| 1 | /* |
| 2 | * Copyright (c) 1988 University of Utah. |
| 3 | * Copyright (c) 1990 The Regents of the University of California. |
| 4 | * All rights reserved. |
| 5 | * |
| 6 | * This code is derived from software contributed to Berkeley by |
| 7 | * the Systems Programming Group of the University of Utah Computer |
| 8 | * Science Department. |
| 9 | * |
| 10 | * %sccs.include.redist.c% |
| 11 | * |
| 12 | * from: Utah $Hdr: vn.c 1.1 91/04/30$ |
| 13 | * |
| 14 | * @(#)vn.c 7.8 (Berkeley) %G% |
| 15 | */ |
| 16 | |
| 17 | /* |
| 18 | * Vnode disk driver. |
| 19 | * |
| 20 | * Block/character interface to a vnode. Allows one to treat a file |
| 21 | * as a disk (e.g. build a filesystem in it, mount it, etc.). |
| 22 | * |
| 23 | * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode |
| 24 | * instead of a simple VOP_RDWR. We do this to avoid distorting the |
| 25 | * local buffer cache. |
| 26 | * |
| 27 | * NOTE 2: There is a security issue involved with this driver. |
| 28 | * Once mounted all access to the contents of the "mapped" file via |
| 29 | * the special file is controlled by the permissions on the special |
| 30 | * file, the protection of the mapped file is ignored (effectively, |
| 31 | * by using root credentials in all transactions). |
| 32 | */ |
| 33 | #include "vn.h" |
| 34 | #if NVN > 0 |
| 35 | |
| 36 | #include "sys/param.h" |
| 37 | #include "sys/systm.h" |
| 38 | #include "sys/namei.h" |
| 39 | #include "sys/proc.h" |
| 40 | #include "sys/errno.h" |
| 41 | #include "sys/dkstat.h" |
| 42 | #include "sys/buf.h" |
| 43 | #include "sys/malloc.h" |
| 44 | #include "sys/ioctl.h" |
| 45 | #include "sys/mount.h" |
| 46 | #include "sys/vnode.h" |
| 47 | #include "sys/specdev.h" |
| 48 | #include "sys/file.h" |
| 49 | #include "sys/uio.h" |
| 50 | |
| 51 | #include "vnioctl.h" |
| 52 | |
| 53 | #ifdef DEBUG |
| 54 | int vndebug = 0x00; |
| 55 | #define VDB_FOLLOW 0x01 |
| 56 | #define VDB_INIT 0x02 |
| 57 | #define VDB_IO 0x04 |
| 58 | #endif |
| 59 | |
| 60 | struct buf vnbuf[NVN]; |
| 61 | struct buf vntab[NVN]; |
| 62 | |
| 63 | #define b_cylin b_resid |
| 64 | |
| 65 | #define vnunit(x) ((minor(x) >> 3) & 0x7) /* for consistency */ |
| 66 | |
| 67 | #define getvnbuf() \ |
| 68 | ((struct buf *)malloc(sizeof(struct buf), M_DEVBUF, M_WAITOK)) |
| 69 | #define putvnbuf(bp) \ |
| 70 | free((caddr_t)(bp), M_DEVBUF) |
| 71 | |
| 72 | struct vn_softc { |
| 73 | int sc_flags; /* flags */ |
| 74 | size_t sc_size; /* size of vn */ |
| 75 | struct vnode *sc_vp; /* vnode */ |
| 76 | struct ucred *sc_cred; /* credentials */ |
| 77 | int sc_maxactive; /* max # of active requests */ |
| 78 | } vn_softc[NVN]; |
| 79 | |
| 80 | /* sc_flags */ |
| 81 | #define VNF_ALIVE 0x01 |
| 82 | #define VNF_INITED 0x02 |
| 83 | |
| 84 | int |
| 85 | vnopen(dev, flags, mode, p) |
| 86 | dev_t dev; |
| 87 | int flags, mode; |
| 88 | struct proc *p; |
| 89 | { |
| 90 | int unit = vnunit(dev); |
| 91 | |
| 92 | #ifdef DEBUG |
| 93 | if (vndebug & VDB_FOLLOW) |
| 94 | printf("vnopen(%x, %x, %x, %x)\n", dev, flags, mode, p); |
| 95 | #endif |
| 96 | if (unit >= NVN) |
| 97 | return(ENXIO); |
| 98 | return(0); |
| 99 | } |
| 100 | |
| 101 | /* |
| 102 | * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY. |
| 103 | * Note that this driver can only be used for swapping over NFS on the hp |
| 104 | * since nfs_strategy on the vax cannot handle u-areas and page tables. |
| 105 | */ |
| 106 | vnstrategy(bp) |
| 107 | register struct buf *bp; |
| 108 | { |
| 109 | int unit = vnunit(bp->b_dev); |
| 110 | register struct vn_softc *vn = &vn_softc[unit]; |
| 111 | register struct buf *nbp; |
| 112 | register int bn, bsize, resid; |
| 113 | register caddr_t addr; |
| 114 | int sz, flags; |
| 115 | extern int vniodone(); |
| 116 | |
| 117 | #ifdef DEBUG |
| 118 | if (vndebug & VDB_FOLLOW) |
| 119 | printf("vnstrategy(%x): unit %d\n", bp, unit); |
| 120 | #endif |
| 121 | if ((vn->sc_flags & VNF_INITED) == 0) { |
| 122 | bp->b_error = ENXIO; |
| 123 | bp->b_flags |= B_ERROR; |
| 124 | biodone(bp); |
| 125 | return; |
| 126 | } |
| 127 | bn = bp->b_blkno; |
| 128 | sz = howmany(bp->b_bcount, DEV_BSIZE); |
| 129 | bp->b_resid = bp->b_bcount; |
| 130 | if (bn < 0 || bn + sz > vn->sc_size) { |
| 131 | if (bn != vn->sc_size) { |
| 132 | bp->b_error = EINVAL; |
| 133 | bp->b_flags |= B_ERROR; |
| 134 | } |
| 135 | biodone(bp); |
| 136 | return; |
| 137 | } |
| 138 | bn = dbtob(bn); |
| 139 | bsize = vn->sc_vp->v_mount->mnt_stat.f_iosize; |
| 140 | addr = bp->b_un.b_addr; |
| 141 | flags = bp->b_flags | B_CALL; |
| 142 | for (resid = bp->b_resid; resid; resid -= sz) { |
| 143 | struct vnode *vp; |
| 144 | daddr_t nbn; |
| 145 | int off, s; |
| 146 | |
| 147 | nbp = getvnbuf(); |
| 148 | off = bn % bsize; |
| 149 | sz = MIN(bsize - off, resid); |
| 150 | (void) VOP_BMAP(vn->sc_vp, bn / bsize, &vp, &nbn); |
| 151 | #ifdef DEBUG |
| 152 | if (vndebug & VDB_IO) |
| 153 | printf("vnstrategy: vp %x/%x bn %x/%x\n", |
| 154 | vn->sc_vp, vp, bn, nbn); |
| 155 | #endif |
| 156 | nbp->b_flags = flags; |
| 157 | nbp->b_bcount = sz; |
| 158 | nbp->b_bufsize = bp->b_bufsize; |
| 159 | nbp->b_error = 0; |
| 160 | if (vp->v_type == VBLK || vp->v_type == VCHR) |
| 161 | nbp->b_dev = vp->v_rdev; |
| 162 | else |
| 163 | nbp->b_dev = NODEV; |
| 164 | nbp->b_un.b_addr = addr; |
| 165 | nbp->b_blkno = nbn + btodb(off); |
| 166 | nbp->b_proc = bp->b_proc; |
| 167 | nbp->b_iodone = vniodone; |
| 168 | nbp->b_vp = vp; |
| 169 | nbp->b_pfcent = (int) bp; /* XXX */ |
| 170 | /* |
| 171 | * Just sort by block number |
| 172 | */ |
| 173 | nbp->b_cylin = nbp->b_blkno; |
| 174 | s = splbio(); |
| 175 | disksort(&vntab[unit], nbp); |
| 176 | if (vntab[unit].b_active < vn->sc_maxactive) { |
| 177 | vntab[unit].b_active++; |
| 178 | vnstart(unit); |
| 179 | } |
| 180 | splx(s); |
| 181 | bn += sz; |
| 182 | addr += sz; |
| 183 | } |
| 184 | } |
| 185 | |
| 186 | /* |
| 187 | * Feed requests sequentially. |
| 188 | * We do it this way to keep from flooding NFS servers if we are connected |
| 189 | * to an NFS file. This places the burden on the client rather than the |
| 190 | * server. |
| 191 | */ |
| 192 | vnstart(unit) |
| 193 | { |
| 194 | register struct vn_softc *vn = &vn_softc[unit]; |
| 195 | register struct buf *bp; |
| 196 | |
| 197 | /* |
| 198 | * Dequeue now since lower level strategy routine might |
| 199 | * queue using same links |
| 200 | */ |
| 201 | bp = vntab[unit].b_actf; |
| 202 | vntab[unit].b_actf = bp->b_actf; |
| 203 | #ifdef DEBUG |
| 204 | if (vndebug & VDB_IO) |
| 205 | printf("vnstart(%d): bp %x vp %x blkno %x addr %x cnt %x\n", |
| 206 | unit, bp, bp->b_vp, bp->b_blkno, bp->b_un.b_addr, |
| 207 | bp->b_bcount); |
| 208 | #endif |
| 209 | VOP_STRATEGY(bp); |
| 210 | } |
| 211 | |
| 212 | vniodone(bp) |
| 213 | register struct buf *bp; |
| 214 | { |
| 215 | register struct buf *pbp = (struct buf *)bp->b_pfcent; /* XXX */ |
| 216 | register int unit = vnunit(pbp->b_dev); |
| 217 | int s; |
| 218 | |
| 219 | s = splbio(); |
| 220 | #ifdef DEBUG |
| 221 | if (vndebug & VDB_IO) |
| 222 | printf("vniodone(%d): bp %x vp %x blkno %x addr %x cnt %x\n", |
| 223 | unit, bp, bp->b_vp, bp->b_blkno, bp->b_un.b_addr, |
| 224 | bp->b_bcount); |
| 225 | #endif |
| 226 | if (bp->b_error) { |
| 227 | #ifdef DEBUG |
| 228 | if (vndebug & VDB_IO) |
| 229 | printf("vniodone: bp %x error %d\n", bp, bp->b_error); |
| 230 | #endif |
| 231 | pbp->b_flags |= B_ERROR; |
| 232 | pbp->b_error = biowait(bp); |
| 233 | } |
| 234 | pbp->b_resid -= bp->b_bcount; |
| 235 | putvnbuf(bp); |
| 236 | if (pbp->b_resid == 0) { |
| 237 | #ifdef DEBUG |
| 238 | if (vndebug & VDB_IO) |
| 239 | printf("vniodone: pbp %x iodone\n", pbp); |
| 240 | #endif |
| 241 | biodone(pbp); |
| 242 | } |
| 243 | if (vntab[unit].b_actf) |
| 244 | vnstart(unit); |
| 245 | else |
| 246 | vntab[unit].b_active--; |
| 247 | splx(s); |
| 248 | } |
| 249 | |
| 250 | vnread(dev, uio, flags, p) |
| 251 | dev_t dev; |
| 252 | struct uio *uio; |
| 253 | int flags; |
| 254 | struct proc *p; |
| 255 | { |
| 256 | register int unit = vnunit(dev); |
| 257 | |
| 258 | #ifdef DEBUG |
| 259 | if (vndebug & VDB_FOLLOW) |
| 260 | printf("vnread(%x, %x, %x, %x)\n", dev, uio, flags, p); |
| 261 | #endif |
| 262 | return(physio(vnstrategy, &vnbuf[unit], dev, B_READ, minphys, uio)); |
| 263 | } |
| 264 | |
| 265 | vnwrite(dev, uio, flags, p) |
| 266 | dev_t dev; |
| 267 | struct uio *uio; |
| 268 | int flags; |
| 269 | struct proc *p; |
| 270 | { |
| 271 | register int unit = vnunit(dev); |
| 272 | |
| 273 | #ifdef DEBUG |
| 274 | if (vndebug & VDB_FOLLOW) |
| 275 | printf("vnwrite(%x, %x, %x, %x)\n", dev, uio, flags, p); |
| 276 | #endif |
| 277 | return(physio(vnstrategy, &vnbuf[unit], dev, B_WRITE, minphys, uio)); |
| 278 | } |
| 279 | |
| 280 | /* ARGSUSED */ |
| 281 | vnioctl(dev, cmd, data, flag, p) |
| 282 | dev_t dev; |
| 283 | u_long cmd; |
| 284 | caddr_t data; |
| 285 | int flag; |
| 286 | struct proc *p; |
| 287 | { |
| 288 | int unit = vnunit(dev); |
| 289 | register struct vn_softc *vn; |
| 290 | struct vn_ioctl *vio; |
| 291 | struct vattr vattr; |
| 292 | struct nameidata nd; |
| 293 | int error; |
| 294 | |
| 295 | #ifdef DEBUG |
| 296 | if (vndebug & VDB_FOLLOW) |
| 297 | printf("vnioctl(%x, %x, %x, %x, %x): unit %d\n", |
| 298 | dev, cmd, data, flag, p, unit); |
| 299 | #endif |
| 300 | error = suser(p->p_ucred, &p->p_acflag); |
| 301 | if (error) |
| 302 | return (error); |
| 303 | if (unit >= NVN) |
| 304 | return (ENXIO); |
| 305 | |
| 306 | vn = &vn_softc[unit]; |
| 307 | vio = (struct vn_ioctl *)data; |
| 308 | switch (cmd) { |
| 309 | |
| 310 | case VNIOCSET: |
| 311 | if (vn->sc_flags & VNF_INITED) |
| 312 | return(EBUSY); |
| 313 | /* |
| 314 | * Always open for read and write. |
| 315 | * This is probably bogus, but it lets vn_open() |
| 316 | * weed out directories, sockets, etc. so we don't |
| 317 | * have to worry about them. |
| 318 | */ |
| 319 | NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vn_file, p); |
| 320 | if (error = vn_open(&nd, FREAD|FWRITE, 0)) |
| 321 | return(error); |
| 322 | if (error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p)) { |
| 323 | VOP_UNLOCK(nd.ni_vp); |
| 324 | (void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p); |
| 325 | return(error); |
| 326 | } |
| 327 | VOP_UNLOCK(nd.ni_vp); |
| 328 | vn->sc_vp = nd.ni_vp; |
| 329 | vn->sc_size = btodb(vattr.va_size); /* note truncation */ |
| 330 | if (error = vnsetcred(vn, p->p_ucred)) { |
| 331 | (void) vn_close(vn->sc_vp, FREAD|FWRITE, p->p_ucred, p); |
| 332 | return(error); |
| 333 | } |
| 334 | vnthrottle(vn, vn->sc_vp); |
| 335 | vio->vn_size = dbtob(vn->sc_size); |
| 336 | vn->sc_flags |= VNF_INITED; |
| 337 | #ifdef DEBUG |
| 338 | if (vndebug & VDB_INIT) |
| 339 | printf("vnioctl: SET vp %x size %x\n", |
| 340 | vn->sc_vp, vn->sc_size); |
| 341 | #endif |
| 342 | break; |
| 343 | |
| 344 | case VNIOCCLR: |
| 345 | if ((vn->sc_flags & VNF_INITED) == 0) |
| 346 | return(ENXIO); |
| 347 | vnclear(vn); |
| 348 | #ifdef DEBUG |
| 349 | if (vndebug & VDB_INIT) |
| 350 | printf("vnioctl: CLRed\n"); |
| 351 | #endif |
| 352 | break; |
| 353 | |
| 354 | default: |
| 355 | return(ENXIO); |
| 356 | } |
| 357 | return(0); |
| 358 | } |
| 359 | |
| 360 | /* |
| 361 | * Duplicate the current processes' credentials. Since we are called only |
| 362 | * as the result of a SET ioctl and only root can do that, any future access |
| 363 | * to this "disk" is essentially as root. Note that credentials may change |
| 364 | * if some other uid can write directly to the mapped file (NFS). |
| 365 | */ |
| 366 | vnsetcred(vn, cred) |
| 367 | register struct vn_softc *vn; |
| 368 | struct ucred cred; |
| 369 | { |
| 370 | struct uio auio; |
| 371 | struct iovec aiov; |
| 372 | char tmpbuf[DEV_BSIZE]; |
| 373 | |
| 374 | vn->sc_cred = crdup(cred); |
| 375 | /* XXX: Horrible kludge to establish credentials for NFS */ |
| 376 | aiov.iov_base = tmpbuf; |
| 377 | aiov.iov_len = MIN(DEV_BSIZE, dbtob(vn->sc_size)); |
| 378 | auio.uio_iov = &aiov; |
| 379 | auio.uio_iovcnt = 1; |
| 380 | auio.uio_offset = 0; |
| 381 | auio.uio_rw = UIO_READ; |
| 382 | auio.uio_segflg = UIO_SYSSPACE; |
| 383 | auio.uio_resid = aiov.iov_len; |
| 384 | return(VOP_READ(vn->sc_vp, &auio, 0, vn->sc_cred)); |
| 385 | } |
| 386 | |
| 387 | /* |
| 388 | * Set maxactive based on FS type |
| 389 | */ |
| 390 | vnthrottle(vn, vp) |
| 391 | register struct vn_softc *vn; |
| 392 | struct vnode *vp; |
| 393 | { |
| 394 | extern struct vnodeops ufs_vnodeops, nfsv2_vnodeops; |
| 395 | |
| 396 | if (vp->v_op == &nfsv2_vnodeops) |
| 397 | vn->sc_maxactive = 2; |
| 398 | else |
| 399 | vn->sc_maxactive = 8; |
| 400 | |
| 401 | if (vn->sc_maxactive < 1) |
| 402 | vn->sc_maxactive = 1; |
| 403 | } |
| 404 | |
| 405 | vnshutdown() |
| 406 | { |
| 407 | register struct vn_softc *vn; |
| 408 | |
| 409 | for (vn = &vn_softc[0]; vn < &vn_softc[NVN]; vn++) |
| 410 | if (vn->sc_flags & VNF_INITED) |
| 411 | vnclear(vn); |
| 412 | } |
| 413 | |
| 414 | vnclear(vn) |
| 415 | register struct vn_softc *vn; |
| 416 | { |
| 417 | register struct vnode *vp = vn->sc_vp; |
| 418 | struct proc *p = curproc; /* XXX */ |
| 419 | |
| 420 | #ifdef DEBUG |
| 421 | if (vndebug & VDB_FOLLOW) |
| 422 | printf("vnclear(%x): vp %x\n", vp); |
| 423 | #endif |
| 424 | vn->sc_flags &= ~VNF_INITED; |
| 425 | if (vp == (struct vnode *)0) |
| 426 | panic("vnioctl: null vp"); |
| 427 | #if 0 |
| 428 | /* XXX - this doesn't work right now */ |
| 429 | (void) VOP_FSYNC(vp, 0, vn->sc_cred, MNT_WAIT, p); |
| 430 | #endif |
| 431 | (void) vn_close(vp, FREAD|FWRITE, vn->sc_cred, p); |
| 432 | crfree(vn->sc_cred); |
| 433 | vn->sc_vp = (struct vnode *)0; |
| 434 | vn->sc_cred = (struct ucred *)0; |
| 435 | vn->sc_size = 0; |
| 436 | } |
| 437 | |
| 438 | vnsize(dev) |
| 439 | dev_t dev; |
| 440 | { |
| 441 | int unit = vnunit(dev); |
| 442 | register struct vn_softc *vn = &vn_softc[unit]; |
| 443 | |
| 444 | if (unit >= NVN || (vn->sc_flags & VNF_INITED) == 0) |
| 445 | return(-1); |
| 446 | return(vn->sc_size); |
| 447 | } |
| 448 | |
| 449 | vndump(dev) |
| 450 | { |
| 451 | return(ENXIO); |
| 452 | } |
| 453 | #endif |