rumpfs.c revision 1.98
1 /* $NetBSD: rumpfs.c,v 1.98 2011/08/07 05:56:32 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 1.98 2011/08/07 05:56:32 hannken Exp $"); 30 31 #include <sys/param.h> 32 #include <sys/atomic.h> 33 #include <sys/buf.h> 34 #include <sys/dirent.h> 35 #include <sys/errno.h> 36 #include <sys/filedesc.h> 37 #include <sys/fcntl.h> 38 #include <sys/kauth.h> 39 #include <sys/malloc.h> 40 #include <sys/module.h> 41 #include <sys/mount.h> 42 #include <sys/namei.h> 43 #include <sys/lock.h> 44 #include <sys/lockf.h> 45 #include <sys/queue.h> 46 #include <sys/stat.h> 47 #include <sys/syscallargs.h> 48 #include <sys/vnode.h> 49 #include <sys/unistd.h> 50 51 #include <miscfs/fifofs/fifo.h> 52 #include <miscfs/specfs/specdev.h> 53 #include <miscfs/genfs/genfs.h> 54 #include <miscfs/genfs/genfs_node.h> 55 56 #include <uvm/uvm_extern.h> 57 58 #include <rump/rumpuser.h> 59 60 #include "rump_private.h" 61 #include "rump_vfs_private.h" 62 63 static int rump_vop_lookup(void *); 64 static int rump_vop_getattr(void *); 65 static int rump_vop_setattr(void *); 66 static int rump_vop_mkdir(void *); 67 static int rump_vop_rmdir(void *); 68 static int rump_vop_remove(void *); 69 static int rump_vop_mknod(void *); 70 static int rump_vop_create(void *); 71 static int rump_vop_inactive(void *); 72 static int rump_vop_reclaim(void *); 73 static int rump_vop_success(void *); 74 static int rump_vop_readdir(void *); 75 static int rump_vop_spec(void *); 76 static int rump_vop_read(void *); 77 static int rump_vop_write(void *); 78 static int rump_vop_open(void *); 79 static int rump_vop_symlink(void *); 80 static int rump_vop_readlink(void *); 81 static int rump_vop_whiteout(void *); 82 static int rump_vop_pathconf(void *); 83 static int rump_vop_bmap(void *); 84 static int rump_vop_strategy(void *); 85 static int rump_vop_advlock(void *); 86 static int rump_vop_access(void *); 87 88 int (**fifo_vnodeop_p)(void *); 89 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = { 90 { &vop_default_desc, vn_default_error }, 91 { 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; 667 668 *vpp = NULL; 669 670 if ((cnp->cn_flags & ISLASTCN) 671 && (dvp->v_mount->mnt_flag & MNT_RDONLY) 672 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 673 return EROFS; 674 675 /* check for dot, return directly if the case */ 676 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 677 vref(dvp); 678 *vpp = dvp; 679 return 0; 680 } 681 682 /* we don't do rename */ 683 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME))) 684 return EOPNOTSUPP; 685 686 /* check for etfs */ 687 if (dvp == rootvnode && 688 (cnp->cn_nameiop == LOOKUP || cnp->cn_nameiop == CREATE)) { 689 bool found; 690 mutex_enter(&etfs_lock); 691 found = etfs_find(cnp->cn_nameptr, &et, false); 692 mutex_exit(&etfs_lock); 693 694 if (found) { 695 rn = et->et_rn; 696 cnp->cn_consume += et->et_keylen - cnp->cn_namelen; 697 if (rn->rn_va.va_type != VDIR) 698 cnp->cn_flags &= ~REQUIREDIR; 699 goto getvnode; 700 } 701 } 702 703 if (rnd->rn_flags & RUMPNODE_DIR_ET) { 704 uint64_t fsize; 705 char *newpath; 706 size_t newpathlen; 707 int hft, error; 708 709 if (dotdot) 710 return EOPNOTSUPP; 711 712 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1; 713 newpath = malloc(newpathlen, M_TEMP, M_WAITOK); 714 715 strlcpy(newpath, rnd->rn_hostpath, newpathlen); 716 strlcat(newpath, "/", newpathlen); 717 strlcat(newpath, cnp->cn_nameptr, newpathlen); 718 719 if (rumpuser_getfileinfo(newpath, &fsize, &hft, &error)) { 720 free(newpath, M_TEMP); 721 return error; 722 } 723 724 /* allow only dirs and regular files */ 725 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) { 726 free(newpath, M_TEMP); 727 return ENOENT; 728 } 729 730 rn = makeprivate(hft_to_vtype(hft), NODEV, fsize, true); 731 rn->rn_flags |= RUMPNODE_CANRECLAIM; 732 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) { 733 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS; 734 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST; 735 } 736 rn->rn_hostpath = newpath; 737 738 goto getvnode; 739 } else { 740 if (dotdot) { 741 if ((rn = rnd->rn_parent) != NULL) 742 goto getvnode; 743 } else { 744 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) { 745 if (rd->rd_namelen == cnp->cn_namelen && 746 strncmp(rd->rd_name, cnp->cn_nameptr, 747 cnp->cn_namelen) == 0) 748 break; 749 } 750 } 751 } 752 753 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE)) 754 return ENOENT; 755 756 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 757 if (dvp->v_mount->mnt_flag & MNT_RDONLY) 758 return EROFS; 759 return EJUSTRETURN; 760 } 761 762 if (RDENT_ISWHITEOUT(rd)) { 763 cnp->cn_flags |= ISWHITEOUT; 764 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) 765 return EJUSTRETURN; 766 return ENOENT; 767 } 768 769 rn = rd->rd_node; 770 771 getvnode: 772 KASSERT(rn); 773 if (dotdot) 774 VOP_UNLOCK(dvp); 775 mutex_enter(&reclock); 776 if ((vp = rn->rn_vp)) { 777 mutex_enter(vp->v_interlock); 778 mutex_exit(&reclock); 779 if (vget(vp, LK_EXCLUSIVE)) { 780 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 781 goto getvnode; 782 } 783 *vpp = vp; 784 } else { 785 mutex_exit(&reclock); 786 rv = makevnode(dvp->v_mount, rn, vpp); 787 } 788 if (dotdot) 789 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 790 791 return rv; 792 } 793 794 int 795 rump_vop_access(void *v) 796 { 797 struct vop_access_args /* { 798 const struct vnodeop_desc *a_desc; 799 struct vnode *a_vp; 800 int a_mode; 801 kauth_cred_t a_cred; 802 } */ *ap = v; 803 struct vnode *vp = ap->a_vp; 804 int mode = ap->a_mode; 805 806 if (mode & VWRITE) { 807 switch (vp->v_type) { 808 case VDIR: 809 case VLNK: 810 case VREG: 811 if ((vp->v_mount->mnt_flag & MNT_RDONLY)) 812 return EROFS; 813 break; 814 default: 815 break; 816 } 817 } 818 819 return 0; 820 } 821 822 static int 823 rump_vop_getattr(void *v) 824 { 825 struct vop_getattr_args /* { 826 struct vnode *a_vp; 827 struct vattr *a_vap; 828 kauth_cred_t a_cred; 829 } */ *ap = v; 830 struct vnode *vp = ap->a_vp; 831 struct rumpfs_node *rn = vp->v_data; 832 struct vattr *vap = ap->a_vap; 833 834 memcpy(vap, &rn->rn_va, sizeof(struct vattr)); 835 vap->va_size = vp->v_size; 836 return 0; 837 } 838 839 static int 840 rump_vop_setattr(void *v) 841 { 842 struct vop_getattr_args /* { 843 struct vnode *a_vp; 844 struct vattr *a_vap; 845 kauth_cred_t a_cred; 846 } */ *ap = v; 847 struct vnode *vp = ap->a_vp; 848 struct vattr *vap = ap->a_vap; 849 struct rumpfs_node *rn = vp->v_data; 850 851 #define SETIFVAL(a,t) if (vap->a != (t)VNOVAL) rn->rn_va.a = vap->a 852 SETIFVAL(va_mode, mode_t); 853 SETIFVAL(va_uid, uid_t); 854 SETIFVAL(va_gid, gid_t); 855 SETIFVAL(va_atime.tv_sec, time_t); 856 SETIFVAL(va_ctime.tv_sec, time_t); 857 SETIFVAL(va_mtime.tv_sec, time_t); 858 SETIFVAL(va_birthtime.tv_sec, time_t); 859 SETIFVAL(va_atime.tv_nsec, long); 860 SETIFVAL(va_ctime.tv_nsec, long); 861 SETIFVAL(va_mtime.tv_nsec, long); 862 SETIFVAL(va_birthtime.tv_nsec, long); 863 SETIFVAL(va_flags, u_long); 864 #undef SETIFVAL 865 866 if (vp->v_type == VREG && 867 vap->va_size != VSIZENOTSET && 868 vap->va_size != rn->rn_dlen) { 869 void *newdata; 870 size_t copylen, newlen; 871 872 newlen = vap->va_size; 873 newdata = rump_hypermalloc(newlen, 0, true, "rumpfs"); 874 875 copylen = MIN(rn->rn_dlen, newlen); 876 memset(newdata, 0, newlen); 877 memcpy(newdata, rn->rn_data, copylen); 878 rump_hyperfree(rn->rn_data, rn->rn_dlen); 879 880 rn->rn_data = newdata; 881 rn->rn_dlen = newlen; 882 uvm_vnp_setsize(vp, newlen); 883 } 884 return 0; 885 } 886 887 static int 888 rump_vop_mkdir(void *v) 889 { 890 struct vop_mkdir_args /* { 891 struct vnode *a_dvp; 892 struct vnode **a_vpp; 893 struct componentname *a_cnp; 894 struct vattr *a_vap; 895 }; */ *ap = v; 896 struct vnode *dvp = ap->a_dvp; 897 struct vnode **vpp = ap->a_vpp; 898 struct componentname *cnp = ap->a_cnp; 899 struct rumpfs_node *rnd = dvp->v_data, *rn; 900 int rv = 0; 901 902 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false); 903 if ((cnp->cn_flags & ISWHITEOUT) != 0) 904 rn->rn_va.va_flags |= UF_OPAQUE; 905 rn->rn_parent = rnd; 906 rv = makevnode(dvp->v_mount, rn, vpp); 907 if (rv) 908 goto out; 909 910 makedir(rnd, cnp, rn); 911 912 out: 913 vput(dvp); 914 return rv; 915 } 916 917 static int 918 rump_vop_rmdir(void *v) 919 { 920 struct vop_rmdir_args /* { 921 struct vnode *a_dvp; 922 struct vnode *a_vp; 923 struct componentname *a_cnp; 924 }; */ *ap = v; 925 struct vnode *dvp = ap->a_dvp; 926 struct vnode *vp = ap->a_vp; 927 struct componentname *cnp = ap->a_cnp; 928 struct rumpfs_node *rnd = dvp->v_data; 929 struct rumpfs_node *rn = vp->v_data; 930 struct rumpfs_dent *rd; 931 int rv = 0; 932 933 LIST_FOREACH(rd, &rn->rn_dir, rd_entries) { 934 if (rd->rd_node != RUMPFS_WHITEOUT) { 935 rv = ENOTEMPTY; 936 goto out; 937 } 938 } 939 while ((rd = LIST_FIRST(&rn->rn_dir)) != NULL) { 940 KASSERT(rd->rd_node == RUMPFS_WHITEOUT); 941 LIST_REMOVE(rd, rd_entries); 942 kmem_free(rd->rd_name, rd->rd_namelen+1); 943 kmem_free(rd, sizeof(*rd)); 944 } 945 946 freedir(rnd, cnp); 947 rn->rn_flags |= RUMPNODE_CANRECLAIM; 948 rn->rn_parent = NULL; 949 950 out: 951 vput(dvp); 952 vput(vp); 953 954 return rv; 955 } 956 957 static int 958 rump_vop_remove(void *v) 959 { 960 struct vop_rmdir_args /* { 961 struct vnode *a_dvp; 962 struct vnode *a_vp; 963 struct componentname *a_cnp; 964 }; */ *ap = v; 965 struct vnode *dvp = ap->a_dvp; 966 struct vnode *vp = ap->a_vp; 967 struct componentname *cnp = ap->a_cnp; 968 struct rumpfs_node *rnd = dvp->v_data; 969 struct rumpfs_node *rn = vp->v_data; 970 int rv = 0; 971 972 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 973 return EOPNOTSUPP; 974 975 if (vp->v_type == VREG) { 976 rump_hyperfree(rn->rn_data, rn->rn_dlen); 977 } 978 979 freedir(rnd, cnp); 980 rn->rn_flags |= RUMPNODE_CANRECLAIM; 981 982 vput(dvp); 983 vput(vp); 984 985 return rv; 986 } 987 988 static int 989 rump_vop_mknod(void *v) 990 { 991 struct vop_mknod_args /* { 992 struct vnode *a_dvp; 993 struct vnode **a_vpp; 994 struct componentname *a_cnp; 995 struct vattr *a_vap; 996 }; */ *ap = v; 997 struct vnode *dvp = ap->a_dvp; 998 struct vnode **vpp = ap->a_vpp; 999 struct componentname *cnp = ap->a_cnp; 1000 struct vattr *va = ap->a_vap; 1001 struct rumpfs_node *rnd = dvp->v_data, *rn; 1002 int rv; 1003 1004 rn = makeprivate(va->va_type, va->va_rdev, DEV_BSIZE, false); 1005 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1006 rn->rn_va.va_flags |= UF_OPAQUE; 1007 rv = makevnode(dvp->v_mount, rn, vpp); 1008 if (rv) 1009 goto out; 1010 1011 makedir(rnd, cnp, rn); 1012 1013 out: 1014 vput(dvp); 1015 return rv; 1016 } 1017 1018 static int 1019 rump_vop_create(void *v) 1020 { 1021 struct vop_create_args /* { 1022 struct vnode *a_dvp; 1023 struct vnode **a_vpp; 1024 struct componentname *a_cnp; 1025 struct vattr *a_vap; 1026 }; */ *ap = v; 1027 struct vnode *dvp = ap->a_dvp; 1028 struct vnode **vpp = ap->a_vpp; 1029 struct componentname *cnp = ap->a_cnp; 1030 struct vattr *va = ap->a_vap; 1031 struct rumpfs_node *rnd = dvp->v_data, *rn; 1032 off_t newsize; 1033 int rv; 1034 1035 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0; 1036 rn = makeprivate(va->va_type, NODEV, newsize, false); 1037 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1038 rn->rn_va.va_flags |= UF_OPAQUE; 1039 rv = makevnode(dvp->v_mount, rn, vpp); 1040 if (rv) 1041 goto out; 1042 1043 makedir(rnd, cnp, rn); 1044 1045 out: 1046 vput(dvp); 1047 return rv; 1048 } 1049 1050 static int 1051 rump_vop_symlink(void *v) 1052 { 1053 struct vop_symlink_args /* { 1054 struct vnode *a_dvp; 1055 struct vnode **a_vpp; 1056 struct componentname *a_cnp; 1057 struct vattr *a_vap; 1058 char *a_target; 1059 }; */ *ap = v; 1060 struct vnode *dvp = ap->a_dvp; 1061 struct vnode **vpp = ap->a_vpp; 1062 struct componentname *cnp = ap->a_cnp; 1063 struct rumpfs_node *rnd = dvp->v_data, *rn; 1064 const char *target = ap->a_target; 1065 size_t linklen; 1066 int rv; 1067 1068 linklen = strlen(target); 1069 KASSERT(linklen < MAXPATHLEN); 1070 rn = makeprivate(VLNK, NODEV, linklen, false); 1071 if ((cnp->cn_flags & ISWHITEOUT) != 0) 1072 rn->rn_va.va_flags |= UF_OPAQUE; 1073 rv = makevnode(dvp->v_mount, rn, vpp); 1074 if (rv) 1075 goto out; 1076 1077 makedir(rnd, cnp, rn); 1078 1079 KASSERT(linklen < MAXPATHLEN); 1080 rn->rn_linktarg = PNBUF_GET(); 1081 rn->rn_linklen = linklen; 1082 strcpy(rn->rn_linktarg, target); 1083 1084 out: 1085 vput(dvp); 1086 return rv; 1087 } 1088 1089 static int 1090 rump_vop_readlink(void *v) 1091 { 1092 struct vop_readlink_args /* { 1093 struct vnode *a_vp; 1094 struct uio *a_uio; 1095 kauth_cred_t a_cred; 1096 }; */ *ap = v; 1097 struct vnode *vp = ap->a_vp; 1098 struct rumpfs_node *rn = vp->v_data; 1099 struct uio *uio = ap->a_uio; 1100 1101 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio); 1102 } 1103 1104 static int 1105 rump_vop_whiteout(void *v) 1106 { 1107 struct vop_whiteout_args /* { 1108 struct vnode *a_dvp; 1109 struct componentname *a_cnp; 1110 int a_flags; 1111 } */ *ap = v; 1112 struct vnode *dvp = ap->a_dvp; 1113 struct rumpfs_node *rnd = dvp->v_data; 1114 struct componentname *cnp = ap->a_cnp; 1115 int flags = ap->a_flags; 1116 1117 switch (flags) { 1118 case LOOKUP: 1119 break; 1120 case CREATE: 1121 makedir(rnd, cnp, RUMPFS_WHITEOUT); 1122 break; 1123 case DELETE: 1124 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */ 1125 freedir(rnd, cnp); 1126 break; 1127 default: 1128 panic("unknown whiteout op %d", flags); 1129 } 1130 1131 return 0; 1132 } 1133 1134 static int 1135 rump_vop_open(void *v) 1136 { 1137 struct vop_open_args /* { 1138 struct vnode *a_vp; 1139 int a_mode; 1140 kauth_cred_t a_cred; 1141 } */ *ap = v; 1142 struct vnode *vp = ap->a_vp; 1143 struct rumpfs_node *rn = vp->v_data; 1144 int mode = ap->a_mode; 1145 int error = EINVAL; 1146 1147 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0) 1148 return 0; 1149 1150 if (mode & FREAD) { 1151 if (rn->rn_readfd != -1) 1152 return 0; 1153 rn->rn_readfd = rumpuser_open(rn->rn_hostpath, 1154 O_RDONLY, &error); 1155 } 1156 1157 if (mode & FWRITE) { 1158 if (rn->rn_writefd != -1) 1159 return 0; 1160 rn->rn_writefd = rumpuser_open(rn->rn_hostpath, 1161 O_WRONLY, &error); 1162 } 1163 1164 return error; 1165 } 1166 1167 /* simple readdir. event omits dotstuff and periods */ 1168 static int 1169 rump_vop_readdir(void *v) 1170 { 1171 struct vop_readdir_args /* { 1172 struct vnode *a_vp; 1173 struct uio *a_uio; 1174 kauth_cred_t a_cred; 1175 int *a_eofflag; 1176 off_t **a_cookies; 1177 int *a_ncookies; 1178 } */ *ap = v; 1179 struct vnode *vp = ap->a_vp; 1180 struct uio *uio = ap->a_uio; 1181 struct rumpfs_node *rnd = vp->v_data; 1182 struct rumpfs_dent *rdent; 1183 unsigned i; 1184 int rv = 0; 1185 1186 /* seek to current entry */ 1187 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir); 1188 (i < uio->uio_offset) && rdent; 1189 i++, rdent = LIST_NEXT(rdent, rd_entries)) 1190 continue; 1191 if (!rdent) 1192 goto out; 1193 1194 /* copy entries */ 1195 for (; rdent && uio->uio_resid > 0; 1196 rdent = LIST_NEXT(rdent, rd_entries), i++) { 1197 struct dirent dent; 1198 1199 strlcpy(dent.d_name, rdent->rd_name, sizeof(dent.d_name)); 1200 dent.d_namlen = strlen(dent.d_name); 1201 dent.d_reclen = _DIRENT_RECLEN(&dent, dent.d_namlen); 1202 1203 if (__predict_false(RDENT_ISWHITEOUT(rdent))) { 1204 dent.d_fileno = INO_WHITEOUT; 1205 dent.d_type = DT_WHT; 1206 } else { 1207 dent.d_fileno = rdent->rd_node->rn_va.va_fileid; 1208 dent.d_type = vtype2dt(rdent->rd_node->rn_va.va_type); 1209 } 1210 1211 if (uio->uio_resid < dent.d_reclen) { 1212 i--; 1213 break; 1214 } 1215 1216 rv = uiomove(&dent, dent.d_reclen, uio); 1217 if (rv) { 1218 i--; 1219 break; 1220 } 1221 } 1222 1223 out: 1224 if (ap->a_cookies) { 1225 *ap->a_ncookies = 0; 1226 *ap->a_cookies = NULL; 1227 } 1228 if (rdent) 1229 *ap->a_eofflag = 0; 1230 else 1231 *ap->a_eofflag = 1; 1232 uio->uio_offset = i; 1233 1234 return rv; 1235 } 1236 1237 static int 1238 etread(struct rumpfs_node *rn, struct uio *uio) 1239 { 1240 uint8_t *buf; 1241 size_t bufsize; 1242 ssize_t n; 1243 int error = 0; 1244 1245 bufsize = uio->uio_resid; 1246 if (bufsize == 0) 1247 return 0; 1248 buf = kmem_alloc(bufsize, KM_SLEEP); 1249 if ((n = rumpuser_pread(rn->rn_readfd, buf, bufsize, 1250 uio->uio_offset + rn->rn_offset, &error)) == -1) 1251 goto out; 1252 KASSERT(n <= bufsize); 1253 error = uiomove(buf, n, uio); 1254 1255 out: 1256 kmem_free(buf, bufsize); 1257 return error; 1258 1259 } 1260 1261 static int 1262 rump_vop_read(void *v) 1263 { 1264 struct vop_read_args /* { 1265 struct vnode *a_vp; 1266 struct uio *a_uio; 1267 int ioflags a_ioflag; 1268 kauth_cred_t a_cred; 1269 }; */ *ap = v; 1270 struct vnode *vp = ap->a_vp; 1271 struct rumpfs_node *rn = vp->v_data; 1272 struct uio *uio = ap->a_uio; 1273 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1274 off_t chunk; 1275 int error = 0; 1276 1277 /* et op? */ 1278 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1279 return etread(rn, uio); 1280 1281 /* otherwise, it's off to ubc with us */ 1282 while (uio->uio_resid > 0) { 1283 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1284 if (chunk == 0) 1285 break; 1286 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1287 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1288 if (error) 1289 break; 1290 } 1291 1292 return error; 1293 } 1294 1295 static int 1296 etwrite(struct rumpfs_node *rn, struct uio *uio) 1297 { 1298 uint8_t *buf; 1299 size_t bufsize; 1300 ssize_t n; 1301 int error = 0; 1302 1303 bufsize = uio->uio_resid; 1304 if (bufsize == 0) 1305 return 0; 1306 buf = kmem_alloc(bufsize, KM_SLEEP); 1307 error = uiomove(buf, bufsize, uio); 1308 if (error) 1309 goto out; 1310 KASSERT(uio->uio_resid == 0); 1311 n = rumpuser_pwrite(rn->rn_writefd, buf, bufsize, 1312 (uio->uio_offset-bufsize) + rn->rn_offset, &error); 1313 if (n >= 0) { 1314 KASSERT(n <= bufsize); 1315 uio->uio_resid = bufsize - n; 1316 } 1317 1318 out: 1319 kmem_free(buf, bufsize); 1320 return error; 1321 } 1322 1323 static int 1324 rump_vop_write(void *v) 1325 { 1326 struct vop_read_args /* { 1327 struct vnode *a_vp; 1328 struct uio *a_uio; 1329 int ioflags a_ioflag; 1330 kauth_cred_t a_cred; 1331 }; */ *ap = v; 1332 struct vnode *vp = ap->a_vp; 1333 struct rumpfs_node *rn = vp->v_data; 1334 struct uio *uio = ap->a_uio; 1335 const int advice = IO_ADV_DECODE(ap->a_ioflag); 1336 void *olddata; 1337 size_t oldlen, newlen; 1338 off_t chunk; 1339 int error = 0; 1340 bool allocd = false; 1341 1342 if (ap->a_ioflag & IO_APPEND) 1343 uio->uio_offset = vp->v_size; 1344 1345 /* consult et? */ 1346 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) 1347 return etwrite(rn, uio); 1348 1349 /* 1350 * Otherwise, it's a case of ubcmove. 1351 */ 1352 1353 /* 1354 * First, make sure we have enough storage. 1355 * 1356 * No, you don't need to tell me it's not very efficient. 1357 * No, it doesn't really support sparse files, just fakes it. 1358 */ 1359 newlen = uio->uio_offset + uio->uio_resid; 1360 oldlen = 0; /* XXXgcc */ 1361 olddata = NULL; 1362 if (rn->rn_dlen < newlen) { 1363 oldlen = rn->rn_dlen; 1364 olddata = rn->rn_data; 1365 1366 rn->rn_data = rump_hypermalloc(newlen, 0, true, "rumpfs"); 1367 rn->rn_dlen = newlen; 1368 memset(rn->rn_data, 0, newlen); 1369 memcpy(rn->rn_data, olddata, oldlen); 1370 allocd = true; 1371 uvm_vnp_setsize(vp, newlen); 1372 } 1373 1374 /* ok, we have enough stooorage. write */ 1375 while (uio->uio_resid > 0) { 1376 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset); 1377 if (chunk == 0) 1378 break; 1379 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice, 1380 UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 1381 if (error) 1382 break; 1383 } 1384 1385 if (allocd) { 1386 if (error) { 1387 rump_hyperfree(rn->rn_data, newlen); 1388 rn->rn_data = olddata; 1389 rn->rn_dlen = oldlen; 1390 uvm_vnp_setsize(vp, oldlen); 1391 } else { 1392 rump_hyperfree(olddata, oldlen); 1393 } 1394 } 1395 1396 return error; 1397 } 1398 1399 static int 1400 rump_vop_bmap(void *v) 1401 { 1402 struct vop_bmap_args /* { 1403 struct vnode *a_vp; 1404 daddr_t a_bn; 1405 struct vnode **a_vpp; 1406 daddr_t *a_bnp; 1407 int *a_runp; 1408 } */ *ap = v; 1409 1410 /* 1:1 mapping */ 1411 if (ap->a_vpp) 1412 *ap->a_vpp = ap->a_vp; 1413 if (ap->a_bnp) 1414 *ap->a_bnp = ap->a_bn; 1415 if (ap->a_runp) 1416 *ap->a_runp = 16; 1417 1418 return 0; 1419 } 1420 1421 static int 1422 rump_vop_strategy(void *v) 1423 { 1424 struct vop_strategy_args /* { 1425 struct vnode *a_vp; 1426 struct buf *a_bp; 1427 } */ *ap = v; 1428 struct vnode *vp = ap->a_vp; 1429 struct rumpfs_node *rn = vp->v_data; 1430 struct buf *bp = ap->a_bp; 1431 off_t copylen, copyoff; 1432 int error; 1433 1434 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) { 1435 error = EINVAL; 1436 goto out; 1437 } 1438 1439 copyoff = bp->b_blkno << DEV_BSHIFT; 1440 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount); 1441 if (BUF_ISWRITE(bp)) { 1442 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen); 1443 } else { 1444 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen); 1445 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen); 1446 } 1447 bp->b_resid = 0; 1448 error = 0; 1449 1450 out: 1451 bp->b_error = error; 1452 biodone(bp); 1453 return 0; 1454 } 1455 1456 static int 1457 rump_vop_pathconf(void *v) 1458 { 1459 struct vop_pathconf_args /* { 1460 struct vnode *a_vp; 1461 int a_name; 1462 register_t *a_retval; 1463 }; */ *ap = v; 1464 int name = ap->a_name; 1465 register_t *retval = ap->a_retval; 1466 1467 switch (name) { 1468 case _PC_LINK_MAX: 1469 *retval = LINK_MAX; 1470 return 0; 1471 case _PC_NAME_MAX: 1472 *retval = NAME_MAX; 1473 return 0; 1474 case _PC_PATH_MAX: 1475 *retval = PATH_MAX; 1476 return 0; 1477 case _PC_PIPE_BUF: 1478 *retval = PIPE_BUF; 1479 return 0; 1480 case _PC_CHOWN_RESTRICTED: 1481 *retval = 1; 1482 return 0; 1483 case _PC_NO_TRUNC: 1484 *retval = 1; 1485 return 0; 1486 case _PC_SYNC_IO: 1487 *retval = 1; 1488 return 0; 1489 case _PC_FILESIZEBITS: 1490 *retval = 43; /* this one goes to 11 */ 1491 return 0; 1492 case _PC_SYMLINK_MAX: 1493 *retval = MAXPATHLEN; 1494 return 0; 1495 case _PC_2_SYMLINKS: 1496 *retval = 1; 1497 return 0; 1498 default: 1499 return EINVAL; 1500 } 1501 } 1502 1503 static int 1504 rump_vop_success(void *v) 1505 { 1506 1507 return 0; 1508 } 1509 1510 static int 1511 rump_vop_inactive(void *v) 1512 { 1513 struct vop_inactive_args /* { 1514 struct vnode *a_vp; 1515 bool *a_recycle; 1516 } */ *ap = v; 1517 struct vnode *vp = ap->a_vp; 1518 struct rumpfs_node *rn = vp->v_data; 1519 int error; 1520 1521 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) { 1522 if (rn->rn_readfd != -1) { 1523 rumpuser_close(rn->rn_readfd, &error); 1524 rn->rn_readfd = -1; 1525 } 1526 if (rn->rn_writefd != -1) { 1527 rumpuser_close(rn->rn_writefd, &error); 1528 rn->rn_writefd = -1; 1529 } 1530 } 1531 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false; 1532 1533 VOP_UNLOCK(vp); 1534 return 0; 1535 } 1536 1537 static int 1538 rump_vop_reclaim(void *v) 1539 { 1540 struct vop_reclaim_args /* { 1541 struct vnode *a_vp; 1542 } */ *ap = v; 1543 struct vnode *vp = ap->a_vp; 1544 struct rumpfs_node *rn = vp->v_data; 1545 1546 mutex_enter(&reclock); 1547 rn->rn_vp = NULL; 1548 mutex_exit(&reclock); 1549 genfs_node_destroy(vp); 1550 vp->v_data = NULL; 1551 1552 if (rn->rn_flags & RUMPNODE_CANRECLAIM) { 1553 if (vp->v_type == VLNK) 1554 PNBUF_PUT(rn->rn_linktarg); 1555 if (rn->rn_hostpath) 1556 free(rn->rn_hostpath, M_TEMP); 1557 kmem_free(rn, sizeof(*rn)); 1558 } 1559 1560 return 0; 1561 } 1562 1563 static int 1564 rump_vop_spec(void *v) 1565 { 1566 struct vop_generic_args *ap = v; 1567 int (**opvec)(void *); 1568 1569 switch (ap->a_desc->vdesc_offset) { 1570 case VOP_ACCESS_DESCOFFSET: 1571 case VOP_GETATTR_DESCOFFSET: 1572 case VOP_SETATTR_DESCOFFSET: 1573 case VOP_LOCK_DESCOFFSET: 1574 case VOP_UNLOCK_DESCOFFSET: 1575 case VOP_ISLOCKED_DESCOFFSET: 1576 case VOP_RECLAIM_DESCOFFSET: 1577 opvec = rump_vnodeop_p; 1578 break; 1579 default: 1580 opvec = spec_vnodeop_p; 1581 break; 1582 } 1583 1584 return VOCALL(opvec, ap->a_desc->vdesc_offset, v); 1585 } 1586 1587 static int 1588 rump_vop_advlock(void *v) 1589 { 1590 struct vop_advlock_args /* { 1591 const struct vnodeop_desc *a_desc; 1592 struct vnode *a_vp; 1593 void *a_id; 1594 int a_op; 1595 struct flock *a_fl; 1596 int a_flags; 1597 } */ *ap = v; 1598 struct vnode *vp = ap->a_vp; 1599 struct rumpfs_node *rn = vp->v_data; 1600 1601 return lf_advlock(ap, &rn->rn_lockf, vp->v_size); 1602 } 1603 1604 /* 1605 * Begin vfs-level stuff 1606 */ 1607 1608 VFS_PROTOS(rumpfs); 1609 struct vfsops rumpfs_vfsops = { 1610 .vfs_name = MOUNT_RUMPFS, 1611 .vfs_min_mount_data = 0, 1612 .vfs_mount = rumpfs_mount, 1613 .vfs_start = (void *)nullop, 1614 .vfs_unmount = rumpfs_unmount, 1615 .vfs_root = rumpfs_root, 1616 .vfs_quotactl = (void *)eopnotsupp, 1617 .vfs_statvfs = genfs_statvfs, 1618 .vfs_sync = (void *)nullop, 1619 .vfs_vget = rumpfs_vget, 1620 .vfs_fhtovp = (void *)eopnotsupp, 1621 .vfs_vptofh = (void *)eopnotsupp, 1622 .vfs_init = rumpfs_init, 1623 .vfs_reinit = NULL, 1624 .vfs_done = rumpfs_done, 1625 .vfs_mountroot = rumpfs_mountroot, 1626 .vfs_snapshot = (void *)eopnotsupp, 1627 .vfs_extattrctl = (void *)eopnotsupp, 1628 .vfs_suspendctl = (void *)eopnotsupp, 1629 .vfs_renamelock_enter = genfs_renamelock_enter, 1630 .vfs_renamelock_exit = genfs_renamelock_exit, 1631 .vfs_opv_descs = rump_opv_descs, 1632 /* vfs_refcount */ 1633 /* vfs_list */ 1634 }; 1635 1636 static int 1637 rumpfs_mountfs(struct mount *mp) 1638 { 1639 struct rumpfs_mount *rfsmp; 1640 struct rumpfs_node *rn; 1641 int error; 1642 1643 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP); 1644 1645 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false); 1646 rn->rn_parent = rn; 1647 if ((error = makevnode(mp, rn, &rfsmp->rfsmp_rvp)) != 0) 1648 return error; 1649 1650 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT; 1651 VOP_UNLOCK(rfsmp->rfsmp_rvp); 1652 1653 mp->mnt_data = rfsmp; 1654 mp->mnt_stat.f_namemax = MAXNAMLEN; 1655 mp->mnt_stat.f_iosize = 512; 1656 mp->mnt_flag |= MNT_LOCAL; 1657 mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO; 1658 mp->mnt_fs_bshift = DEV_BSHIFT; 1659 vfs_getnewfsid(mp); 1660 1661 return 0; 1662 } 1663 1664 int 1665 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen) 1666 { 1667 int error; 1668 1669 if (mp->mnt_flag & MNT_UPDATE) { 1670 return 0; 1671 } 1672 1673 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE, 1674 mp->mnt_op->vfs_name, mp, curlwp); 1675 if (error) 1676 return error; 1677 1678 return rumpfs_mountfs(mp); 1679 } 1680 1681 int 1682 rumpfs_unmount(struct mount *mp, int mntflags) 1683 { 1684 struct rumpfs_mount *rfsmp = mp->mnt_data; 1685 int flags = 0, error; 1686 1687 if (panicstr || mntflags & MNT_FORCE) 1688 flags |= FORCECLOSE; 1689 1690 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0) 1691 return error; 1692 vgone(rfsmp->rfsmp_rvp); /* XXX */ 1693 1694 kmem_free(rfsmp, sizeof(*rfsmp)); 1695 1696 return 0; 1697 } 1698 1699 int 1700 rumpfs_root(struct mount *mp, struct vnode **vpp) 1701 { 1702 struct rumpfs_mount *rfsmp = mp->mnt_data; 1703 1704 vref(rfsmp->rfsmp_rvp); 1705 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY); 1706 *vpp = rfsmp->rfsmp_rvp; 1707 return 0; 1708 } 1709 1710 int 1711 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1712 { 1713 1714 return EOPNOTSUPP; 1715 } 1716 1717 void 1718 rumpfs_init() 1719 { 1720 1721 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET); 1722 1723 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE); 1724 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE); 1725 } 1726 1727 void 1728 rumpfs_done() 1729 { 1730 1731 mutex_destroy(&reclock); 1732 mutex_destroy(&etfs_lock); 1733 } 1734 1735 int 1736 rumpfs_mountroot() 1737 { 1738 struct mount *mp; 1739 int error; 1740 1741 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) { 1742 vrele(rootvp); 1743 return error; 1744 } 1745 1746 if ((error = rumpfs_mountfs(mp)) != 0) 1747 panic("mounting rootfs failed: %d", error); 1748 1749 mutex_enter(&mountlist_lock); 1750 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1751 mutex_exit(&mountlist_lock); 1752 1753 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE, 1754 mp->mnt_op->vfs_name, mp, curlwp); 1755 if (error) 1756 panic("set_statvfs_info failed for rootfs: %d", error); 1757 1758 mp->mnt_flag &= ~MNT_RDONLY; 1759 vfs_unbusy(mp, false, NULL); 1760 1761 return 0; 1762 } 1763
Indexes created Tue Sep 30 20:09:53 GMT 2025