1 /* $NetBSD: umap_vnops.c,v 1.38 2006/04/12 01:09:43 christos 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 1.38 2006/04/12 01:09:43 christos 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(void *); 57 int umap_getattr(void *); 58 int umap_print(void *); 59 int umap_rename(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)(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 { &vop_rmdir_desc, layer_rmdir }, 89 90 { &vop_bwrite_desc, layer_bwrite }, 91 { &vop_bmap_desc, layer_bmap }, 92 { &vop_getpages_desc, layer_getpages }, 93 { &vop_putpages_desc, layer_putpages }, 94 95 { NULL, NULL } 96 }; 97 const struct vnodeopv_desc umapfs_vnodeop_opv_desc = 98 { &umap_vnodeop_p, umap_vnodeop_entries }; 99 100 /* 101 * This is the 08-June-1999 bypass routine. 102 * See layer_vnops.c:layer_bypass for more details. 103 */ 104 int 105 umap_bypass(v) 106 void *v; 107 { 108 struct vop_generic_args /* { 109 struct vnodeop_desc *a_desc; 110 <other random data follows, presumably> 111 } */ *ap = v; 112 int (**our_vnodeop_p)(void *); 113 struct ucred **credpp = 0, *credp = 0; 114 struct ucred *savecredp = 0, *savecompcredp = 0; 115 struct ucred *compcredp = 0; 116 struct vnode **this_vp_p; 117 int error, error1; 118 struct vnode *old_vps[VDESC_MAX_VPS], *vp0; 119 struct vnode **vps_p[VDESC_MAX_VPS]; 120 struct vnode ***vppp; 121 struct vnodeop_desc *descp = ap->a_desc; 122 int reles, i, flags; 123 struct componentname **compnamepp = 0; 124 125 #ifdef SAFETY 126 /* 127 * We require at least one vp. 128 */ 129 if (descp->vdesc_vp_offsets == NULL || 130 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 131 panic("%s: no vp's in map.\n", __func__); 132 #endif 133 134 vps_p[0] = 135 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap); 136 vp0 = *vps_p[0]; 137 flags = MOUNTTOUMAPMOUNT(vp0->v_mount)->umapm_flags; 138 our_vnodeop_p = vp0->v_op; 139 140 if (flags & LAYERFS_MBYPASSDEBUG) 141 printf("%s: %s\n", __func__, descp->vdesc_name); 142 143 /* 144 * Map the vnodes going in. 145 * Later, we'll invoke the operation based on 146 * the first mapped vnode's operation vector. 147 */ 148 reles = descp->vdesc_flags; 149 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 150 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 151 break; /* bail out at end of list */ 152 vps_p[i] = this_vp_p = 153 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i], 154 ap); 155 /* 156 * We're not guaranteed that any but the first vnode 157 * are of our type. Check for and don't map any 158 * that aren't. (We must always map first vp or vclean fails.) 159 */ 160 if (i && (*this_vp_p == NULL || 161 (*this_vp_p)->v_op != our_vnodeop_p)) { 162 old_vps[i] = NULL; 163 } else { 164 old_vps[i] = *this_vp_p; 165 *(vps_p[i]) = UMAPVPTOLOWERVP(*this_vp_p); 166 /* 167 * XXX - Several operations have the side effect 168 * of vrele'ing their vp's. We must account for 169 * that. (This should go away in the future.) 170 */ 171 if (reles & VDESC_VP0_WILLRELE) 172 VREF(*this_vp_p); 173 } 174 175 } 176 177 /* 178 * Fix the credentials. (That's the purpose of this layer.) 179 */ 180 181 if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) { 182 183 credpp = VOPARG_OFFSETTO(struct ucred**, 184 descp->vdesc_cred_offset, ap); 185 186 /* Save old values */ 187 188 savecredp = *credpp; 189 if (savecredp != NOCRED) 190 *credpp = crdup(savecredp); 191 credp = *credpp; 192 193 if ((flags & LAYERFS_MBYPASSDEBUG) && credp->cr_uid != 0) 194 printf("umap_bypass: user was %d, group %d\n", 195 credp->cr_uid, credp->cr_gid); 196 197 /* Map all ids in the credential structure. */ 198 199 umap_mapids(vp0->v_mount, credp); 200 201 if ((flags & LAYERFS_MBYPASSDEBUG) && credp->cr_uid != 0) 202 printf("umap_bypass: user now %d, group %d\n", 203 credp->cr_uid, credp->cr_gid); 204 } 205 206 /* BSD often keeps a credential in the componentname structure 207 * for speed. If there is one, it better get mapped, too. 208 */ 209 210 if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) { 211 212 compnamepp = VOPARG_OFFSETTO(struct componentname**, 213 descp->vdesc_componentname_offset, ap); 214 215 savecompcredp = (*compnamepp)->cn_cred; 216 if (savecompcredp != NOCRED) 217 (*compnamepp)->cn_cred = crdup(savecompcredp); 218 compcredp = (*compnamepp)->cn_cred; 219 220 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 221 printf("umap_bypass: component credit user was %d, group %d\n", 222 compcredp->cr_uid, compcredp->cr_gid); 223 224 /* Map all ids in the credential structure. */ 225 226 umap_mapids(vp0->v_mount, compcredp); 227 228 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 229 printf("umap_bypass: component credit user now %d, group %d\n", 230 compcredp->cr_uid, compcredp->cr_gid); 231 } 232 233 /* 234 * Call the operation on the lower layer 235 * with the modified argument structure. 236 */ 237 error = VCALL(*vps_p[0], descp->vdesc_offset, ap); 238 239 /* 240 * Maintain the illusion of call-by-value 241 * by restoring vnodes in the argument structure 242 * to their original value. 243 */ 244 reles = descp->vdesc_flags; 245 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 246 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 247 break; /* bail out at end of list */ 248 if (old_vps[i]) { 249 *(vps_p[i]) = old_vps[i]; 250 if (reles & VDESC_VP0_WILLUNLOCK) 251 LAYERFS_UPPERUNLOCK(*(vps_p[i]), 0, error1); 252 if (reles & VDESC_VP0_WILLRELE) 253 vrele(*(vps_p[i])); 254 } 255 } 256 257 /* 258 * Map the possible out-going vpp 259 * (Assumes that the lower layer always returns 260 * a VREF'ed vpp unless it gets an error.) 261 */ 262 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 263 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 264 !error) { 265 /* 266 * XXX - even though some ops have vpp returned vp's, 267 * several ops actually vrele this before returning. 268 * We must avoid these ops. 269 * (This should go away when these ops are regularized.) 270 */ 271 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 272 goto out; 273 vppp = VOPARG_OFFSETTO(struct vnode***, 274 descp->vdesc_vpp_offset, ap); 275 /* 276 * Only vop_lookup, vop_create, vop_makedir, vop_bmap, 277 * vop_mknod, and vop_symlink return vpp's. vop_bmap 278 * doesn't call bypass as the lower vpp is fine (we're just 279 * going to do i/o on it). vop_lookup doesn't call bypass 280 * as a lookup on "." would generate a locking error. 281 * So all the calls which get us here have a locked vpp. :-) 282 */ 283 error = layer_node_create(old_vps[0]->v_mount, **vppp, *vppp); 284 if (error) { 285 vput(**vppp); 286 **vppp = NULL; 287 } 288 } 289 290 out: 291 /* 292 * Free duplicate cred structure and restore old one. 293 */ 294 if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) { 295 if ((flags & LAYERFS_MBYPASSDEBUG) && credp && 296 credp->cr_uid != 0) 297 printf("umap_bypass: returning-user was %d\n", 298 credp->cr_uid); 299 300 if (savecredp != NOCRED && credpp) { 301 crfree(credp); 302 *credpp = savecredp; 303 if ((flags & LAYERFS_MBYPASSDEBUG) && credpp && 304 (*credpp)->cr_uid != 0) 305 printf("umap_bypass: returning-user now %d\n\n", 306 savecredp->cr_uid); 307 } 308 } 309 310 if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) { 311 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp && 312 compcredp->cr_uid != 0) 313 printf("umap_bypass: returning-component-user was %d\n", 314 compcredp->cr_uid); 315 316 if (savecompcredp != NOCRED) { 317 crfree(compcredp); 318 (*compnamepp)->cn_cred = savecompcredp; 319 if ((flags & LAYERFS_MBYPASSDEBUG) && savecompcredp && 320 savecompcredp->cr_uid != 0) 321 printf("umap_bypass: returning-component-user now %d\n", 322 savecompcredp->cr_uid); 323 } 324 } 325 326 return (error); 327 } 328 329 /* 330 * This is based on the 08-June-1999 bypass routine. 331 * See layer_vnops.c:layer_bypass for more details. 332 */ 333 int 334 umap_lookup(v) 335 void *v; 336 { 337 struct vop_lookup_args /* { 338 struct vnodeop_desc *a_desc; 339 struct vnode * a_dvp; 340 struct vnode ** a_vpp; 341 struct componentname * a_cnp; 342 } */ *ap = v; 343 struct componentname *cnp = ap->a_cnp; 344 struct ucred *savecompcredp = NULL; 345 struct ucred *compcredp = NULL; 346 struct vnode *dvp, *vp, *ldvp; 347 struct mount *mp; 348 int error; 349 int i, flags, cnf = cnp->cn_flags; 350 351 dvp = ap->a_dvp; 352 mp = dvp->v_mount; 353 354 if ((cnf & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 355 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 356 return (EROFS); 357 358 flags = MOUNTTOUMAPMOUNT(mp)->umapm_flags; 359 ldvp = UMAPVPTOLOWERVP(dvp); 360 361 if (flags & LAYERFS_MBYPASSDEBUG) 362 printf("umap_lookup\n"); 363 364 /* 365 * Fix the credentials. (That's the purpose of this layer.) 366 * 367 * BSD often keeps a credential in the componentname structure 368 * for speed. If there is one, it better get mapped, too. 369 */ 370 371 if ((savecompcredp = cnp->cn_cred)) { 372 compcredp = crdup(savecompcredp); 373 cnp->cn_cred = compcredp; 374 375 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 376 printf("umap_lookup: component credit user was %d, group %d\n", 377 compcredp->cr_uid, compcredp->cr_gid); 378 379 /* Map all ids in the credential structure. */ 380 umap_mapids(mp, compcredp); 381 } 382 383 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp && 384 compcredp->cr_uid != 0) 385 printf("umap_lookup: component credit user now %d, group %d\n", 386 compcredp->cr_uid, compcredp->cr_gid); 387 388 ap->a_dvp = ldvp; 389 error = VCALL(ldvp, ap->a_desc->vdesc_offset, ap); 390 vp = *ap->a_vpp; 391 *ap->a_vpp = NULL; 392 393 if (error == EJUSTRETURN && (cnf & ISLASTCN) && 394 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 395 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 396 error = EROFS; 397 398 /* Do locking fixup as appropriate. See layer_lookup() for info */ 399 if ((cnp->cn_flags & PDIRUNLOCK)) { 400 LAYERFS_UPPERUNLOCK(dvp, 0, i); 401 } 402 if (ldvp == vp) { 403 *ap->a_vpp = dvp; 404 VREF(dvp); 405 vrele(vp); 406 } else if (vp != NULL) { 407 error = layer_node_create(mp, vp, ap->a_vpp); 408 if (error) { 409 vput(vp); 410 if (cnp->cn_flags & PDIRUNLOCK) { 411 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0) 412 cnp->cn_flags &= ~PDIRUNLOCK; 413 } 414 } 415 } 416 417 /* 418 * Free duplicate cred structure and restore old one. 419 */ 420 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp && 421 compcredp->cr_uid != 0) 422 printf("umap_lookup: returning-component-user was %d\n", 423 compcredp->cr_uid); 424 425 if (savecompcredp != NOCRED) { 426 if (compcredp) 427 crfree(compcredp); 428 cnp->cn_cred = savecompcredp; 429 if ((flags & LAYERFS_MBYPASSDEBUG) && savecompcredp && 430 savecompcredp->cr_uid != 0) 431 printf("umap_lookup: returning-component-user now %d\n", 432 savecompcredp->cr_uid); 433 } 434 435 return (error); 436 } 437 438 /* 439 * We handle getattr to change the fsid. 440 */ 441 int 442 umap_getattr(v) 443 void *v; 444 { 445 struct vop_getattr_args /* { 446 struct vnode *a_vp; 447 struct vattr *a_vap; 448 struct ucred *a_cred; 449 struct lwp *a_l; 450 } */ *ap = v; 451 uid_t uid; 452 gid_t gid; 453 int error, tmpid, nentries, gnentries, flags; 454 u_long (*mapdata)[2]; 455 u_long (*gmapdata)[2]; 456 struct vnode **vp1p; 457 const struct vnodeop_desc *descp = ap->a_desc; 458 459 if ((error = umap_bypass(ap)) != 0) 460 return (error); 461 /* Requires that arguments be restored. */ 462 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 463 464 flags = MOUNTTOUMAPMOUNT(ap->a_vp->v_mount)->umapm_flags; 465 /* 466 * Umap needs to map the uid and gid returned by a stat 467 * into the proper values for this site. This involves 468 * finding the returned uid in the mapping information, 469 * translating it into the uid on the other end, 470 * and filling in the proper field in the vattr 471 * structure pointed to by ap->a_vap. The group 472 * is easier, since currently all groups will be 473 * translate to the NULLGROUP. 474 */ 475 476 /* Find entry in map */ 477 478 uid = ap->a_vap->va_uid; 479 gid = ap->a_vap->va_gid; 480 if ((flags & LAYERFS_MBYPASSDEBUG)) 481 printf("umap_getattr: mapped uid = %d, mapped gid = %d\n", uid, 482 gid); 483 484 vp1p = VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap); 485 nentries = MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_nentries; 486 mapdata = (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_mapdata); 487 gnentries = MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gnentries; 488 gmapdata = (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gmapdata); 489 490 /* Reverse map the uid for the vnode. Since it's a reverse 491 map, we can't use umap_mapids() to do it. */ 492 493 tmpid = umap_reverse_findid(uid, mapdata, nentries); 494 495 if (tmpid != -1) { 496 ap->a_vap->va_uid = (uid_t) tmpid; 497 if ((flags & LAYERFS_MBYPASSDEBUG)) 498 printf("umap_getattr: original uid = %d\n", uid); 499 } else 500 ap->a_vap->va_uid = (uid_t) NOBODY; 501 502 /* Reverse map the gid for the vnode. */ 503 504 tmpid = umap_reverse_findid(gid, gmapdata, gnentries); 505 506 if (tmpid != -1) { 507 ap->a_vap->va_gid = (gid_t) tmpid; 508 if ((flags & LAYERFS_MBYPASSDEBUG)) 509 printf("umap_getattr: original gid = %d\n", gid); 510 } else 511 ap->a_vap->va_gid = (gid_t) NULLGROUP; 512 513 return (0); 514 } 515 516 int 517 umap_print(v) 518 void *v; 519 { 520 struct vop_print_args /* { 521 struct vnode *a_vp; 522 } */ *ap = v; 523 struct vnode *vp = ap->a_vp; 524 printf("\ttag VT_UMAPFS, vp=%p, lowervp=%p\n", vp, 525 UMAPVPTOLOWERVP(vp)); 526 return (0); 527 } 528 529 int 530 umap_rename(v) 531 void *v; 532 { 533 struct vop_rename_args /* { 534 struct vnode *a_fdvp; 535 struct vnode *a_fvp; 536 struct componentname *a_fcnp; 537 struct vnode *a_tdvp; 538 struct vnode *a_tvp; 539 struct componentname *a_tcnp; 540 } */ *ap = v; 541 int error, flags; 542 struct componentname *compnamep; 543 struct ucred *compcredp, *savecompcredp; 544 struct vnode *vp; 545 struct vnode *tvp; 546 547 /* 548 * Rename is irregular, having two componentname structures. 549 * We need to map the cre in the second structure, 550 * and then bypass takes care of the rest. 551 */ 552 553 vp = ap->a_fdvp; 554 flags = MOUNTTOUMAPMOUNT(vp->v_mount)->umapm_flags; 555 compnamep = ap->a_tcnp; 556 compcredp = compnamep->cn_cred; 557 558 savecompcredp = compcredp; 559 compcredp = compnamep->cn_cred = crdup(savecompcredp); 560 561 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 562 printf("umap_rename: rename component credit user was %d, group %d\n", 563 compcredp->cr_uid, compcredp->cr_gid); 564 565 /* Map all ids in the credential structure. */ 566 567 umap_mapids(vp->v_mount, compcredp); 568 569 if ((flags & LAYERFS_MBYPASSDEBUG) && compcredp->cr_uid != 0) 570 printf("umap_rename: rename component credit user now %d, group %d\n", 571 compcredp->cr_uid, compcredp->cr_gid); 572 573 tvp = ap->a_tvp; 574 if (tvp) { 575 if (tvp->v_mount != vp->v_mount) 576 tvp = NULL; 577 else 578 vref(tvp); 579 } 580 error = umap_bypass(ap); 581 if (tvp) { 582 if (error == 0) 583 VTOLAYER(tvp)->layer_flags |= LAYERFS_REMOVED; 584 vrele(tvp); 585 } 586 587 /* Restore the additional mapped componentname cred structure. */ 588 589 crfree(compcredp); 590 compnamep->cn_cred = savecompcredp; 591 592 return error; 593 } 594