1 /* $NetBSD: umap_vnops.c,v 1.31 2004/06/16 17:59:53 wrstuden Exp $ */ 2 3 /* 4 * Copyright (c) 1992, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software donated to Berkeley by 8 * the UCLA Ficus project. 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. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)umap_vnops.c 8.6 (Berkeley) 5/22/95 35 */ 36 37 /* 38 * Umap Layer 39 */ 40 41 #include <sys/cdefs.h> 42 __KERNEL_RCSID(0, "$NetBSD: umap_vnops.c,v 1.31 2004/06/16 17:59:53 wrstuden Exp $"); 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/time.h> 47 #include <sys/vnode.h> 48 #include <sys/mount.h> 49 #include <sys/namei.h> 50 #include <sys/malloc.h> 51 #include <sys/buf.h> 52 #include <miscfs/umapfs/umap.h> 53 #include <miscfs/genfs/genfs.h> 54 #include <miscfs/genfs/layer_extern.h> 55 56 int umap_lookup __P((void *)); 57 int umap_getattr __P((void *)); 58 int umap_print __P((void *)); 59 int umap_rename __P((void *)); 60 61 /* 62 * Global vfs data structures 63 */ 64 /* 65 * XXX - strategy, bwrite are hand coded currently. They should 66 * go away with a merged buffer/block cache. 67 * 68 */ 69 int (**umap_vnodeop_p) __P((void *)); 70 const struct vnodeopv_entry_desc umap_vnodeop_entries[] = { 71 { &vop_default_desc, umap_bypass }, 72 73 { &vop_lookup_desc, umap_lookup }, 74 { &vop_getattr_desc, umap_getattr }, 75 { &vop_print_desc, umap_print }, 76 { &vop_rename_desc, umap_rename }, 77 78 { &vop_lock_desc, layer_lock }, 79 { &vop_unlock_desc, layer_unlock }, 80 { &vop_islocked_desc, layer_islocked }, 81 { &vop_fsync_desc, layer_fsync }, 82 { &vop_inactive_desc, layer_inactive }, 83 { &vop_reclaim_desc, layer_reclaim }, 84 { &vop_open_desc, layer_open }, 85 { &vop_setattr_desc, layer_setattr }, 86 { &vop_access_desc, layer_access }, 87 { &vop_remove_desc, layer_remove }, 88 89 { &vop_bwrite_desc, layer_bwrite }, 90 { &vop_bmap_desc, layer_bmap }, 91 { &vop_getpages_desc, layer_getpages }, 92 { &vop_putpages_desc, layer_putpages }, 93 94 { NULL, NULL } 95 }; 96 const struct vnodeopv_desc umapfs_vnodeop_opv_desc = 97 { &umap_vnodeop_p, umap_vnodeop_entries }; 98 99 /* 100 * This is the 08-June-1999 bypass routine. 101 * See layer_vnops.c:layer_bypass for more details. 102 */ 103 int 104 umap_bypass(v) 105 void *v; 106 { 107 struct vop_generic_args /* { 108 struct vnodeop_desc *a_desc; 109 <other random data follows, presumably> 110 } */ *ap = v; 111 int (**our_vnodeop_p) __P((void *)); 112 struct ucred **credpp = 0, *credp = 0; 113 struct ucred *savecredp = 0, *savecompcredp = 0; 114 struct ucred *compcredp = 0; 115 struct vnode **this_vp_p; 116 int error, error1; 117 struct vnode *old_vps[VDESC_MAX_VPS], *vp0; 118 struct vnode **vps_p[VDESC_MAX_VPS]; 119 struct vnode ***vppp; 120 struct vnodeop_desc *descp = ap->a_desc; 121 int reles, i, flags; 122 struct componentname **compnamepp = 0; 123 124 #ifdef SAFETY 125 /* 126 * We require at least one vp. 127 */ 128 if (descp->vdesc_vp_offsets == NULL || 129 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 130 panic("%s: no vp's in map.\n", __func__); 131 #endif 132 133 vps_p[0] = 134 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap); 135 vp0 = *vps_p[0]; 136 flags = MOUNTTOUMAPMOUNT(vp0->v_mount)->umapm_flags; 137 our_vnodeop_p = vp0->v_op; 138 139 if (flags & LAYERFS_MBYPASSDEBUG) 140 printf("%s: %s\n", __func__, descp->vdesc_name); 141 142 /* 143 * Map the vnodes going in. 144 * Later, we'll invoke the operation based on 145 * the first mapped vnode's operation vector. 146 */ 147 reles = descp->vdesc_flags; 148 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 149 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 150 break; /* bail out at end of list */ 151 vps_p[i] = this_vp_p = 152 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i], 153 ap); 154 /* 155 * We're not guaranteed that any but the first vnode 156 * are of our type. Check for and don't map any 157 * that aren't. (We must always map first vp or vclean fails.) 158 */ 159 if (i && (*this_vp_p == NULL || 160 (*this_vp_p)->v_op != our_vnodeop_p)) { 161 old_vps[i] = NULL; 162 } else { 163 old_vps[i] = *this_vp_p; 164 *(vps_p[i]) = UMAPVPTOLOWERVP(*this_vp_p); 165 /* 166 * XXX - Several operations have the side effect 167 * of vrele'ing their vp's. We must account for 168 * that. (This should go away in the future.) 169 */ 170 if (reles & VDESC_VP0_WILLRELE) 171 VREF(*this_vp_p); 172 } 173 174 } 175 176 /* 177 * Fix the credentials. (That's the purpose of this layer.) 178 */ 179 180 if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) { 181 182 credpp = VOPARG_OFFSETTO(struct ucred**, 183 descp->vdesc_cred_offset, ap); 184 185 /* Save old values */ 186 187 savecredp = *credpp; 188 if (savecredp != NOCRED) 189 *credpp = crdup(savecredp); 190 credp = *credpp; 191 192 if ((flags & LAYERFS_MBYPASSDEBUG) && credp->cr_uid != 0) 193 printf("umap_bypass: user was %d, group %d\n", 194 credp->cr_uid, credp->cr_gid); 195 196 /* Map all ids in the credential structure. */ 197 198 umap_mapids(vp0->v_mount, credp); 199 200 if ((flags & LAYERFS_MBYPASSDEBUG) && credp->cr_uid != 0) 201 printf("umap_bypass: user now %d, group %d\n", 202 credp->cr_uid, credp->cr_gid); 203 } 204 205 /* BSD often keeps a credential in the componentname structure 206 * for speed. If there is one, it better get mapped, too. 207 */ 208 209 if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) { 210 211 compnamepp = VOPARG_OFFSETTO(struct componentname**, 212 descp->vdesc_componentname_offset, ap); 213 214 savecompcredp = (*compnamepp)->cn_cred; 215 if (savecompcredp != NOCRED) 216 (*compnamepp)->cn_cred = crdup(savecompcredp); 217 compcredp = (*compnamepp)->cn_cred; 218 219 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 220 printf("umap_bypass: component credit user was %d, group %d\n", 221 compcredp->cr_uid, compcredp->cr_gid); 222 223 /* Map all ids in the credential structure. */ 224 225 umap_mapids(vp0->v_mount, compcredp); 226 227 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 228 printf("umap_bypass: component credit user now %d, group %d\n", 229 compcredp->cr_uid, compcredp->cr_gid); 230 } 231 232 /* 233 * Call the operation on the lower layer 234 * with the modified argument structure. 235 */ 236 error = VCALL(*vps_p[0], descp->vdesc_offset, ap); 237 238 /* 239 * Maintain the illusion of call-by-value 240 * by restoring vnodes in the argument structure 241 * to their original value. 242 */ 243 reles = descp->vdesc_flags; 244 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 245 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 246 break; /* bail out at end of list */ 247 if (old_vps[i]) { 248 *(vps_p[i]) = old_vps[i]; 249 if (reles & VDESC_VP0_WILLUNLOCK) 250 LAYERFS_UPPERUNLOCK(*(vps_p[i]), 0, error1); 251 if (reles & VDESC_VP0_WILLRELE) 252 vrele(*(vps_p[i])); 253 } 254 } 255 256 /* 257 * Map the possible out-going vpp 258 * (Assumes that the lower layer always returns 259 * a VREF'ed vpp unless it gets an error.) 260 */ 261 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 262 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 263 !error) { 264 /* 265 * XXX - even though some ops have vpp returned vp's, 266 * several ops actually vrele this before returning. 267 * We must avoid these ops. 268 * (This should go away when these ops are regularized.) 269 */ 270 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 271 goto out; 272 vppp = VOPARG_OFFSETTO(struct vnode***, 273 descp->vdesc_vpp_offset, ap); 274 /* 275 * Only vop_lookup, vop_create, vop_makedir, vop_bmap, 276 * vop_mknod, and vop_symlink return vpp's. vop_bmap 277 * doesn't call bypass as the lower vpp is fine (we're just 278 * going to do i/o on it). vop_lookup doesn't call bypass 279 * as a lookup on "." would generate a locking error. 280 * So all the calls which get us here have a locked vpp. :-) 281 */ 282 error = layer_node_create(old_vps[0]->v_mount, **vppp, *vppp); 283 if (error) { 284 vput(**vppp); 285 **vppp = NULL; 286 } 287 } 288 289 out: 290 /* 291 * Free duplicate cred structure and restore old one. 292 */ 293 if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) { 294 if ((flags & LAYERFS_MBYPASSDEBUG) && credp && 295 credp->cr_uid != 0) 296 printf("umap_bypass: returning-user was %d\n", 297 credp->cr_uid); 298 299 if (savecredp != NOCRED) { 300 crfree(credp); 301 *credpp = savecredp; 302 if ((flags & LAYERFS_MBYPASSDEBUG) && credpp && 303 (*credpp)->cr_uid != 0) 304 printf("umap_bypass: returning-user now %d\n\n", 305 savecredp->cr_uid); 306 } 307 } 308 309 if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) { 310 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp && 311 compcredp->cr_uid != 0) 312 printf("umap_bypass: returning-component-user was %d\n", 313 compcredp->cr_uid); 314 315 if (savecompcredp != NOCRED) { 316 crfree(compcredp); 317 (*compnamepp)->cn_cred = savecompcredp; 318 if ((flags & LAYERFS_MBYPASSDEBUG) && savecompcredp && 319 savecompcredp->cr_uid != 0) 320 printf("umap_bypass: returning-component-user now %d\n", 321 savecompcredp->cr_uid); 322 } 323 } 324 325 return (error); 326 } 327 328 /* 329 * This is based on the 08-June-1999 bypass routine. 330 * See layer_vnops.c:layer_bypass for more details. 331 */ 332 int 333 umap_lookup(v) 334 void *v; 335 { 336 struct vop_lookup_args /* { 337 struct vnodeop_desc *a_desc; 338 struct vnode * a_dvp; 339 struct vnode ** a_vpp; 340 struct componentname * a_cnp; 341 } */ *ap = v; 342 struct componentname *cnp = ap->a_cnp; 343 struct ucred *savecompcredp = NULL; 344 struct ucred *compcredp = NULL; 345 struct vnode *dvp, *vp, *ldvp; 346 struct mount *mp; 347 int error; 348 int i, flags, cnf = cnp->cn_flags; 349 350 dvp = ap->a_dvp; 351 mp = dvp->v_mount; 352 353 if ((cnf & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 354 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 355 return (EROFS); 356 357 flags = MOUNTTOUMAPMOUNT(mp)->umapm_flags; 358 ldvp = UMAPVPTOLOWERVP(dvp); 359 360 if (flags & LAYERFS_MBYPASSDEBUG) 361 printf("umap_lookup\n"); 362 363 /* 364 * Fix the credentials. (That's the purpose of this layer.) 365 * 366 * BSD often keeps a credential in the componentname structure 367 * for speed. If there is one, it better get mapped, too. 368 */ 369 370 if ((savecompcredp = cnp->cn_cred)) { 371 compcredp = crdup(savecompcredp); 372 cnp->cn_cred = compcredp; 373 374 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 375 printf("umap_lookup: component credit user was %d, group %d\n", 376 compcredp->cr_uid, compcredp->cr_gid); 377 378 /* Map all ids in the credential structure. */ 379 umap_mapids(mp, compcredp); 380 } 381 382 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 383 printf("umap_lookup: component credit user now %d, group %d\n", 384 compcredp->cr_uid, compcredp->cr_gid); 385 386 ap->a_dvp = ldvp; 387 error = VCALL(ldvp, ap->a_desc->vdesc_offset, ap); 388 vp = *ap->a_vpp; 389 *ap->a_vpp = NULL; 390 391 if (error == EJUSTRETURN && (cnf & ISLASTCN) && 392 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 393 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 394 error = EROFS; 395 396 /* Do locking fixup as appropriate. See layer_lookup() for info */ 397 if ((cnp->cn_flags & PDIRUNLOCK)) { 398 LAYERFS_UPPERUNLOCK(dvp, 0, i); 399 } 400 if (ldvp == vp) { 401 *ap->a_vpp = dvp; 402 VREF(dvp); 403 vrele(vp); 404 } else if (vp != NULL) { 405 error = layer_node_create(mp, vp, ap->a_vpp); 406 if (error) { 407 vput(vp); 408 if (cnp->cn_flags & PDIRUNLOCK) { 409 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 410 cnp->cn_flags &= ~PDIRUNLOCK; 411 } 412 } 413 } 414 415 /* 416 * Free duplicate cred structure and restore old one. 417 */ 418 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp && 419 compcredp->cr_uid != 0) 420 printf("umap_lookup: returning-component-user was %d\n", 421 compcredp->cr_uid); 422 423 if (savecompcredp != NOCRED) { 424 crfree(compcredp); 425 cnp->cn_cred = savecompcredp; 426 if ((flags & LAYERFS_MBYPASSDEBUG) && savecompcredp && 427 savecompcredp->cr_uid != 0) 428 printf("umap_lookup: returning-component-user now %d\n", 429 savecompcredp->cr_uid); 430 } 431 432 return (error); 433 } 434 435 /* 436 * We handle getattr to change the fsid. 437 */ 438 int 439 umap_getattr(v) 440 void *v; 441 { 442 struct vop_getattr_args /* { 443 struct vnode *a_vp; 444 struct vattr *a_vap; 445 struct ucred *a_cred; 446 struct proc *a_p; 447 } */ *ap = v; 448 uid_t uid; 449 gid_t gid; 450 int error, tmpid, nentries, gnentries, flags; 451 u_long (*mapdata)[2]; 452 u_long (*gmapdata)[2]; 453 struct vnode **vp1p; 454 const struct vnodeop_desc *descp = ap->a_desc; 455 456 if ((error = umap_bypass(ap)) != 0) 457 return (error); 458 /* Requires that arguments be restored. */ 459 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 460 461 flags = MOUNTTOUMAPMOUNT(ap->a_vp->v_mount)->umapm_flags; 462 /* 463 * Umap needs to map the uid and gid returned by a stat 464 * into the proper values for this site. This involves 465 * finding the returned uid in the mapping information, 466 * translating it into the uid on the other end, 467 * and filling in the proper field in the vattr 468 * structure pointed to by ap->a_vap. The group 469 * is easier, since currently all groups will be 470 * translate to the NULLGROUP. 471 */ 472 473 /* Find entry in map */ 474 475 uid = ap->a_vap->va_uid; 476 gid = ap->a_vap->va_gid; 477 if ((flags & LAYERFS_MBYPASSDEBUG)) 478 printf("umap_getattr: mapped uid = %d, mapped gid = %d\n", uid, 479 gid); 480 481 vp1p = VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap); 482 nentries = MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_nentries; 483 mapdata = (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_mapdata); 484 gnentries = MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gnentries; 485 gmapdata = (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gmapdata); 486 487 /* Reverse map the uid for the vnode. Since it's a reverse 488 map, we can't use umap_mapids() to do it. */ 489 490 tmpid = umap_reverse_findid(uid, mapdata, nentries); 491 492 if (tmpid != -1) { 493 ap->a_vap->va_uid = (uid_t) tmpid; 494 if ((flags & LAYERFS_MBYPASSDEBUG)) 495 printf("umap_getattr: original uid = %d\n", uid); 496 } else 497 ap->a_vap->va_uid = (uid_t) NOBODY; 498 499 /* Reverse map the gid for the vnode. */ 500 501 tmpid = umap_reverse_findid(gid, gmapdata, gnentries); 502 503 if (tmpid != -1) { 504 ap->a_vap->va_gid = (gid_t) tmpid; 505 if ((flags & LAYERFS_MBYPASSDEBUG)) 506 printf("umap_getattr: original gid = %d\n", gid); 507 } else 508 ap->a_vap->va_gid = (gid_t) NULLGROUP; 509 510 return (0); 511 } 512 513 int 514 umap_print(v) 515 void *v; 516 { 517 struct vop_print_args /* { 518 struct vnode *a_vp; 519 } */ *ap = v; 520 struct vnode *vp = ap->a_vp; 521 printf("\ttag VT_UMAPFS, vp=%p, lowervp=%p\n", vp, 522 UMAPVPTOLOWERVP(vp)); 523 return (0); 524 } 525 526 int 527 umap_rename(v) 528 void *v; 529 { 530 struct vop_rename_args /* { 531 struct vnode *a_fdvp; 532 struct vnode *a_fvp; 533 struct componentname *a_fcnp; 534 struct vnode *a_tdvp; 535 struct vnode *a_tvp; 536 struct componentname *a_tcnp; 537 } */ *ap = v; 538 int error, flags; 539 struct componentname *compnamep; 540 struct ucred *compcredp, *savecompcredp; 541 struct vnode *vp; 542 struct vnode *tvp; 543 544 /* 545 * Rename is irregular, having two componentname structures. 546 * We need to map the cre in the second structure, 547 * and then bypass takes care of the rest. 548 */ 549 550 vp = ap->a_fdvp; 551 flags = MOUNTTOUMAPMOUNT(vp->v_mount)->umapm_flags; 552 compnamep = ap->a_tcnp; 553 compcredp = compnamep->cn_cred; 554 555 savecompcredp = compcredp; 556 compcredp = compnamep->cn_cred = crdup(savecompcredp); 557 558 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 559 printf("umap_rename: rename component credit user was %d, group %d\n", 560 compcredp->cr_uid, compcredp->cr_gid); 561 562 /* Map all ids in the credential structure. */ 563 564 umap_mapids(vp->v_mount, compcredp); 565 566 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 567 printf("umap_rename: rename component credit user now %d, group %d\n", 568 compcredp->cr_uid, compcredp->cr_gid); 569 570 tvp = ap->a_tvp; 571 if (tvp) { 572 if (tvp->v_mount != vp->v_mount) 573 tvp = NULL; 574 else 575 vref(tvp); 576 } 577 error = umap_bypass(ap); 578 if (tvp) { 579 if (error == 0) 580 VTOLAYER(tvp)->layer_flags |= LAYERFS_REMOVED; 581 vrele(tvp); 582 } 583 584 /* Restore the additional mapped componentname cred structure. */ 585 586 crfree(compcredp); 587 compnamep->cn_cred = savecompcredp; 588 589 return error; 590 } 591