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  */
  38
  39 #include "opt_quota.h"
  40 #include "opt_suiddir.h"
  41
  42 #include <sys/param.h>
  43 #include <sys/systm.h>
  44 #include <sys/uio.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 SYSCTL_INT(_vm, OID_AUTO, read_shortcut_enable, CTLFLAG_RW,
  66           &vm_read_shortcut_enable, 0, "Direct vm_object vop_read shortcut");
  67
  68 #endif
  69
  70 /*
  71  * vop_helper_access()
  72  *
  73  *      Provide standard UNIX semanics for VOP_ACCESS, but without the quota
  74  *      code.  This procedure was basically pulled out of UFS.
  75  */
  76 int
  77 vop_helper_access(struct vop_access_args *ap, uid_t ino_uid, gid_t ino_gid,
  78                   mode_t ino_mode, u_int32_t ino_flags)
  79 {
  80         struct vnode *vp = ap->a_vp;
  81         struct ucred *cred = ap->a_cred;
  82         mode_t mask, mode = ap->a_mode;
  83         gid_t *gp;
  84         int i;
  85         uid_t proc_uid;
  86         gid_t proc_gid;
  87
  88         if (ap->a_flags & AT_EACCESS) {
  89                 proc_uid = cred->cr_uid;
  90                 proc_gid = cred->cr_gid;
  91         } else {
  92                 proc_uid = cred->cr_ruid;
  93                 proc_gid = cred->cr_rgid;
  94         }
  95
  96         /*
  97          * Disallow write attempts on read-only filesystems;
  98          * unless the file is a socket, fifo, or a block or
  99          * character device resident on the filesystem.
 100          */
 101         if (mode & VWRITE) {
 102                 switch (vp->v_type) {
 103                 case VDIR:
 104                 case VLNK:
 105                 case VREG:
 106                 case VDATABASE:
 107                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
 108                                 return (EROFS);
 109                         break;
 110                 default:
 111                         break;
 112                 }
 113         }
 114
 115         /* If immutable bit set, nobody gets to write it. */
 116         if ((mode & VWRITE) && (ino_flags & IMMUTABLE))
 117                 return (EPERM);
 118
 119         /* Otherwise, user id 0 always gets access. */
 120         if (proc_uid == 0)
 121                 return (0);
 122
 123         mask = 0;
 124
 125         /* Otherwise, check the owner. */
 126         if (proc_uid == ino_uid) {
 127                 if (mode & VEXEC)
 128                         mask |= S_IXUSR;
 129                 if (mode & VREAD)
 130                         mask |= S_IRUSR;
 131                 if (mode & VWRITE)
 132                         mask |= S_IWUSR;
 133                 return ((ino_mode & mask) == mask ? 0 : EACCES);
 134         }
 135
 136         /*
 137          * Otherwise, check the groups.
 138          * We must special-case the primary group to, if needed, check against
 139          * the real gid and not the effective one.
 140          */
 141         if (proc_gid == ino_gid) {
 142                 if (mode & VEXEC)
 143                         mask |= S_IXGRP;
 144                 if (mode & VREAD)
 145                         mask |= S_IRGRP;
 146                 if (mode & VWRITE)
 147                         mask |= S_IWGRP;
 148                 return ((ino_mode & mask) == mask ? 0 : EACCES);
 149         }
 150         for (i = 1, gp = &cred->cr_groups[1]; i < cred->cr_ngroups; i++, gp++)
 151                 if (ino_gid == *gp) {
 152                         if (mode & VEXEC)
 153                                 mask |= S_IXGRP;
 154                         if (mode & VREAD)
 155                                 mask |= S_IRGRP;
 156                         if (mode & VWRITE)
 157                                 mask |= S_IWGRP;
 158                         return ((ino_mode & mask) == mask ? 0 : EACCES);
 159                 }
 160
 161         /* Otherwise, check everyone else. */
 162         if (mode & VEXEC)
 163                 mask |= S_IXOTH;
 164         if (mode & VREAD)
 165                 mask |= S_IROTH;
 166         if (mode & VWRITE)
 167                 mask |= S_IWOTH;
 168         return ((ino_mode & mask) == mask ? 0 : EACCES);
 169 }
 170
 171 int
 172 vop_helper_setattr_flags(u_int32_t *ino_flags, u_int32_t vaflags,
 173                          uid_t uid, struct ucred *cred)
 174 {
 175         int error;
 176
 177         /*
 178          * If uid doesn't match only a privileged user can change the flags
 179          */
 180         if (cred->cr_uid != uid &&
 181             (error = priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0))) {
 182                 return(error);
 183         }
 184         if (cred->cr_uid == 0 &&
 185             (!jailed(cred) || PRISON_CAP_ISSET(cred->cr_prison->pr_caps,
 186                 PRISON_CAP_VFS_CHFLAGS))) {
 187                 if ((*ino_flags & (SF_NOUNLINK|SF_IMMUTABLE|SF_APPEND)) &&
 188                     securelevel > 0)
 189                         return (EPERM);
 190                 *ino_flags = vaflags;
 191         } else {
 192                 if (*ino_flags & (SF_NOUNLINK|SF_IMMUTABLE|SF_APPEND) ||
 193                     (vaflags & UF_SETTABLE) != vaflags)
 194                         return (EPERM);
 195                 *ino_flags &= SF_SETTABLE;
 196                 *ino_flags |= vaflags & UF_SETTABLE;
 197         }
 198         return(0);
 199 }
 200
 201 /*
 202  * This helper function may be used by VFSs to implement UNIX initial
 203  * ownership semantics when creating new objects inside directories.
 204  */
 205 uid_t
 206 vop_helper_create_uid(struct mount *mp, mode_t dmode, uid_t duid,
 207                       struct ucred *cred, mode_t *modep)
 208 {
 209 #ifdef SUIDDIR
 210         if ((mp->mnt_flag & MNT_SUIDDIR) && (dmode & S_ISUID) &&
 211             duid != cred->cr_uid && duid) {
 212                 *modep &= ~07111;
 213                 return(duid);
 214         }
 215 #endif
 216         return(cred->cr_uid);
 217 }
 218
 219 /*
 220  * This helper may be used by VFSs to implement unix chmod semantics.
 221  */
 222 int
 223 vop_helper_chmod(struct vnode *vp, mode_t new_mode, struct ucred *cred,
 224                  uid_t cur_uid, gid_t cur_gid, mode_t *cur_modep)
 225 {
 226         int error;
 227
 228         if (cred->cr_uid != cur_uid) {
 229                 error = priv_check_cred(cred, PRIV_VFS_CHMOD, 0);
 230                 if (error)
 231                         return (error);
 232         }
 233         if (cred->cr_uid) {
 234                 if (vp->v_type != VDIR && (*cur_modep & S_ISTXT))
 235                         return (EFTYPE);
 236                 if (!groupmember(cur_gid, cred) && (*cur_modep & S_ISGID))
 237                         return (EPERM);
 238         }
 239         *cur_modep &= ~ALLPERMS;
 240         *cur_modep |= new_mode & ALLPERMS;
 241         return(0);
 242 }
 243
 244 /*
 245  * This helper may be used by VFSs to implement unix chown semantics.
 246  */
 247 int
 248 vop_helper_chown(struct vnode *vp, uid_t new_uid, gid_t new_gid,
 249                  struct ucred *cred,
 250                  uid_t *cur_uidp, gid_t *cur_gidp, mode_t *cur_modep)
 251 {
 252         gid_t ogid;
 253         uid_t ouid;
 254         int error;
 255
 256         if (new_uid == (uid_t)VNOVAL)
 257                 new_uid = *cur_uidp;
 258         if (new_gid == (gid_t)VNOVAL)
 259                 new_gid = *cur_gidp;
 260
 261         /*
 262          * If we don't own the file, are trying to change the owner
 263          * of the file, or are not a member of the target group,
 264          * the caller must be privileged or the call fails.
 265          */
 266         if ((cred->cr_uid != *cur_uidp || new_uid != *cur_uidp ||
 267             (new_gid != *cur_gidp && !(cred->cr_gid == new_gid ||
 268             groupmember(new_gid, cred)))) &&
 269             (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) {
 270                 return (error);
 271         }
 272         ogid = *cur_gidp;
 273         ouid = *cur_uidp;
 274         /* XXX QUOTA CODE */
 275         *cur_uidp = new_uid;
 276         *cur_gidp = new_gid;
 277         /* XXX QUOTA CODE */
 278
 279         /*
 280          * DragonFly clears both SUID and SGID if either the owner or
 281          * group is changed and root isn't doing it.  If root is doing
 282          * it we do not clear SUID/SGID.
 283          */
 284         if (cred->cr_uid != 0 && (ouid != new_uid || ogid != new_gid))
 285                 *cur_modep &= ~(S_ISUID | S_ISGID);
 286         return(0);
 287 }
 288
 289 #ifdef LWBUF_IS_OPTIMAL
 290
 291 /*
 292  * A VFS can call this function to try to dispose of a read request
 293  * directly from the VM system, pretty much bypassing almost all VFS
 294  * overhead except for atime updates.
 295  *
 296  * If 0 is returned some or all of the uio was handled.  The caller must
 297  * check the uio and handle the remainder.
 298  *
 299  * The caller must fail on a non-zero error.
 300  */
 301 int
 302 vop_helper_read_shortcut(struct vop_read_args *ap)
 303 {
 304         struct vnode *vp;
 305         struct uio *uio;
 306         struct lwbuf *lwb;
 307         struct lwbuf lwb_cache;
 308         vm_object_t obj;
 309         vm_page_t m;
 310         int offset;
 311         int n;
 312         int error;
 313
 314         vp = ap->a_vp;
 315         uio = ap->a_uio;
 316
 317         /*
 318          * We can't short-cut if there is no VM object or this is a special
 319          * UIO_NOCOPY read (typically from VOP_STRATEGY()).  We also can't
 320          * do this if we cannot extract the filesize from the vnode.
 321          */
 322         if (vm_read_shortcut_enable == 0)
 323                 return(0);
 324         if (vp->v_object == NULL || uio->uio_segflg == UIO_NOCOPY)
 325                 return(0);
 326         if (vp->v_filesize == NOOFFSET)
 327                 return(0);
 328         if (uio->uio_resid == 0)
 329                 return(0);
 330
 331         /*
 332          * Iterate the uio on a page-by-page basis
 333          *
 334          * XXX can we leave the object held shared during the uiomove()?
 335          */
 336         obj = vp->v_object;
 337         vm_object_hold_shared(obj);
 338
 339         error = 0;
 340         while (uio->uio_resid && error == 0) {
 341                 offset = (int)uio->uio_offset & PAGE_MASK;
 342                 n = PAGE_SIZE - offset;
 343                 if (n > uio->uio_resid)
 344                         n = uio->uio_resid;
 345                 if (vp->v_filesize < uio->uio_offset)
 346                         break;
 347                 if (uio->uio_offset + n > vp->v_filesize)
 348                         n = vp->v_filesize - uio->uio_offset;
 349                 if (n == 0)
 350                         break;  /* hit EOF */
 351
 352                 m = vm_page_lookup_sbusy_try(obj, OFF_TO_IDX(uio->uio_offset),
 353                                              0, PAGE_SIZE);
 354                 if (error || m == NULL) {
 355                         error = 0;
 356                         break;
 357                 }
 358                 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
 359                         vm_page_sbusy_drop(m);
 360                         break;
 361                 }
 362                 lwb = lwbuf_alloc(m, &lwb_cache);
 363
 364                 /*
 365                  * Use a no-fault uiomove() to avoid deadlocking against
 366                  * our VM object (which could livelock on the same object
 367                  * due to shared-vs-exclusive), or deadlocking against
 368                  * our busied page.  Returns EFAULT on any fault which
 369                  * winds up diving a vnode.
 370                  */
 371                 error = uiomove_nofault((char *)lwbuf_kva(lwb) + offset,
 372                                         n, uio);
 373
 374                 vm_page_flag_set(m, PG_REFERENCED);
 375                 lwbuf_free(lwb);
 376                 vm_page_sbusy_drop(m);
 377         }
 378         vm_object_drop(obj);
 379
 380         /*
 381          * Ignore EFAULT since we used uiomove_nofault(), causes caller
 382          * to fall-back to normal code for this case.
 383          */
 384         if (error == EFAULT)
 385                 error = 0;
 386
 387         return (error);
 388 }
 389
 390 #else
 391
 392 /*
 393  * If lwbuf's aren't optimal then it's best to just use the buffer
 394  * cache.
 395  */
 396 int
 397 vop_helper_read_shortcut(struct vop_read_args *ap)
 398 {
 399         return(0);
 400 }
 401
 402 #endif