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  * $FreeBSD$
   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/limits.h>
   53 #include <sys/lock.h>
   54 #include <sys/malloc.h>
   55 #include <sys/mount.h>
   56 #include <sys/namei.h>
   57 #include <sys/priv.h>
   58 #include <sys/proc.h>
   59 #include <sys/stat.h>
   60 #include <sys/sx.h>
   61 #include <sys/time.h>
   62 #include <sys/ttycom.h>
   63 #include <sys/unistd.h>
   64 #include <sys/vnode.h>
   65 
   66 static struct vop_vector devfs_vnodeops;
   67 static struct vop_vector devfs_specops;
   68 static struct fileops devfs_ops_f;
   69 
   70 #include <fs/devfs/devfs.h>
   71 #include <fs/devfs/devfs_int.h>
   72 
   73 #include <security/mac/mac_framework.h>
   74 
   75 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
   76 
   77 struct mtx      devfs_de_interlock;
   78 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
   79 struct sx       clone_drain_lock;
   80 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
   81 struct mtx      cdevpriv_mtx;
   82 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
   83 
   84 static int
   85 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
   86     int *ref)
   87 {
   88 
   89         *dswp = devvn_refthread(fp->f_vnode, devp, ref);
   90         if (*devp != fp->f_data) {
   91                 if (*dswp != NULL)
   92                         dev_relthread(*devp, *ref);
   93                 return (ENXIO);
   94         }
   95         KASSERT((*devp)->si_refcount > 0,
   96             ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
   97         if (*dswp == NULL)
   98                 return (ENXIO);
   99         curthread->td_fpop = fp;
  100         return (0);
  101 }
  102 
  103 int
  104 devfs_get_cdevpriv(void **datap)
  105 {
  106         struct file *fp;
  107         struct cdev_privdata *p;
  108         int error;
  109 
  110         fp = curthread->td_fpop;
  111         if (fp == NULL)
  112                 return (EBADF);
  113         p = fp->f_cdevpriv;
  114         if (p != NULL) {
  115                 error = 0;
  116                 *datap = p->cdpd_data;
  117         } else
  118                 error = ENOENT;
  119         return (error);
  120 }
  121 
  122 int
  123 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
  124 {
  125         struct file *fp;
  126         struct cdev_priv *cdp;
  127         struct cdev_privdata *p;
  128         int error;
  129 
  130         fp = curthread->td_fpop;
  131         if (fp == NULL)
  132                 return (ENOENT);
  133         cdp = cdev2priv((struct cdev *)fp->f_data);
  134         p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
  135         p->cdpd_data = priv;
  136         p->cdpd_dtr = priv_dtr;
  137         p->cdpd_fp = fp;
  138         mtx_lock(&cdevpriv_mtx);
  139         if (fp->f_cdevpriv == NULL) {
  140                 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
  141                 fp->f_cdevpriv = p;
  142                 mtx_unlock(&cdevpriv_mtx);
  143                 error = 0;
  144         } else {
  145                 mtx_unlock(&cdevpriv_mtx);
  146                 free(p, M_CDEVPDATA);
  147                 error = EBUSY;
  148         }
  149         return (error);
  150 }
  151 
  152 void
  153 devfs_destroy_cdevpriv(struct cdev_privdata *p)
  154 {
  155 
  156         mtx_assert(&cdevpriv_mtx, MA_OWNED);
  157         p->cdpd_fp->f_cdevpriv = NULL;
  158         LIST_REMOVE(p, cdpd_list);
  159         mtx_unlock(&cdevpriv_mtx);
  160         (p->cdpd_dtr)(p->cdpd_data);
  161         free(p, M_CDEVPDATA);
  162 }
  163 
  164 void
  165 devfs_fpdrop(struct file *fp)
  166 {
  167         struct cdev_privdata *p;
  168 
  169         mtx_lock(&cdevpriv_mtx);
  170         if ((p = fp->f_cdevpriv) == NULL) {
  171                 mtx_unlock(&cdevpriv_mtx);
  172                 return;
  173         }
  174         devfs_destroy_cdevpriv(p);
  175 }
  176 
  177 void
  178 devfs_clear_cdevpriv(void)
  179 {
  180         struct file *fp;
  181 
  182         fp = curthread->td_fpop;
  183         if (fp == NULL)
  184                 return;
  185         devfs_fpdrop(fp);
  186 }
  187 
  188 /*
  189  * On success devfs_populate_vp() returns with dmp->dm_lock held.
  190  */
  191 static int
  192 devfs_populate_vp(struct vnode *vp)
  193 {
  194         struct devfs_dirent *de;
  195         struct devfs_mount *dmp;
  196         int locked;
  197 
  198         ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
  199 
  200         dmp = VFSTODEVFS(vp->v_mount);
  201         locked = VOP_ISLOCKED(vp);
  202 
  203         sx_xlock(&dmp->dm_lock);
  204         DEVFS_DMP_HOLD(dmp);
  205 
  206         /* Can't call devfs_populate() with the vnode lock held. */
  207         VOP_UNLOCK(vp, 0);
  208         devfs_populate(dmp);
  209 
  210         sx_xunlock(&dmp->dm_lock);
  211         vn_lock(vp, locked | LK_RETRY);
  212         sx_xlock(&dmp->dm_lock);
  213         if (DEVFS_DMP_DROP(dmp)) {
  214                 sx_xunlock(&dmp->dm_lock);
  215                 devfs_unmount_final(dmp);
  216                 return (EBADF);
  217         }
  218         if ((vp->v_iflag & VI_DOOMED) != 0) {
  219                 sx_xunlock(&dmp->dm_lock);
  220                 return (EBADF);
  221         }
  222         de = vp->v_data;
  223         KASSERT(de != NULL,
  224             ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
  225         if ((de->de_flags & DE_DOOMED) != 0) {
  226                 sx_xunlock(&dmp->dm_lock);
  227                 return (EBADF);
  228         }
  229 
  230         return (0);
  231 }
  232 
  233 static int
  234 devfs_vptocnp(struct vop_vptocnp_args *ap)
  235 {
  236         struct vnode *vp = ap->a_vp;
  237         struct vnode **dvp = ap->a_vpp;
  238         struct devfs_mount *dmp;
  239         char *buf = ap->a_buf;
  240         int *buflen = ap->a_buflen;
  241         struct devfs_dirent *dd, *de;
  242         int i, error;
  243 
  244         dmp = VFSTODEVFS(vp->v_mount);
  245 
  246         error = devfs_populate_vp(vp);
  247         if (error != 0)
  248                 return (error);
  249 
  250         i = *buflen;
  251         dd = vp->v_data;
  252 
  253         if (vp->v_type == VCHR) {
  254                 i -= strlen(dd->de_cdp->cdp_c.si_name);
  255                 if (i < 0) {
  256                         error = ENOMEM;
  257                         goto finished;
  258                 }
  259                 bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
  260                     strlen(dd->de_cdp->cdp_c.si_name));
  261                 de = dd->de_dir;
  262         } else if (vp->v_type == VDIR) {
  263                 if (dd == dmp->dm_rootdir) {
  264                         *dvp = vp;
  265                         vref(*dvp);
  266                         goto finished;
  267                 }
  268                 i -= dd->de_dirent->d_namlen;
  269                 if (i < 0) {
  270                         error = ENOMEM;
  271                         goto finished;
  272                 }
  273                 bcopy(dd->de_dirent->d_name, buf + i,
  274                     dd->de_dirent->d_namlen);
  275                 de = dd;
  276         } else {
  277                 error = ENOENT;
  278                 goto finished;
  279         }
  280         *buflen = i;
  281         de = devfs_parent_dirent(de);
  282         if (de == NULL) {
  283                 error = ENOENT;
  284                 goto finished;
  285         }
  286         mtx_lock(&devfs_de_interlock);
  287         *dvp = de->de_vnode;
  288         if (*dvp != NULL) {
  289                 VI_LOCK(*dvp);
  290                 mtx_unlock(&devfs_de_interlock);
  291                 vholdl(*dvp);
  292                 VI_UNLOCK(*dvp);
  293                 vref(*dvp);
  294                 vdrop(*dvp);
  295         } else {
  296                 mtx_unlock(&devfs_de_interlock);
  297                 error = ENOENT;
  298         }
  299 finished:
  300         sx_xunlock(&dmp->dm_lock);
  301         return (error);
  302 }
  303 
  304 /*
  305  * Construct the fully qualified path name relative to the mountpoint.
  306  * If a NULL cnp is provided, no '/' is appended to the resulting path.
  307  */
  308 char *
  309 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
  310     struct componentname *cnp)
  311 {
  312         int i;
  313         struct devfs_dirent *de;
  314 
  315         sx_assert(&dmp->dm_lock, SA_LOCKED);
  316 
  317         i = SPECNAMELEN;
  318         buf[i] = '\0';
  319         if (cnp != NULL)
  320                 i -= cnp->cn_namelen;
  321         if (i < 0)
  322                  return (NULL);
  323         if (cnp != NULL)
  324                 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
  325         de = dd;
  326         while (de != dmp->dm_rootdir) {
  327                 if (cnp != NULL || i < SPECNAMELEN) {
  328                         i--;
  329                         if (i < 0)
  330                                  return (NULL);
  331                         buf[i] = '/';
  332                 }
  333                 i -= de->de_dirent->d_namlen;
  334                 if (i < 0)
  335                          return (NULL);
  336                 bcopy(de->de_dirent->d_name, buf + i,
  337                     de->de_dirent->d_namlen);
  338                 de = devfs_parent_dirent(de);
  339                 if (de == NULL)
  340                         return (NULL);
  341         }
  342         return (buf + i);
  343 }
  344 
  345 static int
  346 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
  347         struct devfs_dirent *de)
  348 {
  349         int not_found;
  350 
  351         not_found = 0;
  352         if (de->de_flags & DE_DOOMED)
  353                 not_found = 1;
  354         if (DEVFS_DE_DROP(de)) {
  355                 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
  356                 devfs_dirent_free(de);
  357         }
  358         if (DEVFS_DMP_DROP(dmp)) {
  359                 KASSERT(not_found == 1,
  360                         ("DEVFS mount struct freed before dirent"));
  361                 not_found = 2;
  362                 sx_xunlock(&dmp->dm_lock);
  363                 devfs_unmount_final(dmp);
  364         }
  365         if (not_found == 1 || (drop_dm_lock && not_found != 2))
  366                 sx_unlock(&dmp->dm_lock);
  367         return (not_found);
  368 }
  369 
  370 static void
  371 devfs_insmntque_dtr(struct vnode *vp, void *arg)
  372 {
  373         struct devfs_dirent *de;
  374 
  375         de = (struct devfs_dirent *)arg;
  376         mtx_lock(&devfs_de_interlock);
  377         vp->v_data = NULL;
  378         de->de_vnode = NULL;
  379         mtx_unlock(&devfs_de_interlock);
  380         vgone(vp);
  381         vput(vp);
  382 }
  383 
  384 /*
  385  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
  386  * it on return.
  387  */
  388 int
  389 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
  390     struct vnode **vpp)
  391 {
  392         int error;
  393         struct vnode *vp;
  394         struct cdev *dev;
  395         struct devfs_mount *dmp;
  396         struct cdevsw *dsw;
  397 
  398         dmp = VFSTODEVFS(mp);
  399         if (de->de_flags & DE_DOOMED) {
  400                 sx_xunlock(&dmp->dm_lock);
  401                 return (ENOENT);
  402         }
  403 loop:
  404         DEVFS_DE_HOLD(de);
  405         DEVFS_DMP_HOLD(dmp);
  406         mtx_lock(&devfs_de_interlock);
  407         vp = de->de_vnode;
  408         if (vp != NULL) {
  409                 VI_LOCK(vp);
  410                 mtx_unlock(&devfs_de_interlock);
  411                 sx_xunlock(&dmp->dm_lock);
  412                 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
  413                 sx_xlock(&dmp->dm_lock);
  414                 if (devfs_allocv_drop_refs(0, dmp, de)) {
  415                         vput(vp);
  416                         return (ENOENT);
  417                 }
  418                 else if ((vp->v_iflag & VI_DOOMED) != 0) {
  419                         mtx_lock(&devfs_de_interlock);
  420                         if (de->de_vnode == vp) {
  421                                 de->de_vnode = NULL;
  422                                 vp->v_data = NULL;
  423                         }
  424                         mtx_unlock(&devfs_de_interlock);
  425                         vput(vp);
  426                         goto loop;
  427                 }
  428                 sx_xunlock(&dmp->dm_lock);
  429                 *vpp = vp;
  430                 return (0);
  431         }
  432         mtx_unlock(&devfs_de_interlock);
  433         if (de->de_dirent->d_type == DT_CHR) {
  434                 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
  435                         devfs_allocv_drop_refs(1, dmp, de);
  436                         return (ENOENT);
  437                 }
  438                 dev = &de->de_cdp->cdp_c;
  439         } else {
  440                 dev = NULL;
  441         }
  442         error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
  443         if (error != 0) {
  444                 devfs_allocv_drop_refs(1, dmp, de);
  445                 printf("devfs_allocv: failed to allocate new vnode\n");
  446                 return (error);
  447         }
  448 
  449         if (de->de_dirent->d_type == DT_CHR) {
  450                 vp->v_type = VCHR;
  451                 VI_LOCK(vp);
  452                 dev_lock();
  453                 dev_refl(dev);
  454                 /* XXX: v_rdev should be protect by vnode lock */
  455                 vp->v_rdev = dev;
  456                 KASSERT(vp->v_usecount == 1,
  457                     ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
  458                 dev->si_usecount += vp->v_usecount;
  459                 /* Special casing of ttys for deadfs.  Probably redundant. */
  460                 dsw = dev->si_devsw;
  461                 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
  462                         vp->v_vflag |= VV_ISTTY;
  463                 dev_unlock();
  464                 VI_UNLOCK(vp);
  465                 if ((dev->si_flags & SI_ETERNAL) != 0)
  466                         vp->v_vflag |= VV_ETERNALDEV;
  467                 vp->v_op = &devfs_specops;
  468         } else if (de->de_dirent->d_type == DT_DIR) {
  469                 vp->v_type = VDIR;
  470         } else if (de->de_dirent->d_type == DT_LNK) {
  471                 vp->v_type = VLNK;
  472         } else {
  473                 vp->v_type = VBAD;
  474         }
  475         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
  476         VN_LOCK_ASHARE(vp);
  477         mtx_lock(&devfs_de_interlock);
  478         vp->v_data = de;
  479         de->de_vnode = vp;
  480         mtx_unlock(&devfs_de_interlock);
  481         error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
  482         if (error != 0) {
  483                 (void) devfs_allocv_drop_refs(1, dmp, de);
  484                 return (error);
  485         }
  486         if (devfs_allocv_drop_refs(0, dmp, de)) {
  487                 vput(vp);
  488                 return (ENOENT);
  489         }
  490 #ifdef MAC
  491         mac_devfs_vnode_associate(mp, de, vp);
  492 #endif
  493         sx_xunlock(&dmp->dm_lock);
  494         *vpp = vp;
  495         return (0);
  496 }
  497 
  498 static int
  499 devfs_access(struct vop_access_args *ap)
  500 {
  501         struct vnode *vp = ap->a_vp;
  502         struct devfs_dirent *de;
  503         int error;
  504 
  505         de = vp->v_data;
  506         if (vp->v_type == VDIR)
  507                 de = de->de_dir;
  508 
  509         error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
  510             ap->a_accmode, ap->a_cred, NULL);
  511         if (error == 0)
  512                 return (0);
  513         if (error != EACCES)
  514                 return (error);
  515         /* We do, however, allow access to the controlling terminal */
  516         if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
  517                 return (error);
  518         if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp)
  519                 return (0);
  520         return (error);
  521 }
  522 
  523 /* ARGSUSED */
  524 static int
  525 devfs_close(struct vop_close_args *ap)
  526 {
  527         struct vnode *vp = ap->a_vp, *oldvp;
  528         struct thread *td = ap->a_td;
  529         struct cdev *dev = vp->v_rdev;
  530         struct cdevsw *dsw;
  531         int vp_locked, error, ref;
  532 
  533         /*
  534          * XXX: Don't call d_close() if we were called because of
  535          * XXX: insmntque1() failure.
  536          */
  537         if (vp->v_data == NULL)
  538                 return (0);
  539 
  540         /*
  541          * Hack: a tty device that is a controlling terminal
  542          * has a reference from the session structure.
  543          * We cannot easily tell that a character device is
  544          * a controlling terminal, unless it is the closing
  545          * process' controlling terminal.  In that case,
  546          * if the reference count is 2 (this last descriptor
  547          * plus the session), release the reference from the session.
  548          */
  549         oldvp = NULL;
  550         sx_xlock(&proctree_lock);
  551         if (td && vp == td->td_proc->p_session->s_ttyvp) {
  552                 SESS_LOCK(td->td_proc->p_session);
  553                 VI_LOCK(vp);
  554                 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
  555                         td->td_proc->p_session->s_ttyvp = NULL;
  556                         td->td_proc->p_session->s_ttydp = NULL;
  557                         oldvp = vp;
  558                 }
  559                 VI_UNLOCK(vp);
  560                 SESS_UNLOCK(td->td_proc->p_session);
  561         }
  562         sx_xunlock(&proctree_lock);
  563         if (oldvp != NULL)
  564                 vrele(oldvp);
  565         /*
  566          * We do not want to really close the device if it
  567          * is still in use unless we are trying to close it
  568          * forcibly. Since every use (buffer, vnode, swap, cmap)
  569          * holds a reference to the vnode, and because we mark
  570          * any other vnodes that alias this device, when the
  571          * sum of the reference counts on all the aliased
  572          * vnodes descends to one, we are on last close.
  573          */
  574         dsw = dev_refthread(dev, &ref);
  575         if (dsw == NULL)
  576                 return (ENXIO);
  577         VI_LOCK(vp);
  578         if (vp->v_iflag & VI_DOOMED) {
  579                 /* Forced close. */
  580         } else if (dsw->d_flags & D_TRACKCLOSE) {
  581                 /* Keep device updated on status. */
  582         } else if (count_dev(dev) > 1) {
  583                 VI_UNLOCK(vp);
  584                 dev_relthread(dev, ref);
  585                 return (0);
  586         }
  587         vholdl(vp);
  588         VI_UNLOCK(vp);
  589         vp_locked = VOP_ISLOCKED(vp);
  590         VOP_UNLOCK(vp, 0);
  591         KASSERT(dev->si_refcount > 0,
  592             ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
  593         error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
  594         dev_relthread(dev, ref);
  595         vn_lock(vp, vp_locked | LK_RETRY);
  596         vdrop(vp);
  597         return (error);
  598 }
  599 
  600 static int
  601 devfs_close_f(struct file *fp, struct thread *td)
  602 {
  603         int error;
  604         struct file *fpop;
  605 
  606         /*
  607          * NB: td may be NULL if this descriptor is closed due to
  608          * garbage collection from a closed UNIX domain socket.
  609          */
  610         fpop = curthread->td_fpop;
  611         curthread->td_fpop = fp;
  612         error = vnops.fo_close(fp, td);
  613         curthread->td_fpop = fpop;
  614 
  615         /*
  616          * The f_cdevpriv cannot be assigned non-NULL value while we
  617          * are destroying the file.
  618          */
  619         if (fp->f_cdevpriv != NULL)
  620                 devfs_fpdrop(fp);
  621         return (error);
  622 }
  623 
  624 static int
  625 devfs_fsync(struct vop_fsync_args *ap)
  626 {
  627         int error;
  628         struct bufobj *bo;
  629         struct devfs_dirent *de;
  630 
  631         if (!vn_isdisk(ap->a_vp, &error)) {
  632                 bo = &ap->a_vp->v_bufobj;
  633                 de = ap->a_vp->v_data;
  634                 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
  635                         printf("Device %s went missing before all of the data "
  636                             "could be written to it; expect data loss.\n",
  637                             de->de_dirent->d_name);
  638 
  639                         error = vop_stdfsync(ap);
  640                         if (bo->bo_dirty.bv_cnt != 0 || error != 0)
  641                                 panic("devfs_fsync: vop_stdfsync failed.");
  642                 }
  643 
  644                 return (0);
  645         }
  646 
  647         return (vop_stdfsync(ap));
  648 }
  649 
  650 static int
  651 devfs_getattr(struct vop_getattr_args *ap)
  652 {
  653         struct vnode *vp = ap->a_vp;
  654         struct vattr *vap = ap->a_vap;
  655         int error;
  656         struct devfs_dirent *de;
  657         struct devfs_mount *dmp;
  658         struct cdev *dev;
  659 
  660         error = devfs_populate_vp(vp);
  661         if (error != 0)
  662                 return (error);
  663 
  664         dmp = VFSTODEVFS(vp->v_mount);
  665         sx_xunlock(&dmp->dm_lock);
  666 
  667         de = vp->v_data;
  668         KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
  669         if (vp->v_type == VDIR) {
  670                 de = de->de_dir;
  671                 KASSERT(de != NULL,
  672                     ("Null dir dirent in devfs_getattr vp=%p", vp));
  673         }
  674         vap->va_uid = de->de_uid;
  675         vap->va_gid = de->de_gid;
  676         vap->va_mode = de->de_mode;
  677         if (vp->v_type == VLNK)
  678                 vap->va_size = strlen(de->de_symlink);
  679         else if (vp->v_type == VDIR)
  680                 vap->va_size = vap->va_bytes = DEV_BSIZE;
  681         else
  682                 vap->va_size = 0;
  683         if (vp->v_type != VDIR)
  684                 vap->va_bytes = 0;
  685         vap->va_blocksize = DEV_BSIZE;
  686         vap->va_type = vp->v_type;
  687 
  688 #define fix(aa)                                                 \
  689         do {                                                    \
  690                 if ((aa).tv_sec <= 3600) {                      \
  691                         (aa).tv_sec = boottime.tv_sec;          \
  692                         (aa).tv_nsec = boottime.tv_usec * 1000; \
  693                 }                                               \
  694         } while (0)
  695 
  696         if (vp->v_type != VCHR)  {
  697                 fix(de->de_atime);
  698                 vap->va_atime = de->de_atime;
  699                 fix(de->de_mtime);
  700                 vap->va_mtime = de->de_mtime;
  701                 fix(de->de_ctime);
  702                 vap->va_ctime = de->de_ctime;
  703         } else {
  704                 dev = vp->v_rdev;
  705                 fix(dev->si_atime);
  706                 vap->va_atime = dev->si_atime;
  707                 fix(dev->si_mtime);
  708                 vap->va_mtime = dev->si_mtime;
  709                 fix(dev->si_ctime);
  710                 vap->va_ctime = dev->si_ctime;
  711 
  712                 vap->va_rdev = cdev2priv(dev)->cdp_inode;
  713         }
  714         vap->va_gen = 0;
  715         vap->va_flags = 0;
  716         vap->va_filerev = 0;
  717         vap->va_nlink = de->de_links;
  718         vap->va_fileid = de->de_inode;
  719 
  720         return (error);
  721 }
  722 
  723 /* ARGSUSED */
  724 static int
  725 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
  726 {
  727         struct cdev *dev;
  728         struct cdevsw *dsw;
  729         struct vnode *vp;
  730         struct vnode *vpold;
  731         int error, i, ref;
  732         const char *p;
  733         struct fiodgname_arg *fgn;
  734         struct file *fpop;
  735 
  736         fpop = td->td_fpop;
  737         error = devfs_fp_check(fp, &dev, &dsw, &ref);
  738         if (error)
  739                 return (error);
  740 
  741         if (com == FIODTYPE) {
  742                 *(int *)data = dsw->d_flags & D_TYPEMASK;
  743                 td->td_fpop = fpop;
  744                 dev_relthread(dev, ref);
  745                 return (0);
  746         } else if (com == FIODGNAME) {
  747                 fgn = data;
  748                 p = devtoname(dev);
  749                 i = strlen(p) + 1;
  750                 if (i > fgn->len)
  751                         error = EINVAL;
  752                 else
  753                         error = copyout(p, fgn->buf, i);
  754                 td->td_fpop = fpop;
  755                 dev_relthread(dev, ref);
  756                 return (error);
  757         }
  758         error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
  759         td->td_fpop = NULL;
  760         dev_relthread(dev, ref);
  761         if (error == ENOIOCTL)
  762                 error = ENOTTY;
  763         if (error == 0 && com == TIOCSCTTY) {
  764                 vp = fp->f_vnode;
  765 
  766                 /* Do nothing if reassigning same control tty */
  767                 sx_slock(&proctree_lock);
  768                 if (td->td_proc->p_session->s_ttyvp == vp) {
  769                         sx_sunlock(&proctree_lock);
  770                         return (0);
  771                 }
  772 
  773                 vpold = td->td_proc->p_session->s_ttyvp;
  774                 VREF(vp);
  775                 SESS_LOCK(td->td_proc->p_session);
  776                 td->td_proc->p_session->s_ttyvp = vp;
  777                 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
  778                 SESS_UNLOCK(td->td_proc->p_session);
  779 
  780                 sx_sunlock(&proctree_lock);
  781 
  782                 /* Get rid of reference to old control tty */
  783                 if (vpold)
  784                         vrele(vpold);
  785         }
  786         return (error);
  787 }
  788 
  789 /* ARGSUSED */
  790 static int
  791 devfs_kqfilter_f(struct file *fp, struct knote *kn)
  792 {
  793         struct cdev *dev;
  794         struct cdevsw *dsw;
  795         int error, ref;
  796         struct file *fpop;
  797         struct thread *td;
  798 
  799         td = curthread;
  800         fpop = td->td_fpop;
  801         error = devfs_fp_check(fp, &dev, &dsw, &ref);
  802         if (error)
  803                 return (error);
  804         error = dsw->d_kqfilter(dev, kn);
  805         td->td_fpop = fpop;
  806         dev_relthread(dev, ref);
  807         return (error);
  808 }
  809 
  810 static inline int
  811 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
  812 {
  813         struct cdev_priv *cdp;
  814         struct ucred *dcr;
  815         int error;
  816 
  817         cdp = de->de_cdp;
  818         if (cdp == NULL)
  819                 return (0);
  820         dcr = cdp->cdp_c.si_cred;
  821         if (dcr == NULL)
  822                 return (0);
  823 
  824         error = prison_check(td->td_ucred, dcr);
  825         if (error == 0)
  826                 return (0);
  827         /* We do, however, allow access to the controlling terminal */
  828         if (!(td->td_proc->p_flag & P_CONTROLT))
  829                 return (error);
  830         if (td->td_proc->p_session->s_ttydp == cdp)
  831                 return (0);
  832         return (error);
  833 }
  834 
  835 static int
  836 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
  837 {
  838         struct componentname *cnp;
  839         struct vnode *dvp, **vpp;
  840         struct thread *td;
  841         struct devfs_dirent *de, *dd;
  842         struct devfs_dirent **dde;
  843         struct devfs_mount *dmp;
  844         struct cdev *cdev;
  845         int error, flags, nameiop, dvplocked;
  846         char specname[SPECNAMELEN + 1], *pname;
  847 
  848         cnp = ap->a_cnp;
  849         vpp = ap->a_vpp;
  850         dvp = ap->a_dvp;
  851         pname = cnp->cn_nameptr;
  852         td = cnp->cn_thread;
  853         flags = cnp->cn_flags;
  854         nameiop = cnp->cn_nameiop;
  855         dmp = VFSTODEVFS(dvp->v_mount);
  856         dd = dvp->v_data;
  857         *vpp = NULLVP;
  858 
  859         if ((flags & ISLASTCN) && nameiop == RENAME)
  860                 return (EOPNOTSUPP);
  861 
  862         if (dvp->v_type != VDIR)
  863                 return (ENOTDIR);
  864 
  865         if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
  866                 return (EIO);
  867 
  868         error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
  869         if (error)
  870                 return (error);
  871 
  872         if (cnp->cn_namelen == 1 && *pname == '.') {
  873                 if ((flags & ISLASTCN) && nameiop != LOOKUP)
  874                         return (EINVAL);
  875                 *vpp = dvp;
  876                 VREF(dvp);
  877                 return (0);
  878         }
  879 
  880         if (flags & ISDOTDOT) {
  881                 if ((flags & ISLASTCN) && nameiop != LOOKUP)
  882                         return (EINVAL);
  883                 de = devfs_parent_dirent(dd);
  884                 if (de == NULL)
  885                         return (ENOENT);
  886                 dvplocked = VOP_ISLOCKED(dvp);
  887                 VOP_UNLOCK(dvp, 0);
  888                 error = devfs_allocv(de, dvp->v_mount,
  889                     cnp->cn_lkflags & LK_TYPE_MASK, vpp);
  890                 *dm_unlock = 0;
  891                 vn_lock(dvp, dvplocked | LK_RETRY);
  892                 return (error);
  893         }
  894 
  895         dd = dvp->v_data;
  896         de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
  897         while (de == NULL) {    /* While(...) so we can use break */
  898 
  899                 if (nameiop == DELETE)
  900                         return (ENOENT);
  901 
  902                 /*
  903                  * OK, we didn't have an entry for the name we were asked for
  904                  * so we try to see if anybody can create it on demand.
  905                  */
  906                 pname = devfs_fqpn(specname, dmp, dd, cnp);
  907                 if (pname == NULL)
  908                         break;
  909 
  910                 cdev = NULL;
  911                 DEVFS_DMP_HOLD(dmp);
  912                 sx_xunlock(&dmp->dm_lock);
  913                 sx_slock(&clone_drain_lock);
  914                 EVENTHANDLER_INVOKE(dev_clone,
  915                     td->td_ucred, pname, strlen(pname), &cdev);
  916                 sx_sunlock(&clone_drain_lock);
  917 
  918                 if (cdev == NULL)
  919                         sx_xlock(&dmp->dm_lock);
  920                 else if (devfs_populate_vp(dvp) != 0) {
  921                         *dm_unlock = 0;
  922                         sx_xlock(&dmp->dm_lock);
  923                         if (DEVFS_DMP_DROP(dmp)) {
  924                                 sx_xunlock(&dmp->dm_lock);
  925                                 devfs_unmount_final(dmp);
  926                         } else
  927                                 sx_xunlock(&dmp->dm_lock);
  928                         dev_rel(cdev);
  929                         return (ENOENT);
  930                 }
  931                 if (DEVFS_DMP_DROP(dmp)) {
  932                         *dm_unlock = 0;
  933                         sx_xunlock(&dmp->dm_lock);
  934                         devfs_unmount_final(dmp);
  935                         if (cdev != NULL)
  936                                 dev_rel(cdev);
  937                         return (ENOENT);
  938                 }
  939 
  940                 if (cdev == NULL)
  941                         break;
  942 
  943                 dev_lock();
  944                 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
  945                 if (dde != NULL && *dde != NULL)
  946                         de = *dde;
  947                 dev_unlock();
  948                 dev_rel(cdev);
  949                 break;
  950         }
  951 
  952         if (de == NULL || de->de_flags & DE_WHITEOUT) {
  953                 if ((nameiop == CREATE || nameiop == RENAME) &&
  954                     (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
  955                         cnp->cn_flags |= SAVENAME;
  956                         return (EJUSTRETURN);
  957                 }
  958                 return (ENOENT);
  959         }
  960 
  961         if (devfs_prison_check(de, td))
  962                 return (ENOENT);
  963 
  964         if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
  965                 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
  966                 if (error)
  967                         return (error);
  968                 if (*vpp == dvp) {
  969                         VREF(dvp);
  970                         *vpp = dvp;
  971                         return (0);
  972                 }
  973         }
  974         error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
  975             vpp);
  976         *dm_unlock = 0;
  977         return (error);
  978 }
  979 
  980 static int
  981 devfs_lookup(struct vop_lookup_args *ap)
  982 {
  983         int j;
  984         struct devfs_mount *dmp;
  985         int dm_unlock;
  986 
  987         if (devfs_populate_vp(ap->a_dvp) != 0)
  988                 return (ENOTDIR);
  989 
  990         dmp = VFSTODEVFS(ap->a_dvp->v_mount);
  991         dm_unlock = 1;
  992         j = devfs_lookupx(ap, &dm_unlock);
  993         if (dm_unlock == 1)
  994                 sx_xunlock(&dmp->dm_lock);
  995         return (j);
  996 }
  997 
  998 static int
  999 devfs_mknod(struct vop_mknod_args *ap)
 1000 {
 1001         struct componentname *cnp;
 1002         struct vnode *dvp, **vpp;
 1003         struct devfs_dirent *dd, *de;
 1004         struct devfs_mount *dmp;
 1005         int error;
 1006 
 1007         /*
 1008          * The only type of node we should be creating here is a
 1009          * character device, for anything else return EOPNOTSUPP.
 1010          */
 1011         if (ap->a_vap->va_type != VCHR)
 1012                 return (EOPNOTSUPP);
 1013         dvp = ap->a_dvp;
 1014         dmp = VFSTODEVFS(dvp->v_mount);
 1015 
 1016         cnp = ap->a_cnp;
 1017         vpp = ap->a_vpp;
 1018         dd = dvp->v_data;
 1019 
 1020         error = ENOENT;
 1021         sx_xlock(&dmp->dm_lock);
 1022         TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
 1023                 if (cnp->cn_namelen != de->de_dirent->d_namlen)
 1024                         continue;
 1025                 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
 1026                     de->de_dirent->d_namlen) != 0)
 1027                         continue;
 1028                 if (de->de_flags & DE_WHITEOUT)
 1029                         break;
 1030                 goto notfound;
 1031         }
 1032         if (de == NULL)
 1033                 goto notfound;
 1034         de->de_flags &= ~DE_WHITEOUT;
 1035         error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
 1036         return (error);
 1037 notfound:
 1038         sx_xunlock(&dmp->dm_lock);
 1039         return (error);
 1040 }
 1041 
 1042 /* ARGSUSED */
 1043 static int
 1044 devfs_open(struct vop_open_args *ap)
 1045 {
 1046         struct thread *td = ap->a_td;
 1047         struct vnode *vp = ap->a_vp;
 1048         struct cdev *dev = vp->v_rdev;
 1049         struct file *fp = ap->a_fp;
 1050         int error, ref, vlocked;
 1051         struct cdevsw *dsw;
 1052         struct file *fpop;
 1053         struct mtx *mtxp;
 1054 
 1055         if (vp->v_type == VBLK)
 1056                 return (ENXIO);
 1057 
 1058         if (dev == NULL)
 1059                 return (ENXIO);
 1060 
 1061         /* Make this field valid before any I/O in d_open. */
 1062         if (dev->si_iosize_max == 0)
 1063                 dev->si_iosize_max = DFLTPHYS;
 1064 
 1065         dsw = dev_refthread(dev, &ref);
 1066         if (dsw == NULL)
 1067                 return (ENXIO);
 1068         if (fp == NULL && dsw->d_fdopen != NULL) {
 1069                 dev_relthread(dev, ref);
 1070                 return (ENXIO);
 1071         }
 1072 
 1073         vlocked = VOP_ISLOCKED(vp);
 1074         VOP_UNLOCK(vp, 0);
 1075 
 1076         fpop = td->td_fpop;
 1077         td->td_fpop = fp;
 1078         if (fp != NULL) {
 1079                 fp->f_data = dev;
 1080                 fp->f_vnode = vp;
 1081         }
 1082         if (dsw->d_fdopen != NULL)
 1083                 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
 1084         else
 1085                 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
 1086         /* cleanup any cdevpriv upon error */
 1087         if (error != 0)
 1088                 devfs_clear_cdevpriv();
 1089         td->td_fpop = fpop;
 1090 
 1091         vn_lock(vp, vlocked | LK_RETRY);
 1092         dev_relthread(dev, ref);
 1093         if (error != 0) {
 1094                 if (error == ERESTART)
 1095                         error = EINTR;
 1096                 return (error);
 1097         }
 1098 
 1099 #if 0   /* /dev/console */
 1100         KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
 1101 #else
 1102         if (fp == NULL)
 1103                 return (error);
 1104 #endif
 1105         if (fp->f_ops == &badfileops)
 1106                 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
 1107         mtxp = mtx_pool_find(mtxpool_sleep, fp);
 1108 
 1109         /*
 1110          * Hint to the dofilewrite() to not force the buffer draining
 1111          * on the writer to the file.  Most likely, the write would
 1112          * not need normal buffers.
 1113          */
 1114         mtx_lock(mtxp);
 1115         fp->f_vnread_flags |= FDEVFS_VNODE;
 1116         mtx_unlock(mtxp);
 1117         return (error);
 1118 }
 1119 
 1120 static int
 1121 devfs_pathconf(struct vop_pathconf_args *ap)
 1122 {
 1123 
 1124         switch (ap->a_name) {
 1125         case _PC_MAC_PRESENT:
 1126 #ifdef MAC
 1127                 /*
 1128                  * If MAC is enabled, devfs automatically supports
 1129                  * trivial non-persistant label storage.
 1130                  */
 1131                 *ap->a_retval = 1;
 1132 #else
 1133                 *ap->a_retval = 0;
 1134 #endif
 1135                 return (0);
 1136         default:
 1137                 return (vop_stdpathconf(ap));
 1138         }
 1139         /* NOTREACHED */
 1140 }
 1141 
 1142 /* ARGSUSED */
 1143 static int
 1144 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
 1145 {
 1146         struct cdev *dev;
 1147         struct cdevsw *dsw;
 1148         int error, ref;
 1149         struct file *fpop;
 1150 
 1151         fpop = td->td_fpop;
 1152         error = devfs_fp_check(fp, &dev, &dsw, &ref);
 1153         if (error)
 1154                 return (poll_no_poll(events));
 1155         error = dsw->d_poll(dev, events, td);
 1156         td->td_fpop = fpop;
 1157         dev_relthread(dev, ref);
 1158         return(error);
 1159 }
 1160 
 1161 /*
 1162  * Print out the contents of a special device vnode.
 1163  */
 1164 static int
 1165 devfs_print(struct vop_print_args *ap)
 1166 {
 1167 
 1168         printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
 1169         return (0);
 1170 }
 1171 
 1172 static int
 1173 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
 1174     int flags, struct thread *td)
 1175 {
 1176         struct cdev *dev;
 1177         int ioflag, error, ref;
 1178         ssize_t resid;
 1179         struct cdevsw *dsw;
 1180         struct file *fpop;
 1181 
 1182         if (uio->uio_resid > DEVFS_IOSIZE_MAX)
 1183                 return (EINVAL);
 1184         fpop = td->td_fpop;
 1185         error = devfs_fp_check(fp, &dev, &dsw, &ref);
 1186         if (error)
 1187                 return (error);
 1188         resid = uio->uio_resid;
 1189         ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
 1190         if (ioflag & O_DIRECT)
 1191                 ioflag |= IO_DIRECT;
 1192 
 1193         foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
 1194         error = dsw->d_read(dev, uio, ioflag);
 1195         if (uio->uio_resid != resid || (error == 0 && resid != 0))
 1196                 vfs_timestamp(&dev->si_atime);
 1197         td->td_fpop = fpop;
 1198         dev_relthread(dev, ref);
 1199 
 1200         foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
 1201         return (error);
 1202 }
 1203 
 1204 static int
 1205 devfs_readdir(struct vop_readdir_args *ap)
 1206 {
 1207         int error;
 1208         struct uio *uio;
 1209         struct dirent *dp;
 1210         struct devfs_dirent *dd;
 1211         struct devfs_dirent *de;
 1212         struct devfs_mount *dmp;
 1213         off_t off;
 1214         int *tmp_ncookies = NULL;
 1215 
 1216         if (ap->a_vp->v_type != VDIR)
 1217                 return (ENOTDIR);
 1218 
 1219         uio = ap->a_uio;
 1220         if (uio->uio_offset < 0)
 1221                 return (EINVAL);
 1222 
 1223         /*
 1224          * XXX: This is a temporary hack to get around this filesystem not
 1225          * supporting cookies. We store the location of the ncookies pointer
 1226          * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
 1227          * and set the number of cookies to 0. We then set the pointer to
 1228          * NULL so that vfs_read_dirent doesn't try to call realloc() on 
 1229          * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
 1230          * pointer to its original location before returning to the caller.
 1231          */
 1232         if (ap->a_ncookies != NULL) {
 1233                 tmp_ncookies = ap->a_ncookies;
 1234                 *ap->a_ncookies = 0;
 1235                 ap->a_ncookies = NULL;
 1236         }
 1237 
 1238         dmp = VFSTODEVFS(ap->a_vp->v_mount);
 1239         if (devfs_populate_vp(ap->a_vp) != 0) {
 1240                 if (tmp_ncookies != NULL)
 1241                         ap->a_ncookies = tmp_ncookies;
 1242                 return (EIO);
 1243         }
 1244         error = 0;
 1245         de = ap->a_vp->v_data;
 1246         off = 0;
 1247         TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
 1248                 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
 1249                 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
 1250                         continue;
 1251                 if (devfs_prison_check(dd, uio->uio_td))
 1252                         continue;
 1253                 if (dd->de_dirent->d_type == DT_DIR)
 1254                         de = dd->de_dir;
 1255                 else
 1256                         de = dd;
 1257                 dp = dd->de_dirent;
 1258                 if (dp->d_reclen > uio->uio_resid)
 1259                         break;
 1260                 dp->d_fileno = de->de_inode;
 1261                 if (off >= uio->uio_offset) {
 1262                         error = vfs_read_dirent(ap, dp, off);
 1263                         if (error)
 1264                                 break;
 1265                 }
 1266                 off += dp->d_reclen;
 1267         }
 1268         sx_xunlock(&dmp->dm_lock);
 1269         uio->uio_offset = off;
 1270 
 1271         /*
 1272          * Restore ap->a_ncookies if it wasn't originally NULL in the first
 1273          * place.
 1274          */
 1275         if (tmp_ncookies != NULL)
 1276                 ap->a_ncookies = tmp_ncookies;
 1277 
 1278         return (error);
 1279 }
 1280 
 1281 static int
 1282 devfs_readlink(struct vop_readlink_args *ap)
 1283 {
 1284         struct devfs_dirent *de;
 1285 
 1286         de = ap->a_vp->v_data;
 1287         return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
 1288 }
 1289 
 1290 static int
 1291 devfs_reclaim(struct vop_reclaim_args *ap)
 1292 {
 1293         struct vnode *vp = ap->a_vp;
 1294         struct devfs_dirent *de;
 1295         struct cdev *dev;
 1296 
 1297         mtx_lock(&devfs_de_interlock);
 1298         de = vp->v_data;
 1299         if (de != NULL) {
 1300                 de->de_vnode = NULL;
 1301                 vp->v_data = NULL;
 1302         }
 1303         mtx_unlock(&devfs_de_interlock);
 1304 
 1305         vnode_destroy_vobject(vp);
 1306 
 1307         VI_LOCK(vp);
 1308         dev_lock();
 1309         dev = vp->v_rdev;
 1310         vp->v_rdev = NULL;
 1311 
 1312         if (dev == NULL) {
 1313                 dev_unlock();
 1314                 VI_UNLOCK(vp);
 1315                 return (0);
 1316         }
 1317 
 1318         dev->si_usecount -= vp->v_usecount;
 1319         dev_unlock();
 1320         VI_UNLOCK(vp);
 1321         dev_rel(dev);
 1322         return (0);
 1323 }
 1324 
 1325 static int
 1326 devfs_remove(struct vop_remove_args *ap)
 1327 {
 1328         struct vnode *dvp = ap->a_dvp;
 1329         struct vnode *vp = ap->a_vp;
 1330         struct devfs_dirent *dd;
 1331         struct devfs_dirent *de, *de_covered;
 1332         struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
 1333 
 1334         ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
 1335         ASSERT_VOP_ELOCKED(vp, "devfs_remove");
 1336 
 1337         sx_xlock(&dmp->dm_lock);
 1338         dd = ap->a_dvp->v_data;
 1339         de = vp->v_data;
 1340         if (de->de_cdp == NULL) {
 1341                 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
 1342                 if (de->de_dirent->d_type == DT_LNK) {
 1343                         de_covered = devfs_find(dd, de->de_dirent->d_name,
 1344                             de->de_dirent->d_namlen, 0);
 1345                         if (de_covered != NULL)
 1346                                 de_covered->de_flags &= ~DE_COVERED;
 1347                 }
 1348                 /* We need to unlock dvp because devfs_delete() may lock it. */
 1349                 VOP_UNLOCK(vp, 0);
 1350                 if (dvp != vp)
 1351                         VOP_UNLOCK(dvp, 0);
 1352                 devfs_delete(dmp, de, 0);
 1353                 sx_xunlock(&dmp->dm_lock);
 1354                 if (dvp != vp)
 1355                         vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
 1356                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1357         } else {
 1358                 de->de_flags |= DE_WHITEOUT;
 1359                 sx_xunlock(&dmp->dm_lock);
 1360         }
 1361         return (0);
 1362 }
 1363 
 1364 /*
 1365  * Revoke is called on a tty when a terminal session ends.  The vnode
 1366  * is orphaned by setting v_op to deadfs so we need to let go of it
 1367  * as well so that we create a new one next time around.
 1368  *
 1369  */
 1370 static int
 1371 devfs_revoke(struct vop_revoke_args *ap)
 1372 {
 1373         struct vnode *vp = ap->a_vp, *vp2;
 1374         struct cdev *dev;
 1375         struct cdev_priv *cdp;
 1376         struct devfs_dirent *de;
 1377         u_int i;
 1378 
 1379         KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
 1380 
 1381         dev = vp->v_rdev;
 1382         cdp = cdev2priv(dev);
 1383  
 1384         dev_lock();
 1385         cdp->cdp_inuse++;
 1386         dev_unlock();
 1387 
 1388         vhold(vp);
 1389         vgone(vp);
 1390         vdrop(vp);
 1391 
 1392         VOP_UNLOCK(vp,0);
 1393  loop:
 1394         for (;;) {
 1395                 mtx_lock(&devfs_de_interlock);
 1396                 dev_lock();
 1397                 vp2 = NULL;
 1398                 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
 1399                         de = cdp->cdp_dirents[i];
 1400                         if (de == NULL)
 1401                                 continue;
 1402 
 1403                         vp2 = de->de_vnode;
 1404                         if (vp2 != NULL) {
 1405                                 dev_unlock();
 1406                                 VI_LOCK(vp2);
 1407                                 mtx_unlock(&devfs_de_interlock);
 1408                                 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
 1409                                     curthread))
 1410                                         goto loop;
 1411                                 vhold(vp2);
 1412                                 vgone(vp2);
 1413                                 vdrop(vp2);
 1414                                 vput(vp2);
 1415                                 break;
 1416                         } 
 1417                 }
 1418                 if (vp2 != NULL) {
 1419                         continue;
 1420                 }
 1421                 dev_unlock();
 1422                 mtx_unlock(&devfs_de_interlock);
 1423                 break;
 1424         }
 1425         dev_lock();
 1426         cdp->cdp_inuse--;
 1427         if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
 1428                 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
 1429                 dev_unlock();
 1430                 dev_rel(&cdp->cdp_c);
 1431         } else
 1432                 dev_unlock();
 1433 
 1434         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1435         return (0);
 1436 }
 1437 
 1438 static int
 1439 devfs_rioctl(struct vop_ioctl_args *ap)
 1440 {
 1441         struct vnode *vp;
 1442         struct devfs_mount *dmp;
 1443         int error;
 1444 
 1445         vp = ap->a_vp;
 1446         vn_lock(vp, LK_SHARED | LK_RETRY);
 1447         if (vp->v_iflag & VI_DOOMED) {
 1448                 VOP_UNLOCK(vp, 0);
 1449                 return (EBADF);
 1450         }
 1451         dmp = VFSTODEVFS(vp->v_mount);
 1452         sx_xlock(&dmp->dm_lock);
 1453         VOP_UNLOCK(vp, 0);
 1454         DEVFS_DMP_HOLD(dmp);
 1455         devfs_populate(dmp);
 1456         if (DEVFS_DMP_DROP(dmp)) {
 1457                 sx_xunlock(&dmp->dm_lock);
 1458                 devfs_unmount_final(dmp);
 1459                 return (ENOENT);
 1460         }
 1461         error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
 1462         sx_xunlock(&dmp->dm_lock);
 1463         return (error);
 1464 }
 1465 
 1466 static int
 1467 devfs_rread(struct vop_read_args *ap)
 1468 {
 1469 
 1470         if (ap->a_vp->v_type != VDIR)
 1471                 return (EINVAL);
 1472         return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
 1473 }
 1474 
 1475 static int
 1476 devfs_setattr(struct vop_setattr_args *ap)
 1477 {
 1478         struct devfs_dirent *de;
 1479         struct vattr *vap;
 1480         struct vnode *vp;
 1481         struct thread *td;
 1482         int c, error;
 1483         uid_t uid;
 1484         gid_t gid;
 1485 
 1486         vap = ap->a_vap;
 1487         vp = ap->a_vp;
 1488         td = curthread;
 1489         if ((vap->va_type != VNON) ||
 1490             (vap->va_nlink != VNOVAL) ||
 1491             (vap->va_fsid != VNOVAL) ||
 1492             (vap->va_fileid != VNOVAL) ||
 1493             (vap->va_blocksize != VNOVAL) ||
 1494             (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
 1495             (vap->va_rdev != VNOVAL) ||
 1496             ((int)vap->va_bytes != VNOVAL) ||
 1497             (vap->va_gen != VNOVAL)) {
 1498                 return (EINVAL);
 1499         }
 1500 
 1501         de = vp->v_data;
 1502         if (vp->v_type == VDIR)
 1503                 de = de->de_dir;
 1504 
 1505         error = c = 0;
 1506         if (vap->va_uid == (uid_t)VNOVAL)
 1507                 uid = de->de_uid;
 1508         else
 1509                 uid = vap->va_uid;
 1510         if (vap->va_gid == (gid_t)VNOVAL)
 1511                 gid = de->de_gid;
 1512         else
 1513                 gid = vap->va_gid;
 1514         if (uid != de->de_uid || gid != de->de_gid) {
 1515                 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
 1516                     (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
 1517                         error = priv_check(td, PRIV_VFS_CHOWN);
 1518                         if (error)
 1519                                 return (error);
 1520                 }
 1521                 de->de_uid = uid;
 1522                 de->de_gid = gid;
 1523                 c = 1;
 1524         }
 1525 
 1526         if (vap->va_mode != (mode_t)VNOVAL) {
 1527                 if (ap->a_cred->cr_uid != de->de_uid) {
 1528                         error = priv_check(td, PRIV_VFS_ADMIN);
 1529                         if (error)
 1530                                 return (error);
 1531                 }
 1532                 de->de_mode = vap->va_mode;
 1533                 c = 1;
 1534         }
 1535 
 1536         if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
 1537                 /* See the comment in ufs_vnops::ufs_setattr(). */
 1538                 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
 1539                     ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
 1540                     (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
 1541                         return (error);
 1542                 if (vap->va_atime.tv_sec != VNOVAL) {
 1543                         if (vp->v_type == VCHR)
 1544                                 vp->v_rdev->si_atime = vap->va_atime;
 1545                         else
 1546                                 de->de_atime = vap->va_atime;
 1547                 }
 1548                 if (vap->va_mtime.tv_sec != VNOVAL) {
 1549                         if (vp->v_type == VCHR)
 1550                                 vp->v_rdev->si_mtime = vap->va_mtime;
 1551                         else
 1552                                 de->de_mtime = vap->va_mtime;
 1553                 }
 1554                 c = 1;
 1555         }
 1556 
 1557         if (c) {
 1558                 if (vp->v_type == VCHR)
 1559                         vfs_timestamp(&vp->v_rdev->si_ctime);
 1560                 else
 1561                         vfs_timestamp(&de->de_mtime);
 1562         }
 1563         return (0);
 1564 }
 1565 
 1566 #ifdef MAC
 1567 static int
 1568 devfs_setlabel(struct vop_setlabel_args *ap)
 1569 {
 1570         struct vnode *vp;
 1571         struct devfs_dirent *de;
 1572 
 1573         vp = ap->a_vp;
 1574         de = vp->v_data;
 1575 
 1576         mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
 1577         mac_devfs_update(vp->v_mount, de, vp);
 1578 
 1579         return (0);
 1580 }
 1581 #endif
 1582 
 1583 static int
 1584 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
 1585 {
 1586 
 1587         return (vnops.fo_stat(fp, sb, cred, td));
 1588 }
 1589 
 1590 static int
 1591 devfs_symlink(struct vop_symlink_args *ap)
 1592 {
 1593         int i, error;
 1594         struct devfs_dirent *dd;
 1595         struct devfs_dirent *de, *de_covered, *de_dotdot;
 1596         struct devfs_mount *dmp;
 1597 
 1598         error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
 1599         if (error)
 1600                 return(error);
 1601         dmp = VFSTODEVFS(ap->a_dvp->v_mount);
 1602         if (devfs_populate_vp(ap->a_dvp) != 0)
 1603                 return (ENOENT);
 1604 
 1605         dd = ap->a_dvp->v_data;
 1606         de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
 1607         de->de_flags = DE_USER;
 1608         de->de_uid = 0;
 1609         de->de_gid = 0;
 1610         de->de_mode = 0755;
 1611         de->de_inode = alloc_unr(devfs_inos);
 1612         de->de_dir = dd;
 1613         de->de_dirent->d_type = DT_LNK;
 1614         i = strlen(ap->a_target) + 1;
 1615         de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
 1616         bcopy(ap->a_target, de->de_symlink, i);
 1617 #ifdef MAC
 1618         mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
 1619 #endif
 1620         de_covered = devfs_find(dd, de->de_dirent->d_name,
 1621             de->de_dirent->d_namlen, 0);
 1622         if (de_covered != NULL) {
 1623                 if ((de_covered->de_flags & DE_USER) != 0) {
 1624                         devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
 1625                         sx_xunlock(&dmp->dm_lock);
 1626                         return (EEXIST);
 1627                 }
 1628                 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
 1629                     ("devfs_symlink: entry %p already covered", de_covered));
 1630                 de_covered->de_flags |= DE_COVERED;
 1631         }
 1632 
 1633         de_dotdot = TAILQ_FIRST(&dd->de_dlist);         /* "." */
 1634         de_dotdot = TAILQ_NEXT(de_dotdot, de_list);     /* ".." */
 1635         TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
 1636         devfs_dir_ref_de(dmp, dd);
 1637         devfs_rules_apply(dmp, de);
 1638 
 1639         return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
 1640 }
 1641 
 1642 static int
 1643 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
 1644 {
 1645 
 1646         return (vnops.fo_truncate(fp, length, cred, td));
 1647 }
 1648 
 1649 static int
 1650 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
 1651     int flags, struct thread *td)
 1652 {
 1653         struct cdev *dev;
 1654         int error, ioflag, ref;
 1655         ssize_t resid;
 1656         struct cdevsw *dsw;
 1657         struct file *fpop;
 1658 
 1659         if (uio->uio_resid > DEVFS_IOSIZE_MAX)
 1660                 return (EINVAL);
 1661         fpop = td->td_fpop;
 1662         error = devfs_fp_check(fp, &dev, &dsw, &ref);
 1663         if (error)
 1664                 return (error);
 1665         KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
 1666         ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
 1667         if (ioflag & O_DIRECT)
 1668                 ioflag |= IO_DIRECT;
 1669         foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
 1670 
 1671         resid = uio->uio_resid;
 1672 
 1673         error = dsw->d_write(dev, uio, ioflag);
 1674         if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
 1675                 vfs_timestamp(&dev->si_ctime);
 1676                 dev->si_mtime = dev->si_ctime;
 1677         }
 1678         td->td_fpop = fpop;
 1679         dev_relthread(dev, ref);
 1680 
 1681         foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
 1682         return (error);
 1683 }
 1684 
 1685 dev_t
 1686 dev2udev(struct cdev *x)
 1687 {
 1688         if (x == NULL)
 1689                 return (NODEV);
 1690         return (cdev2priv(x)->cdp_inode);
 1691 }
 1692 
 1693 static struct fileops devfs_ops_f = {
 1694         .fo_read =      devfs_read_f,
 1695         .fo_write =     devfs_write_f,
 1696         .fo_truncate =  devfs_truncate_f,
 1697         .fo_ioctl =     devfs_ioctl_f,
 1698         .fo_poll =      devfs_poll_f,
 1699         .fo_kqfilter =  devfs_kqfilter_f,
 1700         .fo_stat =      devfs_stat_f,
 1701         .fo_close =     devfs_close_f,
 1702         .fo_chmod =     vn_chmod,
 1703         .fo_chown =     vn_chown,
 1704         .fo_flags =     DFLAG_PASSABLE | DFLAG_SEEKABLE
 1705 };
 1706 
 1707 static struct vop_vector devfs_vnodeops = {
 1708         .vop_default =          &default_vnodeops,
 1709 
 1710         .vop_access =           devfs_access,
 1711         .vop_getattr =          devfs_getattr,
 1712         .vop_ioctl =            devfs_rioctl,
 1713         .vop_lookup =           devfs_lookup,
 1714         .vop_mknod =            devfs_mknod,
 1715         .vop_pathconf =         devfs_pathconf,
 1716         .vop_read =             devfs_rread,
 1717         .vop_readdir =          devfs_readdir,
 1718         .vop_readlink =         devfs_readlink,
 1719         .vop_reclaim =          devfs_reclaim,
 1720         .vop_remove =           devfs_remove,
 1721         .vop_revoke =           devfs_revoke,
 1722         .vop_setattr =          devfs_setattr,
 1723 #ifdef MAC
 1724         .vop_setlabel =         devfs_setlabel,
 1725 #endif
 1726         .vop_symlink =          devfs_symlink,
 1727         .vop_vptocnp =          devfs_vptocnp,
 1728 };
 1729 
 1730 static struct vop_vector devfs_specops = {
 1731         .vop_default =          &default_vnodeops,
 1732 
 1733         .vop_access =           devfs_access,
 1734         .vop_bmap =             VOP_PANIC,
 1735         .vop_close =            devfs_close,
 1736         .vop_create =           VOP_PANIC,
 1737         .vop_fsync =            devfs_fsync,
 1738         .vop_getattr =          devfs_getattr,
 1739         .vop_link =             VOP_PANIC,
 1740         .vop_mkdir =            VOP_PANIC,
 1741         .vop_mknod =            VOP_PANIC,
 1742         .vop_open =             devfs_open,
 1743         .vop_pathconf =         devfs_pathconf,
 1744         .vop_print =            devfs_print,
 1745         .vop_read =             VOP_PANIC,
 1746         .vop_readdir =          VOP_PANIC,
 1747         .vop_readlink =         VOP_PANIC,
 1748         .vop_reallocblks =      VOP_PANIC,
 1749         .vop_reclaim =          devfs_reclaim,
 1750         .vop_remove =           devfs_remove,
 1751         .vop_rename =           VOP_PANIC,
 1752         .vop_revoke =           devfs_revoke,
 1753         .vop_rmdir =            VOP_PANIC,
 1754         .vop_setattr =          devfs_setattr,
 1755 #ifdef MAC
 1756         .vop_setlabel =         devfs_setlabel,
 1757 #endif
 1758         .vop_strategy =         VOP_PANIC,
 1759         .vop_symlink =          VOP_PANIC,
 1760         .vop_vptocnp =          devfs_vptocnp,
 1761         .vop_write =            VOP_PANIC,
 1762 };
 1763 
 1764 /*
 1765  * Our calling convention to the device drivers used to be that we passed
 1766  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_ 
 1767  * flags instead since that's what open(), close() and ioctl() takes and
 1768  * we don't really want vnode.h in device drivers.
 1769  * We solved the source compatibility by redefining some vnode flags to
 1770  * be the same as the fcntl ones and by sending down the bitwise OR of
 1771  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
 1772  * pulls the rug out under this.
 1773  */
 1774 CTASSERT(O_NONBLOCK == IO_NDELAY);
 1775 CTASSERT(O_FSYNC == IO_SYNC);
Cache object: 71b91bb9c33331bfe4c9f4c3170e8f9c
[ 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
