rumpfs.c revision 1.124
1 /* $NetBSD: rumpfs.c,v 1.124 2014/02/07 15:29:23 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 1.124 2014/02/07 15:29:23 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_v2_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 mutex_enter(&reclock); 801 if ((vp = rn->rn_vp)) { 802 mutex_enter(vp->v_interlock); 803 mutex_exit(&reclock); 804 if (vget(vp, 0)) { 805 goto getvnode; 806 } 807 *vpp = vp; 808 } else { 809 mutex_exit(&reclock); 810 rv = makevnode(dvp->v_mount, rn, vpp); 811 } 812 813 return rv; 814 } 815 816 static int 817 rump_check_possible(struct vnode *vp, struct rumpfs_node *rnode, 818 mode_t mode) 819 { 820 821 if ((mode & VWRITE) == 0) 822 return 0; 823 824 switch (vp->v_type) { 825 case VDIR: 826 case VLNK: 827 case VREG: 828 break; 829 default: 830 /* special file is always writable. */ 831 return 0; 832 } 833 834 return vp->v_mount->mnt_flag & MNT_RDONLY ? EROFS : 0; 835 } 836 837 static int 838 rump_check_permitted(struct vnode *vp, struct rumpfs_node *rnode, 839 mode_t mode, kauth_cred_t cred) 840 { 841 struct vattr *attr = &rnode->rn_va; 842 843 return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(mode, 844 vp->v_type, attr->va_mode), vp, NULL, genfs_can_access(vp->v_type, 845 attr->va_mode, attr->va_uid, attr->va_gid, mode, cred)); 846 } 847 848 int 849 rump_vop_access(void *v) 850 { 851 struct vop_access_args /* { 852 const struct vnodeop_desc *a_desc; 853 struct vnode *a_vp; 854 int a_mode; 855 kauth_cred_t a_cred; 856 } */ *ap = v; 857 struct vnode *vp = ap->a_vp; 858 struct rumpfs_node *rn = vp->v_data; 859 int error; 860 861 error = rump_check_possible(vp, rn, ap->a_mode); 862 if (error) 863 return error; 864 865 error = rump_check_permitted(vp, rn, ap->a_mode, ap->a_cred); 866 867 return error; 868 } 869 870 static int 871 rump_vop_getattr(void *v) 872 { 873 struct vop_getattr_args /* { 874 struct vnode *a_vp; 875 struct vattr *a_vap; 876 kauth_cred_t a_cred; 877 } */ *ap = v; 878 struct vnode *vp = ap->a_vp; 879 struct rumpfs_node *rn = vp->v_data; 880 struct vattr *vap = ap->a_vap; 881 882 memcpy(vap, &rn->rn_va, sizeof(struct vattr)); 883 vap->va_size = vp->v_size; 884 return 0; 885 } 886 887 static int 888 rump_vop_setattr(void *v) 889 { 890 struct vop_setattr_args /* { 891 struct vnode *a_vp; 892 struct vattr *a_vap; 893 kauth_cred_t a_cred; 894 } */ *ap = v; 895 struct vnode *vp = ap->a_vp; 896 struct vattr *vap = ap->a_vap; 897 struct rumpfs_node *rn = vp->v_data; 898 struct vattr *attr = &rn->rn_va; 899 kauth_cred_t cred = ap->a_cred; 900 int error; 901 902 #define CHANGED(a, t) (vap->a != (t)VNOVAL) 903 #define SETIFVAL(a,t) if (CHANGED(a, t)) rn->rn_va.a = vap->a 904 if (CHANGED(va_atime.tv_sec, time_t) || 905 CHANGED(va_ctime.tv_sec, time_t) || 906 CHANGED(va_mtime.tv_sec, time_t) || 907 CHANGED(va_birthtime.tv_sec, time_t) || 908 CHANGED(va_atime.tv_nsec, long) || 909 CHANGED(va_ctime.tv_nsec, long) || 910 CHANGED(va_mtime.tv_nsec, long) || 911 CHANGED(va_birthtime.tv_nsec, long)) { 912 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_TIMES, vp, 913 NULL, genfs_can_chtimes(vp, vap->va_vaflags, attr->va_uid, 914 cred)); 915 if (error) 916 return error; 917 } 918 919 SETIFVAL(va_atime.tv_sec, time_t); 920 SETIFVAL(va_ctime.tv_sec, time_t); 921 SETIFVAL(va_mtime.tv_sec, time_t); 922 SETIFVAL(va_birthtime.tv_sec, time_t); 923 SETIFVAL(va_atime.tv_nsec, long); 924 SETIFVAL(va_ctime.tv_nsec, long); 925 SETIFVAL(va_mtime.tv_nsec, long); 926 SETIFVAL(va_birthtime.tv_nsec, long); 927 928 if (CHANGED(va_flags, u_long)) { 929 /* XXX Can we handle system flags here...? */ 930 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_FLAGS, vp, 931 NULL, genfs_can_chflags(cred, vp->v_type, attr->va_uid, 932 false)); 933 if (error) 934 return error; 935 } 936 937 SETIFVAL(va_flags, u_long); 938 #undef SETIFVAL 939 #undef CHANGED 940 941 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (uid_t)VNOVAL) { 942 uid_t uid = 943 (vap->va_uid != (uid_t)VNOVAL) ? vap->va_uid : attr->va_uid; 944 gid_t gid = 945 (vap->va_gid != (gid_t)VNOVAL) ? vap->va_gid : attr->va_gid; 946 error = kauth_authorize_vnode(cred, 947 KAUTH_VNODE_CHANGE_OWNERSHIP, vp, NULL, 948 genfs_can_chown(cred, attr->va_uid, attr->va_gid, uid, 949 gid)); 950 if (error) 951 return error; 952 attr->va_uid = uid; 953 attr->va_gid = gid; 954 } 955 956 if (vap->va_mode != (mode_t)VNOVAL) { 957 mode_t mode = vap->va_mode; 958 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_SECURITY, 959 vp, NULL, genfs_can_chmod(vp->v_type, cred, attr->va_uid, 960 attr->va_gid, mode)); 961 if (error) 962 return error; 963 attr->va_mode = mode; 964 } 965 966 if (vp->v_type == VREG && 967 vap->va_size != VSIZENOTSET && 968 vap->va_size != rn->rn_dlen) { 969 void *newdata; 970 size_t copylen, newlen; 971 972 newlen = vap->va_size; 973 newdata = rump_hypermalloc(newlen, 0, true, "rumpfs"); 974 975 copylen = MIN(rn->rn_dlen, newlen); 976 memset(newdata, 0, newlen); 977 memcpy(newdata, rn->rn_data, copylen); 978 rump_hyperfree(rn->rn_data, rn->rn_dlen); 979 980 rn->rn_data = newdata; 981 rn->rn_dlen = newlen; 982 uvm_vnp_setsize(vp, newlen); 983 } 984 return 0; 985 } 986 987 static int 988 rump_vop_mkdir(void *v) 989 { 990 struct vop_mkdir_v3_args /* { 991 struct vnode *a_dvp; 992 struct vnode **a_vpp; 993 struct componentname *a_cnp; 994 struct vattr *a_vap; 995 }; */ *ap = v; 996 struct vnode *dvp = ap->a_dvp; 997 struct vnode **vpp = ap->a_vpp; 998 struct componentname *cnp = ap->a_cnp; 999 struct vattr *va = ap->a_vap; 1000 struct rumpfs_node *rnd = dvp->v_data, *rn; 1001 int rv = 0; 1002 1003 rn = makeprivate(VDIR, va->va_mode & ALLPERMS, NODEV, DEV_BSIZE, false); 1004 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1005 rn->rn_va.va_flags |= UF_OPAQUE; 1006 rn->rn_parent = rnd; 1007 rv = makevnode(dvp->v_mount, rn, vpp); 1008 if (rv) 1009 return rv; 1010 1011 makedir(rnd, cnp, rn); 1012 1013 return rv; 1014 } 1015 1016 static int 1017 rump_vop_rmdir(void *v) 1018 { 1019 struct vop_rmdir_args /* { 1020 struct vnode *a_dvp; 1021 struct vnode *a_vp; 1022 struct componentname *a_cnp; 1023 }; */ *ap = v; 1024 struct vnode *dvp = ap->a_dvp; 1025 struct vnode *vp = ap->a_vp; 1026 struct componentname *cnp = ap->a_cnp; 1027 struct rumpfs_node *rnd = dvp->v_data; 1028 struct rumpfs_node *rn = vp->v_data; 1029 struct rumpfs_dent *rd; 1030 int rv = 0; 1031 1032 LIST_FOREACH(rd, &rn->rn_dir, rd_entries) { 1033 if (rd->rd_node != RUMPFS_WHITEOUT) { 1034 rv = ENOTEMPTY; 1035 goto out; 1036 } 1037 } 1038 while ((rd = LIST_FIRST(&rn->rn_dir)) != NULL) { 1039 KASSERT(rd->rd_node == RUMPFS_WHITEOUT); 1040 LIST_REMOVE(rd, rd_entries); 1041 kmem_free(rd->rd_name, rd->rd_namelen+1); 1042 kmem_free(rd, sizeof(*rd)); 1043 } 1044 1045 freedir(rnd, cnp); 1046 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1047 rn->rn_parent = NULL; 1048 1049 out: 1050 vput(dvp); 1051 vput(vp); 1052 1053 return rv; 1054 } 1055 1056 static int 1057 rump_vop_remove(void *v) 1058 { 1059 struct vop_remove_args /* { 1060 struct vnode *a_dvp; 1061 struct vnode *a_vp; 1062 struct componentname *a_cnp; 1063 }; */ *ap = v; 1064 struct vnode *dvp = ap->a_dvp; 1065 struct vnode *vp = ap->a_vp; 1066 struct componentname *cnp = ap->a_cnp; 1067 struct rumpfs_node *rnd = dvp->v_data; 1068 struct rumpfs_node *rn = vp->v_data; 1069 int rv = 0; 1070 1071 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1072 return EOPNOTSUPP; 1073 1074 freedir(rnd, cnp); 1075 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1076 1077 vput(dvp); 1078 vput(vp); 1079 1080 return rv; 1081 } 1082 1083 static int 1084 rump_vop_mknod(void *v) 1085 { 1086 struct vop_mknod_v3_args /* { 1087 struct vnode *a_dvp; 1088 struct vnode **a_vpp; 1089 struct componentname *a_cnp; 1090 struct vattr *a_vap; 1091 }; */ *ap = v; 1092 struct vnode *dvp = ap->a_dvp; 1093 struct vnode **vpp = ap->a_vpp; 1094 struct componentname *cnp = ap->a_cnp; 1095 struct vattr *va = ap->a_vap; 1096 struct rumpfs_node *rnd = dvp->v_data, *rn; 1097 int rv; 1098 1099 rn = makeprivate(va->va_type, va->va_mode & ALLPERMS, va->va_rdev, 1100 DEV_BSIZE, false); 1101 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1102 rn->rn_va.va_flags |= UF_OPAQUE; 1103 rv = makevnode(dvp->v_mount, rn, vpp); 1104 if (rv) 1105 return rv; 1106 1107 makedir(rnd, cnp, rn); 1108 1109 return rv; 1110 } 1111 1112 static int 1113 rump_vop_create(void *v) 1114 { 1115 struct vop_create_v3_args /* { 1116 struct vnode *a_dvp; 1117 struct vnode **a_vpp; 1118 struct componentname *a_cnp; 1119 struct vattr *a_vap; 1120 }; */ *ap = v; 1121 struct vnode *dvp = ap->a_dvp; 1122 struct vnode **vpp = ap->a_vpp; 1123 struct componentname *cnp = ap->a_cnp; 1124 struct vattr *va = ap->a_vap; 1125 struct rumpfs_node *rnd = dvp->v_data, *rn; 1126 off_t newsize; 1127 int rv; 1128 1129 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0; 1130 rn = makeprivate(va->va_type, va->va_mode & ALLPERMS, NODEV, 1131 newsize, false); 1132 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1133 rn->rn_va.va_flags |= UF_OPAQUE; 1134 rv = makevnode(dvp->v_mount, rn, vpp); 1135 if (rv) 1136 return rv; 1137 1138 makedir(rnd, cnp, rn); 1139 1140 return rv; 1141 } 1142 1143 static int 1144 rump_vop_symlink(void *v) 1145 { 1146 struct vop_symlink_v3_args /* { 1147 struct vnode *a_dvp; 1148 struct vnode **a_vpp; 1149 struct componentname *a_cnp; 1150 struct vattr *a_vap; 1151 char *a_target; 1152 }; */ *ap = v; 1153 struct vnode *dvp = ap->a_dvp; 1154 struct vnode **vpp = ap->a_vpp; 1155 struct componentname *cnp = ap->a_cnp; 1156 struct vattr *va = ap->a_vap; 1157 struct rumpfs_node *rnd = dvp->v_data, *rn; 1158 const char *target = ap->a_target; 1159 size_t linklen; 1160 int rv; 1161 1162 linklen = strlen(target); 1163 KASSERT(linklen < MAXPATHLEN); 1164 rn = makeprivate(VLNK, va->va_mode & ALLPERMS, NODEV, linklen, false); 1165 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1166 rn->rn_va.va_flags |= UF_OPAQUE; 1167 rv = makevnode(dvp->v_mount, rn, vpp); 1168 if (rv) 1169 return rv; 1170 1171 makedir(rnd, cnp, rn); 1172 1173 KASSERT(linklen < MAXPATHLEN); 1174 rn->rn_linktarg = PNBUF_GET(); 1175 rn->rn_linklen = linklen; 1176 strcpy(rn->rn_linktarg, target); 1177 1178 return rv; 1179 } 1180 1181 static int 1182 rump_vop_readlink(void *v) 1183 { 1184 struct vop_readlink_args /* { 1185 struct vnode *a_vp; 1186 struct uio *a_uio; 1187 kauth_cred_t a_cred; 1188 }; */ *ap = v; 1189 struct vnode *vp = ap->a_vp; 1190 struct rumpfs_node *rn = vp->v_data; 1191 struct uio *uio = ap->a_uio; 1192 1193 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio); 1194 } 1195 1196 static int 1197 rump_vop_whiteout(void *v) 1198 { 1199 struct vop_whiteout_args /* { 1200 struct vnode *a_dvp; 1201 struct componentname *a_cnp; 1202 int a_flags; 1203 } */ *ap = v; 1204 struct vnode *dvp = ap->a_dvp; 1205 struct rumpfs_node *rnd = dvp->v_data; 1206 struct componentname *cnp = ap->a_cnp; 1207 int flags = ap->a_flags; 1208 1209 switch (flags) { 1210 case LOOKUP: 1211 break; 1212 case CREATE: 1213 makedir(rnd, cnp, RUMPFS_WHITEOUT); 1214 break; 1215 case DELETE: 1216 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */ 1217 freedir(rnd, cnp); 1218 break; 1219 default: 1220 panic("unknown whiteout op %d", flags); 1221 } 1222 1223 return 0; 1224 } 1225 1226 static int 1227 rump_vop_open(void *v) 1228 { 1229 struct vop_open_args /* { 1230 struct vnode *a_vp; 1231 int a_mode; 1232 kauth_cred_t a_cred; 1233 } */ *ap = v; 1234 struct vnode *vp = ap->a_vp; 1235 struct rumpfs_node *rn = vp->v_data; 1236 int mode = ap->a_mode; 1237 int error = EINVAL; 1238 1239 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0) 1240 return 0; 1241 1242 if (mode & FREAD) { 1243 if (rn->rn_readfd != -1) 1244 return 0; 1245 error = rumpuser_open(rn->rn_hostpath, 1246 RUMPUSER_OPEN_RDONLY, &rn->rn_readfd); 1247 } 1248 1249 if (mode & FWRITE) { 1250 if (rn->rn_writefd != -1) 1251 return 0; 1252 error = rumpuser_open(rn->rn_hostpath, 1253 RUMPUSER_OPEN_WRONLY, &rn->rn_writefd); 1254 } 1255 1256 return error; 1257 } 1258 1259 /* simple readdir. even omits dotstuff and periods */ 1260 static int 1261 rump_vop_readdir(void *v) 1262 { 1263 struct vop_readdir_args /* { 1264 struct vnode *a_vp; 1265 struct uio *a_uio; 1266 kauth_cred_t a_cred; 1267 int *a_eofflag; 1268 off_t **a_cookies; 1269 int *a_ncookies; 1270 } */ *ap = v; 1271 struct vnode *vp = ap->a_vp; 1272 struct uio *uio = ap->a_uio; 1273 struct rumpfs_node *rnd = vp->v_data; 1274 struct rumpfs_dent *rdent; 1275 struct dirent *dentp = NULL; 1276 unsigned i; 1277 int rv = 0; 1278 1279 /* seek to current entry */ 1280 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir); 1281 (i < uio->uio_offset) && rdent; 1282 i++, rdent = LIST_NEXT(rdent, rd_entries)) 1283 continue; 1284 if (!rdent) 1285 goto out; 1286 1287 /* copy entries */ 1288 dentp = kmem_alloc(sizeof(*dentp), KM_SLEEP); 1289 for (; rdent && uio->uio_resid > 0; 1290 rdent = LIST_NEXT(rdent, rd_entries), i++) { 1291 strlcpy(dentp->d_name, rdent->rd_name, sizeof(dentp->d_name)); 1292 dentp->d_namlen = strlen(dentp->d_name); 1293 dentp->d_reclen = _DIRENT_RECLEN(dentp, dentp->d_namlen); 1294 1295 if (__predict_false(RDENT_ISWHITEOUT(rdent))) { 1296 dentp->d_fileno = INO_WHITEOUT; 1297 dentp->d_type = DT_WHT; 1298 } else { 1299 dentp->d_fileno = rdent->rd_node->rn_va.va_fileid; 1300 dentp->d_type = vtype2dt(rdent->rd_node->rn_va.va_type); 1301 } 1302 1303 if (uio->uio_resid < dentp->d_reclen) { 1304 i--; 1305 break; 1306 } 1307 1308 rv = uiomove(dentp, dentp->d_reclen, uio); 1309 if (rv) { 1310 i--; 1311 break; 1312 } 1313 } 1314 kmem_free(dentp, sizeof(*dentp)); 1315 dentp = NULL; 1316 1317 out: 1318 KASSERT(dentp == NULL); 1319 if (ap->a_cookies) { 1320 *ap->a_ncookies = 0; 1321 *ap->a_cookies = NULL; 1322 } 1323 if (rdent) 1324 *ap->a_eofflag = 0; 1325 else 1326 *ap->a_eofflag = 1; 1327 uio->uio_offset = i; 1328 1329 return rv; 1330 } 1331 1332 static int 1333 etread(struct rumpfs_node *rn, struct uio *uio) 1334 { 1335 struct rumpuser_iovec iov; 1336 uint8_t *buf; 1337 size_t bufsize, n; 1338 int error = 0; 1339 1340 bufsize = uio->uio_resid; 1341 if (bufsize == 0) 1342 return 0; 1343 buf = kmem_alloc(bufsize, KM_SLEEP); 1344 1345 iov.iov_base = buf; 1346 iov.iov_len = bufsize; 1347 if ((error = rumpuser_iovread(rn->rn_readfd, &iov, 1, 1348 uio->uio_offset + rn->rn_offset, &n)) == 0) { 1349 KASSERT(n <= bufsize); 1350 error = uiomove(buf, n, uio); 1351 } 1352 1353 kmem_free(buf, bufsize); 1354 return error; 1355 } 1356 1357 static int 1358 rump_vop_read(void *v) 1359 { 1360 struct vop_read_args /* { 1361 struct vnode *a_vp; 1362 struct uio *a_uio; 1363 int ioflags a_ioflag; 1364 kauth_cred_t a_cred; 1365 }; */ *ap = v; 1366 struct vnode *vp = ap->a_vp; 1367 struct rumpfs_node *rn = vp->v_data; 1368 struct uio *uio = ap->a_uio; 1369 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1370 off_t chunk; 1371 int error = 0; 1372 1373 if (vp->v_type == VDIR) 1374 return EISDIR; 1375 1376 /* et op? */ 1377 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1378 return etread(rn, uio); 1379 1380 /* otherwise, it's off to ubc with us */ 1381 while (uio->uio_resid > 0) { 1382 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1383 if (chunk == 0) 1384 break; 1385 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1386 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1387 if (error) 1388 break; 1389 } 1390 1391 return error; 1392 } 1393 1394 static int 1395 etwrite(struct rumpfs_node *rn, struct uio *uio) 1396 { 1397 struct rumpuser_iovec iov; 1398 uint8_t *buf; 1399 size_t bufsize, n; 1400 int error = 0; 1401 1402 bufsize = uio->uio_resid; 1403 if (bufsize == 0) 1404 return 0; 1405 buf = kmem_alloc(bufsize, KM_SLEEP); 1406 error = uiomove(buf, bufsize, uio); 1407 if (error) 1408 goto out; 1409 1410 KASSERT(uio->uio_resid == 0); 1411 iov.iov_base = buf; 1412 iov.iov_len = bufsize; 1413 if ((error = rumpuser_iovwrite(rn->rn_writefd, &iov, 1, 1414 (uio->uio_offset-bufsize) + rn->rn_offset, &n)) == 0) { 1415 KASSERT(n <= bufsize); 1416 uio->uio_resid = bufsize - n; 1417 } 1418 1419 out: 1420 kmem_free(buf, bufsize); 1421 return error; 1422 } 1423 1424 static int 1425 rump_vop_write(void *v) 1426 { 1427 struct vop_write_args /* { 1428 struct vnode *a_vp; 1429 struct uio *a_uio; 1430 int ioflags a_ioflag; 1431 kauth_cred_t a_cred; 1432 }; */ *ap = v; 1433 struct vnode *vp = ap->a_vp; 1434 struct rumpfs_node *rn = vp->v_data; 1435 struct uio *uio = ap->a_uio; 1436 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1437 void *olddata; 1438 size_t oldlen, newlen; 1439 off_t chunk; 1440 int error = 0; 1441 bool allocd = false; 1442 1443 if (ap->a_ioflag & IO_APPEND) 1444 uio->uio_offset = vp->v_size; 1445 1446 /* consult et? */ 1447 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1448 return etwrite(rn, uio); 1449 1450 /* 1451 * Otherwise, it's a case of ubcmove. 1452 */ 1453 1454 /* 1455 * First, make sure we have enough storage. 1456 * 1457 * No, you don't need to tell me it's not very efficient. 1458 * No, it doesn't really support sparse files, just fakes it. 1459 */ 1460 newlen = uio->uio_offset + uio->uio_resid; 1461 oldlen = 0; /* XXXgcc */ 1462 olddata = NULL; 1463 if (rn->rn_dlen < newlen) { 1464 oldlen = rn->rn_dlen; 1465 olddata = rn->rn_data; 1466 1467 rn->rn_data = rump_hypermalloc(newlen, 0, true, "rumpfs"); 1468 rn->rn_dlen = newlen; 1469 memset(rn->rn_data, 0, newlen); 1470 memcpy(rn->rn_data, olddata, oldlen); 1471 allocd = true; 1472 uvm_vnp_setsize(vp, newlen); 1473 } 1474 1475 /* ok, we have enough stooorage. write */ 1476 while (uio->uio_resid > 0) { 1477 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1478 if (chunk == 0) 1479 break; 1480 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1481 UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1482 if (error) 1483 break; 1484 } 1485 1486 if (allocd) { 1487 if (error) { 1488 rump_hyperfree(rn->rn_data, newlen); 1489 rn->rn_data = olddata; 1490 rn->rn_dlen = oldlen; 1491 uvm_vnp_setsize(vp, oldlen); 1492 } else { 1493 rump_hyperfree(olddata, oldlen); 1494 } 1495 } 1496 1497 return error; 1498 } 1499 1500 static int 1501 rump_vop_bmap(void *v) 1502 { 1503 struct vop_bmap_args /* { 1504 struct vnode *a_vp; 1505 daddr_t a_bn; 1506 struct vnode **a_vpp; 1507 daddr_t *a_bnp; 1508 int *a_runp; 1509 } */ *ap = v; 1510 1511 /* 1:1 mapping */ 1512 if (ap->a_vpp) 1513 *ap->a_vpp = ap->a_vp; 1514 if (ap->a_bnp) 1515 *ap->a_bnp = ap->a_bn; 1516 if (ap->a_runp) 1517 *ap->a_runp = 16; 1518 1519 return 0; 1520 } 1521 1522 static int 1523 rump_vop_strategy(void *v) 1524 { 1525 struct vop_strategy_args /* { 1526 struct vnode *a_vp; 1527 struct buf *a_bp; 1528 } */ *ap = v; 1529 struct vnode *vp = ap->a_vp; 1530 struct rumpfs_node *rn = vp->v_data; 1531 struct buf *bp = ap->a_bp; 1532 off_t copylen, copyoff; 1533 int error; 1534 1535 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) { 1536 error = EINVAL; 1537 goto out; 1538 } 1539 1540 copyoff = bp->b_blkno << DEV_BSHIFT; 1541 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount); 1542 if (BUF_ISWRITE(bp)) { 1543 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen); 1544 } else { 1545 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen); 1546 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen); 1547 } 1548 bp->b_resid = 0; 1549 error = 0; 1550 1551 out: 1552 bp->b_error = error; 1553 biodone(bp); 1554 return 0; 1555 } 1556 1557 static int 1558 rump_vop_pathconf(void *v) 1559 { 1560 struct vop_pathconf_args /* { 1561 struct vnode *a_vp; 1562 int a_name; 1563 register_t *a_retval; 1564 }; */ *ap = v; 1565 int name = ap->a_name; 1566 register_t *retval = ap->a_retval; 1567 1568 switch (name) { 1569 case _PC_LINK_MAX: 1570 *retval = LINK_MAX; 1571 return 0; 1572 case _PC_NAME_MAX: 1573 *retval = RUMPFS_MAXNAMLEN; 1574 return 0; 1575 case _PC_PATH_MAX: 1576 *retval = PATH_MAX; 1577 return 0; 1578 case _PC_PIPE_BUF: 1579 *retval = PIPE_BUF; 1580 return 0; 1581 case _PC_CHOWN_RESTRICTED: 1582 *retval = 1; 1583 return 0; 1584 case _PC_NO_TRUNC: 1585 *retval = 1; 1586 return 0; 1587 case _PC_SYNC_IO: 1588 *retval = 1; 1589 return 0; 1590 case _PC_FILESIZEBITS: 1591 *retval = 43; /* this one goes to 11 */ 1592 return 0; 1593 case _PC_SYMLINK_MAX: 1594 *retval = MAXPATHLEN; 1595 return 0; 1596 case _PC_2_SYMLINKS: 1597 *retval = 1; 1598 return 0; 1599 default: 1600 return EINVAL; 1601 } 1602 } 1603 1604 static int 1605 rump_vop_success(void *v) 1606 { 1607 1608 return 0; 1609 } 1610 1611 static int 1612 rump_vop_inactive(void *v) 1613 { 1614 struct vop_inactive_args /* { 1615 struct vnode *a_vp; 1616 bool *a_recycle; 1617 } */ *ap = v; 1618 struct vnode *vp = ap->a_vp; 1619 struct rumpfs_node *rn = vp->v_data; 1620 1621 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) { 1622 if (rn->rn_readfd != -1) { 1623 rumpuser_close(rn->rn_readfd); 1624 rn->rn_readfd = -1; 1625 } 1626 if (rn->rn_writefd != -1) { 1627 rumpuser_close(rn->rn_writefd); 1628 rn->rn_writefd = -1; 1629 } 1630 } 1631 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false; 1632 1633 VOP_UNLOCK(vp); 1634 return 0; 1635 } 1636 1637 static int 1638 rump_vop_reclaim(void *v) 1639 { 1640 struct vop_reclaim_args /* { 1641 struct vnode *a_vp; 1642 } */ *ap = v; 1643 struct vnode *vp = ap->a_vp; 1644 struct rumpfs_node *rn = vp->v_data; 1645 1646 mutex_enter(&reclock); 1647 rn->rn_vp = NULL; 1648 mutex_exit(&reclock); 1649 genfs_node_destroy(vp); 1650 vp->v_data = NULL; 1651 1652 if (rn->rn_flags & RUMPNODE_CANRECLAIM) { 1653 if (vp->v_type == VREG 1654 && (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0 1655 && rn->rn_data) { 1656 rump_hyperfree(rn->rn_data, rn->rn_dlen); 1657 rn->rn_data = NULL; 1658 } 1659 1660 if (vp->v_type == VLNK) 1661 PNBUF_PUT(rn->rn_linktarg); 1662 if (rn->rn_hostpath) 1663 free(rn->rn_hostpath, M_TEMP); 1664 kmem_free(rn, sizeof(*rn)); 1665 } 1666 1667 return 0; 1668 } 1669 1670 static int 1671 rump_vop_spec(void *v) 1672 { 1673 struct vop_generic_args *ap = v; 1674 int (**opvec)(void *); 1675 1676 switch (ap->a_desc->vdesc_offset) { 1677 case VOP_ACCESS_DESCOFFSET: 1678 case VOP_GETATTR_DESCOFFSET: 1679 case VOP_SETATTR_DESCOFFSET: 1680 case VOP_LOCK_DESCOFFSET: 1681 case VOP_UNLOCK_DESCOFFSET: 1682 case VOP_ISLOCKED_DESCOFFSET: 1683 case VOP_RECLAIM_DESCOFFSET: 1684 opvec = rump_vnodeop_p; 1685 break; 1686 default: 1687 opvec = spec_vnodeop_p; 1688 break; 1689 } 1690 1691 return VOCALL(opvec, ap->a_desc->vdesc_offset, v); 1692 } 1693 1694 static int 1695 rump_vop_advlock(void *v) 1696 { 1697 struct vop_advlock_args /* { 1698 const struct vnodeop_desc *a_desc; 1699 struct vnode *a_vp; 1700 void *a_id; 1701 int a_op; 1702 struct flock *a_fl; 1703 int a_flags; 1704 } */ *ap = v; 1705 struct vnode *vp = ap->a_vp; 1706 struct rumpfs_node *rn = vp->v_data; 1707 1708 return lf_advlock(ap, &rn->rn_lockf, vp->v_size); 1709 } 1710 1711 /* 1712 * Begin vfs-level stuff 1713 */ 1714 1715 VFS_PROTOS(rumpfs); 1716 struct vfsops rumpfs_vfsops = { 1717 .vfs_name = MOUNT_RUMPFS, 1718 .vfs_min_mount_data = 0, 1719 .vfs_mount = rumpfs_mount, 1720 .vfs_start = (void *)nullop, 1721 .vfs_unmount = rumpfs_unmount, 1722 .vfs_root = rumpfs_root, 1723 .vfs_quotactl = (void *)eopnotsupp, 1724 .vfs_statvfs = genfs_statvfs, 1725 .vfs_sync = (void *)nullop, 1726 .vfs_vget = rumpfs_vget, 1727 .vfs_fhtovp = (void *)eopnotsupp, 1728 .vfs_vptofh = (void *)eopnotsupp, 1729 .vfs_init = rumpfs_init, 1730 .vfs_reinit = NULL, 1731 .vfs_done = rumpfs_done, 1732 .vfs_mountroot = rumpfs_mountroot, 1733 .vfs_snapshot = (void *)eopnotsupp, 1734 .vfs_extattrctl = (void *)eopnotsupp, 1735 .vfs_suspendctl = (void *)eopnotsupp, 1736 .vfs_renamelock_enter = genfs_renamelock_enter, 1737 .vfs_renamelock_exit = genfs_renamelock_exit, 1738 .vfs_opv_descs = rump_opv_descs, 1739 /* vfs_refcount */ 1740 /* vfs_list */ 1741 }; 1742 1743 static int 1744 rumpfs_mountfs(struct mount *mp) 1745 { 1746 struct rumpfs_mount *rfsmp; 1747 struct rumpfs_node *rn; 1748 int error; 1749 1750 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP); 1751 1752 rn = makeprivate(VDIR, RUMPFS_DEFAULTMODE, NODEV, DEV_BSIZE, false); 1753 rn->rn_parent = rn; 1754 if ((error = makevnode(mp, rn, &rfsmp->rfsmp_rvp)) != 0) 1755 return error; 1756 1757 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT; 1758 1759 mp->mnt_data = rfsmp; 1760 mp->mnt_stat.f_namemax = RUMPFS_MAXNAMLEN; 1761 mp->mnt_stat.f_iosize = 512; 1762 mp->mnt_flag |= MNT_LOCAL; 1763 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO; 1764 mp->mnt_fs_bshift = DEV_BSHIFT; 1765 vfs_getnewfsid(mp); 1766 1767 return 0; 1768 } 1769 1770 int 1771 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen) 1772 { 1773 int error; 1774 1775 if (mp->mnt_flag & MNT_UPDATE) { 1776 return 0; 1777 } 1778 1779 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE, 1780 mp->mnt_op->vfs_name, mp, curlwp); 1781 if (error) 1782 return error; 1783 1784 return rumpfs_mountfs(mp); 1785 } 1786 1787 int 1788 rumpfs_unmount(struct mount *mp, int mntflags) 1789 { 1790 struct rumpfs_mount *rfsmp = mp->mnt_data; 1791 int flags = 0, error; 1792 1793 if (panicstr || mntflags & MNT_FORCE) 1794 flags |= FORCECLOSE; 1795 1796 if (rfsmp->rfsmp_rvp->v_usecount > 1 && (flags & FORCECLOSE) == 0) 1797 return EBUSY; 1798 1799 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0) 1800 return error; 1801 vgone(rfsmp->rfsmp_rvp); 1802 1803 kmem_free(rfsmp, sizeof(*rfsmp)); 1804 1805 return 0; 1806 } 1807 1808 int 1809 rumpfs_root(struct mount *mp, struct vnode **vpp) 1810 { 1811 struct rumpfs_mount *rfsmp = mp->mnt_data; 1812 1813 vref(rfsmp->rfsmp_rvp); 1814 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY); 1815 *vpp = rfsmp->rfsmp_rvp; 1816 return 0; 1817 } 1818 1819 int 1820 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1821 { 1822 1823 return EOPNOTSUPP; 1824 } 1825 1826 void 1827 rumpfs_init() 1828 { 1829 1830 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET); 1831 1832 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE); 1833 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE); 1834 } 1835 1836 void 1837 rumpfs_done() 1838 { 1839 1840 mutex_destroy(&reclock); 1841 mutex_destroy(&etfs_lock); 1842 } 1843 1844 int 1845 rumpfs_mountroot() 1846 { 1847 struct mount *mp; 1848 int error; 1849 1850 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) { 1851 vrele(rootvp); 1852 return error; 1853 } 1854 1855 if ((error = rumpfs_mountfs(mp)) != 0) 1856 panic("mounting rootfs failed: %d", error); 1857 1858 mountlist_append(mp); 1859 1860 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE, 1861 mp->mnt_op->vfs_name, mp, curlwp); 1862 if (error) 1863 panic("set_statvfs_info failed for rootfs: %d", error); 1864 1865 mp->mnt_flag &= ~MNT_RDONLY; 1866 vfs_unbusy(mp, false, NULL); 1867 1868 return 0; 1869 } 1870
Indexes created Tue Sep 30 20:09:53 GMT 2025