rumpfs.c revision 1.81 1 /* $NetBSD: rumpfs.c,v 1.81 2011/01/04 09:49:16 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 1.81 2011/01/04 09:49:16 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
86 int (**fifo_vnodeop_p)(void *);
87 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = {
88 { &vop_default_desc, vn_default_error },
89 { NULL, NULL }
90 };
91 const struct vnodeopv_desc fifo_vnodeop_opv_desc =
92 { &fifo_vnodeop_p, fifo_vnodeop_entries };
93
94 int (**rump_vnodeop_p)(void *);
95 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = {
96 { &vop_default_desc, vn_default_error },
97 { &vop_lookup_desc, rump_vop_lookup },
98 { &vop_getattr_desc, rump_vop_getattr },
99 { &vop_setattr_desc, rump_vop_setattr },
100 { &vop_mkdir_desc, rump_vop_mkdir },
101 { &vop_rmdir_desc, rump_vop_rmdir },
102 { &vop_remove_desc, rump_vop_remove },
103 { &vop_mknod_desc, rump_vop_mknod },
104 { &vop_create_desc, rump_vop_create },
105 { &vop_symlink_desc, rump_vop_symlink },
106 { &vop_readlink_desc, rump_vop_readlink },
107 { &vop_access_desc, rump_vop_success },
108 { &vop_readdir_desc, rump_vop_readdir },
109 { &vop_read_desc, rump_vop_read },
110 { &vop_write_desc, rump_vop_write },
111 { &vop_open_desc, rump_vop_open },
112 { &vop_close_desc, genfs_nullop },
113 { &vop_seek_desc, genfs_seek },
114 { &vop_getpages_desc, genfs_getpages },
115 { &vop_putpages_desc, genfs_putpages },
116 { &vop_whiteout_desc, rump_vop_whiteout },
117 { &vop_fsync_desc, rump_vop_success },
118 { &vop_lock_desc, genfs_lock },
119 { &vop_unlock_desc, genfs_unlock },
120 { &vop_islocked_desc, genfs_islocked },
121 { &vop_inactive_desc, rump_vop_inactive },
122 { &vop_reclaim_desc, rump_vop_reclaim },
123 { &vop_link_desc, genfs_eopnotsupp },
124 { &vop_pathconf_desc, rump_vop_pathconf },
125 { &vop_bmap_desc, rump_vop_bmap },
126 { &vop_strategy_desc, rump_vop_strategy },
127 { NULL, NULL }
128 };
129 const struct vnodeopv_desc rump_vnodeop_opv_desc =
130 { &rump_vnodeop_p, rump_vnodeop_entries };
131
132 int (**rump_specop_p)(void *);
133 const struct vnodeopv_entry_desc rump_specop_entries[] = {
134 { &vop_default_desc, rump_vop_spec },
135 { NULL, NULL }
136 };
137 const struct vnodeopv_desc rump_specop_opv_desc =
138 { &rump_specop_p, rump_specop_entries };
139
140 const struct vnodeopv_desc * const rump_opv_descs[] = {
141 &rump_vnodeop_opv_desc,
142 &rump_specop_opv_desc,
143 NULL
144 };
145
146 #define RUMPFS_WHITEOUT NULL
147 #define RDENT_ISWHITEOUT(rdp) (rdp->rd_node == RUMPFS_WHITEOUT)
148 struct rumpfs_dent {
149 char *rd_name;
150 int rd_namelen;
151 struct rumpfs_node *rd_node;
152
153 LIST_ENTRY(rumpfs_dent) rd_entries;
154 };
155
156 struct genfs_ops rumpfs_genfsops = {
157 .gop_size = genfs_size,
158 .gop_write = genfs_gop_write,
159
160 /* optional */
161 .gop_alloc = NULL,
162 .gop_markupdate = NULL,
163 };
164
165 struct rumpfs_node {
166 struct genfs_node rn_gn;
167 struct vattr rn_va;
168 struct vnode *rn_vp;
169 char *rn_hostpath;
170 int rn_flags;
171
172 union {
173 struct { /* VREG */
174 int readfd;
175 int writefd;
176 uint64_t offset;
177 } reg;
178 struct {
179 void *data;
180 size_t dlen;
181 } reg_noet;
182 struct { /* VDIR */
183 LIST_HEAD(, rumpfs_dent) dents;
184 struct rumpfs_node *parent;
185 int flags;
186 } dir;
187 struct {
188 char *target;
189 size_t len;
190 } link;
191 } rn_u;
192 };
193 #define rn_readfd rn_u.reg.readfd
194 #define rn_writefd rn_u.reg.writefd
195 #define rn_offset rn_u.reg.offset
196 #define rn_data rn_u.reg_noet.data
197 #define rn_dlen rn_u.reg_noet.dlen
198 #define rn_dir rn_u.dir.dents
199 #define rn_parent rn_u.dir.parent
200 #define rn_linktarg rn_u.link.target
201 #define rn_linklen rn_u.link.len
202
203 #define RUMPNODE_CANRECLAIM 0x01
204 #define RUMPNODE_DIR_ET 0x02
205 #define RUMPNODE_DIR_ETSUBS 0x04
206 #define RUMPNODE_ET_PHONE_HOST 0x10
207
208 struct rumpfs_mount {
209 struct vnode *rfsmp_rvp;
210 };
211
212 static struct rumpfs_node *makeprivate(enum vtype, dev_t, off_t, bool);
213
214 /*
215 * Extra Terrestrial stuff. We map a given key (pathname) to a file on
216 * the host FS. ET phones home only from the root node of rumpfs.
217 *
218 * When an etfs node is removed, a vnode potentially behind it is not
219 * immediately recycled.
220 */
221
222 struct etfs {
223 char et_key[MAXPATHLEN];
224 size_t et_keylen;
225 bool et_prefixkey;
226 bool et_removing;
227 devminor_t et_blkmin;
228
229 LIST_ENTRY(etfs) et_entries;
230
231 struct rumpfs_node *et_rn;
232 };
233 static kmutex_t etfs_lock;
234 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list);
235
236 static enum vtype
237 ettype_to_vtype(enum rump_etfs_type et)
238 {
239 enum vtype vt;
240
241 switch (et) {
242 case RUMP_ETFS_REG:
243 vt = VREG;
244 break;
245 case RUMP_ETFS_BLK:
246 vt = VBLK;
247 break;
248 case RUMP_ETFS_CHR:
249 vt = VCHR;
250 break;
251 case RUMP_ETFS_DIR:
252 vt = VDIR;
253 break;
254 case RUMP_ETFS_DIR_SUBDIRS:
255 vt = VDIR;
256 break;
257 default:
258 panic("invalid et type: %d", et);
259 }
260
261 return vt;
262 }
263
264 static enum vtype
265 hft_to_vtype(int hft)
266 {
267 enum vtype vt;
268
269 switch (hft) {
270 case RUMPUSER_FT_OTHER:
271 vt = VNON;
272 break;
273 case RUMPUSER_FT_DIR:
274 vt = VDIR;
275 break;
276 case RUMPUSER_FT_REG:
277 vt = VREG;
278 break;
279 case RUMPUSER_FT_BLK:
280 vt = VBLK;
281 break;
282 case RUMPUSER_FT_CHR:
283 vt = VCHR;
284 break;
285 default:
286 vt = VNON;
287 break;
288 }
289
290 return vt;
291 }
292
293 static bool
294 etfs_find(const char *key, struct etfs **etp, bool forceprefix)
295 {
296 struct etfs *et;
297 size_t keylen = strlen(key);
298
299 KASSERT(mutex_owned(&etfs_lock));
300
301 LIST_FOREACH(et, &etfs_list, et_entries) {
302 if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix)
303 && strncmp(key, et->et_key, et->et_keylen) == 0) {
304 if (etp)
305 *etp = et;
306 return true;
307 }
308 }
309
310 return false;
311 }
312
313 #define REGDIR(ftype) \
314 ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS)
315 static int
316 doregister(const char *key, const char *hostpath,
317 enum rump_etfs_type ftype, uint64_t begin, uint64_t size)
318 {
319 char buf[9];
320 struct etfs *et;
321 struct rumpfs_node *rn;
322 uint64_t fsize;
323 dev_t rdev = NODEV;
324 devminor_t dmin = -1;
325 int hft, error;
326
327 if (key[0] != '/') {
328 return EINVAL;
329 }
330 while (key[0] == '/') {
331 key++;
332 }
333
334 if (rumpuser_getfileinfo(hostpath, &fsize, &hft, &error))
335 return error;
336
337 /* etfs directory requires a directory on the host */
338 if (REGDIR(ftype)) {
339 if (hft != RUMPUSER_FT_DIR)
340 return ENOTDIR;
341 if (begin != 0)
342 return EISDIR;
343 if (size != RUMP_ETFS_SIZE_ENDOFF)
344 return EISDIR;
345 size = fsize;
346 } else {
347 if (begin > fsize)
348 return EINVAL;
349 if (size == RUMP_ETFS_SIZE_ENDOFF)
350 size = fsize - begin;
351 if (begin + size > fsize)
352 return EINVAL;
353 }
354
355 if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) {
356 error = rumpblk_register(hostpath, &dmin, begin, size);
357 if (error != 0) {
358 return error;
359 }
360 rdev = makedev(RUMPBLK_DEVMAJOR, dmin);
361 }
362
363 et = kmem_alloc(sizeof(*et), KM_SLEEP);
364 strcpy(et->et_key, key);
365 et->et_keylen = strlen(et->et_key);
366 et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), rdev, size, true);
367 et->et_removing = false;
368 et->et_blkmin = dmin;
369
370 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST;
371
372 if (ftype == RUMP_ETFS_REG || REGDIR(ftype) || et->et_blkmin != -1) {
373 size_t len = strlen(hostpath)+1;
374
375 rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
376 memcpy(rn->rn_hostpath, hostpath, len);
377 rn->rn_offset = begin;
378 }
379
380 if (REGDIR(ftype)) {
381 rn->rn_flags |= RUMPNODE_DIR_ET;
382 et->et_prefixkey = true;
383 } else {
384 et->et_prefixkey = false;
385 }
386
387 if (ftype == RUMP_ETFS_DIR_SUBDIRS)
388 rn->rn_flags |= RUMPNODE_DIR_ETSUBS;
389
390 mutex_enter(&etfs_lock);
391 if (etfs_find(key, NULL, REGDIR(ftype))) {
392 mutex_exit(&etfs_lock);
393 if (et->et_blkmin != -1)
394 rumpblk_deregister(hostpath);
395 if (et->et_rn->rn_hostpath != NULL)
396 free(et->et_rn->rn_hostpath, M_TEMP);
397 kmem_free(et->et_rn, sizeof(*et->et_rn));
398 kmem_free(et, sizeof(*et));
399 return EEXIST;
400 }
401 LIST_INSERT_HEAD(&etfs_list, et, et_entries);
402 mutex_exit(&etfs_lock);
403
404 if (ftype == RUMP_ETFS_BLK) {
405 format_bytes(buf, sizeof(buf), size);
406 aprint_verbose("/%s: hostpath %s (%s)\n", key, hostpath, buf);
407 }
408
409 return 0;
410 }
411 #undef REGDIR
412
413 int
414 rump_etfs_register(const char *key, const char *hostpath,
415 enum rump_etfs_type ftype)
416 {
417
418 return doregister(key, hostpath, ftype, 0, RUMP_ETFS_SIZE_ENDOFF);
419 }
420
421 int
422 rump_etfs_register_withsize(const char *key, const char *hostpath,
423 enum rump_etfs_type ftype, uint64_t begin, uint64_t size)
424 {
425
426 return doregister(key, hostpath, ftype, begin, size);
427 }
428
429 /* remove etfs mapping. caller's responsibility to make sure it's not in use */
430 int
431 rump_etfs_remove(const char *key)
432 {
433 struct etfs *et;
434 size_t keylen;
435 int rv;
436
437 if (key[0] != '/') {
438 return EINVAL;
439 }
440 while (key[0] == '/') {
441 key++;
442 }
443
444 keylen = strlen(key);
445
446 mutex_enter(&etfs_lock);
447 LIST_FOREACH(et, &etfs_list, et_entries) {
448 if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) {
449 if (et->et_removing)
450 et = NULL;
451 else
452 et->et_removing = true;
453 break;
454 }
455 }
456 mutex_exit(&etfs_lock);
457 if (!et)
458 return ENOENT;
459
460 /*
461 * ok, we know what we want to remove and have signalled there
462 * actually are men at work. first, unregister from rumpblk
463 */
464 if (et->et_blkmin != -1) {
465 rv = rumpblk_deregister(et->et_rn->rn_hostpath);
466 } else {
467 rv = 0;
468 }
469 KASSERT(rv == 0);
470
471 /* then do the actual removal */
472 mutex_enter(&etfs_lock);
473 LIST_REMOVE(et, et_entries);
474 mutex_exit(&etfs_lock);
475
476 /* node is unreachable, safe to nuke all device copies */
477 if (et->et_blkmin != -1)
478 vdevgone(RUMPBLK_DEVMAJOR, et->et_blkmin, et->et_blkmin, VBLK);
479
480 if (et->et_rn->rn_hostpath != NULL)
481 free(et->et_rn->rn_hostpath, M_TEMP);
482 kmem_free(et->et_rn, sizeof(*et->et_rn));
483 kmem_free(et, sizeof(*et));
484
485 return 0;
486 }
487
488 /*
489 * rumpfs
490 */
491
492 #define INO_WHITEOUT 1
493 static int lastino = 2;
494 static kmutex_t reclock;
495
496 static struct rumpfs_node *
497 makeprivate(enum vtype vt, dev_t rdev, off_t size, bool et)
498 {
499 struct rumpfs_node *rn;
500 struct vattr *va;
501 struct timespec ts;
502
503 rn = kmem_zalloc(sizeof(*rn), KM_SLEEP);
504
505 switch (vt) {
506 case VDIR:
507 LIST_INIT(&rn->rn_dir);
508 break;
509 case VREG:
510 if (et) {
511 rn->rn_readfd = -1;
512 rn->rn_writefd = -1;
513 }
514 break;
515 default:
516 break;
517 }
518
519 nanotime(&ts);
520
521 va = &rn->rn_va;
522 va->va_type = vt;
523 va->va_mode = 0755;
524 if (vt == VDIR)
525 va->va_nlink = 2;
526 else
527 va->va_nlink = 1;
528 va->va_uid = 0;
529 va->va_gid = 0;
530 va->va_fsid =
531 va->va_fileid = atomic_inc_uint_nv(&lastino);
532 va->va_size = size;
533 va->va_blocksize = 512;
534 va->va_atime = ts;
535 va->va_mtime = ts;
536 va->va_ctime = ts;
537 va->va_birthtime = ts;
538 va->va_gen = 0;
539 va->va_flags = 0;
540 va->va_rdev = rdev;
541 va->va_bytes = 512;
542 va->va_filerev = 0;
543 va->va_vaflags = 0;
544
545 return rn;
546 }
547
548 static int
549 makevnode(struct mount *mp, struct rumpfs_node *rn, struct vnode **vpp)
550 {
551 struct vnode *vp;
552 int (**vpops)(void *);
553 struct vattr *va = &rn->rn_va;
554 int rv;
555
556 KASSERT(!mutex_owned(&reclock));
557
558 if (va->va_type == VCHR || va->va_type == VBLK) {
559 vpops = rump_specop_p;
560 } else {
561 vpops = rump_vnodeop_p;
562 }
563
564 rv = getnewvnode(VT_RUMP, mp, vpops, &vp);
565 if (rv)
566 return rv;
567
568 vp->v_size = vp->v_writesize = va->va_size;
569 vp->v_type = va->va_type;
570
571 if (vpops == rump_specop_p) {
572 spec_node_init(vp, va->va_rdev);
573 }
574 vp->v_data = rn;
575
576 genfs_node_init(vp, &rumpfs_genfsops);
577 vn_lock(vp, LK_RETRY | LK_EXCLUSIVE);
578 mutex_enter(&reclock);
579 rn->rn_vp = vp;
580 mutex_exit(&reclock);
581
582 *vpp = vp;
583
584 return 0;
585 }
586
587
588 static void
589 makedir(struct rumpfs_node *rnd,
590 struct componentname *cnp, struct rumpfs_node *rn)
591 {
592 struct rumpfs_dent *rdent;
593
594 rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP);
595 rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP);
596 rdent->rd_node = rn;
597 strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1);
598 rdent->rd_namelen = strlen(rdent->rd_name);
599
600 LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries);
601 }
602
603 static void
604 freedir(struct rumpfs_node *rnd, struct componentname *cnp)
605 {
606 struct rumpfs_dent *rd = NULL;
607
608 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) {
609 if (rd->rd_namelen == cnp->cn_namelen &&
610 strncmp(rd->rd_name, cnp->cn_nameptr,
611 cnp->cn_namelen) == 0)
612 break;
613 }
614 if (rd == NULL)
615 panic("could not find directory entry: %s", cnp->cn_nameptr);
616
617 LIST_REMOVE(rd, rd_entries);
618 kmem_free(rd->rd_name, rd->rd_namelen+1);
619 kmem_free(rd, sizeof(*rd));
620 }
621
622 /*
623 * Simple lookup for rump file systems.
624 *
625 * uhm, this is twisted. C F C C, hope of C C F C looming
626 */
627 static int
628 rump_vop_lookup(void *v)
629 {
630 struct vop_lookup_args /* {
631 struct vnode *a_dvp;
632 struct vnode **a_vpp;
633 struct componentname *a_cnp;
634 }; */ *ap = v;
635 struct componentname *cnp = ap->a_cnp;
636 struct vnode *dvp = ap->a_dvp;
637 struct vnode **vpp = ap->a_vpp;
638 struct vnode *vp;
639 struct rumpfs_node *rnd = dvp->v_data, *rn;
640 struct rumpfs_dent *rd = NULL;
641 struct etfs *et;
642 bool dotdot = (cnp->cn_flags & ISDOTDOT) != 0;
643 int rv = 0;
644
645 /* check for dot, return directly if the case */
646 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
647 vref(dvp);
648 *vpp = dvp;
649 return 0;
650 }
651
652 /* we don't do rename */
653 if (!(((cnp->cn_flags & ISLASTCN) == 0) || (cnp->cn_nameiop != RENAME)))
654 return EOPNOTSUPP;
655
656 /* check for etfs */
657 if (dvp == rootvnode && cnp->cn_nameiop == LOOKUP) {
658 bool found;
659 mutex_enter(&etfs_lock);
660 found = etfs_find(cnp->cn_nameptr, &et, false);
661 mutex_exit(&etfs_lock);
662
663 if (found) {
664 rn = et->et_rn;
665 cnp->cn_consume += et->et_keylen - cnp->cn_namelen;
666 if (rn->rn_va.va_type != VDIR)
667 cnp->cn_flags &= ~REQUIREDIR;
668 goto getvnode;
669 }
670 }
671
672 if (rnd->rn_flags & RUMPNODE_DIR_ET) {
673 uint64_t fsize;
674 char *newpath;
675 size_t newpathlen;
676 int hft, error;
677
678 if (dotdot)
679 return EOPNOTSUPP;
680
681 newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1;
682 newpath = malloc(newpathlen, M_TEMP, M_WAITOK);
683
684 strlcpy(newpath, rnd->rn_hostpath, newpathlen);
685 strlcat(newpath, "/", newpathlen);
686 strlcat(newpath, cnp->cn_nameptr, newpathlen);
687
688 if (rumpuser_getfileinfo(newpath, &fsize, &hft, &error)) {
689 free(newpath, M_TEMP);
690 return error;
691 }
692
693 /* allow only dirs and regular files */
694 if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) {
695 free(newpath, M_TEMP);
696 return ENOENT;
697 }
698
699 rn = makeprivate(hft_to_vtype(hft), NODEV, fsize, true);
700 rn->rn_flags |= RUMPNODE_CANRECLAIM;
701 if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) {
702 rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS;
703 rn->rn_flags |= RUMPNODE_ET_PHONE_HOST;
704 }
705 rn->rn_hostpath = newpath;
706
707 goto getvnode;
708 } else {
709 if (dotdot) {
710 rn = rnd->rn_parent;
711 goto getvnode;
712 } else {
713 LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) {
714 if (rd->rd_namelen == cnp->cn_namelen &&
715 strncmp(rd->rd_name, cnp->cn_nameptr,
716 cnp->cn_namelen) == 0)
717 break;
718 }
719 }
720 }
721
722 if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE))
723 return ENOENT;
724
725 if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) {
726 return EJUSTRETURN;
727 }
728
729 rn = rd->rd_node;
730
731 getvnode:
732 KASSERT(rn);
733 if (dotdot)
734 VOP_UNLOCK(dvp);
735 mutex_enter(&reclock);
736 if ((vp = rn->rn_vp)) {
737 mutex_enter(&vp->v_interlock);
738 mutex_exit(&reclock);
739 if (vget(vp, LK_EXCLUSIVE)) {
740 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
741 goto getvnode;
742 }
743 *vpp = vp;
744 } else {
745 mutex_exit(&reclock);
746 rv = makevnode(dvp->v_mount, rn, vpp);
747 }
748 if (dotdot)
749 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
750
751 return rv;
752 }
753
754 static int
755 rump_vop_getattr(void *v)
756 {
757 struct vop_getattr_args /* {
758 struct vnode *a_vp;
759 struct vattr *a_vap;
760 kauth_cred_t a_cred;
761 } */ *ap = v;
762 struct vnode *vp = ap->a_vp;
763 struct rumpfs_node *rn = vp->v_data;
764 struct vattr *vap = ap->a_vap;
765
766 memcpy(vap, &rn->rn_va, sizeof(struct vattr));
767 vap->va_size = vp->v_size;
768 return 0;
769 }
770
771 static int
772 rump_vop_setattr(void *v)
773 {
774 struct vop_getattr_args /* {
775 struct vnode *a_vp;
776 struct vattr *a_vap;
777 kauth_cred_t a_cred;
778 } */ *ap = v;
779 struct vnode *vp = ap->a_vp;
780 struct vattr *vap = ap->a_vap;
781 struct rumpfs_node *rn = vp->v_data;
782
783 #define SETIFVAL(a,t) if (vap->a != (t)VNOVAL) rn->rn_va.a = vap->a
784 SETIFVAL(va_mode, mode_t);
785 SETIFVAL(va_uid, uid_t);
786 SETIFVAL(va_gid, gid_t);
787 SETIFVAL(va_atime.tv_sec, time_t);
788 SETIFVAL(va_ctime.tv_sec, time_t);
789 SETIFVAL(va_mtime.tv_sec, time_t);
790 SETIFVAL(va_birthtime.tv_sec, time_t);
791 SETIFVAL(va_atime.tv_nsec, long);
792 SETIFVAL(va_ctime.tv_nsec, long);
793 SETIFVAL(va_mtime.tv_nsec, long);
794 SETIFVAL(va_birthtime.tv_nsec, long);
795 SETIFVAL(va_flags, u_long);
796 #undef SETIFVAL
797
798 if (vp->v_type == VREG && vap->va_size != VSIZENOTSET)
799 uvm_vnp_setsize(vp, vap->va_size);
800 return 0;
801 }
802
803 static int
804 rump_vop_mkdir(void *v)
805 {
806 struct vop_mkdir_args /* {
807 struct vnode *a_dvp;
808 struct vnode **a_vpp;
809 struct componentname *a_cnp;
810 struct vattr *a_vap;
811 }; */ *ap = v;
812 struct vnode *dvp = ap->a_dvp;
813 struct vnode **vpp = ap->a_vpp;
814 struct componentname *cnp = ap->a_cnp;
815 struct rumpfs_node *rnd = dvp->v_data, *rn;
816 int rv = 0;
817
818 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false);
819 rn->rn_parent = rnd;
820 rv = makevnode(dvp->v_mount, rn, vpp);
821 if (rv)
822 goto out;
823
824 makedir(rnd, cnp, rn);
825
826 out:
827 vput(dvp);
828 return rv;
829 }
830
831 static int
832 rump_vop_rmdir(void *v)
833 {
834 struct vop_rmdir_args /* {
835 struct vnode *a_dvp;
836 struct vnode *a_vp;
837 struct componentname *a_cnp;
838 }; */ *ap = v;
839 struct vnode *dvp = ap->a_dvp;
840 struct vnode *vp = ap->a_vp;
841 struct componentname *cnp = ap->a_cnp;
842 struct rumpfs_node *rnd = dvp->v_data;
843 struct rumpfs_node *rn = vp->v_data;
844 int rv = 0;
845
846 if (!LIST_EMPTY(&rn->rn_dir)) {
847 rv = ENOTEMPTY;
848 goto out;
849 }
850
851 freedir(rnd, cnp);
852 rn->rn_flags |= RUMPNODE_CANRECLAIM;
853
854 out:
855 vput(dvp);
856 vput(vp);
857
858 return rv;
859 }
860
861 static int
862 rump_vop_remove(void *v)
863 {
864 struct vop_rmdir_args /* {
865 struct vnode *a_dvp;
866 struct vnode *a_vp;
867 struct componentname *a_cnp;
868 }; */ *ap = v;
869 struct vnode *dvp = ap->a_dvp;
870 struct vnode *vp = ap->a_vp;
871 struct componentname *cnp = ap->a_cnp;
872 struct rumpfs_node *rnd = dvp->v_data;
873 struct rumpfs_node *rn = vp->v_data;
874 int rv = 0;
875
876 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)
877 return EOPNOTSUPP;
878
879 if (vp->v_type == VREG) {
880 rump_hyperfree(rn->rn_data, rn->rn_dlen);
881 }
882
883 freedir(rnd, cnp);
884 rn->rn_flags |= RUMPNODE_CANRECLAIM;
885
886 vput(dvp);
887 vput(vp);
888
889 return rv;
890 }
891
892 static int
893 rump_vop_mknod(void *v)
894 {
895 struct vop_mknod_args /* {
896 struct vnode *a_dvp;
897 struct vnode **a_vpp;
898 struct componentname *a_cnp;
899 struct vattr *a_vap;
900 }; */ *ap = v;
901 struct vnode *dvp = ap->a_dvp;
902 struct vnode **vpp = ap->a_vpp;
903 struct componentname *cnp = ap->a_cnp;
904 struct vattr *va = ap->a_vap;
905 struct rumpfs_node *rnd = dvp->v_data, *rn;
906 int rv;
907
908 rn = makeprivate(va->va_type, va->va_rdev, DEV_BSIZE, false);
909 rv = makevnode(dvp->v_mount, rn, vpp);
910 if (rv)
911 goto out;
912
913 makedir(rnd, cnp, rn);
914
915 out:
916 vput(dvp);
917 return rv;
918 }
919
920 static int
921 rump_vop_create(void *v)
922 {
923 struct vop_create_args /* {
924 struct vnode *a_dvp;
925 struct vnode **a_vpp;
926 struct componentname *a_cnp;
927 struct vattr *a_vap;
928 }; */ *ap = v;
929 struct vnode *dvp = ap->a_dvp;
930 struct vnode **vpp = ap->a_vpp;
931 struct componentname *cnp = ap->a_cnp;
932 struct vattr *va = ap->a_vap;
933 struct rumpfs_node *rnd = dvp->v_data, *rn;
934 off_t newsize;
935 int rv;
936
937 newsize = va->va_type == VSOCK ? DEV_BSIZE : 0;
938 rn = makeprivate(va->va_type, NODEV, newsize, false);
939 rv = makevnode(dvp->v_mount, rn, vpp);
940 if (rv)
941 goto out;
942
943 makedir(rnd, cnp, rn);
944
945 out:
946 vput(dvp);
947 return rv;
948 }
949
950 static int
951 rump_vop_symlink(void *v)
952 {
953 struct vop_symlink_args /* {
954 struct vnode *a_dvp;
955 struct vnode **a_vpp;
956 struct componentname *a_cnp;
957 struct vattr *a_vap;
958 char *a_target;
959 }; */ *ap = v;
960 struct vnode *dvp = ap->a_dvp;
961 struct vnode **vpp = ap->a_vpp;
962 struct componentname *cnp = ap->a_cnp;
963 struct rumpfs_node *rnd = dvp->v_data, *rn;
964 const char *target = ap->a_target;
965 size_t linklen;
966 int rv;
967
968 linklen = strlen(target);
969 KASSERT(linklen < MAXPATHLEN);
970 rn = makeprivate(VLNK, NODEV, linklen, false);
971 rv = makevnode(dvp->v_mount, rn, vpp);
972 if (rv)
973 goto out;
974
975 makedir(rnd, cnp, rn);
976
977 KASSERT(linklen < MAXPATHLEN);
978 rn->rn_linktarg = PNBUF_GET();
979 rn->rn_linklen = linklen;
980 strcpy(rn->rn_linktarg, target);
981
982 out:
983 vput(dvp);
984 return rv;
985 }
986
987 static int
988 rump_vop_readlink(void *v)
989 {
990 struct vop_readlink_args /* {
991 struct vnode *a_vp;
992 struct uio *a_uio;
993 kauth_cred_t a_cred;
994 }; */ *ap = v;
995 struct vnode *vp = ap->a_vp;
996 struct rumpfs_node *rn = vp->v_data;
997 struct uio *uio = ap->a_uio;
998
999 return uiomove(rn->rn_linktarg, rn->rn_linklen, uio);
1000 }
1001
1002 static int
1003 rump_vop_whiteout(void *v)
1004 {
1005 struct vop_whiteout_args /* {
1006 struct vnode *a_dvp;
1007 struct componentname *a_cnp;
1008 int a_flags;
1009 } */ *ap = v;
1010 struct vnode *dvp = ap->a_dvp;
1011 struct rumpfs_node *rnd = dvp->v_data;
1012 struct componentname *cnp = ap->a_cnp;
1013 int flags = ap->a_flags;
1014
1015 switch (flags) {
1016 case LOOKUP:
1017 break;
1018 case CREATE:
1019 makedir(rnd, cnp, RUMPFS_WHITEOUT);
1020 break;
1021 case DELETE:
1022 cnp->cn_flags &= ~DOWHITEOUT; /* cargo culting never fails ? */
1023 freedir(rnd, cnp);
1024 break;
1025 default:
1026 panic("unknown whiteout op %d", flags);
1027 }
1028
1029 return 0;
1030 }
1031
1032 static int
1033 rump_vop_open(void *v)
1034 {
1035 struct vop_open_args /* {
1036 struct vnode *a_vp;
1037 int a_mode;
1038 kauth_cred_t a_cred;
1039 } */ *ap = v;
1040 struct vnode *vp = ap->a_vp;
1041 struct rumpfs_node *rn = vp->v_data;
1042 int mode = ap->a_mode;
1043 int error = EINVAL;
1044
1045 if (vp->v_type != VREG || (rn->rn_flags & RUMPNODE_ET_PHONE_HOST) == 0)
1046 return 0;
1047
1048 if (mode & FREAD) {
1049 if (rn->rn_readfd != -1)
1050 return 0;
1051 rn->rn_readfd = rumpuser_open(rn->rn_hostpath,
1052 O_RDONLY, &error);
1053 }
1054
1055 if (mode & FWRITE) {
1056 if (rn->rn_writefd != -1)
1057 return 0;
1058 rn->rn_writefd = rumpuser_open(rn->rn_hostpath,
1059 O_WRONLY, &error);
1060 }
1061
1062 return error;
1063 }
1064
1065 /* simple readdir. event omits dotstuff and periods */
1066 static int
1067 rump_vop_readdir(void *v)
1068 {
1069 struct vop_readdir_args /* {
1070 struct vnode *a_vp;
1071 struct uio *a_uio;
1072 kauth_cred_t a_cred;
1073 int *a_eofflag;
1074 off_t **a_cookies;
1075 int *a_ncookies;
1076 } */ *ap = v;
1077 struct vnode *vp = ap->a_vp;
1078 struct uio *uio = ap->a_uio;
1079 struct rumpfs_node *rnd = vp->v_data;
1080 struct rumpfs_dent *rdent;
1081 unsigned i;
1082 int rv = 0;
1083
1084 /* seek to current entry */
1085 for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir);
1086 (i < uio->uio_offset) && rdent;
1087 i++, rdent = LIST_NEXT(rdent, rd_entries))
1088 continue;
1089 if (!rdent)
1090 goto out;
1091
1092 /* copy entries */
1093 for (; rdent && uio->uio_resid > 0;
1094 rdent = LIST_NEXT(rdent, rd_entries), i++) {
1095 struct dirent dent;
1096
1097 strlcpy(dent.d_name, rdent->rd_name, sizeof(dent.d_name));
1098 dent.d_namlen = strlen(dent.d_name);
1099 dent.d_reclen = _DIRENT_RECLEN(&dent, dent.d_namlen);
1100
1101 if (__predict_false(RDENT_ISWHITEOUT(rdent))) {
1102 dent.d_fileno = INO_WHITEOUT;
1103 dent.d_type = DT_WHT;
1104 } else {
1105 dent.d_fileno = rdent->rd_node->rn_va.va_fileid;
1106 dent.d_type = vtype2dt(rdent->rd_node->rn_va.va_type);
1107 }
1108
1109 if (uio->uio_resid < dent.d_reclen) {
1110 i--;
1111 break;
1112 }
1113
1114 rv = uiomove(&dent, dent.d_reclen, uio);
1115 if (rv) {
1116 i--;
1117 break;
1118 }
1119 }
1120
1121 out:
1122 if (ap->a_cookies) {
1123 *ap->a_ncookies = 0;
1124 *ap->a_cookies = NULL;
1125 }
1126 if (rdent)
1127 *ap->a_eofflag = 0;
1128 else
1129 *ap->a_eofflag = 1;
1130 uio->uio_offset = i;
1131
1132 return rv;
1133 }
1134
1135 static int
1136 etread(struct rumpfs_node *rn, struct uio *uio)
1137 {
1138 uint8_t *buf;
1139 size_t bufsize;
1140 ssize_t n;
1141 int error = 0;
1142
1143 bufsize = uio->uio_resid;
1144 buf = kmem_alloc(bufsize, KM_SLEEP);
1145 if ((n = rumpuser_pread(rn->rn_readfd, buf, bufsize,
1146 uio->uio_offset + rn->rn_offset, &error)) == -1)
1147 goto out;
1148 KASSERT(n <= bufsize);
1149 error = uiomove(buf, n, uio);
1150
1151 out:
1152 kmem_free(buf, bufsize);
1153 return error;
1154
1155 }
1156
1157 static int
1158 rump_vop_read(void *v)
1159 {
1160 struct vop_read_args /* {
1161 struct vnode *a_vp;
1162 struct uio *a_uio;
1163 int ioflags a_ioflag;
1164 kauth_cred_t a_cred;
1165 }; */ *ap = v;
1166 struct vnode *vp = ap->a_vp;
1167 struct rumpfs_node *rn = vp->v_data;
1168 struct uio *uio = ap->a_uio;
1169 const int advice = IO_ADV_DECODE(ap->a_ioflag);
1170 off_t chunk;
1171 int error = 0;
1172
1173 /* et op? */
1174 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)
1175 return etread(rn, uio);
1176
1177 /* otherwise, it's off to ubc with us */
1178 while (uio->uio_resid > 0) {
1179 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset);
1180 if (chunk == 0)
1181 break;
1182 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice,
1183 UBC_READ | UBC_PARTIALOK | UBC_WANT_UNMAP(vp)?UBC_UNMAP:0);
1184 if (error)
1185 break;
1186 }
1187
1188 return error;
1189 }
1190
1191 static int
1192 etwrite(struct rumpfs_node *rn, struct uio *uio)
1193 {
1194 uint8_t *buf;
1195 size_t bufsize;
1196 ssize_t n;
1197 int error = 0;
1198
1199 bufsize = uio->uio_resid;
1200 buf = kmem_alloc(bufsize, KM_SLEEP);
1201 error = uiomove(buf, bufsize, uio);
1202 if (error)
1203 goto out;
1204 KASSERT(uio->uio_resid == 0);
1205 n = rumpuser_pwrite(rn->rn_writefd, buf, bufsize,
1206 (uio->uio_offset-bufsize) + rn->rn_offset, &error);
1207 if (n >= 0) {
1208 KASSERT(n <= bufsize);
1209 uio->uio_resid = bufsize - n;
1210 }
1211
1212 out:
1213 kmem_free(buf, bufsize);
1214 return error;
1215 }
1216
1217 static int
1218 rump_vop_write(void *v)
1219 {
1220 struct vop_read_args /* {
1221 struct vnode *a_vp;
1222 struct uio *a_uio;
1223 int ioflags a_ioflag;
1224 kauth_cred_t a_cred;
1225 }; */ *ap = v;
1226 struct vnode *vp = ap->a_vp;
1227 struct rumpfs_node *rn = vp->v_data;
1228 struct uio *uio = ap->a_uio;
1229 const int advice = IO_ADV_DECODE(ap->a_ioflag);
1230 void *olddata;
1231 size_t oldlen, newlen;
1232 off_t chunk;
1233 int error = 0;
1234 bool allocd = false;
1235
1236 if (ap->a_ioflag & IO_APPEND)
1237 uio->uio_offset = vp->v_size;
1238
1239 /* consult et? */
1240 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST)
1241 return etwrite(rn, uio);
1242
1243 /*
1244 * Otherwise, it's a case of ubcmove.
1245 */
1246
1247 /*
1248 * First, make sure we have enough storage.
1249 *
1250 * No, you don't need to tell me it's not very efficient.
1251 * No, it doesn't really support sparse files, just fakes it.
1252 */
1253 newlen = uio->uio_offset + uio->uio_resid;
1254 oldlen = 0; /* XXXgcc */
1255 olddata = NULL;
1256 if (rn->rn_dlen < newlen) {
1257 oldlen = rn->rn_dlen;
1258 olddata = rn->rn_data;
1259
1260 rn->rn_data = rump_hypermalloc(newlen, 0, true, "rumpfs");
1261 rn->rn_dlen = newlen;
1262 memset(rn->rn_data, 0, newlen);
1263 memcpy(rn->rn_data, olddata, oldlen);
1264 allocd = true;
1265 uvm_vnp_setsize(vp, newlen);
1266 }
1267
1268 /* ok, we have enough stooorage. write */
1269 while (uio->uio_resid > 0) {
1270 chunk = MIN(uio->uio_resid, (off_t)rn->rn_dlen-uio->uio_offset);
1271 if (chunk == 0)
1272 break;
1273 error = ubc_uiomove(&vp->v_uobj, uio, chunk, advice,
1274 UBC_WRITE | UBC_PARTIALOK | UBC_WANT_UNMAP(vp)?UBC_UNMAP:0);
1275 if (error)
1276 break;
1277 }
1278
1279 if (allocd) {
1280 if (error) {
1281 rump_hyperfree(rn->rn_data, newlen);
1282 rn->rn_data = olddata;
1283 rn->rn_dlen = oldlen;
1284 uvm_vnp_setsize(vp, oldlen);
1285 } else {
1286 rump_hyperfree(olddata, oldlen);
1287 }
1288 }
1289
1290 return error;
1291 }
1292
1293 static int
1294 rump_vop_bmap(void *v)
1295 {
1296 struct vop_bmap_args /* {
1297 struct vnode *a_vp;
1298 daddr_t a_bn;
1299 struct vnode **a_vpp;
1300 daddr_t *a_bnp;
1301 int *a_runp;
1302 } */ *ap = v;
1303
1304 /* 1:1 mapping */
1305 if (ap->a_vpp)
1306 *ap->a_vpp = ap->a_vp;
1307 if (ap->a_bnp)
1308 *ap->a_bnp = ap->a_bn;
1309 if (ap->a_runp)
1310 *ap->a_runp = 16;
1311
1312 return 0;
1313 }
1314
1315 static int
1316 rump_vop_strategy(void *v)
1317 {
1318 struct vop_strategy_args /* {
1319 struct vnode *a_vp;
1320 struct buf *a_bp;
1321 } */ *ap = v;
1322 struct vnode *vp = ap->a_vp;
1323 struct rumpfs_node *rn = vp->v_data;
1324 struct buf *bp = ap->a_bp;
1325 off_t copylen, copyoff;
1326 int error;
1327
1328 if (vp->v_type != VREG || rn->rn_flags & RUMPNODE_ET_PHONE_HOST) {
1329 error = EINVAL;
1330 goto out;
1331 }
1332
1333 copyoff = bp->b_blkno << DEV_BSHIFT;
1334 copylen = MIN(rn->rn_dlen - copyoff, bp->b_bcount);
1335 if (BUF_ISWRITE(bp)) {
1336 memcpy((uint8_t *)rn->rn_data + copyoff, bp->b_data, copylen);
1337 } else {
1338 memset((uint8_t*)bp->b_data + copylen, 0, bp->b_bcount-copylen);
1339 memcpy(bp->b_data, (uint8_t *)rn->rn_data + copyoff, copylen);
1340 }
1341 bp->b_resid = 0;
1342 error = 0;
1343
1344 out:
1345 bp->b_error = error;
1346 biodone(bp);
1347 return 0;
1348 }
1349
1350 static int
1351 rump_vop_pathconf(void *v)
1352 {
1353 struct vop_pathconf_args /* {
1354 struct vnode *a_vp;
1355 int a_name;
1356 register_t *a_retval;
1357 }; */ *ap = v;
1358 int name = ap->a_name;
1359 register_t *retval = ap->a_retval;
1360
1361 switch (name) {
1362 case _PC_LINK_MAX:
1363 *retval = LINK_MAX;
1364 return 0;
1365 case _PC_NAME_MAX:
1366 *retval = NAME_MAX;
1367 return 0;
1368 case _PC_PATH_MAX:
1369 *retval = PATH_MAX;
1370 return 0;
1371 case _PC_PIPE_BUF:
1372 *retval = PIPE_BUF;
1373 return 0;
1374 case _PC_CHOWN_RESTRICTED:
1375 *retval = 1;
1376 return 0;
1377 case _PC_NO_TRUNC:
1378 *retval = 1;
1379 return 0;
1380 case _PC_SYNC_IO:
1381 *retval = 1;
1382 return 0;
1383 case _PC_FILESIZEBITS:
1384 *retval = 43; /* this one goes to 11 */
1385 return 0;
1386 case _PC_SYMLINK_MAX:
1387 *retval = MAXPATHLEN;
1388 return 0;
1389 case _PC_2_SYMLINKS:
1390 *retval = 1;
1391 return 0;
1392 default:
1393 return EINVAL;
1394 }
1395 }
1396
1397 static int
1398 rump_vop_success(void *v)
1399 {
1400
1401 return 0;
1402 }
1403
1404 static int
1405 rump_vop_inactive(void *v)
1406 {
1407 struct vop_inactive_args /* {
1408 struct vnode *a_vp;
1409 bool *a_recycle;
1410 } */ *ap = v;
1411 struct vnode *vp = ap->a_vp;
1412 struct rumpfs_node *rn = vp->v_data;
1413 int error;
1414
1415 if (rn->rn_flags & RUMPNODE_ET_PHONE_HOST && vp->v_type == VREG) {
1416 if (rn->rn_readfd != -1) {
1417 rumpuser_close(rn->rn_readfd, &error);
1418 rn->rn_readfd = -1;
1419 }
1420 if (rn->rn_writefd != -1) {
1421 rumpuser_close(rn->rn_writefd, &error);
1422 rn->rn_writefd = -1;
1423 }
1424 }
1425 *ap->a_recycle = (rn->rn_flags & RUMPNODE_CANRECLAIM) ? true : false;
1426
1427 VOP_UNLOCK(vp);
1428 return 0;
1429 }
1430
1431 static int
1432 rump_vop_reclaim(void *v)
1433 {
1434 struct vop_reclaim_args /* {
1435 struct vnode *a_vp;
1436 } */ *ap = v;
1437 struct vnode *vp = ap->a_vp;
1438 struct rumpfs_node *rn = vp->v_data;
1439
1440 mutex_enter(&reclock);
1441 rn->rn_vp = NULL;
1442 mutex_exit(&reclock);
1443 genfs_node_destroy(vp);
1444 vp->v_data = NULL;
1445
1446 if (rn->rn_flags & RUMPNODE_CANRECLAIM) {
1447 if (vp->v_type == VLNK)
1448 PNBUF_PUT(rn->rn_linktarg);
1449 if (rn->rn_hostpath)
1450 free(rn->rn_hostpath, M_TEMP);
1451 kmem_free(rn, sizeof(*rn));
1452 }
1453
1454 return 0;
1455 }
1456
1457 static int
1458 rump_vop_spec(void *v)
1459 {
1460 struct vop_generic_args *ap = v;
1461 int (**opvec)(void *);
1462
1463 switch (ap->a_desc->vdesc_offset) {
1464 case VOP_ACCESS_DESCOFFSET:
1465 case VOP_GETATTR_DESCOFFSET:
1466 case VOP_SETATTR_DESCOFFSET:
1467 case VOP_LOCK_DESCOFFSET:
1468 case VOP_UNLOCK_DESCOFFSET:
1469 case VOP_ISLOCKED_DESCOFFSET:
1470 case VOP_RECLAIM_DESCOFFSET:
1471 opvec = rump_vnodeop_p;
1472 break;
1473 default:
1474 opvec = spec_vnodeop_p;
1475 break;
1476 }
1477
1478 return VOCALL(opvec, ap->a_desc->vdesc_offset, v);
1479 }
1480
1481 /*
1482 * Begin vfs-level stuff
1483 */
1484
1485 VFS_PROTOS(rumpfs);
1486 struct vfsops rumpfs_vfsops = {
1487 .vfs_name = MOUNT_RUMPFS,
1488 .vfs_min_mount_data = 0,
1489 .vfs_mount = rumpfs_mount,
1490 .vfs_start = (void *)nullop,
1491 .vfs_unmount = rumpfs_unmount,
1492 .vfs_root = rumpfs_root,
1493 .vfs_quotactl = (void *)eopnotsupp,
1494 .vfs_statvfs = genfs_statvfs,
1495 .vfs_sync = (void *)nullop,
1496 .vfs_vget = rumpfs_vget,
1497 .vfs_fhtovp = (void *)eopnotsupp,
1498 .vfs_vptofh = (void *)eopnotsupp,
1499 .vfs_init = rumpfs_init,
1500 .vfs_reinit = NULL,
1501 .vfs_done = rumpfs_done,
1502 .vfs_mountroot = rumpfs_mountroot,
1503 .vfs_snapshot = (void *)eopnotsupp,
1504 .vfs_extattrctl = (void *)eopnotsupp,
1505 .vfs_suspendctl = (void *)eopnotsupp,
1506 .vfs_renamelock_enter = genfs_renamelock_enter,
1507 .vfs_renamelock_exit = genfs_renamelock_exit,
1508 .vfs_opv_descs = rump_opv_descs,
1509 /* vfs_refcount */
1510 /* vfs_list */
1511 };
1512
1513 static int
1514 rumpfs_mountfs(struct mount *mp)
1515 {
1516 struct rumpfs_mount *rfsmp;
1517 struct rumpfs_node *rn;
1518 int error;
1519
1520 rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP);
1521
1522 rn = makeprivate(VDIR, NODEV, DEV_BSIZE, false);
1523 rn->rn_parent = rn;
1524 if ((error = makevnode(mp, rn, &rfsmp->rfsmp_rvp)) != 0)
1525 return error;
1526
1527 rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT;
1528 VOP_UNLOCK(rfsmp->rfsmp_rvp);
1529
1530 mp->mnt_data = rfsmp;
1531 mp->mnt_stat.f_namemax = MAXNAMLEN;
1532 mp->mnt_stat.f_iosize = 512;
1533 mp->mnt_flag |= MNT_LOCAL;
1534 mp->mnt_iflag |= IMNT_MPSAFE;
1535 mp->mnt_fs_bshift = DEV_BSHIFT;
1536 vfs_getnewfsid(mp);
1537
1538 return 0;
1539 }
1540
1541 int
1542 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen)
1543 {
1544 int error;
1545
1546 error = set_statvfs_info(mntpath, UIO_USERSPACE, "rumpfs", UIO_SYSSPACE,
1547 mp->mnt_op->vfs_name, mp, curlwp);
1548 if (error)
1549 return error;
1550
1551 return rumpfs_mountfs(mp);
1552 }
1553
1554 int
1555 rumpfs_unmount(struct mount *mp, int mntflags)
1556 {
1557 struct rumpfs_mount *rfsmp = mp->mnt_data;
1558 int flags = 0, error;
1559
1560 if (panicstr || mntflags & MNT_FORCE)
1561 flags |= FORCECLOSE;
1562
1563 if ((error = vflush(mp, rfsmp->rfsmp_rvp, flags)) != 0)
1564 return error;
1565 vgone(rfsmp->rfsmp_rvp); /* XXX */
1566
1567 kmem_free(rfsmp, sizeof(*rfsmp));
1568
1569 return 0;
1570 }
1571
1572 int
1573 rumpfs_root(struct mount *mp, struct vnode **vpp)
1574 {
1575 struct rumpfs_mount *rfsmp = mp->mnt_data;
1576
1577 vref(rfsmp->rfsmp_rvp);
1578 vn_lock(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY);
1579 *vpp = rfsmp->rfsmp_rvp;
1580 return 0;
1581 }
1582
1583 int
1584 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1585 {
1586
1587 return EOPNOTSUPP;
1588 }
1589
1590 void
1591 rumpfs_init()
1592 {
1593
1594 CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET);
1595
1596 mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE);
1597 mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE);
1598 }
1599
1600 void
1601 rumpfs_done()
1602 {
1603
1604 mutex_destroy(&reclock);
1605 mutex_destroy(&etfs_lock);
1606 }
1607
1608 int
1609 rumpfs_mountroot()
1610 {
1611 struct mount *mp;
1612 int error;
1613
1614 if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) {
1615 vrele(rootvp);
1616 return error;
1617 }
1618
1619 if ((error = rumpfs_mountfs(mp)) != 0)
1620 panic("mounting rootfs failed: %d", error);
1621
1622 mutex_enter(&mountlist_lock);
1623 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1624 mutex_exit(&mountlist_lock);
1625
1626 error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE,
1627 mp->mnt_op->vfs_name, mp, curlwp);
1628 if (error)
1629 panic("set_statvfs_info failed for rootfs: %d", error);
1630
1631 vfs_unbusy(mp, false, NULL);
1632
1633 return 0;
1634 }
1635