rumpfs.c revision 1.64
1 /* $NetBSD: rumpfs.c,v 1.64 2010/09/06 14:50:34 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 2009 Antti Kantee. All Rights Reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 16 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 18 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 21 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __KERNEL_RCSID(0, "$NetBSD: rumpfs.c,v 1.64 2010/09/06 14:50:34 pooka Exp $"); 30 31 #include <sys/param.h> 32 #include <sys/atomic.h> 33 #include <sys/dirent.h> 34 #include <sys/errno.h> 35 #include <sys/filedesc.h> 36 #include <sys/fcntl.h> 37 #include <sys/kauth.h> 38 #include <sys/malloc.h> 39 #include <sys/module.h> 40 #include <sys/mount.h> 41 #include <sys/namei.h> 42 #include <sys/lock.h> 43 #include <sys/lockf.h> 44 #include <sys/queue.h> 45 #include <sys/stat.h> 46 #include <sys/syscallargs.h> 47 #include <sys/vnode.h> 48 49 #include <miscfs/fifofs/fifo.h> 50 #include <miscfs/specfs/specdev.h> 51 #include <miscfs/genfs/genfs.h> 52 53 #include <rump/rumpuser.h> 54 55 #include "rump_private.h" 56 #include "rump_vfs_private.h" 57 58 static int rump_vop_lookup(void *); 59 static int rump_vop_getattr(void *); 60 static int rump_vop_mkdir(void *); 61 static int rump_vop_rmdir(void *); 62 static int rump_vop_mknod(void *); 63 static int rump_vop_create(void *); 64 static int rump_vop_inactive(void *); 65 static int rump_vop_reclaim(void *); 66 static int rump_vop_success(void *); 67 static int rump_vop_readdir(void *); 68 static int rump_vop_spec(void *); 69 static int rump_vop_read(void *); 70 static int rump_vop_write(void *); 71 static int rump_vop_open(void *); 72 static int rump_vop_symlink(void *); 73 static int rump_vop_readlink(void *); 74 static int rump_vop_whiteout(void *); 75 76 int (**fifo_vnodeop_p)(void *); 77 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = { 78 { &vop_default_desc, vn_default_error }, 79 { NULL, NULL } 80 }; 81 const struct vnodeopv_desc fifo_vnodeop_opv_desc = 82 { &fifo_vnodeop_p, fifo_vnodeop_entries }; 83 84 int (**rump_vnodeop_p)(void *); 85 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = { 86 { &vop_default_desc, vn_default_error }, 87 { &vop_lookup_desc, rump_vop_lookup }, 88 { &vop_getattr_desc, rump_vop_getattr }, 89 { &vop_mkdir_desc, rump_vop_mkdir }, 90 { &vop_rmdir_desc, rump_vop_rmdir }, 91 { &vop_mknod_desc, rump_vop_mknod }, 92 { &vop_create_desc, rump_vop_create }, 93 { &vop_symlink_desc, rump_vop_symlink }, 94 { &vop_readlink_desc, rump_vop_readlink }, 95 { &vop_access_desc, rump_vop_success }, 96 { &vop_readdir_desc, rump_vop_readdir }, 97 { &vop_read_desc, rump_vop_read }, 98 { &vop_write_desc, rump_vop_write }, 99 { &vop_open_desc, rump_vop_open }, 100 { &vop_seek_desc, genfs_seek }, 101 { &vop_putpages_desc, genfs_null_putpages }, 102 { &vop_whiteout_desc, rump_vop_whiteout }, 103 { &vop_fsync_desc, rump_vop_success }, 104 { &vop_lock_desc, genfs_lock }, 105 { &vop_unlock_desc, genfs_unlock }, 106 { &vop_islocked_desc, genfs_islocked }, 107 { &vop_inactive_desc, rump_vop_inactive }, 108 { &vop_reclaim_desc, rump_vop_reclaim }, 109 { &vop_remove_desc, genfs_eopnotsupp }, 110 { &vop_link_desc, genfs_eopnotsupp }, 111 { NULL, NULL } 112 }; 113 const struct vnodeopv_desc rump_vnodeop_opv_desc = 114 { &rump_vnodeop_p, rump_vnodeop_entries }; 115 116 int (**rump_specop_p)(void *); 117 const struct vnodeopv_entry_desc rump_specop_entries[] = { 118 { &vop_default_desc, rump_vop_spec }, 119 { NULL, NULL } 120 }; 121 const struct vnodeopv_desc rump_specop_opv_desc = 122 { &rump_specop_p, rump_specop_entries }; 123 124 const struct vnodeopv_desc * const rump_opv_descs[] = { 125 &rump_vnodeop_opv_desc, 126 &rump_specop_opv_desc, 127 NULL 128 }; 129 130 #define RUMPFS_WHITEOUT NULL 131 #define RDENT_ISWHITEOUT(rdp) (rdp->rd_node == RUMPFS_WHITEOUT) 132 struct rumpfs_dent { 133 char *rd_name; 134 int rd_namelen; 135 struct rumpfs_node *rd_node; 136 137 LIST_ENTRY(rumpfs_dent) rd_entries; 138 }; 139 140 struct rumpfs_node { 141 struct vattr rn_va; 142 struct vnode *rn_vp; 143 char *rn_hostpath; 144 int rn_flags; 145 146 union { 147 struct { /* VREG */ 148 int readfd; 149 int writefd; 150 uint64_t offset; 151 } reg; 152 struct { /* VDIR */ 153 LIST_HEAD(, rumpfs_dent) dents; 154 struct rumpfs_node *parent; 155 int flags; 156 } dir; 157 struct { 158 char *target; 159 size_t len; 160 } link; 161 } rn_u; 162 }; 163 #define rn_readfd rn_u.reg.readfd 164 #define rn_writefd rn_u.reg.writefd 165 #define rn_offset rn_u.reg.offset 166 #define rn_dir rn_u.dir.dents 167 #define rn_parent rn_u.dir.parent 168 #define rn_linktarg rn_u.link.target 169 #define rn_linklen rn_u.link.len 170 171 #define RUMPNODE_CANRECLAIM 0x01 172 #define RUMPNODE_DIR_ET 0x02 173 #define RUMPNODE_DIR_ETSUBS 0x04 174 175 struct rumpfs_mount { 176 struct vnode *rfsmp_rvp; 177 }; 178 179 static struct rumpfs_node *makeprivate(enum vtype, dev_t, off_t); 180 181 /* 182 * Extra Terrestrial stuff. We map a given key (pathname) to a file on 183 * the host FS. ET phones home only from the root node of rumpfs. 184 * 185 * When an etfs node is removed, a vnode potentially behind it is not 186 * immediately recycled. 187 */ 188 189 struct etfs { 190 char et_key[MAXPATHLEN]; 191 size_t et_keylen; 192 bool et_prefixkey; 193 bool et_removing; 194 devminor_t et_blkmin; 195 196 LIST_ENTRY(etfs) et_entries; 197 198 struct rumpfs_node *et_rn; 199 }; 200 static kmutex_t etfs_lock; 201 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list); 202 203 static enum vtype 204 ettype_to_vtype(enum rump_etfs_type et) 205 { 206 enum vtype vt; 207 208 switch (et) { 209 case RUMP_ETFS_REG: 210 vt = VREG; 211 break; 212 case RUMP_ETFS_BLK: 213 vt = VBLK; 214 break; 215 case RUMP_ETFS_CHR: 216 vt = VCHR; 217 break; 218 case RUMP_ETFS_DIR: 219 vt = VDIR; 220 break; 221 case RUMP_ETFS_DIR_SUBDIRS: 222 vt = VDIR; 223 break; 224 default: 225 panic("invalid et type: %d", et); 226 } 227 228 return vt; 229 } 230 231 static enum vtype 232 hft_to_vtype(int hft) 233 { 234 enum vtype vt; 235 236 switch (hft) { 237 case RUMPUSER_FT_OTHER: 238 vt = VNON; 239 break; 240 case RUMPUSER_FT_DIR: 241 vt = VDIR; 242 break; 243 case RUMPUSER_FT_REG: 244 vt = VREG; 245 break; 246 case RUMPUSER_FT_BLK: 247 vt = VBLK; 248 break; 249 case RUMPUSER_FT_CHR: 250 vt = VCHR; 251 break; 252 default: 253 vt = VNON; 254 break; 255 } 256 257 return vt; 258 } 259 260 static bool 261 etfs_find(const char *key, struct etfs **etp, bool forceprefix) 262 { 263 struct etfs *et; 264 size_t keylen = strlen(key); 265 266 KASSERT(mutex_owned(&etfs_lock)); 267 268 LIST_FOREACH(et, &etfs_list, et_entries) { 269 if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix) 270 && strncmp(key, et->et_key, et->et_keylen) == 0) { 271 if (etp) 272 *etp = et; 273 return true; 274 } 275 } 276 277 return false; 278 } 279 280 #define REGDIR(ftype) \ 281 ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS) 282 static int 283 doregister(const char *key, const char *hostpath, 284 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 285 { 286 struct etfs *et; 287 struct rumpfs_node *rn; 288 uint64_t fsize; 289 dev_t rdev = NODEV; 290 devminor_t dmin = -1; 291 int hft, error; 292 293 if (rumpuser_getfileinfo(hostpath, &fsize, &hft, &error)) 294 return error; 295 296 /* etfs directory requires a directory on the host */ 297 if (REGDIR(ftype)) { 298 if (hft != RUMPUSER_FT_DIR) 299 return ENOTDIR; 300 if (begin != 0) 301 return EISDIR; 302 if (size != RUMP_ETFS_SIZE_ENDOFF) 303 return EISDIR; 304 size = fsize; 305 } else { 306 if (begin > fsize) 307 return EINVAL; 308 if (size == RUMP_ETFS_SIZE_ENDOFF) 309 size = fsize - begin; 310 if (begin + size > fsize) 311 return EINVAL; 312 } 313 314 if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) { 315 error = rumpblk_register(hostpath, &dmin, begin, size); 316 if (error != 0) { 317 return error; 318 } 319 rdev = makedev(RUMPBLK_DEVMAJOR, dmin); 320 } 321 322 et = kmem_alloc(sizeof(*et), KM_SLEEP); 323 strcpy(et->et_key, key); 324 et->et_keylen = strlen(et->et_key); 325 et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), rdev, size); 326 et->et_removing = false; 327 et->et_blkmin = dmin; 328 329 if (ftype == RUMP_ETFS_REG || REGDIR(ftype) || et->et_blkmin != -1) { 330 size_t len = strlen(hostpath)+1; 331 332 rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 333 memcpy(rn->rn_hostpath, hostpath, len); 334 rn->rn_offset = begin; 335 } 336 337 if (REGDIR(ftype)) { 338 rn->rn_flags |= RUMPNODE_DIR_ET; 339 et->et_prefixkey = true; 340 } else { 341 et->et_prefixkey = false; 342 } 343 344 if (ftype == RUMP_ETFS_DIR_SUBDIRS) 345 rn->rn_flags |= RUMPNODE_DIR_ETSUBS; 346 347 mutex_enter(&etfs_lock); 348 if (etfs_find(key, NULL, REGDIR(ftype))) { 349 mutex_exit(&etfs_lock); 350 if (et->et_blkmin != -1) 351 rumpblk_deregister(hostpath); 352 if (et->et_rn->rn_hostpath != NULL) 353 free(et->et_rn->rn_hostpath, M_TEMP); 354 kmem_free(et->et_rn, sizeof(*et->et_rn)); 355 kmem_free(et, sizeof(*et)); 356 return EEXIST; 357 } 358 LIST_INSERT_HEAD(&etfs_list, et, et_entries); 359 mutex_exit(&etfs_lock); 360 361 return 0; 362 } 363 #undef REGDIR 364 365 int 366 rump_etfs_register(const char *key, const char *hostpath, 367 enum rump_etfs_type ftype) 368 { 369 370 return doregister(key, hostpath, ftype, 0, RUMP_ETFS_SIZE_ENDOFF); 371 } 372 373 int 374 rump_etfs_register_withsize(const char *key, const char *hostpath, 375 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 376 { 377 378 return doregister(key, hostpath, ftype, begin, size); 379 } 380 381 /* remove etfs mapping. caller's responsibility to make sure it's not in use */ 382 int 383 rump_etfs_remove(const char *key) 384 { 385 struct etfs *et; 386 size_t keylen = strlen(key); 387 int rv; 388 389 mutex_enter(&etfs_lock); 390 LIST_FOREACH(et, &etfs_list, et_entries) { 391 if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) { 392 if (et->et_removing) 393 et = NULL; 394 else 395 et->et_removing = true; 396 break; 397 } 398 } 399 mutex_exit(&etfs_lock); 400 if (!et) 401 return ENOENT; 402 403 /* 404 * ok, we know what we want to remove and have signalled there 405 * actually are men at work. first, unregister from rumpblk 406 */ 407 if (et->et_blkmin != -1) { 408 rv = rumpblk_deregister(et->et_rn->rn_hostpath); 409 } else { 410 rv = 0; 411 } 412 KASSERT(rv == 0); 413 414 /* then do the actual removal */ 415 mutex_enter(&etfs_lock); 416 LIST_REMOVE(et, et_entries); 417 mutex_exit(&etfs_lock); 418 419 /* node is unreachable, safe to nuke all device copies */ 420 if (et->et_blkmin != -1) 421 vdevgone(RUMPBLK_DEVMAJOR, et->et_blkmin, et->et_blkmin, VBLK); 422 423 if (et->et_rn->rn_hostpath != NULL) 424 free(et->et_rn->rn_hostpath, M_TEMP); 425 kmem_free(et->et_rn, sizeof(*et->et_rn)); 426 kmem_free(et, sizeof(*et)); 427 428 return 0; 429 } 430 431 /* 432 * rumpfs 433 */ 434 435 #define INO_WHITEOUT 1 436 static int lastino = 2; 437 static kmutex_t reclock; 438 439 static struct rumpfs_node * 440 makeprivate(enum vtype vt, dev_t rdev, off_t size) 441 { 442 struct rumpfs_node *rn; 443 struct vattr *va; 444 struct timespec ts; 445 446 rn = kmem_zalloc(sizeof(*rn), KM_SLEEP); 447 448 switch (vt) { 449 case VDIR: 450 LIST_INIT(&rn->rn_dir); 451 break; 452 case VREG: 453 rn->rn_readfd = -1; 454 rn->rn_writefd = -1; 455 break; 456 default: 457 break; 458 } 459 460 nanotime(&ts); 461 462 va = &rn->rn_va; 463 va->va_type = vt; 464 va->va_mode = 0755; 465 if (vt == VDIR) 466 va->va_nlink = 2; 467 else 468 va->va_nlink = 1; 469 va->va_uid = 0; 470 va->va_gid = 0; 471 va->va_fsid = 472 va->va_fileid = atomic_inc_uint_nv(&lastino); 473 va->va_size = size; 474 va->va_blocksize = 512; 475 va->va_atime = ts; 476 va->va_mtime = ts; 477 va->va_ctime = ts; 478 va->va_birthtime = ts; 479 va->va_gen = 0; 480 va->va_flags = 0; 481 va->va_rdev = rdev; 482 va->va_bytes = 512; 483 va->va_filerev = 0; 484 va->va_vaflags = 0; 485 486 return rn; 487 } 488 489 static int 490 makevnode(struct mount *mp, struct rumpfs_node *rn, struct vnode **vpp) 491 { 492 struct vnode *vp; 493 int (**vpops)(void *); 494 struct vattr *va = &rn->rn_va; 495 int rv; 496 497 KASSERT(!mutex_owned(&reclock)); 498 499 if (va->va_type == VCHR || va->va_type == VBLK) { 500 vpops = rump_specop_p; 501 } else { 502 vpops = rump_vnodeop_p; 503 } 504 if (vpops != rump_specop_p && va->va_type != VDIR 505 && !(va->va_type == VREG && rn->rn_hostpath != NULL) 506 && va->va_type != VSOCK && va->va_type != VLNK) 507 return EOPNOTSUPP; 508 509 rv = getnewvnode(VT_RUMP, mp, vpops, &vp); 510 if (rv) 511 return rv; 512 513 vp->v_size = vp->v_writesize = va->va_size; 514 vp->v_type = va->va_type; 515 516 if (vpops == rump_specop_p) { 517 spec_node_init(vp, va->va_rdev); 518 } 519 vp->v_data = rn; 520 521 vn_lock(vp, LK_RETRY | LK_EXCLUSIVE); 522 mutex_enter(&reclock); 523 rn->rn_vp = vp; 524 mutex_exit(&reclock); 525 526 *vpp = vp; 527 528 return 0; 529 } 530 531 532 static void 533 makedir(struct rumpfs_node *rnd, 534 struct componentname *cnp, struct rumpfs_node *rn) 535 { 536 struct rumpfs_dent *rdent; 537 538 rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP); 539 rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP); 540 rdent->rd_node = rn; 541 strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1); 542 rdent->rd_namelen = strlen(rdent->rd_name); 543 544 LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries); 545 } 546 547 static void 548 freedir(struct rumpfs_node *rnd, struct componentname *cnp) 549 { 550 struct rumpfs_dent *rd = NULL; 551 552 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 553 if (rd->rd_namelen == cnp->cn_namelen && 554 strncmp(rd->rd_name, cnp->cn_nameptr, 555 cnp->cn_namelen) == 0) 556 break; 557 } 558 if (rd == NULL) 559 panic("could not find directory entry: %s", cnp->cn_nameptr); 560 561 LIST_REMOVE(rd, rd_entries); 562 kmem_free(rd->rd_name, rd->rd_namelen+1); 563 kmem_free(rd, sizeof(*rd)); 564 } 565 566 /* 567 * Simple lookup for rump file systems. 568 * 569 * uhm, this is twisted. C F C C, hope of C C F C looming 570 */ 571 static int 572 rump_vop_lookup(void *v) 573 { 574 struct vop_lookup_args /* { 575 struct vnode *a_dvp; 576 struct vnode **a_vpp; 577 struct componentname *a_cnp; 578 }; */ *ap = v; 579 struct componentname *cnp = ap->a_cnp; 580 struct vnode *dvp = ap->a_dvp; 581 struct vnode **vpp = ap->a_vpp; 582 struct vnode *vp; 583 struct rumpfs_node *rnd = dvp->v_data, *rn; 584 struct rumpfs_dent *rd = NULL; 585 struct etfs *et; 586 bool dotdot = (cnp->cn_flags & ISDOTDOT) != 0; 587 int rv = 0; 588 589 /* check for dot, return directly if the case */ 590 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 591 vref(dvp); 592 *vpp = dvp; 593 return 0; 594 } 595 596 /* we handle only some "non-special" cases */ 597 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME))) 598 return EOPNOTSUPP; 599 600 /* check for etfs */ 601 if (dvp == rootvnode && cnp->cn_nameiop == LOOKUP) { 602 bool found; 603 mutex_enter(&etfs_lock); 604 found = etfs_find(cnp->cn_pnbuf, &et, false); 605 mutex_exit(&etfs_lock); 606 607 if (found) { 608 char *offset; 609 610 offset = strstr(cnp->cn_pnbuf, et->et_key); 611 KASSERT(offset); 612 613 rn = et->et_rn; 614 cnp->cn_consume += et->et_keylen 615 - (cnp->cn_nameptr - offset) - cnp->cn_namelen; 616 if (rn->rn_va.va_type != VDIR) 617 cnp->cn_flags &= ~REQUIREDIR; 618 goto getvnode; 619 } 620 } 621 622 if (rnd->rn_flags & RUMPNODE_DIR_ET) { 623 uint64_t fsize; 624 char *newpath; 625 size_t newpathlen; 626 int hft, error; 627 628 if (dotdot) 629 return EOPNOTSUPP; 630 631 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1; 632 newpath = malloc(newpathlen, M_TEMP, M_WAITOK); 633 634 strlcpy(newpath, rnd->rn_hostpath, newpathlen); 635 strlcat(newpath, "/", newpathlen); 636 strlcat(newpath, cnp->cn_nameptr, newpathlen); 637 638 if (rumpuser_getfileinfo(newpath, &fsize, &hft, &error)) { 639 free(newpath, M_TEMP); 640 return error; 641 } 642 643 /* allow only dirs and regular files */ 644 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) { 645 free(newpath, M_TEMP); 646 return ENOENT; 647 } 648 649 rn = makeprivate(hft_to_vtype(hft), NODEV, fsize); 650 rn->rn_flags |= RUMPNODE_CANRECLAIM; 651 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) { 652 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS; 653 } 654 rn->rn_hostpath = newpath; 655 656 goto getvnode; 657 } else { 658 if (dotdot) { 659 rn = rnd->rn_parent; 660 goto getvnode; 661 } else { 662 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 663 if (rd->rd_namelen == cnp->cn_namelen && 664 strncmp(rd->rd_name, cnp->cn_nameptr, 665 cnp->cn_namelen) == 0) 666 break; 667 } 668 } 669 } 670 671 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE)) 672 return ENOENT; 673 674 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 675 cnp->cn_flags |= SAVENAME; 676 return EJUSTRETURN; 677 } 678 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == DELETE) 679 cnp->cn_flags |= SAVENAME; 680 681 rn = rd->rd_node; 682 683 getvnode: 684 KASSERT(rn); 685 if (dotdot) 686 VOP_UNLOCK(dvp); 687 mutex_enter(&reclock); 688 if ((vp = rn->rn_vp)) { 689 mutex_enter(&vp->v_interlock); 690 mutex_exit(&reclock); 691 if (vget(vp, LK_EXCLUSIVE)) { 692 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 693 goto getvnode; 694 } 695 *vpp = vp; 696 } else { 697 mutex_exit(&reclock); 698 rv = makevnode(dvp->v_mount, rn, vpp); 699 } 700 if (dotdot) 701 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 702 703 return rv; 704 } 705 706 static int 707 rump_vop_getattr(void *v) 708 { 709 struct vop_getattr_args /* { 710 struct vnode *a_vp; 711 struct vattr *a_vap; 712 kauth_cred_t a_cred; 713 } */ *ap = v; 714 struct rumpfs_node *rn = ap->a_vp->v_data; 715 716 memcpy(ap->a_vap, &rn->rn_va, sizeof(struct vattr)); 717 return 0; 718 } 719 720 static int 721 rump_vop_mkdir(void *v) 722 { 723 struct vop_mkdir_args /* { 724 struct vnode *a_dvp; 725 struct vnode **a_vpp; 726 struct componentname *a_cnp; 727 struct vattr *a_vap; 728 }; */ *ap = v; 729 struct vnode *dvp = ap->a_dvp; 730 struct vnode **vpp = ap->a_vpp; 731 struct componentname *cnp = ap->a_cnp; 732 struct rumpfs_node *rnd = dvp->v_data, *rn; 733 int rv = 0; 734 735 rn = makeprivate(VDIR, NODEV, DEV_BSIZE); 736 rn->rn_parent = rnd; 737 rv = makevnode(dvp->v_mount, rn, vpp); 738 if (rv) 739 goto out; 740 741 makedir(rnd, cnp, rn); 742 743 out: 744 PNBUF_PUT(cnp->cn_pnbuf); 745 vput(dvp); 746 return rv; 747 } 748 749 static int 750 rump_vop_rmdir(void *v) 751 { 752 struct vop_rmdir_args /* { 753 struct vnode *a_dvp; 754 struct vnode *a_vp; 755 struct componentname *a_cnp; 756 }; */ *ap = v; 757 struct vnode *dvp = ap->a_dvp; 758 struct vnode *vp = ap->a_vp; 759 struct componentname *cnp = ap->a_cnp; 760 struct rumpfs_node *rnd = dvp->v_data; 761 struct rumpfs_node *rn = vp->v_data; 762 int rv = 0; 763 764 if (!LIST_EMPTY(&rn->rn_dir)) { 765 rv = ENOTEMPTY; 766 goto out; 767 } 768 769 freedir(rnd, cnp); 770 rn->rn_flags |= RUMPNODE_CANRECLAIM; 771 772 out: 773 PNBUF_PUT(cnp->cn_pnbuf); 774 vput(dvp); 775 vput(vp); 776 777 return rv; 778 } 779 780 static int 781 rump_vop_mknod(void *v) 782 { 783 struct vop_mknod_args /* { 784 struct vnode *a_dvp; 785 struct vnode **a_vpp; 786 struct componentname *a_cnp; 787 struct vattr *a_vap; 788 }; */ *ap = v; 789 struct vnode *dvp = ap->a_dvp; 790 struct vnode **vpp = ap->a_vpp; 791 struct componentname *cnp = ap->a_cnp; 792 struct vattr *va = ap->a_vap; 793 struct rumpfs_node *rnd = dvp->v_data, *rn; 794 int rv; 795 796 rn = makeprivate(va->va_type, va->va_rdev, DEV_BSIZE); 797 rv = makevnode(dvp->v_mount, rn, vpp); 798 if (rv) 799 goto out; 800 801 makedir(rnd, cnp, rn); 802 803 out: 804 PNBUF_PUT(cnp->cn_pnbuf); 805 vput(dvp); 806 return rv; 807 } 808 809 static int 810 rump_vop_create(void *v) 811 { 812 struct vop_create_args /* { 813 struct vnode *a_dvp; 814 struct vnode **a_vpp; 815 struct componentname *a_cnp; 816 struct vattr *a_vap; 817 }; */ *ap = v; 818 struct vnode *dvp = ap->a_dvp; 819 struct vnode **vpp = ap->a_vpp; 820 struct componentname *cnp = ap->a_cnp; 821 struct vattr *va = ap->a_vap; 822 struct rumpfs_node *rnd = dvp->v_data, *rn; 823 int rv; 824 825 if (va->va_type != VSOCK) { 826 rv = EOPNOTSUPP; 827 goto out; 828 } 829 rn = makeprivate(VSOCK, NODEV, DEV_BSIZE); 830 rv = makevnode(dvp->v_mount, rn, vpp); 831 if (rv) 832 goto out; 833 834 makedir(rnd, cnp, rn); 835 836 out: 837 PNBUF_PUT(cnp->cn_pnbuf); 838 vput(dvp); 839 return rv; 840 } 841 842 static int 843 rump_vop_symlink(void *v) 844 { 845 struct vop_symlink_args /* { 846 struct vnode *a_dvp; 847 struct vnode **a_vpp; 848 struct componentname *a_cnp; 849 struct vattr *a_vap; 850 char *a_target; 851 }; */ *ap = v; 852 struct vnode *dvp = ap->a_dvp; 853 struct vnode **vpp = ap->a_vpp; 854 struct componentname *cnp = ap->a_cnp; 855 struct rumpfs_node *rnd = dvp->v_data, *rn; 856 const char *target = ap->a_target; 857 size_t linklen; 858 int rv; 859 860 linklen = strlen(target); 861 KASSERT(linklen < MAXPATHLEN); 862 rn = makeprivate(VLNK, NODEV, linklen); 863 rv = makevnode(dvp->v_mount, rn, vpp); 864 if (rv) 865 goto out; 866 867 makedir(rnd, cnp, rn); 868 869 KASSERT(linklen < MAXPATHLEN); 870 rn->rn_linktarg = PNBUF_GET(); 871 rn->rn_linklen = linklen; 872 strcpy(rn->rn_linktarg, target); 873 874 out: 875 vput(dvp); 876 return rv; 877 } 878 879 static int 880 rump_vop_readlink(void *v) 881 { 882 struct vop_readlink_args /* { 883 struct vnode *a_vp; 884 struct uio *a_uio; 885 kauth_cred_t a_cred; 886 }; */ *ap = v; 887 struct vnode *vp = ap->a_vp; 888 struct rumpfs_node *rn = vp->v_data; 889 struct uio *uio = ap->a_uio; 890 891 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio); 892 } 893 894 static int 895 rump_vop_whiteout(void *v) 896 { 897 struct vop_whiteout_args /* { 898 struct vnode *a_dvp; 899 struct componentname *a_cnp; 900 int a_flags; 901 } */ *ap = v; 902 struct vnode *dvp = ap->a_dvp; 903 struct rumpfs_node *rnd = dvp->v_data; 904 struct componentname *cnp = ap->a_cnp; 905 int flags = ap->a_flags; 906 907 switch (flags) { 908 case LOOKUP: 909 break; 910 case CREATE: 911 makedir(rnd, cnp, RUMPFS_WHITEOUT); 912 break; 913 case DELETE: 914 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */ 915 freedir(rnd, cnp); 916 break; 917 default: 918 panic("unknown whiteout op %d", flags); 919 } 920 921 return 0; 922 } 923 924 static int 925 rump_vop_open(void *v) 926 { 927 struct vop_open_args /* { 928 struct vnode *a_vp; 929 int a_mode; 930 kauth_cred_t a_cred; 931 } */ *ap = v; 932 struct vnode *vp = ap->a_vp; 933 struct rumpfs_node *rn = vp->v_data; 934 int mode = ap->a_mode; 935 int error = EINVAL; 936 937 if (vp->v_type != VREG) 938 return 0; 939 940 if (mode & FREAD) { 941 if (rn->rn_readfd != -1) 942 return 0; 943 rn->rn_readfd = rumpuser_open(rn->rn_hostpath, 944 O_RDONLY, &error); 945 } 946 947 if (mode & FWRITE) { 948 if (rn->rn_writefd != -1) 949 return 0; 950 rn->rn_writefd = rumpuser_open(rn->rn_hostpath, 951 O_WRONLY, &error); 952 } 953 954 return error; 955 } 956 957 /* simple readdir. event omits dotstuff and periods */ 958 static int 959 rump_vop_readdir(void *v) 960 { 961 struct vop_readdir_args /* { 962 struct vnode *a_vp; 963 struct uio *a_uio; 964 kauth_cred_t a_cred; 965 int *a_eofflag; 966 off_t **a_cookies; 967 int *a_ncookies; 968 } */ *ap = v; 969 struct vnode *vp = ap->a_vp; 970 struct uio *uio = ap->a_uio; 971 struct rumpfs_node *rnd = vp->v_data; 972 struct rumpfs_dent *rdent; 973 unsigned i; 974 int rv = 0; 975 976 /* seek to current entry */ 977 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir); 978 (i < uio->uio_offset) && rdent; 979 i++, rdent = LIST_NEXT(rdent, rd_entries)) 980 continue; 981 if (!rdent) 982 goto out; 983 984 /* copy entries */ 985 for (; rdent && uio->uio_resid > 0; 986 rdent = LIST_NEXT(rdent, rd_entries), i++) { 987 struct dirent dent; 988 989 strlcpy(dent.d_name, rdent->rd_name, sizeof(dent.d_name)); 990 dent.d_namlen = strlen(dent.d_name); 991 dent.d_reclen = _DIRENT_RECLEN(&dent, dent.d_namlen); 992 993 if (__predict_false(RDENT_ISWHITEOUT(rdent))) { 994 dent.d_fileno = INO_WHITEOUT; 995 dent.d_type = DT_WHT; 996 } else { 997 dent.d_fileno = rdent->rd_node->rn_va.va_fileid; 998 dent.d_type = vtype2dt(rdent->rd_node->rn_va.va_type); 999 } 1000 1001 if (uio->uio_resid < dent.d_reclen) { 1002 i--; 1003 break; 1004 } 1005 1006 rv = uiomove(&dent, dent.d_reclen, uio); 1007 if (rv) { 1008 i--; 1009 break; 1010 } 1011 } 1012 1013 out: 1014 if (ap->a_cookies) { 1015 *ap->a_ncookies = 0; 1016 *ap->a_cookies = NULL; 1017 } 1018 if (rdent) 1019 *ap->a_eofflag = 0; 1020 else 1021 *ap->a_eofflag = 1; 1022 uio->uio_offset = i; 1023 1024 return rv; 1025 } 1026 1027 static int 1028 rump_vop_read(void *v) 1029 { 1030 struct vop_read_args /* { 1031 struct vnode *a_vp; 1032 struct uio *a_uio; 1033 int ioflags a_ioflag; 1034 kauth_cred_t a_cred; 1035 }; */ *ap = v; 1036 struct vnode *vp = ap->a_vp; 1037 struct rumpfs_node *rn = vp->v_data; 1038 struct uio *uio = ap->a_uio; 1039 uint8_t *buf; 1040 size_t bufsize; 1041 ssize_t n; 1042 int error = 0; 1043 1044 bufsize = uio->uio_resid; 1045 buf = kmem_alloc(bufsize, KM_SLEEP); 1046 if ((n = rumpuser_pread(rn->rn_readfd, buf, bufsize, 1047 uio->uio_offset + rn->rn_offset, &error)) == -1) 1048 goto out; 1049 KASSERT(n <= bufsize); 1050 error = uiomove(buf, n, uio); 1051 1052 out: 1053 kmem_free(buf, bufsize); 1054 return error; 1055 } 1056 1057 static int 1058 rump_vop_write(void *v) 1059 { 1060 struct vop_read_args /* { 1061 struct vnode *a_vp; 1062 struct uio *a_uio; 1063 int ioflags a_ioflag; 1064 kauth_cred_t a_cred; 1065 }; */ *ap = v; 1066 struct vnode *vp = ap->a_vp; 1067 struct rumpfs_node *rn = vp->v_data; 1068 struct uio *uio = ap->a_uio; 1069 uint8_t *buf; 1070 size_t bufsize; 1071 ssize_t n; 1072 int error = 0; 1073 1074 bufsize = uio->uio_resid; 1075 buf = kmem_alloc(bufsize, KM_SLEEP); 1076 error = uiomove(buf, bufsize, uio); 1077 if (error) 1078 goto out; 1079 KASSERT(uio->uio_resid == 0); 1080 n = rumpuser_pwrite(rn->rn_writefd, buf, bufsize, 1081 (uio->uio_offset-bufsize) + rn->rn_offset, &error); 1082 if (n >= 0) { 1083 KASSERT(n <= bufsize); 1084 uio->uio_resid = bufsize - n; 1085 } 1086 1087 out: 1088 kmem_free(buf, bufsize); 1089 return error; 1090 } 1091 1092 static int 1093 rump_vop_success(void *v) 1094 { 1095 1096 return 0; 1097 } 1098 1099 static int 1100 rump_vop_inactive(void *v) 1101 { 1102 struct vop_inactive_args /* { 1103 struct vnode *a_vp; 1104 bool *a_recycle; 1105 } */ *ap = v; 1106 struct vnode *vp = ap->a_vp; 1107 struct rumpfs_node *rn = vp->v_data; 1108 int error; 1109 1110 if (vp->v_type == VREG) { 1111 if (rn->rn_readfd != -1) { 1112 rumpuser_close(rn->rn_readfd, &error); 1113 rn->rn_readfd = -1; 1114 } 1115 if (rn->rn_writefd != -1) { 1116 rumpuser_close(rn->rn_writefd, &error); 1117 rn->rn_writefd = -1; 1118 } 1119 } 1120 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false; 1121 1122 VOP_UNLOCK(vp); 1123 return 0; 1124 } 1125 1126 static int 1127 rump_vop_reclaim(void *v) 1128 { 1129 struct vop_reclaim_args /* { 1130 struct vnode *a_vp; 1131 } */ *ap = v; 1132 struct vnode *vp = ap->a_vp; 1133 struct rumpfs_node *rn = vp->v_data; 1134 1135 mutex_enter(&reclock); 1136 rn->rn_vp = NULL; 1137 mutex_exit(&reclock); 1138 vp->v_data = NULL; 1139 1140 if (rn->rn_flags & RUMPNODE_CANRECLAIM) { 1141 if (vp->v_type == VLNK) 1142 PNBUF_PUT(rn->rn_linktarg); 1143 if (rn->rn_hostpath) 1144 free(rn->rn_hostpath, M_TEMP); 1145 kmem_free(rn, sizeof(*rn)); 1146 } 1147 1148 return 0; 1149 } 1150 1151 static int 1152 rump_vop_spec(void *v) 1153 { 1154 struct vop_generic_args *ap = v; 1155 int (**opvec)(void *); 1156 1157 switch (ap->a_desc->vdesc_offset) { 1158 case VOP_ACCESS_DESCOFFSET: 1159 case VOP_GETATTR_DESCOFFSET: 1160 case VOP_LOCK_DESCOFFSET: 1161 case VOP_UNLOCK_DESCOFFSET: 1162 case VOP_RECLAIM_DESCOFFSET: 1163 opvec = rump_vnodeop_p; 1164 break; 1165 default: 1166 opvec = spec_vnodeop_p; 1167 break; 1168 } 1169 1170 return VOCALL(opvec, ap->a_desc->vdesc_offset, v); 1171 } 1172 1173 /* 1174 * Begin vfs-level stuff 1175 */ 1176 1177 VFS_PROTOS(rumpfs); 1178 struct vfsops rumpfs_vfsops = { 1179 .vfs_name = MOUNT_RUMPFS, 1180 .vfs_min_mount_data = 0, 1181 .vfs_mount = rumpfs_mount, 1182 .vfs_start = (void *)nullop, 1183 .vfs_unmount = rumpfs_unmount, 1184 .vfs_root = rumpfs_root, 1185 .vfs_quotactl = (void *)eopnotsupp, 1186 .vfs_statvfs = genfs_statvfs, 1187 .vfs_sync = (void *)nullop, 1188 .vfs_vget = rumpfs_vget, 1189 .vfs_fhtovp = (void *)eopnotsupp, 1190 .vfs_vptofh = (void *)eopnotsupp, 1191 .vfs_init = rumpfs_init, 1192 .vfs_reinit = NULL, 1193 .vfs_done = rumpfs_done, 1194 .vfs_mountroot = rumpfs_mountroot, 1195 .vfs_snapshot = (void *)eopnotsupp, 1196 .vfs_extattrctl = (void *)eopnotsupp, 1197 .vfs_suspendctl = (void *)eopnotsupp, 1198 .vfs_opv_descs = rump_opv_descs, 1199 /* vfs_refcount */ 1200 /* vfs_list */ 1201 }; 1202 1203 int 1204 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen) 1205 { 1206 1207 return EOPNOTSUPP; 1208 } 1209 1210 int 1211 rumpfs_unmount(struct mount *mp, int flags) 1212 { 1213 1214 /* if going for it, just lie about it */ 1215 if (panicstr) 1216 return 0; 1217 1218 return EOPNOTSUPP; /* ;) */ 1219 } 1220 1221 int 1222 rumpfs_root(struct mount *mp, struct vnode **vpp) 1223 { 1224 struct rumpfs_mount *rfsmp = mp->mnt_data; 1225 1226 vref(rfsmp->rfsmp_rvp); 1227 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY); 1228 *vpp = rfsmp->rfsmp_rvp; 1229 return 0; 1230 } 1231 1232 int 1233 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1234 { 1235 1236 return EOPNOTSUPP; 1237 } 1238 1239 void 1240 rumpfs_init() 1241 { 1242 1243 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET); 1244 1245 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE); 1246 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE); 1247 } 1248 1249 void 1250 rumpfs_done() 1251 { 1252 1253 mutex_destroy(&reclock); 1254 mutex_destroy(&etfs_lock); 1255 } 1256 1257 int 1258 rumpfs_mountroot() 1259 { 1260 struct mount *mp; 1261 struct rumpfs_mount *rfsmp; 1262 struct rumpfs_node *rn; 1263 int error; 1264 1265 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) { 1266 vrele(rootvp); 1267 return error; 1268 } 1269 1270 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP); 1271 1272 rn = makeprivate(VDIR, NODEV, DEV_BSIZE); 1273 rn->rn_parent = rn; 1274 error = makevnode(mp, rn, &rfsmp->rfsmp_rvp); 1275 if (error) 1276 panic("could not create root vnode: %d", error); 1277 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT; 1278 VOP_UNLOCK(rfsmp->rfsmp_rvp); 1279 1280 mutex_enter(&mountlist_lock); 1281 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1282 mutex_exit(&mountlist_lock); 1283 1284 mp->mnt_data = rfsmp; 1285 mp->mnt_stat.f_namemax = MAXNAMLEN; 1286 mp->mnt_stat.f_iosize = 512; 1287 mp->mnt_flag |= MNT_LOCAL; 1288 mp->mnt_iflag |= IMNT_MPSAFE; 1289 vfs_getnewfsid(mp); 1290 1291 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE, 1292 mp->mnt_op->vfs_name, mp, curlwp); 1293 if (error) 1294 panic("set statvfsinfo for rootfs failed"); 1295 1296 vfs_unbusy(mp, false, NULL); 1297 1298 return 0; 1299 } 1300
Indexes created Tue Sep 30 11:09:46 GMT 2025