rumpfs.c revision 1.123
1 /* $NetBSD: rumpfs.c,v 1.123 2014/01/23 10:13:57 hannken Exp $ */ 2 3 /* 4 * Copyright (c) 2009, 2010, 2011 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 1.123 2014/01/23 10:13:57 hannken Exp $"); 30 31 #include <sys/param.h> 32 #include <sys/atomic.h> 33 #include <sys/buf.h> 34 #include <sys/dirent.h> 35 #include <sys/errno.h> 36 #include <sys/filedesc.h> 37 #include <sys/fcntl.h> 38 #include <sys/kauth.h> 39 #include <sys/malloc.h> 40 #include <sys/module.h> 41 #include <sys/mount.h> 42 #include <sys/namei.h> 43 #include <sys/lock.h> 44 #include <sys/lockf.h> 45 #include <sys/queue.h> 46 #include <sys/stat.h> 47 #include <sys/syscallargs.h> 48 #include <sys/vnode.h> 49 #include <sys/unistd.h> 50 51 #include <miscfs/fifofs/fifo.h> 52 #include <miscfs/specfs/specdev.h> 53 #include <miscfs/genfs/genfs.h> 54 #include <miscfs/genfs/genfs_node.h> 55 56 #include <uvm/uvm_extern.h> 57 58 #include <rump/rumpuser.h> 59 60 #include "rump_private.h" 61 #include "rump_vfs_private.h" 62 63 static int rump_vop_lookup(void *); 64 static int rump_vop_getattr(void *); 65 static int rump_vop_setattr(void *); 66 static int rump_vop_mkdir(void *); 67 static int rump_vop_rmdir(void *); 68 static int rump_vop_remove(void *); 69 static int rump_vop_mknod(void *); 70 static int rump_vop_create(void *); 71 static int rump_vop_inactive(void *); 72 static int rump_vop_reclaim(void *); 73 static int rump_vop_success(void *); 74 static int rump_vop_readdir(void *); 75 static int rump_vop_spec(void *); 76 static int rump_vop_read(void *); 77 static int rump_vop_write(void *); 78 static int rump_vop_open(void *); 79 static int rump_vop_symlink(void *); 80 static int rump_vop_readlink(void *); 81 static int rump_vop_whiteout(void *); 82 static int rump_vop_pathconf(void *); 83 static int rump_vop_bmap(void *); 84 static int rump_vop_strategy(void *); 85 static int rump_vop_advlock(void *); 86 static int rump_vop_access(void *); 87 88 int (**fifo_vnodeop_p)(void *); 89 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = { 90 { &vop_default_desc, vn_default_error }, 91 { &vop_putpages_desc, genfs_null_putpages }, 92 { NULL, NULL } 93 }; 94 const struct vnodeopv_desc fifo_vnodeop_opv_desc = 95 { &fifo_vnodeop_p, fifo_vnodeop_entries }; 96 97 int (**rump_vnodeop_p)(void *); 98 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = { 99 { &vop_default_desc, vn_default_error }, 100 { &vop_lookup_desc, rump_vop_lookup }, 101 { &vop_getattr_desc, rump_vop_getattr }, 102 { &vop_setattr_desc, rump_vop_setattr }, 103 { &vop_mkdir_desc, rump_vop_mkdir }, 104 { &vop_rmdir_desc, rump_vop_rmdir }, 105 { &vop_remove_desc, rump_vop_remove }, 106 { &vop_mknod_desc, rump_vop_mknod }, 107 { &vop_create_desc, rump_vop_create }, 108 { &vop_symlink_desc, rump_vop_symlink }, 109 { &vop_readlink_desc, rump_vop_readlink }, 110 { &vop_access_desc, rump_vop_access }, 111 { &vop_readdir_desc, rump_vop_readdir }, 112 { &vop_read_desc, rump_vop_read }, 113 { &vop_write_desc, rump_vop_write }, 114 { &vop_open_desc, rump_vop_open }, 115 { &vop_close_desc, genfs_nullop }, 116 { &vop_seek_desc, genfs_seek }, 117 { &vop_getpages_desc, genfs_getpages }, 118 { &vop_putpages_desc, genfs_putpages }, 119 { &vop_whiteout_desc, rump_vop_whiteout }, 120 { &vop_fsync_desc, rump_vop_success }, 121 { &vop_lock_desc, genfs_lock }, 122 { &vop_unlock_desc, genfs_unlock }, 123 { &vop_islocked_desc, genfs_islocked }, 124 { &vop_inactive_desc, rump_vop_inactive }, 125 { &vop_reclaim_desc, rump_vop_reclaim }, 126 { &vop_link_desc, genfs_eopnotsupp }, 127 { &vop_pathconf_desc, rump_vop_pathconf }, 128 { &vop_bmap_desc, rump_vop_bmap }, 129 { &vop_strategy_desc, rump_vop_strategy }, 130 { &vop_advlock_desc, rump_vop_advlock }, 131 { NULL, NULL } 132 }; 133 const struct vnodeopv_desc rump_vnodeop_opv_desc = 134 { &rump_vnodeop_p, rump_vnodeop_entries }; 135 136 int (**rump_specop_p)(void *); 137 const struct vnodeopv_entry_desc rump_specop_entries[] = { 138 { &vop_default_desc, rump_vop_spec }, 139 { NULL, NULL } 140 }; 141 const struct vnodeopv_desc rump_specop_opv_desc = 142 { &rump_specop_p, rump_specop_entries }; 143 144 const struct vnodeopv_desc * const rump_opv_descs[] = { 145 &rump_vnodeop_opv_desc, 146 &rump_specop_opv_desc, 147 NULL 148 }; 149 150 #define RUMPFS_WHITEOUT ((void *)-1) 151 #define RDENT_ISWHITEOUT(rdp) (rdp->rd_node == RUMPFS_WHITEOUT) 152 struct rumpfs_dent { 153 char *rd_name; 154 int rd_namelen; 155 struct rumpfs_node *rd_node; 156 157 LIST_ENTRY(rumpfs_dent) rd_entries; 158 }; 159 160 struct genfs_ops rumpfs_genfsops = { 161 .gop_size = genfs_size, 162 .gop_write = genfs_gop_write, 163 164 /* optional */ 165 .gop_alloc = NULL, 166 .gop_markupdate = NULL, 167 }; 168 169 struct rumpfs_node { 170 struct genfs_node rn_gn; 171 struct vattr rn_va; 172 struct vnode *rn_vp; 173 char *rn_hostpath; 174 int rn_flags; 175 struct lockf *rn_lockf; 176 177 union { 178 struct { /* VREG */ 179 int readfd; 180 int writefd; 181 uint64_t offset; 182 } reg; 183 struct { 184 void *data; 185 size_t dlen; 186 } reg_noet; 187 struct { /* VDIR */ 188 LIST_HEAD(, rumpfs_dent) dents; 189 struct rumpfs_node *parent; 190 int flags; 191 } dir; 192 struct { 193 char *target; 194 size_t len; 195 } link; 196 } rn_u; 197 }; 198 #define rn_readfd rn_u.reg.readfd 199 #define rn_writefd rn_u.reg.writefd 200 #define rn_offset rn_u.reg.offset 201 #define rn_data rn_u.reg_noet.data 202 #define rn_dlen rn_u.reg_noet.dlen 203 #define rn_dir rn_u.dir.dents 204 #define rn_parent rn_u.dir.parent 205 #define rn_linktarg rn_u.link.target 206 #define rn_linklen rn_u.link.len 207 208 #define RUMPNODE_CANRECLAIM 0x01 209 #define RUMPNODE_DIR_ET 0x02 210 #define RUMPNODE_DIR_ETSUBS 0x04 211 #define RUMPNODE_ET_PHONE_HOST 0x10 212 213 struct rumpfs_mount { 214 struct vnode *rfsmp_rvp; 215 }; 216 217 #define INO_WHITEOUT 1 218 static int lastino = 2; 219 static kmutex_t reclock; 220 221 #define RUMPFS_DEFAULTMODE 0755 222 static void freedir(struct rumpfs_node *, struct componentname *); 223 static struct rumpfs_node *makeprivate(enum vtype, mode_t, dev_t, off_t, bool); 224 225 /* 226 * Extra Terrestrial stuff. We map a given key (pathname) to a file on 227 * the host FS. ET phones home only from the root node of rumpfs. 228 * 229 * When an etfs node is removed, a vnode potentially behind it is not 230 * immediately recycled. 231 */ 232 233 struct etfs { 234 char et_key[MAXPATHLEN]; 235 size_t et_keylen; 236 bool et_prefixkey; 237 bool et_removing; 238 devminor_t et_blkmin; 239 240 LIST_ENTRY(etfs) et_entries; 241 242 struct rumpfs_node *et_rn; 243 }; 244 static kmutex_t etfs_lock; 245 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list); 246 247 static enum vtype 248 ettype_to_vtype(enum rump_etfs_type et) 249 { 250 enum vtype vt; 251 252 switch (et) { 253 case RUMP_ETFS_REG: 254 vt = VREG; 255 break; 256 case RUMP_ETFS_BLK: 257 vt = VBLK; 258 break; 259 case RUMP_ETFS_CHR: 260 vt = VCHR; 261 break; 262 case RUMP_ETFS_DIR: 263 vt = VDIR; 264 break; 265 case RUMP_ETFS_DIR_SUBDIRS: 266 vt = VDIR; 267 break; 268 default: 269 panic("invalid et type: %d", et); 270 } 271 272 return vt; 273 } 274 275 static enum vtype 276 hft_to_vtype(int hft) 277 { 278 enum vtype vt; 279 280 switch (hft) { 281 case RUMPUSER_FT_OTHER: 282 vt = VNON; 283 break; 284 case RUMPUSER_FT_DIR: 285 vt = VDIR; 286 break; 287 case RUMPUSER_FT_REG: 288 vt = VREG; 289 break; 290 case RUMPUSER_FT_BLK: 291 vt = VBLK; 292 break; 293 case RUMPUSER_FT_CHR: 294 vt = VCHR; 295 break; 296 default: 297 vt = VNON; 298 break; 299 } 300 301 return vt; 302 } 303 304 static bool 305 etfs_find(const char *key, struct etfs **etp, bool forceprefix) 306 { 307 struct etfs *et; 308 size_t keylen = strlen(key); 309 310 KASSERT(mutex_owned(&etfs_lock)); 311 312 LIST_FOREACH(et, &etfs_list, et_entries) { 313 if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix) 314 && strncmp(key, et->et_key, et->et_keylen) == 0) { 315 if (etp) 316 *etp = et; 317 return true; 318 } 319 } 320 321 return false; 322 } 323 324 #define REGDIR(ftype) \ 325 ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS) 326 static int 327 doregister(const char *key, const char *hostpath, 328 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 329 { 330 char buf[9]; 331 struct etfs *et; 332 struct rumpfs_node *rn; 333 uint64_t fsize; 334 dev_t rdev = NODEV; 335 devminor_t dmin = -1; 336 int hft, error; 337 338 if (key[0] != '/') { 339 return EINVAL; 340 } 341 while (key[0] == '/') { 342 key++; 343 } 344 345 if ((error = rumpuser_getfileinfo(hostpath, &fsize, &hft)) != 0) 346 return error; 347 348 /* etfs directory requires a directory on the host */ 349 if (REGDIR(ftype)) { 350 if (hft != RUMPUSER_FT_DIR) 351 return ENOTDIR; 352 if (begin != 0) 353 return EISDIR; 354 if (size != RUMP_ETFS_SIZE_ENDOFF) 355 return EISDIR; 356 size = fsize; 357 } else { 358 if (begin > fsize) 359 return EINVAL; 360 if (size == RUMP_ETFS_SIZE_ENDOFF) 361 size = fsize - begin; 362 if (begin + size > fsize) 363 return EINVAL; 364 } 365 366 if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) { 367 error = rumpblk_register(hostpath, &dmin, begin, size); 368 if (error != 0) { 369 return error; 370 } 371 rdev = makedev(RUMPBLK_DEVMAJOR, dmin); 372 } 373 374 et = kmem_alloc(sizeof(*et), KM_SLEEP); 375 strcpy(et->et_key, key); 376 et->et_keylen = strlen(et->et_key); 377 et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), RUMPFS_DEFAULTMODE, 378 rdev, size, true); 379 et->et_removing = false; 380 et->et_blkmin = dmin; 381 382 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 383 384 if (ftype == RUMP_ETFS_REG || REGDIR(ftype) || et->et_blkmin != -1) { 385 size_t len = strlen(hostpath)+1; 386 387 rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 388 memcpy(rn->rn_hostpath, hostpath, len); 389 rn->rn_offset = begin; 390 } 391 392 if (REGDIR(ftype)) { 393 rn->rn_flags |= RUMPNODE_DIR_ET; 394 et->et_prefixkey = true; 395 } else { 396 et->et_prefixkey = false; 397 } 398 399 if (ftype == RUMP_ETFS_DIR_SUBDIRS) 400 rn->rn_flags |= RUMPNODE_DIR_ETSUBS; 401 402 mutex_enter(&etfs_lock); 403 if (etfs_find(key, NULL, REGDIR(ftype))) { 404 mutex_exit(&etfs_lock); 405 if (et->et_blkmin != -1) 406 rumpblk_deregister(hostpath); 407 if (et->et_rn->rn_hostpath != NULL) 408 free(et->et_rn->rn_hostpath, M_TEMP); 409 kmem_free(et->et_rn, sizeof(*et->et_rn)); 410 kmem_free(et, sizeof(*et)); 411 return EEXIST; 412 } 413 LIST_INSERT_HEAD(&etfs_list, et, et_entries); 414 mutex_exit(&etfs_lock); 415 416 if (ftype == RUMP_ETFS_BLK) { 417 format_bytes(buf, sizeof(buf), size); 418 aprint_verbose("/%s: hostpath %s (%s)\n", key, hostpath, buf); 419 } 420 421 return 0; 422 } 423 #undef REGDIR 424 425 int 426 rump_etfs_register(const char *key, const char *hostpath, 427 enum rump_etfs_type ftype) 428 { 429 430 return doregister(key, hostpath, ftype, 0, RUMP_ETFS_SIZE_ENDOFF); 431 } 432 433 int 434 rump_etfs_register_withsize(const char *key, const char *hostpath, 435 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 436 { 437 438 return doregister(key, hostpath, ftype, begin, size); 439 } 440 441 /* remove etfs mapping. caller's responsibility to make sure it's not in use */ 442 int 443 rump_etfs_remove(const char *key) 444 { 445 struct etfs *et; 446 size_t keylen; 447 int rv; 448 449 if (key[0] != '/') { 450 return EINVAL; 451 } 452 while (key[0] == '/') { 453 key++; 454 } 455 456 keylen = strlen(key); 457 458 mutex_enter(&etfs_lock); 459 LIST_FOREACH(et, &etfs_list, et_entries) { 460 if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) { 461 if (et->et_removing) 462 et = NULL; 463 else 464 et->et_removing = true; 465 break; 466 } 467 } 468 mutex_exit(&etfs_lock); 469 if (!et) 470 return ENOENT; 471 472 /* 473 * ok, we know what we want to remove and have signalled there 474 * actually are men at work. first, unregister from rumpblk 475 */ 476 if (et->et_blkmin != -1) { 477 rv = rumpblk_deregister(et->et_rn->rn_hostpath); 478 } else { 479 rv = 0; 480 } 481 KASSERT(rv == 0); 482 483 /* then do the actual removal */ 484 mutex_enter(&etfs_lock); 485 LIST_REMOVE(et, et_entries); 486 mutex_exit(&etfs_lock); 487 488 /* node is unreachable, safe to nuke all device copies */ 489 if (et->et_blkmin != -1) { 490 vdevgone(RUMPBLK_DEVMAJOR, et->et_blkmin, et->et_blkmin, VBLK); 491 } else { 492 struct vnode *vp; 493 494 mutex_enter(&reclock); 495 if ((vp = et->et_rn->rn_vp) != NULL) 496 mutex_enter(vp->v_interlock); 497 mutex_exit(&reclock); 498 if (vp && vget(vp, 0) == 0) 499 vgone(vp); 500 } 501 502 if (et->et_rn->rn_hostpath != NULL) 503 free(et->et_rn->rn_hostpath, M_TEMP); 504 kmem_free(et->et_rn, sizeof(*et->et_rn)); 505 kmem_free(et, sizeof(*et)); 506 507 return 0; 508 } 509 510 /* 511 * rumpfs 512 */ 513 514 static struct rumpfs_node * 515 makeprivate(enum vtype vt, mode_t mode, dev_t rdev, off_t size, bool et) 516 { 517 struct rumpfs_node *rn; 518 struct vattr *va; 519 struct timespec ts; 520 521 KASSERT((mode & ~ALLPERMS) == 0); 522 rn = kmem_zalloc(sizeof(*rn), KM_SLEEP); 523 524 switch (vt) { 525 case VDIR: 526 LIST_INIT(&rn->rn_dir); 527 break; 528 case VREG: 529 if (et) { 530 rn->rn_readfd = -1; 531 rn->rn_writefd = -1; 532 } 533 break; 534 default: 535 break; 536 } 537 538 nanotime(&ts); 539 540 va = &rn->rn_va; 541 va->va_type = vt; 542 va->va_mode = mode; 543 if (vt == VDIR) 544 va->va_nlink = 2; 545 else 546 va->va_nlink = 1; 547 va->va_uid = 0; 548 va->va_gid = 0; 549 va->va_fsid = 550 va->va_fileid = atomic_inc_uint_nv(&lastino); 551 va->va_size = size; 552 va->va_blocksize = 512; 553 va->va_atime = ts; 554 va->va_mtime = ts; 555 va->va_ctime = ts; 556 va->va_birthtime = ts; 557 va->va_gen = 0; 558 va->va_flags = 0; 559 va->va_rdev = rdev; 560 va->va_bytes = 512; 561 va->va_filerev = 0; 562 va->va_vaflags = 0; 563 564 return rn; 565 } 566 567 static int 568 makevnode(struct mount *mp, struct rumpfs_node *rn, struct vnode **vpp) 569 { 570 struct vnode *vp; 571 int (**vpops)(void *); 572 struct vattr *va = &rn->rn_va; 573 int rv; 574 575 KASSERT(!mutex_owned(&reclock)); 576 577 if (va->va_type == VCHR || va->va_type == VBLK) { 578 vpops = rump_specop_p; 579 } else { 580 vpops = rump_vnodeop_p; 581 } 582 583 rv = getnewvnode(VT_RUMP, mp, vpops, NULL, &vp); 584 if (rv) 585 return rv; 586 587 vp->v_size = vp->v_writesize = va->va_size; 588 vp->v_type = va->va_type; 589 590 if (vpops == rump_specop_p) { 591 spec_node_init(vp, va->va_rdev); 592 } 593 vp->v_data = rn; 594 595 genfs_node_init(vp, &rumpfs_genfsops); 596 mutex_enter(&reclock); 597 rn->rn_vp = vp; 598 mutex_exit(&reclock); 599 600 *vpp = vp; 601 602 return 0; 603 } 604 605 606 static void 607 makedir(struct rumpfs_node *rnd, 608 struct componentname *cnp, struct rumpfs_node *rn) 609 { 610 struct rumpfs_dent *rdent; 611 612 rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP); 613 rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP); 614 rdent->rd_node = rn; 615 strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1); 616 rdent->rd_namelen = strlen(rdent->rd_name); 617 618 if ((cnp->cn_flags & ISWHITEOUT) != 0) { 619 KASSERT((cnp->cn_flags & DOWHITEOUT) == 0); 620 freedir(rnd, cnp); 621 } 622 LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries); 623 } 624 625 static void 626 freedir(struct rumpfs_node *rnd, struct componentname *cnp) 627 { 628 struct rumpfs_dent *rd = NULL; 629 630 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 631 if (rd->rd_namelen == cnp->cn_namelen && 632 strncmp(rd->rd_name, cnp->cn_nameptr, 633 cnp->cn_namelen) == 0) 634 break; 635 } 636 if (rd == NULL) 637 panic("could not find directory entry: %s", cnp->cn_nameptr); 638 639 if (cnp->cn_flags & DOWHITEOUT) { 640 rd->rd_node = RUMPFS_WHITEOUT; 641 } else { 642 LIST_REMOVE(rd, rd_entries); 643 kmem_free(rd->rd_name, rd->rd_namelen+1); 644 kmem_free(rd, sizeof(*rd)); 645 } 646 } 647 648 /* 649 * Simple lookup for rump file systems. 650 * 651 * uhm, this is twisted. C F C C, hope of C C F C looming 652 */ 653 static int 654 rump_vop_lookup(void *v) 655 { 656 struct vop_lookup_args /* { 657 struct vnode *a_dvp; 658 struct vnode **a_vpp; 659 struct componentname *a_cnp; 660 }; */ *ap = v; 661 struct componentname *cnp = ap->a_cnp; 662 struct vnode *dvp = ap->a_dvp; 663 struct vnode **vpp = ap->a_vpp; 664 struct vnode *vp; 665 struct rumpfs_node *rnd = dvp->v_data, *rn; 666 struct rumpfs_dent *rd = NULL; 667 struct etfs *et; 668 bool dotdot = (cnp->cn_flags & ISDOTDOT) != 0; 669 int rv = 0; 670 const char *cp; 671 672 *vpp = NULL; 673 674 rv = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 675 if (rv) 676 return rv; 677 678 if ((cnp->cn_flags & ISLASTCN) 679 && (dvp->v_mount->mnt_flag & MNT_RDONLY) 680 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 681 return EROFS; 682 683 /* check for dot, return directly if the case */ 684 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 685 vref(dvp); 686 *vpp = dvp; 687 return 0; 688 } 689 690 /* we don't do rename */ 691 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME))) 692 return EOPNOTSUPP; 693 694 /* check for etfs */ 695 if (dvp == rootvnode && 696 (cnp->cn_nameiop == LOOKUP || cnp->cn_nameiop == CREATE)) { 697 bool found; 698 mutex_enter(&etfs_lock); 699 found = etfs_find(cnp->cn_nameptr, &et, false); 700 mutex_exit(&etfs_lock); 701 702 if (found) { 703 rn = et->et_rn; 704 cnp->cn_consume += et->et_keylen - cnp->cn_namelen; 705 /* 706 * consume trailing slashes if any and clear 707 * REQUIREDIR if we consumed the full path. 708 */ 709 cp = &cnp->cn_nameptr[cnp->cn_namelen]; 710 cp += cnp->cn_consume; 711 KASSERT(*cp == '\0' || *cp == '/'); 712 if (*cp == '\0' && rn->rn_va.va_type != VDIR) 713 cnp->cn_flags &= ~REQUIREDIR; 714 while (*cp++ == '/') 715 cnp->cn_consume++; 716 goto getvnode; 717 } 718 } 719 720 if (rnd->rn_flags & RUMPNODE_DIR_ET) { 721 uint64_t fsize; 722 char *newpath; 723 size_t newpathlen; 724 int hft, error; 725 726 if (dotdot) 727 return EOPNOTSUPP; 728 729 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1; 730 newpath = malloc(newpathlen, M_TEMP, M_WAITOK); 731 732 strlcpy(newpath, rnd->rn_hostpath, newpathlen); 733 strlcat(newpath, "/", newpathlen); 734 strlcat(newpath, cnp->cn_nameptr, newpathlen); 735 736 if ((error = rumpuser_getfileinfo(newpath, &fsize, &hft)) != 0){ 737 free(newpath, M_TEMP); 738 return error; 739 } 740 741 /* allow only dirs and regular files */ 742 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) { 743 free(newpath, M_TEMP); 744 return ENOENT; 745 } 746 747 rn = makeprivate(hft_to_vtype(hft), RUMPFS_DEFAULTMODE, 748 NODEV, fsize, true); 749 rn->rn_flags |= RUMPNODE_CANRECLAIM; 750 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) { 751 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS; 752 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 753 } 754 rn->rn_hostpath = newpath; 755 756 goto getvnode; 757 } else { 758 if (dotdot) { 759 if ((rn = rnd->rn_parent) != NULL) 760 goto getvnode; 761 } else { 762 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 763 if (rd->rd_namelen == cnp->cn_namelen && 764 strncmp(rd->rd_name, cnp->cn_nameptr, 765 cnp->cn_namelen) == 0) 766 break; 767 } 768 } 769 } 770 771 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE)) 772 return ENOENT; 773 774 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 775 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 776 return EROFS; 777 rv = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred); 778 if (rv) 779 return rv; 780 return EJUSTRETURN; 781 } 782 783 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == DELETE) { 784 rv = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred); 785 if (rv) 786 return rv; 787 } 788 789 if (RDENT_ISWHITEOUT(rd)) { 790 cnp->cn_flags |= ISWHITEOUT; 791 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) 792 return EJUSTRETURN; 793 return ENOENT; 794 } 795 796 rn = rd->rd_node; 797 798 getvnode: 799 KASSERT(rn); 800 if (dotdot) 801 VOP_UNLOCK(dvp); 802 mutex_enter(&reclock); 803 if ((vp = rn->rn_vp)) { 804 mutex_enter(vp->v_interlock); 805 mutex_exit(&reclock); 806 if (vget(vp, LK_EXCLUSIVE)) { 807 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 808 goto getvnode; 809 } 810 *vpp = vp; 811 } else { 812 mutex_exit(&reclock); 813 rv = makevnode(dvp->v_mount, rn, vpp); 814 if (rv == 0) { 815 rv = vn_lock(*vpp, LK_EXCLUSIVE); 816 if (rv != 0) { 817 vrele(*vpp); 818 *vpp = NULL; 819 } 820 } 821 } 822 if (dotdot) 823 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 824 825 return rv; 826 } 827 828 static int 829 rump_check_possible(struct vnode *vp, struct rumpfs_node *rnode, 830 mode_t mode) 831 { 832 833 if ((mode & VWRITE) == 0) 834 return 0; 835 836 switch (vp->v_type) { 837 case VDIR: 838 case VLNK: 839 case VREG: 840 break; 841 default: 842 /* special file is always writable. */ 843 return 0; 844 } 845 846 return vp->v_mount->mnt_flag & MNT_RDONLY ? EROFS : 0; 847 } 848 849 static int 850 rump_check_permitted(struct vnode *vp, struct rumpfs_node *rnode, 851 mode_t mode, kauth_cred_t cred) 852 { 853 struct vattr *attr = &rnode->rn_va; 854 855 return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(mode, 856 vp->v_type, attr->va_mode), vp, NULL, genfs_can_access(vp->v_type, 857 attr->va_mode, attr->va_uid, attr->va_gid, mode, cred)); 858 } 859 860 int 861 rump_vop_access(void *v) 862 { 863 struct vop_access_args /* { 864 const struct vnodeop_desc *a_desc; 865 struct vnode *a_vp; 866 int a_mode; 867 kauth_cred_t a_cred; 868 } */ *ap = v; 869 struct vnode *vp = ap->a_vp; 870 struct rumpfs_node *rn = vp->v_data; 871 int error; 872 873 error = rump_check_possible(vp, rn, ap->a_mode); 874 if (error) 875 return error; 876 877 error = rump_check_permitted(vp, rn, ap->a_mode, ap->a_cred); 878 879 return error; 880 } 881 882 static int 883 rump_vop_getattr(void *v) 884 { 885 struct vop_getattr_args /* { 886 struct vnode *a_vp; 887 struct vattr *a_vap; 888 kauth_cred_t a_cred; 889 } */ *ap = v; 890 struct vnode *vp = ap->a_vp; 891 struct rumpfs_node *rn = vp->v_data; 892 struct vattr *vap = ap->a_vap; 893 894 memcpy(vap, &rn->rn_va, sizeof(struct vattr)); 895 vap->va_size = vp->v_size; 896 return 0; 897 } 898 899 static int 900 rump_vop_setattr(void *v) 901 { 902 struct vop_setattr_args /* { 903 struct vnode *a_vp; 904 struct vattr *a_vap; 905 kauth_cred_t a_cred; 906 } */ *ap = v; 907 struct vnode *vp = ap->a_vp; 908 struct vattr *vap = ap->a_vap; 909 struct rumpfs_node *rn = vp->v_data; 910 struct vattr *attr = &rn->rn_va; 911 kauth_cred_t cred = ap->a_cred; 912 int error; 913 914 #define CHANGED(a, t) (vap->a != (t)VNOVAL) 915 #define SETIFVAL(a,t) if (CHANGED(a, t)) rn->rn_va.a = vap->a 916 if (CHANGED(va_atime.tv_sec, time_t) || 917 CHANGED(va_ctime.tv_sec, time_t) || 918 CHANGED(va_mtime.tv_sec, time_t) || 919 CHANGED(va_birthtime.tv_sec, time_t) || 920 CHANGED(va_atime.tv_nsec, long) || 921 CHANGED(va_ctime.tv_nsec, long) || 922 CHANGED(va_mtime.tv_nsec, long) || 923 CHANGED(va_birthtime.tv_nsec, long)) { 924 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_TIMES, vp, 925 NULL, genfs_can_chtimes(vp, vap->va_vaflags, attr->va_uid, 926 cred)); 927 if (error) 928 return error; 929 } 930 931 SETIFVAL(va_atime.tv_sec, time_t); 932 SETIFVAL(va_ctime.tv_sec, time_t); 933 SETIFVAL(va_mtime.tv_sec, time_t); 934 SETIFVAL(va_birthtime.tv_sec, time_t); 935 SETIFVAL(va_atime.tv_nsec, long); 936 SETIFVAL(va_ctime.tv_nsec, long); 937 SETIFVAL(va_mtime.tv_nsec, long); 938 SETIFVAL(va_birthtime.tv_nsec, long); 939 940 if (CHANGED(va_flags, u_long)) { 941 /* XXX Can we handle system flags here...? */ 942 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_FLAGS, vp, 943 NULL, genfs_can_chflags(cred, vp->v_type, attr->va_uid, 944 false)); 945 if (error) 946 return error; 947 } 948 949 SETIFVAL(va_flags, u_long); 950 #undef SETIFVAL 951 #undef CHANGED 952 953 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (uid_t)VNOVAL) { 954 uid_t uid = 955 (vap->va_uid != (uid_t)VNOVAL) ? vap->va_uid : attr->va_uid; 956 gid_t gid = 957 (vap->va_gid != (gid_t)VNOVAL) ? vap->va_gid : attr->va_gid; 958 error = kauth_authorize_vnode(cred, 959 KAUTH_VNODE_CHANGE_OWNERSHIP, vp, NULL, 960 genfs_can_chown(cred, attr->va_uid, attr->va_gid, uid, 961 gid)); 962 if (error) 963 return error; 964 attr->va_uid = uid; 965 attr->va_gid = gid; 966 } 967 968 if (vap->va_mode != (mode_t)VNOVAL) { 969 mode_t mode = vap->va_mode; 970 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_SECURITY, 971 vp, NULL, genfs_can_chmod(vp->v_type, cred, attr->va_uid, 972 attr->va_gid, mode)); 973 if (error) 974 return error; 975 attr->va_mode = mode; 976 } 977 978 if (vp->v_type == VREG && 979 vap->va_size != VSIZENOTSET && 980 vap->va_size != rn->rn_dlen) { 981 void *newdata; 982 size_t copylen, newlen; 983 984 newlen = vap->va_size; 985 newdata = rump_hypermalloc(newlen, 0, true, "rumpfs"); 986 987 copylen = MIN(rn->rn_dlen, newlen); 988 memset(newdata, 0, newlen); 989 memcpy(newdata, rn->rn_data, copylen); 990 rump_hyperfree(rn->rn_data, rn->rn_dlen); 991 992 rn->rn_data = newdata; 993 rn->rn_dlen = newlen; 994 uvm_vnp_setsize(vp, newlen); 995 } 996 return 0; 997 } 998 999 static int 1000 rump_vop_mkdir(void *v) 1001 { 1002 struct vop_mkdir_v3_args /* { 1003 struct vnode *a_dvp; 1004 struct vnode **a_vpp; 1005 struct componentname *a_cnp; 1006 struct vattr *a_vap; 1007 }; */ *ap = v; 1008 struct vnode *dvp = ap->a_dvp; 1009 struct vnode **vpp = ap->a_vpp; 1010 struct componentname *cnp = ap->a_cnp; 1011 struct vattr *va = ap->a_vap; 1012 struct rumpfs_node *rnd = dvp->v_data, *rn; 1013 int rv = 0; 1014 1015 rn = makeprivate(VDIR, va->va_mode & ALLPERMS, NODEV, DEV_BSIZE, false); 1016 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1017 rn->rn_va.va_flags |= UF_OPAQUE; 1018 rn->rn_parent = rnd; 1019 rv = makevnode(dvp->v_mount, rn, vpp); 1020 if (rv) 1021 return rv; 1022 1023 makedir(rnd, cnp, rn); 1024 1025 return rv; 1026 } 1027 1028 static int 1029 rump_vop_rmdir(void *v) 1030 { 1031 struct vop_rmdir_args /* { 1032 struct vnode *a_dvp; 1033 struct vnode *a_vp; 1034 struct componentname *a_cnp; 1035 }; */ *ap = v; 1036 struct vnode *dvp = ap->a_dvp; 1037 struct vnode *vp = ap->a_vp; 1038 struct componentname *cnp = ap->a_cnp; 1039 struct rumpfs_node *rnd = dvp->v_data; 1040 struct rumpfs_node *rn = vp->v_data; 1041 struct rumpfs_dent *rd; 1042 int rv = 0; 1043 1044 LIST_FOREACH(rd, &rn->rn_dir, rd_entries) { 1045 if (rd->rd_node != RUMPFS_WHITEOUT) { 1046 rv = ENOTEMPTY; 1047 goto out; 1048 } 1049 } 1050 while ((rd = LIST_FIRST(&rn->rn_dir)) != NULL) { 1051 KASSERT(rd->rd_node == RUMPFS_WHITEOUT); 1052 LIST_REMOVE(rd, rd_entries); 1053 kmem_free(rd->rd_name, rd->rd_namelen+1); 1054 kmem_free(rd, sizeof(*rd)); 1055 } 1056 1057 freedir(rnd, cnp); 1058 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1059 rn->rn_parent = NULL; 1060 1061 out: 1062 vput(dvp); 1063 vput(vp); 1064 1065 return rv; 1066 } 1067 1068 static int 1069 rump_vop_remove(void *v) 1070 { 1071 struct vop_remove_args /* { 1072 struct vnode *a_dvp; 1073 struct vnode *a_vp; 1074 struct componentname *a_cnp; 1075 }; */ *ap = v; 1076 struct vnode *dvp = ap->a_dvp; 1077 struct vnode *vp = ap->a_vp; 1078 struct componentname *cnp = ap->a_cnp; 1079 struct rumpfs_node *rnd = dvp->v_data; 1080 struct rumpfs_node *rn = vp->v_data; 1081 int rv = 0; 1082 1083 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1084 return EOPNOTSUPP; 1085 1086 freedir(rnd, cnp); 1087 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1088 1089 vput(dvp); 1090 vput(vp); 1091 1092 return rv; 1093 } 1094 1095 static int 1096 rump_vop_mknod(void *v) 1097 { 1098 struct vop_mknod_v3_args /* { 1099 struct vnode *a_dvp; 1100 struct vnode **a_vpp; 1101 struct componentname *a_cnp; 1102 struct vattr *a_vap; 1103 }; */ *ap = v; 1104 struct vnode *dvp = ap->a_dvp; 1105 struct vnode **vpp = ap->a_vpp; 1106 struct componentname *cnp = ap->a_cnp; 1107 struct vattr *va = ap->a_vap; 1108 struct rumpfs_node *rnd = dvp->v_data, *rn; 1109 int rv; 1110 1111 rn = makeprivate(va->va_type, va->va_mode & ALLPERMS, va->va_rdev, 1112 DEV_BSIZE, false); 1113 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1114 rn->rn_va.va_flags |= UF_OPAQUE; 1115 rv = makevnode(dvp->v_mount, rn, vpp); 1116 if (rv) 1117 return rv; 1118 1119 makedir(rnd, cnp, rn); 1120 1121 return rv; 1122 } 1123 1124 static int 1125 rump_vop_create(void *v) 1126 { 1127 struct vop_create_v3_args /* { 1128 struct vnode *a_dvp; 1129 struct vnode **a_vpp; 1130 struct componentname *a_cnp; 1131 struct vattr *a_vap; 1132 }; */ *ap = v; 1133 struct vnode *dvp = ap->a_dvp; 1134 struct vnode **vpp = ap->a_vpp; 1135 struct componentname *cnp = ap->a_cnp; 1136 struct vattr *va = ap->a_vap; 1137 struct rumpfs_node *rnd = dvp->v_data, *rn; 1138 off_t newsize; 1139 int rv; 1140 1141 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0; 1142 rn = makeprivate(va->va_type, va->va_mode & ALLPERMS, NODEV, 1143 newsize, false); 1144 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1145 rn->rn_va.va_flags |= UF_OPAQUE; 1146 rv = makevnode(dvp->v_mount, rn, vpp); 1147 if (rv) 1148 return rv; 1149 1150 makedir(rnd, cnp, rn); 1151 1152 return rv; 1153 } 1154 1155 static int 1156 rump_vop_symlink(void *v) 1157 { 1158 struct vop_symlink_v3_args /* { 1159 struct vnode *a_dvp; 1160 struct vnode **a_vpp; 1161 struct componentname *a_cnp; 1162 struct vattr *a_vap; 1163 char *a_target; 1164 }; */ *ap = v; 1165 struct vnode *dvp = ap->a_dvp; 1166 struct vnode **vpp = ap->a_vpp; 1167 struct componentname *cnp = ap->a_cnp; 1168 struct vattr *va = ap->a_vap; 1169 struct rumpfs_node *rnd = dvp->v_data, *rn; 1170 const char *target = ap->a_target; 1171 size_t linklen; 1172 int rv; 1173 1174 linklen = strlen(target); 1175 KASSERT(linklen < MAXPATHLEN); 1176 rn = makeprivate(VLNK, va->va_mode & ALLPERMS, NODEV, linklen, false); 1177 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1178 rn->rn_va.va_flags |= UF_OPAQUE; 1179 rv = makevnode(dvp->v_mount, rn, vpp); 1180 if (rv) 1181 return rv; 1182 1183 makedir(rnd, cnp, rn); 1184 1185 KASSERT(linklen < MAXPATHLEN); 1186 rn->rn_linktarg = PNBUF_GET(); 1187 rn->rn_linklen = linklen; 1188 strcpy(rn->rn_linktarg, target); 1189 1190 return rv; 1191 } 1192 1193 static int 1194 rump_vop_readlink(void *v) 1195 { 1196 struct vop_readlink_args /* { 1197 struct vnode *a_vp; 1198 struct uio *a_uio; 1199 kauth_cred_t a_cred; 1200 }; */ *ap = v; 1201 struct vnode *vp = ap->a_vp; 1202 struct rumpfs_node *rn = vp->v_data; 1203 struct uio *uio = ap->a_uio; 1204 1205 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio); 1206 } 1207 1208 static int 1209 rump_vop_whiteout(void *v) 1210 { 1211 struct vop_whiteout_args /* { 1212 struct vnode *a_dvp; 1213 struct componentname *a_cnp; 1214 int a_flags; 1215 } */ *ap = v; 1216 struct vnode *dvp = ap->a_dvp; 1217 struct rumpfs_node *rnd = dvp->v_data; 1218 struct componentname *cnp = ap->a_cnp; 1219 int flags = ap->a_flags; 1220 1221 switch (flags) { 1222 case LOOKUP: 1223 break; 1224 case CREATE: 1225 makedir(rnd, cnp, RUMPFS_WHITEOUT); 1226 break; 1227 case DELETE: 1228 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */ 1229 freedir(rnd, cnp); 1230 break; 1231 default: 1232 panic("unknown whiteout op %d", flags); 1233 } 1234 1235 return 0; 1236 } 1237 1238 static int 1239 rump_vop_open(void *v) 1240 { 1241 struct vop_open_args /* { 1242 struct vnode *a_vp; 1243 int a_mode; 1244 kauth_cred_t a_cred; 1245 } */ *ap = v; 1246 struct vnode *vp = ap->a_vp; 1247 struct rumpfs_node *rn = vp->v_data; 1248 int mode = ap->a_mode; 1249 int error = EINVAL; 1250 1251 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0) 1252 return 0; 1253 1254 if (mode & FREAD) { 1255 if (rn->rn_readfd != -1) 1256 return 0; 1257 error = rumpuser_open(rn->rn_hostpath, 1258 RUMPUSER_OPEN_RDONLY, &rn->rn_readfd); 1259 } 1260 1261 if (mode & FWRITE) { 1262 if (rn->rn_writefd != -1) 1263 return 0; 1264 error = rumpuser_open(rn->rn_hostpath, 1265 RUMPUSER_OPEN_WRONLY, &rn->rn_writefd); 1266 } 1267 1268 return error; 1269 } 1270 1271 /* simple readdir. even omits dotstuff and periods */ 1272 static int 1273 rump_vop_readdir(void *v) 1274 { 1275 struct vop_readdir_args /* { 1276 struct vnode *a_vp; 1277 struct uio *a_uio; 1278 kauth_cred_t a_cred; 1279 int *a_eofflag; 1280 off_t **a_cookies; 1281 int *a_ncookies; 1282 } */ *ap = v; 1283 struct vnode *vp = ap->a_vp; 1284 struct uio *uio = ap->a_uio; 1285 struct rumpfs_node *rnd = vp->v_data; 1286 struct rumpfs_dent *rdent; 1287 struct dirent *dentp = NULL; 1288 unsigned i; 1289 int rv = 0; 1290 1291 /* seek to current entry */ 1292 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir); 1293 (i < uio->uio_offset) && rdent; 1294 i++, rdent = LIST_NEXT(rdent, rd_entries)) 1295 continue; 1296 if (!rdent) 1297 goto out; 1298 1299 /* copy entries */ 1300 dentp = kmem_alloc(sizeof(*dentp), KM_SLEEP); 1301 for (; rdent && uio->uio_resid > 0; 1302 rdent = LIST_NEXT(rdent, rd_entries), i++) { 1303 strlcpy(dentp->d_name, rdent->rd_name, sizeof(dentp->d_name)); 1304 dentp->d_namlen = strlen(dentp->d_name); 1305 dentp->d_reclen = _DIRENT_RECLEN(dentp, dentp->d_namlen); 1306 1307 if (__predict_false(RDENT_ISWHITEOUT(rdent))) { 1308 dentp->d_fileno = INO_WHITEOUT; 1309 dentp->d_type = DT_WHT; 1310 } else { 1311 dentp->d_fileno = rdent->rd_node->rn_va.va_fileid; 1312 dentp->d_type = vtype2dt(rdent->rd_node->rn_va.va_type); 1313 } 1314 1315 if (uio->uio_resid < dentp->d_reclen) { 1316 i--; 1317 break; 1318 } 1319 1320 rv = uiomove(dentp, dentp->d_reclen, uio); 1321 if (rv) { 1322 i--; 1323 break; 1324 } 1325 } 1326 kmem_free(dentp, sizeof(*dentp)); 1327 dentp = NULL; 1328 1329 out: 1330 KASSERT(dentp == NULL); 1331 if (ap->a_cookies) { 1332 *ap->a_ncookies = 0; 1333 *ap->a_cookies = NULL; 1334 } 1335 if (rdent) 1336 *ap->a_eofflag = 0; 1337 else 1338 *ap->a_eofflag = 1; 1339 uio->uio_offset = i; 1340 1341 return rv; 1342 } 1343 1344 static int 1345 etread(struct rumpfs_node *rn, struct uio *uio) 1346 { 1347 struct rumpuser_iovec iov; 1348 uint8_t *buf; 1349 size_t bufsize, n; 1350 int error = 0; 1351 1352 bufsize = uio->uio_resid; 1353 if (bufsize == 0) 1354 return 0; 1355 buf = kmem_alloc(bufsize, KM_SLEEP); 1356 1357 iov.iov_base = buf; 1358 iov.iov_len = bufsize; 1359 if ((error = rumpuser_iovread(rn->rn_readfd, &iov, 1, 1360 uio->uio_offset + rn->rn_offset, &n)) == 0) { 1361 KASSERT(n <= bufsize); 1362 error = uiomove(buf, n, uio); 1363 } 1364 1365 kmem_free(buf, bufsize); 1366 return error; 1367 } 1368 1369 static int 1370 rump_vop_read(void *v) 1371 { 1372 struct vop_read_args /* { 1373 struct vnode *a_vp; 1374 struct uio *a_uio; 1375 int ioflags a_ioflag; 1376 kauth_cred_t a_cred; 1377 }; */ *ap = v; 1378 struct vnode *vp = ap->a_vp; 1379 struct rumpfs_node *rn = vp->v_data; 1380 struct uio *uio = ap->a_uio; 1381 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1382 off_t chunk; 1383 int error = 0; 1384 1385 if (vp->v_type == VDIR) 1386 return EISDIR; 1387 1388 /* et op? */ 1389 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1390 return etread(rn, uio); 1391 1392 /* otherwise, it's off to ubc with us */ 1393 while (uio->uio_resid > 0) { 1394 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1395 if (chunk == 0) 1396 break; 1397 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1398 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1399 if (error) 1400 break; 1401 } 1402 1403 return error; 1404 } 1405 1406 static int 1407 etwrite(struct rumpfs_node *rn, struct uio *uio) 1408 { 1409 struct rumpuser_iovec iov; 1410 uint8_t *buf; 1411 size_t bufsize, n; 1412 int error = 0; 1413 1414 bufsize = uio->uio_resid; 1415 if (bufsize == 0) 1416 return 0; 1417 buf = kmem_alloc(bufsize, KM_SLEEP); 1418 error = uiomove(buf, bufsize, uio); 1419 if (error) 1420 goto out; 1421 1422 KASSERT(uio->uio_resid == 0); 1423 iov.iov_base = buf; 1424 iov.iov_len = bufsize; 1425 if ((error = rumpuser_iovwrite(rn->rn_writefd, &iov, 1, 1426 (uio->uio_offset-bufsize) + rn->rn_offset, &n)) == 0) { 1427 KASSERT(n <= bufsize); 1428 uio->uio_resid = bufsize - n; 1429 } 1430 1431 out: 1432 kmem_free(buf, bufsize); 1433 return error; 1434 } 1435 1436 static int 1437 rump_vop_write(void *v) 1438 { 1439 struct vop_write_args /* { 1440 struct vnode *a_vp; 1441 struct uio *a_uio; 1442 int ioflags a_ioflag; 1443 kauth_cred_t a_cred; 1444 }; */ *ap = v; 1445 struct vnode *vp = ap->a_vp; 1446 struct rumpfs_node *rn = vp->v_data; 1447 struct uio *uio = ap->a_uio; 1448 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1449 void *olddata; 1450 size_t oldlen, newlen; 1451 off_t chunk; 1452 int error = 0; 1453 bool allocd = false; 1454 1455 if (ap->a_ioflag & IO_APPEND) 1456 uio->uio_offset = vp->v_size; 1457 1458 /* consult et? */ 1459 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1460 return etwrite(rn, uio); 1461 1462 /* 1463 * Otherwise, it's a case of ubcmove. 1464 */ 1465 1466 /* 1467 * First, make sure we have enough storage. 1468 * 1469 * No, you don't need to tell me it's not very efficient. 1470 * No, it doesn't really support sparse files, just fakes it. 1471 */ 1472 newlen = uio->uio_offset + uio->uio_resid; 1473 oldlen = 0; /* XXXgcc */ 1474 olddata = NULL; 1475 if (rn->rn_dlen < newlen) { 1476 oldlen = rn->rn_dlen; 1477 olddata = rn->rn_data; 1478 1479 rn->rn_data = rump_hypermalloc(newlen, 0, true, "rumpfs"); 1480 rn->rn_dlen = newlen; 1481 memset(rn->rn_data, 0, newlen); 1482 memcpy(rn->rn_data, olddata, oldlen); 1483 allocd = true; 1484 uvm_vnp_setsize(vp, newlen); 1485 } 1486 1487 /* ok, we have enough stooorage. write */ 1488 while (uio->uio_resid > 0) { 1489 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1490 if (chunk == 0) 1491 break; 1492 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1493 UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1494 if (error) 1495 break; 1496 } 1497 1498 if (allocd) { 1499 if (error) { 1500 rump_hyperfree(rn->rn_data, newlen); 1501 rn->rn_data = olddata; 1502 rn->rn_dlen = oldlen; 1503 uvm_vnp_setsize(vp, oldlen); 1504 } else { 1505 rump_hyperfree(olddata, oldlen); 1506 } 1507 } 1508 1509 return error; 1510 } 1511 1512 static int 1513 rump_vop_bmap(void *v) 1514 { 1515 struct vop_bmap_args /* { 1516 struct vnode *a_vp; 1517 daddr_t a_bn; 1518 struct vnode **a_vpp; 1519 daddr_t *a_bnp; 1520 int *a_runp; 1521 } */ *ap = v; 1522 1523 /* 1:1 mapping */ 1524 if (ap->a_vpp) 1525 *ap->a_vpp = ap->a_vp; 1526 if (ap->a_bnp) 1527 *ap->a_bnp = ap->a_bn; 1528 if (ap->a_runp) 1529 *ap->a_runp = 16; 1530 1531 return 0; 1532 } 1533 1534 static int 1535 rump_vop_strategy(void *v) 1536 { 1537 struct vop_strategy_args /* { 1538 struct vnode *a_vp; 1539 struct buf *a_bp; 1540 } */ *ap = v; 1541 struct vnode *vp = ap->a_vp; 1542 struct rumpfs_node *rn = vp->v_data; 1543 struct buf *bp = ap->a_bp; 1544 off_t copylen, copyoff; 1545 int error; 1546 1547 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) { 1548 error = EINVAL; 1549 goto out; 1550 } 1551 1552 copyoff = bp->b_blkno << DEV_BSHIFT; 1553 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount); 1554 if (BUF_ISWRITE(bp)) { 1555 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen); 1556 } else { 1557 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen); 1558 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen); 1559 } 1560 bp->b_resid = 0; 1561 error = 0; 1562 1563 out: 1564 bp->b_error = error; 1565 biodone(bp); 1566 return 0; 1567 } 1568 1569 static int 1570 rump_vop_pathconf(void *v) 1571 { 1572 struct vop_pathconf_args /* { 1573 struct vnode *a_vp; 1574 int a_name; 1575 register_t *a_retval; 1576 }; */ *ap = v; 1577 int name = ap->a_name; 1578 register_t *retval = ap->a_retval; 1579 1580 switch (name) { 1581 case _PC_LINK_MAX: 1582 *retval = LINK_MAX; 1583 return 0; 1584 case _PC_NAME_MAX: 1585 *retval = RUMPFS_MAXNAMLEN; 1586 return 0; 1587 case _PC_PATH_MAX: 1588 *retval = PATH_MAX; 1589 return 0; 1590 case _PC_PIPE_BUF: 1591 *retval = PIPE_BUF; 1592 return 0; 1593 case _PC_CHOWN_RESTRICTED: 1594 *retval = 1; 1595 return 0; 1596 case _PC_NO_TRUNC: 1597 *retval = 1; 1598 return 0; 1599 case _PC_SYNC_IO: 1600 *retval = 1; 1601 return 0; 1602 case _PC_FILESIZEBITS: 1603 *retval = 43; /* this one goes to 11 */ 1604 return 0; 1605 case _PC_SYMLINK_MAX: 1606 *retval = MAXPATHLEN; 1607 return 0; 1608 case _PC_2_SYMLINKS: 1609 *retval = 1; 1610 return 0; 1611 default: 1612 return EINVAL; 1613 } 1614 } 1615 1616 static int 1617 rump_vop_success(void *v) 1618 { 1619 1620 return 0; 1621 } 1622 1623 static int 1624 rump_vop_inactive(void *v) 1625 { 1626 struct vop_inactive_args /* { 1627 struct vnode *a_vp; 1628 bool *a_recycle; 1629 } */ *ap = v; 1630 struct vnode *vp = ap->a_vp; 1631 struct rumpfs_node *rn = vp->v_data; 1632 1633 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) { 1634 if (rn->rn_readfd != -1) { 1635 rumpuser_close(rn->rn_readfd); 1636 rn->rn_readfd = -1; 1637 } 1638 if (rn->rn_writefd != -1) { 1639 rumpuser_close(rn->rn_writefd); 1640 rn->rn_writefd = -1; 1641 } 1642 } 1643 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false; 1644 1645 VOP_UNLOCK(vp); 1646 return 0; 1647 } 1648 1649 static int 1650 rump_vop_reclaim(void *v) 1651 { 1652 struct vop_reclaim_args /* { 1653 struct vnode *a_vp; 1654 } */ *ap = v; 1655 struct vnode *vp = ap->a_vp; 1656 struct rumpfs_node *rn = vp->v_data; 1657 1658 mutex_enter(&reclock); 1659 rn->rn_vp = NULL; 1660 mutex_exit(&reclock); 1661 genfs_node_destroy(vp); 1662 vp->v_data = NULL; 1663 1664 if (rn->rn_flags & RUMPNODE_CANRECLAIM) { 1665 if (vp->v_type == VREG 1666 && (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0 1667 && rn->rn_data) { 1668 rump_hyperfree(rn->rn_data, rn->rn_dlen); 1669 rn->rn_data = NULL; 1670 } 1671 1672 if (vp->v_type == VLNK) 1673 PNBUF_PUT(rn->rn_linktarg); 1674 if (rn->rn_hostpath) 1675 free(rn->rn_hostpath, M_TEMP); 1676 kmem_free(rn, sizeof(*rn)); 1677 } 1678 1679 return 0; 1680 } 1681 1682 static int 1683 rump_vop_spec(void *v) 1684 { 1685 struct vop_generic_args *ap = v; 1686 int (**opvec)(void *); 1687 1688 switch (ap->a_desc->vdesc_offset) { 1689 case VOP_ACCESS_DESCOFFSET: 1690 case VOP_GETATTR_DESCOFFSET: 1691 case VOP_SETATTR_DESCOFFSET: 1692 case VOP_LOCK_DESCOFFSET: 1693 case VOP_UNLOCK_DESCOFFSET: 1694 case VOP_ISLOCKED_DESCOFFSET: 1695 case VOP_RECLAIM_DESCOFFSET: 1696 opvec = rump_vnodeop_p; 1697 break; 1698 default: 1699 opvec = spec_vnodeop_p; 1700 break; 1701 } 1702 1703 return VOCALL(opvec, ap->a_desc->vdesc_offset, v); 1704 } 1705 1706 static int 1707 rump_vop_advlock(void *v) 1708 { 1709 struct vop_advlock_args /* { 1710 const struct vnodeop_desc *a_desc; 1711 struct vnode *a_vp; 1712 void *a_id; 1713 int a_op; 1714 struct flock *a_fl; 1715 int a_flags; 1716 } */ *ap = v; 1717 struct vnode *vp = ap->a_vp; 1718 struct rumpfs_node *rn = vp->v_data; 1719 1720 return lf_advlock(ap, &rn->rn_lockf, vp->v_size); 1721 } 1722 1723 /* 1724 * Begin vfs-level stuff 1725 */ 1726 1727 VFS_PROTOS(rumpfs); 1728 struct vfsops rumpfs_vfsops = { 1729 .vfs_name = MOUNT_RUMPFS, 1730 .vfs_min_mount_data = 0, 1731 .vfs_mount = rumpfs_mount, 1732 .vfs_start = (void *)nullop, 1733 .vfs_unmount = rumpfs_unmount, 1734 .vfs_root = rumpfs_root, 1735 .vfs_quotactl = (void *)eopnotsupp, 1736 .vfs_statvfs = genfs_statvfs, 1737 .vfs_sync = (void *)nullop, 1738 .vfs_vget = rumpfs_vget, 1739 .vfs_fhtovp = (void *)eopnotsupp, 1740 .vfs_vptofh = (void *)eopnotsupp, 1741 .vfs_init = rumpfs_init, 1742 .vfs_reinit = NULL, 1743 .vfs_done = rumpfs_done, 1744 .vfs_mountroot = rumpfs_mountroot, 1745 .vfs_snapshot = (void *)eopnotsupp, 1746 .vfs_extattrctl = (void *)eopnotsupp, 1747 .vfs_suspendctl = (void *)eopnotsupp, 1748 .vfs_renamelock_enter = genfs_renamelock_enter, 1749 .vfs_renamelock_exit = genfs_renamelock_exit, 1750 .vfs_opv_descs = rump_opv_descs, 1751 /* vfs_refcount */ 1752 /* vfs_list */ 1753 }; 1754 1755 static int 1756 rumpfs_mountfs(struct mount *mp) 1757 { 1758 struct rumpfs_mount *rfsmp; 1759 struct rumpfs_node *rn; 1760 int error; 1761 1762 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP); 1763 1764 rn = makeprivate(VDIR, RUMPFS_DEFAULTMODE, NODEV, DEV_BSIZE, false); 1765 rn->rn_parent = rn; 1766 if ((error = makevnode(mp, rn, &rfsmp->rfsmp_rvp)) != 0) 1767 return error; 1768 1769 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT; 1770 1771 mp->mnt_data = rfsmp; 1772 mp->mnt_stat.f_namemax = RUMPFS_MAXNAMLEN; 1773 mp->mnt_stat.f_iosize = 512; 1774 mp->mnt_flag |= MNT_LOCAL; 1775 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO; 1776 mp->mnt_fs_bshift = DEV_BSHIFT; 1777 vfs_getnewfsid(mp); 1778 1779 return 0; 1780 } 1781 1782 int 1783 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen) 1784 { 1785 int error; 1786 1787 if (mp->mnt_flag & MNT_UPDATE) { 1788 return 0; 1789 } 1790 1791 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE, 1792 mp->mnt_op->vfs_name, mp, curlwp); 1793 if (error) 1794 return error; 1795 1796 return rumpfs_mountfs(mp); 1797 } 1798 1799 int 1800 rumpfs_unmount(struct mount *mp, int mntflags) 1801 { 1802 struct rumpfs_mount *rfsmp = mp->mnt_data; 1803 int flags = 0, error; 1804 1805 if (panicstr || mntflags & MNT_FORCE) 1806 flags |= FORCECLOSE; 1807 1808 if (rfsmp->rfsmp_rvp->v_usecount > 1 && (flags & FORCECLOSE) == 0) 1809 return EBUSY; 1810 1811 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0) 1812 return error; 1813 vgone(rfsmp->rfsmp_rvp); 1814 1815 kmem_free(rfsmp, sizeof(*rfsmp)); 1816 1817 return 0; 1818 } 1819 1820 int 1821 rumpfs_root(struct mount *mp, struct vnode **vpp) 1822 { 1823 struct rumpfs_mount *rfsmp = mp->mnt_data; 1824 1825 vref(rfsmp->rfsmp_rvp); 1826 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY); 1827 *vpp = rfsmp->rfsmp_rvp; 1828 return 0; 1829 } 1830 1831 int 1832 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1833 { 1834 1835 return EOPNOTSUPP; 1836 } 1837 1838 void 1839 rumpfs_init() 1840 { 1841 1842 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET); 1843 1844 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE); 1845 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE); 1846 } 1847 1848 void 1849 rumpfs_done() 1850 { 1851 1852 mutex_destroy(&reclock); 1853 mutex_destroy(&etfs_lock); 1854 } 1855 1856 int 1857 rumpfs_mountroot() 1858 { 1859 struct mount *mp; 1860 int error; 1861 1862 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) { 1863 vrele(rootvp); 1864 return error; 1865 } 1866 1867 if ((error = rumpfs_mountfs(mp)) != 0) 1868 panic("mounting rootfs failed: %d", error); 1869 1870 mountlist_append(mp); 1871 1872 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE, 1873 mp->mnt_op->vfs_name, mp, curlwp); 1874 if (error) 1875 panic("set_statvfs_info failed for rootfs: %d", error); 1876 1877 mp->mnt_flag &= ~MNT_RDONLY; 1878 vfs_unbusy(mp, false, NULL); 1879 1880 return 0; 1881 } 1882
Indexes created Wed Oct 01 07:09:59 GMT 2025