sys/vfs/umapfs/umap_vnops.c

   1 /*
   2  * Copyright (c) 1992, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software donated to Berkeley by
   6  * the UCLA Ficus project.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. 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  *      @(#)umap_vnops.c        8.6 (Berkeley) 5/22/95
  37  * $FreeBSD: src/sys/miscfs/umapfs/umap_vnops.c,v 1.30 1999/08/30 07:08:04 bde Exp $
  38  * $DragonFly: src/sys/vfs/umapfs/Attic/umap_vnops.c,v 1.7 2004/04/24 04:32:05 drhodus Exp $
  39  */
  40
  41 /*
  42  * Umap Layer
  43  */
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/kernel.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/vnode.h>
  50 #include <sys/mount.h>
  51 #include <sys/proc.h>
  52 #include <sys/namei.h>
  53 #include <sys/malloc.h>
  54 #include <sys/buf.h>
  55 #include "umap.h"
  56 #include <vfs/nullfs/null.h>
  57
  58 static int umap_bug_bypass = 0;   /* for debugging: enables bypass printf'ing */
  59 SYSCTL_INT(_debug, OID_AUTO, umapfs_bug_bypass, CTLFLAG_RW,
  60         &umap_bug_bypass, 0, "");
  61
  62 static int      umap_bypass (struct vop_generic_args *ap);
  63 static int      umap_getattr (struct vop_getattr_args *ap);
  64 static int      umap_inactive (struct vop_inactive_args *ap);
  65 static int      umap_lock (struct vop_lock_args *ap);
  66 static int      umap_print (struct vop_print_args *ap);
  67 static int      umap_reclaim (struct vop_reclaim_args *ap);
  68 static int      umap_rename (struct vop_rename_args *ap);
  69 static int      umap_unlock (struct vop_unlock_args *ap);
  70
  71 /*
  72  * This is the 10-Apr-92 bypass routine.
  73  * See null_vnops.c:null_bypass for more details.
  74  */
  75 static int
  76 umap_bypass(ap)
  77         struct vop_generic_args /* {
  78                 struct vnodeop_desc *a_desc;
  79                 <other random data follows, presumably>
  80         } */ *ap;
  81 {
  82         struct ucred **credpp = 0, *credp = 0;
  83         struct ucred *savecredp = 0, *savecompcredp = 0;
  84         struct ucred *compcredp = 0;
  85         struct vnode **this_vp_p;
  86         int error;
  87         struct vnode *old_vps[VDESC_MAX_VPS];
  88         struct vnode *vp1 = 0;
  89         struct vnode **vps_p[VDESC_MAX_VPS];
  90         struct vnode ***vppp;
  91         struct vnodeop_desc *descp = ap->a_desc;
  92         int reles, i;
  93         struct componentname **compnamepp = 0;
  94
  95         if (umap_bug_bypass)
  96                 printf ("umap_bypass: %s\n", descp->vdesc_name);
  97
  98 #ifdef DIAGNOSTIC
  99         /*
 100          * We require at least one vp.
 101          */
 102         if (descp->vdesc_vp_offsets == NULL ||
 103             descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
 104                 panic ("umap_bypass: no vp's in map");
 105 #endif
 106
 107         /*
 108          * Map the vnodes going in.
 109          * Later, we'll invoke the operation based on
 110          * the first mapped vnode's operation vector.
 111          */
 112         reles = descp->vdesc_flags;
 113         for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
 114                 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
 115                         break;   /* bail out at end of list */
 116                 vps_p[i] = this_vp_p =
 117                         VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i], ap);
 118
 119                 if (i == 0) {
 120                         vp1 = *vps_p[0];
 121                 }
 122
 123                 /*
 124                  * We're not guaranteed that any but the first vnode
 125                  * are of our type.  Check for and don't map any
 126                  * that aren't.  (Must map first vp or vclean fails.)
 127                  */
 128
 129                 if (i && (*this_vp_p)->v_op != umap_vnodeop_p) {
 130                         old_vps[i] = NULL;
 131                 } else {
 132                         old_vps[i] = *this_vp_p;
 133                         *(vps_p[i]) = UMAPVPTOLOWERVP(*this_vp_p);
 134                         if (reles & 1)
 135                                 vref(*this_vp_p);
 136                 }
 137
 138         }
 139
 140         /*
 141          * Fix the credentials.  (That's the purpose of this layer.)
 142          */
 143
 144         if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) {
 145
 146                 credpp = VOPARG_OFFSETTO(struct ucred**,
 147                     descp->vdesc_cred_offset, ap);
 148
 149                 /* Save old values */
 150
 151                 savecredp = (*credpp);
 152                 if (savecredp != NOCRED)
 153                         (*credpp) = crdup(savecredp);
 154                 credp = *credpp;
 155
 156                 if (umap_bug_bypass && credp->cr_uid != 0)
 157                         printf("umap_bypass: user was %lu, group %lu\n",
 158                             (u_long)credp->cr_uid, (u_long)credp->cr_gid);
 159
 160                 /* Map all ids in the credential structure. */
 161
 162                 umap_mapids(vp1->v_mount, credp);
 163
 164                 if (umap_bug_bypass && credp->cr_uid != 0)
 165                         printf("umap_bypass: user now %lu, group %lu\n",
 166                             (u_long)credp->cr_uid, (u_long)credp->cr_gid);
 167         }
 168
 169         /* BSD often keeps a credential in the componentname structure
 170          * for speed.  If there is one, it better get mapped, too.
 171          */
 172
 173         if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) {
 174
 175                 compnamepp = VOPARG_OFFSETTO(struct componentname**,
 176                     descp->vdesc_componentname_offset, ap);
 177
 178                 compcredp = (*compnamepp)->cn_cred;
 179                 savecompcredp = compcredp;
 180                 if (savecompcredp != NOCRED)
 181                         (*compnamepp)->cn_cred = crdup(savecompcredp);
 182                 compcredp = (*compnamepp)->cn_cred;
 183
 184                 if (umap_bug_bypass && compcredp->cr_uid != 0)
 185                         printf(
 186                     "umap_bypass: component credit user was %lu, group %lu\n",
 187                             (u_long)compcredp->cr_uid,
 188                             (u_long)compcredp->cr_gid);
 189
 190                 /* Map all ids in the credential structure. */
 191
 192                 umap_mapids(vp1->v_mount, compcredp);
 193
 194                 if (umap_bug_bypass && compcredp->cr_uid != 0)
 195                         printf(
 196                     "umap_bypass: component credit user now %lu, group %lu\n",
 197                             (u_long)compcredp->cr_uid,
 198                             (u_long)compcredp->cr_gid);
 199         }
 200
 201         /*
 202          * Call the operation on the lower layer
 203          * with the modified argument structure.
 204          */
 205         error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap);
 206
 207         /*
 208          * Maintain the illusion of call-by-value
 209          * by restoring vnodes in the argument structure
 210          * to their original value.
 211          */
 212         reles = descp->vdesc_flags;
 213         for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
 214                 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
 215                         break;   /* bail out at end of list */
 216                 if (old_vps[i]) {
 217                         *(vps_p[i]) = old_vps[i];
 218                         if (reles & 1)
 219                                 vrele(*(vps_p[i]));
 220                 };
 221         };
 222
 223         /*
 224          * Map the possible out-going vpp
 225          * (Assumes that the lower layer always returns
 226          * a vref'ed vpp unless it gets an error.)
 227          */
 228         if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
 229             !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
 230             !error) {
 231                 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
 232                         goto out;
 233                 vppp = VOPARG_OFFSETTO(struct vnode***,
 234                                  descp->vdesc_vpp_offset, ap);
 235                 if (*vppp)
 236                         error = umap_node_create(old_vps[0]->v_mount, **vppp, *vppp);
 237         };
 238
 239  out:
 240         /*
 241          * Free duplicate cred structure and restore old one.
 242          */
 243         if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) {
 244                 if (umap_bug_bypass && credp && credp->cr_uid != 0)
 245                         printf("umap_bypass: returning-user was %lu\n",
 246                             (u_long)credp->cr_uid);
 247
 248                 if (savecredp != NOCRED) {
 249                         crfree(credp);
 250                         (*credpp) = savecredp;
 251                         if (umap_bug_bypass && credpp && (*credpp)->cr_uid != 0)
 252                                 printf(
 253                                     "umap_bypass: returning-user now %lu\n\n",
 254                                     (u_long)(*credpp)->cr_uid);
 255                 }
 256         }
 257
 258         if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) {
 259                 if (umap_bug_bypass && compcredp && compcredp->cr_uid != 0)
 260                         printf(
 261                             "umap_bypass: returning-component-user was %lu\n",
 262                             (u_long)compcredp->cr_uid);
 263
 264                 if (savecompcredp != NOCRED) {
 265                         crfree(compcredp);
 266                         (*compnamepp)->cn_cred = savecompcredp;
 267                         if (umap_bug_bypass && credpp && (*credpp)->cr_uid != 0)
 268                                 printf(
 269                             "umap_bypass: returning-component-user now %lu\n",
 270                                     (u_long)compcredp->cr_uid);
 271                 }
 272         }
 273
 274         return (error);
 275 }
 276
 277
 278 /*
 279  *  We handle getattr to change the fsid.
 280  */
 281 static int
 282 umap_getattr(ap)
 283         struct vop_getattr_args /* {
 284                 struct vnode *a_vp;
 285                 struct vattr *a_vap;
 286                 struct ucred *a_cred;
 287                 struct thread *a_td;
 288         } */ *ap;
 289 {
 290         short uid, gid;
 291         int error, tmpid, nentries, gnentries;
 292         u_long (*mapdata)[2], (*gmapdata)[2];
 293         struct vnode **vp1p;
 294         struct vnodeop_desc *descp = ap->a_desc;
 295
 296         error = umap_bypass((struct vop_generic_args *)ap);
 297         if (error)
 298                 return (error);
 299
 300         /*
 301          * Umap needs to map the uid and gid returned by a stat
 302          * into the proper values for this site.  This involves
 303          * finding the returned uid in the mapping information,
 304          * translating it into the uid on the other end,
 305          * and filling in the proper field in the vattr
 306          * structure pointed to by ap->a_vap.  The group
 307          * is easier, since currently all groups will be
 308          * translate to the NULLGROUP.
 309          */
 310
 311         /* Find entry in map */
 312
 313         uid = ap->a_vap->va_uid;
 314         gid = ap->a_vap->va_gid;
 315         if (umap_bug_bypass)
 316                 printf("umap_getattr: mapped uid = %d, mapped gid = %d\n", uid,
 317                     gid);
 318
 319         vp1p = VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap);
 320         nentries =  MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_nentries;
 321         mapdata =  (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_mapdata);
 322         gnentries =  MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gnentries;
 323         gmapdata =  (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gmapdata);
 324
 325         /* Reverse map the uid for the vnode.  Since it's a reverse
 326                 map, we can't use umap_mapids() to do it. */
 327
 328         tmpid = umap_reverse_findid(uid, mapdata, nentries);
 329
 330         if (tmpid != -1) {
 331
 332                 ap->a_vap->va_uid = (uid_t) tmpid;
 333                 if (umap_bug_bypass)
 334                         printf("umap_getattr: original uid = %d\n", uid);
 335         } else
 336                 ap->a_vap->va_uid = (uid_t) NOBODY;
 337
 338         /* Reverse map the gid for the vnode. */
 339
 340         tmpid = umap_reverse_findid(gid, gmapdata, gnentries);
 341
 342         if (tmpid != -1) {
 343
 344                 ap->a_vap->va_gid = (gid_t) tmpid;
 345                 if (umap_bug_bypass)
 346                         printf("umap_getattr: original gid = %d\n", gid);
 347         } else
 348                 ap->a_vap->va_gid = (gid_t) NULLGROUP;
 349
 350         return (0);
 351 }
 352
 353 /*
 354  * We need to process our own vnode lock and then clear the
 355  * interlock flag as it applies only to our vnode, not the
 356  * vnodes below us on the stack.
 357  */
 358 static int
 359 umap_lock(ap)
 360         struct vop_lock_args /* {
 361                 struct vnode *a_vp;
 362                 lwkt_tokref_t a_vlock;
 363                 int a_flags;
 364                 struct thread *a_td;
 365         } */ *ap;
 366 {
 367
 368         vop_nolock(ap);
 369         if ((ap->a_flags & LK_TYPE_MASK) == LK_DRAIN)
 370                 return (0);
 371         ap->a_flags &= ~LK_INTERLOCK;
 372         return (null_bypass((struct vop_generic_args *)ap));
 373 }
 374
 375 /*
 376  * We need to process our own vnode unlock and then clear the
 377  * interlock flag as it applies only to our vnode, not the
 378  * vnodes below us on the stack.
 379  */
 380 int
 381 umap_unlock(ap)
 382         struct vop_unlock_args /* {
 383                 struct vnode *a_vp;
 384                 int a_flags;
 385                 struct thread *a_td;
 386         } */ *ap;
 387 {
 388         vop_nounlock(ap);
 389         ap->a_flags &= ~LK_INTERLOCK;
 390         return (null_bypass((struct vop_generic_args *)ap));
 391 }
 392
 393 static int
 394 umap_inactive(ap)
 395         struct vop_inactive_args /* {
 396                 struct vnode *a_vp;
 397                 struct thread *a_td;
 398         } */ *ap;
 399 {
 400         struct vnode *vp = ap->a_vp;
 401         struct umap_node *xp = VTOUMAP(vp);
 402         struct vnode *lowervp = xp->umap_lowervp;
 403         /*
 404          * Do nothing (and _don't_ bypass).
 405          * Wait to vrele lowervp until reclaim,
 406          * so that until then our umap_node is in the
 407          * cache and reusable.
 408          *
 409          */
 410         VOP_INACTIVE(lowervp, ap->a_td);
 411         VOP_UNLOCK(ap->a_vp, NULL, 0, ap->a_td);
 412         return (0);
 413 }
 414
 415 static int
 416 umap_reclaim(ap)
 417         struct vop_reclaim_args /* {
 418                 struct vnode *a_vp;
 419         } */ *ap;
 420 {
 421         struct vnode *vp = ap->a_vp;
 422         struct umap_node *xp = VTOUMAP(vp);
 423         struct vnode *lowervp = xp->umap_lowervp;
 424
 425         /* After this assignment, this node will not be re-used. */
 426         xp->umap_lowervp = NULL;
 427         LIST_REMOVE(xp, umap_hash);
 428         FREE(vp->v_data, M_TEMP);
 429         vp->v_data = NULL;
 430         vrele(lowervp);
 431         return (0);
 432 }
 433
 434 static int
 435 umap_print(ap)
 436         struct vop_print_args /* {
 437                 struct vnode *a_vp;
 438         } */ *ap;
 439 {
 440         struct vnode *vp = ap->a_vp;
 441         printf("\ttag VT_UMAPFS, vp=%p, lowervp=%p\n", vp, UMAPVPTOLOWERVP(vp));
 442         return (0);
 443 }
 444
 445 static int
 446 umap_rename(ap)
 447         struct vop_rename_args  /* {
 448                 struct vnode *a_fdvp;
 449                 struct vnode *a_fvp;
 450                 struct componentname *a_fcnp;
 451                 struct vnode *a_tdvp;
 452                 struct vnode *a_tvp;
 453                 struct componentname *a_tcnp;
 454         } */ *ap;
 455 {
 456         int error;
 457         struct componentname *compnamep;
 458         struct ucred *compcredp, *savecompcredp;
 459         struct vnode *vp;
 460
 461         /*
 462          * Rename is irregular, having two componentname structures.
 463          * We need to map the cre in the second structure,
 464          * and then bypass takes care of the rest.
 465          */
 466
 467         vp = ap->a_fdvp;
 468         compnamep = ap->a_tcnp;
 469         compcredp = compnamep->cn_cred;
 470
 471         savecompcredp = compcredp;
 472         compcredp = compnamep->cn_cred = crdup(savecompcredp);
 473
 474         if (umap_bug_bypass && compcredp->cr_uid != 0)
 475                 printf(
 476             "umap_rename: rename component credit user was %lu, group %lu\n",
 477                     (u_long)compcredp->cr_uid, (u_long)compcredp->cr_gid);
 478
 479         /* Map all ids in the credential structure. */
 480
 481         umap_mapids(vp->v_mount, compcredp);
 482
 483         if (umap_bug_bypass && compcredp->cr_uid != 0)
 484                 printf(
 485             "umap_rename: rename component credit user now %lu, group %lu\n",
 486                     (u_long)compcredp->cr_uid, (u_long)compcredp->cr_gid);
 487
 488         error = umap_bypass((struct vop_generic_args *)ap);
 489
 490         /* Restore the additional mapped componentname cred structure. */
 491
 492         crfree(compcredp);
 493         compnamep->cn_cred = savecompcredp;
 494
 495         return error;
 496 }
 497
 498 /*
 499  * Global vfs data structures
 500  */
 501 /*
 502  * XXX - strategy, bwrite are hand coded currently.  They should
 503  * go away with a merged buffer/block cache.
 504  *
 505  */
 506 vop_t **umap_vnodeop_p;
 507 static struct vnodeopv_entry_desc umap_vnodeop_entries[] = {
 508         { &vop_default_desc,            (vop_t *) umap_bypass },
 509         { &vop_getattr_desc,            (vop_t *) umap_getattr },
 510         { &vop_inactive_desc,           (vop_t *) umap_inactive },
 511         { &vop_lock_desc,               (vop_t *) umap_lock },
 512         { &vop_print_desc,              (vop_t *) umap_print },
 513         { &vop_reclaim_desc,            (vop_t *) umap_reclaim },
 514         { &vop_rename_desc,             (vop_t *) umap_rename },
 515         { &vop_unlock_desc,             (vop_t *) umap_unlock },
 516         { NULL, NULL }
 517 };
 518 static struct vnodeopv_desc umap_vnodeop_opv_desc =
 519         { &umap_vnodeop_p, umap_vnodeop_entries };
 520
 521 VNODEOP_SET(umap_vnodeop_opv_desc);