sys/kern/vfs_helper.c

   1 /*
   2  * (The copyright below applies to ufs_access())
   3  *
   4  * Copyright (c) 1982, 1986, 1989, 1993, 1995
   5  *      The Regents of the University of California.  All rights reserved.
   6  * (c) UNIX System Laboratories, Inc.
   7  * All or some portions of this file are derived from material licensed
   8  * to the University of California by American Telephone and Telegraph
   9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  10  * the permission of UNIX System Laboratories, Inc.
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions and the following disclaimer.
  17  * 2. Redistributions in binary form must reproduce the above copyright
  18  *    notice, this list of conditions and the following disclaimer in the
  19  *    documentation and/or other materials provided with the distribution.
  20  * 3. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  *
  36  * @(#)ufs_vnops.c      8.27 (Berkeley) 5/27/95
  37  * $DragonFly: src/sys/kern/vfs_helper.c,v 1.5 2008/05/25 18:34:46 dillon Exp $
  38  */
  39
  40 #include "opt_quota.h"
  41 #include "opt_suiddir.h"
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/conf.h>
  46 #include <sys/kernel.h>
  47 #include <sys/fcntl.h>
  48 #include <sys/stat.h>
  49 #include <sys/mount.h>
  50 #include <sys/unistd.h>
  51 #include <sys/vnode.h>
  52 #include <sys/file.h>           /* XXX */
  53 #include <sys/proc.h>
  54 #include <sys/priv.h>
  55 #include <sys/jail.h>
  56 #include <sys/sysctl.h>
  57 #include <sys/sfbuf.h>
  58 #include <vm/vm_extern.h>
  59 #include <vm/vm_object.h>
  60 #include <vm/vm_page2.h>
  61
  62 #ifdef LWBUF_IS_OPTIMAL
  63
  64 static int vm_read_shortcut_enable = 1;
  65 static long vm_read_shortcut_count;
  66 static long vm_read_shortcut_failed;
  67 SYSCTL_INT(_vm, OID_AUTO, read_shortcut_enable, CTLFLAG_RW,
  68           &vm_read_shortcut_enable, 0, "Direct vm_object vop_read shortcut");
  69 SYSCTL_LONG(_vm, OID_AUTO, read_shortcut_count, CTLFLAG_RW,
  70           &vm_read_shortcut_count, 0, "Statistics");
  71 SYSCTL_LONG(_vm, OID_AUTO, read_shortcut_failed, CTLFLAG_RW,
  72           &vm_read_shortcut_failed, 0, "Statistics");
  73
  74 #endif
  75
  76 /*
  77  * vop_helper_access()
  78  *
  79  *      Provide standard UNIX semanics for VOP_ACCESS, but without the quota
  80  *      code.  This procedure was basically pulled out of UFS.
  81  */
  82 int
  83 vop_helper_access(struct vop_access_args *ap, uid_t ino_uid, gid_t ino_gid,
  84                   mode_t ino_mode, u_int32_t ino_flags)
  85 {
  86         struct vnode *vp = ap->a_vp;
  87         struct ucred *cred = ap->a_cred;
  88         mode_t mask, mode = ap->a_mode;
  89         gid_t *gp;
  90         int i;
  91         uid_t proc_uid;
  92         gid_t proc_gid;
  93
  94         if (ap->a_flags & AT_EACCESS) {
  95                 proc_uid = cred->cr_uid;
  96                 proc_gid = cred->cr_gid;
  97         } else {
  98                 proc_uid = cred->cr_ruid;
  99                 proc_gid = cred->cr_rgid;
 100         }
 101
 102         /*
 103          * Disallow write attempts on read-only filesystems;
 104          * unless the file is a socket, fifo, or a block or
 105          * character device resident on the filesystem.
 106          */
 107         if (mode & VWRITE) {
 108                 switch (vp->v_type) {
 109                 case VDIR:
 110                 case VLNK:
 111                 case VREG:
 112                 case VDATABASE:
 113                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
 114                                 return (EROFS);
 115                         break;
 116                 default:
 117                         break;
 118                 }
 119         }
 120
 121         /* If immutable bit set, nobody gets to write it. */
 122         if ((mode & VWRITE) && (ino_flags & IMMUTABLE))
 123                 return (EPERM);
 124
 125         /* Otherwise, user id 0 always gets access. */
 126         if (proc_uid == 0)
 127                 return (0);
 128
 129         mask = 0;
 130
 131         /* Otherwise, check the owner. */
 132         if (proc_uid == ino_uid) {
 133                 if (mode & VEXEC)
 134                         mask |= S_IXUSR;
 135                 if (mode & VREAD)
 136                         mask |= S_IRUSR;
 137                 if (mode & VWRITE)
 138                         mask |= S_IWUSR;
 139                 return ((ino_mode & mask) == mask ? 0 : EACCES);
 140         }
 141
 142         /*
 143          * Otherwise, check the groups.
 144          * We must special-case the primary group to, if needed, check against
 145          * the real gid and not the effective one.
 146          */
 147         if (proc_gid == ino_gid) {
 148                 if (mode & VEXEC)
 149                         mask |= S_IXGRP;
 150                 if (mode & VREAD)
 151                         mask |= S_IRGRP;
 152                 if (mode & VWRITE)
 153                         mask |= S_IWGRP;
 154                 return ((ino_mode & mask) == mask ? 0 : EACCES);
 155         }
 156         for (i = 1, gp = &cred->cr_groups[1]; i < cred->cr_ngroups; i++, gp++)
 157                 if (ino_gid == *gp) {
 158                         if (mode & VEXEC)
 159                                 mask |= S_IXGRP;
 160                         if (mode & VREAD)
 161                                 mask |= S_IRGRP;
 162                         if (mode & VWRITE)
 163                                 mask |= S_IWGRP;
 164                         return ((ino_mode & mask) == mask ? 0 : EACCES);
 165                 }
 166
 167         /* Otherwise, check everyone else. */
 168         if (mode & VEXEC)
 169                 mask |= S_IXOTH;
 170         if (mode & VREAD)
 171                 mask |= S_IROTH;
 172         if (mode & VWRITE)
 173                 mask |= S_IWOTH;
 174         return ((ino_mode & mask) == mask ? 0 : EACCES);
 175 }
 176
 177 int
 178 vop_helper_setattr_flags(u_int32_t *ino_flags, u_int32_t vaflags,
 179                          uid_t uid, struct ucred *cred)
 180 {
 181         int error;
 182
 183         /*
 184          * If uid doesn't match only a privileged user can change the flags
 185          */
 186         if (cred->cr_uid != uid &&
 187             (error = priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0))) {
 188                 return(error);
 189         }
 190         if (cred->cr_uid == 0 &&
 191             (!jailed(cred)|| jail_chflags_allowed)) {
 192                 if ((*ino_flags & (SF_NOUNLINK|SF_IMMUTABLE|SF_APPEND)) &&
 193                     securelevel > 0)
 194                         return (EPERM);
 195                 *ino_flags = vaflags;
 196         } else {
 197                 if (*ino_flags & (SF_NOUNLINK|SF_IMMUTABLE|SF_APPEND) ||
 198                     (vaflags & UF_SETTABLE) != vaflags)
 199                         return (EPERM);
 200                 *ino_flags &= SF_SETTABLE;
 201                 *ino_flags |= vaflags & UF_SETTABLE;
 202         }
 203         return(0);
 204 }
 205
 206 /*
 207  * This helper function may be used by VFSs to implement UNIX initial
 208  * ownership semantics when creating new objects inside directories.
 209  */
 210 uid_t
 211 vop_helper_create_uid(struct mount *mp, mode_t dmode, uid_t duid,
 212                       struct ucred *cred, mode_t *modep)
 213 {
 214 #ifdef SUIDDIR
 215         if ((mp->mnt_flag & MNT_SUIDDIR) && (dmode & S_ISUID) &&
 216             duid != cred->cr_uid && duid) {
 217                 *modep &= ~07111;
 218                 return(duid);
 219         }
 220 #endif
 221         return(cred->cr_uid);
 222 }
 223
 224 /*
 225  * This helper may be used by VFSs to implement unix chmod semantics.
 226  */
 227 int
 228 vop_helper_chmod(struct vnode *vp, mode_t new_mode, struct ucred *cred,
 229                  uid_t cur_uid, gid_t cur_gid, mode_t *cur_modep)
 230 {
 231         int error;
 232
 233         if (cred->cr_uid != cur_uid) {
 234                 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0);
 235                 if (error)
 236                         return (error);
 237         }
 238         if (cred->cr_uid) {
 239                 if (vp->v_type != VDIR && (*cur_modep & S_ISTXT))
 240                         return (EFTYPE);
 241                 if (!groupmember(cur_gid, cred) && (*cur_modep & S_ISGID))
 242                         return (EPERM);
 243         }
 244         *cur_modep &= ~ALLPERMS;
 245         *cur_modep |= new_mode & ALLPERMS;
 246         return(0);
 247 }
 248
 249 /*
 250  * This helper may be used by VFSs to implement unix chown semantics.
 251  */
 252 int
 253 vop_helper_chown(struct vnode *vp, uid_t new_uid, gid_t new_gid,
 254                  struct ucred *cred,
 255                  uid_t *cur_uidp, gid_t *cur_gidp, mode_t *cur_modep)
 256 {
 257         gid_t ogid;
 258         uid_t ouid;
 259         int error;
 260
 261         if (new_uid == (uid_t)VNOVAL)
 262                 new_uid = *cur_uidp;
 263         if (new_gid == (gid_t)VNOVAL)
 264                 new_gid = *cur_gidp;
 265
 266         /*
 267          * If we don't own the file, are trying to change the owner
 268          * of the file, or are not a member of the target group,
 269          * the caller must be privileged or the call fails.
 270          */
 271         if ((cred->cr_uid != *cur_uidp || new_uid != *cur_uidp ||
 272             (new_gid != *cur_gidp && !(cred->cr_gid == new_gid ||
 273             groupmember(new_gid, cred)))) &&
 274             (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) {
 275                 return (error);
 276         }
 277         ogid = *cur_gidp;
 278         ouid = *cur_uidp;
 279         /* XXX QUOTA CODE */
 280         *cur_uidp = new_uid;
 281         *cur_gidp = new_gid;
 282         /* XXX QUOTA CODE */
 283
 284         /*
 285          * DragonFly clears both SUID and SGID if either the owner or
 286          * group is changed and root isn't doing it.  If root is doing
 287          * it we do not clear SUID/SGID.
 288          */
 289         if (cred->cr_uid != 0 && (ouid != new_uid || ogid != new_gid))
 290                 *cur_modep &= ~(S_ISUID | S_ISGID);
 291         return(0);
 292 }
 293
 294 #ifdef LWBUF_IS_OPTIMAL
 295
 296 /*
 297  * A VFS can call this function to try to dispose of a read request
 298  * directly from the VM system, pretty much bypassing almost all VFS
 299  * overhead except for atime updates.
 300  *
 301  * If 0 is returned some or all of the uio was handled.  The caller must
 302  * check the uio and handle the remainder.
 303  *
 304  * The caller must fail on a non-zero error.
 305  */
 306 int
 307 vop_helper_read_shortcut(struct vop_read_args *ap)
 308 {
 309         struct vnode *vp;
 310         struct uio *uio;
 311         struct lwbuf *lwb;
 312         struct lwbuf lwb_cache;
 313         vm_object_t obj;
 314         vm_page_t m;
 315         int offset;
 316         int n;
 317         int error;
 318
 319         vp = ap->a_vp;
 320         uio = ap->a_uio;
 321
 322         /*
 323          * We can't short-cut if there is no VM object or this is a special
 324          * UIO_NOCOPY read (typically from VOP_STRATEGY()).  We also can't
 325          * do this if we cannot extract the filesize from the vnode.
 326          */
 327         if (vm_read_shortcut_enable == 0)
 328                 return(0);
 329         if (vp->v_object == NULL || uio->uio_segflg == UIO_NOCOPY)
 330                 return(0);
 331         if (vp->v_filesize == NOOFFSET)
 332                 return(0);
 333         if (uio->uio_resid == 0)
 334                 return(0);
 335
 336         /*
 337          * Iterate the uio on a page-by-page basis
 338          *
 339          * XXX can we leave the object held shared during the uiomove()?
 340          */
 341         ++vm_read_shortcut_count;
 342         obj = vp->v_object;
 343         vm_object_hold_shared(obj);
 344
 345         error = 0;
 346         while (uio->uio_resid && error == 0) {
 347                 offset = (int)uio->uio_offset & PAGE_MASK;
 348                 n = PAGE_SIZE - offset;
 349                 if (n > uio->uio_resid)
 350                         n = uio->uio_resid;
 351                 if (vp->v_filesize < uio->uio_offset)
 352                         break;
 353                 if (uio->uio_offset + n > vp->v_filesize)
 354                         n = vp->v_filesize - uio->uio_offset;
 355                 if (n == 0)
 356                         break;  /* hit EOF */
 357
 358                 m = vm_page_lookup_busy_try(obj, OFF_TO_IDX(uio->uio_offset),
 359                                             FALSE, &error);
 360                 if (error || m == NULL) {
 361                         ++vm_read_shortcut_failed;
 362                         error = 0;
 363                         break;
 364                 }
 365                 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
 366                         ++vm_read_shortcut_failed;
 367                         vm_page_wakeup(m);
 368                         break;
 369                 }
 370                 lwb = lwbuf_alloc(m, &lwb_cache);
 371
 372                 /*
 373                  * Use a no-fault uiomove() to avoid deadlocking against
 374                  * our VM object (which could livelock on the same object
 375                  * due to shared-vs-exclusive), or deadlocking against
 376                  * our busied page.  Returns EFAULT on any fault which
 377                  * winds up diving a vnode.
 378                  */
 379                 error = uiomove_nofault((char *)lwbuf_kva(lwb) + offset,
 380                                         n, uio);
 381
 382                 vm_page_flag_set(m, PG_REFERENCED);
 383                 lwbuf_free(lwb);
 384                 vm_page_wakeup(m);
 385         }
 386         vm_object_drop(obj);
 387
 388         /*
 389          * Ignore EFAULT since we used uiomove_nofault(), causes caller
 390          * to fall-back to normal code for this case.
 391          */
 392         if (error == EFAULT)
 393                 error = 0;
 394
 395         return (error);
 396 }
 397
 398 #else
 399
 400 /*
 401  * If lwbuf's aren't optimal then it's best to just use the buffer
 402  * cache.
 403  */
 404 int
 405 vop_helper_read_shortcut(struct vop_read_args *ap)
 406 {
 407         return(0);
 408 }
 409
 410 #endif