sys/fs/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  * 4. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  *
  32  *      @(#)umap_vnops.c        8.6 (Berkeley) 5/22/95
  33  * $FreeBSD$
  34  */
  35
  36 /*
  37  * Umap Layer
  38  */
  39
  40 #include <sys/param.h>
  41 #include <sys/systm.h>
  42 #include <sys/kernel.h>
  43 #include <sys/lock.h>
  44 #include <sys/malloc.h>
  45 #include <sys/mount.h>
  46 #include <sys/namei.h>
  47 #include <sys/sysctl.h>
  48 #include <sys/vnode.h>
  49
  50 #include <fs/umapfs/umap.h>
  51 #include <fs/nullfs/null.h>
  52
  53 static int umap_bug_bypass = 0;   /* for debugging: enables bypass printf'ing */
  54 SYSCTL_INT(_debug, OID_AUTO, umapfs_bug_bypass, CTLFLAG_RW,
  55         &umap_bug_bypass, 0, "");
  56
  57 static vop_generic_t    umap_bypass;
  58 static vop_getattr_t    umap_getattr;
  59 static vop_inactive_t   umap_inactive;
  60 static vop_lock_t       umap_lock;
  61 static vop_print_t      umap_print;
  62 static vop_reclaim_t    umap_reclaim;
  63 static vop_rename_t     umap_rename;
  64 static vop_unlock_t     umap_unlock;
  65
  66 /*
  67  * This is the 10-Apr-92 bypass routine.
  68  * See null_vnops.c:null_bypass for more details.
  69  */
  70 static int
  71 umap_bypass(ap)
  72         struct vop_generic_args /* {
  73                 struct vnodeop_desc *a_desc;
  74                 <other random data follows, presumably>
  75         } */ *ap;
  76 {
  77         struct ucred **credpp = 0, *credp = 0;
  78         struct ucred *savecredp = 0, *savecompcredp = 0;
  79         struct ucred *compcredp = 0;
  80         struct vnode **this_vp_p;
  81         int error;
  82         struct vnode *old_vps[VDESC_MAX_VPS];
  83         struct vnode *vp1 = 0;
  84         struct vnode **vps_p[VDESC_MAX_VPS];
  85         struct vnode ***vppp;
  86         struct vnodeop_desc *descp = ap->a_desc;
  87         int reles, i;
  88         struct componentname **compnamepp = 0;
  89
  90         if (umap_bug_bypass)
  91                 printf ("umap_bypass: %s\n", descp->vdesc_name);
  92
  93 #ifdef DIAGNOSTIC
  94         /*
  95          * We require at least one vp.
  96          */
  97         if (descp->vdesc_vp_offsets == NULL ||
  98             descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
  99                 panic ("umap_bypass: no vp's in map");
 100 #endif
 101
 102         /*
 103          * Map the vnodes going in.
 104          * Later, we'll invoke the operation based on
 105          * the first mapped vnode's operation vector.
 106          */
 107         reles = descp->vdesc_flags;
 108         for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
 109                 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
 110                         break;   /* bail out at end of list */
 111                 vps_p[i] = this_vp_p =
 112                         VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i], ap);
 113
 114                 if (i == 0) {
 115                         vp1 = *vps_p[0];
 116                 }
 117
 118                 /*
 119                  * We're not guaranteed that any but the first vnode
 120                  * are of our type.  Check for and don't map any
 121                  * that aren't.  (Must map first vp or vclean fails.)
 122                  */
 123
 124                 if (i && (*this_vp_p)->v_op != umap_vnodeop_p) {
 125                         old_vps[i] = NULL;
 126                 } else {
 127                         old_vps[i] = *this_vp_p;
 128                         *(vps_p[i]) = UMAPVPTOLOWERVP(*this_vp_p);
 129                         if (reles & 1)
 130                                 VREF(*this_vp_p);
 131                 }
 132
 133         }
 134
 135         /*
 136          * Fix the credentials.  (That's the purpose of this layer.)
 137          */
 138
 139         if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) {
 140
 141                 credpp = VOPARG_OFFSETTO(struct ucred**,
 142                     descp->vdesc_cred_offset, ap);
 143
 144                 /* Save old values */
 145
 146                 savecredp = (*credpp);
 147                 if (savecredp != NOCRED)
 148                         (*credpp) = crdup(savecredp);
 149                 credp = *credpp;
 150
 151                 if (umap_bug_bypass && credp->cr_uid != 0)
 152                         printf("umap_bypass: user was %lu, group %lu\n",
 153                             (u_long)credp->cr_uid, (u_long)credp->cr_gid);
 154
 155                 /* Map all ids in the credential structure. */
 156
 157                 umap_mapids(vp1->v_mount, credp);
 158
 159                 if (umap_bug_bypass && credp->cr_uid != 0)
 160                         printf("umap_bypass: user now %lu, group %lu\n",
 161                             (u_long)credp->cr_uid, (u_long)credp->cr_gid);
 162         }
 163
 164         /* BSD often keeps a credential in the componentname structure
 165          * for speed.  If there is one, it better get mapped, too.
 166          */
 167
 168         if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) {
 169
 170                 compnamepp = VOPARG_OFFSETTO(struct componentname**,
 171                     descp->vdesc_componentname_offset, ap);
 172
 173                 compcredp = (*compnamepp)->cn_cred;
 174                 savecompcredp = compcredp;
 175                 if (savecompcredp != NOCRED)
 176                         (*compnamepp)->cn_cred = crdup(savecompcredp);
 177                 compcredp = (*compnamepp)->cn_cred;
 178
 179                 if (umap_bug_bypass && compcredp->cr_uid != 0)
 180                         printf(
 181                     "umap_bypass: component credit user was %lu, group %lu\n",
 182                             (u_long)compcredp->cr_uid,
 183                             (u_long)compcredp->cr_gid);
 184
 185                 /* Map all ids in the credential structure. */
 186
 187                 umap_mapids(vp1->v_mount, compcredp);
 188
 189                 if (umap_bug_bypass && compcredp->cr_uid != 0)
 190                         printf(
 191                     "umap_bypass: component credit user now %lu, group %lu\n",
 192                             (u_long)compcredp->cr_uid,
 193                             (u_long)compcredp->cr_gid);
 194         }
 195
 196         /*
 197          * Call the operation on the lower layer
 198          * with the modified argument structure.
 199          */
 200         error = VCALL(ap);
 201
 202         /*
 203          * Maintain the illusion of call-by-value
 204          * by restoring vnodes in the argument structure
 205          * to their original value.
 206          */
 207         reles = descp->vdesc_flags;
 208         for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
 209                 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
 210                         break;   /* bail out at end of list */
 211                 if (old_vps[i]) {
 212                         *(vps_p[i]) = old_vps[i];
 213                         if (reles & 1)
 214                                 vrele(*(vps_p[i]));
 215                 };
 216         };
 217
 218         /*
 219          * Map the possible out-going vpp
 220          * (Assumes that the lower layer always returns
 221          * a VREF'ed vpp unless it gets an error.)
 222          */
 223         if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
 224             !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
 225             !error) {
 226                 if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
 227                         goto out;
 228                 vppp = VOPARG_OFFSETTO(struct vnode***,
 229                                  descp->vdesc_vpp_offset, ap);
 230                 if (*vppp)
 231                         error = umap_node_create(old_vps[0]->v_mount, **vppp, *vppp);
 232         };
 233
 234  out:
 235         /*
 236          * Free duplicate cred structure and restore old one.
 237          */
 238         if (descp->vdesc_cred_offset != VDESC_NO_OFFSET) {
 239                 if (umap_bug_bypass && credp && credp->cr_uid != 0)
 240                         printf("umap_bypass: returning-user was %lu\n",
 241                             (u_long)credp->cr_uid);
 242
 243                 if (savecredp != NOCRED) {
 244                         crfree(credp);
 245                         (*credpp) = savecredp;
 246                         if (umap_bug_bypass && credpp && (*credpp)->cr_uid != 0)
 247                                 printf(
 248                                     "umap_bypass: returning-user now %lu\n\n",
 249                                     (u_long)(*credpp)->cr_uid);
 250                 }
 251         }
 252
 253         if (descp->vdesc_componentname_offset != VDESC_NO_OFFSET) {
 254                 if (umap_bug_bypass && compcredp && compcredp->cr_uid != 0)
 255                         printf(
 256                             "umap_bypass: returning-component-user was %lu\n",
 257                             (u_long)compcredp->cr_uid);
 258
 259                 if (savecompcredp != NOCRED) {
 260                         crfree(compcredp);
 261                         (*compnamepp)->cn_cred = savecompcredp;
 262                         if (umap_bug_bypass && credpp && (*credpp)->cr_uid != 0)
 263                                 printf(
 264                             "umap_bypass: returning-component-user now %lu\n",
 265                                     (u_long)compcredp->cr_uid);
 266                 }
 267         }
 268
 269         return (error);
 270 }
 271
 272
 273 /*
 274  *  We handle getattr to change the fsid.
 275  */
 276 static int
 277 umap_getattr(ap)
 278         struct vop_getattr_args /* {
 279                 struct vnode *a_vp;
 280                 struct vattr *a_vap;
 281                 struct ucred *a_cred;
 282                 struct thread *a_td;
 283         } */ *ap;
 284 {
 285         short uid, gid;
 286         int error, tmpid, nentries, gnentries;
 287         u_long (*mapdata)[2], (*gmapdata)[2];
 288         struct vnode **vp1p;
 289         struct vnodeop_desc *descp = ap->a_desc;
 290
 291         error = umap_bypass((struct vop_generic_args *)ap);
 292         if (error)
 293                 return (error);
 294
 295         /*
 296          * Umap needs to map the uid and gid returned by a stat
 297          * into the proper values for this site.  This involves
 298          * finding the returned uid in the mapping information,
 299          * translating it into the uid on the other end,
 300          * and filling in the proper field in the vattr
 301          * structure pointed to by ap->a_vap.  The group
 302          * is easier, since currently all groups will be
 303          * translate to the NULLGROUP.
 304          */
 305
 306         /* Find entry in map */
 307
 308         uid = ap->a_vap->va_uid;
 309         gid = ap->a_vap->va_gid;
 310         if (umap_bug_bypass)
 311                 printf("umap_getattr: mapped uid = %d, mapped gid = %d\n", uid,
 312                     gid);
 313
 314         vp1p = VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap);
 315         nentries =  MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_nentries;
 316         mapdata =  (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_mapdata);
 317         gnentries =  MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gnentries;
 318         gmapdata =  (MOUNTTOUMAPMOUNT((*vp1p)->v_mount)->info_gmapdata);
 319
 320         /* Reverse map the uid for the vnode.  Since it's a reverse
 321                 map, we can't use umap_mapids() to do it. */
 322
 323         tmpid = umap_reverse_findid(uid, mapdata, nentries);
 324
 325         if (tmpid != -1) {
 326
 327                 ap->a_vap->va_uid = (uid_t) tmpid;
 328                 if (umap_bug_bypass)
 329                         printf("umap_getattr: original uid = %d\n", uid);
 330         } else
 331                 ap->a_vap->va_uid = (uid_t) NOBODY;
 332
 333         /* Reverse map the gid for the vnode. */
 334
 335         tmpid = umap_reverse_findid(gid, gmapdata, gnentries);
 336
 337         if (tmpid != -1) {
 338
 339                 ap->a_vap->va_gid = (gid_t) tmpid;
 340                 if (umap_bug_bypass)
 341                         printf("umap_getattr: original gid = %d\n", gid);
 342         } else
 343                 ap->a_vap->va_gid = (gid_t) NULLGROUP;
 344
 345         return (0);
 346 }
 347
 348 /*
 349  * We need to process our own vnode lock and then clear the
 350  * interlock flag as it applies only to our vnode, not the
 351  * vnodes below us on the stack.
 352  */
 353 static int
 354 umap_lock(ap)
 355         struct vop_lock_args /* {
 356                 struct vnode *a_vp;
 357                 int a_flags;
 358                 struct thread *a_td;
 359         } */ *ap;
 360 {
 361
 362         /*
 363          * vop_nolock no longer exists.  I could have pasted the code
 364          * in so that it compiles, but that would be doing our users a
 365          * great disservice.  umapfs is about 5 years behind the nullfs
 366          * code that it is derived from.  The stub locking here guarantees
 367          * a deadlock the moment a VOP_INACTIVE arrives.  There is no point
 368          * pasting the code that makes it compile either, because that just
 369          * makes it Even More Wrong.
 370          */
 371         vop_nolock(ap);
 372         if ((ap->a_flags & LK_TYPE_MASK) == LK_DRAIN)
 373                 return (0);
 374         ap->a_flags &= ~LK_INTERLOCK;
 375         return (null_bypass((struct vop_generic_args *)ap));
 376 }
 377
 378 /*
 379  * We need to process our own vnode unlock and then clear the
 380  * interlock flag as it applies only to our vnode, not the
 381  * vnodes below us on the stack.
 382  */
 383 static int
 384 umap_unlock(ap)
 385         struct vop_unlock_args /* {
 386                 struct vnode *a_vp;
 387                 int a_flags;
 388                 struct thread *a_td;
 389         } */ *ap;
 390 {
 391         vop_nounlock(ap);
 392         ap->a_flags &= ~LK_INTERLOCK;
 393         return (null_bypass((struct vop_generic_args *)ap));
 394 }
 395
 396 static int
 397 umap_inactive(ap)
 398         struct vop_inactive_args /* {
 399                 struct vnode *a_vp;
 400                 struct thread *a_td;
 401         } */ *ap;
 402 {
 403         struct vnode *vp = ap->a_vp;
 404         struct umap_node *xp = VTOUMAP(vp);
 405         struct vnode *lowervp = xp->umap_lowervp;
 406         /*
 407          * Do nothing (and _don't_ bypass).
 408          * Wait to vrele lowervp until reclaim,
 409          * so that until then our umap_node is in the
 410          * cache and reusable.
 411          *
 412          */
 413         VOP_INACTIVE(lowervp, ap->a_td);
 414         return (0);
 415 }
 416
 417 static int
 418 umap_reclaim(ap)
 419         struct vop_reclaim_args /* {
 420                 struct vnode *a_vp;
 421         } */ *ap;
 422 {
 423         struct vnode *vp = ap->a_vp;
 424         struct umap_node *xp = VTOUMAP(vp);
 425         struct vnode *lowervp = xp->umap_lowervp;
 426
 427         /* After this assignment, this node will not be re-used. */
 428         xp->umap_lowervp = NULL;
 429         LIST_REMOVE(xp, umap_hash);
 430         FREE(vp->v_data, M_TEMP);
 431         vp->v_data = NULL;
 432         vp->v_object = NULL;
 433         vrele(lowervp);
 434         return (0);
 435 }
 436
 437 static int
 438 umap_print(ap)
 439         struct vop_print_args /* {
 440                 struct vnode *a_vp;
 441         } */ *ap;
 442 {
 443         struct vnode *vp = ap->a_vp;
 444         printf("\tvp=%p, lowervp=%p\n", vp, UMAPVPTOLOWERVP(vp));
 445         return (0);
 446 }
 447
 448 static int
 449 umap_rename(ap)
 450         struct vop_rename_args  /* {
 451                 struct vnode *a_fdvp;
 452                 struct vnode *a_fvp;
 453                 struct componentname *a_fcnp;
 454                 struct vnode *a_tdvp;
 455                 struct vnode *a_tvp;
 456                 struct componentname *a_tcnp;
 457         } */ *ap;
 458 {
 459         int error;
 460         struct componentname *compnamep;
 461         struct ucred *compcredp, *savecompcredp;
 462         struct vnode *vp;
 463
 464         /*
 465          * Rename is irregular, having two componentname structures.
 466          * We need to map the cre in the second structure,
 467          * and then bypass takes care of the rest.
 468          */
 469
 470         vp = ap->a_fdvp;
 471         compnamep = ap->a_tcnp;
 472         compcredp = compnamep->cn_cred;
 473
 474         savecompcredp = compcredp;
 475         compcredp = compnamep->cn_cred = crdup(savecompcredp);
 476
 477         if (umap_bug_bypass && compcredp->cr_uid != 0)
 478                 printf(
 479             "umap_rename: rename component credit user was %lu, group %lu\n",
 480                     (u_long)compcredp->cr_uid, (u_long)compcredp->cr_gid);
 481
 482         /* Map all ids in the credential structure. */
 483
 484         umap_mapids(vp->v_mount, compcredp);
 485
 486         if (umap_bug_bypass && compcredp->cr_uid != 0)
 487                 printf(
 488             "umap_rename: rename component credit user now %lu, group %lu\n",
 489                     (u_long)compcredp->cr_uid, (u_long)compcredp->cr_gid);
 490
 491         error = umap_bypass((struct vop_generic_args *)ap);
 492
 493         /* Restore the additional mapped componentname cred structure. */
 494
 495         crfree(compcredp);
 496         compnamep->cn_cred = savecompcredp;
 497
 498         return error;
 499 }
 500
 501 /*
 502  * Global vfs data structures
 503  */
 504 /*
 505  * XXX - strategy, bwrite are hand coded currently.  They should
 506  * go away with a merged buffer/block cache.
 507  *
 508  */
 509 static struct vop_vector umap_vnodeops = {
 510         .vop_default =          umap_bypass,
 511
 512         .vop_getattr =          umap_getattr,
 513         .vop_inactive =         umap_inactive,
 514         .vop_lock =             umap_lock,
 515         .vop_print =            umap_print,
 516         .vop_reclaim =          umap_reclaim,
 517         .vop_rename =           umap_rename,
 518         .vop_unlock =           umap_unlock,
 519 };