| 1 | /* |
| 2 | * Copyright (c) 1994 Jan-Simon Pendry |
| 3 | * Copyright (c) 1994 |
| 4 | * The Regents of the University of California. All rights reserved. |
| 5 | * |
| 6 | * This code is derived from software contributed to Berkeley by |
| 7 | * Jan-Simon Pendry. |
| 8 | * |
| 9 | * %sccs.include.redist.c% |
| 10 | * |
| 11 | * @(#)union_subr.c 8.16 (Berkeley) %G% |
| 12 | */ |
| 13 | |
| 14 | #include <sys/param.h> |
| 15 | #include <sys/systm.h> |
| 16 | #include <sys/time.h> |
| 17 | #include <sys/kernel.h> |
| 18 | #include <sys/vnode.h> |
| 19 | #include <sys/namei.h> |
| 20 | #include <sys/malloc.h> |
| 21 | #include <sys/file.h> |
| 22 | #include <sys/filedesc.h> |
| 23 | #include <sys/queue.h> |
| 24 | #include <sys/mount.h> |
| 25 | #include <sys/stat.h> |
| 26 | #include <vm/vm.h> /* for vnode_pager_setsize */ |
| 27 | #include <miscfs/union/union.h> |
| 28 | |
| 29 | #ifdef DIAGNOSTIC |
| 30 | #include <sys/proc.h> |
| 31 | #endif |
| 32 | |
| 33 | /* must be power of two, otherwise change UNION_HASH() */ |
| 34 | #define NHASH 32 |
| 35 | |
| 36 | /* unsigned int ... */ |
| 37 | #define UNION_HASH(u, l) \ |
| 38 | (((((unsigned long) (u)) + ((unsigned long) l)) >> 8) & (NHASH-1)) |
| 39 | |
| 40 | static LIST_HEAD(unhead, union_node) unhead[NHASH]; |
| 41 | static int unvplock[NHASH]; |
| 42 | |
| 43 | int |
| 44 | union_init() |
| 45 | { |
| 46 | int i; |
| 47 | |
| 48 | for (i = 0; i < NHASH; i++) |
| 49 | LIST_INIT(&unhead[i]); |
| 50 | bzero((caddr_t) unvplock, sizeof(unvplock)); |
| 51 | } |
| 52 | |
| 53 | static int |
| 54 | union_list_lock(ix) |
| 55 | int ix; |
| 56 | { |
| 57 | |
| 58 | if (unvplock[ix] & UN_LOCKED) { |
| 59 | unvplock[ix] |= UN_WANT; |
| 60 | sleep((caddr_t) &unvplock[ix], PINOD); |
| 61 | return (1); |
| 62 | } |
| 63 | |
| 64 | unvplock[ix] |= UN_LOCKED; |
| 65 | |
| 66 | return (0); |
| 67 | } |
| 68 | |
| 69 | static void |
| 70 | union_list_unlock(ix) |
| 71 | int ix; |
| 72 | { |
| 73 | |
| 74 | unvplock[ix] &= ~UN_LOCKED; |
| 75 | |
| 76 | if (unvplock[ix] & UN_WANT) { |
| 77 | unvplock[ix] &= ~UN_WANT; |
| 78 | wakeup((caddr_t) &unvplock[ix]); |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | void |
| 83 | union_updatevp(un, uppervp, lowervp) |
| 84 | struct union_node *un; |
| 85 | struct vnode *uppervp; |
| 86 | struct vnode *lowervp; |
| 87 | { |
| 88 | int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp); |
| 89 | int nhash = UNION_HASH(uppervp, lowervp); |
| 90 | int docache = (lowervp != NULLVP || uppervp != NULLVP); |
| 91 | |
| 92 | /* |
| 93 | * Ensure locking is ordered from lower to higher |
| 94 | * to avoid deadlocks. |
| 95 | */ |
| 96 | if (nhash < ohash) { |
| 97 | int t = ohash; |
| 98 | ohash = nhash; |
| 99 | nhash = t; |
| 100 | } |
| 101 | |
| 102 | if (ohash != nhash) |
| 103 | while (union_list_lock(ohash)) |
| 104 | continue; |
| 105 | |
| 106 | while (union_list_lock(nhash)) |
| 107 | continue; |
| 108 | |
| 109 | if (ohash != nhash || !docache) { |
| 110 | if (un->un_flags & UN_CACHED) { |
| 111 | un->un_flags &= ~UN_CACHED; |
| 112 | LIST_REMOVE(un, un_cache); |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | if (ohash != nhash) |
| 117 | union_list_unlock(ohash); |
| 118 | |
| 119 | if (un->un_lowervp != lowervp) { |
| 120 | if (un->un_lowervp) { |
| 121 | vrele(un->un_lowervp); |
| 122 | if (un->un_path) { |
| 123 | free(un->un_path, M_TEMP); |
| 124 | un->un_path = 0; |
| 125 | } |
| 126 | if (un->un_dirvp) { |
| 127 | vrele(un->un_dirvp); |
| 128 | un->un_dirvp = NULLVP; |
| 129 | } |
| 130 | } |
| 131 | un->un_lowervp = lowervp; |
| 132 | un->un_lowersz = VNOVAL; |
| 133 | } |
| 134 | |
| 135 | if (un->un_uppervp != uppervp) { |
| 136 | if (un->un_uppervp) |
| 137 | vrele(un->un_uppervp); |
| 138 | |
| 139 | un->un_uppervp = uppervp; |
| 140 | un->un_uppersz = VNOVAL; |
| 141 | } |
| 142 | |
| 143 | if (docache && (ohash != nhash)) { |
| 144 | LIST_INSERT_HEAD(&unhead[nhash], un, un_cache); |
| 145 | un->un_flags |= UN_CACHED; |
| 146 | } |
| 147 | |
| 148 | union_list_unlock(nhash); |
| 149 | } |
| 150 | |
| 151 | void |
| 152 | union_newlower(un, lowervp) |
| 153 | struct union_node *un; |
| 154 | struct vnode *lowervp; |
| 155 | { |
| 156 | |
| 157 | union_updatevp(un, un->un_uppervp, lowervp); |
| 158 | } |
| 159 | |
| 160 | void |
| 161 | union_newupper(un, uppervp) |
| 162 | struct union_node *un; |
| 163 | struct vnode *uppervp; |
| 164 | { |
| 165 | |
| 166 | union_updatevp(un, uppervp, un->un_lowervp); |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | * Keep track of size changes in the underlying vnodes. |
| 171 | * If the size changes, then callback to the vm layer |
| 172 | * giving priority to the upper layer size. |
| 173 | */ |
| 174 | void |
| 175 | union_newsize(vp, uppersz, lowersz) |
| 176 | struct vnode *vp; |
| 177 | off_t uppersz, lowersz; |
| 178 | { |
| 179 | struct union_node *un; |
| 180 | off_t sz; |
| 181 | |
| 182 | /* only interested in regular files */ |
| 183 | if (vp->v_type != VREG) |
| 184 | return; |
| 185 | |
| 186 | un = VTOUNION(vp); |
| 187 | sz = VNOVAL; |
| 188 | |
| 189 | if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) { |
| 190 | un->un_uppersz = uppersz; |
| 191 | if (sz == VNOVAL) |
| 192 | sz = un->un_uppersz; |
| 193 | } |
| 194 | |
| 195 | if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) { |
| 196 | un->un_lowersz = lowersz; |
| 197 | if (sz == VNOVAL) |
| 198 | sz = un->un_lowersz; |
| 199 | } |
| 200 | |
| 201 | if (sz != VNOVAL) { |
| 202 | #ifdef UNION_DIAGNOSTIC |
| 203 | printf("union: %s size now %ld\n", |
| 204 | uppersz != VNOVAL ? "upper" : "lower", (long) sz); |
| 205 | #endif |
| 206 | vnode_pager_setsize(vp, sz); |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | /* |
| 211 | * allocate a union_node/vnode pair. the vnode is |
| 212 | * referenced and locked. the new vnode is returned |
| 213 | * via (vpp). (mp) is the mountpoint of the union filesystem, |
| 214 | * (dvp) is the parent directory where the upper layer object |
| 215 | * should exist (but doesn't) and (cnp) is the componentname |
| 216 | * information which is partially copied to allow the upper |
| 217 | * layer object to be created at a later time. (uppervp) |
| 218 | * and (lowervp) reference the upper and lower layer objects |
| 219 | * being mapped. either, but not both, can be nil. |
| 220 | * if supplied, (uppervp) is locked. |
| 221 | * the reference is either maintained in the new union_node |
| 222 | * object which is allocated, or they are vrele'd. |
| 223 | * |
| 224 | * all union_nodes are maintained on a singly-linked |
| 225 | * list. new nodes are only allocated when they cannot |
| 226 | * be found on this list. entries on the list are |
| 227 | * removed when the vfs reclaim entry is called. |
| 228 | * |
| 229 | * a single lock is kept for the entire list. this is |
| 230 | * needed because the getnewvnode() function can block |
| 231 | * waiting for a vnode to become free, in which case there |
| 232 | * may be more than one process trying to get the same |
| 233 | * vnode. this lock is only taken if we are going to |
| 234 | * call getnewvnode, since the kernel itself is single-threaded. |
| 235 | * |
| 236 | * if an entry is found on the list, then call vget() to |
| 237 | * take a reference. this is done because there may be |
| 238 | * zero references to it and so it needs to removed from |
| 239 | * the vnode free list. |
| 240 | */ |
| 241 | int |
| 242 | union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp, docache) |
| 243 | struct vnode **vpp; |
| 244 | struct mount *mp; |
| 245 | struct vnode *undvp; /* parent union vnode */ |
| 246 | struct vnode *dvp; /* may be null */ |
| 247 | struct componentname *cnp; /* may be null */ |
| 248 | struct vnode *uppervp; /* may be null */ |
| 249 | struct vnode *lowervp; /* may be null */ |
| 250 | int docache; |
| 251 | { |
| 252 | int error; |
| 253 | struct union_node *un; |
| 254 | struct union_node **pp; |
| 255 | struct vnode *xlowervp = NULLVP; |
| 256 | struct union_mount *um = MOUNTTOUNIONMOUNT(mp); |
| 257 | int hash; |
| 258 | int vflag; |
| 259 | int try; |
| 260 | |
| 261 | if (uppervp == NULLVP && lowervp == NULLVP) |
| 262 | panic("union: unidentifiable allocation"); |
| 263 | |
| 264 | if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) { |
| 265 | xlowervp = lowervp; |
| 266 | lowervp = NULLVP; |
| 267 | } |
| 268 | |
| 269 | /* detect the root vnode (and aliases) */ |
| 270 | vflag = 0; |
| 271 | if ((uppervp == um->um_uppervp) && |
| 272 | ((lowervp == NULLVP) || lowervp == um->um_lowervp)) { |
| 273 | if (lowervp == NULLVP) { |
| 274 | lowervp = um->um_lowervp; |
| 275 | if (lowervp != NULLVP) |
| 276 | VREF(lowervp); |
| 277 | } |
| 278 | vflag = VROOT; |
| 279 | } |
| 280 | |
| 281 | loop: |
| 282 | if (!docache) { |
| 283 | un = 0; |
| 284 | } else for (try = 0; try < 3; try++) { |
| 285 | switch (try) { |
| 286 | case 0: |
| 287 | if (lowervp == NULLVP) |
| 288 | continue; |
| 289 | hash = UNION_HASH(uppervp, lowervp); |
| 290 | break; |
| 291 | |
| 292 | case 1: |
| 293 | if (uppervp == NULLVP) |
| 294 | continue; |
| 295 | hash = UNION_HASH(uppervp, NULLVP); |
| 296 | break; |
| 297 | |
| 298 | case 2: |
| 299 | if (lowervp == NULLVP) |
| 300 | continue; |
| 301 | hash = UNION_HASH(NULLVP, lowervp); |
| 302 | break; |
| 303 | } |
| 304 | |
| 305 | while (union_list_lock(hash)) |
| 306 | continue; |
| 307 | |
| 308 | for (un = unhead[hash].lh_first; un != 0; |
| 309 | un = un->un_cache.le_next) { |
| 310 | if ((un->un_lowervp == lowervp || |
| 311 | un->un_lowervp == NULLVP) && |
| 312 | (un->un_uppervp == uppervp || |
| 313 | un->un_uppervp == NULLVP) && |
| 314 | (UNIONTOV(un)->v_mount == mp)) { |
| 315 | if (vget(UNIONTOV(un), 0)) { |
| 316 | union_list_unlock(hash); |
| 317 | goto loop; |
| 318 | } |
| 319 | break; |
| 320 | } |
| 321 | } |
| 322 | |
| 323 | union_list_unlock(hash); |
| 324 | |
| 325 | if (un) |
| 326 | break; |
| 327 | } |
| 328 | |
| 329 | if (un) { |
| 330 | /* |
| 331 | * Obtain a lock on the union_node. |
| 332 | * uppervp is locked, though un->un_uppervp |
| 333 | * may not be. this doesn't break the locking |
| 334 | * hierarchy since in the case that un->un_uppervp |
| 335 | * is not yet locked it will be vrele'd and replaced |
| 336 | * with uppervp. |
| 337 | */ |
| 338 | |
| 339 | if ((dvp != NULLVP) && (uppervp == dvp)) { |
| 340 | /* |
| 341 | * Access ``.'', so (un) will already |
| 342 | * be locked. Since this process has |
| 343 | * the lock on (uppervp) no other |
| 344 | * process can hold the lock on (un). |
| 345 | */ |
| 346 | #ifdef DIAGNOSTIC |
| 347 | if ((un->un_flags & UN_LOCKED) == 0) |
| 348 | panic("union: . not locked"); |
| 349 | else if (curproc && un->un_pid != curproc->p_pid && |
| 350 | un->un_pid > -1 && curproc->p_pid > -1) |
| 351 | panic("union: allocvp not lock owner"); |
| 352 | #endif |
| 353 | } else { |
| 354 | if (un->un_flags & UN_LOCKED) { |
| 355 | vrele(UNIONTOV(un)); |
| 356 | un->un_flags |= UN_WANT; |
| 357 | sleep((caddr_t) &un->un_flags, PINOD); |
| 358 | goto loop; |
| 359 | } |
| 360 | un->un_flags |= UN_LOCKED; |
| 361 | |
| 362 | #ifdef DIAGNOSTIC |
| 363 | if (curproc) |
| 364 | un->un_pid = curproc->p_pid; |
| 365 | else |
| 366 | un->un_pid = -1; |
| 367 | #endif |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * At this point, the union_node is locked, |
| 372 | * un->un_uppervp may not be locked, and uppervp |
| 373 | * is locked or nil. |
| 374 | */ |
| 375 | |
| 376 | /* |
| 377 | * Save information about the upper layer. |
| 378 | */ |
| 379 | if (uppervp != un->un_uppervp) { |
| 380 | union_newupper(un, uppervp); |
| 381 | } else if (uppervp) { |
| 382 | vrele(uppervp); |
| 383 | } |
| 384 | |
| 385 | if (un->un_uppervp) { |
| 386 | un->un_flags |= UN_ULOCK; |
| 387 | un->un_flags &= ~UN_KLOCK; |
| 388 | } |
| 389 | |
| 390 | /* |
| 391 | * Save information about the lower layer. |
| 392 | * This needs to keep track of pathname |
| 393 | * and directory information which union_vn_create |
| 394 | * might need. |
| 395 | */ |
| 396 | if (lowervp != un->un_lowervp) { |
| 397 | union_newlower(un, lowervp); |
| 398 | if (cnp && (lowervp != NULLVP)) { |
| 399 | un->un_hash = cnp->cn_hash; |
| 400 | un->un_path = malloc(cnp->cn_namelen+1, |
| 401 | M_TEMP, M_WAITOK); |
| 402 | bcopy(cnp->cn_nameptr, un->un_path, |
| 403 | cnp->cn_namelen); |
| 404 | un->un_path[cnp->cn_namelen] = '\0'; |
| 405 | VREF(dvp); |
| 406 | un->un_dirvp = dvp; |
| 407 | } |
| 408 | } else if (lowervp) { |
| 409 | vrele(lowervp); |
| 410 | } |
| 411 | *vpp = UNIONTOV(un); |
| 412 | return (0); |
| 413 | } |
| 414 | |
| 415 | if (docache) { |
| 416 | /* |
| 417 | * otherwise lock the vp list while we call getnewvnode |
| 418 | * since that can block. |
| 419 | */ |
| 420 | hash = UNION_HASH(uppervp, lowervp); |
| 421 | |
| 422 | if (union_list_lock(hash)) |
| 423 | goto loop; |
| 424 | } |
| 425 | |
| 426 | error = getnewvnode(VT_UNION, mp, union_vnodeop_p, vpp); |
| 427 | if (error) { |
| 428 | if (uppervp) { |
| 429 | if (dvp == uppervp) |
| 430 | vrele(uppervp); |
| 431 | else |
| 432 | vput(uppervp); |
| 433 | } |
| 434 | if (lowervp) |
| 435 | vrele(lowervp); |
| 436 | |
| 437 | goto out; |
| 438 | } |
| 439 | |
| 440 | MALLOC((*vpp)->v_data, void *, sizeof(struct union_node), |
| 441 | M_TEMP, M_WAITOK); |
| 442 | |
| 443 | (*vpp)->v_flag |= vflag; |
| 444 | if (uppervp) |
| 445 | (*vpp)->v_type = uppervp->v_type; |
| 446 | else |
| 447 | (*vpp)->v_type = lowervp->v_type; |
| 448 | un = VTOUNION(*vpp); |
| 449 | un->un_vnode = *vpp; |
| 450 | un->un_uppervp = uppervp; |
| 451 | un->un_uppersz = VNOVAL; |
| 452 | un->un_lowervp = lowervp; |
| 453 | un->un_lowersz = VNOVAL; |
| 454 | un->un_pvp = undvp; |
| 455 | if (undvp != NULLVP) |
| 456 | VREF(undvp); |
| 457 | un->un_dircache = 0; |
| 458 | un->un_openl = 0; |
| 459 | un->un_flags = UN_LOCKED; |
| 460 | if (un->un_uppervp) |
| 461 | un->un_flags |= UN_ULOCK; |
| 462 | #ifdef DIAGNOSTIC |
| 463 | if (curproc) |
| 464 | un->un_pid = curproc->p_pid; |
| 465 | else |
| 466 | un->un_pid = -1; |
| 467 | #endif |
| 468 | if (cnp && (lowervp != NULLVP)) { |
| 469 | un->un_hash = cnp->cn_hash; |
| 470 | un->un_path = malloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK); |
| 471 | bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen); |
| 472 | un->un_path[cnp->cn_namelen] = '\0'; |
| 473 | VREF(dvp); |
| 474 | un->un_dirvp = dvp; |
| 475 | } else { |
| 476 | un->un_hash = 0; |
| 477 | un->un_path = 0; |
| 478 | un->un_dirvp = 0; |
| 479 | } |
| 480 | |
| 481 | if (docache) { |
| 482 | LIST_INSERT_HEAD(&unhead[hash], un, un_cache); |
| 483 | un->un_flags |= UN_CACHED; |
| 484 | } |
| 485 | |
| 486 | if (xlowervp) |
| 487 | vrele(xlowervp); |
| 488 | |
| 489 | out: |
| 490 | if (docache) |
| 491 | union_list_unlock(hash); |
| 492 | |
| 493 | return (error); |
| 494 | } |
| 495 | |
| 496 | int |
| 497 | union_freevp(vp) |
| 498 | struct vnode *vp; |
| 499 | { |
| 500 | struct union_node *un = VTOUNION(vp); |
| 501 | |
| 502 | if (un->un_flags & UN_CACHED) { |
| 503 | un->un_flags &= ~UN_CACHED; |
| 504 | LIST_REMOVE(un, un_cache); |
| 505 | } |
| 506 | |
| 507 | if (un->un_pvp != NULLVP) |
| 508 | vrele(un->un_pvp); |
| 509 | if (un->un_uppervp != NULLVP) |
| 510 | vrele(un->un_uppervp); |
| 511 | if (un->un_lowervp != NULLVP) |
| 512 | vrele(un->un_lowervp); |
| 513 | if (un->un_dirvp != NULLVP) |
| 514 | vrele(un->un_dirvp); |
| 515 | if (un->un_path) |
| 516 | free(un->un_path, M_TEMP); |
| 517 | |
| 518 | FREE(vp->v_data, M_TEMP); |
| 519 | vp->v_data = 0; |
| 520 | |
| 521 | return (0); |
| 522 | } |
| 523 | |
| 524 | /* |
| 525 | * copyfile. copy the vnode (fvp) to the vnode (tvp) |
| 526 | * using a sequence of reads and writes. both (fvp) |
| 527 | * and (tvp) are locked on entry and exit. |
| 528 | */ |
| 529 | int |
| 530 | union_copyfile(fvp, tvp, cred, p) |
| 531 | struct vnode *fvp; |
| 532 | struct vnode *tvp; |
| 533 | struct ucred *cred; |
| 534 | struct proc *p; |
| 535 | { |
| 536 | char *buf; |
| 537 | struct uio uio; |
| 538 | struct iovec iov; |
| 539 | int error = 0; |
| 540 | |
| 541 | /* |
| 542 | * strategy: |
| 543 | * allocate a buffer of size MAXBSIZE. |
| 544 | * loop doing reads and writes, keeping track |
| 545 | * of the current uio offset. |
| 546 | * give up at the first sign of trouble. |
| 547 | */ |
| 548 | |
| 549 | uio.uio_procp = p; |
| 550 | uio.uio_segflg = UIO_SYSSPACE; |
| 551 | uio.uio_offset = 0; |
| 552 | |
| 553 | VOP_UNLOCK(fvp); /* XXX */ |
| 554 | VOP_LEASE(fvp, p, cred, LEASE_READ); |
| 555 | VOP_LOCK(fvp); /* XXX */ |
| 556 | VOP_UNLOCK(tvp); /* XXX */ |
| 557 | VOP_LEASE(tvp, p, cred, LEASE_WRITE); |
| 558 | VOP_LOCK(tvp); /* XXX */ |
| 559 | |
| 560 | buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); |
| 561 | |
| 562 | /* ugly loop follows... */ |
| 563 | do { |
| 564 | off_t offset = uio.uio_offset; |
| 565 | |
| 566 | uio.uio_iov = &iov; |
| 567 | uio.uio_iovcnt = 1; |
| 568 | iov.iov_base = buf; |
| 569 | iov.iov_len = MAXBSIZE; |
| 570 | uio.uio_resid = iov.iov_len; |
| 571 | uio.uio_rw = UIO_READ; |
| 572 | error = VOP_READ(fvp, &uio, 0, cred); |
| 573 | |
| 574 | if (error == 0) { |
| 575 | uio.uio_iov = &iov; |
| 576 | uio.uio_iovcnt = 1; |
| 577 | iov.iov_base = buf; |
| 578 | iov.iov_len = MAXBSIZE - uio.uio_resid; |
| 579 | uio.uio_offset = offset; |
| 580 | uio.uio_rw = UIO_WRITE; |
| 581 | uio.uio_resid = iov.iov_len; |
| 582 | |
| 583 | if (uio.uio_resid == 0) |
| 584 | break; |
| 585 | |
| 586 | do { |
| 587 | error = VOP_WRITE(tvp, &uio, 0, cred); |
| 588 | } while ((uio.uio_resid > 0) && (error == 0)); |
| 589 | } |
| 590 | |
| 591 | } while (error == 0); |
| 592 | |
| 593 | free(buf, M_TEMP); |
| 594 | return (error); |
| 595 | } |
| 596 | |
| 597 | /* |
| 598 | * (un) is assumed to be locked on entry and remains |
| 599 | * locked on exit. |
| 600 | */ |
| 601 | int |
| 602 | union_copyup(un, docopy, cred, p) |
| 603 | struct union_node *un; |
| 604 | int docopy; |
| 605 | struct ucred *cred; |
| 606 | struct proc *p; |
| 607 | { |
| 608 | int error; |
| 609 | struct vnode *lvp, *uvp; |
| 610 | |
| 611 | error = union_vn_create(&uvp, un, p); |
| 612 | if (error) |
| 613 | return (error); |
| 614 | |
| 615 | /* at this point, uppervp is locked */ |
| 616 | union_newupper(un, uvp); |
| 617 | un->un_flags |= UN_ULOCK; |
| 618 | |
| 619 | lvp = un->un_lowervp; |
| 620 | |
| 621 | if (docopy) { |
| 622 | /* |
| 623 | * XX - should not ignore errors |
| 624 | * from VOP_CLOSE |
| 625 | */ |
| 626 | VOP_LOCK(lvp); |
| 627 | error = VOP_OPEN(lvp, FREAD, cred, p); |
| 628 | if (error == 0) { |
| 629 | error = union_copyfile(lvp, uvp, cred, p); |
| 630 | VOP_UNLOCK(lvp); |
| 631 | (void) VOP_CLOSE(lvp, FREAD); |
| 632 | } |
| 633 | #ifdef UNION_DIAGNOSTIC |
| 634 | if (error == 0) |
| 635 | uprintf("union: copied up %s\n", un->un_path); |
| 636 | #endif |
| 637 | |
| 638 | } |
| 639 | un->un_flags &= ~UN_ULOCK; |
| 640 | VOP_UNLOCK(uvp); |
| 641 | union_vn_close(uvp, FWRITE, cred, p); |
| 642 | VOP_LOCK(uvp); |
| 643 | un->un_flags |= UN_ULOCK; |
| 644 | |
| 645 | /* |
| 646 | * Subsequent IOs will go to the top layer, so |
| 647 | * call close on the lower vnode and open on the |
| 648 | * upper vnode to ensure that the filesystem keeps |
| 649 | * its references counts right. This doesn't do |
| 650 | * the right thing with (cred) and (FREAD) though. |
| 651 | * Ignoring error returns is not right, either. |
| 652 | */ |
| 653 | if (error == 0) { |
| 654 | int i; |
| 655 | |
| 656 | for (i = 0; i < un->un_openl; i++) { |
| 657 | (void) VOP_CLOSE(lvp, FREAD); |
| 658 | (void) VOP_OPEN(uvp, FREAD, cred, p); |
| 659 | } |
| 660 | un->un_openl = 0; |
| 661 | } |
| 662 | |
| 663 | return (error); |
| 664 | |
| 665 | } |
| 666 | |
| 667 | static int |
| 668 | union_relookup(um, dvp, vpp, cnp, cn, path, pathlen) |
| 669 | struct union_mount *um; |
| 670 | struct vnode *dvp; |
| 671 | struct vnode **vpp; |
| 672 | struct componentname *cnp; |
| 673 | struct componentname *cn; |
| 674 | char *path; |
| 675 | int pathlen; |
| 676 | { |
| 677 | int error; |
| 678 | |
| 679 | /* |
| 680 | * A new componentname structure must be faked up because |
| 681 | * there is no way to know where the upper level cnp came |
| 682 | * from or what it is being used for. This must duplicate |
| 683 | * some of the work done by NDINIT, some of the work done |
| 684 | * by namei, some of the work done by lookup and some of |
| 685 | * the work done by VOP_LOOKUP when given a CREATE flag. |
| 686 | * Conclusion: Horrible. |
| 687 | * |
| 688 | * The pathname buffer will be FREEed by VOP_MKDIR. |
| 689 | */ |
| 690 | cn->cn_namelen = pathlen; |
| 691 | cn->cn_pnbuf = malloc(cn->cn_namelen+1, M_NAMEI, M_WAITOK); |
| 692 | bcopy(path, cn->cn_pnbuf, cn->cn_namelen); |
| 693 | cn->cn_pnbuf[cn->cn_namelen] = '\0'; |
| 694 | |
| 695 | cn->cn_nameiop = CREATE; |
| 696 | cn->cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN); |
| 697 | cn->cn_proc = cnp->cn_proc; |
| 698 | if (um->um_op == UNMNT_ABOVE) |
| 699 | cn->cn_cred = cnp->cn_cred; |
| 700 | else |
| 701 | cn->cn_cred = um->um_cred; |
| 702 | cn->cn_nameptr = cn->cn_pnbuf; |
| 703 | cn->cn_hash = cnp->cn_hash; |
| 704 | cn->cn_consume = cnp->cn_consume; |
| 705 | |
| 706 | VREF(dvp); |
| 707 | error = relookup(dvp, vpp, cn); |
| 708 | if (!error) |
| 709 | vrele(dvp); |
| 710 | |
| 711 | return (error); |
| 712 | } |
| 713 | |
| 714 | /* |
| 715 | * Create a shadow directory in the upper layer. |
| 716 | * The new vnode is returned locked. |
| 717 | * |
| 718 | * (um) points to the union mount structure for access to the |
| 719 | * the mounting process's credentials. |
| 720 | * (dvp) is the directory in which to create the shadow directory. |
| 721 | * it is unlocked on entry and exit. |
| 722 | * (cnp) is the componentname to be created. |
| 723 | * (vpp) is the returned newly created shadow directory, which |
| 724 | * is returned locked. |
| 725 | */ |
| 726 | int |
| 727 | union_mkshadow(um, dvp, cnp, vpp) |
| 728 | struct union_mount *um; |
| 729 | struct vnode *dvp; |
| 730 | struct componentname *cnp; |
| 731 | struct vnode **vpp; |
| 732 | { |
| 733 | int error; |
| 734 | struct vattr va; |
| 735 | struct proc *p = cnp->cn_proc; |
| 736 | struct componentname cn; |
| 737 | |
| 738 | error = union_relookup(um, dvp, vpp, cnp, &cn, |
| 739 | cnp->cn_nameptr, cnp->cn_namelen); |
| 740 | if (error) |
| 741 | return (error); |
| 742 | |
| 743 | if (*vpp) { |
| 744 | VOP_ABORTOP(dvp, &cn); |
| 745 | VOP_UNLOCK(dvp); |
| 746 | vrele(*vpp); |
| 747 | *vpp = NULLVP; |
| 748 | return (EEXIST); |
| 749 | } |
| 750 | |
| 751 | /* |
| 752 | * policy: when creating the shadow directory in the |
| 753 | * upper layer, create it owned by the user who did |
| 754 | * the mount, group from parent directory, and mode |
| 755 | * 777 modified by umask (ie mostly identical to the |
| 756 | * mkdir syscall). (jsp, kb) |
| 757 | */ |
| 758 | |
| 759 | VATTR_NULL(&va); |
| 760 | va.va_type = VDIR; |
| 761 | va.va_mode = um->um_cmode; |
| 762 | |
| 763 | /* VOP_LEASE: dvp is locked */ |
| 764 | VOP_LEASE(dvp, p, cn.cn_cred, LEASE_WRITE); |
| 765 | |
| 766 | error = VOP_MKDIR(dvp, vpp, &cn, &va); |
| 767 | return (error); |
| 768 | } |
| 769 | |
| 770 | /* |
| 771 | * Create a whiteout entry in the upper layer. |
| 772 | * |
| 773 | * (um) points to the union mount structure for access to the |
| 774 | * the mounting process's credentials. |
| 775 | * (dvp) is the directory in which to create the whiteout. |
| 776 | * it is locked on entry and exit. |
| 777 | * (cnp) is the componentname to be created. |
| 778 | */ |
| 779 | int |
| 780 | union_mkwhiteout(um, dvp, cnp, path) |
| 781 | struct union_mount *um; |
| 782 | struct vnode *dvp; |
| 783 | struct componentname *cnp; |
| 784 | char *path; |
| 785 | { |
| 786 | int error; |
| 787 | struct vattr va; |
| 788 | struct proc *p = cnp->cn_proc; |
| 789 | struct vnode *wvp; |
| 790 | struct componentname cn; |
| 791 | |
| 792 | VOP_UNLOCK(dvp); |
| 793 | error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path)); |
| 794 | if (error) { |
| 795 | VOP_LOCK(dvp); |
| 796 | return (error); |
| 797 | } |
| 798 | |
| 799 | if (wvp) { |
| 800 | VOP_ABORTOP(dvp, &cn); |
| 801 | vrele(dvp); |
| 802 | vrele(wvp); |
| 803 | return (EEXIST); |
| 804 | } |
| 805 | |
| 806 | /* VOP_LEASE: dvp is locked */ |
| 807 | VOP_LEASE(dvp, p, p->p_ucred, LEASE_WRITE); |
| 808 | |
| 809 | error = VOP_WHITEOUT(dvp, &cn, CREATE); |
| 810 | if (error) |
| 811 | VOP_ABORTOP(dvp, &cn); |
| 812 | |
| 813 | vrele(dvp); |
| 814 | |
| 815 | return (error); |
| 816 | } |
| 817 | |
| 818 | /* |
| 819 | * union_vn_create: creates and opens a new shadow file |
| 820 | * on the upper union layer. this function is similar |
| 821 | * in spirit to calling vn_open but it avoids calling namei(). |
| 822 | * the problem with calling namei is that a) it locks too many |
| 823 | * things, and b) it doesn't start at the "right" directory, |
| 824 | * whereas relookup is told where to start. |
| 825 | */ |
| 826 | int |
| 827 | union_vn_create(vpp, un, p) |
| 828 | struct vnode **vpp; |
| 829 | struct union_node *un; |
| 830 | struct proc *p; |
| 831 | { |
| 832 | struct vnode *vp; |
| 833 | struct ucred *cred = p->p_ucred; |
| 834 | struct vattr vat; |
| 835 | struct vattr *vap = &vat; |
| 836 | int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL); |
| 837 | int error; |
| 838 | int cmode = UN_FILEMODE & ~p->p_fd->fd_cmask; |
| 839 | char *cp; |
| 840 | struct componentname cn; |
| 841 | |
| 842 | *vpp = NULLVP; |
| 843 | |
| 844 | /* |
| 845 | * Build a new componentname structure (for the same |
| 846 | * reasons outlines in union_mkshadow). |
| 847 | * The difference here is that the file is owned by |
| 848 | * the current user, rather than by the person who |
| 849 | * did the mount, since the current user needs to be |
| 850 | * able to write the file (that's why it is being |
| 851 | * copied in the first place). |
| 852 | */ |
| 853 | cn.cn_namelen = strlen(un->un_path); |
| 854 | cn.cn_pnbuf = (caddr_t) malloc(cn.cn_namelen, M_NAMEI, M_WAITOK); |
| 855 | bcopy(un->un_path, cn.cn_pnbuf, cn.cn_namelen+1); |
| 856 | cn.cn_nameiop = CREATE; |
| 857 | cn.cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN); |
| 858 | cn.cn_proc = p; |
| 859 | cn.cn_cred = p->p_ucred; |
| 860 | cn.cn_nameptr = cn.cn_pnbuf; |
| 861 | cn.cn_hash = un->un_hash; |
| 862 | cn.cn_consume = 0; |
| 863 | |
| 864 | VREF(un->un_dirvp); |
| 865 | if (error = relookup(un->un_dirvp, &vp, &cn)) |
| 866 | return (error); |
| 867 | vrele(un->un_dirvp); |
| 868 | |
| 869 | if (vp) { |
| 870 | VOP_ABORTOP(un->un_dirvp, &cn); |
| 871 | if (un->un_dirvp == vp) |
| 872 | vrele(un->un_dirvp); |
| 873 | else |
| 874 | vput(un->un_dirvp); |
| 875 | vrele(vp); |
| 876 | return (EEXIST); |
| 877 | } |
| 878 | |
| 879 | /* |
| 880 | * Good - there was no race to create the file |
| 881 | * so go ahead and create it. The permissions |
| 882 | * on the file will be 0666 modified by the |
| 883 | * current user's umask. Access to the file, while |
| 884 | * it is unioned, will require access to the top *and* |
| 885 | * bottom files. Access when not unioned will simply |
| 886 | * require access to the top-level file. |
| 887 | * TODO: confirm choice of access permissions. |
| 888 | */ |
| 889 | VATTR_NULL(vap); |
| 890 | vap->va_type = VREG; |
| 891 | vap->va_mode = cmode; |
| 892 | VOP_LEASE(un->un_dirvp, p, cred, LEASE_WRITE); |
| 893 | if (error = VOP_CREATE(un->un_dirvp, &vp, &cn, vap)) |
| 894 | return (error); |
| 895 | |
| 896 | if (error = VOP_OPEN(vp, fmode, cred, p)) { |
| 897 | vput(vp); |
| 898 | return (error); |
| 899 | } |
| 900 | |
| 901 | vp->v_writecount++; |
| 902 | *vpp = vp; |
| 903 | return (0); |
| 904 | } |
| 905 | |
| 906 | int |
| 907 | union_vn_close(vp, fmode, cred, p) |
| 908 | struct vnode *vp; |
| 909 | int fmode; |
| 910 | struct ucred *cred; |
| 911 | struct proc *p; |
| 912 | { |
| 913 | |
| 914 | if (fmode & FWRITE) |
| 915 | --vp->v_writecount; |
| 916 | return (VOP_CLOSE(vp, fmode)); |
| 917 | } |
| 918 | |
| 919 | void |
| 920 | union_removed_upper(un) |
| 921 | struct union_node *un; |
| 922 | { |
| 923 | |
| 924 | if (un->un_flags & UN_ULOCK) { |
| 925 | un->un_flags &= ~UN_ULOCK; |
| 926 | VOP_UNLOCK(un->un_uppervp); |
| 927 | } |
| 928 | |
| 929 | if (un->un_flags & UN_CACHED) { |
| 930 | un->un_flags &= ~UN_CACHED; |
| 931 | LIST_REMOVE(un, un_cache); |
| 932 | } |
| 933 | } |
| 934 | |
| 935 | #if 0 |
| 936 | struct vnode * |
| 937 | union_lowervp(vp) |
| 938 | struct vnode *vp; |
| 939 | { |
| 940 | struct union_node *un = VTOUNION(vp); |
| 941 | |
| 942 | if ((un->un_lowervp != NULLVP) && |
| 943 | (vp->v_type == un->un_lowervp->v_type)) { |
| 944 | if (vget(un->un_lowervp, 0) == 0) |
| 945 | return (un->un_lowervp); |
| 946 | } |
| 947 | |
| 948 | return (NULLVP); |
| 949 | } |
| 950 | #endif |
| 951 | |
| 952 | /* |
| 953 | * determine whether a whiteout is needed |
| 954 | * during a remove/rmdir operation. |
| 955 | */ |
| 956 | int |
| 957 | union_dowhiteout(un, cred, p) |
| 958 | struct union_node *un; |
| 959 | struct ucred *cred; |
| 960 | struct proc *p; |
| 961 | { |
| 962 | struct vattr va; |
| 963 | |
| 964 | if (un->un_lowervp != NULLVP) |
| 965 | return (1); |
| 966 | |
| 967 | if (VOP_GETATTR(un->un_uppervp, &va, cred, p) == 0 && |
| 968 | (va.va_flags & OPAQUE)) |
| 969 | return (1); |
| 970 | |
| 971 | return (0); |
| 972 | } |
| 973 | |
| 974 | static void |
| 975 | union_dircache_r(vp, vppp, cntp) |
| 976 | struct vnode *vp; |
| 977 | struct vnode ***vppp; |
| 978 | int *cntp; |
| 979 | { |
| 980 | struct union_node *un; |
| 981 | |
| 982 | if (vp->v_op != union_vnodeop_p) { |
| 983 | if (vppp) { |
| 984 | VREF(vp); |
| 985 | *(*vppp)++ = vp; |
| 986 | if (--(*cntp) == 0) |
| 987 | panic("union: dircache table too small"); |
| 988 | } else { |
| 989 | (*cntp)++; |
| 990 | } |
| 991 | |
| 992 | return; |
| 993 | } |
| 994 | |
| 995 | un = VTOUNION(vp); |
| 996 | if (un->un_uppervp != NULLVP) |
| 997 | union_dircache_r(un->un_uppervp, vppp, cntp); |
| 998 | if (un->un_lowervp != NULLVP) |
| 999 | union_dircache_r(un->un_lowervp, vppp, cntp); |
| 1000 | } |
| 1001 | |
| 1002 | struct vnode * |
| 1003 | union_dircache(vp) |
| 1004 | struct vnode *vp; |
| 1005 | { |
| 1006 | int cnt; |
| 1007 | struct vnode *nvp; |
| 1008 | struct vnode **vpp; |
| 1009 | struct vnode **dircache = VTOUNION(vp)->un_dircache; |
| 1010 | struct union_node *un; |
| 1011 | int error; |
| 1012 | |
| 1013 | if (dircache == 0) { |
| 1014 | cnt = 0; |
| 1015 | union_dircache_r(vp, 0, &cnt); |
| 1016 | cnt++; |
| 1017 | dircache = (struct vnode **) |
| 1018 | malloc(cnt * sizeof(struct vnode *), |
| 1019 | M_TEMP, M_WAITOK); |
| 1020 | vpp = dircache; |
| 1021 | union_dircache_r(vp, &vpp, &cnt); |
| 1022 | *vpp = NULLVP; |
| 1023 | vpp = dircache + 1; |
| 1024 | } else { |
| 1025 | vpp = dircache; |
| 1026 | do { |
| 1027 | if (*vpp++ == VTOUNION(vp)->un_uppervp) |
| 1028 | break; |
| 1029 | } while (*vpp != NULLVP); |
| 1030 | } |
| 1031 | |
| 1032 | if (*vpp == NULLVP) |
| 1033 | return (NULLVP); |
| 1034 | |
| 1035 | VOP_LOCK(*vpp); |
| 1036 | VREF(*vpp); |
| 1037 | error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, 0, *vpp, NULLVP, 0); |
| 1038 | if (error) |
| 1039 | return (NULLVP); |
| 1040 | VTOUNION(vp)->un_dircache = 0; |
| 1041 | un = VTOUNION(nvp); |
| 1042 | un->un_dircache = dircache; |
| 1043 | |
| 1044 | return (nvp); |
| 1045 | } |