rumpfs.c revision 1.150
1 /* $NetBSD: rumpfs.c,v 1.150 2017/05/28 16:37:16 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 1.150 2017/05/28 16:37:16 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/fstrans.h> 50 #include <sys/unistd.h> 51 52 #include <miscfs/fifofs/fifo.h> 53 #include <miscfs/specfs/specdev.h> 54 #include <miscfs/genfs/genfs.h> 55 #include <miscfs/genfs/genfs_node.h> 56 57 #include <uvm/uvm_extern.h> 58 59 #include <rump-sys/kern.h> 60 #include <rump-sys/vfs.h> 61 62 #include <rump/rumpfs.h> 63 #include <rump/rumpuser.h> 64 65 static int rump_vop_lookup(void *); 66 static int rump_vop_getattr(void *); 67 static int rump_vop_setattr(void *); 68 static int rump_vop_mkdir(void *); 69 static int rump_vop_rmdir(void *); 70 static int rump_vop_remove(void *); 71 static int rump_vop_mknod(void *); 72 static int rump_vop_create(void *); 73 static int rump_vop_inactive(void *); 74 static int rump_vop_reclaim(void *); 75 static int rump_vop_success(void *); 76 static int rump_vop_readdir(void *); 77 static int rump_vop_spec(void *); 78 static int rump_vop_read(void *); 79 static int rump_vop_write(void *); 80 static int rump_vop_open(void *); 81 static int rump_vop_symlink(void *); 82 static int rump_vop_readlink(void *); 83 static int rump_vop_whiteout(void *); 84 static int rump_vop_pathconf(void *); 85 static int rump_vop_bmap(void *); 86 static int rump_vop_strategy(void *); 87 static int rump_vop_advlock(void *); 88 static int rump_vop_access(void *); 89 static int rump_vop_fcntl(void *); 90 91 int (**fifo_vnodeop_p)(void *); 92 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = { 93 { &vop_default_desc, vn_default_error }, 94 { &vop_putpages_desc, genfs_null_putpages }, 95 { NULL, NULL } 96 }; 97 const struct vnodeopv_desc fifo_vnodeop_opv_desc = 98 { &fifo_vnodeop_p, fifo_vnodeop_entries }; 99 100 int (**rump_vnodeop_p)(void *); 101 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = { 102 { &vop_default_desc, vn_default_error }, 103 { &vop_lookup_desc, rump_vop_lookup }, 104 { &vop_getattr_desc, rump_vop_getattr }, 105 { &vop_setattr_desc, rump_vop_setattr }, 106 { &vop_mkdir_desc, rump_vop_mkdir }, 107 { &vop_rmdir_desc, rump_vop_rmdir }, 108 { &vop_remove_desc, rump_vop_remove }, 109 { &vop_mknod_desc, rump_vop_mknod }, 110 { &vop_create_desc, rump_vop_create }, 111 { &vop_symlink_desc, rump_vop_symlink }, 112 { &vop_readlink_desc, rump_vop_readlink }, 113 { &vop_access_desc, rump_vop_access }, 114 { &vop_readdir_desc, rump_vop_readdir }, 115 { &vop_read_desc, rump_vop_read }, 116 { &vop_write_desc, rump_vop_write }, 117 { &vop_open_desc, rump_vop_open }, 118 { &vop_close_desc, genfs_nullop }, 119 { &vop_seek_desc, genfs_seek }, 120 { &vop_getpages_desc, genfs_getpages }, 121 { &vop_putpages_desc, genfs_putpages }, 122 { &vop_whiteout_desc, rump_vop_whiteout }, 123 { &vop_fsync_desc, rump_vop_success }, 124 { &vop_lock_desc, genfs_lock }, 125 { &vop_unlock_desc, genfs_unlock }, 126 { &vop_islocked_desc, genfs_islocked }, 127 { &vop_inactive_desc, rump_vop_inactive }, 128 { &vop_reclaim_desc, rump_vop_reclaim }, 129 { &vop_link_desc, genfs_eopnotsupp }, 130 { &vop_pathconf_desc, rump_vop_pathconf }, 131 { &vop_bmap_desc, rump_vop_bmap }, 132 { &vop_strategy_desc, rump_vop_strategy }, 133 { &vop_advlock_desc, rump_vop_advlock }, 134 { &vop_fcntl_desc, rump_vop_fcntl }, 135 { NULL, NULL } 136 }; 137 const struct vnodeopv_desc rump_vnodeop_opv_desc = 138 { &rump_vnodeop_p, rump_vnodeop_entries }; 139 140 int (**rump_specop_p)(void *); 141 const struct vnodeopv_entry_desc rump_specop_entries[] = { 142 { &vop_default_desc, rump_vop_spec }, 143 { NULL, NULL } 144 }; 145 const struct vnodeopv_desc rump_specop_opv_desc = 146 { &rump_specop_p, rump_specop_entries }; 147 148 const struct vnodeopv_desc * const rump_opv_descs[] = { 149 &rump_vnodeop_opv_desc, 150 &rump_specop_opv_desc, 151 NULL 152 }; 153 154 #define RUMPFS_WHITEOUT ((void *)-1) 155 #define RDENT_ISWHITEOUT(rdp) (rdp->rd_node == RUMPFS_WHITEOUT) 156 struct rumpfs_dent { 157 char *rd_name; 158 int rd_namelen; 159 struct rumpfs_node *rd_node; 160 161 LIST_ENTRY(rumpfs_dent) rd_entries; 162 }; 163 164 struct genfs_ops rumpfs_genfsops = { 165 .gop_size = genfs_size, 166 .gop_write = genfs_gop_write, 167 168 /* optional */ 169 .gop_alloc = NULL, 170 .gop_markupdate = NULL, 171 }; 172 173 struct rumpfs_node { 174 struct genfs_node rn_gn; 175 struct vattr rn_va; 176 struct vnode *rn_vp; 177 char *rn_hostpath; 178 int rn_flags; 179 struct lockf *rn_lockf; 180 181 union { 182 struct { /* VREG */ 183 int readfd; 184 int writefd; 185 uint64_t offset; 186 } reg; 187 struct { 188 void *data; 189 size_t dlen; 190 } reg_noet; 191 struct { /* VDIR */ 192 LIST_HEAD(, rumpfs_dent) dents; 193 struct rumpfs_node *parent; 194 int flags; 195 } dir; 196 struct { 197 char *target; 198 size_t len; 199 } link; 200 } rn_u; 201 }; 202 #define rn_readfd rn_u.reg.readfd 203 #define rn_writefd rn_u.reg.writefd 204 #define rn_offset rn_u.reg.offset 205 #define rn_data rn_u.reg_noet.data 206 #define rn_dlen rn_u.reg_noet.dlen 207 #define rn_dir rn_u.dir.dents 208 #define rn_parent rn_u.dir.parent 209 #define rn_linktarg rn_u.link.target 210 #define rn_linklen rn_u.link.len 211 212 #define RUMPNODE_CANRECLAIM 0x01 213 #define RUMPNODE_DIR_ET 0x02 214 #define RUMPNODE_DIR_ETSUBS 0x04 215 #define RUMPNODE_ET_PHONE_HOST 0x10 216 #define RUMPNODE_EXTSTORAGE 0x20 217 218 struct rumpfs_mount { 219 struct vnode *rfsmp_rvp; 220 }; 221 222 #define INO_WHITEOUT 1 223 static int lastino = 2; 224 static kmutex_t reclock; 225 226 #define RUMPFS_DEFAULTMODE 0755 227 static void freedir(struct rumpfs_node *, struct componentname *); 228 static struct rumpfs_node *makeprivate(enum vtype, mode_t, dev_t, off_t, bool); 229 static void freeprivate(struct rumpfs_node *); 230 231 /* 232 * Extra Terrestrial stuff. We map a given key (pathname) to a file on 233 * the host FS. ET phones home only from the root node of rumpfs. 234 * 235 * When an etfs node is removed, a vnode potentially behind it is not 236 * immediately recycled. 237 */ 238 239 struct etfs { 240 char et_key[MAXPATHLEN]; 241 size_t et_keylen; 242 bool et_prefixkey; 243 bool et_removing; 244 devminor_t et_blkmin; 245 246 LIST_ENTRY(etfs) et_entries; 247 248 struct rumpfs_node *et_rn; 249 }; 250 static kmutex_t etfs_lock; 251 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list); 252 253 static enum vtype 254 ettype_to_vtype(enum rump_etfs_type et) 255 { 256 enum vtype vt; 257 258 switch (et) { 259 case RUMP_ETFS_REG: 260 vt = VREG; 261 break; 262 case RUMP_ETFS_BLK: 263 vt = VBLK; 264 break; 265 case RUMP_ETFS_CHR: 266 vt = VCHR; 267 break; 268 case RUMP_ETFS_DIR: 269 vt = VDIR; 270 break; 271 case RUMP_ETFS_DIR_SUBDIRS: 272 vt = VDIR; 273 break; 274 default: 275 panic("invalid et type: %d", et); 276 } 277 278 return vt; 279 } 280 281 static enum vtype 282 hft_to_vtype(int hft) 283 { 284 enum vtype vt; 285 286 switch (hft) { 287 case RUMPUSER_FT_OTHER: 288 vt = VNON; 289 break; 290 case RUMPUSER_FT_DIR: 291 vt = VDIR; 292 break; 293 case RUMPUSER_FT_REG: 294 vt = VREG; 295 break; 296 case RUMPUSER_FT_BLK: 297 vt = VBLK; 298 break; 299 case RUMPUSER_FT_CHR: 300 vt = VCHR; 301 break; 302 default: 303 vt = VNON; 304 break; 305 } 306 307 return vt; 308 } 309 310 static bool 311 etfs_find(const char *key, struct etfs **etp, bool forceprefix) 312 { 313 struct etfs *et; 314 size_t keylen = strlen(key); 315 316 KASSERT(mutex_owned(&etfs_lock)); 317 318 LIST_FOREACH(et, &etfs_list, et_entries) { 319 if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix) 320 && strncmp(key, et->et_key, et->et_keylen) == 0) { 321 if (etp) 322 *etp = et; 323 return true; 324 } 325 } 326 327 return false; 328 } 329 330 #define REGDIR(ftype) \ 331 ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS) 332 static int 333 etfsregister(const char *key, const char *hostpath, 334 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 335 { 336 char buf[9]; 337 struct etfs *et; 338 struct rumpfs_node *rn; 339 uint64_t fsize; 340 dev_t rdev = NODEV; 341 devminor_t dmin = -1; 342 int hft, error; 343 344 if (key[0] != '/') { 345 return EINVAL; 346 } 347 while (key[0] == '/') { 348 key++; 349 } 350 351 if ((error = rumpuser_getfileinfo(hostpath, &fsize, &hft)) != 0) 352 return error; 353 354 /* etfs directory requires a directory on the host */ 355 if (REGDIR(ftype)) { 356 if (hft != RUMPUSER_FT_DIR) 357 return ENOTDIR; 358 if (begin != 0) 359 return EISDIR; 360 if (size != RUMP_ETFS_SIZE_ENDOFF) 361 return EISDIR; 362 size = fsize; 363 } else { 364 if (begin > fsize) 365 return EINVAL; 366 if (size == RUMP_ETFS_SIZE_ENDOFF) 367 size = fsize - begin; 368 if (begin + size > fsize) 369 return EINVAL; 370 } 371 372 if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) { 373 error = rumpblk_register(hostpath, &dmin, begin, size); 374 if (error != 0) { 375 return error; 376 } 377 rdev = makedev(RUMPBLK_DEVMAJOR, dmin); 378 } 379 380 et = kmem_alloc(sizeof(*et), KM_SLEEP); 381 strcpy(et->et_key, key); 382 et->et_keylen = strlen(et->et_key); 383 et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), RUMPFS_DEFAULTMODE, 384 rdev, size, true); 385 et->et_removing = false; 386 et->et_blkmin = dmin; 387 388 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 389 390 if (ftype == RUMP_ETFS_REG || REGDIR(ftype) || et->et_blkmin != -1) { 391 size_t len = strlen(hostpath)+1; 392 393 rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 394 memcpy(rn->rn_hostpath, hostpath, len); 395 rn->rn_offset = begin; 396 } 397 398 if (REGDIR(ftype)) { 399 rn->rn_flags |= RUMPNODE_DIR_ET; 400 et->et_prefixkey = true; 401 } else { 402 et->et_prefixkey = false; 403 } 404 405 if (ftype == RUMP_ETFS_DIR_SUBDIRS) 406 rn->rn_flags |= RUMPNODE_DIR_ETSUBS; 407 408 mutex_enter(&etfs_lock); 409 if (etfs_find(key, NULL, REGDIR(ftype))) { 410 mutex_exit(&etfs_lock); 411 if (et->et_blkmin != -1) 412 rumpblk_deregister(hostpath); 413 if (et->et_rn->rn_hostpath != NULL) 414 free(et->et_rn->rn_hostpath, M_TEMP); 415 freeprivate(et->et_rn); 416 kmem_free(et, sizeof(*et)); 417 return EEXIST; 418 } 419 LIST_INSERT_HEAD(&etfs_list, et, et_entries); 420 mutex_exit(&etfs_lock); 421 422 if (ftype == RUMP_ETFS_BLK) { 423 format_bytes(buf, sizeof(buf), size); 424 aprint_verbose("/%s: hostpath %s (%s)\n", key, hostpath, buf); 425 } 426 427 return 0; 428 } 429 #undef REGDIR 430 431 /* remove etfs mapping. caller's responsibility to make sure it's not in use */ 432 static int 433 etfsremove(const char *key) 434 { 435 struct etfs *et; 436 size_t keylen; 437 int rv __diagused; 438 439 if (key[0] != '/') { 440 return EINVAL; 441 } 442 while (key[0] == '/') { 443 key++; 444 } 445 446 keylen = strlen(key); 447 448 mutex_enter(&etfs_lock); 449 LIST_FOREACH(et, &etfs_list, et_entries) { 450 if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) { 451 if (et->et_removing) 452 et = NULL; 453 else 454 et->et_removing = true; 455 break; 456 } 457 } 458 mutex_exit(&etfs_lock); 459 if (!et) 460 return ENOENT; 461 462 /* 463 * ok, we know what we want to remove and have signalled there 464 * actually are men at work. first, unregister from rumpblk 465 */ 466 if (et->et_blkmin != -1) { 467 rv = rumpblk_deregister(et->et_rn->rn_hostpath); 468 } else { 469 rv = 0; 470 } 471 KASSERT(rv == 0); 472 473 /* then do the actual removal */ 474 mutex_enter(&etfs_lock); 475 LIST_REMOVE(et, et_entries); 476 mutex_exit(&etfs_lock); 477 478 /* node is unreachable, safe to nuke all device copies */ 479 if (et->et_blkmin != -1) { 480 vdevgone(RUMPBLK_DEVMAJOR, et->et_blkmin, et->et_blkmin, VBLK); 481 } else { 482 struct vnode *vp; 483 struct mount *mp; 484 struct rumpfs_node *rn; 485 486 mutex_enter(&reclock); 487 if ((vp = et->et_rn->rn_vp) != NULL) { 488 mp = vp->v_mount; 489 rn = vp->v_data; 490 KASSERT(rn == et->et_rn); 491 } else { 492 mp = NULL; 493 } 494 mutex_exit(&reclock); 495 if (mp && vcache_get(mp, &rn, sizeof(rn), &vp) == 0) { 496 rv = vfs_suspend(mp, 0); 497 KASSERT(rv == 0); 498 vgone(vp); 499 vfs_resume(mp); 500 } 501 } 502 503 if (et->et_rn->rn_hostpath != NULL) 504 free(et->et_rn->rn_hostpath, M_TEMP); 505 freeprivate(et->et_rn); 506 kmem_free(et, sizeof(*et)); 507 508 return 0; 509 } 510 511 /* 512 * rumpfs 513 */ 514 515 static struct rumpfs_node * 516 makeprivate(enum vtype vt, mode_t mode, dev_t rdev, off_t size, bool et) 517 { 518 struct rumpfs_node *rn; 519 struct vattr *va; 520 struct timespec ts; 521 522 KASSERT((mode & ~ALLPERMS) == 0); 523 rn = kmem_zalloc(sizeof(*rn), KM_SLEEP); 524 525 switch (vt) { 526 case VDIR: 527 LIST_INIT(&rn->rn_dir); 528 break; 529 case VREG: 530 if (et) { 531 rn->rn_readfd = -1; 532 rn->rn_writefd = -1; 533 } 534 break; 535 default: 536 break; 537 } 538 539 nanotime(&ts); 540 541 va = &rn->rn_va; 542 va->va_type = vt; 543 va->va_mode = mode; 544 if (vt == VDIR) 545 va->va_nlink = 2; 546 else 547 va->va_nlink = 1; 548 va->va_uid = 0; 549 va->va_gid = 0; 550 va->va_fsid = 551 va->va_fileid = atomic_inc_uint_nv(&lastino); 552 va->va_size = size; 553 va->va_blocksize = 512; 554 va->va_atime = ts; 555 va->va_mtime = ts; 556 va->va_ctime = ts; 557 va->va_birthtime = ts; 558 va->va_gen = 0; 559 va->va_flags = 0; 560 va->va_rdev = rdev; 561 va->va_bytes = 512; 562 va->va_filerev = 0; 563 va->va_vaflags = 0; 564 565 return rn; 566 } 567 568 static void 569 freeprivate(struct rumpfs_node *rn) 570 { 571 572 kmem_free(rn, sizeof(*rn)); 573 } 574 575 static void 576 makedir(struct rumpfs_node *rnd, 577 struct componentname *cnp, struct rumpfs_node *rn) 578 { 579 struct rumpfs_dent *rdent; 580 581 rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP); 582 rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP); 583 rdent->rd_node = rn; 584 strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1); 585 rdent->rd_namelen = strlen(rdent->rd_name); 586 587 if ((cnp->cn_flags & ISWHITEOUT) != 0) { 588 KASSERT((cnp->cn_flags & DOWHITEOUT) == 0); 589 freedir(rnd, cnp); 590 } 591 LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries); 592 } 593 594 static void 595 freedir(struct rumpfs_node *rnd, struct componentname *cnp) 596 { 597 struct rumpfs_dent *rd = NULL; 598 599 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 600 if (rd->rd_namelen == cnp->cn_namelen && 601 strncmp(rd->rd_name, cnp->cn_nameptr, 602 cnp->cn_namelen) == 0) 603 break; 604 } 605 if (rd == NULL) 606 panic("could not find directory entry: %s", cnp->cn_nameptr); 607 608 if (cnp->cn_flags & DOWHITEOUT) { 609 rd->rd_node = RUMPFS_WHITEOUT; 610 } else { 611 LIST_REMOVE(rd, rd_entries); 612 kmem_free(rd->rd_name, rd->rd_namelen+1); 613 kmem_free(rd, sizeof(*rd)); 614 } 615 } 616 617 /* 618 * Simple lookup for rump file systems. 619 * 620 * uhm, this is twisted. C F C C, hope of C C F C looming 621 */ 622 static int 623 rump_vop_lookup(void *v) 624 { 625 struct vop_lookup_v2_args /* { 626 struct vnode *a_dvp; 627 struct vnode **a_vpp; 628 struct componentname *a_cnp; 629 }; */ *ap = v; 630 struct componentname *cnp = ap->a_cnp; 631 struct vnode *dvp = ap->a_dvp; 632 struct vnode **vpp = ap->a_vpp; 633 struct rumpfs_node *rnd = dvp->v_data, *rn; 634 struct rumpfs_dent *rd = NULL; 635 struct etfs *et; 636 bool dotdot = (cnp->cn_flags & ISDOTDOT) != 0; 637 int rv = 0; 638 const char *cp; 639 640 *vpp = NULL; 641 642 rv = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 643 if (rv) 644 return rv; 645 646 if ((cnp->cn_flags & ISLASTCN) 647 && (dvp->v_mount->mnt_flag & MNT_RDONLY) 648 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 649 return EROFS; 650 651 /* check for dot, return directly if the case */ 652 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 653 vref(dvp); 654 *vpp = dvp; 655 return 0; 656 } 657 658 /* we don't do rename */ 659 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME))) 660 return EOPNOTSUPP; 661 662 /* check for etfs */ 663 if (dvp == rootvnode && 664 (cnp->cn_nameiop == LOOKUP || cnp->cn_nameiop == CREATE)) { 665 bool found; 666 mutex_enter(&etfs_lock); 667 found = etfs_find(cnp->cn_nameptr, &et, false); 668 mutex_exit(&etfs_lock); 669 670 if (found) { 671 rn = et->et_rn; 672 cnp->cn_consume += et->et_keylen - cnp->cn_namelen; 673 /* 674 * consume trailing slashes if any and clear 675 * REQUIREDIR if we consumed the full path. 676 */ 677 cp = &cnp->cn_nameptr[cnp->cn_namelen]; 678 cp += cnp->cn_consume; 679 KASSERT(*cp == '\0' || *cp == '/'); 680 if (*cp == '\0' && rn->rn_va.va_type != VDIR) 681 cnp->cn_flags &= ~REQUIREDIR; 682 while (*cp++ == '/') 683 cnp->cn_consume++; 684 goto getvnode; 685 } 686 } 687 688 if (rnd->rn_flags & RUMPNODE_DIR_ET) { 689 uint64_t fsize; 690 char *newpath; 691 size_t newpathlen; 692 int hft, error; 693 694 if (dotdot) 695 return EOPNOTSUPP; 696 697 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1; 698 newpath = malloc(newpathlen, M_TEMP, M_WAITOK); 699 700 strlcpy(newpath, rnd->rn_hostpath, newpathlen); 701 strlcat(newpath, "/", newpathlen); 702 strlcat(newpath, cnp->cn_nameptr, newpathlen); 703 704 if ((error = rumpuser_getfileinfo(newpath, &fsize, &hft)) != 0){ 705 free(newpath, M_TEMP); 706 return error; 707 } 708 709 /* allow only dirs and regular files */ 710 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) { 711 free(newpath, M_TEMP); 712 return ENOENT; 713 } 714 715 rn = makeprivate(hft_to_vtype(hft), RUMPFS_DEFAULTMODE, 716 NODEV, fsize, true); 717 rn->rn_flags |= RUMPNODE_CANRECLAIM; 718 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) { 719 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS; 720 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 721 } 722 rn->rn_hostpath = newpath; 723 724 goto getvnode; 725 } else { 726 if (dotdot) { 727 if ((rn = rnd->rn_parent) != NULL) 728 goto getvnode; 729 } else { 730 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 731 if (rd->rd_namelen == cnp->cn_namelen && 732 strncmp(rd->rd_name, cnp->cn_nameptr, 733 cnp->cn_namelen) == 0) 734 break; 735 } 736 } 737 } 738 739 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE)) 740 return ENOENT; 741 742 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 743 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 744 return EROFS; 745 rv = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred); 746 if (rv) 747 return rv; 748 return EJUSTRETURN; 749 } 750 751 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == DELETE) { 752 rv = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred); 753 if (rv) 754 return rv; 755 } 756 757 if (RDENT_ISWHITEOUT(rd)) { 758 cnp->cn_flags |= ISWHITEOUT; 759 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) 760 return EJUSTRETURN; 761 return ENOENT; 762 } 763 764 rn = rd->rd_node; 765 766 getvnode: 767 KASSERT(rn); 768 rv = vcache_get(dvp->v_mount, &rn, sizeof(rn), vpp); 769 if (rv) { 770 if (rnd->rn_flags & RUMPNODE_DIR_ET) 771 freeprivate(rn); 772 return rv; 773 } 774 775 return 0; 776 } 777 778 static int 779 rump_check_possible(struct vnode *vp, struct rumpfs_node *rnode, 780 mode_t mode) 781 { 782 783 if ((mode & VWRITE) == 0) 784 return 0; 785 786 switch (vp->v_type) { 787 case VDIR: 788 case VLNK: 789 case VREG: 790 break; 791 default: 792 /* special file is always writable. */ 793 return 0; 794 } 795 796 return vp->v_mount->mnt_flag & MNT_RDONLY ? EROFS : 0; 797 } 798 799 static int 800 rump_check_permitted(struct vnode *vp, struct rumpfs_node *rnode, 801 mode_t mode, kauth_cred_t cred) 802 { 803 struct vattr *attr = &rnode->rn_va; 804 805 return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(mode, 806 vp->v_type, attr->va_mode), vp, NULL, genfs_can_access(vp->v_type, 807 attr->va_mode, attr->va_uid, attr->va_gid, mode, cred)); 808 } 809 810 int 811 rump_vop_access(void *v) 812 { 813 struct vop_access_args /* { 814 const struct vnodeop_desc *a_desc; 815 struct vnode *a_vp; 816 int a_mode; 817 kauth_cred_t a_cred; 818 } */ *ap = v; 819 struct vnode *vp = ap->a_vp; 820 struct rumpfs_node *rn = vp->v_data; 821 int error; 822 823 error = rump_check_possible(vp, rn, ap->a_mode); 824 if (error) 825 return error; 826 827 error = rump_check_permitted(vp, rn, ap->a_mode, ap->a_cred); 828 829 return error; 830 } 831 832 static int 833 rump_vop_getattr(void *v) 834 { 835 struct vop_getattr_args /* { 836 struct vnode *a_vp; 837 struct vattr *a_vap; 838 kauth_cred_t a_cred; 839 } */ *ap = v; 840 struct vnode *vp = ap->a_vp; 841 struct rumpfs_node *rn = vp->v_data; 842 struct vattr *vap = ap->a_vap; 843 844 memcpy(vap, &rn->rn_va, sizeof(struct vattr)); 845 vap->va_size = vp->v_size; 846 return 0; 847 } 848 849 static int 850 rump_vop_setattr(void *v) 851 { 852 struct vop_setattr_args /* { 853 struct vnode *a_vp; 854 struct vattr *a_vap; 855 kauth_cred_t a_cred; 856 } */ *ap = v; 857 struct vnode *vp = ap->a_vp; 858 struct vattr *vap = ap->a_vap; 859 struct rumpfs_node *rn = vp->v_data; 860 struct vattr *attr = &rn->rn_va; 861 kauth_cred_t cred = ap->a_cred; 862 int error; 863 864 #define CHANGED(a, t) (vap->a != (t)VNOVAL) 865 #define SETIFVAL(a,t) if (CHANGED(a, t)) rn->rn_va.a = vap->a 866 if (CHANGED(va_atime.tv_sec, time_t) || 867 CHANGED(va_ctime.tv_sec, time_t) || 868 CHANGED(va_mtime.tv_sec, time_t) || 869 CHANGED(va_birthtime.tv_sec, time_t) || 870 CHANGED(va_atime.tv_nsec, long) || 871 CHANGED(va_ctime.tv_nsec, long) || 872 CHANGED(va_mtime.tv_nsec, long) || 873 CHANGED(va_birthtime.tv_nsec, long)) { 874 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_TIMES, vp, 875 NULL, genfs_can_chtimes(vp, vap->va_vaflags, attr->va_uid, 876 cred)); 877 if (error) 878 return error; 879 } 880 881 SETIFVAL(va_atime.tv_sec, time_t); 882 SETIFVAL(va_ctime.tv_sec, time_t); 883 SETIFVAL(va_mtime.tv_sec, time_t); 884 SETIFVAL(va_birthtime.tv_sec, time_t); 885 SETIFVAL(va_atime.tv_nsec, long); 886 SETIFVAL(va_ctime.tv_nsec, long); 887 SETIFVAL(va_mtime.tv_nsec, long); 888 SETIFVAL(va_birthtime.tv_nsec, long); 889 890 if (CHANGED(va_flags, u_long)) { 891 /* XXX Can we handle system flags here...? */ 892 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_FLAGS, vp, 893 NULL, genfs_can_chflags(cred, vp->v_type, attr->va_uid, 894 false)); 895 if (error) 896 return error; 897 } 898 899 SETIFVAL(va_flags, u_long); 900 #undef SETIFVAL 901 #undef CHANGED 902 903 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (uid_t)VNOVAL) { 904 uid_t uid = 905 (vap->va_uid != (uid_t)VNOVAL) ? vap->va_uid : attr->va_uid; 906 gid_t gid = 907 (vap->va_gid != (gid_t)VNOVAL) ? vap->va_gid : attr->va_gid; 908 error = kauth_authorize_vnode(cred, 909 KAUTH_VNODE_CHANGE_OWNERSHIP, vp, NULL, 910 genfs_can_chown(cred, attr->va_uid, attr->va_gid, uid, 911 gid)); 912 if (error) 913 return error; 914 attr->va_uid = uid; 915 attr->va_gid = gid; 916 } 917 918 if (vap->va_mode != (mode_t)VNOVAL) { 919 mode_t mode = vap->va_mode; 920 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_SECURITY, 921 vp, NULL, genfs_can_chmod(vp->v_type, cred, attr->va_uid, 922 attr->va_gid, mode)); 923 if (error) 924 return error; 925 attr->va_mode = mode; 926 } 927 928 if (vp->v_type == VREG && 929 vap->va_size != VSIZENOTSET && 930 vap->va_size != rn->rn_dlen && 931 (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0) { 932 void *newdata; 933 size_t copylen, newlen; 934 935 newlen = vap->va_size; 936 newdata = rump_hypermalloc(newlen, 0, false, "rumpfs"); 937 if (newdata == NULL) 938 return ENOSPC; 939 940 copylen = MIN(rn->rn_dlen, newlen); 941 memcpy(newdata, rn->rn_data, copylen); 942 memset((char *)newdata + copylen, 0, newlen - copylen); 943 944 if ((rn->rn_flags & RUMPNODE_EXTSTORAGE) == 0) { 945 rump_hyperfree(rn->rn_data, rn->rn_dlen); 946 } else { 947 rn->rn_flags &= ~RUMPNODE_EXTSTORAGE; 948 } 949 950 rn->rn_data = newdata; 951 rn->rn_dlen = newlen; 952 uvm_vnp_setsize(vp, newlen); 953 } 954 return 0; 955 } 956 957 static int 958 rump_vop_mkdir(void *v) 959 { 960 struct vop_mkdir_v3_args /* { 961 struct vnode *a_dvp; 962 struct vnode **a_vpp; 963 struct componentname *a_cnp; 964 struct vattr *a_vap; 965 }; */ *ap = v; 966 struct vnode *dvp = ap->a_dvp; 967 struct vnode **vpp = ap->a_vpp; 968 struct componentname *cnp = ap->a_cnp; 969 struct vattr *va = ap->a_vap; 970 struct rumpfs_node *rnd = dvp->v_data, *rn; 971 int rv = 0; 972 973 rn = makeprivate(VDIR, va->va_mode & ALLPERMS, NODEV, DEV_BSIZE, false); 974 if ((cnp->cn_flags & ISWHITEOUT) != 0) 975 rn->rn_va.va_flags |= UF_OPAQUE; 976 rn->rn_parent = rnd; 977 rv = vcache_get(dvp->v_mount, &rn, sizeof(rn), vpp); 978 if (rv) { 979 freeprivate(rn); 980 return rv; 981 } 982 983 makedir(rnd, cnp, rn); 984 985 return rv; 986 } 987 988 static int 989 rump_vop_rmdir(void *v) 990 { 991 struct vop_rmdir_v2_args /* { 992 struct vnode *a_dvp; 993 struct vnode *a_vp; 994 struct componentname *a_cnp; 995 }; */ *ap = v; 996 struct vnode *dvp = ap->a_dvp; 997 struct vnode *vp = ap->a_vp; 998 struct componentname *cnp = ap->a_cnp; 999 struct rumpfs_node *rnd = dvp->v_data; 1000 struct rumpfs_node *rn = vp->v_data; 1001 struct rumpfs_dent *rd; 1002 int rv = 0; 1003 1004 LIST_FOREACH(rd, &rn->rn_dir, rd_entries) { 1005 if (rd->rd_node != RUMPFS_WHITEOUT) { 1006 rv = ENOTEMPTY; 1007 goto out; 1008 } 1009 } 1010 while ((rd = LIST_FIRST(&rn->rn_dir)) != NULL) { 1011 KASSERT(rd->rd_node == RUMPFS_WHITEOUT); 1012 LIST_REMOVE(rd, rd_entries); 1013 kmem_free(rd->rd_name, rd->rd_namelen+1); 1014 kmem_free(rd, sizeof(*rd)); 1015 } 1016 1017 freedir(rnd, cnp); 1018 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1019 rn->rn_parent = NULL; 1020 rn->rn_va.va_nlink = 0; 1021 1022 out: 1023 vput(vp); 1024 return rv; 1025 } 1026 1027 static int 1028 rump_vop_remove(void *v) 1029 { 1030 struct vop_remove_v2_args /* { 1031 struct vnode *a_dvp; 1032 struct vnode *a_vp; 1033 struct componentname *a_cnp; 1034 }; */ *ap = v; 1035 struct vnode *dvp = ap->a_dvp; 1036 struct vnode *vp = ap->a_vp; 1037 struct componentname *cnp = ap->a_cnp; 1038 struct rumpfs_node *rnd = dvp->v_data; 1039 struct rumpfs_node *rn = vp->v_data; 1040 int rv = 0; 1041 1042 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1043 return EOPNOTSUPP; 1044 1045 freedir(rnd, cnp); 1046 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1047 rn->rn_va.va_nlink = 0; 1048 1049 vput(vp); 1050 return rv; 1051 } 1052 1053 static int 1054 rump_vop_mknod(void *v) 1055 { 1056 struct vop_mknod_v3_args /* { 1057 struct vnode *a_dvp; 1058 struct vnode **a_vpp; 1059 struct componentname *a_cnp; 1060 struct vattr *a_vap; 1061 }; */ *ap = v; 1062 struct vnode *dvp = ap->a_dvp; 1063 struct vnode **vpp = ap->a_vpp; 1064 struct componentname *cnp = ap->a_cnp; 1065 struct vattr *va = ap->a_vap; 1066 struct rumpfs_node *rnd = dvp->v_data, *rn; 1067 int rv; 1068 1069 rn = makeprivate(va->va_type, va->va_mode & ALLPERMS, va->va_rdev, 1070 DEV_BSIZE, false); 1071 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1072 rn->rn_va.va_flags |= UF_OPAQUE; 1073 rv = vcache_get(dvp->v_mount, &rn, sizeof(rn), vpp); 1074 if (rv) { 1075 freeprivate(rn); 1076 return rv; 1077 } 1078 1079 makedir(rnd, cnp, rn); 1080 1081 return rv; 1082 } 1083 1084 static int 1085 rump_vop_create(void *v) 1086 { 1087 struct vop_create_v3_args /* { 1088 struct vnode *a_dvp; 1089 struct vnode **a_vpp; 1090 struct componentname *a_cnp; 1091 struct vattr *a_vap; 1092 }; */ *ap = v; 1093 struct vnode *dvp = ap->a_dvp; 1094 struct vnode **vpp = ap->a_vpp; 1095 struct componentname *cnp = ap->a_cnp; 1096 struct vattr *va = ap->a_vap; 1097 struct rumpfs_node *rnd = dvp->v_data, *rn; 1098 off_t newsize; 1099 int rv; 1100 1101 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0; 1102 rn = makeprivate(va->va_type, va->va_mode & ALLPERMS, NODEV, 1103 newsize, false); 1104 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1105 rn->rn_va.va_flags |= UF_OPAQUE; 1106 rv = vcache_get(dvp->v_mount, &rn, sizeof(rn), vpp); 1107 if (rv) { 1108 freeprivate(rn); 1109 return rv; 1110 } 1111 1112 makedir(rnd, cnp, rn); 1113 1114 return rv; 1115 } 1116 1117 static int 1118 rump_vop_symlink(void *v) 1119 { 1120 struct vop_symlink_v3_args /* { 1121 struct vnode *a_dvp; 1122 struct vnode **a_vpp; 1123 struct componentname *a_cnp; 1124 struct vattr *a_vap; 1125 char *a_target; 1126 }; */ *ap = v; 1127 struct vnode *dvp = ap->a_dvp; 1128 struct vnode **vpp = ap->a_vpp; 1129 struct componentname *cnp = ap->a_cnp; 1130 struct vattr *va = ap->a_vap; 1131 struct rumpfs_node *rnd = dvp->v_data, *rn; 1132 const char *target = ap->a_target; 1133 size_t linklen; 1134 int rv; 1135 1136 linklen = strlen(target); 1137 KASSERT(linklen < MAXPATHLEN); 1138 rn = makeprivate(VLNK, va->va_mode & ALLPERMS, NODEV, linklen, false); 1139 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1140 rn->rn_va.va_flags |= UF_OPAQUE; 1141 rv = vcache_get(dvp->v_mount, &rn, sizeof(rn), vpp); 1142 if (rv) { 1143 freeprivate(rn); 1144 return rv; 1145 } 1146 1147 makedir(rnd, cnp, rn); 1148 1149 KASSERT(linklen < MAXPATHLEN); 1150 rn->rn_linktarg = PNBUF_GET(); 1151 rn->rn_linklen = linklen; 1152 strcpy(rn->rn_linktarg, target); 1153 1154 return rv; 1155 } 1156 1157 static int 1158 rump_vop_readlink(void *v) 1159 { 1160 struct vop_readlink_args /* { 1161 struct vnode *a_vp; 1162 struct uio *a_uio; 1163 kauth_cred_t a_cred; 1164 }; */ *ap = v; 1165 struct vnode *vp = ap->a_vp; 1166 struct rumpfs_node *rn = vp->v_data; 1167 struct uio *uio = ap->a_uio; 1168 1169 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio); 1170 } 1171 1172 static int 1173 rump_vop_whiteout(void *v) 1174 { 1175 struct vop_whiteout_args /* { 1176 struct vnode *a_dvp; 1177 struct componentname *a_cnp; 1178 int a_flags; 1179 } */ *ap = v; 1180 struct vnode *dvp = ap->a_dvp; 1181 struct rumpfs_node *rnd = dvp->v_data; 1182 struct componentname *cnp = ap->a_cnp; 1183 int flags = ap->a_flags; 1184 1185 switch (flags) { 1186 case LOOKUP: 1187 break; 1188 case CREATE: 1189 makedir(rnd, cnp, RUMPFS_WHITEOUT); 1190 break; 1191 case DELETE: 1192 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */ 1193 freedir(rnd, cnp); 1194 break; 1195 default: 1196 panic("unknown whiteout op %d", flags); 1197 } 1198 1199 return 0; 1200 } 1201 1202 static int 1203 rump_vop_open(void *v) 1204 { 1205 struct vop_open_args /* { 1206 struct vnode *a_vp; 1207 int a_mode; 1208 kauth_cred_t a_cred; 1209 } */ *ap = v; 1210 struct vnode *vp = ap->a_vp; 1211 struct rumpfs_node *rn = vp->v_data; 1212 int mode = ap->a_mode; 1213 int error = EINVAL; 1214 1215 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0) 1216 return 0; 1217 1218 if (mode & FREAD) { 1219 if (rn->rn_readfd != -1) 1220 return 0; 1221 error = rumpuser_open(rn->rn_hostpath, 1222 RUMPUSER_OPEN_RDONLY, &rn->rn_readfd); 1223 } 1224 1225 if (mode & FWRITE) { 1226 if (rn->rn_writefd != -1) 1227 return 0; 1228 error = rumpuser_open(rn->rn_hostpath, 1229 RUMPUSER_OPEN_WRONLY, &rn->rn_writefd); 1230 } 1231 1232 return error; 1233 } 1234 1235 /* simple readdir. even omits dotstuff and periods */ 1236 static int 1237 rump_vop_readdir(void *v) 1238 { 1239 struct vop_readdir_args /* { 1240 struct vnode *a_vp; 1241 struct uio *a_uio; 1242 kauth_cred_t a_cred; 1243 int *a_eofflag; 1244 off_t **a_cookies; 1245 int *a_ncookies; 1246 } */ *ap = v; 1247 struct vnode *vp = ap->a_vp; 1248 struct uio *uio = ap->a_uio; 1249 struct rumpfs_node *rnd = vp->v_data; 1250 struct rumpfs_dent *rdent; 1251 struct dirent *dentp = NULL; 1252 unsigned i; 1253 int rv = 0; 1254 1255 /* seek to current entry */ 1256 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir); 1257 (i < uio->uio_offset) && rdent; 1258 i++, rdent = LIST_NEXT(rdent, rd_entries)) 1259 continue; 1260 if (!rdent) 1261 goto out; 1262 1263 /* copy entries */ 1264 dentp = kmem_alloc(sizeof(*dentp), KM_SLEEP); 1265 for (; rdent && uio->uio_resid > 0; 1266 rdent = LIST_NEXT(rdent, rd_entries), i++) { 1267 strlcpy(dentp->d_name, rdent->rd_name, sizeof(dentp->d_name)); 1268 dentp->d_namlen = strlen(dentp->d_name); 1269 dentp->d_reclen = _DIRENT_RECLEN(dentp, dentp->d_namlen); 1270 1271 if (__predict_false(RDENT_ISWHITEOUT(rdent))) { 1272 dentp->d_fileno = INO_WHITEOUT; 1273 dentp->d_type = DT_WHT; 1274 } else { 1275 dentp->d_fileno = rdent->rd_node->rn_va.va_fileid; 1276 dentp->d_type = vtype2dt(rdent->rd_node->rn_va.va_type); 1277 } 1278 1279 if (uio->uio_resid < dentp->d_reclen) { 1280 i--; 1281 break; 1282 } 1283 1284 rv = uiomove(dentp, dentp->d_reclen, uio); 1285 if (rv) { 1286 i--; 1287 break; 1288 } 1289 } 1290 kmem_free(dentp, sizeof(*dentp)); 1291 dentp = NULL; 1292 1293 out: 1294 KASSERT(dentp == NULL); 1295 if (ap->a_cookies) { 1296 *ap->a_ncookies = 0; 1297 *ap->a_cookies = NULL; 1298 } 1299 if (rdent) 1300 *ap->a_eofflag = 0; 1301 else 1302 *ap->a_eofflag = 1; 1303 uio->uio_offset = i; 1304 1305 return rv; 1306 } 1307 1308 static int 1309 etread(struct rumpfs_node *rn, struct uio *uio) 1310 { 1311 struct rumpuser_iovec iov; 1312 uint8_t *buf; 1313 size_t bufsize, n; 1314 int error = 0; 1315 1316 bufsize = uio->uio_resid; 1317 if (bufsize == 0) 1318 return 0; 1319 buf = kmem_alloc(bufsize, KM_SLEEP); 1320 1321 iov.iov_base = buf; 1322 iov.iov_len = bufsize; 1323 if ((error = rumpuser_iovread(rn->rn_readfd, &iov, 1, 1324 uio->uio_offset + rn->rn_offset, &n)) == 0) { 1325 KASSERT(n <= bufsize); 1326 error = uiomove(buf, n, uio); 1327 } 1328 1329 kmem_free(buf, bufsize); 1330 return error; 1331 } 1332 1333 static int 1334 rump_vop_read(void *v) 1335 { 1336 struct vop_read_args /* { 1337 struct vnode *a_vp; 1338 struct uio *a_uio; 1339 int ioflags a_ioflag; 1340 kauth_cred_t a_cred; 1341 }; */ *ap = v; 1342 struct vnode *vp = ap->a_vp; 1343 struct rumpfs_node *rn = vp->v_data; 1344 struct uio *uio = ap->a_uio; 1345 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1346 off_t chunk; 1347 int error = 0; 1348 1349 if (vp->v_type == VDIR) 1350 return EISDIR; 1351 1352 /* et op? */ 1353 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1354 return etread(rn, uio); 1355 1356 /* otherwise, it's off to ubc with us */ 1357 while (uio->uio_resid > 0) { 1358 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1359 if (chunk == 0) 1360 break; 1361 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1362 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1363 if (error) 1364 break; 1365 } 1366 1367 return error; 1368 } 1369 1370 static int 1371 etwrite(struct rumpfs_node *rn, struct uio *uio) 1372 { 1373 struct rumpuser_iovec iov; 1374 uint8_t *buf; 1375 size_t bufsize, n; 1376 int error = 0; 1377 1378 bufsize = uio->uio_resid; 1379 if (bufsize == 0) 1380 return 0; 1381 buf = kmem_alloc(bufsize, KM_SLEEP); 1382 error = uiomove(buf, bufsize, uio); 1383 if (error) 1384 goto out; 1385 1386 KASSERT(uio->uio_resid == 0); 1387 iov.iov_base = buf; 1388 iov.iov_len = bufsize; 1389 if ((error = rumpuser_iovwrite(rn->rn_writefd, &iov, 1, 1390 (uio->uio_offset-bufsize) + rn->rn_offset, &n)) == 0) { 1391 KASSERT(n <= bufsize); 1392 uio->uio_resid = bufsize - n; 1393 } 1394 1395 out: 1396 kmem_free(buf, bufsize); 1397 return error; 1398 } 1399 1400 static int 1401 rump_vop_write(void *v) 1402 { 1403 struct vop_write_args /* { 1404 struct vnode *a_vp; 1405 struct uio *a_uio; 1406 int ioflags a_ioflag; 1407 kauth_cred_t a_cred; 1408 }; */ *ap = v; 1409 struct vnode *vp = ap->a_vp; 1410 struct rumpfs_node *rn = vp->v_data; 1411 struct uio *uio = ap->a_uio; 1412 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1413 void *olddata; 1414 size_t oldlen, newlen; 1415 off_t chunk; 1416 int error = 0; 1417 bool allocd = false; 1418 1419 if (ap->a_ioflag & IO_APPEND) 1420 uio->uio_offset = vp->v_size; 1421 1422 /* consult et? */ 1423 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1424 return etwrite(rn, uio); 1425 1426 /* 1427 * Otherwise, it's a case of ubcmove. 1428 */ 1429 1430 /* 1431 * First, make sure we have enough storage. 1432 * 1433 * No, you don't need to tell me it's not very efficient. 1434 * No, it doesn't really support sparse files, just fakes it. 1435 */ 1436 newlen = uio->uio_offset + uio->uio_resid; 1437 oldlen = 0; /* XXXgcc */ 1438 olddata = NULL; 1439 if (rn->rn_dlen < newlen) { 1440 oldlen = rn->rn_dlen; 1441 olddata = rn->rn_data; 1442 1443 rn->rn_data = rump_hypermalloc(newlen, 0, false, "rumpfs"); 1444 if (rn->rn_data == NULL) 1445 return ENOSPC; 1446 rn->rn_dlen = newlen; 1447 memset(rn->rn_data, 0, newlen); 1448 memcpy(rn->rn_data, olddata, oldlen); 1449 allocd = true; 1450 uvm_vnp_setsize(vp, newlen); 1451 } 1452 1453 /* ok, we have enough stooorage. write */ 1454 while (uio->uio_resid > 0) { 1455 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1456 if (chunk == 0) 1457 break; 1458 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1459 UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1460 if (error) 1461 break; 1462 } 1463 1464 if (allocd) { 1465 if (error) { 1466 rump_hyperfree(rn->rn_data, newlen); 1467 rn->rn_data = olddata; 1468 rn->rn_dlen = oldlen; 1469 uvm_vnp_setsize(vp, oldlen); 1470 } else { 1471 if ((rn->rn_flags & RUMPNODE_EXTSTORAGE) == 0) { 1472 rump_hyperfree(olddata, oldlen); 1473 } else { 1474 rn->rn_flags &= ~RUMPNODE_EXTSTORAGE; 1475 } 1476 } 1477 } 1478 1479 return error; 1480 } 1481 1482 static int 1483 rump_vop_bmap(void *v) 1484 { 1485 struct vop_bmap_args /* { 1486 struct vnode *a_vp; 1487 daddr_t a_bn; 1488 struct vnode **a_vpp; 1489 daddr_t *a_bnp; 1490 int *a_runp; 1491 } */ *ap = v; 1492 1493 /* 1:1 mapping */ 1494 if (ap->a_vpp) 1495 *ap->a_vpp = ap->a_vp; 1496 if (ap->a_bnp) 1497 *ap->a_bnp = ap->a_bn; 1498 if (ap->a_runp) 1499 *ap->a_runp = 16; 1500 1501 return 0; 1502 } 1503 1504 static int 1505 rump_vop_strategy(void *v) 1506 { 1507 struct vop_strategy_args /* { 1508 struct vnode *a_vp; 1509 struct buf *a_bp; 1510 } */ *ap = v; 1511 struct vnode *vp = ap->a_vp; 1512 struct rumpfs_node *rn = vp->v_data; 1513 struct buf *bp = ap->a_bp; 1514 off_t copylen, copyoff; 1515 int error; 1516 1517 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) { 1518 error = EINVAL; 1519 goto out; 1520 } 1521 1522 copyoff = bp->b_blkno << DEV_BSHIFT; 1523 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount); 1524 if (BUF_ISWRITE(bp)) { 1525 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen); 1526 } else { 1527 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen); 1528 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen); 1529 } 1530 bp->b_resid = 0; 1531 error = 0; 1532 1533 out: 1534 bp->b_error = error; 1535 biodone(bp); 1536 return 0; 1537 } 1538 1539 static int 1540 rump_vop_pathconf(void *v) 1541 { 1542 struct vop_pathconf_args /* { 1543 struct vnode *a_vp; 1544 int a_name; 1545 register_t *a_retval; 1546 }; */ *ap = v; 1547 int name = ap->a_name; 1548 register_t *retval = ap->a_retval; 1549 1550 switch (name) { 1551 case _PC_LINK_MAX: 1552 *retval = LINK_MAX; 1553 return 0; 1554 case _PC_NAME_MAX: 1555 *retval = RUMPFS_MAXNAMLEN; 1556 return 0; 1557 case _PC_PATH_MAX: 1558 *retval = PATH_MAX; 1559 return 0; 1560 case _PC_PIPE_BUF: 1561 *retval = PIPE_BUF; 1562 return 0; 1563 case _PC_CHOWN_RESTRICTED: 1564 *retval = 1; 1565 return 0; 1566 case _PC_NO_TRUNC: 1567 *retval = 1; 1568 return 0; 1569 case _PC_SYNC_IO: 1570 *retval = 1; 1571 return 0; 1572 case _PC_FILESIZEBITS: 1573 *retval = 43; /* this one goes to 11 */ 1574 return 0; 1575 case _PC_SYMLINK_MAX: 1576 *retval = MAXPATHLEN; 1577 return 0; 1578 case _PC_2_SYMLINKS: 1579 *retval = 1; 1580 return 0; 1581 default: 1582 return EINVAL; 1583 } 1584 } 1585 1586 static int 1587 rump_vop_success(void *v) 1588 { 1589 1590 return 0; 1591 } 1592 1593 static int 1594 rump_vop_inactive(void *v) 1595 { 1596 struct vop_inactive_v2_args /* { 1597 struct vnode *a_vp; 1598 bool *a_recycle; 1599 } */ *ap = v; 1600 struct vnode *vp = ap->a_vp; 1601 struct rumpfs_node *rn = vp->v_data; 1602 1603 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) { 1604 if (rn->rn_readfd != -1) { 1605 rumpuser_close(rn->rn_readfd); 1606 rn->rn_readfd = -1; 1607 } 1608 if (rn->rn_writefd != -1) { 1609 rumpuser_close(rn->rn_writefd); 1610 rn->rn_writefd = -1; 1611 } 1612 } 1613 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false; 1614 1615 return 0; 1616 } 1617 1618 static int 1619 rump_vop_reclaim(void *v) 1620 { 1621 struct vop_reclaim_v2_args /* { 1622 struct vnode *a_vp; 1623 } */ *ap = v; 1624 struct vnode *vp = ap->a_vp; 1625 struct rumpfs_node *rn = vp->v_data; 1626 1627 VOP_UNLOCK(vp); 1628 1629 mutex_enter(&reclock); 1630 rn->rn_vp = NULL; 1631 mutex_exit(&reclock); 1632 genfs_node_destroy(vp); 1633 vp->v_data = NULL; 1634 1635 if (rn->rn_flags & RUMPNODE_CANRECLAIM) { 1636 if (vp->v_type == VREG 1637 && (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0 1638 && rn->rn_data) { 1639 if ((rn->rn_flags & RUMPNODE_EXTSTORAGE) == 0) { 1640 rump_hyperfree(rn->rn_data, rn->rn_dlen); 1641 } else { 1642 rn->rn_flags &= ~RUMPNODE_EXTSTORAGE; 1643 } 1644 rn->rn_data = NULL; 1645 } 1646 1647 if (vp->v_type == VLNK) 1648 PNBUF_PUT(rn->rn_linktarg); 1649 if (rn->rn_hostpath) 1650 free(rn->rn_hostpath, M_TEMP); 1651 freeprivate(rn); 1652 } 1653 1654 return 0; 1655 } 1656 1657 static int 1658 rump_vop_spec(void *v) 1659 { 1660 struct vop_generic_args *ap = v; 1661 int (**opvec)(void *); 1662 1663 switch (ap->a_desc->vdesc_offset) { 1664 case VOP_ACCESS_DESCOFFSET: 1665 case VOP_GETATTR_DESCOFFSET: 1666 case VOP_SETATTR_DESCOFFSET: 1667 case VOP_LOCK_DESCOFFSET: 1668 case VOP_UNLOCK_DESCOFFSET: 1669 case VOP_ISLOCKED_DESCOFFSET: 1670 case VOP_INACTIVE_DESCOFFSET: 1671 case VOP_RECLAIM_DESCOFFSET: 1672 opvec = rump_vnodeop_p; 1673 break; 1674 default: 1675 opvec = spec_vnodeop_p; 1676 break; 1677 } 1678 1679 return VOCALL(opvec, ap->a_desc->vdesc_offset, v); 1680 } 1681 1682 static int 1683 rump_vop_advlock(void *v) 1684 { 1685 struct vop_advlock_args /* { 1686 const struct vnodeop_desc *a_desc; 1687 struct vnode *a_vp; 1688 void *a_id; 1689 int a_op; 1690 struct flock *a_fl; 1691 int a_flags; 1692 } */ *ap = v; 1693 struct vnode *vp = ap->a_vp; 1694 struct rumpfs_node *rn = vp->v_data; 1695 1696 return lf_advlock(ap, &rn->rn_lockf, vp->v_size); 1697 } 1698 1699 static int 1700 rump_vop_fcntl(void *v) 1701 { 1702 struct vop_fcntl_args /* { 1703 struct vnode *a_vp; 1704 u_int a_command; 1705 void *a_data; 1706 int a_fflag; 1707 kauth_cred_t a_cred; 1708 } */ *ap = v; 1709 struct proc *p = curproc; 1710 struct vnode *vp = ap->a_vp; 1711 struct rumpfs_node *rn = vp->v_data; 1712 u_int cmd = ap->a_command; 1713 int fflag = ap->a_fflag; 1714 struct rumpfs_extstorage *rfse = ap->a_data; 1715 int error = 0; 1716 1717 /* none of the current rumpfs fcntlops are defined for remotes */ 1718 if (!RUMP_LOCALPROC_P(p)) 1719 return EINVAL; 1720 1721 switch (cmd) { 1722 case RUMPFS_FCNTL_EXTSTORAGE_ADD: 1723 break; 1724 default: 1725 return EINVAL; 1726 } 1727 1728 if ((fflag & FWRITE) == 0) 1729 return EBADF; 1730 1731 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)) 1732 return EINVAL; 1733 1734 if (rfse->rfse_flags != 0) 1735 return EINVAL; 1736 1737 /* 1738 * Ok, we are good to go. Process. 1739 */ 1740 1741 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1742 1743 KASSERT(cmd == RUMPFS_FCNTL_EXTSTORAGE_ADD); 1744 if (rn->rn_data && (rn->rn_flags & RUMPNODE_EXTSTORAGE) == 0) { 1745 rump_hyperfree(rn->rn_data, rn->rn_dlen); 1746 } 1747 1748 rn->rn_data = rfse->rfse_data; 1749 rn->rn_dlen = rfse->rfse_dlen; 1750 uvm_vnp_setsize(vp, rn->rn_dlen); 1751 rn->rn_flags |= RUMPNODE_EXTSTORAGE; 1752 1753 VOP_UNLOCK(vp); 1754 1755 return error; 1756 } 1757 1758 /* 1759 * Begin vfs-level stuff 1760 */ 1761 1762 VFS_PROTOS(rumpfs); 1763 struct vfsops rumpfs_vfsops = { 1764 .vfs_name = MOUNT_RUMPFS, 1765 .vfs_min_mount_data = 0, 1766 .vfs_mount = rumpfs_mount, 1767 .vfs_start = (void *)nullop, 1768 .vfs_unmount = rumpfs_unmount, 1769 .vfs_root = rumpfs_root, 1770 .vfs_quotactl = (void *)eopnotsupp, 1771 .vfs_statvfs = genfs_statvfs, 1772 .vfs_sync = (void *)nullop, 1773 .vfs_vget = rumpfs_vget, 1774 .vfs_loadvnode = rumpfs_loadvnode, 1775 .vfs_fhtovp = (void *)eopnotsupp, 1776 .vfs_vptofh = (void *)eopnotsupp, 1777 .vfs_init = rumpfs_init, 1778 .vfs_reinit = NULL, 1779 .vfs_done = rumpfs_done, 1780 .vfs_mountroot = rumpfs_mountroot, 1781 .vfs_snapshot = (void *)eopnotsupp, 1782 .vfs_extattrctl = (void *)eopnotsupp, 1783 .vfs_suspendctl = genfs_suspendctl, 1784 .vfs_renamelock_enter = genfs_renamelock_enter, 1785 .vfs_renamelock_exit = genfs_renamelock_exit, 1786 .vfs_opv_descs = rump_opv_descs, 1787 /* vfs_refcount */ 1788 /* vfs_list */ 1789 }; 1790 1791 static int 1792 rumpfs_mountfs(struct mount *mp) 1793 { 1794 struct rumpfs_mount *rfsmp; 1795 struct rumpfs_node *rn; 1796 int error; 1797 1798 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP); 1799 1800 rn = makeprivate(VDIR, RUMPFS_DEFAULTMODE, NODEV, DEV_BSIZE, false); 1801 rn->rn_parent = rn; 1802 if ((error = vcache_get(mp, &rn, sizeof(rn), &rfsmp->rfsmp_rvp)) 1803 != 0) { 1804 freeprivate(rn); 1805 kmem_free(rfsmp, sizeof(*rfsmp)); 1806 return error; 1807 } 1808 1809 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT; 1810 1811 mp->mnt_data = rfsmp; 1812 mp->mnt_stat.f_namemax = RUMPFS_MAXNAMLEN; 1813 mp->mnt_stat.f_iosize = 512; 1814 mp->mnt_flag |= MNT_LOCAL; 1815 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO; 1816 mp->mnt_fs_bshift = DEV_BSHIFT; 1817 vfs_getnewfsid(mp); 1818 1819 return 0; 1820 } 1821 1822 int 1823 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen) 1824 { 1825 int error, flags; 1826 1827 if (mp->mnt_flag & MNT_GETARGS) { 1828 return 0; 1829 } 1830 if (mp->mnt_flag & MNT_UPDATE) { 1831 if ((mp->mnt_iflag & IMNT_WANTRDONLY)) { 1832 /* Changing from read/write to read-only. */ 1833 flags = WRITECLOSE; 1834 if ((mp->mnt_flag & MNT_FORCE)) 1835 flags |= FORCECLOSE; 1836 error = vflush(mp, NULL, flags); 1837 if (error) 1838 return error; 1839 } 1840 return 0; 1841 } 1842 1843 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE, 1844 mp->mnt_op->vfs_name, mp, curlwp); 1845 if (error) 1846 return error; 1847 1848 return rumpfs_mountfs(mp); 1849 } 1850 1851 int 1852 rumpfs_unmount(struct mount *mp, int mntflags) 1853 { 1854 struct rumpfs_mount *rfsmp = mp->mnt_data; 1855 int flags = 0, error; 1856 1857 if (panicstr || mntflags & MNT_FORCE) 1858 flags |= FORCECLOSE; 1859 1860 if (rfsmp->rfsmp_rvp->v_usecount > 1 && (flags & FORCECLOSE) == 0) 1861 return EBUSY; 1862 1863 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0) 1864 return error; 1865 vgone(rfsmp->rfsmp_rvp); 1866 1867 kmem_free(rfsmp, sizeof(*rfsmp)); 1868 1869 return 0; 1870 } 1871 1872 int 1873 rumpfs_root(struct mount *mp, struct vnode **vpp) 1874 { 1875 struct rumpfs_mount *rfsmp = mp->mnt_data; 1876 1877 vref(rfsmp->rfsmp_rvp); 1878 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY); 1879 *vpp = rfsmp->rfsmp_rvp; 1880 return 0; 1881 } 1882 1883 int 1884 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1885 { 1886 1887 return EOPNOTSUPP; 1888 } 1889 1890 int 1891 rumpfs_loadvnode(struct mount *mp, struct vnode *vp, 1892 const void *key, size_t key_len, const void **new_key) 1893 { 1894 struct rumpfs_node *rn; 1895 struct vattr *va; 1896 1897 KASSERT(!mutex_owned(&reclock)); 1898 1899 KASSERT(key_len == sizeof(rn)); 1900 memcpy(&rn, key, key_len); 1901 1902 va = &rn->rn_va; 1903 1904 vp->v_tag = VT_RUMP; 1905 vp->v_type = va->va_type; 1906 switch (vp->v_type) { 1907 case VCHR: 1908 case VBLK: 1909 vp->v_op = rump_specop_p; 1910 spec_node_init(vp, va->va_rdev); 1911 break; 1912 default: 1913 vp->v_op = rump_vnodeop_p; 1914 break; 1915 } 1916 vp->v_size = vp->v_writesize = va->va_size; 1917 vp->v_data = rn; 1918 1919 genfs_node_init(vp, &rumpfs_genfsops); 1920 mutex_enter(&reclock); 1921 rn->rn_vp = vp; 1922 mutex_exit(&reclock); 1923 1924 *new_key = &vp->v_data; 1925 1926 return 0; 1927 } 1928 1929 void 1930 rumpfs_init() 1931 { 1932 extern rump_etfs_register_withsize_fn rump__etfs_register; 1933 extern rump_etfs_remove_fn rump__etfs_remove; 1934 extern struct rump_boot_etfs *ebstart; 1935 struct rump_boot_etfs *eb; 1936 1937 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET); 1938 1939 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE); 1940 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE); 1941 1942 rump__etfs_register = etfsregister; 1943 rump__etfs_remove = etfsremove; 1944 1945 for (eb = ebstart; eb; eb = eb->_eb_next) { 1946 eb->eb_status = etfsregister(eb->eb_key, eb->eb_hostpath, 1947 eb->eb_type, eb->eb_begin, eb->eb_size); 1948 } 1949 } 1950 1951 void 1952 rumpfs_done() 1953 { 1954 1955 mutex_destroy(&reclock); 1956 mutex_destroy(&etfs_lock); 1957 } 1958 1959 int 1960 rumpfs_mountroot() 1961 { 1962 struct mount *mp; 1963 int error; 1964 1965 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) { 1966 vrele(rootvp); 1967 return error; 1968 } 1969 1970 if ((error = rumpfs_mountfs(mp)) != 0) 1971 panic("mounting rootfs failed: %d", error); 1972 1973 mountlist_append(mp); 1974 1975 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE, 1976 mp->mnt_op->vfs_name, mp, curlwp); 1977 if (error) 1978 panic("set_statvfs_info failed for rootfs: %d", error); 1979 1980 mp->mnt_flag &= ~MNT_RDONLY; 1981 vfs_unbusy(mp); 1982 1983 return 0; 1984 } 1985
Indexes created Wed Oct 01 15:09:59 GMT 2025