 error = EINVAL;
 2297         }
 2298         if (error == 0 && noneg && offset < 0)
 2299                 error = EINVAL;
 2300         if (error != 0)
 2301                 goto drop;
 2302         VFS_KNOTE_UNLOCKED(vp, 0);
 2303         *(off_t *)(td->td_retval) = offset;
 2304 drop:
 2305         foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
 2306         return (error);
 2307 }
 2308 
 2309 int
 2310 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
 2311     struct thread *td)
 2312 {
 2313         int error;
 2314 
 2315         /*
 2316          * Grant permission if the caller is the owner of the file, or
 2317          * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
 2318          * on the file.  If the time pointer is null, then write
 2319          * permission on the file is also sufficient.
 2320          *
 2321          * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
 2322          * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
 2323          * will be allowed to set the times [..] to the current
 2324          * server time.
 2325          */
 2326         error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
 2327         if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
 2328                 error = VOP_ACCESS(vp, VWRITE, cred, td);
 2329         return (error);
 2330 }

Cache object: 1e89e2c7ff1bf47db73160d15520e003


[ 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.