devfs_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) 2000-2004
    3  *      Poul-Henning Kamp.  All rights reserved.
    4  * Copyright (c) 1989, 1992-1993, 1995
    5  *      The Regents of the University of California.  All rights reserved.
    6  *
    7  * This code is derived from software donated to Berkeley by
    8  * Jan-Simon Pendry.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Neither the name of the University nor the names of its contributors
   16  *    may be used to endorse or promote products derived from this software
   17  *    without specific prior written permission.
   18  *
   19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   29  * SUCH DAMAGE.
   30  *
   31  *      @(#)kernfs_vnops.c      8.15 (Berkeley) 5/21/95
   32  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
   33  *
   34  * $FreeBSD: releng/9.0/sys/fs/devfs/devfs_vnops.c 227211 2011-11-06 13:06:31Z kib $
   35  */
   36 
   37 /*
   38  * TODO:
   39  *      mkdir: want it ?
   40  */
   41 
   42 #include <sys/param.h>
   43 #include <sys/systm.h>
   44 #include <sys/conf.h>
   45 #include <sys/dirent.h>
   46 #include <sys/fcntl.h>
   47 #include <sys/file.h>
   48 #include <sys/filedesc.h>
   49 #include <sys/filio.h>
   50 #include <sys/jail.h>
   51 #include <sys/kernel.h>
   52 #include <sys/lock.h>
   53 #include <sys/malloc.h>
   54 #include <sys/mount.h>
   55 #include <sys/namei.h>
   56 #include <sys/priv.h>
   57 #include <sys/proc.h>
   58 #include <sys/stat.h>
   59 #include <sys/sx.h>
   60 #include <sys/time.h>
   61 #include <sys/ttycom.h>
   62 #include <sys/unistd.h>
   63 #include <sys/vnode.h>
   64 
   65 static struct vop_vector devfs_vnodeops;
   66 static struct vop_vector devfs_specops;
   67 static struct fileops devfs_ops_f;
   68 
   69 #include <fs/devfs/devfs.h>
   70 #include <fs/devfs/devfs_int.h>
   71 
   72 #include <security/mac/mac_framework.h>
   73 
   74 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
   75 
   76 struct mtx      devfs_de_interlock;
   77 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
   78 struct sx       clone_drain_lock;
   79 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
   80 struct mtx      cdevpriv_mtx;
   81 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
   82 
   83 static int
   84 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
   85     int *ref)
   86 {
   87 
   88         *dswp = devvn_refthread(fp->f_vnode, devp, ref);
   89         if (*devp != fp->f_data) {
   90                 if (*dswp != NULL)
   91                         dev_relthread(*devp, *ref);
   92                 return (ENXIO);
   93         }
   94         KASSERT((*devp)->si_refcount > 0,
   95             ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
   96         if (*dswp == NULL)
   97                 return (ENXIO);
   98         curthread->td_fpop = fp;
   99         return (0);
  100 }
  101 
  102 int
  103 devfs_get_cdevpriv(void **datap)
  104 {
  105         struct file *fp;
  106         struct cdev_privdata *p;
  107         int error;
  108 
  109         fp = curthread->td_fpop;
  110         if (fp == NULL)
  111                 return (EBADF);
  112         p = fp->f_cdevpriv;
  113         if (p != NULL) {
  114                 error = 0;
  115                 *datap = p->cdpd_data;
  116         } else
  117                 error = ENOENT;
  118         return (error);
  119 }
  120 
  121 int
  122 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
  123 {
  124         struct file *fp;
  125         struct cdev_priv *cdp;
  126         struct cdev_privdata *p;
  127         int error;
  128 
  129         fp = curthread->td_fpop;
  130         if (fp == NULL)
  131                 return (ENOENT);
  132         cdp = cdev2priv((struct cdev *)fp->f_data);
  133         p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
  134         p->cdpd_data = priv;
  135         p->cdpd_dtr = priv_dtr;
  136         p->cdpd_fp = fp;
  137         mtx_lock(&cdevpriv_mtx);
  138         if (fp->f_cdevpriv == NULL) {
  139                 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
  140                 fp->f_cdevpriv = p;
  141                 mtx_unlock(&cdevpriv_mtx);
  142                 error = 0;
  143         } else {
  144                 mtx_unlock(&cdevpriv_mtx);
  145                 free(p, M_CDEVPDATA);
  146                 error = EBUSY;
  147         }
  148         return (error);
  149 }
  150 
  151 void
  152 devfs_destroy_cdevpriv(struct cdev_privdata *p)
  153 {
  154 
  155         mtx_assert(&cdevpriv_mtx, MA_OWNED);
  156         p->cdpd_fp->f_cdevpriv = NULL;
  157         LIST_REMOVE(p, cdpd_list);
  158         mtx_unlock(&cdevpriv_mtx);
  159         (p->cdpd_dtr)(p->cdpd_data);
  160         free(p, M_CDEVPDATA);
  161 }
  162 
  163 void
  164 devfs_fpdrop(struct file *fp)
  165 {
  166         struct cdev_privdata *p;
  167 
  168         mtx_lock(&cdevpriv_mtx);
  169         if ((p = fp->f_cdevpriv) == NULL) {
  170                 mtx_unlock(&cdevpriv_mtx);
  171                 return;
  172         }
  173         devfs_destroy_cdevpriv(p);
  174 }
  175 
  176 void
  177 devfs_clear_cdevpriv(void)
  178 {
  179         struct file *fp;
  180 
  181         fp = curthread->td_fpop;
  182         if (fp == NULL)
  183                 return;
  184         devfs_fpdrop(fp);
  185 }
  186 
  187 /*
  188  * On success devfs_populate_vp() returns with dmp->dm_lock held.
  189  */
  190 static int
  191 devfs_populate_vp(struct vnode *vp)
  192 {
  193         struct devfs_dirent *de;
  194         struct devfs_mount *dmp;
  195         int locked;
  196 
  197         ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
  198 
  199         dmp = VFSTODEVFS(vp->v_mount);
  200         locked = VOP_ISLOCKED(vp);
  201 
  202         sx_xlock(&dmp->dm_lock);
  203         DEVFS_DMP_HOLD(dmp);
  204 
  205         /* Can't call devfs_populate() with the vnode lock held. */
  206         VOP_UNLOCK(vp, 0);
  207         devfs_populate(dmp);
  208 
  209         sx_xunlock(&dmp->dm_lock);
  210         vn_lock(vp, locked | LK_RETRY);
  211         sx_xlock(&dmp->dm_lock);
  212         if (DEVFS_DMP_DROP(dmp)) {
  213                 sx_xunlock(&dmp->dm_lock);
  214                 devfs_unmount_final(dmp);
  215                 return (EBADF);
  216         }
  217         if ((vp->v_iflag & VI_DOOMED) != 0) {
  218                 sx_xunlock(&dmp->dm_lock);
  219                 return (EBADF);
  220         }
  221         de = vp->v_data;
  222         KASSERT(de != NULL,
  223             ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
  224         if ((de->de_flags & DE_DOOMED) != 0) {
  225                 sx_xunlock(&dmp->dm_lock);
  226                 return (EBADF);
  227         }
  228 
  229         return (0);
  230 }
  231 
  232 static int
  233 devfs_vptocnp(struct vop_vptocnp_args *ap)
  234 {
  235         struct vnode *vp = ap->a_vp;
  236         struct vnode **dvp = ap->a_vpp;
  237         struct devfs_mount *dmp;
  238         char *buf = ap->a_buf;
  239         int *buflen = ap->a_buflen;
  240         struct devfs_dirent *dd, *de;
  241         int i, error;
  242 
  243         dmp = VFSTODEVFS(vp->v_mount);
  244 
  245         error = devfs_populate_vp(vp);
  246         if (error != 0)
  247                 return (error);
  248 
  249         i = *buflen;
  250         dd = vp->v_data;
  251 
  252         if (vp->v_type == VCHR) {
  253                 i -= strlen(dd->de_cdp->cdp_c.si_name);
  254                 if (i < 0) {
  255                         error = ENOMEM;
  256                         goto finished;
  257                 }
  258                 bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
  259                     strlen(dd->de_cdp->cdp_c.si_name));
  260                 de = dd->de_dir;
  261         } else if (vp->v_type == VDIR) {
  262                 if (dd == dmp->dm_rootdir) {
  263                         *dvp = vp;
  264                         vhold(*dvp);
  265                         goto finished;
  266                 }
  267                 i -= dd->de_dirent->d_namlen;
  268                 if (i < 0) {
  269                         error = ENOMEM;
  270                         goto finished;
  271                 }
  272                 bcopy(dd->de_dirent->d_name, buf + i,
  273                     dd->de_dirent->d_namlen);
  274                 de = dd;
  275         } else {
  276                 error = ENOENT;
  277                 goto finished;
  278         }
  279         *buflen = i;
  280         de = devfs_parent_dirent(de);
  281         if (de == NULL) {
  282                 error = ENOENT;
  283                 goto finished;
  284         }
  285         mtx_lock(&devfs_de_interlock);
  286         *dvp = de->de_vnode;
  287         if (*dvp != NULL) {
  288                 VI_LOCK(*dvp);
  289                 mtx_unlock(&devfs_de_interlock);
  290                 vholdl(*dvp);
  291                 VI_UNLOCK(*dvp);
  292         } else {
  293                 mtx_unlock(&devfs_de_interlock);
  294                 error = ENOENT;
  295         }
  296 finished:
  297         sx_xunlock(&dmp->dm_lock);
  298         return (error);
  299 }
  300 
  301 /*
  302  * Construct the fully qualified path name relative to the mountpoint.
  303  * If a NULL cnp is provided, no '/' is appended to the resulting path.
  304  */
  305 char *
  306 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
  307     struct componentname *cnp)
  308 {
  309         int i;
  310         struct devfs_dirent *de;
  311 
  312         sx_assert(&dmp->dm_lock, SA_LOCKED);
  313 
  314         i = SPECNAMELEN;
  315         buf[i] = '\0';
  316         if (cnp != NULL)
  317                 i -= cnp->cn_namelen;
  318         if (i < 0)
  319                  return (NULL);
  320         if (cnp != NULL)
  321                 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
  322         de = dd;
  323         while (de != dmp->dm_rootdir) {
  324                 if (cnp != NULL || i < SPECNAMELEN) {
  325                         i--;
  326                         if (i < 0)
  327                                  return (NULL);
  328                         buf[i] = '/';
  329                 }
  330                 i -= de->de_dirent->d_namlen;
  331                 if (i < 0)
  332                          return (NULL);
  333                 bcopy(de->de_dirent->d_name, buf + i,
  334                     de->de_dirent->d_namlen);
  335                 de = devfs_parent_dirent(de);
  336                 if (de == NULL)
  337                         return (NULL);
  338         }
  339         return (buf + i);
  340 }
  341 
  342 static int
  343 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
  344         struct devfs_dirent *de)
  345 {
  346         int not_found;
  347 
  348         not_found = 0;
  349         if (de->de_flags & DE_DOOMED)
  350                 not_found = 1;
  351         if (DEVFS_DE_DROP(de)) {
  352                 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
  353                 devfs_dirent_free(de);
  354         }
  355         if (DEVFS_DMP_DROP(dmp)) {
  356                 KASSERT(not_found == 1,
  357                         ("DEVFS mount struct freed before dirent"));
  358                 not_found = 2;
  359                 sx_xunlock(&dmp->dm_lock);
  360                 devfs_unmount_final(dmp);
  361         }
  362         if (not_found == 1 || (drop_dm_lock && not_found != 2))
  363                 sx_unlock(&dmp->dm_lock);
  364         return (not_found);
  365 }
  366 
  367 static void
  368 devfs_insmntque_dtr(struct vnode *vp, void *arg)
  369 {
  370         struct devfs_dirent *de;
  371 
  372         de = (struct devfs_dirent *)arg;
  373         mtx_lock(&devfs_de_interlock);
  374         vp->v_data = NULL;
  375         de->de_vnode = NULL;
  376         mtx_unlock(&devfs_de_interlock);
  377         vgone(vp);
  378         vput(vp);
  379 }
  380 
  381 /*
  382  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
  383  * it on return.
  384  */
  385 int
  386 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
  387     struct vnode **vpp)
  388 {
  389         int error;
  390         struct vnode *vp;
  391         struct cdev *dev;
  392         struct devfs_mount *dmp;
  393         struct cdevsw *dsw;
  394 
  395         dmp = VFSTODEVFS(mp);
  396         if (de->de_flags & DE_DOOMED) {
  397                 sx_xunlock(&dmp->dm_lock);
  398                 return (ENOENT);
  399         }
  400 loop:
  401         DEVFS_DE_HOLD(de);
  402         DEVFS_DMP_HOLD(dmp);
  403         mtx_lock(&devfs_de_interlock);
  404         vp = de->de_vnode;
  405         if (vp != NULL) {
  406                 VI_LOCK(vp);
  407                 mtx_unlock(&devfs_de_interlock);
  408                 sx_xunlock(&dmp->dm_lock);
  409                 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
  410                 sx_xlock(&dmp->dm_lock);
  411                 if (devfs_allocv_drop_refs(0, dmp, de)) {
  412                         vput(vp);
  413                         return (ENOENT);
  414                 }
  415                 else if ((vp->v_iflag & VI_DOOMED) != 0) {
  416                         mtx_lock(&devfs_de_interlock);
  417                         if (de->de_vnode == vp) {
  418                                 de->de_vnode = NULL;
  419                                 vp->v_data = NULL;
  420                         }
  421                         mtx_unlock(&devfs_de_interlock);
  422                         vput(vp);
  423                         goto loop;
  424                 }
  425                 sx_xunlock(&dmp->dm_lock);
  426                 *vpp = vp;
  427                 return (0);
  428         }
  429         mtx_unlock(&devfs_de_interlock);
  430         if (de->de_dirent->d_type == DT_CHR) {
  431                 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
  432                         devfs_allocv_drop_refs(1, dmp, de);
  433                         return (ENOENT);
  434                 }
  435                 dev = &de->de_cdp->cdp_c;
  436         } else {
  437                 dev = NULL;
  438         }
  439         error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
  440         if (error != 0) {
  441                 devfs_allocv_drop_refs(1, dmp, de);
  442                 printf("devfs_allocv: failed to allocate new vnode\n");
  443                 return (error);
  444         }
  445 
  446         if (de->de_dirent->d_type == DT_CHR) {
  447                 vp->v_type = VCHR;
  448                 VI_LOCK(vp);
  449                 dev_lock();
  450                 dev_refl(dev);
  451                 /* XXX: v_rdev should be protect by vnode lock */
  452                 vp->v_rdev = dev;
  453                 KASSERT(vp->v_usecount == 1,
  454                     ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
  455                 dev->si_usecount += vp->v_usecount;
  456                 /* Special casing of ttys for deadfs.  Probably redundant. */
  457                 dsw = dev->si_devsw;
  458                 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
  459                         vp->v_vflag |= VV_ISTTY;
  460                 dev_unlock();
  461                 VI_UNLOCK(vp);
  462                 if ((dev->si_flags & SI_ETERNAL) != 0)
  463                         vp->v_vflag |= VV_ETERNALDEV;
  464                 vp->v_op = &devfs_specops;
  465         } else if (de->de_dirent->d_type == DT_DIR) {
  466                 vp->v_type = VDIR;
  467         } else if (de->de_dirent->d_type == DT_LNK) {
  468                 vp->v_type = VLNK;
  469         } else {
  470                 vp->v_type = VBAD;
  471         }
  472         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
  473         VN_LOCK_ASHARE(vp);
  474         mtx_lock(&devfs_de_interlock);
  475         vp->v_data = de;
  476         de->de_vnode = vp;
  477         mtx_unlock(&devfs_de_interlock);
  478         error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
  479         if (error != 0) {
  480                 (void) devfs_allocv_drop_refs(1, dmp, de);
  481                 return (error);
  482         }
  483         if (devfs_allocv_drop_refs(0, dmp, de)) {
  484                 vput(vp);
  485                 return (ENOENT);
  486         }
  487 #ifdef MAC
  488         mac_devfs_vnode_associate(mp, de, vp);
  489 #endif
  490         sx_xunlock(&dmp->dm_lock);
  491         *vpp = vp;
  492         return (0);
  493 }
  494 
  495 static int
  496 devfs_access(struct vop_access_args *ap)
  497 {
  498         struct vnode *vp = ap->a_vp;
  499         struct devfs_dirent *de;
  500         int error;
  501 
  502         de = vp->v_data;
  503         if (vp->v_type == VDIR)
  504                 de = de->de_dir;
  505 
  506         error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
  507             ap->a_accmode, ap->a_cred, NULL);
  508         if (error == 0)
  509                 return (0);
  510         if (error != EACCES)
  511                 return (error);
  512         /* We do, however, allow access to the controlling terminal */
  513         if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
  514                 return (error);
  515         if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp)
  516                 return (0);
  517         return (error);
  518 }
  519 
  520 /* ARGSUSED */
  521 static int
  522 devfs_close(struct vop_close_args *ap)
  523 {
  524         struct vnode *vp = ap->a_vp, *oldvp;
  525         struct thread *td = ap->a_td;
  526         struct cdev *dev = vp->v_rdev;
  527         struct cdevsw *dsw;
  528         int vp_locked, error, ref;
  529 
  530         /*
  531          * XXX: Don't call d_close() if we were called because of
  532          * XXX: insmntque1() failure.
  533          */
  534         if (vp->v_data == NULL)
  535                 return (0);
  536 
  537         /*
  538          * Hack: a tty device that is a controlling terminal
  539          * has a reference from the session structure.
  540          * We cannot easily tell that a character device is
  541          * a controlling terminal, unless it is the closing
  542          * process' controlling terminal.  In that case,
  543          * if the reference count is 2 (this last descriptor
  544          * plus the session), release the reference from the session.
  545          */
  546         oldvp = NULL;
  547         sx_xlock(&proctree_lock);
  548         if (td && vp == td->td_proc->p_session->s_ttyvp) {
  549                 SESS_LOCK(td->td_proc->p_session);
  550                 VI_LOCK(vp);
  551                 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
  552                         td->td_proc->p_session->s_ttyvp = NULL;
  553                         td->td_proc->p_session->s_ttydp = NULL;
  554                         oldvp = vp;
  555                 }
  556                 VI_UNLOCK(vp);
  557                 SESS_UNLOCK(td->td_proc->p_session);
  558         }
  559         sx_xunlock(&proctree_lock);
  560         if (oldvp != NULL)
  561                 vrele(oldvp);
  562         /*
  563          * We do not want to really close the device if it
  564          * is still in use unless we are trying to close it
  565          * forcibly. Since every use (buffer, vnode, swap, cmap)
  566          * holds a reference to the vnode, and because we mark
  567          * any other vnodes that alias this device, when the
  568          * sum of the reference counts on all the aliased
  569          * vnodes descends to one, we are on last close.
  570          */
  571         dsw = dev_refthread(dev, &ref);
  572         if (dsw == NULL)
  573                 return (ENXIO);
  574         VI_LOCK(vp);
  575         if (vp->v_iflag & VI_DOOMED) {
  576                 /* Forced close. */
  577         } else if (dsw->d_flags & D_TRACKCLOSE) {
  578                 /* Keep device updated on status. */
  579         } else if (count_dev(dev) > 1) {
  580                 VI_UNLOCK(vp);
  581                 dev_relthread(dev, ref);
  582                 return (0);
  583         }
  584         vholdl(vp);
  585         VI_UNLOCK(vp);
  586         vp_locked = VOP_ISLOCKED(vp);
  587         VOP_UNLOCK(vp, 0);
  588         KASSERT(dev->si_refcount > 0,
  589             ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
  590         error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
  591         dev_relthread(dev, ref);
  592         vn_lock(vp, vp_locked | LK_RETRY);
  593         vdrop(vp);
  594         return (error);
  595 }
  596 
  597 static int
  598 devfs_close_f(struct file *fp, struct thread *td)
  599 {
  600         int error;
  601         struct file *fpop;
  602 
  603         fpop = td->td_fpop;
  604         td->td_fpop = fp;
  605         error = vnops.fo_close(fp, td);
  606         td->td_fpop = fpop;
  607         return (error);
  608 }
  609 
  610 static int
  611 devfs_fsync(struct vop_fsync_args *ap)
  612 {
  613         int error;
  614         struct bufobj *bo;
  615         struct devfs_dirent *de;
  616 
  617         if (!vn_isdisk(ap->a_vp, &error)) {
  618                 bo = &ap->a_vp->v_bufobj;
  619                 de = ap->a_vp->v_data;
  620                 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
  621                         printf("Device %s went missing before all of the data "
  622                             "could be written to it; expect data loss.\n",
  623                             de->de_dirent->d_name);
  624 
  625                         error = vop_stdfsync(ap);
  626                         if (bo->bo_dirty.bv_cnt != 0 || error != 0)
  627                                 panic("devfs_fsync: vop_stdfsync failed.");
  628                 }
  629 
  630                 return (0);
  631         }
  632 
  633         return (vop_stdfsync(ap));
  634 }
  635 
  636 static int
  637 devfs_getattr(struct vop_getattr_args *ap)
  638 {
  639         struct vnode *vp = ap->a_vp;
  640         struct vattr *vap = ap->a_vap;
  641         int error;
  642         struct devfs_dirent *de;
  643         struct devfs_mount *dmp;
  644         struct cdev *dev;
  645 
  646         error = devfs_populate_vp(vp);
  647         if (error != 0)
  648                 return (error);
  649 
  650         dmp = VFSTODEVFS(vp->v_mount);
  651         sx_xunlock(&dmp->dm_lock);
  652 
  653         de = vp->v_data;
  654         KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
  655         if (vp->v_type == VDIR) {
  656                 de = de->de_dir;
  657                 KASSERT(de != NULL,
  658                     ("Null dir dirent in devfs_getattr vp=%p", vp));
  659         }
  660         vap->va_uid = de->de_uid;
  661         vap->va_gid = de->de_gid;
  662         vap->va_mode = de->de_mode;
  663         if (vp->v_type == VLNK)
  664                 vap->va_size = strlen(de->de_symlink);
  665         else if (vp->v_type == VDIR)
  666                 vap->va_size = vap->va_bytes = DEV_BSIZE;
  667         else
  668                 vap->va_size = 0;
  669         if (vp->v_type != VDIR)
  670                 vap->va_bytes = 0;
  671         vap->va_blocksize = DEV_BSIZE;
  672         vap->va_type = vp->v_type;
  673 
  674 #define fix(aa)                                                 \
  675         do {                                                    \
  676                 if ((aa).tv_sec <= 3600) {                      \
  677                         (aa).tv_sec = boottime.tv_sec;          \
  678                         (aa).tv_nsec = boottime.tv_usec * 1000; \
  679                 }                                               \
  680         } while (0)
  681 
  682         if (vp->v_type != VCHR)  {
  683                 fix(de->de_atime);
  684                 vap->va_atime = de->de_atime;
  685                 fix(de->de_mtime);
  686                 vap->va_mtime = de->de_mtime;
  687                 fix(de->de_ctime);
  688                 vap->va_ctime = de->de_ctime;
  689         } else {
  690                 dev = vp->v_rdev;
  691                 fix(dev->si_atime);
  692                 vap->va_atime = dev->si_atime;
  693                 fix(dev->si_mtime);
  694                 vap->va_mtime = dev->si_mtime;
  695                 fix(dev->si_ctime);
  696                 vap->va_ctime = dev->si_ctime;
  697 
  698                 vap->va_rdev = cdev2priv(dev)->cdp_inode;
  699         }
  700         vap->va_gen = 0;
  701         vap->va_flags = 0;
  702         vap->va_filerev = 0;
  703         vap->va_nlink = de->de_links;
  704         vap->va_fileid = de->de_inode;
  705 
  706         return (error);
  707 }
  708 
  709 /* ARGSUSED */
  710 static int
  711 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
  712 {
  713         struct cdev *dev;
  714         struct cdevsw *dsw;
  715         struct vnode *vp;
  716         struct vnode *vpold;
  717         int error, i, ref;
  718         const char *p;
  719         struct fiodgname_arg *fgn;
  720         struct file *fpop;
  721 
  722         fpop = td->td_fpop;
  723         error = devfs_fp_check(fp, &dev, &dsw, &ref);
  724         if (error)
  725                 return (error);
  726 
  727         if (com == FIODTYPE) {
  728                 *(int *)data = dsw->d_flags & D_TYPEMASK;
  729                 td->td_fpop = fpop;
  730                 dev_relthread(dev, ref);
  731                 return (0);
  732         } else if (com == FIODGNAME) {
  733                 fgn = data;
  734                 p = devtoname(dev);
  735                 i = strlen(p) + 1;
  736                 if (i > fgn->len)
  737                         error = EINVAL;
  738                 else
  739                         error = copyout(p, fgn->buf, i);
  740                 td->td_fpop = fpop;
  741                 dev_relthread(dev, ref);
  742                 return (error);
  743         }
  744         error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
  745         td->td_fpop = NULL;
  746         dev_relthread(dev, ref);
  747         if (error == ENOIOCTL)
  748                 error = ENOTTY;
  749         if (error == 0 && com == TIOCSCTTY) {
  750                 vp = fp->f_vnode;
  751 
  752                 /* Do nothing if reassigning same control tty */
  753                 sx_slock(&proctree_lock);
  754                 if (td->td_proc->p_session->s_ttyvp == vp) {
  755                         sx_sunlock(&proctree_lock);
  756                         return (0);
  757                 }
  758 
  759                 vpold = td->td_proc->p_session->s_ttyvp;
  760                 VREF(vp);
  761                 SESS_LOCK(td->td_proc->p_session);
  762                 td->td_proc->p_session->s_ttyvp = vp;
  763                 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
  764                 SESS_UNLOCK(td->td_proc->p_session);
  765 
  766                 sx_sunlock(&proctree_lock);
  767 
  768                 /* Get rid of reference to old control tty */
  769                 if (vpold)
  770                         vrele(vpold);
  771         }
  772         return (error);
  773 }
  774 
  775 /* ARGSUSED */
  776 static int
  777 devfs_kqfilter_f(struct file *fp, struct knote *kn)
  778 {
  779         struct cdev *dev;
  780         struct cdevsw *dsw;
  781         int error, ref;
  782         struct file *fpop;
  783         struct thread *td;
  784 
  785         td = curthread;
  786         fpop = td->td_fpop;
  787         error = devfs_fp_check(fp, &dev, &dsw, &ref);
  788         if (error)
  789                 return (error);
  790         error = dsw->d_kqfilter(dev, kn);
  791         td->td_fpop = fpop;
  792         dev_relthread(dev, ref);
  793         return (error);
  794 }
  795 
  796 static inline int
  797 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
  798 {
  799         struct cdev_priv *cdp;
  800         struct ucred *dcr;
  801         int error;
  802 
  803         cdp = de->de_cdp;
  804         if (cdp == NULL)
  805                 return (0);
  806         dcr = cdp->cdp_c.si_cred;
  807         if (dcr == NULL)
  808                 return (0);
  809 
  810         error = prison_check(td->td_ucred, dcr);
  811         if (error == 0)
  812                 return (0);
  813         /* We do, however, allow access to the controlling terminal */
  814         if (!(td->td_proc->p_flag & P_CONTROLT))
  815                 return (error);
  816         if (td->td_proc->p_session->s_ttydp == cdp)
  817                 return (0);
  818         return (error);
  819 }
  820 
  821 static int
  822 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
  823 {
  824         struct componentname *cnp;
  825         struct vnode *dvp, **vpp;
  826         struct thread *td;
  827         struct devfs_dirent *de, *dd;
  828         struct devfs_dirent **dde;
  829         struct devfs_mount *dmp;
  830         struct cdev *cdev;
  831         int error, flags, nameiop, dvplocked;
  832         char specname[SPECNAMELEN + 1], *pname;
  833 
  834         cnp = ap->a_cnp;
  835         vpp = ap->a_vpp;
  836         dvp = ap->a_dvp;
  837         pname = cnp->cn_nameptr;
  838         td = cnp->cn_thread;
  839         flags = cnp->cn_flags;
  840         nameiop = cnp->cn_nameiop;
  841         dmp = VFSTODEVFS(dvp->v_mount);
  842         dd = dvp->v_data;
  843         *vpp = NULLVP;
  844 
  845         if ((flags & ISLASTCN) && nameiop == RENAME)
  846                 return (EOPNOTSUPP);
  847 
  848         if (dvp->v_type != VDIR)
  849                 return (ENOTDIR);
  850 
  851         if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
  852                 return (EIO);
  853 
  854         error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
  855         if (error)
  856                 return (error);
  857 
  858         if (cnp->cn_namelen == 1 && *pname == '.') {
  859                 if ((flags & ISLASTCN) && nameiop != LOOKUP)
  860                         return (EINVAL);
  861                 *vpp = dvp;
  862                 VREF(dvp);
  863                 return (0);
  864         }
  865 
  866         if (flags & ISDOTDOT) {
  867                 if ((flags & ISLASTCN) && nameiop != LOOKUP)
  868                         return (EINVAL);
  869                 de = devfs_parent_dirent(dd);
  870                 if (de == NULL)
  871                         return (ENOENT);
  872                 dvplocked = VOP_ISLOCKED(dvp);
  873                 VOP_UNLOCK(dvp, 0);
  874                 error = devfs_allocv(de, dvp->v_mount,
  875                     cnp->cn_lkflags & LK_TYPE_MASK, vpp);
  876                 *dm_unlock = 0;
  877                 vn_lock(dvp, dvplocked | LK_RETRY);
  878                 return (error);
  879         }
  880 
  881         dd = dvp->v_data;
  882         de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
  883         while (de == NULL) {    /* While(...) so we can use break */
  884 
  885                 if (nameiop == DELETE)
  886                         return (ENOENT);
  887 
  888                 /*
  889                  * OK, we didn't have an entry for the name we were asked for
  890                  * so we try to see if anybody can create it on demand.
  891                  */
  892                 pname = devfs_fqpn(specname, dmp, dd, cnp);
  893                 if (pname == NULL)
  894                         break;
  895 
  896                 cdev = NULL;
  897                 DEVFS_DMP_HOLD(dmp);
  898                 sx_xunlock(&dmp->dm_lock);
  899                 sx_slock(&clone_drain_lock);
  900                 EVENTHANDLER_INVOKE(dev_clone,
  901                     td->td_ucred, pname, strlen(pname), &cdev);
  902                 sx_sunlock(&clone_drain_lock);
  903 
  904                 if (cdev == NULL)
  905                         sx_xlock(&dmp->dm_lock);
  906                 else if (devfs_populate_vp(dvp) != 0) {
  907                         *dm_unlock = 0;
  908                         sx_xlock(&dmp->dm_lock);
  909                         if (DEVFS_DMP_DROP(dmp)) {
  910                                 sx_xunlock(&dmp->dm_lock);
  911                                 devfs_unmount_final(dmp);
  912                         } else
  913                                 sx_xunlock(&dmp->dm_lock);
  914                         dev_rel(cdev);
  915                         return (ENOENT);
  916                 }
  917                 if (DEVFS_DMP_DROP(dmp)) {
  918                         *dm_unlock = 0;
  919                         sx_xunlock(&dmp->dm_lock);
  920                         devfs_unmount_final(dmp);
  921                         if (cdev != NULL)
  922                                 dev_rel(cdev);
  923                         return (ENOENT);
  924                 }
  925 
  926                 if (cdev == NULL)
  927                         break;
  928 
  929                 dev_lock();
  930                 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
  931                 if (dde != NULL && *dde != NULL)
  932                         de = *dde;
  933                 dev_unlock();
  934                 dev_rel(cdev);
  935                 break;
  936         }
  937 
  938         if (de == NULL || de->de_flags & DE_WHITEOUT) {
  939                 if ((nameiop == CREATE || nameiop == RENAME) &&
  940                     (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
  941                         cnp->cn_flags |= SAVENAME;
  942                         return (EJUSTRETURN);
  943                 }
  944                 return (ENOENT);
  945         }
  946 
  947         if (devfs_prison_check(de, td))
  948                 return (ENOENT);
  949 
  950         if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
  951                 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
  952                 if (error)
  953                         return (error);
  954                 if (*vpp == dvp) {
  955                         VREF(dvp);
  956                         *vpp = dvp;
  957                         return (0);
  958                 }
  959         }
  960         error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
  961             vpp);
  962         *dm_unlock = 0;
  963         return (error);
  964 }
  965 
  966 static int
  967 devfs_lookup(struct vop_lookup_args *ap)
  968 {
  969         int j;
  970         struct devfs_mount *dmp;
  971         int dm_unlock;
  972 
  973         if (devfs_populate_vp(ap->a_dvp) != 0)
  974                 return (ENOTDIR);
  975 
  976         dmp = VFSTODEVFS(ap->a_dvp->v_mount);
  977         dm_unlock = 1;
  978         j = devfs_lookupx(ap, &dm_unlock);
  979         if (dm_unlock == 1)
  980                 sx_xunlock(&dmp->dm_lock);
  981         return (j);
  982 }
  983 
  984 static int
  985 devfs_mknod(struct vop_mknod_args *ap)
  986 {
  987         struct componentname *cnp;
  988         struct vnode *dvp, **vpp;
  989         struct devfs_dirent *dd, *de;
  990         struct devfs_mount *dmp;
  991         int error;
  992 
  993         /*
  994          * The only type of node we should be creating here is a
  995          * character device, for anything else return EOPNOTSUPP.
  996          */
  997         if (ap->a_vap->va_type != VCHR)
  998                 return (EOPNOTSUPP);
  999         dvp = ap->a_dvp;
 1000         dmp = VFSTODEVFS(dvp->v_mount);
 1001 
 1002         cnp = ap->a_cnp;
 1003         vpp = ap->a_vpp;
 1004         dd = dvp->v_data;
 1005 
 1006         error = ENOENT;
 1007         sx_xlock(&dmp->dm_lock);
 1008         TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
 1009                 if (cnp->cn_namelen != de->de_dirent->d_namlen)
 1010                         continue;
 1011                 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
 1012                     de->de_dirent->d_namlen) != 0)
 1013                         continue;
 1014                 if (de->de_flags & DE_WHITEOUT)
 1015                         break;
 1016                 goto notfound;
 1017         }
 1018         if (de == NULL)
 1019                 goto notfound;
 1020         de->de_flags &= ~DE_WHITEOUT;
 1021         error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
 1022         return (error);
 1023 notfound:
 1024         sx_xunlock(&dmp->dm_lock);
 1025         return (error);
 1026 }
 1027 
 1028 /* ARGSUSED */
 1029 static int
 1030 devfs_open(struct vop_open_args *ap)
 1031 {
 1032         struct thread *td = ap->a_td;
 1033         struct vnode *vp = ap->a_vp;
 1034         struct cdev *dev = vp->v_rdev;
 1035         struct file *fp = ap->a_fp;
 1036         int error, ref, vlocked;
 1037         struct cdevsw *dsw;
 1038         struct file *fpop;
 1039 
 1040         if (vp->v_type == VBLK)
 1041                 return (ENXIO);
 1042 
 1043         if (dev == NULL)
 1044                 return (ENXIO);
 1045 
 1046         /* Make this field valid before any I/O in d_open. */
 1047         if (dev->si_iosize_max == 0)
 1048                 dev->si_iosize_max = DFLTPHYS;
 1049 
 1050         dsw = dev_refthread(dev, &ref);
 1051         if (dsw == NULL)
 1052                 return (ENXIO);
 1053         if (fp == NULL && dsw->d_fdopen != NULL) {
 1054                 dev_relthread(dev, ref);
 1055                 return (ENXIO);
 1056         }
 1057 
 1058         vlocked = VOP_ISLOCKED(vp);
 1059         VOP_UNLOCK(vp, 0);
 1060 
 1061         fpop = td->td_fpop;
 1062         td->td_fpop = fp;
 1063         if (fp != NULL) {
 1064                 fp->f_data = dev;
 1065                 fp->f_vnode = vp;
 1066         }
 1067         if (dsw->d_fdopen != NULL)
 1068                 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
 1069         else
 1070                 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
 1071         td->td_fpop = fpop;
 1072 
 1073         vn_lock(vp, vlocked | LK_RETRY);
 1074         dev_relthread(dev, ref);
 1075         if (error)
 1076                 return (error);
 1077 
 1078 #if 0   /* /dev/console */
 1079         KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
 1080 #else
 1081         if (fp == NULL)
 1082                 return (error);
 1083 #endif
 1084         if (fp->f_ops == &badfileops)
 1085                 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
 1086         return (error);
 1087 }
 1088 
 1089 static int
 1090 devfs_pathconf(struct vop_pathconf_args *ap)
 1091 {
 1092 
 1093         switch (ap->a_name) {
 1094         case _PC_MAC_PRESENT:
 1095 #ifdef MAC
 1096                 /*
 1097                  * If MAC is enabled, devfs automatically supports
 1098                  * trivial non-persistant label storage.
 1099                  */
 1100                 *ap->a_retval = 1;
 1101 #else
 1102                 *ap->a_retval = 0;
 1103 #endif
 1104                 return (0);
 1105         default:
 1106                 return (vop_stdpathconf(ap));
 1107         }
 1108         /* NOTREACHED */
 1109 }
 1110 
 1111 /* ARGSUSED */
 1112 static int
 1113 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
 1114 {
 1115         struct cdev *dev;
 1116         struct cdevsw *dsw;
 1117         int error, ref;
 1118         struct file *fpop;
 1119 
 1120         fpop = td->td_fpop;
 1121         error = devfs_fp_check(fp, &dev, &dsw, &ref);
 1122         if (error)
 1123                 return (poll_no_poll(events));
 1124         error = dsw->d_poll(dev, events, td);
 1125         td->td_fpop = fpop;
 1126         dev_relthread(dev, ref);
 1127         return(error);
 1128 }
 1129 
 1130 /*
 1131  * Print out the contents of a special device vnode.
 1132  */
 1133 static int
 1134 devfs_print(struct vop_print_args *ap)
 1135 {
 1136 
 1137         printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
 1138         return (0);
 1139 }
 1140 
 1141 /* ARGSUSED */
 1142 static int
 1143 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
 1144 {
 1145         struct cdev *dev;
 1146         int ioflag, error, ref, resid;
 1147         struct cdevsw *dsw;
 1148         struct file *fpop;
 1149 
 1150         fpop = td->td_fpop;
 1151         error = devfs_fp_check(fp, &dev, &dsw, &ref);
 1152         if (error)
 1153                 return (error);
 1154         resid = uio->uio_resid;
 1155         ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
 1156         if (ioflag & O_DIRECT)
 1157                 ioflag |= IO_DIRECT;
 1158 
 1159         if ((flags & FOF_OFFSET) == 0)
 1160                 uio->uio_offset = fp->f_offset;
 1161 
 1162         error = dsw->d_read(dev, uio, ioflag);
 1163         if (uio->uio_resid != resid || (error == 0 && resid != 0))
 1164                 vfs_timestamp(&dev->si_atime);
 1165         td->td_fpop = fpop;
 1166         dev_relthread(dev, ref);
 1167 
 1168         if ((flags & FOF_OFFSET) == 0)
 1169                 fp->f_offset = uio->uio_offset;
 1170         fp->f_nextoff = uio->uio_offset;
 1171         return (error);
 1172 }
 1173 
 1174 static int
 1175 devfs_readdir(struct vop_readdir_args *ap)
 1176 {
 1177         int error;
 1178         struct uio *uio;
 1179         struct dirent *dp;
 1180         struct devfs_dirent *dd;
 1181         struct devfs_dirent *de;
 1182         struct devfs_mount *dmp;
 1183         off_t off;
 1184         int *tmp_ncookies = NULL;
 1185 
 1186         if (ap->a_vp->v_type != VDIR)
 1187                 return (ENOTDIR);
 1188 
 1189         uio = ap->a_uio;
 1190         if (uio->uio_offset < 0)
 1191                 return (EINVAL);
 1192 
 1193         /*
 1194          * XXX: This is a temporary hack to get around this filesystem not
 1195          * supporting cookies. We store the location of the ncookies pointer
 1196          * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
 1197          * and set the number of cookies to 0. We then set the pointer to
 1198          * NULL so that vfs_read_dirent doesn't try to call realloc() on 
 1199          * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
 1200          * pointer to its original location before returning to the caller.
 1201          */
 1202         if (ap->a_ncookies != NULL) {
 1203                 tmp_ncookies = ap->a_ncookies;
 1204                 *ap->a_ncookies = 0;
 1205                 ap->a_ncookies = NULL;
 1206         }
 1207 
 1208         dmp = VFSTODEVFS(ap->a_vp->v_mount);
 1209         if (devfs_populate_vp(ap->a_vp) != 0) {
 1210                 if (tmp_ncookies != NULL)
 1211                         ap->a_ncookies = tmp_ncookies;
 1212                 return (EIO);
 1213         }
 1214         error = 0;
 1215         de = ap->a_vp->v_data;
 1216         off = 0;
 1217         TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
 1218                 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
 1219                 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
 1220                         continue;
 1221                 if (devfs_prison_check(dd, uio->uio_td))
 1222                         continue;
 1223                 if (dd->de_dirent->d_type == DT_DIR)
 1224                         de = dd->de_dir;
 1225                 else
 1226                         de = dd;
 1227                 dp = dd->de_dirent;
 1228                 if (dp->d_reclen > uio->uio_resid)
 1229                         break;
 1230                 dp->d_fileno = de->de_inode;
 1231                 if (off >= uio->uio_offset) {
 1232                         error = vfs_read_dirent(ap, dp, off);
 1233                         if (error)
 1234                                 break;
 1235                 }
 1236                 off += dp->d_reclen;
 1237         }
 1238         sx_xunlock(&dmp->dm_lock);
 1239         uio->uio_offset = off;
 1240 
 1241         /*
 1242          * Restore ap->a_ncookies if it wasn't originally NULL in the first
 1243          * place.
 1244          */
 1245         if (tmp_ncookies != NULL)
 1246                 ap->a_ncookies = tmp_ncookies;
 1247 
 1248         return (error);
 1249 }
 1250 
 1251 static int
 1252 devfs_readlink(struct vop_readlink_args *ap)
 1253 {
 1254         struct devfs_dirent *de;
 1255 
 1256         de = ap->a_vp->v_data;
 1257         return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
 1258 }
 1259 
 1260 static int
 1261 devfs_reclaim(struct vop_reclaim_args *ap)
 1262 {
 1263         struct vnode *vp = ap->a_vp;
 1264         struct devfs_dirent *de;
 1265         struct cdev *dev;
 1266 
 1267         mtx_lock(&devfs_de_interlock);
 1268         de = vp->v_data;
 1269         if (de != NULL) {
 1270                 de->de_vnode = NULL;
 1271                 vp->v_data = NULL;
 1272         }
 1273         mtx_unlock(&devfs_de_interlock);
 1274 
 1275         vnode_destroy_vobject(vp);
 1276 
 1277         VI_LOCK(vp);
 1278         dev_lock();
 1279         dev = vp->v_rdev;
 1280         vp->v_rdev = NULL;
 1281 
 1282         if (dev == NULL) {
 1283                 dev_unlock();
 1284                 VI_UNLOCK(vp);
 1285                 return (0);
 1286         }
 1287 
 1288         dev->si_usecount -= vp->v_usecount;
 1289         dev_unlock();
 1290         VI_UNLOCK(vp);
 1291         dev_rel(dev);
 1292         return (0);
 1293 }
 1294 
 1295 static int
 1296 devfs_remove(struct vop_remove_args *ap)
 1297 {
 1298         struct vnode *dvp = ap->a_dvp;
 1299         struct vnode *vp = ap->a_vp;
 1300         struct devfs_dirent *dd;
 1301         struct devfs_dirent *de, *de_covered;
 1302         struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
 1303 
 1304         ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
 1305         ASSERT_VOP_ELOCKED(vp, "devfs_remove");
 1306 
 1307         sx_xlock(&dmp->dm_lock);
 1308         dd = ap->a_dvp->v_data;
 1309         de = vp->v_data;
 1310         if (de->de_cdp == NULL) {
 1311                 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
 1312                 if (de->de_dirent->d_type == DT_LNK) {
 1313                         de_covered = devfs_find(dd, de->de_dirent->d_name,
 1314                             de->de_dirent->d_namlen, 0);
 1315                         if (de_covered != NULL)
 1316                                 de_covered->de_flags &= ~DE_COVERED;
 1317                 }
 1318                 /* We need to unlock dvp because devfs_delete() may lock it. */
 1319                 VOP_UNLOCK(vp, 0);
 1320                 if (dvp != vp)
 1321                         VOP_UNLOCK(dvp, 0);
 1322                 devfs_delete(dmp, de, 0);
 1323                 sx_xunlock(&dmp->dm_lock);
 1324                 if (dvp != vp)
 1325                         vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
 1326                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1327         } else {
 1328                 de->de_flags |= DE_WHITEOUT;
 1329                 sx_xunlock(&dmp->dm_lock);
 1330         }
 1331         return (0);
 1332 }
 1333 
 1334 /*
 1335  * Revoke is called on a tty when a terminal session ends.  The vnode
 1336  * is orphaned by setting v_op to deadfs so we need to let go of it
 1337  * as well so that we create a new one next time around.
 1338  *
 1339  */
 1340 static int
 1341 devfs_revoke(struct vop_revoke_args *ap)
 1342 {
 1343         struct vnode *vp = ap->a_vp, *vp2;
 1344         struct cdev *dev;
 1345         struct cdev_priv *cdp;
 1346         struct devfs_dirent *de;
 1347         int i;
 1348 
 1349         KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
 1350 
 1351         dev = vp->v_rdev;
 1352         cdp = cdev2priv(dev);
 1353  
 1354         dev_lock();
 1355         cdp->cdp_inuse++;
 1356         dev_unlock();
 1357 
 1358         vhold(vp);
 1359         vgone(vp);
 1360         vdrop(vp);
 1361 
 1362         VOP_UNLOCK(vp,0);
 1363  loop:
 1364         for (;;) {
 1365                 mtx_lock(&devfs_de_interlock);
 1366                 dev_lock();
 1367                 vp2 = NULL;
 1368                 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
 1369                         de = cdp->cdp_dirents[i];
 1370                         if (de == NULL)
 1371                                 continue;
 1372 
 1373                         vp2 = de->de_vnode;
 1374                         if (vp2 != NULL) {
 1375                                 dev_unlock();
 1376                                 VI_LOCK(vp2);
 1377                                 mtx_unlock(&devfs_de_interlock);
 1378                                 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
 1379                                     curthread))
 1380                                         goto loop;
 1381                                 vhold(vp2);
 1382                                 vgone(vp2);
 1383                                 vdrop(vp2);
 1384                                 vput(vp2);
 1385                                 break;
 1386                         } 
 1387                 }
 1388                 if (vp2 != NULL) {
 1389                         continue;
 1390                 }
 1391                 dev_unlock();
 1392                 mtx_unlock(&devfs_de_interlock);
 1393                 break;
 1394         }
 1395         dev_lock();
 1396         cdp->cdp_inuse--;
 1397         if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
 1398                 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
 1399                 dev_unlock();
 1400                 dev_rel(&cdp->cdp_c);
 1401         } else
 1402                 dev_unlock();
 1403 
 1404         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1405         return (0);
 1406 }
 1407 
 1408 static int
 1409 devfs_rioctl(struct vop_ioctl_args *ap)
 1410 {
 1411         struct vnode *vp;
 1412         struct devfs_mount *dmp;
 1413         int error;
 1414 
 1415         vp = ap->a_vp;
 1416         vn_lock(vp, LK_SHARED | LK_RETRY);
 1417         if (vp->v_iflag & VI_DOOMED) {
 1418                 VOP_UNLOCK(vp, 0);
 1419                 return (EBADF);
 1420         }
 1421         dmp = VFSTODEVFS(vp->v_mount);
 1422         sx_xlock(&dmp->dm_lock);
 1423         VOP_UNLOCK(vp, 0);
 1424         DEVFS_DMP_HOLD(dmp);
 1425         devfs_populate(dmp);
 1426         if (DEVFS_DMP_DROP(dmp)) {
 1427                 sx_xunlock(&dmp->dm_lock);
 1428                 devfs_unmount_final(dmp);
 1429                 return (ENOENT);
 1430         }
 1431         error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
 1432         sx_xunlock(&dmp->dm_lock);
 1433         return (error);
 1434 }
 1435 
 1436 static int
 1437 devfs_rread(struct vop_read_args *ap)
 1438 {
 1439 
 1440         if (ap->a_vp->v_type != VDIR)
 1441                 return (EINVAL);
 1442         return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
 1443 }
 1444 
 1445 static int
 1446 devfs_setattr(struct vop_setattr_args *ap)
 1447 {
 1448         struct devfs_dirent *de;
 1449         struct vattr *vap;
 1450         struct vnode *vp;
 1451         struct thread *td;
 1452         int c, error;
 1453         uid_t uid;
 1454         gid_t gid;
 1455 
 1456         vap = ap->a_vap;
 1457         vp = ap->a_vp;
 1458         td = curthread;
 1459         if ((vap->va_type != VNON) ||
 1460             (vap->va_nlink != VNOVAL) ||
 1461             (vap->va_fsid != VNOVAL) ||
 1462             (vap->va_fileid != VNOVAL) ||
 1463             (vap->va_blocksize != VNOVAL) ||
 1464             (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
 1465             (vap->va_rdev != VNOVAL) ||
 1466             ((int)vap->va_bytes != VNOVAL) ||
 1467             (vap->va_gen != VNOVAL)) {
 1468                 return (EINVAL);
 1469         }
 1470 
 1471         de = vp->v_data;
 1472         if (vp->v_type == VDIR)
 1473                 de = de->de_dir;
 1474 
 1475         error = c = 0;
 1476         if (vap->va_uid == (uid_t)VNOVAL)
 1477                 uid = de->de_uid;
 1478         else
 1479                 uid = vap->va_uid;
 1480         if (vap->va_gid == (gid_t)VNOVAL)
 1481                 gid = de->de_gid;
 1482         else
 1483                 gid = vap->va_gid;
 1484         if (uid != de->de_uid || gid != de->de_gid) {
 1485                 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
 1486                     (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
 1487                         error = priv_check(td, PRIV_VFS_CHOWN);
 1488                         if (error)
 1489                                 return (error);
 1490                 }
 1491                 de->de_uid = uid;
 1492                 de->de_gid = gid;
 1493                 c = 1;
 1494         }
 1495 
 1496         if (vap->va_mode != (mode_t)VNOVAL) {
 1497                 if (ap->a_cred->cr_uid != de->de_uid) {
 1498                         error = priv_check(td, PRIV_VFS_ADMIN);
 1499                         if (error)
 1500                                 return (error);
 1501                 }
 1502                 de->de_mode = vap->va_mode;
 1503                 c = 1;
 1504         }
 1505 
 1506         if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
 1507                 /* See the comment in ufs_vnops::ufs_setattr(). */
 1508                 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
 1509                     ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
 1510                     (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
 1511                         return (error);
 1512                 if (vap->va_atime.tv_sec != VNOVAL) {
 1513                         if (vp->v_type == VCHR)
 1514                                 vp->v_rdev->si_atime = vap->va_atime;
 1515                         else
 1516                                 de->de_atime = vap->va_atime;
 1517                 }
 1518                 if (vap->va_mtime.tv_sec != VNOVAL) {
 1519                         if (vp->v_type == VCHR)
 1520                                 vp->v_rdev->si_mtime = vap->va_mtime;
 1521                         else
 1522                                 de->de_mtime = vap->va_mtime;
 1523                 }
 1524                 c = 1;
 1525         }
 1526 
 1527         if (c) {
 1528                 if (vp->v_type == VCHR)
 1529                         vfs_timestamp(&vp->v_rdev->si_ctime);
 1530                 else
 1531                         vfs_timestamp(&de->de_mtime);
 1532         }
 1533         return (0);
 1534 }
 1535 
 1536 #ifdef MAC
 1537 static int
 1538 devfs_setlabel(struct vop_setlabel_args *ap)
 1539 {
 1540         struct vnode *vp;
 1541         struct devfs_dirent *de;
 1542 
 1543         vp = ap->a_vp;
 1544         de = vp->v_data;
 1545 
 1546         mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
 1547         mac_devfs_update(vp->v_mount, de, vp);
 1548 
 1549         return (0);
 1550 }
 1551 #endif
 1552 
 1553 static int
 1554 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
 1555 {
 1556 
 1557         return (vnops.fo_stat(fp, sb, cred, td));
 1558 }
 1559 
 1560 static int
 1561 devfs_symlink(struct vop_symlink_args *ap)
 1562 {
 1563         int i, error;
 1564         struct devfs_dirent *dd;
 1565         struct devfs_dirent *de, *de_covered, *de_dotdot;
 1566         struct devfs_mount *dmp;
 1567 
 1568         error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
 1569         if (error)
 1570                 return(error);
 1571         dmp = VFSTODEVFS(ap->a_dvp->v_mount);
 1572         if (devfs_populate_vp(ap->a_dvp) != 0)
 1573                 return (ENOENT);
 1574 
 1575         dd = ap->a_dvp->v_data;
 1576         de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
 1577         de->de_flags = DE_USER;
 1578         de->de_uid = 0;
 1579         de->de_gid = 0;
 1580         de->de_mode = 0755;
 1581         de->de_inode = alloc_unr(devfs_inos);
 1582         de->de_dir = dd;
 1583         de->de_dirent->d_type = DT_LNK;
 1584         i = strlen(ap->a_target) + 1;
 1585         de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
 1586         bcopy(ap->a_target, de->de_symlink, i);
 1587 #ifdef MAC
 1588         mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
 1589 #endif
 1590         de_covered = devfs_find(dd, de->de_dirent->d_name,
 1591             de->de_dirent->d_namlen, 0);
 1592         if (de_covered != NULL) {
 1593                 if ((de_covered->de_flags & DE_USER) != 0) {
 1594                         devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
 1595                         sx_xunlock(&dmp->dm_lock);
 1596                         return (EEXIST);
 1597                 }
 1598                 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
 1599                     ("devfs_symlink: entry %p already covered", de_covered));
 1600                 de_covered->de_flags |= DE_COVERED;
 1601         }
 1602 
 1603         de_dotdot = TAILQ_FIRST(&dd->de_dlist);         /* "." */
 1604         de_dotdot = TAILQ_NEXT(de_dotdot, de_list);     /* ".." */
 1605         TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
 1606         devfs_dir_ref_de(dmp, dd);
 1607         devfs_rules_apply(dmp, de);
 1608 
 1609         return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
 1610 }
 1611 
 1612 static int
 1613 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
 1614 {
 1615 
 1616         return (vnops.fo_truncate(fp, length, cred, td));
 1617 }
 1618 
 1619 /* ARGSUSED */
 1620 static int
 1621 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
 1622 {
 1623         struct cdev *dev;
 1624         int error, ioflag, ref, resid;
 1625         struct cdevsw *dsw;
 1626         struct file *fpop;
 1627 
 1628         fpop = td->td_fpop;
 1629         error = devfs_fp_check(fp, &dev, &dsw, &ref);
 1630         if (error)
 1631                 return (error);
 1632         KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
 1633         ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
 1634         if (ioflag & O_DIRECT)
 1635                 ioflag |= IO_DIRECT;
 1636         if ((flags & FOF_OFFSET) == 0)
 1637                 uio->uio_offset = fp->f_offset;
 1638 
 1639         resid = uio->uio_resid;
 1640 
 1641         error = dsw->d_write(dev, uio, ioflag);
 1642         if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
 1643                 vfs_timestamp(&dev->si_ctime);
 1644                 dev->si_mtime = dev->si_ctime;
 1645         }
 1646         td->td_fpop = fpop;
 1647         dev_relthread(dev, ref);
 1648 
 1649         if ((flags & FOF_OFFSET) == 0)
 1650                 fp->f_offset = uio->uio_offset;
 1651         fp->f_nextoff = uio->uio_offset;
 1652         return (error);
 1653 }
 1654 
 1655 dev_t
 1656 dev2udev(struct cdev *x)
 1657 {
 1658         if (x == NULL)
 1659                 return (NODEV);
 1660         return (cdev2priv(x)->cdp_inode);
 1661 }
 1662 
 1663 static struct fileops devfs_ops_f = {
 1664         .fo_read =      devfs_read_f,
 1665         .fo_write =     devfs_write_f,
 1666         .fo_truncate =  devfs_truncate_f,
 1667         .fo_ioctl =     devfs_ioctl_f,
 1668         .fo_poll =      devfs_poll_f,
 1669         .fo_kqfilter =  devfs_kqfilter_f,
 1670         .fo_stat =      devfs_stat_f,
 1671         .fo_close =     devfs_close_f,
 1672         .fo_chmod =     vn_chmod,
 1673         .fo_chown =     vn_chown,
 1674         .fo_flags =     DFLAG_PASSABLE | DFLAG_SEEKABLE
 1675 };
 1676 
 1677 static struct vop_vector devfs_vnodeops = {
 1678         .vop_default =          &default_vnodeops,
 1679 
 1680         .vop_access =           devfs_access,
 1681         .vop_getattr =          devfs_getattr,
 1682         .vop_ioctl =            devfs_rioctl,
 1683         .vop_lookup =           devfs_lookup,
 1684         .vop_mknod =            devfs_mknod,
 1685         .vop_pathconf =         devfs_pathconf,
 1686         .vop_read =             devfs_rread,
 1687         .vop_readdir =          devfs_readdir,
 1688         .vop_readlink =         devfs_readlink,
 1689         .vop_reclaim =          devfs_reclaim,
 1690         .vop_remove =           devfs_remove,
 1691         .vop_revoke =           devfs_revoke,
 1692         .vop_setattr =          devfs_setattr,
 1693 #ifdef MAC
 1694         .vop_setlabel =         devfs_setlabel,
 1695 #endif
 1696         .vop_symlink =          devfs_symlink,
 1697         .vop_vptocnp =          devfs_vptocnp,
 1698 };
 1699 
 1700 static struct vop_vector devfs_specops = {
 1701         .vop_default =          &default_vnodeops,
 1702 
 1703         .vop_access =           devfs_access,
 1704         .vop_bmap =             VOP_PANIC,
 1705         .vop_close =            devfs_close,
 1706         .vop_create =           VOP_PANIC,
 1707         .vop_fsync =            devfs_fsync,
 1708         .vop_getattr =          devfs_getattr,
 1709         .vop_link =             VOP_PANIC,
 1710         .vop_mkdir =            VOP_PANIC,
 1711         .vop_mknod =            VOP_PANIC,
 1712         .vop_open =             devfs_open,
 1713         .vop_pathconf =         devfs_pathconf,
 1714         .vop_print =            devfs_print,
 1715         .vop_read =             VOP_PANIC,
 1716         .vop_readdir =          VOP_PANIC,
 1717         .vop_readlink =         VOP_PANIC,
 1718         .vop_reallocblks =      VOP_PANIC,
 1719         .vop_reclaim =          devfs_reclaim,
 1720         .vop_remove =           devfs_remove,
 1721         .vop_rename =           VOP_PANIC,
 1722         .vop_revoke =           devfs_revoke,
 1723         .vop_rmdir =            VOP_PANIC,
 1724         .vop_setattr =          devfs_setattr,
 1725 #ifdef MAC
 1726         .vop_setlabel =         devfs_setlabel,
 1727 #endif
 1728         .vop_strategy =         VOP_PANIC,
 1729         .vop_symlink =          VOP_PANIC,
 1730         .vop_vptocnp =          devfs_vptocnp,
 1731         .vop_write =            VOP_PANIC,
 1732 };
 1733 
 1734 /*
 1735  * Our calling convention to the device drivers used to be that we passed
 1736  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 
 1737  * flags instead since that's what open(), close() and ioctl() takes and
 1738  * we don't really want vnode.h in device drivers.
 1739  * We solved the source compatibility by redefining some vnode flags to
 1740  * be the same as the fcntl ones and by sending down the bitwise OR of
 1741  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
 1742  * pulls the rug out under this.
 1743  */
 1744 CTASSERT(O_NONBLOCK == IO_NDELAY);
 1745 CTASSERT(O_FSYNC == IO_SYNC);
Cache object: e25535858f37de6c71cdbf1bf92a25cf
[ 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.
FreeBSD/Linux Kernel Cross Reference sys/fs/devfs/devfs_vnops.c

FreeBSD/Linux Kernel Cross Reference
sys/fs/devfs/devfs_vnops.c
