rumpfs.c revision 1.84 1 /* $NetBSD: rumpfs.c,v 1.84 2011/01/12 17:20:54 pooka Exp $ */
2
3 /*
4 * Copyright (c) 2009, 2010 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 1.84 2011/01/12 17:20:54 pooka 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
87 int (**fifo_vnodeop_p)(void *);
88 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = {
89 { &vop_default_desc, vn_default_error },
90 { NULL, NULL }
91 };
92 const struct vnodeopv_desc fifo_vnodeop_opv_desc =
93 { &fifo_vnodeop_p, fifo_vnodeop_entries };
94
95 int (**rump_vnodeop_p)(void *);
96 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = {
97 { &vop_default_desc, vn_default_error },
98 { &vop_lookup_desc, rump_vop_lookup },
99 { &vop_getattr_desc, rump_vop_getattr },
100 { &vop_setattr_desc, rump_vop_setattr },
101 { &vop_mkdir_desc, rump_vop_mkdir },
102 { &vop_rmdir_desc, rump_vop_rmdir },
103 { &vop_remove_desc, rump_vop_remove },
104 { &vop_mknod_desc, rump_vop_mknod },
105 { &vop_create_desc, rump_vop_create },
106 { &vop_symlink_desc, rump_vop_symlink },
107 { &vop_readlink_desc, rump_vop_readlink },
108 { &vop_access_desc, rump_vop_success },
109 { &vop_readdir_desc, rump_vop_readdir },
110 { &vop_read_desc, rump_vop_read },
111 { &vop_write_desc, rump_vop_write },
112 { &vop_open_desc, rump_vop_open },
113 { &vop_close_desc, genfs_nullop },
114 { &vop_seek_desc, genfs_seek },
115 { &vop_getpages_desc, genfs_getpages },
116 { &vop_putpages_desc, genfs_putpages },
117 { &vop_whiteout_desc, rump_vop_whiteout },
118 { &vop_fsync_desc, rump_vop_success },
119 { &vop_lock_desc, genfs_lock },
120 { &vop_unlock_desc, genfs_unlock },
121 { &vop_islocked_desc, genfs_islocked },
122 { &vop_inactive_desc, rump_vop_inactive },
123 { &vop_reclaim_desc, rump_vop_reclaim },
124 { &vop_link_desc, genfs_eopnotsupp },
125 { &vop_pathconf_desc, rump_vop_pathconf },
126 { &vop_bmap_desc, rump_vop_bmap },
127 { &vop_strategy_desc, rump_vop_strategy },
128 { &vop_advlock_desc, rump_vop_advlock },
129 { NULL, NULL }
130 };
131 const struct vnodeopv_desc rump_vnodeop_opv_desc =
132 { &rump_vnodeop_p, rump_vnodeop_entries };
133
134 int (**rump_specop_p)(void *);
135 const struct vnodeopv_entry_desc rump_specop_entries[] = {
136 { &vop_default_desc, rump_vop_spec },
137 { NULL, NULL }
138 };
139 const struct vnodeopv_desc rump_specop_opv_desc =
140 { &rump_specop_p, rump_specop_entries };
141
142 const struct vnodeopv_desc * const rump_opv_descs[] = {
143 &rump_vnodeop_opv_desc,
144 &rump_specop_opv_desc,
145 NULL
146 };
147
148 #define RUMPFS_WHITEOUT NULL
149 #define RDENT_ISWHITEOUT(rdp) (rdp->rd_node == RUMPFS_WHITEOUT)
150 struct rumpfs_dent {
151 char *rd_name;
152 int rd_namelen;
153 struct rumpfs_node *rd_node;
154
155 LIST_ENTRY(rumpfs_dent) rd_entries;
156 };
157
158 struct genfs_ops rumpfs_genfsops = {
159 .gop_size = genfs_size,
160 .gop_write = genfs_gop_write,
161
162 /* optional */
163 .gop_alloc = NULL,
164 .gop_markupdate = NULL,
165 };
166
167 struct rumpfs_node {
168 struct genfs_node rn_gn;
169 struct vattr rn_va;
170 struct vnode *rn_vp;
171 char *rn_hostpath;
172 int rn_flags;
173 struct lockf *rn_lockf;
174
175 union {
176 struct { /* VREG */
177 int readfd;
178 int writefd;
179 uint64_t offset;
180 } reg;
181 struct {
182 void *data;
183 size_t dlen;
184 } reg_noet;
185 struct { /* VDIR */
186 LIST_HEAD(, rumpfs_dent) dents;
187 struct rumpfs_node *parent;
188 int flags;
189 } dir;
190 struct {
191 char *target;
192 size_t len;
193 } link;
194 } rn_u;
195 };
196 #define rn_readfd rn_u.reg.readfd
197 #define rn_writefd rn_u.reg.writefd
198 #define rn_offset rn_u.reg.offset
199 #define rn_data rn_u.reg_noet.data
200 #define rn_dlen rn_u.reg_noet.dlen
201 #define rn_dir rn_u.dir.dents
202 #define rn_parent rn_u.dir.parent
203 #define rn_linktarg rn_u.link.target
204 #define rn_linklen rn_u.link.len
205
206 #define RUMPNODE_CANRECLAIM 0x01
207 #define RUMPNODE_DIR_ET 0x02
208 #define RUMPNODE_DIR_ETSUBS 0x04
209 #define RUMPNODE_ET_PHONE_HOST 0x10
210
211 struct rumpfs_mount {
212 struct vnode *rfsmp_rvp;
213 };
214
215 static struct rumpfs_node *makeprivate(enum vtype, dev_t, off_t, bool);
216
217 /*
218 * Extra Terrestrial stuff. We map a given key (pathname) to a file on
219 * the host FS. ET phones home only from the root node of rumpfs.
220 *
221 * When an etfs node is removed, a vnode potentially behind it is not
222 * immediately recycled.
223 */
224
225 struct etfs {
226 char et_key[MAXPATHLEN];
227 size_t et_keylen;
228 bool et_prefixkey;
229 bool et_removing;
230 devminor_t et_blkmin;
231
232 LIST_ENTRY(etfs) et_entries;
233
234 struct rumpfs_node *et_rn;
235 };
236 static kmutex_t etfs_lock;
237 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list);
238
239 static enum vtype
240 ettype_to_vtype(enum rump_etfs_type et)
241 {
242 enum vtype vt;
243
244 switch (et) {
245 case RUMP_ETFS_REG:
246 vt = VREG;
247 break;
248 case RUMP_ETFS_BLK:
249 vt = VBLK;
250 break;
251 case RUMP_ETFS_CHR:
252 vt = VCHR;
253 break;
254 case RUMP_ETFS_DIR:
255 vt = VDIR;
256 break;
257 case RUMP_ETFS_DIR_SUBDIRS:
258 vt = VDIR;
259 break;
260 default:
261 panic("invalid et type: %d", et);
262 }
263
264 return vt;
265 }
266
267 static enum vtype
268 hft_to_vtype(int hft)
269 {
270 enum vtype vt;
271
272 switch (hft) {
273 case RUMPUSER_FT_OTHER:
274 vt = VNON;
275 break;
276 case RUMPUSER_FT_DIR:
277 vt = VDIR;
278 break;
279 case RUMPUSER_FT_REG:
280 vt = VREG;
281 break;
282 case RUMPUSER_FT_BLK:
283 vt = VBLK;
284 break;
285 case RUMPUSER_FT_CHR:
286 vt = VCHR;
287 break;
288 default:
289 vt = VNON;
290 break;
291 }
292
293 return vt;
294 }
295
296 static bool
297 etfs_find(const char *key, struct etfs **etp, bool forceprefix)
298 {
299 struct etfs *et;
300 size_t keylen = strlen(key);
301
302 KASSERT(mutex_owned(&etfs_lock));
303
304 LIST_FOREACH(et, &etfs_list, et_entries) {
305 if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix)
306 && strncmp(key, et->et_key, et->et_keylen) == 0) {
307 if (etp)
308 *etp = et;
309 return true;
310 }
311 }
312
313 return false;
314 }
315
316 #define REGDIR(ftype) \
317 ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS)
318 static int
319 doregister(const char *key, const char *hostpath,
320 enum rump_etfs_type ftype, uint64_t begin, uint64_t size)
321 {
322 char buf[9];
323 struct etfs *et;
324 struct rumpfs_node *rn;
325 uint64_t fsize;
326 dev_t rdev = NODEV;
327 devminor_t dmin = -1;
328 int hft, error;
329
330 if (key[0] != '/') {
331 return EINVAL;
332 }
333 while (key[0] == '/') {
334 key++;
335 }
336
337 if (rumpuser_getfileinfo(hostpath, &fsize, &hft, &error))
338 return error;
339
340 /* etfs directory requires a directory on the host */
341 if (REGDIR(ftype)) {
342 if (hft != RUMPUSER_FT_DIR)
343 return ENOTDIR;
344 if (begin != 0)
345 return EISDIR;
346 if (size != RUMP_ETFS_SIZE_ENDOFF)
347 return EISDIR;
348 size = fsize;
349 } else {
350 if (begin > fsize)
351 return EINVAL;
352 if (size == RUMP_ETFS_SIZE_ENDOFF)
353 size = fsize - begin;
354 if (begin + size > fsize)
355 return EINVAL;
356 }
357
358 if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) {
359 error = rumpblk_register(hostpath, &dmin, begin, size);
360 if (error != 0) {
361 return error;
362 }
363 rdev = makedev(RUMPBLK_DEVMAJOR, dmin);
364 }
365
366 et = kmem_alloc(sizeof(*et), KM_SLEEP);
367 strcpy(et->et_key, key);
368 et->et_keylen = strlen(et->et_key);
369 et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), rdev, size, true);
370 et->et_removing = false;
371 et->et_blkmin = dmin;
372
373 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST;
374
375 if (ftype == RUMP_ETFS_REG || REGDIR(ftype) || et->et_blkmin != -1) {
376 size_t len = strlen(hostpath)+1;
377
378 rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
379 memcpy(rn->rn_hostpath, hostpath, len);
380 rn->rn_offset = begin;
381 }
382
383 if (REGDIR(ftype)) {
384 rn->rn_flags |= RUMPNODE_DIR_ET;
385 et->et_prefixkey = true;
386 } else {
387 et->et_prefixkey = false;
388 }
389
390 if (ftype == RUMP_ETFS_DIR_SUBDIRS)
391 rn->rn_flags |= RUMPNODE_DIR_ETSUBS;
392
393 mutex_enter(&etfs_lock);
394 if (etfs_find(key, NULL, REGDIR(ftype))) {
395 mutex_exit(&etfs_lock);
396 if (et->et_blkmin != -1)
397 rumpblk_deregister(hostpath);
398 if (et->et_rn->rn_hostpath != NULL)
399 free(et->et_rn->rn_hostpath, M_TEMP);
400 kmem_free(et->et_rn, sizeof(*et->et_rn));
401 kmem_free(et, sizeof(*et));
402 return EEXIST;
403 }
404 LIST_INSERT_HEAD(&etfs_list, et, et_entries);
405 mutex_exit(&etfs_lock);
406
407 if (ftype == RUMP_ETFS_BLK) {
408 format_bytes(buf, sizeof(buf), size);
409 aprint_verbose("/%s: hostpath %s (%s)\n", key, hostpath, buf);
410 }
411
412 return 0;
413 }
414 #undef REGDIR
415
416 int
417 rump_etfs_register(const char *key, const char *hostpath,
418 enum rump_etfs_type ftype)
419 {
420
421 return doregister(key, hostpath, ftype, 0, RUMP_ETFS_SIZE_ENDOFF);
422 }
423
424 int
425 rump_etfs_register_withsize(const char *key, const char *hostpath,
426 enum rump_etfs_type ftype, uint64_t begin, uint64_t size)
427 {
428
429 return doregister(key, hostpath, ftype, begin, size);
430 }
431
432 /* remove etfs mapping. caller's responsibility to make sure it's not in use */
433 int
434 rump_etfs_remove(const char *key)
435 {
436 struct etfs *et;
437 size_t keylen;
438 int rv;
439
440 if (key[0] != '/') {
441 return EINVAL;
442 }
443 while (key[0] == '/') {
444 key++;
445 }
446
447 keylen = strlen(key);
448
449 mutex_enter(&etfs_lock);
450 LIST_FOREACH(et, &etfs_list, et_entries) {
451 if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) {
452 if (et->et_removing)
453 et = NULL;
454 else
455 et->et_removing = true;
456 break;
457 }
458 }
459 mutex_exit(&etfs_lock);
460 if (!et)
461 return ENOENT;
462
463 /*
464 * ok, we know what we want to remove and have signalled there
465 * actually are men at work. first, unregister from rumpblk
466 */
467 if (et->et_blkmin != -1) {
468 rv = rumpblk_deregister(et->et_rn->rn_hostpath);
469 } else {
470 rv = 0;
471 }
472 KASSERT(rv == 0);
473
474 /* then do the actual removal */
475 mutex_enter(&etfs_lock);
476 LIST_REMOVE(et, et_entries);
477 mutex_exit(&etfs_lock);
478
479 /* node is unreachable, safe to nuke all device copies */
480 if (et->et_blkmin != -1)
481 vdevgone(RUMPBLK_DEVMAJOR, et->et_blkmin, et->et_blkmin, VBLK);
482
483 if (et->et_rn->rn_hostpath != NULL)
484 free(et->et_rn->rn_hostpath, M_TEMP);
485 kmem_free(et->et_rn, sizeof(*et->et_rn));
486 kmem_free(et, sizeof(*et));
487
488 return 0;
489 }
490
491 /*
492 * rumpfs
493 */
494
495 #define INO_WHITEOUT 1
496 static int lastino = 2;
497 static kmutex_t reclock;
498
499 static struct rumpfs_node *
500 makeprivate(enum vtype vt, dev_t rdev, off_t size, bool et)
501 {
502 struct rumpfs_node *rn;
503 struct vattr *va;
504 struct timespec ts;
505
506 rn = kmem_zalloc(sizeof(*rn), KM_SLEEP);
507
508 switch (vt) {
509 case VDIR:
510 LIST_INIT(&rn->rn_dir);
511 break;
512 case VREG:
513 if (et) {
514 rn->rn_readfd = -1;
515 rn->rn_writefd = -1;
516 }
517 break;
518 default:
519 break;
520 }
521
522 nanotime(&ts);
523
524 va = &rn->rn_va;
525 va->va_type = vt;
526 va->va_mode = 0755;
527 if (vt == VDIR)
528 va->va_nlink = 2;
529 else
530 va->va_nlink = 1;
531 va->va_uid = 0;
532 va->va_gid = 0;
533 va->va_fsid =
534 va->va_fileid = atomic_inc_uint_nv(&lastino);
535 va->va_size = size;
536 va->va_blocksize = 512;
537 va->va_atime = ts;
538 va->va_mtime = ts;
539 va->va_ctime = ts;
540 va->va_birthtime = ts;
541 va->va_gen = 0;
542 va->va_flags = 0;
543 va->va_rdev = rdev;
544 va->va_bytes = 512;
545 va->va_filerev = 0;
546 va->va_vaflags = 0;
547
548 return rn;
549 }
550
551 static int
552 makevnode(struct mount *mp, struct rumpfs_node *rn, struct vnode **vpp)
553 {
554 struct vnode *vp;
555 int (**vpops)(void *);
556 struct vattr *va = &rn->rn_va;
557 int rv;
558
559 KASSERT(!mutex_owned(&reclock));
560
561 if (va->va_type == VCHR || va->va_type == VBLK) {
562 vpops = rump_specop_p;
563 } else {
564 vpops = rump_vnodeop_p;
565 }
566
567 rv = getnewvnode(VT_RUMP, mp, vpops, &vp);
568 if (rv)
569 return rv;
570
571 vp->v_size = vp->v_writesize = va->va_size;
572 vp->v_type = va->va_type;
573
574 if (vpops == rump_specop_p) {
575 spec_node_init(vp, va->va_rdev);
576 }
577 vp->v_data = rn;
578
579 genfs_node_init(vp, &rumpfs_genfsops);
580 vn_lock(vp, LK_RETRY | LK_EXCLUSIVE);
581 mutex_enter(&reclock);
582 rn->rn_vp = vp;
583 mutex_exit(&reclock);
584
585 *vpp = vp;
586
587 return 0;
588 }
589
590
591 static void
592 makedir(struct rumpfs_node *rnd,
593 struct componentname *cnp, struct rumpfs_node *rn)
594 {
595 struct rumpfs_dent *rdent;
596
597 rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP);
598 rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP);
599 rdent->rd_node = rn;
600 strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1);
601 rdent->rd_namelen = strlen(rdent->rd_name);
602
603 LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries);
604 }
605
606 static void
607 freedir(struct rumpfs_node *rnd, struct componentname *cnp)
608 {
609 struct rumpfs_dent *rd = NULL;
610
611 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) {
612 if (rd->rd_namelen == cnp->cn_namelen &&
613 strncmp(rd->rd_name, cnp->cn_nameptr,
614 cnp->cn_namelen) == 0)
615 break;
616 }
617 if (rd == NULL)
618 panic("could not find directory entry: %s", cnp->cn_nameptr);
619
620 LIST_REMOVE(rd, rd_entries);
621 kmem_free(rd->rd_name, rd->rd_namelen+1);
622 kmem_free(rd, sizeof(*rd));
623 }
624
625 /*
626 * Simple lookup for rump file systems.
627 *
628 * uhm, this is twisted. C F C C, hope of C C F C looming
629 */
630 static int
631 rump_vop_lookup(void *v)
632 {
633 struct vop_lookup_args /* {
634 struct vnode *a_dvp;
635 struct vnode **a_vpp;
636 struct componentname *a_cnp;
637 }; */ *ap = v;
638 struct componentname *cnp = ap->a_cnp;
639 struct vnode *dvp = ap->a_dvp;
640 struct vnode **vpp = ap->a_vpp;
641 struct vnode *vp;
642 struct rumpfs_node *rnd = dvp->v_data, *rn;
643 struct rumpfs_dent *rd = NULL;
644 struct etfs *et;
645 bool dotdot = (cnp->cn_flags & ISDOTDOT) != 0;
646 int rv = 0;
647
648 /* check for dot, return directly if the case */
649 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
650 vref(dvp);
651 *vpp = dvp;
652 return 0;
653 }
654
655 /* we don't do rename */
656 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME)))
657 return EOPNOTSUPP;
658
659 /* check for etfs */
660 if (dvp == rootvnode && cnp->cn_nameiop == LOOKUP) {
661 bool found;
662 mutex_enter(&etfs_lock);
663 found = etfs_find(cnp->cn_nameptr, &et, false);
664 mutex_exit(&etfs_lock);
665
666 if (found) {
667 rn = et->et_rn;
668 cnp->cn_consume += et->et_keylen - cnp->cn_namelen;
669 if (rn->rn_va.va_type != VDIR)
670 cnp->cn_flags &= ~REQUIREDIR;
671 goto getvnode;
672 }
673 }
674
675 if (rnd->rn_flags & RUMPNODE_DIR_ET) {
676 uint64_t fsize;
677 char *newpath;
678 size_t newpathlen;
679 int hft, error;
680
681 if (dotdot)
682 return EOPNOTSUPP;
683
684 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1;
685 newpath = malloc(newpathlen, M_TEMP, M_WAITOK);
686
687 strlcpy(newpath, rnd->rn_hostpath, newpathlen);
688 strlcat(newpath, "/", newpathlen);
689 strlcat(newpath, cnp->cn_nameptr, newpathlen);
690
691 if (rumpuser_getfileinfo(newpath, &fsize, &hft, &error)) {
692 free(newpath, M_TEMP);
693 return error;
694 }
695
696 /* allow only dirs and regular files */
697 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) {
698 free(newpath, M_TEMP);
699 return ENOENT;
700 }
701
702 rn = makeprivate(hft_to_vtype(hft), NODEV, fsize, true);
703 rn->rn_flags |= RUMPNODE_CANRECLAIM;
704 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) {
705 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS;
706 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST;
707 }
708 rn->rn_hostpath = newpath;
709
710 goto getvnode;
711 } else {
712 if (dotdot) {
713 rn = rnd->rn_parent;
714 goto getvnode;
715 } else {
716 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) {
717 if (rd->rd_namelen == cnp->cn_namelen &&
718 strncmp(rd->rd_name, cnp->cn_nameptr,
719 cnp->cn_namelen) == 0)
720 break;
721 }
722 }
723 }
724
725 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE))
726 return ENOENT;
727
728 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) {
729 return EJUSTRETURN;
730 }
731
732 rn = rd->rd_node;
733
734 getvnode:
735 KASSERT(rn);
736 if (dotdot)
737 VOP_UNLOCK(dvp);
738 mutex_enter(&reclock);
739 if ((vp = rn->rn_vp)) {
740 mutex_enter(&vp->v_interlock);
741 mutex_exit(&reclock);
742 if (vget(vp, LK_EXCLUSIVE)) {
743 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
744 goto getvnode;
745 }
746 *vpp = vp;
747 } else {
748 mutex_exit(&reclock);
749 rv = makevnode(dvp->v_mount, rn, vpp);
750 }
751 if (dotdot)
752 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
753
754 return rv;
755 }
756
757 static int
758 rump_vop_getattr(void *v)
759 {
760 struct vop_getattr_args /* {
761 struct vnode *a_vp;
762 struct vattr *a_vap;
763 kauth_cred_t a_cred;
764 } */ *ap = v;
765 struct vnode *vp = ap->a_vp;
766 struct rumpfs_node *rn = vp->v_data;
767 struct vattr *vap = ap->a_vap;
768
769 memcpy(vap, &rn->rn_va, sizeof(struct vattr));
770 vap->va_size = vp->v_size;
771 return 0;
772 }
773
774 static int
775 rump_vop_setattr(void *v)
776 {
777 struct vop_getattr_args /* {
778 struct vnode *a_vp;
779 struct vattr *a_vap;
780 kauth_cred_t a_cred;
781 } */ *ap = v;
782 struct vnode *vp = ap->a_vp;
783 struct vattr *vap = ap->a_vap;
784 struct rumpfs_node *rn = vp->v_data;
785
786 #define SETIFVAL(a,t) if (vap->a != (t)VNOVAL) rn->rn_va.a = vap->a
787 SETIFVAL(va_mode, mode_t);
788 SETIFVAL(va_uid, uid_t);
789 SETIFVAL(va_gid, gid_t);
790 SETIFVAL(va_atime.tv_sec, time_t);
791 SETIFVAL(va_ctime.tv_sec, time_t);
792 SETIFVAL(va_mtime.tv_sec, time_t);
793 SETIFVAL(va_birthtime.tv_sec, time_t);
794 SETIFVAL(va_atime.tv_nsec, long);
795 SETIFVAL(va_ctime.tv_nsec, long);
796 SETIFVAL(va_mtime.tv_nsec, long);
797 SETIFVAL(va_birthtime.tv_nsec, long);
798 SETIFVAL(va_flags, u_long);
799 #undef SETIFVAL
800
801 if (vp->v_type == VREG && vap->va_size != VSIZENOTSET)
802 uvm_vnp_setsize(vp, vap->va_size);
803 return 0;
804 }
805
806 static int
807 rump_vop_mkdir(void *v)
808 {
809 struct vop_mkdir_args /* {
810 struct vnode *a_dvp;
811 struct vnode **a_vpp;
812 struct componentname *a_cnp;
813 struct vattr *a_vap;
814 }; */ *ap = v;
815 struct vnode *dvp = ap->a_dvp;
816 struct vnode **vpp = ap->a_vpp;
817 struct componentname *cnp = ap->a_cnp;
818 struct rumpfs_node *rnd = dvp->v_data, *rn;
819 int rv = 0;
820
821 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false);
822 rn->rn_parent = rnd;
823 rv = makevnode(dvp->v_mount, rn, vpp);
824 if (rv)
825 goto out;
826
827 makedir(rnd, cnp, rn);
828
829 out:
830 vput(dvp);
831 return rv;
832 }
833
834 static int
835 rump_vop_rmdir(void *v)
836 {
837 struct vop_rmdir_args /* {
838 struct vnode *a_dvp;
839 struct vnode *a_vp;
840 struct componentname *a_cnp;
841 }; */ *ap = v;
842 struct vnode *dvp = ap->a_dvp;
843 struct vnode *vp = ap->a_vp;
844 struct componentname *cnp = ap->a_cnp;
845 struct rumpfs_node *rnd = dvp->v_data;
846 struct rumpfs_node *rn = vp->v_data;
847 int rv = 0;
848
849 if (!LIST_EMPTY(&rn->rn_dir)) {
850 rv = ENOTEMPTY;
851 goto out;
852 }
853
854 freedir(rnd, cnp);
855 rn->rn_flags |= RUMPNODE_CANRECLAIM;
856
857 out:
858 vput(dvp);
859 vput(vp);
860
861 return rv;
862 }
863
864 static int
865 rump_vop_remove(void *v)
866 {
867 struct vop_rmdir_args /* {
868 struct vnode *a_dvp;
869 struct vnode *a_vp;
870 struct componentname *a_cnp;
871 }; */ *ap = v;
872 struct vnode *dvp = ap->a_dvp;
873 struct vnode *vp = ap->a_vp;
874 struct componentname *cnp = ap->a_cnp;
875 struct rumpfs_node *rnd = dvp->v_data;
876 struct rumpfs_node *rn = vp->v_data;
877 int rv = 0;
878
879 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)
880 return EOPNOTSUPP;
881
882 if (vp->v_type == VREG) {
883 rump_hyperfree(rn->rn_data, rn->rn_dlen);
884 }
885
886 freedir(rnd, cnp);
887 rn->rn_flags |= RUMPNODE_CANRECLAIM;
888
889 vput(dvp);
890 vput(vp);
891
892 return rv;
893 }
894
895 static int
896 rump_vop_mknod(void *v)
897 {
898 struct vop_mknod_args /* {
899 struct vnode *a_dvp;
900 struct vnode **a_vpp;
901 struct componentname *a_cnp;
902 struct vattr *a_vap;
903 }; */ *ap = v;
904 struct vnode *dvp = ap->a_dvp;
905 struct vnode **vpp = ap->a_vpp;
906 struct componentname *cnp = ap->a_cnp;
907 struct vattr *va = ap->a_vap;
908 struct rumpfs_node *rnd = dvp->v_data, *rn;
909 int rv;
910
911 rn = makeprivate(va->va_type, va->va_rdev, DEV_BSIZE, false);
912 rv = makevnode(dvp->v_mount, rn, vpp);
913 if (rv)
914 goto out;
915
916 makedir(rnd, cnp, rn);
917
918 out:
919 vput(dvp);
920 return rv;
921 }
922
923 static int
924 rump_vop_create(void *v)
925 {
926 struct vop_create_args /* {
927 struct vnode *a_dvp;
928 struct vnode **a_vpp;
929 struct componentname *a_cnp;
930 struct vattr *a_vap;
931 }; */ *ap = v;
932 struct vnode *dvp = ap->a_dvp;
933 struct vnode **vpp = ap->a_vpp;
934 struct componentname *cnp = ap->a_cnp;
935 struct vattr *va = ap->a_vap;
936 struct rumpfs_node *rnd = dvp->v_data, *rn;
937 off_t newsize;
938 int rv;
939
940 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0;
941 rn = makeprivate(va->va_type, NODEV, newsize, false);
942 rv = makevnode(dvp->v_mount, rn, vpp);
943 if (rv)
944 goto out;
945
946 makedir(rnd, cnp, rn);
947
948 out:
949 vput(dvp);
950 return rv;
951 }
952
953 static int
954 rump_vop_symlink(void *v)
955 {
956 struct vop_symlink_args /* {
957 struct vnode *a_dvp;
958 struct vnode **a_vpp;
959 struct componentname *a_cnp;
960 struct vattr *a_vap;
961 char *a_target;
962 }; */ *ap = v;
963 struct vnode *dvp = ap->a_dvp;
964 struct vnode **vpp = ap->a_vpp;
965 struct componentname *cnp = ap->a_cnp;
966 struct rumpfs_node *rnd = dvp->v_data, *rn;
967 const char *target = ap->a_target;
968 size_t linklen;
969 int rv;
970
971 linklen = strlen(target);
972 KASSERT(linklen < MAXPATHLEN);
973 rn = makeprivate(VLNK, NODEV, linklen, false);
974 rv = makevnode(dvp->v_mount, rn, vpp);
975 if (rv)
976 goto out;
977
978 makedir(rnd, cnp, rn);
979
980 KASSERT(linklen < MAXPATHLEN);
981 rn->rn_linktarg = PNBUF_GET();
982 rn->rn_linklen = linklen;
983 strcpy(rn->rn_linktarg, target);
984
985 out:
986 vput(dvp);
987 return rv;
988 }
989
990 static int
991 rump_vop_readlink(void *v)
992 {
993 struct vop_readlink_args /* {
994 struct vnode *a_vp;
995 struct uio *a_uio;
996 kauth_cred_t a_cred;
997 }; */ *ap = v;
998 struct vnode *vp = ap->a_vp;
999 struct rumpfs_node *rn = vp->v_data;
1000 struct uio *uio = ap->a_uio;
1001
1002 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio);
1003 }
1004
1005 static int
1006 rump_vop_whiteout(void *v)
1007 {
1008 struct vop_whiteout_args /* {
1009 struct vnode *a_dvp;
1010 struct componentname *a_cnp;
1011 int a_flags;
1012 } */ *ap = v;
1013 struct vnode *dvp = ap->a_dvp;
1014 struct rumpfs_node *rnd = dvp->v_data;
1015 struct componentname *cnp = ap->a_cnp;
1016 int flags = ap->a_flags;
1017
1018 switch (flags) {
1019 case LOOKUP:
1020 break;
1021 case CREATE:
1022 makedir(rnd, cnp, RUMPFS_WHITEOUT);
1023 break;
1024 case DELETE:
1025 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */
1026 freedir(rnd, cnp);
1027 break;
1028 default:
1029 panic("unknown whiteout op %d", flags);
1030 }
1031
1032 return 0;
1033 }
1034
1035 static int
1036 rump_vop_open(void *v)
1037 {
1038 struct vop_open_args /* {
1039 struct vnode *a_vp;
1040 int a_mode;
1041 kauth_cred_t a_cred;
1042 } */ *ap = v;
1043 struct vnode *vp = ap->a_vp;
1044 struct rumpfs_node *rn = vp->v_data;
1045 int mode = ap->a_mode;
1046 int error = EINVAL;
1047
1048 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0)
1049 return 0;
1050
1051 if (mode & FREAD) {
1052 if (rn->rn_readfd != -1)
1053 return 0;
1054 rn->rn_readfd = rumpuser_open(rn->rn_hostpath,
1055 O_RDONLY, &error);
1056 }
1057
1058 if (mode & FWRITE) {
1059 if (rn->rn_writefd != -1)
1060 return 0;
1061 rn->rn_writefd = rumpuser_open(rn->rn_hostpath,
1062 O_WRONLY, &error);
1063 }
1064
1065 return error;
1066 }
1067
1068 /* simple readdir. event omits dotstuff and periods */
1069 static int
1070 rump_vop_readdir(void *v)
1071 {
1072 struct vop_readdir_args /* {
1073 struct vnode *a_vp;
1074 struct uio *a_uio;
1075 kauth_cred_t a_cred;
1076 int *a_eofflag;
1077 off_t **a_cookies;
1078 int *a_ncookies;
1079 } */ *ap = v;
1080 struct vnode *vp = ap->a_vp;
1081 struct uio *uio = ap->a_uio;
1082 struct rumpfs_node *rnd = vp->v_data;
1083 struct rumpfs_dent *rdent;
1084 unsigned i;
1085 int rv = 0;
1086
1087 /* seek to current entry */
1088 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir);
1089 (i < uio->uio_offset) && rdent;
1090 i++, rdent = LIST_NEXT(rdent, rd_entries))
1091 continue;
1092 if (!rdent)
1093 goto out;
1094
1095 /* copy entries */
1096 for (; rdent && uio->uio_resid > 0;
1097 rdent = LIST_NEXT(rdent, rd_entries), i++) {
1098 struct dirent dent;
1099
1100 strlcpy(dent.d_name, rdent->rd_name, sizeof(dent.d_name));
1101 dent.d_namlen = strlen(dent.d_name);
1102 dent.d_reclen = _DIRENT_RECLEN(&dent, dent.d_namlen);
1103
1104 if (__predict_false(RDENT_ISWHITEOUT(rdent))) {
1105 dent.d_fileno = INO_WHITEOUT;
1106 dent.d_type = DT_WHT;
1107 } else {
1108 dent.d_fileno = rdent->rd_node->rn_va.va_fileid;
1109 dent.d_type = vtype2dt(rdent->rd_node->rn_va.va_type);
1110 }
1111
1112 if (uio->uio_resid < dent.d_reclen) {
1113 i--;
1114 break;
1115 }
1116
1117 rv = uiomove(&dent, dent.d_reclen, uio);
1118 if (rv) {
1119 i--;
1120 break;
1121 }
1122 }
1123
1124 out:
1125 if (ap->a_cookies) {
1126 *ap->a_ncookies = 0;
1127 *ap->a_cookies = NULL;
1128 }
1129 if (rdent)
1130 *ap->a_eofflag = 0;
1131 else
1132 *ap->a_eofflag = 1;
1133 uio->uio_offset = i;
1134
1135 return rv;
1136 }
1137
1138 static int
1139 etread(struct rumpfs_node *rn, struct uio *uio)
1140 {
1141 uint8_t *buf;
1142 size_t bufsize;
1143 ssize_t n;
1144 int error = 0;
1145
1146 bufsize = uio->uio_resid;
1147 buf = kmem_alloc(bufsize, KM_SLEEP);
1148 if ((n = rumpuser_pread(rn->rn_readfd, buf, bufsize,
1149 uio->uio_offset + rn->rn_offset, &error)) == -1)
1150 goto out;
1151 KASSERT(n <= bufsize);
1152 error = uiomove(buf, n, uio);
1153
1154 out:
1155 kmem_free(buf, bufsize);
1156 return error;
1157
1158 }
1159
1160 static int
1161 rump_vop_read(void *v)
1162 {
1163 struct vop_read_args /* {
1164 struct vnode *a_vp;
1165 struct uio *a_uio;
1166 int ioflags a_ioflag;
1167 kauth_cred_t a_cred;
1168 }; */ *ap = v;
1169 struct vnode *vp = ap->a_vp;
1170 struct rumpfs_node *rn = vp->v_data;
1171 struct uio *uio = ap->a_uio;
1172 const int advice = IO_ADV_DECODE(ap->a_ioflag);
1173 off_t chunk;
1174 int error = 0;
1175
1176 /* et op? */
1177 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)
1178 return etread(rn, uio);
1179
1180 /* otherwise, it's off to ubc with us */
1181 while (uio->uio_resid > 0) {
1182 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset);
1183 if (chunk == 0)
1184 break;
1185 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice,
1186 UBC_READ | UBC_PARTIALOK | UBC_WANT_UNMAP(vp)?UBC_UNMAP:0);
1187 if (error)
1188 break;
1189 }
1190
1191 return error;
1192 }
1193
1194 static int
1195 etwrite(struct rumpfs_node *rn, struct uio *uio)
1196 {
1197 uint8_t *buf;
1198 size_t bufsize;
1199 ssize_t n;
1200 int error = 0;
1201
1202 bufsize = uio->uio_resid;
1203 buf = kmem_alloc(bufsize, KM_SLEEP);
1204 error = uiomove(buf, bufsize, uio);
1205 if (error)
1206 goto out;
1207 KASSERT(uio->uio_resid == 0);
1208 n = rumpuser_pwrite(rn->rn_writefd, buf, bufsize,
1209 (uio->uio_offset-bufsize) + rn->rn_offset, &error);
1210 if (n >= 0) {
1211 KASSERT(n <= bufsize);
1212 uio->uio_resid = bufsize - n;
1213 }
1214
1215 out:
1216 kmem_free(buf, bufsize);
1217 return error;
1218 }
1219
1220 static int
1221 rump_vop_write(void *v)
1222 {
1223 struct vop_read_args /* {
1224 struct vnode *a_vp;
1225 struct uio *a_uio;
1226 int ioflags a_ioflag;
1227 kauth_cred_t a_cred;
1228 }; */ *ap = v;
1229 struct vnode *vp = ap->a_vp;
1230 struct rumpfs_node *rn = vp->v_data;
1231 struct uio *uio = ap->a_uio;
1232 const int advice = IO_ADV_DECODE(ap->a_ioflag);
1233 void *olddata;
1234 size_t oldlen, newlen;
1235 off_t chunk;
1236 int error = 0;
1237 bool allocd = false;
1238
1239 if (ap->a_ioflag & IO_APPEND)
1240 uio->uio_offset = vp->v_size;
1241
1242 /* consult et? */
1243 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)
1244 return etwrite(rn, uio);
1245
1246 /*
1247 * Otherwise, it's a case of ubcmove.
1248 */
1249
1250 /*
1251 * First, make sure we have enough storage.
1252 *
1253 * No, you don't need to tell me it's not very efficient.
1254 * No, it doesn't really support sparse files, just fakes it.
1255 */
1256 newlen = uio->uio_offset + uio->uio_resid;
1257 oldlen = 0; /* XXXgcc */
1258 olddata = NULL;
1259 if (rn->rn_dlen < newlen) {
1260 oldlen = rn->rn_dlen;
1261 olddata = rn->rn_data;
1262
1263 rn->rn_data = rump_hypermalloc(newlen, 0, true, "rumpfs");
1264 rn->rn_dlen = newlen;
1265 memset(rn->rn_data, 0, newlen);
1266 memcpy(rn->rn_data, olddata, oldlen);
1267 allocd = true;
1268 uvm_vnp_setsize(vp, newlen);
1269 }
1270
1271 /* ok, we have enough stooorage. write */
1272 while (uio->uio_resid > 0) {
1273 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset);
1274 if (chunk == 0)
1275 break;
1276 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice,
1277 UBC_WRITE | UBC_PARTIALOK | UBC_WANT_UNMAP(vp)?UBC_UNMAP:0);
1278 if (error)
1279 break;
1280 }
1281
1282 if (allocd) {
1283 if (error) {
1284 rump_hyperfree(rn->rn_data, newlen);
1285 rn->rn_data = olddata;
1286 rn->rn_dlen = oldlen;
1287 uvm_vnp_setsize(vp, oldlen);
1288 } else {
1289 rump_hyperfree(olddata, oldlen);
1290 }
1291 }
1292
1293 return error;
1294 }
1295
1296 static int
1297 rump_vop_bmap(void *v)
1298 {
1299 struct vop_bmap_args /* {
1300 struct vnode *a_vp;
1301 daddr_t a_bn;
1302 struct vnode **a_vpp;
1303 daddr_t *a_bnp;
1304 int *a_runp;
1305 } */ *ap = v;
1306
1307 /* 1:1 mapping */
1308 if (ap->a_vpp)
1309 *ap->a_vpp = ap->a_vp;
1310 if (ap->a_bnp)
1311 *ap->a_bnp = ap->a_bn;
1312 if (ap->a_runp)
1313 *ap->a_runp = 16;
1314
1315 return 0;
1316 }
1317
1318 static int
1319 rump_vop_strategy(void *v)
1320 {
1321 struct vop_strategy_args /* {
1322 struct vnode *a_vp;
1323 struct buf *a_bp;
1324 } */ *ap = v;
1325 struct vnode *vp = ap->a_vp;
1326 struct rumpfs_node *rn = vp->v_data;
1327 struct buf *bp = ap->a_bp;
1328 off_t copylen, copyoff;
1329 int error;
1330
1331 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) {
1332 error = EINVAL;
1333 goto out;
1334 }
1335
1336 copyoff = bp->b_blkno << DEV_BSHIFT;
1337 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount);
1338 if (BUF_ISWRITE(bp)) {
1339 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen);
1340 } else {
1341 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen);
1342 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen);
1343 }
1344 bp->b_resid = 0;
1345 error = 0;
1346
1347 out:
1348 bp->b_error = error;
1349 biodone(bp);
1350 return 0;
1351 }
1352
1353 static int
1354 rump_vop_pathconf(void *v)
1355 {
1356 struct vop_pathconf_args /* {
1357 struct vnode *a_vp;
1358 int a_name;
1359 register_t *a_retval;
1360 }; */ *ap = v;
1361 int name = ap->a_name;
1362 register_t *retval = ap->a_retval;
1363
1364 switch (name) {
1365 case _PC_LINK_MAX:
1366 *retval = LINK_MAX;
1367 return 0;
1368 case _PC_NAME_MAX:
1369 *retval = NAME_MAX;
1370 return 0;
1371 case _PC_PATH_MAX:
1372 *retval = PATH_MAX;
1373 return 0;
1374 case _PC_PIPE_BUF:
1375 *retval = PIPE_BUF;
1376 return 0;
1377 case _PC_CHOWN_RESTRICTED:
1378 *retval = 1;
1379 return 0;
1380 case _PC_NO_TRUNC:
1381 *retval = 1;
1382 return 0;
1383 case _PC_SYNC_IO:
1384 *retval = 1;
1385 return 0;
1386 case _PC_FILESIZEBITS:
1387 *retval = 43; /* this one goes to 11 */
1388 return 0;
1389 case _PC_SYMLINK_MAX:
1390 *retval = MAXPATHLEN;
1391 return 0;
1392 case _PC_2_SYMLINKS:
1393 *retval = 1;
1394 return 0;
1395 default:
1396 return EINVAL;
1397 }
1398 }
1399
1400 static int
1401 rump_vop_success(void *v)
1402 {
1403
1404 return 0;
1405 }
1406
1407 static int
1408 rump_vop_inactive(void *v)
1409 {
1410 struct vop_inactive_args /* {
1411 struct vnode *a_vp;
1412 bool *a_recycle;
1413 } */ *ap = v;
1414 struct vnode *vp = ap->a_vp;
1415 struct rumpfs_node *rn = vp->v_data;
1416 int error;
1417
1418 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) {
1419 if (rn->rn_readfd != -1) {
1420 rumpuser_close(rn->rn_readfd, &error);
1421 rn->rn_readfd = -1;
1422 }
1423 if (rn->rn_writefd != -1) {
1424 rumpuser_close(rn->rn_writefd, &error);
1425 rn->rn_writefd = -1;
1426 }
1427 }
1428 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false;
1429
1430 VOP_UNLOCK(vp);
1431 return 0;
1432 }
1433
1434 static int
1435 rump_vop_reclaim(void *v)
1436 {
1437 struct vop_reclaim_args /* {
1438 struct vnode *a_vp;
1439 } */ *ap = v;
1440 struct vnode *vp = ap->a_vp;
1441 struct rumpfs_node *rn = vp->v_data;
1442
1443 mutex_enter(&reclock);
1444 rn->rn_vp = NULL;
1445 mutex_exit(&reclock);
1446 genfs_node_destroy(vp);
1447 vp->v_data = NULL;
1448
1449 if (rn->rn_flags & RUMPNODE_CANRECLAIM) {
1450 if (vp->v_type == VLNK)
1451 PNBUF_PUT(rn->rn_linktarg);
1452 if (rn->rn_hostpath)
1453 free(rn->rn_hostpath, M_TEMP);
1454 kmem_free(rn, sizeof(*rn));
1455 }
1456
1457 return 0;
1458 }
1459
1460 static int
1461 rump_vop_spec(void *v)
1462 {
1463 struct vop_generic_args *ap = v;
1464 int (**opvec)(void *);
1465
1466 switch (ap->a_desc->vdesc_offset) {
1467 case VOP_ACCESS_DESCOFFSET:
1468 case VOP_GETATTR_DESCOFFSET:
1469 case VOP_SETATTR_DESCOFFSET:
1470 case VOP_LOCK_DESCOFFSET:
1471 case VOP_UNLOCK_DESCOFFSET:
1472 case VOP_ISLOCKED_DESCOFFSET:
1473 case VOP_RECLAIM_DESCOFFSET:
1474 opvec = rump_vnodeop_p;
1475 break;
1476 default:
1477 opvec = spec_vnodeop_p;
1478 break;
1479 }
1480
1481 return VOCALL(opvec, ap->a_desc->vdesc_offset, v);
1482 }
1483
1484 static int
1485 rump_vop_advlock(void *v)
1486 {
1487 struct vop_advlock_args /* {
1488 const struct vnodeop_desc *a_desc;
1489 struct vnode *a_vp;
1490 void *a_id;
1491 int a_op;
1492 struct flock *a_fl;
1493 int a_flags;
1494 } */ *ap = v;
1495 struct vnode *vp = ap->a_vp;
1496 struct rumpfs_node *rn = vp->v_data;
1497
1498 return lf_advlock(ap, &rn->rn_lockf, vp->v_size);
1499 }
1500
1501 /*
1502 * Begin vfs-level stuff
1503 */
1504
1505 VFS_PROTOS(rumpfs);
1506 struct vfsops rumpfs_vfsops = {
1507 .vfs_name = MOUNT_RUMPFS,
1508 .vfs_min_mount_data = 0,
1509 .vfs_mount = rumpfs_mount,
1510 .vfs_start = (void *)nullop,
1511 .vfs_unmount = rumpfs_unmount,
1512 .vfs_root = rumpfs_root,
1513 .vfs_quotactl = (void *)eopnotsupp,
1514 .vfs_statvfs = genfs_statvfs,
1515 .vfs_sync = (void *)nullop,
1516 .vfs_vget = rumpfs_vget,
1517 .vfs_fhtovp = (void *)eopnotsupp,
1518 .vfs_vptofh = (void *)eopnotsupp,
1519 .vfs_init = rumpfs_init,
1520 .vfs_reinit = NULL,
1521 .vfs_done = rumpfs_done,
1522 .vfs_mountroot = rumpfs_mountroot,
1523 .vfs_snapshot = (void *)eopnotsupp,
1524 .vfs_extattrctl = (void *)eopnotsupp,
1525 .vfs_suspendctl = (void *)eopnotsupp,
1526 .vfs_renamelock_enter = genfs_renamelock_enter,
1527 .vfs_renamelock_exit = genfs_renamelock_exit,
1528 .vfs_opv_descs = rump_opv_descs,
1529 /* vfs_refcount */
1530 /* vfs_list */
1531 };
1532
1533 static int
1534 rumpfs_mountfs(struct mount *mp)
1535 {
1536 struct rumpfs_mount *rfsmp;
1537 struct rumpfs_node *rn;
1538 int error;
1539
1540 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP);
1541
1542 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false);
1543 rn->rn_parent = rn;
1544 if ((error = makevnode(mp, rn, &rfsmp->rfsmp_rvp)) != 0)
1545 return error;
1546
1547 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT;
1548 VOP_UNLOCK(rfsmp->rfsmp_rvp);
1549
1550 mp->mnt_data = rfsmp;
1551 mp->mnt_stat.f_namemax = MAXNAMLEN;
1552 mp->mnt_stat.f_iosize = 512;
1553 mp->mnt_flag |= MNT_LOCAL;
1554 mp->mnt_iflag |= IMNT_MPSAFE;
1555 mp->mnt_fs_bshift = DEV_BSHIFT;
1556 vfs_getnewfsid(mp);
1557
1558 return 0;
1559 }
1560
1561 int
1562 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen)
1563 {
1564 int error;
1565
1566 if (mp->mnt_flag & MNT_RDONLY) {
1567 printf("rumpfs does not support r/o mounts\n");
1568 return EOPNOTSUPP;
1569 }
1570
1571 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE,
1572 mp->mnt_op->vfs_name, mp, curlwp);
1573 if (error)
1574 return error;
1575
1576 return rumpfs_mountfs(mp);
1577 }
1578
1579 int
1580 rumpfs_unmount(struct mount *mp, int mntflags)
1581 {
1582 struct rumpfs_mount *rfsmp = mp->mnt_data;
1583 int flags = 0, error;
1584
1585 if (panicstr || mntflags & MNT_FORCE)
1586 flags |= FORCECLOSE;
1587
1588 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0)
1589 return error;
1590 vgone(rfsmp->rfsmp_rvp); /* XXX */
1591
1592 kmem_free(rfsmp, sizeof(*rfsmp));
1593
1594 return 0;
1595 }
1596
1597 int
1598 rumpfs_root(struct mount *mp, struct vnode **vpp)
1599 {
1600 struct rumpfs_mount *rfsmp = mp->mnt_data;
1601
1602 vref(rfsmp->rfsmp_rvp);
1603 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY);
1604 *vpp = rfsmp->rfsmp_rvp;
1605 return 0;
1606 }
1607
1608 int
1609 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1610 {
1611
1612 return EOPNOTSUPP;
1613 }
1614
1615 void
1616 rumpfs_init()
1617 {
1618
1619 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET);
1620
1621 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE);
1622 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE);
1623 }
1624
1625 void
1626 rumpfs_done()
1627 {
1628
1629 mutex_destroy(&reclock);
1630 mutex_destroy(&etfs_lock);
1631 }
1632
1633 int
1634 rumpfs_mountroot()
1635 {
1636 struct mount *mp;
1637 int error;
1638
1639 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) {
1640 vrele(rootvp);
1641 return error;
1642 }
1643
1644 if ((error = rumpfs_mountfs(mp)) != 0)
1645 panic("mounting rootfs failed: %d", error);
1646
1647 mutex_enter(&mountlist_lock);
1648 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1649 mutex_exit(&mountlist_lock);
1650
1651 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE,
1652 mp->mnt_op->vfs_name, mp, curlwp);
1653 if (error)
1654 panic("set_statvfs_info failed for rootfs: %d", error);
1655
1656 mp->mnt_flag &= ~MNT_RDONLY;
1657 vfs_unbusy(mp, false, NULL);
1658
1659 return 0;
1660 }
1661