rumpfs.c revision 1.106
1 /* $NetBSD: rumpfs.c,v 1.106 2012/01/31 19:00:03 njoly 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 1.106 2012/01/31 19:00:03 njoly 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 { NULL, NULL } 92 }; 93 const struct vnodeopv_desc fifo_vnodeop_opv_desc = 94 { &fifo_vnodeop_p, fifo_vnodeop_entries }; 95 96 int (**rump_vnodeop_p)(void *); 97 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = { 98 { &vop_default_desc, vn_default_error }, 99 { &vop_lookup_desc, rump_vop_lookup }, 100 { &vop_getattr_desc, rump_vop_getattr }, 101 { &vop_setattr_desc, rump_vop_setattr }, 102 { &vop_mkdir_desc, rump_vop_mkdir }, 103 { &vop_rmdir_desc, rump_vop_rmdir }, 104 { &vop_remove_desc, rump_vop_remove }, 105 { &vop_mknod_desc, rump_vop_mknod }, 106 { &vop_create_desc, rump_vop_create }, 107 { &vop_symlink_desc, rump_vop_symlink }, 108 { &vop_readlink_desc, rump_vop_readlink }, 109 { &vop_access_desc, rump_vop_access }, 110 { &vop_readdir_desc, rump_vop_readdir }, 111 { &vop_read_desc, rump_vop_read }, 112 { &vop_write_desc, rump_vop_write }, 113 { &vop_open_desc, rump_vop_open }, 114 { &vop_close_desc, genfs_nullop }, 115 { &vop_seek_desc, genfs_seek }, 116 { &vop_getpages_desc, genfs_getpages }, 117 { &vop_putpages_desc, genfs_putpages }, 118 { &vop_whiteout_desc, rump_vop_whiteout }, 119 { &vop_fsync_desc, rump_vop_success }, 120 { &vop_lock_desc, genfs_lock }, 121 { &vop_unlock_desc, genfs_unlock }, 122 { &vop_islocked_desc, genfs_islocked }, 123 { &vop_inactive_desc, rump_vop_inactive }, 124 { &vop_reclaim_desc, rump_vop_reclaim }, 125 { &vop_link_desc, genfs_eopnotsupp }, 126 { &vop_pathconf_desc, rump_vop_pathconf }, 127 { &vop_bmap_desc, rump_vop_bmap }, 128 { &vop_strategy_desc, rump_vop_strategy }, 129 { &vop_advlock_desc, rump_vop_advlock }, 130 { NULL, NULL } 131 }; 132 const struct vnodeopv_desc rump_vnodeop_opv_desc = 133 { &rump_vnodeop_p, rump_vnodeop_entries }; 134 135 int (**rump_specop_p)(void *); 136 const struct vnodeopv_entry_desc rump_specop_entries[] = { 137 { &vop_default_desc, rump_vop_spec }, 138 { NULL, NULL } 139 }; 140 const struct vnodeopv_desc rump_specop_opv_desc = 141 { &rump_specop_p, rump_specop_entries }; 142 143 const struct vnodeopv_desc * const rump_opv_descs[] = { 144 &rump_vnodeop_opv_desc, 145 &rump_specop_opv_desc, 146 NULL 147 }; 148 149 #define RUMPFS_WHITEOUT ((void *)-1) 150 #define RDENT_ISWHITEOUT(rdp) (rdp->rd_node == RUMPFS_WHITEOUT) 151 struct rumpfs_dent { 152 char *rd_name; 153 int rd_namelen; 154 struct rumpfs_node *rd_node; 155 156 LIST_ENTRY(rumpfs_dent) rd_entries; 157 }; 158 159 struct genfs_ops rumpfs_genfsops = { 160 .gop_size = genfs_size, 161 .gop_write = genfs_gop_write, 162 163 /* optional */ 164 .gop_alloc = NULL, 165 .gop_markupdate = NULL, 166 }; 167 168 struct rumpfs_node { 169 struct genfs_node rn_gn; 170 struct vattr rn_va; 171 struct vnode *rn_vp; 172 char *rn_hostpath; 173 int rn_flags; 174 struct lockf *rn_lockf; 175 176 union { 177 struct { /* VREG */ 178 int readfd; 179 int writefd; 180 uint64_t offset; 181 } reg; 182 struct { 183 void *data; 184 size_t dlen; 185 } reg_noet; 186 struct { /* VDIR */ 187 LIST_HEAD(, rumpfs_dent) dents; 188 struct rumpfs_node *parent; 189 int flags; 190 } dir; 191 struct { 192 char *target; 193 size_t len; 194 } link; 195 } rn_u; 196 }; 197 #define rn_readfd rn_u.reg.readfd 198 #define rn_writefd rn_u.reg.writefd 199 #define rn_offset rn_u.reg.offset 200 #define rn_data rn_u.reg_noet.data 201 #define rn_dlen rn_u.reg_noet.dlen 202 #define rn_dir rn_u.dir.dents 203 #define rn_parent rn_u.dir.parent 204 #define rn_linktarg rn_u.link.target 205 #define rn_linklen rn_u.link.len 206 207 #define RUMPNODE_CANRECLAIM 0x01 208 #define RUMPNODE_DIR_ET 0x02 209 #define RUMPNODE_DIR_ETSUBS 0x04 210 #define RUMPNODE_ET_PHONE_HOST 0x10 211 212 struct rumpfs_mount { 213 struct vnode *rfsmp_rvp; 214 }; 215 216 #define INO_WHITEOUT 1 217 static int lastino = 2; 218 static kmutex_t reclock; 219 220 static void freedir(struct rumpfs_node *, struct componentname *); 221 static struct rumpfs_node *makeprivate(enum vtype, dev_t, off_t, bool); 222 223 /* 224 * Extra Terrestrial stuff. We map a given key (pathname) to a file on 225 * the host FS. ET phones home only from the root node of rumpfs. 226 * 227 * When an etfs node is removed, a vnode potentially behind it is not 228 * immediately recycled. 229 */ 230 231 struct etfs { 232 char et_key[MAXPATHLEN]; 233 size_t et_keylen; 234 bool et_prefixkey; 235 bool et_removing; 236 devminor_t et_blkmin; 237 238 LIST_ENTRY(etfs) et_entries; 239 240 struct rumpfs_node *et_rn; 241 }; 242 static kmutex_t etfs_lock; 243 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list); 244 245 static enum vtype 246 ettype_to_vtype(enum rump_etfs_type et) 247 { 248 enum vtype vt; 249 250 switch (et) { 251 case RUMP_ETFS_REG: 252 vt = VREG; 253 break; 254 case RUMP_ETFS_BLK: 255 vt = VBLK; 256 break; 257 case RUMP_ETFS_CHR: 258 vt = VCHR; 259 break; 260 case RUMP_ETFS_DIR: 261 vt = VDIR; 262 break; 263 case RUMP_ETFS_DIR_SUBDIRS: 264 vt = VDIR; 265 break; 266 default: 267 panic("invalid et type: %d", et); 268 } 269 270 return vt; 271 } 272 273 static enum vtype 274 hft_to_vtype(int hft) 275 { 276 enum vtype vt; 277 278 switch (hft) { 279 case RUMPUSER_FT_OTHER: 280 vt = VNON; 281 break; 282 case RUMPUSER_FT_DIR: 283 vt = VDIR; 284 break; 285 case RUMPUSER_FT_REG: 286 vt = VREG; 287 break; 288 case RUMPUSER_FT_BLK: 289 vt = VBLK; 290 break; 291 case RUMPUSER_FT_CHR: 292 vt = VCHR; 293 break; 294 default: 295 vt = VNON; 296 break; 297 } 298 299 return vt; 300 } 301 302 static bool 303 etfs_find(const char *key, struct etfs **etp, bool forceprefix) 304 { 305 struct etfs *et; 306 size_t keylen = strlen(key); 307 308 KASSERT(mutex_owned(&etfs_lock)); 309 310 LIST_FOREACH(et, &etfs_list, et_entries) { 311 if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix) 312 && strncmp(key, et->et_key, et->et_keylen) == 0) { 313 if (etp) 314 *etp = et; 315 return true; 316 } 317 } 318 319 return false; 320 } 321 322 #define REGDIR(ftype) \ 323 ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS) 324 static int 325 doregister(const char *key, const char *hostpath, 326 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 327 { 328 char buf[9]; 329 struct etfs *et; 330 struct rumpfs_node *rn; 331 uint64_t fsize; 332 dev_t rdev = NODEV; 333 devminor_t dmin = -1; 334 int hft, error; 335 336 if (key[0] != '/') { 337 return EINVAL; 338 } 339 while (key[0] == '/') { 340 key++; 341 } 342 343 if (rumpuser_getfileinfo(hostpath, &fsize, &hft, &error)) 344 return error; 345 346 /* etfs directory requires a directory on the host */ 347 if (REGDIR(ftype)) { 348 if (hft != RUMPUSER_FT_DIR) 349 return ENOTDIR; 350 if (begin != 0) 351 return EISDIR; 352 if (size != RUMP_ETFS_SIZE_ENDOFF) 353 return EISDIR; 354 size = fsize; 355 } else { 356 if (begin > fsize) 357 return EINVAL; 358 if (size == RUMP_ETFS_SIZE_ENDOFF) 359 size = fsize - begin; 360 if (begin + size > fsize) 361 return EINVAL; 362 } 363 364 if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) { 365 error = rumpblk_register(hostpath, &dmin, begin, size); 366 if (error != 0) { 367 return error; 368 } 369 rdev = makedev(RUMPBLK_DEVMAJOR, dmin); 370 } 371 372 et = kmem_alloc(sizeof(*et), KM_SLEEP); 373 strcpy(et->et_key, key); 374 et->et_keylen = strlen(et->et_key); 375 et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), rdev, size, true); 376 et->et_removing = false; 377 et->et_blkmin = dmin; 378 379 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 380 381 if (ftype == RUMP_ETFS_REG || REGDIR(ftype) || et->et_blkmin != -1) { 382 size_t len = strlen(hostpath)+1; 383 384 rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 385 memcpy(rn->rn_hostpath, hostpath, len); 386 rn->rn_offset = begin; 387 } 388 389 if (REGDIR(ftype)) { 390 rn->rn_flags |= RUMPNODE_DIR_ET; 391 et->et_prefixkey = true; 392 } else { 393 et->et_prefixkey = false; 394 } 395 396 if (ftype == RUMP_ETFS_DIR_SUBDIRS) 397 rn->rn_flags |= RUMPNODE_DIR_ETSUBS; 398 399 mutex_enter(&etfs_lock); 400 if (etfs_find(key, NULL, REGDIR(ftype))) { 401 mutex_exit(&etfs_lock); 402 if (et->et_blkmin != -1) 403 rumpblk_deregister(hostpath); 404 if (et->et_rn->rn_hostpath != NULL) 405 free(et->et_rn->rn_hostpath, M_TEMP); 406 kmem_free(et->et_rn, sizeof(*et->et_rn)); 407 kmem_free(et, sizeof(*et)); 408 return EEXIST; 409 } 410 LIST_INSERT_HEAD(&etfs_list, et, et_entries); 411 mutex_exit(&etfs_lock); 412 413 if (ftype == RUMP_ETFS_BLK) { 414 format_bytes(buf, sizeof(buf), size); 415 aprint_verbose("/%s: hostpath %s (%s)\n", key, hostpath, buf); 416 } 417 418 return 0; 419 } 420 #undef REGDIR 421 422 int 423 rump_etfs_register(const char *key, const char *hostpath, 424 enum rump_etfs_type ftype) 425 { 426 427 return doregister(key, hostpath, ftype, 0, RUMP_ETFS_SIZE_ENDOFF); 428 } 429 430 int 431 rump_etfs_register_withsize(const char *key, const char *hostpath, 432 enum rump_etfs_type ftype, uint64_t begin, uint64_t size) 433 { 434 435 return doregister(key, hostpath, ftype, begin, size); 436 } 437 438 /* remove etfs mapping. caller's responsibility to make sure it's not in use */ 439 int 440 rump_etfs_remove(const char *key) 441 { 442 struct etfs *et; 443 size_t keylen; 444 int rv; 445 446 if (key[0] != '/') { 447 return EINVAL; 448 } 449 while (key[0] == '/') { 450 key++; 451 } 452 453 keylen = strlen(key); 454 455 mutex_enter(&etfs_lock); 456 LIST_FOREACH(et, &etfs_list, et_entries) { 457 if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) { 458 if (et->et_removing) 459 et = NULL; 460 else 461 et->et_removing = true; 462 break; 463 } 464 } 465 mutex_exit(&etfs_lock); 466 if (!et) 467 return ENOENT; 468 469 /* 470 * ok, we know what we want to remove and have signalled there 471 * actually are men at work. first, unregister from rumpblk 472 */ 473 if (et->et_blkmin != -1) { 474 rv = rumpblk_deregister(et->et_rn->rn_hostpath); 475 } else { 476 rv = 0; 477 } 478 KASSERT(rv == 0); 479 480 /* then do the actual removal */ 481 mutex_enter(&etfs_lock); 482 LIST_REMOVE(et, et_entries); 483 mutex_exit(&etfs_lock); 484 485 /* node is unreachable, safe to nuke all device copies */ 486 if (et->et_blkmin != -1) { 487 vdevgone(RUMPBLK_DEVMAJOR, et->et_blkmin, et->et_blkmin, VBLK); 488 } else { 489 struct vnode *vp; 490 491 mutex_enter(&reclock); 492 if ((vp = et->et_rn->rn_vp) != NULL) 493 mutex_enter(vp->v_interlock); 494 mutex_exit(&reclock); 495 if (vp && vget(vp, 0) == 0) 496 vgone(vp); 497 } 498 499 if (et->et_rn->rn_hostpath != NULL) 500 free(et->et_rn->rn_hostpath, M_TEMP); 501 kmem_free(et->et_rn, sizeof(*et->et_rn)); 502 kmem_free(et, sizeof(*et)); 503 504 return 0; 505 } 506 507 /* 508 * rumpfs 509 */ 510 511 static struct rumpfs_node * 512 makeprivate(enum vtype vt, dev_t rdev, off_t size, bool et) 513 { 514 struct rumpfs_node *rn; 515 struct vattr *va; 516 struct timespec ts; 517 518 rn = kmem_zalloc(sizeof(*rn), KM_SLEEP); 519 520 switch (vt) { 521 case VDIR: 522 LIST_INIT(&rn->rn_dir); 523 break; 524 case VREG: 525 if (et) { 526 rn->rn_readfd = -1; 527 rn->rn_writefd = -1; 528 } 529 break; 530 default: 531 break; 532 } 533 534 nanotime(&ts); 535 536 va = &rn->rn_va; 537 va->va_type = vt; 538 va->va_mode = 0755; 539 if (vt == VDIR) 540 va->va_nlink = 2; 541 else 542 va->va_nlink = 1; 543 va->va_uid = 0; 544 va->va_gid = 0; 545 va->va_fsid = 546 va->va_fileid = atomic_inc_uint_nv(&lastino); 547 va->va_size = size; 548 va->va_blocksize = 512; 549 va->va_atime = ts; 550 va->va_mtime = ts; 551 va->va_ctime = ts; 552 va->va_birthtime = ts; 553 va->va_gen = 0; 554 va->va_flags = 0; 555 va->va_rdev = rdev; 556 va->va_bytes = 512; 557 va->va_filerev = 0; 558 va->va_vaflags = 0; 559 560 return rn; 561 } 562 563 static int 564 makevnode(struct mount *mp, struct rumpfs_node *rn, struct vnode **vpp) 565 { 566 struct vnode *vp; 567 int (**vpops)(void *); 568 struct vattr *va = &rn->rn_va; 569 int rv; 570 571 KASSERT(!mutex_owned(&reclock)); 572 573 if (va->va_type == VCHR || va->va_type == VBLK) { 574 vpops = rump_specop_p; 575 } else { 576 vpops = rump_vnodeop_p; 577 } 578 579 rv = getnewvnode(VT_RUMP, mp, vpops, NULL, &vp); 580 if (rv) 581 return rv; 582 583 vp->v_size = vp->v_writesize = va->va_size; 584 vp->v_type = va->va_type; 585 586 if (vpops == rump_specop_p) { 587 spec_node_init(vp, va->va_rdev); 588 } 589 vp->v_data = rn; 590 591 genfs_node_init(vp, &rumpfs_genfsops); 592 vn_lock(vp, LK_RETRY | LK_EXCLUSIVE); 593 mutex_enter(&reclock); 594 rn->rn_vp = vp; 595 mutex_exit(&reclock); 596 597 *vpp = vp; 598 599 return 0; 600 } 601 602 603 static void 604 makedir(struct rumpfs_node *rnd, 605 struct componentname *cnp, struct rumpfs_node *rn) 606 { 607 struct rumpfs_dent *rdent; 608 609 rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP); 610 rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP); 611 rdent->rd_node = rn; 612 strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1); 613 rdent->rd_namelen = strlen(rdent->rd_name); 614 615 if ((cnp->cn_flags & ISWHITEOUT) != 0) { 616 KASSERT((cnp->cn_flags & DOWHITEOUT) == 0); 617 freedir(rnd, cnp); 618 } 619 LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries); 620 } 621 622 static void 623 freedir(struct rumpfs_node *rnd, struct componentname *cnp) 624 { 625 struct rumpfs_dent *rd = NULL; 626 627 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 628 if (rd->rd_namelen == cnp->cn_namelen && 629 strncmp(rd->rd_name, cnp->cn_nameptr, 630 cnp->cn_namelen) == 0) 631 break; 632 } 633 if (rd == NULL) 634 panic("could not find directory entry: %s", cnp->cn_nameptr); 635 636 if (cnp->cn_flags & DOWHITEOUT) { 637 rd->rd_node = RUMPFS_WHITEOUT; 638 } else { 639 LIST_REMOVE(rd, rd_entries); 640 kmem_free(rd->rd_name, rd->rd_namelen+1); 641 kmem_free(rd, sizeof(*rd)); 642 } 643 } 644 645 /* 646 * Simple lookup for rump file systems. 647 * 648 * uhm, this is twisted. C F C C, hope of C C F C looming 649 */ 650 static int 651 rump_vop_lookup(void *v) 652 { 653 struct vop_lookup_args /* { 654 struct vnode *a_dvp; 655 struct vnode **a_vpp; 656 struct componentname *a_cnp; 657 }; */ *ap = v; 658 struct componentname *cnp = ap->a_cnp; 659 struct vnode *dvp = ap->a_dvp; 660 struct vnode **vpp = ap->a_vpp; 661 struct vnode *vp; 662 struct rumpfs_node *rnd = dvp->v_data, *rn; 663 struct rumpfs_dent *rd = NULL; 664 struct etfs *et; 665 bool dotdot = (cnp->cn_flags & ISDOTDOT) != 0; 666 int rv = 0; 667 const char *cp; 668 669 *vpp = NULL; 670 671 rv = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred); 672 if (rv) 673 return rv; 674 675 if ((cnp->cn_flags & ISLASTCN) 676 && (dvp->v_mount->mnt_flag & MNT_RDONLY) 677 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 678 return EROFS; 679 680 /* check for dot, return directly if the case */ 681 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 682 vref(dvp); 683 *vpp = dvp; 684 return 0; 685 } 686 687 /* we don't do rename */ 688 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME))) 689 return EOPNOTSUPP; 690 691 /* check for etfs */ 692 if (dvp == rootvnode && 693 (cnp->cn_nameiop == LOOKUP || cnp->cn_nameiop == CREATE)) { 694 bool found; 695 mutex_enter(&etfs_lock); 696 found = etfs_find(cnp->cn_nameptr, &et, false); 697 mutex_exit(&etfs_lock); 698 699 if (found) { 700 rn = et->et_rn; 701 cnp->cn_consume += et->et_keylen - cnp->cn_namelen; 702 /* 703 * consume trailing slashes if any and clear 704 * REQUIREDIR if we consumed the full path. 705 */ 706 cp = &cnp->cn_nameptr[cnp->cn_namelen]; 707 cp += cnp->cn_consume; 708 KASSERT(*cp == '\0' || *cp == '/'); 709 if (*cp == '\0' && rn->rn_va.va_type != VDIR) 710 cnp->cn_flags &= ~REQUIREDIR; 711 while (*cp++ == '/') 712 cnp->cn_consume++; 713 goto getvnode; 714 } 715 } 716 717 if (rnd->rn_flags & RUMPNODE_DIR_ET) { 718 uint64_t fsize; 719 char *newpath; 720 size_t newpathlen; 721 int hft, error; 722 723 if (dotdot) 724 return EOPNOTSUPP; 725 726 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1; 727 newpath = malloc(newpathlen, M_TEMP, M_WAITOK); 728 729 strlcpy(newpath, rnd->rn_hostpath, newpathlen); 730 strlcat(newpath, "/", newpathlen); 731 strlcat(newpath, cnp->cn_nameptr, newpathlen); 732 733 if (rumpuser_getfileinfo(newpath, &fsize, &hft, &error)) { 734 free(newpath, M_TEMP); 735 return error; 736 } 737 738 /* allow only dirs and regular files */ 739 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) { 740 free(newpath, M_TEMP); 741 return ENOENT; 742 } 743 744 rn = makeprivate(hft_to_vtype(hft), NODEV, fsize, true); 745 rn->rn_flags |= RUMPNODE_CANRECLAIM; 746 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) { 747 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS; 748 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 749 } 750 rn->rn_hostpath = newpath; 751 752 goto getvnode; 753 } else { 754 if (dotdot) { 755 if ((rn = rnd->rn_parent) != NULL) 756 goto getvnode; 757 } else { 758 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 759 if (rd->rd_namelen == cnp->cn_namelen && 760 strncmp(rd->rd_name, cnp->cn_nameptr, 761 cnp->cn_namelen) == 0) 762 break; 763 } 764 } 765 } 766 767 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE)) 768 return ENOENT; 769 770 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 771 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 772 return EROFS; 773 rv = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred); 774 if (rv) 775 return rv; 776 return EJUSTRETURN; 777 } 778 779 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == DELETE) { 780 rv = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred); 781 if (rv) 782 return rv; 783 } 784 785 if (RDENT_ISWHITEOUT(rd)) { 786 cnp->cn_flags |= ISWHITEOUT; 787 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) 788 return EJUSTRETURN; 789 return ENOENT; 790 } 791 792 rn = rd->rd_node; 793 794 getvnode: 795 KASSERT(rn); 796 if (dotdot) 797 VOP_UNLOCK(dvp); 798 mutex_enter(&reclock); 799 if ((vp = rn->rn_vp)) { 800 mutex_enter(vp->v_interlock); 801 mutex_exit(&reclock); 802 if (vget(vp, LK_EXCLUSIVE)) { 803 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 804 goto getvnode; 805 } 806 *vpp = vp; 807 } else { 808 mutex_exit(&reclock); 809 rv = makevnode(dvp->v_mount, rn, vpp); 810 } 811 if (dotdot) 812 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 813 814 return rv; 815 } 816 817 static int 818 rump_check_possible(struct vnode *vp, struct rumpfs_node *rnode, 819 mode_t mode) 820 { 821 822 if ((mode & VWRITE) == 0) 823 return 0; 824 825 switch (vp->v_type) { 826 case VDIR: 827 case VLNK: 828 case VREG: 829 break; 830 default: 831 /* special file is always writable. */ 832 return 0; 833 } 834 835 return vp->v_mount->mnt_flag & MNT_RDONLY ? EROFS : 0; 836 } 837 838 static int 839 rump_check_permitted(struct vnode *vp, struct rumpfs_node *rnode, 840 mode_t mode, kauth_cred_t cred) 841 { 842 struct vattr *attr = &rnode->rn_va; 843 844 return genfs_can_access(vp->v_type, attr->va_mode, attr->va_uid, 845 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_getattr_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 SETIFVAL(a,t) if (vap->a != (t)VNOVAL) rn->rn_va.a = vap->a 903 SETIFVAL(va_atime.tv_sec, time_t); 904 SETIFVAL(va_ctime.tv_sec, time_t); 905 SETIFVAL(va_mtime.tv_sec, time_t); 906 SETIFVAL(va_birthtime.tv_sec, time_t); 907 SETIFVAL(va_atime.tv_nsec, long); 908 SETIFVAL(va_ctime.tv_nsec, long); 909 SETIFVAL(va_mtime.tv_nsec, long); 910 SETIFVAL(va_birthtime.tv_nsec, long); 911 SETIFVAL(va_flags, u_long); 912 #undef SETIFVAL 913 914 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (uid_t)VNOVAL) { 915 uid_t uid = 916 (vap->va_uid != (uid_t)VNOVAL) ? vap->va_uid : attr->va_uid; 917 gid_t gid = 918 (vap->va_gid != (gid_t)VNOVAL) ? vap->va_gid : attr->va_gid; 919 error = kauth_authorize_vnode(cred, 920 KAUTH_VNODE_CHANGE_OWNERSHIP, vp, NULL, 921 genfs_can_chown(vp, cred, attr->va_uid, attr->va_gid, uid, 922 gid)); 923 if (error) 924 return error; 925 attr->va_uid = uid; 926 attr->va_gid = gid; 927 } 928 929 if (vap->va_mode != (mode_t)VNOVAL) { 930 mode_t mode = vap->va_mode; 931 error = kauth_authorize_vnode(cred, KAUTH_VNODE_WRITE_SECURITY, 932 vp, NULL, genfs_can_chmod(vp, cred, attr->va_uid, 933 attr->va_gid, mode)); 934 if (error) 935 return error; 936 attr->va_mode = mode; 937 } 938 939 if (vp->v_type == VREG && 940 vap->va_size != VSIZENOTSET && 941 vap->va_size != rn->rn_dlen) { 942 void *newdata; 943 size_t copylen, newlen; 944 945 newlen = vap->va_size; 946 newdata = rump_hypermalloc(newlen, 0, true, "rumpfs"); 947 948 copylen = MIN(rn->rn_dlen, newlen); 949 memset(newdata, 0, newlen); 950 memcpy(newdata, rn->rn_data, copylen); 951 rump_hyperfree(rn->rn_data, rn->rn_dlen); 952 953 rn->rn_data = newdata; 954 rn->rn_dlen = newlen; 955 uvm_vnp_setsize(vp, newlen); 956 } 957 return 0; 958 } 959 960 static int 961 rump_vop_mkdir(void *v) 962 { 963 struct vop_mkdir_args /* { 964 struct vnode *a_dvp; 965 struct vnode **a_vpp; 966 struct componentname *a_cnp; 967 struct vattr *a_vap; 968 }; */ *ap = v; 969 struct vnode *dvp = ap->a_dvp; 970 struct vnode **vpp = ap->a_vpp; 971 struct componentname *cnp = ap->a_cnp; 972 struct rumpfs_node *rnd = dvp->v_data, *rn; 973 int rv = 0; 974 975 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false); 976 if ((cnp->cn_flags & ISWHITEOUT) != 0) 977 rn->rn_va.va_flags |= UF_OPAQUE; 978 rn->rn_parent = rnd; 979 rv = makevnode(dvp->v_mount, rn, vpp); 980 if (rv) 981 goto out; 982 983 makedir(rnd, cnp, rn); 984 985 out: 986 vput(dvp); 987 return rv; 988 } 989 990 static int 991 rump_vop_rmdir(void *v) 992 { 993 struct vop_rmdir_args /* { 994 struct vnode *a_dvp; 995 struct vnode *a_vp; 996 struct componentname *a_cnp; 997 }; */ *ap = v; 998 struct vnode *dvp = ap->a_dvp; 999 struct vnode *vp = ap->a_vp; 1000 struct componentname *cnp = ap->a_cnp; 1001 struct rumpfs_node *rnd = dvp->v_data; 1002 struct rumpfs_node *rn = vp->v_data; 1003 struct rumpfs_dent *rd; 1004 int rv = 0; 1005 1006 LIST_FOREACH(rd, &rn->rn_dir, rd_entries) { 1007 if (rd->rd_node != RUMPFS_WHITEOUT) { 1008 rv = ENOTEMPTY; 1009 goto out; 1010 } 1011 } 1012 while ((rd = LIST_FIRST(&rn->rn_dir)) != NULL) { 1013 KASSERT(rd->rd_node == RUMPFS_WHITEOUT); 1014 LIST_REMOVE(rd, rd_entries); 1015 kmem_free(rd->rd_name, rd->rd_namelen+1); 1016 kmem_free(rd, sizeof(*rd)); 1017 } 1018 1019 freedir(rnd, cnp); 1020 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1021 rn->rn_parent = NULL; 1022 1023 out: 1024 vput(dvp); 1025 vput(vp); 1026 1027 return rv; 1028 } 1029 1030 static int 1031 rump_vop_remove(void *v) 1032 { 1033 struct vop_rmdir_args /* { 1034 struct vnode *a_dvp; 1035 struct vnode *a_vp; 1036 struct componentname *a_cnp; 1037 }; */ *ap = v; 1038 struct vnode *dvp = ap->a_dvp; 1039 struct vnode *vp = ap->a_vp; 1040 struct componentname *cnp = ap->a_cnp; 1041 struct rumpfs_node *rnd = dvp->v_data; 1042 struct rumpfs_node *rn = vp->v_data; 1043 int rv = 0; 1044 1045 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1046 return EOPNOTSUPP; 1047 1048 if (vp->v_type == VREG) { 1049 rump_hyperfree(rn->rn_data, rn->rn_dlen); 1050 } 1051 1052 freedir(rnd, cnp); 1053 rn->rn_flags |= RUMPNODE_CANRECLAIM; 1054 1055 vput(dvp); 1056 vput(vp); 1057 1058 return rv; 1059 } 1060 1061 static int 1062 rump_vop_mknod(void *v) 1063 { 1064 struct vop_mknod_args /* { 1065 struct vnode *a_dvp; 1066 struct vnode **a_vpp; 1067 struct componentname *a_cnp; 1068 struct vattr *a_vap; 1069 }; */ *ap = v; 1070 struct vnode *dvp = ap->a_dvp; 1071 struct vnode **vpp = ap->a_vpp; 1072 struct componentname *cnp = ap->a_cnp; 1073 struct vattr *va = ap->a_vap; 1074 struct rumpfs_node *rnd = dvp->v_data, *rn; 1075 int rv; 1076 1077 rn = makeprivate(va->va_type, va->va_rdev, DEV_BSIZE, false); 1078 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1079 rn->rn_va.va_flags |= UF_OPAQUE; 1080 rv = makevnode(dvp->v_mount, rn, vpp); 1081 if (rv) 1082 goto out; 1083 1084 makedir(rnd, cnp, rn); 1085 1086 out: 1087 vput(dvp); 1088 return rv; 1089 } 1090 1091 static int 1092 rump_vop_create(void *v) 1093 { 1094 struct vop_create_args /* { 1095 struct vnode *a_dvp; 1096 struct vnode **a_vpp; 1097 struct componentname *a_cnp; 1098 struct vattr *a_vap; 1099 }; */ *ap = v; 1100 struct vnode *dvp = ap->a_dvp; 1101 struct vnode **vpp = ap->a_vpp; 1102 struct componentname *cnp = ap->a_cnp; 1103 struct vattr *va = ap->a_vap; 1104 struct rumpfs_node *rnd = dvp->v_data, *rn; 1105 off_t newsize; 1106 int rv; 1107 1108 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0; 1109 rn = makeprivate(va->va_type, NODEV, newsize, false); 1110 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1111 rn->rn_va.va_flags |= UF_OPAQUE; 1112 rv = makevnode(dvp->v_mount, rn, vpp); 1113 if (rv) 1114 goto out; 1115 1116 makedir(rnd, cnp, rn); 1117 1118 out: 1119 vput(dvp); 1120 return rv; 1121 } 1122 1123 static int 1124 rump_vop_symlink(void *v) 1125 { 1126 struct vop_symlink_args /* { 1127 struct vnode *a_dvp; 1128 struct vnode **a_vpp; 1129 struct componentname *a_cnp; 1130 struct vattr *a_vap; 1131 char *a_target; 1132 }; */ *ap = v; 1133 struct vnode *dvp = ap->a_dvp; 1134 struct vnode **vpp = ap->a_vpp; 1135 struct componentname *cnp = ap->a_cnp; 1136 struct rumpfs_node *rnd = dvp->v_data, *rn; 1137 const char *target = ap->a_target; 1138 size_t linklen; 1139 int rv; 1140 1141 linklen = strlen(target); 1142 KASSERT(linklen < MAXPATHLEN); 1143 rn = makeprivate(VLNK, NODEV, linklen, false); 1144 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1145 rn->rn_va.va_flags |= UF_OPAQUE; 1146 rv = makevnode(dvp->v_mount, rn, vpp); 1147 if (rv) 1148 goto out; 1149 1150 makedir(rnd, cnp, rn); 1151 1152 KASSERT(linklen < MAXPATHLEN); 1153 rn->rn_linktarg = PNBUF_GET(); 1154 rn->rn_linklen = linklen; 1155 strcpy(rn->rn_linktarg, target); 1156 1157 out: 1158 vput(dvp); 1159 return rv; 1160 } 1161 1162 static int 1163 rump_vop_readlink(void *v) 1164 { 1165 struct vop_readlink_args /* { 1166 struct vnode *a_vp; 1167 struct uio *a_uio; 1168 kauth_cred_t a_cred; 1169 }; */ *ap = v; 1170 struct vnode *vp = ap->a_vp; 1171 struct rumpfs_node *rn = vp->v_data; 1172 struct uio *uio = ap->a_uio; 1173 1174 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio); 1175 } 1176 1177 static int 1178 rump_vop_whiteout(void *v) 1179 { 1180 struct vop_whiteout_args /* { 1181 struct vnode *a_dvp; 1182 struct componentname *a_cnp; 1183 int a_flags; 1184 } */ *ap = v; 1185 struct vnode *dvp = ap->a_dvp; 1186 struct rumpfs_node *rnd = dvp->v_data; 1187 struct componentname *cnp = ap->a_cnp; 1188 int flags = ap->a_flags; 1189 1190 switch (flags) { 1191 case LOOKUP: 1192 break; 1193 case CREATE: 1194 makedir(rnd, cnp, RUMPFS_WHITEOUT); 1195 break; 1196 case DELETE: 1197 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */ 1198 freedir(rnd, cnp); 1199 break; 1200 default: 1201 panic("unknown whiteout op %d", flags); 1202 } 1203 1204 return 0; 1205 } 1206 1207 static int 1208 rump_vop_open(void *v) 1209 { 1210 struct vop_open_args /* { 1211 struct vnode *a_vp; 1212 int a_mode; 1213 kauth_cred_t a_cred; 1214 } */ *ap = v; 1215 struct vnode *vp = ap->a_vp; 1216 struct rumpfs_node *rn = vp->v_data; 1217 int mode = ap->a_mode; 1218 int error = EINVAL; 1219 1220 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0) 1221 return 0; 1222 1223 if (mode & FREAD) { 1224 if (rn->rn_readfd != -1) 1225 return 0; 1226 rn->rn_readfd = rumpuser_open(rn->rn_hostpath, 1227 O_RDONLY, &error); 1228 } 1229 1230 if (mode & FWRITE) { 1231 if (rn->rn_writefd != -1) 1232 return 0; 1233 rn->rn_writefd = rumpuser_open(rn->rn_hostpath, 1234 O_WRONLY, &error); 1235 } 1236 1237 return error; 1238 } 1239 1240 /* simple readdir. event omits dotstuff and periods */ 1241 static int 1242 rump_vop_readdir(void *v) 1243 { 1244 struct vop_readdir_args /* { 1245 struct vnode *a_vp; 1246 struct uio *a_uio; 1247 kauth_cred_t a_cred; 1248 int *a_eofflag; 1249 off_t **a_cookies; 1250 int *a_ncookies; 1251 } */ *ap = v; 1252 struct vnode *vp = ap->a_vp; 1253 struct uio *uio = ap->a_uio; 1254 struct rumpfs_node *rnd = vp->v_data; 1255 struct rumpfs_dent *rdent; 1256 unsigned i; 1257 int rv = 0; 1258 1259 /* seek to current entry */ 1260 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir); 1261 (i < uio->uio_offset) && rdent; 1262 i++, rdent = LIST_NEXT(rdent, rd_entries)) 1263 continue; 1264 if (!rdent) 1265 goto out; 1266 1267 /* copy entries */ 1268 for (; rdent && uio->uio_resid > 0; 1269 rdent = LIST_NEXT(rdent, rd_entries), i++) { 1270 struct dirent dent; 1271 1272 strlcpy(dent.d_name, rdent->rd_name, sizeof(dent.d_name)); 1273 dent.d_namlen = strlen(dent.d_name); 1274 dent.d_reclen = _DIRENT_RECLEN(&dent, dent.d_namlen); 1275 1276 if (__predict_false(RDENT_ISWHITEOUT(rdent))) { 1277 dent.d_fileno = INO_WHITEOUT; 1278 dent.d_type = DT_WHT; 1279 } else { 1280 dent.d_fileno = rdent->rd_node->rn_va.va_fileid; 1281 dent.d_type = vtype2dt(rdent->rd_node->rn_va.va_type); 1282 } 1283 1284 if (uio->uio_resid < dent.d_reclen) { 1285 i--; 1286 break; 1287 } 1288 1289 rv = uiomove(&dent, dent.d_reclen, uio); 1290 if (rv) { 1291 i--; 1292 break; 1293 } 1294 } 1295 1296 out: 1297 if (ap->a_cookies) { 1298 *ap->a_ncookies = 0; 1299 *ap->a_cookies = NULL; 1300 } 1301 if (rdent) 1302 *ap->a_eofflag = 0; 1303 else 1304 *ap->a_eofflag = 1; 1305 uio->uio_offset = i; 1306 1307 return rv; 1308 } 1309 1310 static int 1311 etread(struct rumpfs_node *rn, struct uio *uio) 1312 { 1313 uint8_t *buf; 1314 size_t bufsize; 1315 ssize_t n; 1316 int error = 0; 1317 1318 bufsize = uio->uio_resid; 1319 if (bufsize == 0) 1320 return 0; 1321 buf = kmem_alloc(bufsize, KM_SLEEP); 1322 if ((n = rumpuser_pread(rn->rn_readfd, buf, bufsize, 1323 uio->uio_offset + rn->rn_offset, &error)) == -1) 1324 goto out; 1325 KASSERT(n <= bufsize); 1326 error = uiomove(buf, n, uio); 1327 1328 out: 1329 kmem_free(buf, bufsize); 1330 return error; 1331 1332 } 1333 1334 static int 1335 rump_vop_read(void *v) 1336 { 1337 struct vop_read_args /* { 1338 struct vnode *a_vp; 1339 struct uio *a_uio; 1340 int ioflags a_ioflag; 1341 kauth_cred_t a_cred; 1342 }; */ *ap = v; 1343 struct vnode *vp = ap->a_vp; 1344 struct rumpfs_node *rn = vp->v_data; 1345 struct uio *uio = ap->a_uio; 1346 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1347 off_t chunk; 1348 int error = 0; 1349 1350 if (vp->v_type == VDIR) 1351 return EISDIR; 1352 1353 /* et op? */ 1354 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1355 return etread(rn, uio); 1356 1357 /* otherwise, it's off to ubc with us */ 1358 while (uio->uio_resid > 0) { 1359 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1360 if (chunk == 0) 1361 break; 1362 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1363 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1364 if (error) 1365 break; 1366 } 1367 1368 return error; 1369 } 1370 1371 static int 1372 etwrite(struct rumpfs_node *rn, struct uio *uio) 1373 { 1374 uint8_t *buf; 1375 size_t bufsize; 1376 ssize_t n; 1377 int error = 0; 1378 1379 bufsize = uio->uio_resid; 1380 if (bufsize == 0) 1381 return 0; 1382 buf = kmem_alloc(bufsize, KM_SLEEP); 1383 error = uiomove(buf, bufsize, uio); 1384 if (error) 1385 goto out; 1386 KASSERT(uio->uio_resid == 0); 1387 n = rumpuser_pwrite(rn->rn_writefd, buf, bufsize, 1388 (uio->uio_offset-bufsize) + rn->rn_offset, &error); 1389 if (n >= 0) { 1390 KASSERT(n <= bufsize); 1391 uio->uio_resid = bufsize - n; 1392 } 1393 1394 out: 1395 kmem_free(buf, bufsize); 1396 return error; 1397 } 1398 1399 static int 1400 rump_vop_write(void *v) 1401 { 1402 struct vop_read_args /* { 1403 struct vnode *a_vp; 1404 struct uio *a_uio; 1405 int ioflags a_ioflag; 1406 kauth_cred_t a_cred; 1407 }; */ *ap = v; 1408 struct vnode *vp = ap->a_vp; 1409 struct rumpfs_node *rn = vp->v_data; 1410 struct uio *uio = ap->a_uio; 1411 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1412 void *olddata; 1413 size_t oldlen, newlen; 1414 off_t chunk; 1415 int error = 0; 1416 bool allocd = false; 1417 1418 if (ap->a_ioflag & IO_APPEND) 1419 uio->uio_offset = vp->v_size; 1420 1421 /* consult et? */ 1422 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1423 return etwrite(rn, uio); 1424 1425 /* 1426 * Otherwise, it's a case of ubcmove. 1427 */ 1428 1429 /* 1430 * First, make sure we have enough storage. 1431 * 1432 * No, you don't need to tell me it's not very efficient. 1433 * No, it doesn't really support sparse files, just fakes it. 1434 */ 1435 newlen = uio->uio_offset + uio->uio_resid; 1436 oldlen = 0; /* XXXgcc */ 1437 olddata = NULL; 1438 if (rn->rn_dlen < newlen) { 1439 oldlen = rn->rn_dlen; 1440 olddata = rn->rn_data; 1441 1442 rn->rn_data = rump_hypermalloc(newlen, 0, true, "rumpfs"); 1443 rn->rn_dlen = newlen; 1444 memset(rn->rn_data, 0, newlen); 1445 memcpy(rn->rn_data, olddata, oldlen); 1446 allocd = true; 1447 uvm_vnp_setsize(vp, newlen); 1448 } 1449 1450 /* ok, we have enough stooorage. write */ 1451 while (uio->uio_resid > 0) { 1452 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1453 if (chunk == 0) 1454 break; 1455 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1456 UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1457 if (error) 1458 break; 1459 } 1460 1461 if (allocd) { 1462 if (error) { 1463 rump_hyperfree(rn->rn_data, newlen); 1464 rn->rn_data = olddata; 1465 rn->rn_dlen = oldlen; 1466 uvm_vnp_setsize(vp, oldlen); 1467 } else { 1468 rump_hyperfree(olddata, oldlen); 1469 } 1470 } 1471 1472 return error; 1473 } 1474 1475 static int 1476 rump_vop_bmap(void *v) 1477 { 1478 struct vop_bmap_args /* { 1479 struct vnode *a_vp; 1480 daddr_t a_bn; 1481 struct vnode **a_vpp; 1482 daddr_t *a_bnp; 1483 int *a_runp; 1484 } */ *ap = v; 1485 1486 /* 1:1 mapping */ 1487 if (ap->a_vpp) 1488 *ap->a_vpp = ap->a_vp; 1489 if (ap->a_bnp) 1490 *ap->a_bnp = ap->a_bn; 1491 if (ap->a_runp) 1492 *ap->a_runp = 16; 1493 1494 return 0; 1495 } 1496 1497 static int 1498 rump_vop_strategy(void *v) 1499 { 1500 struct vop_strategy_args /* { 1501 struct vnode *a_vp; 1502 struct buf *a_bp; 1503 } */ *ap = v; 1504 struct vnode *vp = ap->a_vp; 1505 struct rumpfs_node *rn = vp->v_data; 1506 struct buf *bp = ap->a_bp; 1507 off_t copylen, copyoff; 1508 int error; 1509 1510 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) { 1511 error = EINVAL; 1512 goto out; 1513 } 1514 1515 copyoff = bp->b_blkno << DEV_BSHIFT; 1516 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount); 1517 if (BUF_ISWRITE(bp)) { 1518 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen); 1519 } else { 1520 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen); 1521 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen); 1522 } 1523 bp->b_resid = 0; 1524 error = 0; 1525 1526 out: 1527 bp->b_error = error; 1528 biodone(bp); 1529 return 0; 1530 } 1531 1532 static int 1533 rump_vop_pathconf(void *v) 1534 { 1535 struct vop_pathconf_args /* { 1536 struct vnode *a_vp; 1537 int a_name; 1538 register_t *a_retval; 1539 }; */ *ap = v; 1540 int name = ap->a_name; 1541 register_t *retval = ap->a_retval; 1542 1543 switch (name) { 1544 case _PC_LINK_MAX: 1545 *retval = LINK_MAX; 1546 return 0; 1547 case _PC_NAME_MAX: 1548 *retval = RUMPFS_MAXNAMLEN; 1549 return 0; 1550 case _PC_PATH_MAX: 1551 *retval = PATH_MAX; 1552 return 0; 1553 case _PC_PIPE_BUF: 1554 *retval = PIPE_BUF; 1555 return 0; 1556 case _PC_CHOWN_RESTRICTED: 1557 *retval = 1; 1558 return 0; 1559 case _PC_NO_TRUNC: 1560 *retval = 1; 1561 return 0; 1562 case _PC_SYNC_IO: 1563 *retval = 1; 1564 return 0; 1565 case _PC_FILESIZEBITS: 1566 *retval = 43; /* this one goes to 11 */ 1567 return 0; 1568 case _PC_SYMLINK_MAX: 1569 *retval = MAXPATHLEN; 1570 return 0; 1571 case _PC_2_SYMLINKS: 1572 *retval = 1; 1573 return 0; 1574 default: 1575 return EINVAL; 1576 } 1577 } 1578 1579 static int 1580 rump_vop_success(void *v) 1581 { 1582 1583 return 0; 1584 } 1585 1586 static int 1587 rump_vop_inactive(void *v) 1588 { 1589 struct vop_inactive_args /* { 1590 struct vnode *a_vp; 1591 bool *a_recycle; 1592 } */ *ap = v; 1593 struct vnode *vp = ap->a_vp; 1594 struct rumpfs_node *rn = vp->v_data; 1595 int error; 1596 1597 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) { 1598 if (rn->rn_readfd != -1) { 1599 rumpuser_close(rn->rn_readfd, &error); 1600 rn->rn_readfd = -1; 1601 } 1602 if (rn->rn_writefd != -1) { 1603 rumpuser_close(rn->rn_writefd, &error); 1604 rn->rn_writefd = -1; 1605 } 1606 } 1607 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false; 1608 1609 VOP_UNLOCK(vp); 1610 return 0; 1611 } 1612 1613 static int 1614 rump_vop_reclaim(void *v) 1615 { 1616 struct vop_reclaim_args /* { 1617 struct vnode *a_vp; 1618 } */ *ap = v; 1619 struct vnode *vp = ap->a_vp; 1620 struct rumpfs_node *rn = vp->v_data; 1621 1622 mutex_enter(&reclock); 1623 rn->rn_vp = NULL; 1624 mutex_exit(&reclock); 1625 genfs_node_destroy(vp); 1626 vp->v_data = NULL; 1627 1628 if (rn->rn_flags & RUMPNODE_CANRECLAIM) { 1629 if (vp->v_type == VLNK) 1630 PNBUF_PUT(rn->rn_linktarg); 1631 if (rn->rn_hostpath) 1632 free(rn->rn_hostpath, M_TEMP); 1633 kmem_free(rn, sizeof(*rn)); 1634 } 1635 1636 return 0; 1637 } 1638 1639 static int 1640 rump_vop_spec(void *v) 1641 { 1642 struct vop_generic_args *ap = v; 1643 int (**opvec)(void *); 1644 1645 switch (ap->a_desc->vdesc_offset) { 1646 case VOP_ACCESS_DESCOFFSET: 1647 case VOP_GETATTR_DESCOFFSET: 1648 case VOP_SETATTR_DESCOFFSET: 1649 case VOP_LOCK_DESCOFFSET: 1650 case VOP_UNLOCK_DESCOFFSET: 1651 case VOP_ISLOCKED_DESCOFFSET: 1652 case VOP_RECLAIM_DESCOFFSET: 1653 opvec = rump_vnodeop_p; 1654 break; 1655 default: 1656 opvec = spec_vnodeop_p; 1657 break; 1658 } 1659 1660 return VOCALL(opvec, ap->a_desc->vdesc_offset, v); 1661 } 1662 1663 static int 1664 rump_vop_advlock(void *v) 1665 { 1666 struct vop_advlock_args /* { 1667 const struct vnodeop_desc *a_desc; 1668 struct vnode *a_vp; 1669 void *a_id; 1670 int a_op; 1671 struct flock *a_fl; 1672 int a_flags; 1673 } */ *ap = v; 1674 struct vnode *vp = ap->a_vp; 1675 struct rumpfs_node *rn = vp->v_data; 1676 1677 return lf_advlock(ap, &rn->rn_lockf, vp->v_size); 1678 } 1679 1680 /* 1681 * Begin vfs-level stuff 1682 */ 1683 1684 VFS_PROTOS(rumpfs); 1685 struct vfsops rumpfs_vfsops = { 1686 .vfs_name = MOUNT_RUMPFS, 1687 .vfs_min_mount_data = 0, 1688 .vfs_mount = rumpfs_mount, 1689 .vfs_start = (void *)nullop, 1690 .vfs_unmount = rumpfs_unmount, 1691 .vfs_root = rumpfs_root, 1692 .vfs_quotactl = (void *)eopnotsupp, 1693 .vfs_statvfs = genfs_statvfs, 1694 .vfs_sync = (void *)nullop, 1695 .vfs_vget = rumpfs_vget, 1696 .vfs_fhtovp = (void *)eopnotsupp, 1697 .vfs_vptofh = (void *)eopnotsupp, 1698 .vfs_init = rumpfs_init, 1699 .vfs_reinit = NULL, 1700 .vfs_done = rumpfs_done, 1701 .vfs_mountroot = rumpfs_mountroot, 1702 .vfs_snapshot = (void *)eopnotsupp, 1703 .vfs_extattrctl = (void *)eopnotsupp, 1704 .vfs_suspendctl = (void *)eopnotsupp, 1705 .vfs_renamelock_enter = genfs_renamelock_enter, 1706 .vfs_renamelock_exit = genfs_renamelock_exit, 1707 .vfs_opv_descs = rump_opv_descs, 1708 /* vfs_refcount */ 1709 /* vfs_list */ 1710 }; 1711 1712 static int 1713 rumpfs_mountfs(struct mount *mp) 1714 { 1715 struct rumpfs_mount *rfsmp; 1716 struct rumpfs_node *rn; 1717 int error; 1718 1719 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP); 1720 1721 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false); 1722 rn->rn_parent = rn; 1723 if ((error = makevnode(mp, rn, &rfsmp->rfsmp_rvp)) != 0) 1724 return error; 1725 1726 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT; 1727 VOP_UNLOCK(rfsmp->rfsmp_rvp); 1728 1729 mp->mnt_data = rfsmp; 1730 mp->mnt_stat.f_namemax = RUMPFS_MAXNAMLEN; 1731 mp->mnt_stat.f_iosize = 512; 1732 mp->mnt_flag |= MNT_LOCAL; 1733 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO; 1734 mp->mnt_fs_bshift = DEV_BSHIFT; 1735 vfs_getnewfsid(mp); 1736 1737 return 0; 1738 } 1739 1740 int 1741 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen) 1742 { 1743 int error; 1744 1745 if (mp->mnt_flag & MNT_UPDATE) { 1746 return 0; 1747 } 1748 1749 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE, 1750 mp->mnt_op->vfs_name, mp, curlwp); 1751 if (error) 1752 return error; 1753 1754 return rumpfs_mountfs(mp); 1755 } 1756 1757 int 1758 rumpfs_unmount(struct mount *mp, int mntflags) 1759 { 1760 struct rumpfs_mount *rfsmp = mp->mnt_data; 1761 int flags = 0, error; 1762 1763 if (panicstr || mntflags & MNT_FORCE) 1764 flags |= FORCECLOSE; 1765 1766 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0) 1767 return error; 1768 vgone(rfsmp->rfsmp_rvp); /* XXX */ 1769 1770 kmem_free(rfsmp, sizeof(*rfsmp)); 1771 1772 return 0; 1773 } 1774 1775 int 1776 rumpfs_root(struct mount *mp, struct vnode **vpp) 1777 { 1778 struct rumpfs_mount *rfsmp = mp->mnt_data; 1779 1780 vref(rfsmp->rfsmp_rvp); 1781 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY); 1782 *vpp = rfsmp->rfsmp_rvp; 1783 return 0; 1784 } 1785 1786 int 1787 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1788 { 1789 1790 return EOPNOTSUPP; 1791 } 1792 1793 void 1794 rumpfs_init() 1795 { 1796 1797 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET); 1798 1799 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE); 1800 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE); 1801 } 1802 1803 void 1804 rumpfs_done() 1805 { 1806 1807 mutex_destroy(&reclock); 1808 mutex_destroy(&etfs_lock); 1809 } 1810 1811 int 1812 rumpfs_mountroot() 1813 { 1814 struct mount *mp; 1815 int error; 1816 1817 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) { 1818 vrele(rootvp); 1819 return error; 1820 } 1821 1822 if ((error = rumpfs_mountfs(mp)) != 0) 1823 panic("mounting rootfs failed: %d", error); 1824 1825 mutex_enter(&mountlist_lock); 1826 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1827 mutex_exit(&mountlist_lock); 1828 1829 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE, 1830 mp->mnt_op->vfs_name, mp, curlwp); 1831 if (error) 1832 panic("set_statvfs_info failed for rootfs: %d", error); 1833 1834 mp->mnt_flag &= ~MNT_RDONLY; 1835 vfs_unbusy(mp, false, NULL); 1836 1837 return 0; 1838 } 1839
Indexes created Tue Sep 30 07:10:03 GMT 2025