1 | /* $NetBSD: puffs_vfsops.c,v 1.118 2015/12/20 01:34:00 christos Exp $ */ |
2 | |
3 | /* |
4 | * Copyright (c) 2005, 2006 Antti Kantee. All Rights Reserved. |
5 | * |
6 | * Development of this software was supported by the |
7 | * Google Summer of Code program and the Ulla Tuominen Foundation. |
8 | * The Google SoC project was mentored by Bill Studenmund. |
9 | * |
10 | * Redistribution and use in source and binary forms, with or without |
11 | * modification, are permitted provided that the following conditions |
12 | * are met: |
13 | * 1. Redistributions of source code must retain the above copyright |
14 | * notice, this list of conditions and the following disclaimer. |
15 | * 2. Redistributions in binary form must reproduce the above copyright |
16 | * notice, this list of conditions and the following disclaimer in the |
17 | * documentation and/or other materials provided with the distribution. |
18 | * |
19 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS |
20 | * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
21 | * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
22 | * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
23 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
24 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
25 | * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
26 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
27 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
28 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
29 | * SUCH DAMAGE. |
30 | */ |
31 | |
32 | #include <sys/cdefs.h> |
33 | __KERNEL_RCSID(0, "$NetBSD: puffs_vfsops.c,v 1.118 2015/12/20 01:34:00 christos Exp $" ); |
34 | |
35 | #include <sys/param.h> |
36 | #include <sys/kernel.h> |
37 | #include <sys/mount.h> |
38 | #include <sys/extattr.h> |
39 | #include <sys/queue.h> |
40 | #include <sys/vnode.h> |
41 | #include <sys/dirent.h> |
42 | #include <sys/kauth.h> |
43 | #include <sys/proc.h> |
44 | #include <sys/module.h> |
45 | #include <sys/kthread.h> |
46 | |
47 | #include <uvm/uvm.h> |
48 | |
49 | #include <dev/putter/putter_sys.h> |
50 | |
51 | #include <miscfs/genfs/genfs.h> |
52 | |
53 | #include <fs/puffs/puffs_msgif.h> |
54 | #include <fs/puffs/puffs_sys.h> |
55 | |
56 | #include <lib/libkern/libkern.h> |
57 | |
58 | #include <nfs/nfsproto.h> /* for fh sizes */ |
59 | |
60 | MODULE(MODULE_CLASS_VFS, puffs, "putter" ); |
61 | |
62 | VFS_PROTOS(puffs_vfsop); |
63 | |
64 | static struct putter_ops puffs_putter = { |
65 | .pop_getout = puffs_msgif_getout, |
66 | .pop_releaseout = puffs_msgif_releaseout, |
67 | .pop_waitcount = puffs_msgif_waitcount, |
68 | .pop_dispatch = puffs_msgif_dispatch, |
69 | .pop_close = puffs_msgif_close, |
70 | }; |
71 | |
72 | static const struct genfs_ops puffs_genfsops = { |
73 | .gop_size = puffs_gop_size, |
74 | .gop_write = genfs_gop_write, |
75 | .gop_markupdate = puffs_gop_markupdate, |
76 | #if 0 |
77 | .gop_alloc, should ask userspace |
78 | #endif |
79 | }; |
80 | |
81 | /* |
82 | * Try to ensure data structures used by the puffs protocol |
83 | * do not unexpectedly change. |
84 | */ |
85 | #if defined(__i386__) && defined(__ELF__) |
86 | CTASSERT(sizeof(struct puffs_kargs) == 3928); |
87 | CTASSERT(sizeof(struct vattr) == 136); |
88 | CTASSERT(sizeof(struct puffs_req) == 44); |
89 | #endif |
90 | |
91 | int |
92 | puffs_vfsop_mount(struct mount *mp, const char *path, void *data, |
93 | size_t *data_len) |
94 | { |
95 | struct puffs_mount *pmp = NULL; |
96 | struct puffs_kargs *args; |
97 | char fstype[_VFS_NAMELEN]; |
98 | char *p; |
99 | int error = 0, i; |
100 | pid_t mntpid = curlwp->l_proc->p_pid; |
101 | |
102 | if (data == NULL) |
103 | return EINVAL; |
104 | if (*data_len < sizeof *args) |
105 | return EINVAL; |
106 | |
107 | if (mp->mnt_flag & MNT_GETARGS) { |
108 | pmp = MPTOPUFFSMP(mp); |
109 | *(struct puffs_kargs *)data = pmp->pmp_args; |
110 | *data_len = sizeof *args; |
111 | return 0; |
112 | } |
113 | |
114 | /* update is not supported currently */ |
115 | if (mp->mnt_flag & MNT_UPDATE) |
116 | return EOPNOTSUPP; |
117 | |
118 | args = (struct puffs_kargs *)data; |
119 | |
120 | if (args->pa_vers != PUFFSVERSION) { |
121 | printf("puffs_mount: development version mismatch: " |
122 | "kernel %d, lib %d\n" , PUFFSVERSION, args->pa_vers); |
123 | error = EINVAL; |
124 | goto out; |
125 | } |
126 | |
127 | if ((args->pa_flags & ~PUFFS_KFLAG_MASK) != 0) { |
128 | printf("puffs_mount: invalid KFLAGs 0x%x\n" , args->pa_flags); |
129 | error = EINVAL; |
130 | goto out; |
131 | } |
132 | if ((args->pa_fhflags & ~PUFFS_FHFLAG_MASK) != 0) { |
133 | printf("puffs_mount: invalid FHFLAGs 0x%x\n" , args->pa_fhflags); |
134 | error = EINVAL; |
135 | goto out; |
136 | } |
137 | |
138 | for (i = 0; i < __arraycount(args->pa_spare); i++) { |
139 | if (args->pa_spare[i] != 0) { |
140 | printf("puffs_mount: pa_spare[%d] = 0x%x\n" , |
141 | i, args->pa_spare[i]); |
142 | error = EINVAL; |
143 | goto out; |
144 | } |
145 | } |
146 | |
147 | /* use dummy value for passthrough */ |
148 | if (args->pa_fhflags & PUFFS_FHFLAG_PASSTHROUGH) |
149 | args->pa_fhsize = sizeof(struct fid); |
150 | |
151 | /* sanitize file handle length */ |
152 | if (PUFFS_TOFHSIZE(args->pa_fhsize) > FHANDLE_SIZE_MAX) { |
153 | printf("puffs_mount: handle size %zu too large\n" , |
154 | args->pa_fhsize); |
155 | error = EINVAL; |
156 | goto out; |
157 | } |
158 | /* sanity check file handle max sizes */ |
159 | if (args->pa_fhsize && args->pa_fhflags & PUFFS_FHFLAG_PROTOMASK) { |
160 | size_t kfhsize = PUFFS_TOFHSIZE(args->pa_fhsize); |
161 | |
162 | if (args->pa_fhflags & PUFFS_FHFLAG_NFSV2) { |
163 | if (NFSX_FHTOOBIG_P(kfhsize, 0)) { |
164 | printf("puffs_mount: fhsize larger than " |
165 | "NFSv2 max %d\n" , |
166 | PUFFS_FROMFHSIZE(NFSX_V2FH)); |
167 | error = EINVAL; |
168 | goto out; |
169 | } |
170 | } |
171 | |
172 | if (args->pa_fhflags & PUFFS_FHFLAG_NFSV3) { |
173 | if (NFSX_FHTOOBIG_P(kfhsize, 1)) { |
174 | printf("puffs_mount: fhsize larger than " |
175 | "NFSv3 max %d\n" , |
176 | PUFFS_FROMFHSIZE(NFSX_V3FHMAX)); |
177 | error = EINVAL; |
178 | goto out; |
179 | } |
180 | } |
181 | } |
182 | |
183 | /* don't allow non-printing characters (like my sweet umlauts.. snif) */ |
184 | args->pa_typename[sizeof(args->pa_typename)-1] = '\0'; |
185 | for (p = args->pa_typename; *p; p++) |
186 | if (*p < ' ' || *p > '~') |
187 | *p = '.'; |
188 | |
189 | args->pa_mntfromname[sizeof(args->pa_mntfromname)-1] = '\0'; |
190 | for (p = args->pa_mntfromname; *p; p++) |
191 | if (*p < ' ' || *p > '~') |
192 | *p = '.'; |
193 | |
194 | /* build real name */ |
195 | (void)strlcpy(fstype, PUFFS_TYPEPREFIX, sizeof(fstype)); |
196 | (void)strlcat(fstype, args->pa_typename, sizeof(fstype)); |
197 | |
198 | /* inform user server if it got the max request size it wanted */ |
199 | if (args->pa_maxmsglen == 0 || args->pa_maxmsglen > PUFFS_MSG_MAXSIZE) |
200 | args->pa_maxmsglen = PUFFS_MSG_MAXSIZE; |
201 | else if (args->pa_maxmsglen < 2*PUFFS_MSGSTRUCT_MAX) |
202 | args->pa_maxmsglen = 2*PUFFS_MSGSTRUCT_MAX; |
203 | |
204 | (void)strlcpy(args->pa_typename, fstype, sizeof(args->pa_typename)); |
205 | |
206 | error = set_statvfs_info(path, UIO_USERSPACE, args->pa_mntfromname, |
207 | UIO_SYSSPACE, fstype, mp, curlwp); |
208 | if (error) |
209 | goto out; |
210 | mp->mnt_stat.f_iosize = DEV_BSIZE; |
211 | mp->mnt_stat.f_namemax = args->pa_svfsb.f_namemax; |
212 | |
213 | /* |
214 | * We can't handle the VFS_STATVFS() mount_domount() does |
215 | * after VFS_MOUNT() because we'd deadlock, so handle it |
216 | * here already. |
217 | */ |
218 | copy_statvfs_info(&args->pa_svfsb, mp); |
219 | (void)memcpy(&mp->mnt_stat, &args->pa_svfsb, sizeof(mp->mnt_stat)); |
220 | |
221 | KASSERT(curlwp != uvm.pagedaemon_lwp); |
222 | pmp = kmem_zalloc(sizeof(struct puffs_mount), KM_SLEEP); |
223 | |
224 | mp->mnt_fs_bshift = DEV_BSHIFT; |
225 | mp->mnt_dev_bshift = DEV_BSHIFT; |
226 | mp->mnt_flag &= ~MNT_LOCAL; /* we don't really know, so ... */ |
227 | mp->mnt_data = pmp; |
228 | |
229 | #if 0 |
230 | /* |
231 | * XXX: puffs code is MPSAFE. However, VFS really isn't. |
232 | * Currently, there is nothing which protects an inode from |
233 | * reclaim while there are threads inside the file system. |
234 | * This means that in the event of a server crash, an MPSAFE |
235 | * mount is likely to end up accessing invalid memory. For the |
236 | * non-mpsafe case, the kernel lock, general structure of |
237 | * puffs and pmp_refcount protect the threads during escape. |
238 | * |
239 | * Fixing this will require: |
240 | * a) fixing vfs |
241 | * OR |
242 | * b) adding a small sleep to puffs_msgif_close() between |
243 | * userdead() and dounmount(). |
244 | * (well, this isn't really a fix, but would solve |
245 | * 99.999% of the race conditions). |
246 | * |
247 | * Also, in the event of "b", unmount -f should be used, |
248 | * like with any other file system, sparingly and only when |
249 | * it is "known" to be safe. |
250 | */ |
251 | mp->mnt_iflags |= IMNT_MPSAFE; |
252 | #endif |
253 | |
254 | pmp->pmp_status = PUFFSTAT_MOUNTING; |
255 | pmp->pmp_mp = mp; |
256 | pmp->pmp_msg_maxsize = args->pa_maxmsglen; |
257 | pmp->pmp_args = *args; |
258 | |
259 | /* |
260 | * Inform the fileops processing code that we have a mountpoint. |
261 | * If it doesn't know about anyone with our pid/fd having the |
262 | * device open, punt |
263 | */ |
264 | if ((pmp->pmp_pi |
265 | = putter_attach(mntpid, args->pa_fd, pmp, &puffs_putter)) == NULL) { |
266 | error = ENOENT; |
267 | goto out; |
268 | } |
269 | |
270 | /* XXX: check parameters */ |
271 | pmp->pmp_root_cookie = args->pa_root_cookie; |
272 | switch (args->pa_root_vtype) { |
273 | case VNON: case VREG: case VDIR: case VBLK: |
274 | case VCHR: case VLNK: case VSOCK: case VFIFO: |
275 | break; |
276 | default: |
277 | error = EINVAL; |
278 | goto out; |
279 | } |
280 | pmp->pmp_root_vtype = args->pa_root_vtype; |
281 | |
282 | if (args->pa_root_vsize < 0) { |
283 | error = EINVAL; |
284 | goto out; |
285 | } |
286 | pmp->pmp_root_vsize = args->pa_root_vsize; |
287 | |
288 | pmp->pmp_root_rdev = args->pa_root_rdev; |
289 | pmp->pmp_docompat = args->pa_time32; |
290 | |
291 | mutex_init(&pmp->pmp_lock, MUTEX_DEFAULT, IPL_NONE); |
292 | mutex_init(&pmp->pmp_sopmtx, MUTEX_DEFAULT, IPL_NONE); |
293 | cv_init(&pmp->pmp_msg_waiter_cv, "puffsget" ); |
294 | cv_init(&pmp->pmp_refcount_cv, "puffsref" ); |
295 | cv_init(&pmp->pmp_unmounting_cv, "puffsum" ); |
296 | cv_init(&pmp->pmp_sopcv, "puffsop" ); |
297 | TAILQ_INIT(&pmp->pmp_msg_touser); |
298 | TAILQ_INIT(&pmp->pmp_msg_replywait); |
299 | TAILQ_INIT(&pmp->pmp_sopfastreqs); |
300 | TAILQ_INIT(&pmp->pmp_sopnodereqs); |
301 | |
302 | if ((error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, |
303 | puffs_sop_thread, pmp, NULL, "puffsop" )) != 0) |
304 | goto out; |
305 | pmp->pmp_sopthrcount = 1; |
306 | |
307 | DPRINTF(("puffs_mount: mount point at %p, puffs specific at %p\n" , |
308 | mp, MPTOPUFFSMP(mp))); |
309 | |
310 | vfs_getnewfsid(mp); |
311 | |
312 | out: |
313 | if (error && pmp && pmp->pmp_pi) |
314 | putter_detach(pmp->pmp_pi); |
315 | if (error && pmp) |
316 | kmem_free(pmp, sizeof(struct puffs_mount)); |
317 | return error; |
318 | } |
319 | |
320 | int |
321 | puffs_vfsop_start(struct mount *mp, int flags) |
322 | { |
323 | struct puffs_mount *pmp = MPTOPUFFSMP(mp); |
324 | |
325 | KASSERT(pmp->pmp_status == PUFFSTAT_MOUNTING); |
326 | pmp->pmp_status = PUFFSTAT_RUNNING; |
327 | |
328 | return 0; |
329 | } |
330 | |
331 | int |
332 | puffs_vfsop_unmount(struct mount *mp, int mntflags) |
333 | { |
334 | PUFFS_MSG_VARS(vfs, unmount); |
335 | struct puffs_mount *pmp; |
336 | int error, force; |
337 | |
338 | error = 0; |
339 | force = mntflags & MNT_FORCE; |
340 | pmp = MPTOPUFFSMP(mp); |
341 | |
342 | DPRINTF(("puffs_unmount: detach filesystem from vfs, current " |
343 | "status 0x%x\n" , pmp->pmp_status)); |
344 | |
345 | /* |
346 | * flush all the vnodes. VOP_RECLAIM() takes care that the |
347 | * root vnode does not get flushed until unmount. The |
348 | * userspace root node cookie is stored in the mount |
349 | * structure, so we can always re-instantiate a root vnode, |
350 | * should userspace unmount decide it doesn't want to |
351 | * cooperate. |
352 | */ |
353 | error = vflush(mp, NULLVP, force ? FORCECLOSE : 0); |
354 | if (error) |
355 | goto out; |
356 | |
357 | /* |
358 | * If we are not DYING, we should ask userspace's opinion |
359 | * about the situation |
360 | */ |
361 | mutex_enter(&pmp->pmp_lock); |
362 | if (pmp->pmp_status != PUFFSTAT_DYING) { |
363 | pmp->pmp_unmounting = 1; |
364 | mutex_exit(&pmp->pmp_lock); |
365 | |
366 | PUFFS_MSG_ALLOC(vfs, unmount); |
367 | puffs_msg_setinfo(park_unmount, |
368 | PUFFSOP_VFS, PUFFS_VFS_UNMOUNT, NULL); |
369 | unmount_msg->pvfsr_flags = mntflags; |
370 | |
371 | PUFFS_MSG_ENQUEUEWAIT(pmp, park_unmount, error); |
372 | PUFFS_MSG_RELEASE(unmount); |
373 | |
374 | error = checkerr(pmp, error, __func__); |
375 | DPRINTF(("puffs_unmount: error %d force %d\n" , error, force)); |
376 | |
377 | mutex_enter(&pmp->pmp_lock); |
378 | pmp->pmp_unmounting = 0; |
379 | cv_broadcast(&pmp->pmp_unmounting_cv); |
380 | } |
381 | |
382 | /* |
383 | * if userspace cooperated or we really need to die, |
384 | * screw what userland thinks and just die. |
385 | */ |
386 | if (error == 0 || force) { |
387 | struct puffs_sopreq *psopr; |
388 | |
389 | /* tell waiters & other resources to go unwait themselves */ |
390 | puffs_userdead(pmp); |
391 | putter_detach(pmp->pmp_pi); |
392 | |
393 | /* |
394 | * Wait until there are no more users for the mount resource. |
395 | * Notice that this is hooked against transport_close |
396 | * and return from touser. In an ideal world, it would |
397 | * be hooked against final return from all operations. |
398 | * But currently it works well enough, since nobody |
399 | * does weird blocking voodoo after return from touser(). |
400 | */ |
401 | while (pmp->pmp_refcount != 0) |
402 | cv_wait(&pmp->pmp_refcount_cv, &pmp->pmp_lock); |
403 | mutex_exit(&pmp->pmp_lock); |
404 | |
405 | /* |
406 | * Release kernel thread now that there is nothing |
407 | * it would be wanting to lock. |
408 | */ |
409 | KASSERT(curlwp != uvm.pagedaemon_lwp); |
410 | psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP); |
411 | psopr->psopr_sopreq = PUFFS_SOPREQSYS_EXIT; |
412 | mutex_enter(&pmp->pmp_sopmtx); |
413 | if (pmp->pmp_sopthrcount == 0) { |
414 | mutex_exit(&pmp->pmp_sopmtx); |
415 | kmem_free(psopr, sizeof(*psopr)); |
416 | mutex_enter(&pmp->pmp_sopmtx); |
417 | KASSERT(pmp->pmp_sopthrcount == 0); |
418 | } else { |
419 | TAILQ_INSERT_TAIL(&pmp->pmp_sopfastreqs, |
420 | psopr, psopr_entries); |
421 | cv_signal(&pmp->pmp_sopcv); |
422 | } |
423 | while (pmp->pmp_sopthrcount > 0) |
424 | cv_wait(&pmp->pmp_sopcv, &pmp->pmp_sopmtx); |
425 | mutex_exit(&pmp->pmp_sopmtx); |
426 | |
427 | /* free resources now that we hopefully have no waiters left */ |
428 | cv_destroy(&pmp->pmp_unmounting_cv); |
429 | cv_destroy(&pmp->pmp_refcount_cv); |
430 | cv_destroy(&pmp->pmp_msg_waiter_cv); |
431 | cv_destroy(&pmp->pmp_sopcv); |
432 | mutex_destroy(&pmp->pmp_lock); |
433 | mutex_destroy(&pmp->pmp_sopmtx); |
434 | |
435 | kmem_free(pmp, sizeof(struct puffs_mount)); |
436 | error = 0; |
437 | } else { |
438 | mutex_exit(&pmp->pmp_lock); |
439 | } |
440 | |
441 | out: |
442 | DPRINTF(("puffs_unmount: return %d\n" , error)); |
443 | return error; |
444 | } |
445 | |
446 | /* |
447 | * This doesn't need to travel to userspace |
448 | */ |
449 | int |
450 | puffs_vfsop_root(struct mount *mp, struct vnode **vpp) |
451 | { |
452 | struct puffs_mount *pmp = MPTOPUFFSMP(mp); |
453 | int rv; |
454 | |
455 | rv = puffs_cookie2vnode(pmp, pmp->pmp_root_cookie, vpp); |
456 | KASSERT(rv != PUFFS_NOSUCHCOOKIE); |
457 | if (rv != 0) |
458 | return rv; |
459 | rv = vn_lock(*vpp, LK_EXCLUSIVE); |
460 | if (rv != 0) { |
461 | vrele(*vpp); |
462 | *vpp = NULL; |
463 | return rv; |
464 | } |
465 | return 0; |
466 | } |
467 | |
468 | int |
469 | puffs_vfsop_statvfs(struct mount *mp, struct statvfs *sbp) |
470 | { |
471 | PUFFS_MSG_VARS(vfs, statvfs); |
472 | struct puffs_mount *pmp; |
473 | int error = 0; |
474 | |
475 | pmp = MPTOPUFFSMP(mp); |
476 | |
477 | /* |
478 | * If we are mounting, it means that the userspace counterpart |
479 | * is calling mount(2), but mount(2) also calls statvfs. So |
480 | * requesting statvfs from userspace would mean a deadlock. |
481 | * Compensate. |
482 | */ |
483 | if (__predict_false(pmp->pmp_status == PUFFSTAT_MOUNTING)) |
484 | return EINPROGRESS; |
485 | |
486 | PUFFS_MSG_ALLOC(vfs, statvfs); |
487 | puffs_msg_setinfo(park_statvfs, PUFFSOP_VFS, PUFFS_VFS_STATVFS, NULL); |
488 | |
489 | PUFFS_MSG_ENQUEUEWAIT(pmp, park_statvfs, error); |
490 | error = checkerr(pmp, error, __func__); |
491 | statvfs_msg->pvfsr_sb.f_iosize = DEV_BSIZE; |
492 | |
493 | /* |
494 | * Try to produce a sensible result even in the event |
495 | * of userspace error. |
496 | * |
497 | * XXX: cache the copy in non-error case |
498 | */ |
499 | if (!error) { |
500 | copy_statvfs_info(&statvfs_msg->pvfsr_sb, mp); |
501 | (void)memcpy(sbp, &statvfs_msg->pvfsr_sb, |
502 | sizeof(struct statvfs)); |
503 | } else { |
504 | copy_statvfs_info(sbp, mp); |
505 | } |
506 | |
507 | PUFFS_MSG_RELEASE(statvfs); |
508 | return error; |
509 | } |
510 | |
511 | static bool |
512 | pageflush_selector(void *cl, struct vnode *vp) |
513 | { |
514 | return vp->v_type == VREG && |
515 | !(LIST_EMPTY(&vp->v_dirtyblkhd) && UVM_OBJ_IS_CLEAN(&vp->v_uobj)); |
516 | } |
517 | |
518 | static int |
519 | pageflush(struct mount *mp, kauth_cred_t cred, int waitfor) |
520 | { |
521 | struct puffs_node *pn; |
522 | struct vnode *vp; |
523 | struct vnode_iterator *marker; |
524 | int error, rv, fsyncwait; |
525 | |
526 | error = 0; |
527 | fsyncwait = (waitfor == MNT_WAIT) ? FSYNC_WAIT : 0; |
528 | |
529 | /* |
530 | * Sync all cached data from regular vnodes (which are not |
531 | * currently locked, see below). After this we call VFS_SYNC |
532 | * for the fs server, which should handle data and metadata for |
533 | * all the nodes it knows to exist. |
534 | */ |
535 | vfs_vnode_iterator_init(mp, &marker); |
536 | while ((vp = vfs_vnode_iterator_next(marker, pageflush_selector, |
537 | NULL))) |
538 | { |
539 | /* |
540 | * Here we try to get a reference to the vnode and to |
541 | * lock it. This is mostly cargo-culted, but I will |
542 | * offer an explanation to why I believe this might |
543 | * actually do the right thing. |
544 | * |
545 | * If the vnode is a goner, we quite obviously don't need |
546 | * to sync it. |
547 | * |
548 | * If the vnode was busy, we don't need to sync it because |
549 | * this is never called with MNT_WAIT except from |
550 | * dounmount(), when we are wait-flushing all the dirty |
551 | * vnodes through other routes in any case. So there, |
552 | * sync() doesn't actually sync. Happy now? |
553 | */ |
554 | error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); |
555 | if (error) { |
556 | vrele(vp); |
557 | continue; |
558 | } |
559 | pn = VPTOPP(vp); |
560 | /* hmm.. is the FAF thing entirely sensible? */ |
561 | if (waitfor == MNT_LAZY) { |
562 | mutex_enter(vp->v_interlock); |
563 | pn->pn_stat |= PNODE_FAF; |
564 | mutex_exit(vp->v_interlock); |
565 | } |
566 | rv = VOP_FSYNC(vp, cred, fsyncwait, 0, 0); |
567 | if (waitfor == MNT_LAZY) { |
568 | mutex_enter(vp->v_interlock); |
569 | pn->pn_stat &= ~PNODE_FAF; |
570 | mutex_exit(vp->v_interlock); |
571 | } |
572 | if (rv) |
573 | error = rv; |
574 | vput(vp); |
575 | } |
576 | vfs_vnode_iterator_destroy(marker); |
577 | |
578 | return error; |
579 | } |
580 | |
581 | int |
582 | puffs_vfsop_sync(struct mount *mp, int waitfor, struct kauth_cred *cred) |
583 | { |
584 | PUFFS_MSG_VARS(vfs, sync); |
585 | struct puffs_mount *pmp = MPTOPUFFSMP(mp); |
586 | int error, rv; |
587 | |
588 | error = pageflush(mp, cred, waitfor); |
589 | |
590 | /* sync fs */ |
591 | PUFFS_MSG_ALLOC(vfs, sync); |
592 | sync_msg->pvfsr_waitfor = waitfor; |
593 | puffs_credcvt(&sync_msg->pvfsr_cred, cred); |
594 | puffs_msg_setinfo(park_sync, PUFFSOP_VFS, PUFFS_VFS_SYNC, NULL); |
595 | |
596 | PUFFS_MSG_ENQUEUEWAIT(pmp, park_sync, rv); |
597 | rv = checkerr(pmp, rv, __func__); |
598 | if (rv) |
599 | error = rv; |
600 | |
601 | PUFFS_MSG_RELEASE(sync); |
602 | return error; |
603 | } |
604 | |
605 | int |
606 | puffs_vfsop_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp) |
607 | { |
608 | PUFFS_MSG_VARS(vfs, fhtonode); |
609 | struct puffs_mount *pmp = MPTOPUFFSMP(mp); |
610 | struct vnode *vp; |
611 | void *fhdata; |
612 | size_t argsize, fhlen; |
613 | int error; |
614 | |
615 | if (pmp->pmp_args.pa_fhsize == 0) |
616 | return EOPNOTSUPP; |
617 | |
618 | if (pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_PASSTHROUGH) { |
619 | fhlen = fhp->fid_len; |
620 | fhdata = fhp; |
621 | } else { |
622 | fhlen = PUFFS_FROMFHSIZE(fhp->fid_len); |
623 | fhdata = fhp->fid_data; |
624 | |
625 | if (pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_DYNAMIC) { |
626 | if (pmp->pmp_args.pa_fhsize < fhlen) |
627 | return EINVAL; |
628 | } else { |
629 | if (pmp->pmp_args.pa_fhsize != fhlen) |
630 | return EINVAL; |
631 | } |
632 | } |
633 | |
634 | argsize = sizeof(struct puffs_vfsmsg_fhtonode) + fhlen; |
635 | puffs_msgmem_alloc(argsize, &park_fhtonode, (void *)&fhtonode_msg, 1); |
636 | fhtonode_msg->pvfsr_dsize = fhlen; |
637 | memcpy(fhtonode_msg->pvfsr_data, fhdata, fhlen); |
638 | puffs_msg_setinfo(park_fhtonode, PUFFSOP_VFS, PUFFS_VFS_FHTOVP, NULL); |
639 | |
640 | PUFFS_MSG_ENQUEUEWAIT(pmp, park_fhtonode, error); |
641 | error = checkerr(pmp, error, __func__); |
642 | if (error) |
643 | goto out; |
644 | |
645 | error = puffs_getvnode(mp, fhtonode_msg->pvfsr_fhcookie, |
646 | fhtonode_msg->pvfsr_vtype, fhtonode_msg->pvfsr_size, |
647 | fhtonode_msg->pvfsr_rdev, &vp); |
648 | if (error) |
649 | goto out; |
650 | vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); |
651 | |
652 | *vpp = vp; |
653 | out: |
654 | puffs_msgmem_release(park_fhtonode); |
655 | return error; |
656 | } |
657 | |
658 | int |
659 | puffs_vfsop_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size) |
660 | { |
661 | PUFFS_MSG_VARS(vfs, nodetofh); |
662 | struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); |
663 | size_t argsize, fhlen; |
664 | int error; |
665 | |
666 | if (pmp->pmp_args.pa_fhsize == 0) |
667 | return EOPNOTSUPP; |
668 | |
669 | /* if file handles are static len, we can test len immediately */ |
670 | if (((pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_DYNAMIC) == 0) |
671 | && ((pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_PASSTHROUGH) == 0) |
672 | && (PUFFS_FROMFHSIZE(*fh_size) < pmp->pmp_args.pa_fhsize)) { |
673 | *fh_size = PUFFS_TOFHSIZE(pmp->pmp_args.pa_fhsize); |
674 | return E2BIG; |
675 | } |
676 | |
677 | if (pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_PASSTHROUGH) |
678 | fhlen = *fh_size; |
679 | else |
680 | fhlen = PUFFS_FROMFHSIZE(*fh_size); |
681 | |
682 | argsize = sizeof(struct puffs_vfsmsg_nodetofh) + fhlen; |
683 | puffs_msgmem_alloc(argsize, &park_nodetofh, (void *)&nodetofh_msg, 1); |
684 | nodetofh_msg->pvfsr_fhcookie = VPTOPNC(vp); |
685 | nodetofh_msg->pvfsr_dsize = fhlen; |
686 | puffs_msg_setinfo(park_nodetofh, PUFFSOP_VFS, PUFFS_VFS_VPTOFH, NULL); |
687 | |
688 | PUFFS_MSG_ENQUEUEWAIT(pmp, park_nodetofh, error); |
689 | error = checkerr(pmp, error, __func__); |
690 | |
691 | if (pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_PASSTHROUGH) |
692 | fhlen = nodetofh_msg->pvfsr_dsize; |
693 | else if (pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_DYNAMIC) |
694 | fhlen = PUFFS_TOFHSIZE(nodetofh_msg->pvfsr_dsize); |
695 | else |
696 | fhlen = PUFFS_TOFHSIZE(pmp->pmp_args.pa_fhsize); |
697 | |
698 | if (error) { |
699 | if (error == E2BIG) |
700 | *fh_size = fhlen; |
701 | goto out; |
702 | } |
703 | |
704 | if (fhlen > FHANDLE_SIZE_MAX) { |
705 | puffs_senderr(pmp, PUFFS_ERR_VPTOFH, E2BIG, |
706 | "file handle too big" , VPTOPNC(vp)); |
707 | error = EPROTO; |
708 | goto out; |
709 | } |
710 | |
711 | if (*fh_size < fhlen) { |
712 | *fh_size = fhlen; |
713 | error = E2BIG; |
714 | goto out; |
715 | } |
716 | *fh_size = fhlen; |
717 | |
718 | if (fhp) { |
719 | if (pmp->pmp_args.pa_fhflags & PUFFS_FHFLAG_PASSTHROUGH) { |
720 | memcpy(fhp, nodetofh_msg->pvfsr_data, fhlen); |
721 | } else { |
722 | fhp->fid_len = *fh_size; |
723 | memcpy(fhp->fid_data, nodetofh_msg->pvfsr_data, |
724 | nodetofh_msg->pvfsr_dsize); |
725 | } |
726 | } |
727 | |
728 | out: |
729 | puffs_msgmem_release(park_nodetofh); |
730 | return error; |
731 | } |
732 | |
733 | int |
734 | puffs_vfsop_loadvnode(struct mount *mp, struct vnode *vp, |
735 | const void *key, size_t key_len, const void **new_key) |
736 | { |
737 | struct puffs_mount *pmp; |
738 | struct puffs_node *pnode; |
739 | |
740 | KASSERT(key_len == sizeof(puffs_cookie_t)); |
741 | |
742 | pmp = MPTOPUFFSMP(mp); |
743 | |
744 | /* Allocate and initialize the pnode. */ |
745 | pnode = pool_get(&puffs_pnpool, PR_WAITOK); |
746 | memset(pnode, 0, sizeof(struct puffs_node)); |
747 | |
748 | pnode->pn_vp = vp; |
749 | memcpy(&pnode->pn_cookie, key, key_len); |
750 | pnode->pn_refcount = 1; |
751 | mutex_init(&pnode->pn_mtx, MUTEX_DEFAULT, IPL_NONE); |
752 | mutex_init(&pnode->pn_sizemtx, MUTEX_DEFAULT, IPL_NONE); |
753 | selinit(&pnode->pn_sel); |
754 | vp->v_tag = VT_PUFFS; |
755 | vp->v_type = VNON; |
756 | vp->v_op = puffs_vnodeop_p; |
757 | if (pnode->pn_cookie == pmp->pmp_root_cookie) |
758 | vp->v_vflag |= VV_ROOT; |
759 | vp->v_data = pnode; |
760 | |
761 | genfs_node_init(vp, &puffs_genfsops); |
762 | uvm_vnp_setsize(vp, 0); |
763 | |
764 | *new_key = &pnode->pn_cookie; |
765 | return 0; |
766 | } |
767 | |
768 | void |
769 | puffs_vfsop_init(void) |
770 | { |
771 | |
772 | /* some checks depend on this */ |
773 | KASSERT(VNOVAL == VSIZENOTSET); |
774 | |
775 | pool_init(&puffs_pnpool, sizeof(struct puffs_node), 0, 0, 0, |
776 | "puffpnpl" , &pool_allocator_nointr, IPL_NONE); |
777 | pool_init(&puffs_vapool, sizeof(struct vattr), 0, 0, 0, |
778 | "puffvapl" , &pool_allocator_nointr, IPL_NONE); |
779 | puffs_msgif_init(); |
780 | } |
781 | |
782 | void |
783 | puffs_vfsop_done(void) |
784 | { |
785 | |
786 | puffs_msgif_destroy(); |
787 | pool_destroy(&puffs_pnpool); |
788 | pool_destroy(&puffs_vapool); |
789 | } |
790 | |
791 | int |
792 | puffs_vfsop_snapshot(struct mount *mp, struct vnode *vp, struct timespec *ts) |
793 | { |
794 | |
795 | return EOPNOTSUPP; |
796 | } |
797 | |
798 | int |
799 | puffs_vfsop_extattrctl(struct mount *mp, int cmd, struct vnode *vp, |
800 | int attrnamespace, const char *attrname) |
801 | { |
802 | PUFFS_MSG_VARS(vfs, extattrctl); |
803 | struct puffs_mount *pmp = MPTOPUFFSMP(mp); |
804 | struct puffs_node *pnp; |
805 | puffs_cookie_t pnc; |
806 | int error, flags; |
807 | |
808 | if (vp) { |
809 | /* doesn't make sense for puffs servers */ |
810 | if (vp->v_mount != mp) |
811 | return EXDEV; |
812 | pnp = vp->v_data; |
813 | pnc = pnp->pn_cookie; |
814 | flags = PUFFS_EXTATTRCTL_HASNODE; |
815 | } else { |
816 | pnp = pnc = NULL; |
817 | flags = 0; |
818 | } |
819 | |
820 | PUFFS_MSG_ALLOC(vfs, extattrctl); |
821 | extattrctl_msg->pvfsr_cmd = cmd; |
822 | extattrctl_msg->pvfsr_attrnamespace = attrnamespace; |
823 | extattrctl_msg->pvfsr_flags = flags; |
824 | if (attrname) { |
825 | strlcpy(extattrctl_msg->pvfsr_attrname, attrname, |
826 | sizeof(extattrctl_msg->pvfsr_attrname)); |
827 | extattrctl_msg->pvfsr_flags |= PUFFS_EXTATTRCTL_HASATTRNAME; |
828 | } |
829 | puffs_msg_setinfo(park_extattrctl, |
830 | PUFFSOP_VFS, PUFFS_VFS_EXTATTRCTL, pnc); |
831 | |
832 | puffs_msg_enqueue(pmp, park_extattrctl); |
833 | if (vp) { |
834 | mutex_enter(&pnp->pn_mtx); |
835 | puffs_referencenode(pnp); |
836 | mutex_exit(&pnp->pn_mtx); |
837 | VOP_UNLOCK(vp); |
838 | } |
839 | error = puffs_msg_wait2(pmp, park_extattrctl, pnp, NULL); |
840 | PUFFS_MSG_RELEASE(extattrctl); |
841 | if (vp) { |
842 | puffs_releasenode(pnp); |
843 | } |
844 | |
845 | return checkerr(pmp, error, __func__); |
846 | } |
847 | |
848 | const struct vnodeopv_desc * const puffs_vnodeopv_descs[] = { |
849 | &puffs_vnodeop_opv_desc, |
850 | &puffs_specop_opv_desc, |
851 | &puffs_fifoop_opv_desc, |
852 | &puffs_msgop_opv_desc, |
853 | NULL, |
854 | }; |
855 | |
856 | struct vfsops puffs_vfsops = { |
857 | .vfs_name = MOUNT_PUFFS, |
858 | .vfs_min_mount_data = sizeof (struct puffs_kargs), |
859 | .vfs_mount = puffs_vfsop_mount, |
860 | .vfs_start = puffs_vfsop_start, |
861 | .vfs_unmount = puffs_vfsop_unmount, |
862 | .vfs_root = puffs_vfsop_root, |
863 | .vfs_quotactl = (void *)eopnotsupp, |
864 | .vfs_statvfs = puffs_vfsop_statvfs, |
865 | .vfs_sync = puffs_vfsop_sync, |
866 | .vfs_vget = (void *)eopnotsupp, |
867 | .vfs_loadvnode = puffs_vfsop_loadvnode, |
868 | .vfs_fhtovp = puffs_vfsop_fhtovp, |
869 | .vfs_vptofh = puffs_vfsop_vptofh, |
870 | .vfs_init = puffs_vfsop_init, |
871 | .vfs_done = puffs_vfsop_done, |
872 | .vfs_snapshot = puffs_vfsop_snapshot, |
873 | .vfs_extattrctl = puffs_vfsop_extattrctl, |
874 | .vfs_suspendctl = (void *)eopnotsupp, |
875 | .vfs_renamelock_enter = genfs_renamelock_enter, |
876 | .vfs_renamelock_exit = genfs_renamelock_exit, |
877 | .vfs_fsync = (void *)eopnotsupp, |
878 | .vfs_opv_descs = puffs_vnodeopv_descs |
879 | }; |
880 | |
881 | static int |
882 | puffs_modcmd(modcmd_t cmd, void *arg) |
883 | { |
884 | |
885 | switch (cmd) { |
886 | case MODULE_CMD_INIT: |
887 | return vfs_attach(&puffs_vfsops); |
888 | case MODULE_CMD_FINI: |
889 | return vfs_detach(&puffs_vfsops); |
890 | default: |
891 | return ENOTTY; |
892 | } |
893 | } |
894 | |