nfs_vnops.c revision 1.294 1 /* $NetBSD: nfs_vnops.c,v 1.294 2012/04/27 18:12:01 drochner Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
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 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)nfs_vnops.c 8.19 (Berkeley) 7/31/95
35 */
36
37 /*
38 * vnode op calls for Sun NFS version 2 and 3
39 */
40
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.294 2012/04/27 18:12:01 drochner Exp $");
43
44 #ifdef _KERNEL_OPT
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 #endif
48
49 #include <sys/param.h>
50 #include <sys/proc.h>
51 #include <sys/kernel.h>
52 #include <sys/systm.h>
53 #include <sys/resourcevar.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/condvar.h>
57 #include <sys/disk.h>
58 #include <sys/malloc.h>
59 #include <sys/kmem.h>
60 #include <sys/mbuf.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/vnode.h>
64 #include <sys/dirent.h>
65 #include <sys/fcntl.h>
66 #include <sys/hash.h>
67 #include <sys/lockf.h>
68 #include <sys/stat.h>
69 #include <sys/unistd.h>
70 #include <sys/kauth.h>
71 #include <sys/cprng.h>
72
73 #include <uvm/uvm_extern.h>
74 #include <uvm/uvm.h>
75
76 #include <miscfs/fifofs/fifo.h>
77 #include <miscfs/genfs/genfs.h>
78 #include <miscfs/genfs/genfs_node.h>
79 #include <miscfs/specfs/specdev.h>
80
81 #include <nfs/rpcv2.h>
82 #include <nfs/nfsproto.h>
83 #include <nfs/nfs.h>
84 #include <nfs/nfsnode.h>
85 #include <nfs/nfsmount.h>
86 #include <nfs/xdr_subs.h>
87 #include <nfs/nfsm_subs.h>
88 #include <nfs/nfs_var.h>
89
90 #include <net/if.h>
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93
94 /*
95 * Global vfs data structures for nfs
96 */
97 int (**nfsv2_vnodeop_p)(void *);
98 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
99 { &vop_default_desc, vn_default_error },
100 { &vop_lookup_desc, nfs_lookup }, /* lookup */
101 { &vop_create_desc, nfs_create }, /* create */
102 { &vop_mknod_desc, nfs_mknod }, /* mknod */
103 { &vop_open_desc, nfs_open }, /* open */
104 { &vop_close_desc, nfs_close }, /* close */
105 { &vop_access_desc, nfs_access }, /* access */
106 { &vop_getattr_desc, nfs_getattr }, /* getattr */
107 { &vop_setattr_desc, nfs_setattr }, /* setattr */
108 { &vop_read_desc, nfs_read }, /* read */
109 { &vop_write_desc, nfs_write }, /* write */
110 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
111 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
112 { &vop_poll_desc, nfs_poll }, /* poll */
113 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
114 { &vop_revoke_desc, nfs_revoke }, /* revoke */
115 { &vop_mmap_desc, nfs_mmap }, /* mmap */
116 { &vop_fsync_desc, nfs_fsync }, /* fsync */
117 { &vop_seek_desc, nfs_seek }, /* seek */
118 { &vop_remove_desc, nfs_remove }, /* remove */
119 { &vop_link_desc, nfs_link }, /* link */
120 { &vop_rename_desc, nfs_rename }, /* rename */
121 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
122 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
123 { &vop_symlink_desc, nfs_symlink }, /* symlink */
124 { &vop_readdir_desc, nfs_readdir }, /* readdir */
125 { &vop_readlink_desc, nfs_readlink }, /* readlink */
126 { &vop_abortop_desc, nfs_abortop }, /* abortop */
127 { &vop_inactive_desc, nfs_inactive }, /* inactive */
128 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
129 { &vop_lock_desc, nfs_lock }, /* lock */
130 { &vop_unlock_desc, nfs_unlock }, /* unlock */
131 { &vop_bmap_desc, nfs_bmap }, /* bmap */
132 { &vop_strategy_desc, nfs_strategy }, /* strategy */
133 { &vop_print_desc, nfs_print }, /* print */
134 { &vop_islocked_desc, nfs_islocked }, /* islocked */
135 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
136 { &vop_advlock_desc, nfs_advlock }, /* advlock */
137 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
138 { &vop_getpages_desc, nfs_getpages }, /* getpages */
139 { &vop_putpages_desc, genfs_putpages }, /* putpages */
140 { NULL, NULL }
141 };
142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
144
145 /*
146 * Special device vnode ops
147 */
148 int (**spec_nfsv2nodeop_p)(void *);
149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
150 { &vop_default_desc, vn_default_error },
151 { &vop_lookup_desc, spec_lookup }, /* lookup */
152 { &vop_create_desc, spec_create }, /* create */
153 { &vop_mknod_desc, spec_mknod }, /* mknod */
154 { &vop_open_desc, spec_open }, /* open */
155 { &vop_close_desc, nfsspec_close }, /* close */
156 { &vop_access_desc, nfsspec_access }, /* access */
157 { &vop_getattr_desc, nfs_getattr }, /* getattr */
158 { &vop_setattr_desc, nfs_setattr }, /* setattr */
159 { &vop_read_desc, nfsspec_read }, /* read */
160 { &vop_write_desc, nfsspec_write }, /* write */
161 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
162 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
163 { &vop_poll_desc, spec_poll }, /* poll */
164 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
165 { &vop_revoke_desc, spec_revoke }, /* revoke */
166 { &vop_mmap_desc, spec_mmap }, /* mmap */
167 { &vop_fsync_desc, spec_fsync }, /* fsync */
168 { &vop_seek_desc, spec_seek }, /* seek */
169 { &vop_remove_desc, spec_remove }, /* remove */
170 { &vop_link_desc, spec_link }, /* link */
171 { &vop_rename_desc, spec_rename }, /* rename */
172 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
173 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
174 { &vop_symlink_desc, spec_symlink }, /* symlink */
175 { &vop_readdir_desc, spec_readdir }, /* readdir */
176 { &vop_readlink_desc, spec_readlink }, /* readlink */
177 { &vop_abortop_desc, spec_abortop }, /* abortop */
178 { &vop_inactive_desc, nfs_inactive }, /* inactive */
179 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
180 { &vop_lock_desc, nfs_lock }, /* lock */
181 { &vop_unlock_desc, nfs_unlock }, /* unlock */
182 { &vop_bmap_desc, spec_bmap }, /* bmap */
183 { &vop_strategy_desc, spec_strategy }, /* strategy */
184 { &vop_print_desc, nfs_print }, /* print */
185 { &vop_islocked_desc, nfs_islocked }, /* islocked */
186 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
187 { &vop_advlock_desc, spec_advlock }, /* advlock */
188 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */
189 { &vop_getpages_desc, spec_getpages }, /* getpages */
190 { &vop_putpages_desc, spec_putpages }, /* putpages */
191 { NULL, NULL }
192 };
193 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
194 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
195
196 int (**fifo_nfsv2nodeop_p)(void *);
197 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
198 { &vop_default_desc, vn_default_error },
199 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */
200 { &vop_create_desc, vn_fifo_bypass }, /* create */
201 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */
202 { &vop_open_desc, vn_fifo_bypass }, /* open */
203 { &vop_close_desc, nfsfifo_close }, /* close */
204 { &vop_access_desc, nfsspec_access }, /* access */
205 { &vop_getattr_desc, nfs_getattr }, /* getattr */
206 { &vop_setattr_desc, nfs_setattr }, /* setattr */
207 { &vop_read_desc, nfsfifo_read }, /* read */
208 { &vop_write_desc, nfsfifo_write }, /* write */
209 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
210 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
211 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
212 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
213 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */
214 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */
215 { &vop_fsync_desc, nfs_fsync }, /* fsync */
216 { &vop_seek_desc, vn_fifo_bypass }, /* seek */
217 { &vop_remove_desc, vn_fifo_bypass }, /* remove */
218 { &vop_link_desc, vn_fifo_bypass }, /* link */
219 { &vop_rename_desc, vn_fifo_bypass }, /* rename */
220 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */
221 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */
222 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */
223 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */
224 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */
225 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */
226 { &vop_inactive_desc, nfs_inactive }, /* inactive */
227 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
228 { &vop_lock_desc, nfs_lock }, /* lock */
229 { &vop_unlock_desc, nfs_unlock }, /* unlock */
230 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */
231 { &vop_strategy_desc, genfs_badop }, /* strategy */
232 { &vop_print_desc, nfs_print }, /* print */
233 { &vop_islocked_desc, nfs_islocked }, /* islocked */
234 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
235 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */
236 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
237 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */
238 { NULL, NULL }
239 };
240 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
241 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
242
243 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
244 size_t, kauth_cred_t, struct lwp *);
245 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *);
246
247 /*
248 * Global variables
249 */
250 extern u_int32_t nfs_true, nfs_false;
251 extern u_int32_t nfs_xdrneg1;
252 extern const nfstype nfsv3_type[9];
253
254 int nfs_numasync = 0;
255 #define DIRHDSIZ _DIRENT_NAMEOFF(dp)
256 #define UIO_ADVANCE(uio, siz) \
257 (void)((uio)->uio_resid -= (siz), \
258 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
259 (uio)->uio_iov->iov_len -= (siz))
260
261 static void nfs_cache_enter(struct vnode *, struct vnode *,
262 struct componentname *);
263
264 static void
265 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
266 struct componentname *cnp)
267 {
268 struct nfsnode *dnp = VTONFS(dvp);
269
270 if (vp != NULL) {
271 struct nfsnode *np = VTONFS(vp);
272
273 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
274 }
275
276 if (!timespecisset(&dnp->n_nctime))
277 dnp->n_nctime = dnp->n_vattr->va_mtime;
278
279 cache_enter(dvp, vp, cnp);
280 }
281
282 /*
283 * nfs null call from vfs.
284 */
285 int
286 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l)
287 {
288 char *bpos, *dpos;
289 int error = 0;
290 struct mbuf *mreq, *mrep, *md, *mb;
291 struct nfsnode *np = VTONFS(vp);
292
293 nfsm_reqhead(np, NFSPROC_NULL, 0);
294 nfsm_request(np, NFSPROC_NULL, l, cred);
295 nfsm_reqdone;
296 return (error);
297 }
298
299 /*
300 * nfs access vnode op.
301 * For nfs version 2, just return ok. File accesses may fail later.
302 * For nfs version 3, use the access rpc to check accessibility. If file modes
303 * are changed on the server, accesses might still fail later.
304 */
305 int
306 nfs_access(void *v)
307 {
308 struct vop_access_args /* {
309 struct vnode *a_vp;
310 int a_mode;
311 kauth_cred_t a_cred;
312 } */ *ap = v;
313 struct vnode *vp = ap->a_vp;
314 #ifndef NFS_V2_ONLY
315 u_int32_t *tl;
316 char *cp;
317 int32_t t1, t2;
318 char *bpos, *dpos, *cp2;
319 int error = 0, attrflag;
320 struct mbuf *mreq, *mrep, *md, *mb;
321 u_int32_t mode, rmode;
322 const int v3 = NFS_ISV3(vp);
323 #endif
324 int cachevalid;
325 struct nfsnode *np = VTONFS(vp);
326 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
327
328 cachevalid = (np->n_accstamp != -1 &&
329 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
330 np->n_accuid == kauth_cred_geteuid(ap->a_cred));
331
332 /*
333 * Check access cache first. If this request has been made for this
334 * uid shortly before, use the cached result.
335 */
336 if (cachevalid) {
337 if (!np->n_accerror) {
338 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
339 return np->n_accerror;
340 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
341 return np->n_accerror;
342 }
343
344 #ifndef NFS_V2_ONLY
345 /*
346 * For nfs v3, do an access rpc, otherwise you are stuck emulating
347 * ufs_access() locally using the vattr. This may not be correct,
348 * since the server may apply other access criteria such as
349 * client uid-->server uid mapping that we do not know about, but
350 * this is better than just returning anything that is lying about
351 * in the cache.
352 */
353 if (v3) {
354 nfsstats.rpccnt[NFSPROC_ACCESS]++;
355 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
356 nfsm_fhtom(np, v3);
357 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
358 if (ap->a_mode & VREAD)
359 mode = NFSV3ACCESS_READ;
360 else
361 mode = 0;
362 if (vp->v_type != VDIR) {
363 if (ap->a_mode & VWRITE)
364 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
365 if (ap->a_mode & VEXEC)
366 mode |= NFSV3ACCESS_EXECUTE;
367 } else {
368 if (ap->a_mode & VWRITE)
369 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
370 NFSV3ACCESS_DELETE);
371 if (ap->a_mode & VEXEC)
372 mode |= NFSV3ACCESS_LOOKUP;
373 }
374 *tl = txdr_unsigned(mode);
375 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
376 nfsm_postop_attr(vp, attrflag, 0);
377 if (!error) {
378 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
379 rmode = fxdr_unsigned(u_int32_t, *tl);
380 /*
381 * The NFS V3 spec does not clarify whether or not
382 * the returned access bits can be a superset of
383 * the ones requested, so...
384 */
385 if ((rmode & mode) != mode)
386 error = EACCES;
387 }
388 nfsm_reqdone;
389 } else
390 #endif
391 return (nfsspec_access(ap));
392 #ifndef NFS_V2_ONLY
393 /*
394 * Disallow write attempts on filesystems mounted read-only;
395 * unless the file is a socket, fifo, or a block or character
396 * device resident on the filesystem.
397 */
398 if (!error && (ap->a_mode & VWRITE) &&
399 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
400 switch (vp->v_type) {
401 case VREG:
402 case VDIR:
403 case VLNK:
404 error = EROFS;
405 default:
406 break;
407 }
408 }
409
410 if (!error || error == EACCES) {
411 /*
412 * If we got the same result as for a previous,
413 * different request, OR it in. Don't update
414 * the timestamp in that case.
415 */
416 if (cachevalid && np->n_accstamp != -1 &&
417 error == np->n_accerror) {
418 if (!error)
419 np->n_accmode |= ap->a_mode;
420 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
421 np->n_accmode = ap->a_mode;
422 } else {
423 np->n_accstamp = time_uptime;
424 np->n_accuid = kauth_cred_geteuid(ap->a_cred);
425 np->n_accmode = ap->a_mode;
426 np->n_accerror = error;
427 }
428 }
429
430 return (error);
431 #endif
432 }
433
434 /*
435 * nfs open vnode op
436 * Check to see if the type is ok
437 * and that deletion is not in progress.
438 * For paged in text files, you will need to flush the page cache
439 * if consistency is lost.
440 */
441 /* ARGSUSED */
442 int
443 nfs_open(void *v)
444 {
445 struct vop_open_args /* {
446 struct vnode *a_vp;
447 int a_mode;
448 kauth_cred_t a_cred;
449 } */ *ap = v;
450 struct vnode *vp = ap->a_vp;
451 struct nfsnode *np = VTONFS(vp);
452 int error;
453
454 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
455 return (EACCES);
456 }
457
458 if (ap->a_mode & FREAD) {
459 if (np->n_rcred != NULL)
460 kauth_cred_free(np->n_rcred);
461 np->n_rcred = ap->a_cred;
462 kauth_cred_hold(np->n_rcred);
463 }
464 if (ap->a_mode & FWRITE) {
465 if (np->n_wcred != NULL)
466 kauth_cred_free(np->n_wcred);
467 np->n_wcred = ap->a_cred;
468 kauth_cred_hold(np->n_wcred);
469 }
470
471 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
472 if (error)
473 return error;
474
475 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
476
477 return (0);
478 }
479
480 /*
481 * nfs close vnode op
482 * What an NFS client should do upon close after writing is a debatable issue.
483 * Most NFS clients push delayed writes to the server upon close, basically for
484 * two reasons:
485 * 1 - So that any write errors may be reported back to the client process
486 * doing the close system call. By far the two most likely errors are
487 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
488 * 2 - To put a worst case upper bound on cache inconsistency between
489 * multiple clients for the file.
490 * There is also a consistency problem for Version 2 of the protocol w.r.t.
491 * not being able to tell if other clients are writing a file concurrently,
492 * since there is no way of knowing if the changed modify time in the reply
493 * is only due to the write for this client.
494 * (NFS Version 3 provides weak cache consistency data in the reply that
495 * should be sufficient to detect and handle this case.)
496 *
497 * The current code does the following:
498 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
499 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
500 * or commit them (this satisfies 1 and 2 except for the
501 * case where the server crashes after this close but
502 * before the commit RPC, which is felt to be "good
503 * enough". Changing the last argument to nfs_flush() to
504 * a 1 would force a commit operation, if it is felt a
505 * commit is necessary now.
506 */
507 /* ARGSUSED */
508 int
509 nfs_close(void *v)
510 {
511 struct vop_close_args /* {
512 struct vnodeop_desc *a_desc;
513 struct vnode *a_vp;
514 int a_fflag;
515 kauth_cred_t a_cred;
516 } */ *ap = v;
517 struct vnode *vp = ap->a_vp;
518 struct nfsnode *np = VTONFS(vp);
519 int error = 0;
520 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
521
522 if (vp->v_type == VREG) {
523 if (np->n_flag & NMODIFIED) {
524 #ifndef NFS_V2_ONLY
525 if (NFS_ISV3(vp)) {
526 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
527 np->n_flag &= ~NMODIFIED;
528 } else
529 #endif
530 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
531 NFS_INVALIDATE_ATTRCACHE(np);
532 }
533 if (np->n_flag & NWRITEERR) {
534 np->n_flag &= ~NWRITEERR;
535 error = np->n_error;
536 }
537 }
538 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
539 return (error);
540 }
541
542 /*
543 * nfs getattr call from vfs.
544 */
545 int
546 nfs_getattr(void *v)
547 {
548 struct vop_getattr_args /* {
549 struct vnode *a_vp;
550 struct vattr *a_vap;
551 kauth_cred_t a_cred;
552 } */ *ap = v;
553 struct vnode *vp = ap->a_vp;
554 struct nfsnode *np = VTONFS(vp);
555 char *cp;
556 u_int32_t *tl;
557 int32_t t1, t2;
558 char *bpos, *dpos;
559 int error = 0;
560 struct mbuf *mreq, *mrep, *md, *mb;
561 const int v3 = NFS_ISV3(vp);
562
563 /*
564 * Update local times for special files.
565 */
566 if (np->n_flag & (NACC | NUPD))
567 np->n_flag |= NCHG;
568
569 /*
570 * if we have delayed truncation, do it now.
571 */
572 nfs_delayedtruncate(vp);
573
574 /*
575 * First look in the cache.
576 */
577 if (nfs_getattrcache(vp, ap->a_vap) == 0)
578 return (0);
579 nfsstats.rpccnt[NFSPROC_GETATTR]++;
580 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
581 nfsm_fhtom(np, v3);
582 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
583 if (!error) {
584 nfsm_loadattr(vp, ap->a_vap, 0);
585 if (vp->v_type == VDIR &&
586 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
587 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
588 }
589 nfsm_reqdone;
590 return (error);
591 }
592
593 /*
594 * nfs setattr call.
595 */
596 int
597 nfs_setattr(void *v)
598 {
599 struct vop_setattr_args /* {
600 struct vnodeop_desc *a_desc;
601 struct vnode *a_vp;
602 struct vattr *a_vap;
603 kauth_cred_t a_cred;
604 } */ *ap = v;
605 struct vnode *vp = ap->a_vp;
606 struct nfsnode *np = VTONFS(vp);
607 struct vattr *vap = ap->a_vap;
608 int error = 0;
609 u_quad_t tsize = 0;
610
611 /*
612 * Setting of flags is not supported.
613 */
614 if (vap->va_flags != VNOVAL)
615 return (EOPNOTSUPP);
616
617 /*
618 * Disallow write attempts if the filesystem is mounted read-only.
619 */
620 if ((vap->va_uid != (uid_t)VNOVAL ||
621 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
622 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
623 (vp->v_mount->mnt_flag & MNT_RDONLY))
624 return (EROFS);
625 if (vap->va_size != VNOVAL) {
626 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
627 return EFBIG;
628 }
629 switch (vp->v_type) {
630 case VDIR:
631 return (EISDIR);
632 case VCHR:
633 case VBLK:
634 case VSOCK:
635 case VFIFO:
636 if (vap->va_mtime.tv_sec == VNOVAL &&
637 vap->va_atime.tv_sec == VNOVAL &&
638 vap->va_mode == (mode_t)VNOVAL &&
639 vap->va_uid == (uid_t)VNOVAL &&
640 vap->va_gid == (gid_t)VNOVAL)
641 return (0);
642 vap->va_size = VNOVAL;
643 break;
644 default:
645 /*
646 * Disallow write attempts if the filesystem is
647 * mounted read-only.
648 */
649 if (vp->v_mount->mnt_flag & MNT_RDONLY)
650 return (EROFS);
651 genfs_node_wrlock(vp);
652 uvm_vnp_setsize(vp, vap->va_size);
653 tsize = np->n_size;
654 np->n_size = vap->va_size;
655 if (vap->va_size == 0)
656 error = nfs_vinvalbuf(vp, 0,
657 ap->a_cred, curlwp, 1);
658 else
659 error = nfs_vinvalbuf(vp, V_SAVE,
660 ap->a_cred, curlwp, 1);
661 if (error) {
662 uvm_vnp_setsize(vp, tsize);
663 genfs_node_unlock(vp);
664 return (error);
665 }
666 np->n_vattr->va_size = vap->va_size;
667 }
668 } else {
669 /*
670 * flush files before setattr because a later write of
671 * cached data might change timestamps or reset sugid bits
672 */
673 if ((vap->va_mtime.tv_sec != VNOVAL ||
674 vap->va_atime.tv_sec != VNOVAL ||
675 vap->va_mode != VNOVAL) &&
676 vp->v_type == VREG &&
677 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
678 curlwp, 1)) == EINTR)
679 return (error);
680 }
681 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
682 if (vap->va_size != VNOVAL) {
683 if (error) {
684 np->n_size = np->n_vattr->va_size = tsize;
685 uvm_vnp_setsize(vp, np->n_size);
686 }
687 genfs_node_unlock(vp);
688 }
689 VN_KNOTE(vp, NOTE_ATTRIB);
690 return (error);
691 }
692
693 /*
694 * Do an nfs setattr rpc.
695 */
696 int
697 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l)
698 {
699 struct nfsv2_sattr *sp;
700 char *cp;
701 int32_t t1, t2;
702 char *bpos, *dpos;
703 u_int32_t *tl;
704 int error = 0;
705 struct mbuf *mreq, *mrep, *md, *mb;
706 const int v3 = NFS_ISV3(vp);
707 struct nfsnode *np = VTONFS(vp);
708 #ifndef NFS_V2_ONLY
709 int wccflag = NFSV3_WCCRATTR;
710 char *cp2;
711 #endif
712
713 nfsstats.rpccnt[NFSPROC_SETATTR]++;
714 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
715 nfsm_fhtom(np, v3);
716 #ifndef NFS_V2_ONLY
717 if (v3) {
718 nfsm_v3attrbuild(vap, true);
719 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
720 *tl = nfs_false;
721 } else {
722 #endif
723 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
724 if (vap->va_mode == (mode_t)VNOVAL)
725 sp->sa_mode = nfs_xdrneg1;
726 else
727 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
728 if (vap->va_uid == (uid_t)VNOVAL)
729 sp->sa_uid = nfs_xdrneg1;
730 else
731 sp->sa_uid = txdr_unsigned(vap->va_uid);
732 if (vap->va_gid == (gid_t)VNOVAL)
733 sp->sa_gid = nfs_xdrneg1;
734 else
735 sp->sa_gid = txdr_unsigned(vap->va_gid);
736 sp->sa_size = txdr_unsigned(vap->va_size);
737 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
738 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
739 #ifndef NFS_V2_ONLY
740 }
741 #endif
742 nfsm_request(np, NFSPROC_SETATTR, l, cred);
743 #ifndef NFS_V2_ONLY
744 if (v3) {
745 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
746 } else
747 #endif
748 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
749 nfsm_reqdone;
750 return (error);
751 }
752
753 /*
754 * nfs lookup call, one step at a time...
755 * First look in cache
756 * If not found, do the rpc.
757 */
758 int
759 nfs_lookup(void *v)
760 {
761 struct vop_lookup_args /* {
762 struct vnodeop_desc *a_desc;
763 struct vnode *a_dvp;
764 struct vnode **a_vpp;
765 struct componentname *a_cnp;
766 } */ *ap = v;
767 struct componentname *cnp = ap->a_cnp;
768 struct vnode *dvp = ap->a_dvp;
769 struct vnode **vpp = ap->a_vpp;
770 int flags;
771 struct vnode *newvp;
772 u_int32_t *tl;
773 char *cp;
774 int32_t t1, t2;
775 char *bpos, *dpos, *cp2;
776 struct mbuf *mreq, *mrep, *md, *mb;
777 long len;
778 nfsfh_t *fhp;
779 struct nfsnode *np;
780 int error = 0, attrflag, fhsize;
781 const int v3 = NFS_ISV3(dvp);
782
783 flags = cnp->cn_flags;
784
785 *vpp = NULLVP;
786 newvp = NULLVP;
787 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
788 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
789 return (EROFS);
790 if (dvp->v_type != VDIR)
791 return (ENOTDIR);
792
793 /*
794 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
795 */
796 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
797 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
798 if (error)
799 return error;
800 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
801 return EISDIR;
802 vref(dvp);
803 *vpp = dvp;
804 return 0;
805 }
806
807 np = VTONFS(dvp);
808
809 /*
810 * Before performing an RPC, check the name cache to see if
811 * the directory/name pair we are looking for is known already.
812 * If the directory/name pair is found in the name cache,
813 * we have to ensure the directory has not changed from
814 * the time the cache entry has been created. If it has,
815 * the cache entry has to be ignored.
816 */
817 error = cache_lookup_raw(dvp, vpp, cnp);
818 KASSERT(dvp != *vpp);
819 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0);
820 if (error >= 0) {
821 struct vattr vattr;
822 int err2;
823
824 if (error && error != ENOENT) {
825 *vpp = NULLVP;
826 return error;
827 }
828
829 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
830 if (err2 != 0) {
831 if (error == 0)
832 vrele(*vpp);
833 *vpp = NULLVP;
834 return err2;
835 }
836
837 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
838 || timespeccmp(&vattr.va_mtime,
839 &VTONFS(dvp)->n_nctime, !=)) {
840 if (error == 0) {
841 vrele(*vpp);
842 *vpp = NULLVP;
843 }
844 cache_purge1(dvp, NULL, PURGE_CHILDREN);
845 timespecclear(&np->n_nctime);
846 goto dorpc;
847 }
848
849 if (error == ENOENT) {
850 goto noentry;
851 }
852
853 /*
854 * investigate the vnode returned by cache_lookup_raw.
855 * if it isn't appropriate, do an rpc.
856 */
857 newvp = *vpp;
858 if ((flags & ISDOTDOT) != 0) {
859 VOP_UNLOCK(dvp);
860 }
861 error = vn_lock(newvp, LK_EXCLUSIVE);
862 if ((flags & ISDOTDOT) != 0) {
863 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
864 }
865 if (error != 0) {
866 /* newvp has been reclaimed. */
867 vrele(newvp);
868 *vpp = NULLVP;
869 goto dorpc;
870 }
871 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
872 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
873 nfsstats.lookupcache_hits++;
874 KASSERT(newvp->v_type != VNON);
875 return (0);
876 }
877 cache_purge1(newvp, NULL, PURGE_PARENTS);
878 vput(newvp);
879 *vpp = NULLVP;
880 }
881 dorpc:
882 #if 0
883 /*
884 * because nfsv3 has the same CREATE semantics as ours,
885 * we don't have to perform LOOKUPs beforehand.
886 *
887 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
888 * XXX although we have no way to know if O_EXCL is requested or not.
889 */
890
891 if (v3 && cnp->cn_nameiop == CREATE &&
892 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
893 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
894 return (EJUSTRETURN);
895 }
896 #endif /* 0 */
897
898 error = 0;
899 newvp = NULLVP;
900 nfsstats.lookupcache_misses++;
901 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
902 len = cnp->cn_namelen;
903 nfsm_reqhead(np, NFSPROC_LOOKUP,
904 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
905 nfsm_fhtom(np, v3);
906 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
907 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
908 if (error) {
909 nfsm_postop_attr(dvp, attrflag, 0);
910 m_freem(mrep);
911 goto nfsmout;
912 }
913 nfsm_getfh(fhp, fhsize, v3);
914
915 /*
916 * Handle RENAME case...
917 */
918 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
919 if (NFS_CMPFH(np, fhp, fhsize)) {
920 m_freem(mrep);
921 return (EISDIR);
922 }
923 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
924 if (error) {
925 m_freem(mrep);
926 return error;
927 }
928 newvp = NFSTOV(np);
929 #ifndef NFS_V2_ONLY
930 if (v3) {
931 nfsm_postop_attr(newvp, attrflag, 0);
932 nfsm_postop_attr(dvp, attrflag, 0);
933 } else
934 #endif
935 nfsm_loadattr(newvp, (struct vattr *)0, 0);
936 *vpp = newvp;
937 m_freem(mrep);
938 goto validate;
939 }
940
941 /*
942 * The postop attr handling is duplicated for each if case,
943 * because it should be done while dvp is locked (unlocking
944 * dvp is different for each case).
945 */
946
947 if (NFS_CMPFH(np, fhp, fhsize)) {
948 /*
949 * as we handle "." lookup locally, this should be
950 * a broken server.
951 */
952 vref(dvp);
953 newvp = dvp;
954 #ifndef NFS_V2_ONLY
955 if (v3) {
956 nfsm_postop_attr(newvp, attrflag, 0);
957 nfsm_postop_attr(dvp, attrflag, 0);
958 } else
959 #endif
960 nfsm_loadattr(newvp, (struct vattr *)0, 0);
961 } else if (flags & ISDOTDOT) {
962 /*
963 * ".." lookup
964 */
965 VOP_UNLOCK(dvp);
966 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
967 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
968 if (error) {
969 m_freem(mrep);
970 return error;
971 }
972 newvp = NFSTOV(np);
973
974 #ifndef NFS_V2_ONLY
975 if (v3) {
976 nfsm_postop_attr(newvp, attrflag, 0);
977 nfsm_postop_attr(dvp, attrflag, 0);
978 } else
979 #endif
980 nfsm_loadattr(newvp, (struct vattr *)0, 0);
981 } else {
982 /*
983 * Other lookups.
984 */
985 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
986 if (error) {
987 m_freem(mrep);
988 return error;
989 }
990 newvp = NFSTOV(np);
991 #ifndef NFS_V2_ONLY
992 if (v3) {
993 nfsm_postop_attr(newvp, attrflag, 0);
994 nfsm_postop_attr(dvp, attrflag, 0);
995 } else
996 #endif
997 nfsm_loadattr(newvp, (struct vattr *)0, 0);
998 }
999 if ((cnp->cn_flags & MAKEENTRY) &&
1000 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1001 nfs_cache_enter(dvp, newvp, cnp);
1002 }
1003 *vpp = newvp;
1004 nfsm_reqdone;
1005 if (error) {
1006 /*
1007 * We get here only because of errors returned by
1008 * the RPC. Otherwise we'll have returned above
1009 * (the nfsm_* macros will jump to nfsm_reqdone
1010 * on error).
1011 */
1012 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1013 cnp->cn_nameiop != CREATE) {
1014 nfs_cache_enter(dvp, NULL, cnp);
1015 }
1016 if (newvp != NULLVP) {
1017 if (newvp == dvp) {
1018 vrele(newvp);
1019 } else {
1020 vput(newvp);
1021 }
1022 }
1023 noentry:
1024 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1025 (flags & ISLASTCN) && error == ENOENT) {
1026 if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1027 error = EROFS;
1028 } else {
1029 error = EJUSTRETURN;
1030 }
1031 }
1032 *vpp = NULL;
1033 return error;
1034 }
1035
1036 validate:
1037 /*
1038 * make sure we have valid type and size.
1039 */
1040
1041 newvp = *vpp;
1042 if (newvp->v_type == VNON) {
1043 struct vattr vattr; /* dummy */
1044
1045 KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1046 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1047 if (error) {
1048 vput(newvp);
1049 *vpp = NULL;
1050 }
1051 }
1052
1053 return error;
1054 }
1055
1056 /*
1057 * nfs read call.
1058 * Just call nfs_bioread() to do the work.
1059 */
1060 int
1061 nfs_read(void *v)
1062 {
1063 struct vop_read_args /* {
1064 struct vnode *a_vp;
1065 struct uio *a_uio;
1066 int a_ioflag;
1067 kauth_cred_t a_cred;
1068 } */ *ap = v;
1069 struct vnode *vp = ap->a_vp;
1070
1071 if (vp->v_type != VREG)
1072 return EISDIR;
1073 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1074 }
1075
1076 /*
1077 * nfs readlink call
1078 */
1079 int
1080 nfs_readlink(void *v)
1081 {
1082 struct vop_readlink_args /* {
1083 struct vnode *a_vp;
1084 struct uio *a_uio;
1085 kauth_cred_t a_cred;
1086 } */ *ap = v;
1087 struct vnode *vp = ap->a_vp;
1088 struct nfsnode *np = VTONFS(vp);
1089
1090 if (vp->v_type != VLNK)
1091 return (EPERM);
1092
1093 if (np->n_rcred != NULL) {
1094 kauth_cred_free(np->n_rcred);
1095 }
1096 np->n_rcred = ap->a_cred;
1097 kauth_cred_hold(np->n_rcred);
1098
1099 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1100 }
1101
1102 /*
1103 * Do a readlink rpc.
1104 * Called by nfs_doio() from below the buffer cache.
1105 */
1106 int
1107 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
1108 {
1109 u_int32_t *tl;
1110 char *cp;
1111 int32_t t1, t2;
1112 char *bpos, *dpos, *cp2;
1113 int error = 0;
1114 uint32_t len;
1115 struct mbuf *mreq, *mrep, *md, *mb;
1116 const int v3 = NFS_ISV3(vp);
1117 struct nfsnode *np = VTONFS(vp);
1118 #ifndef NFS_V2_ONLY
1119 int attrflag;
1120 #endif
1121
1122 nfsstats.rpccnt[NFSPROC_READLINK]++;
1123 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1124 nfsm_fhtom(np, v3);
1125 nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1126 #ifndef NFS_V2_ONLY
1127 if (v3)
1128 nfsm_postop_attr(vp, attrflag, 0);
1129 #endif
1130 if (!error) {
1131 #ifndef NFS_V2_ONLY
1132 if (v3) {
1133 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1134 len = fxdr_unsigned(uint32_t, *tl);
1135 if (len > NFS_MAXPATHLEN) {
1136 /*
1137 * this pathname is too long for us.
1138 */
1139 m_freem(mrep);
1140 /* Solaris returns EINVAL. should we follow? */
1141 error = ENAMETOOLONG;
1142 goto nfsmout;
1143 }
1144 } else
1145 #endif
1146 {
1147 nfsm_strsiz(len, NFS_MAXPATHLEN);
1148 }
1149 nfsm_mtouio(uiop, len);
1150 }
1151 nfsm_reqdone;
1152 return (error);
1153 }
1154
1155 /*
1156 * nfs read rpc call
1157 * Ditto above
1158 */
1159 int
1160 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1161 {
1162 u_int32_t *tl;
1163 char *cp;
1164 int32_t t1, t2;
1165 char *bpos, *dpos, *cp2;
1166 struct mbuf *mreq, *mrep, *md, *mb;
1167 struct nfsmount *nmp;
1168 int error = 0, len, retlen, tsiz, eof, byte_count;
1169 const int v3 = NFS_ISV3(vp);
1170 struct nfsnode *np = VTONFS(vp);
1171 #ifndef NFS_V2_ONLY
1172 int attrflag;
1173 #endif
1174
1175 #ifndef nolint
1176 eof = 0;
1177 #endif
1178 nmp = VFSTONFS(vp->v_mount);
1179 tsiz = uiop->uio_resid;
1180 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1181 return (EFBIG);
1182 iostat_busy(nmp->nm_stats);
1183 byte_count = 0; /* count bytes actually transferred */
1184 while (tsiz > 0) {
1185 nfsstats.rpccnt[NFSPROC_READ]++;
1186 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1187 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1188 nfsm_fhtom(np, v3);
1189 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1190 #ifndef NFS_V2_ONLY
1191 if (v3) {
1192 txdr_hyper(uiop->uio_offset, tl);
1193 *(tl + 2) = txdr_unsigned(len);
1194 } else
1195 #endif
1196 {
1197 *tl++ = txdr_unsigned(uiop->uio_offset);
1198 *tl++ = txdr_unsigned(len);
1199 *tl = 0;
1200 }
1201 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1202 #ifndef NFS_V2_ONLY
1203 if (v3) {
1204 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1205 if (error) {
1206 m_freem(mrep);
1207 goto nfsmout;
1208 }
1209 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1210 eof = fxdr_unsigned(int, *(tl + 1));
1211 } else
1212 #endif
1213 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1214 nfsm_strsiz(retlen, nmp->nm_rsize);
1215 nfsm_mtouio(uiop, retlen);
1216 m_freem(mrep);
1217 tsiz -= retlen;
1218 byte_count += retlen;
1219 #ifndef NFS_V2_ONLY
1220 if (v3) {
1221 if (eof || retlen == 0)
1222 tsiz = 0;
1223 } else
1224 #endif
1225 if (retlen < len)
1226 tsiz = 0;
1227 }
1228 nfsmout:
1229 iostat_unbusy(nmp->nm_stats, byte_count, 1);
1230 return (error);
1231 }
1232
1233 struct nfs_writerpc_context {
1234 kmutex_t nwc_lock;
1235 kcondvar_t nwc_cv;
1236 int nwc_mbufcount;
1237 };
1238
1239 /*
1240 * free mbuf used to refer protected pages while write rpc call.
1241 * called at splvm.
1242 */
1243 static void
1244 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1245 {
1246 struct nfs_writerpc_context *ctx = arg;
1247
1248 KASSERT(m != NULL);
1249 KASSERT(ctx != NULL);
1250 pool_cache_put(mb_cache, m);
1251 mutex_enter(&ctx->nwc_lock);
1252 if (--ctx->nwc_mbufcount == 0) {
1253 cv_signal(&ctx->nwc_cv);
1254 }
1255 mutex_exit(&ctx->nwc_lock);
1256 }
1257
1258 /*
1259 * nfs write call
1260 */
1261 int
1262 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp)
1263 {
1264 u_int32_t *tl;
1265 char *cp;
1266 int32_t t1, t2;
1267 char *bpos, *dpos;
1268 struct mbuf *mreq, *mrep, *md, *mb;
1269 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1270 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1271 const int v3 = NFS_ISV3(vp);
1272 int committed = NFSV3WRITE_FILESYNC;
1273 struct nfsnode *np = VTONFS(vp);
1274 struct nfs_writerpc_context ctx;
1275 int byte_count;
1276 size_t origresid;
1277 #ifndef NFS_V2_ONLY
1278 char *cp2;
1279 int rlen, commit;
1280 #endif
1281
1282 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1283 panic("writerpc readonly vp %p", vp);
1284 }
1285
1286 #ifdef DIAGNOSTIC
1287 if (uiop->uio_iovcnt != 1)
1288 panic("nfs: writerpc iovcnt > 1");
1289 #endif
1290 tsiz = uiop->uio_resid;
1291 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1292 return EFBIG;
1293
1294 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1295 cv_init(&ctx.nwc_cv, "nfsmblk");
1296 ctx.nwc_mbufcount = 1;
1297
1298 retry:
1299 origresid = uiop->uio_resid;
1300 KASSERT(origresid == uiop->uio_iov->iov_len);
1301 iostat_busy(nmp->nm_stats);
1302 byte_count = 0; /* count of bytes actually written */
1303 while (tsiz > 0) {
1304 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1305 uint32_t backup;
1306 bool stalewriteverf = false;
1307
1308 nfsstats.rpccnt[NFSPROC_WRITE]++;
1309 len = min(tsiz, nmp->nm_wsize);
1310 datalen = pageprotected ? 0 : nfsm_rndup(len);
1311 nfsm_reqhead(np, NFSPROC_WRITE,
1312 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1313 nfsm_fhtom(np, v3);
1314 #ifndef NFS_V2_ONLY
1315 if (v3) {
1316 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1317 txdr_hyper(uiop->uio_offset, tl);
1318 tl += 2;
1319 *tl++ = txdr_unsigned(len);
1320 *tl++ = txdr_unsigned(*iomode);
1321 *tl = txdr_unsigned(len);
1322 } else
1323 #endif
1324 {
1325 u_int32_t x;
1326
1327 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1328 /* Set both "begin" and "current" to non-garbage. */
1329 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1330 *tl++ = x; /* "begin offset" */
1331 *tl++ = x; /* "current offset" */
1332 x = txdr_unsigned(len);
1333 *tl++ = x; /* total to this offset */
1334 *tl = x; /* size of this write */
1335
1336 }
1337 if (pageprotected) {
1338 /*
1339 * since we know pages can't be modified during i/o,
1340 * no need to copy them for us.
1341 */
1342 struct mbuf *m;
1343 struct iovec *iovp = uiop->uio_iov;
1344
1345 m = m_get(M_WAIT, MT_DATA);
1346 MCLAIM(m, &nfs_mowner);
1347 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1348 nfs_writerpc_extfree, &ctx);
1349 m->m_flags |= M_EXT_ROMAP;
1350 m->m_len = len;
1351 mb->m_next = m;
1352 /*
1353 * no need to maintain mb and bpos here
1354 * because no one care them later.
1355 */
1356 #if 0
1357 mb = m;
1358 bpos = mtod(void *, mb) + mb->m_len;
1359 #endif
1360 UIO_ADVANCE(uiop, len);
1361 uiop->uio_offset += len;
1362 mutex_enter(&ctx.nwc_lock);
1363 ctx.nwc_mbufcount++;
1364 mutex_exit(&ctx.nwc_lock);
1365 nfs_zeropad(mb, 0, nfsm_padlen(len));
1366 } else {
1367 nfsm_uiotom(uiop, len);
1368 }
1369 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1370 #ifndef NFS_V2_ONLY
1371 if (v3) {
1372 wccflag = NFSV3_WCCCHK;
1373 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1374 if (!error) {
1375 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1376 + NFSX_V3WRITEVERF);
1377 rlen = fxdr_unsigned(int, *tl++);
1378 if (rlen == 0) {
1379 error = NFSERR_IO;
1380 m_freem(mrep);
1381 break;
1382 } else if (rlen < len) {
1383 backup = len - rlen;
1384 UIO_ADVANCE(uiop, -backup);
1385 uiop->uio_offset -= backup;
1386 len = rlen;
1387 }
1388 commit = fxdr_unsigned(int, *tl++);
1389
1390 /*
1391 * Return the lowest committment level
1392 * obtained by any of the RPCs.
1393 */
1394 if (committed == NFSV3WRITE_FILESYNC)
1395 committed = commit;
1396 else if (committed == NFSV3WRITE_DATASYNC &&
1397 commit == NFSV3WRITE_UNSTABLE)
1398 committed = commit;
1399 mutex_enter(&nmp->nm_lock);
1400 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1401 memcpy(nmp->nm_writeverf, tl,
1402 NFSX_V3WRITEVERF);
1403 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1404 } else if ((nmp->nm_iflag &
1405 NFSMNT_STALEWRITEVERF) ||
1406 memcmp(tl, nmp->nm_writeverf,
1407 NFSX_V3WRITEVERF)) {
1408 memcpy(nmp->nm_writeverf, tl,
1409 NFSX_V3WRITEVERF);
1410 /*
1411 * note NFSMNT_STALEWRITEVERF
1412 * if we're the first thread to
1413 * notice it.
1414 */
1415 if ((nmp->nm_iflag &
1416 NFSMNT_STALEWRITEVERF) == 0) {
1417 stalewriteverf = true;
1418 nmp->nm_iflag |=
1419 NFSMNT_STALEWRITEVERF;
1420 }
1421 }
1422 mutex_exit(&nmp->nm_lock);
1423 }
1424 } else
1425 #endif
1426 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1427 if (wccflag)
1428 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1429 m_freem(mrep);
1430 if (error)
1431 break;
1432 tsiz -= len;
1433 byte_count += len;
1434 if (stalewriteverf) {
1435 *stalewriteverfp = true;
1436 stalewriteverf = false;
1437 if (committed == NFSV3WRITE_UNSTABLE &&
1438 len != origresid) {
1439 /*
1440 * if our write requests weren't atomic but
1441 * unstable, datas in previous iterations
1442 * might have already been lost now.
1443 * then, we should resend them to nfsd.
1444 */
1445 backup = origresid - tsiz;
1446 UIO_ADVANCE(uiop, -backup);
1447 uiop->uio_offset -= backup;
1448 tsiz = origresid;
1449 goto retry;
1450 }
1451 }
1452 }
1453 nfsmout:
1454 iostat_unbusy(nmp->nm_stats, byte_count, 0);
1455 if (pageprotected) {
1456 /*
1457 * wait until mbufs go away.
1458 * retransmitted mbufs can survive longer than rpc requests
1459 * themselves.
1460 */
1461 mutex_enter(&ctx.nwc_lock);
1462 ctx.nwc_mbufcount--;
1463 while (ctx.nwc_mbufcount > 0) {
1464 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1465 }
1466 mutex_exit(&ctx.nwc_lock);
1467 }
1468 mutex_destroy(&ctx.nwc_lock);
1469 cv_destroy(&ctx.nwc_cv);
1470 *iomode = committed;
1471 if (error)
1472 uiop->uio_resid = tsiz;
1473 return (error);
1474 }
1475
1476 /*
1477 * nfs mknod rpc
1478 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1479 * mode set to specify the file type and the size field for rdev.
1480 */
1481 int
1482 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap)
1483 {
1484 struct nfsv2_sattr *sp;
1485 u_int32_t *tl;
1486 char *cp;
1487 int32_t t1, t2;
1488 struct vnode *newvp = (struct vnode *)0;
1489 struct nfsnode *dnp, *np;
1490 char *cp2;
1491 char *bpos, *dpos;
1492 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1493 struct mbuf *mreq, *mrep, *md, *mb;
1494 u_int32_t rdev;
1495 const int v3 = NFS_ISV3(dvp);
1496
1497 if (vap->va_type == VCHR || vap->va_type == VBLK)
1498 rdev = txdr_unsigned(vap->va_rdev);
1499 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1500 rdev = nfs_xdrneg1;
1501 else {
1502 VOP_ABORTOP(dvp, cnp);
1503 vput(dvp);
1504 return (EOPNOTSUPP);
1505 }
1506 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1507 dnp = VTONFS(dvp);
1508 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1509 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1510 nfsm_fhtom(dnp, v3);
1511 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1512 #ifndef NFS_V2_ONLY
1513 if (v3) {
1514 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1515 *tl++ = vtonfsv3_type(vap->va_type);
1516 nfsm_v3attrbuild(vap, false);
1517 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1518 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1519 *tl++ = txdr_unsigned(major(vap->va_rdev));
1520 *tl = txdr_unsigned(minor(vap->va_rdev));
1521 }
1522 } else
1523 #endif
1524 {
1525 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1526 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1527 sp->sa_uid = nfs_xdrneg1;
1528 sp->sa_gid = nfs_xdrneg1;
1529 sp->sa_size = rdev;
1530 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1531 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1532 }
1533 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1534 if (!error) {
1535 nfsm_mtofh(dvp, newvp, v3, gotvp);
1536 if (!gotvp) {
1537 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1538 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1539 if (!error)
1540 newvp = NFSTOV(np);
1541 }
1542 }
1543 #ifndef NFS_V2_ONLY
1544 if (v3)
1545 nfsm_wcc_data(dvp, wccflag, 0, !error);
1546 #endif
1547 nfsm_reqdone;
1548 if (error) {
1549 if (newvp)
1550 vput(newvp);
1551 } else {
1552 if (cnp->cn_flags & MAKEENTRY)
1553 nfs_cache_enter(dvp, newvp, cnp);
1554 *vpp = newvp;
1555 }
1556 VTONFS(dvp)->n_flag |= NMODIFIED;
1557 if (!wccflag)
1558 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1559 vput(dvp);
1560 return (error);
1561 }
1562
1563 /*
1564 * nfs mknod vop
1565 * just call nfs_mknodrpc() to do the work.
1566 */
1567 /* ARGSUSED */
1568 int
1569 nfs_mknod(void *v)
1570 {
1571 struct vop_mknod_args /* {
1572 struct vnode *a_dvp;
1573 struct vnode **a_vpp;
1574 struct componentname *a_cnp;
1575 struct vattr *a_vap;
1576 } */ *ap = v;
1577 struct vnode *dvp = ap->a_dvp;
1578 struct componentname *cnp = ap->a_cnp;
1579 int error;
1580
1581 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1582 VN_KNOTE(dvp, NOTE_WRITE);
1583 if (error == 0 || error == EEXIST)
1584 cache_purge1(dvp, cnp, 0);
1585 return (error);
1586 }
1587
1588 /*
1589 * nfs file create call
1590 */
1591 int
1592 nfs_create(void *v)
1593 {
1594 struct vop_create_args /* {
1595 struct vnode *a_dvp;
1596 struct vnode **a_vpp;
1597 struct componentname *a_cnp;
1598 struct vattr *a_vap;
1599 } */ *ap = v;
1600 struct vnode *dvp = ap->a_dvp;
1601 struct vattr *vap = ap->a_vap;
1602 struct componentname *cnp = ap->a_cnp;
1603 struct nfsv2_sattr *sp;
1604 u_int32_t *tl;
1605 char *cp;
1606 int32_t t1, t2;
1607 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1608 struct vnode *newvp = (struct vnode *)0;
1609 char *bpos, *dpos, *cp2;
1610 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1611 struct mbuf *mreq, *mrep, *md, *mb;
1612 const int v3 = NFS_ISV3(dvp);
1613 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED;
1614
1615 /*
1616 * Oops, not for me..
1617 */
1618 if (vap->va_type == VSOCK)
1619 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1620
1621 KASSERT(vap->va_type == VREG);
1622
1623 #ifdef VA_EXCLUSIVE
1624 if (vap->va_vaflags & VA_EXCLUSIVE) {
1625 excl_mode = NFSV3CREATE_EXCLUSIVE;
1626 }
1627 #endif
1628 again:
1629 error = 0;
1630 nfsstats.rpccnt[NFSPROC_CREATE]++;
1631 dnp = VTONFS(dvp);
1632 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1633 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1634 nfsm_fhtom(dnp, v3);
1635 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1636 #ifndef NFS_V2_ONLY
1637 if (v3) {
1638 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1639 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1640 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1641 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1642 *tl++ = cprng_fast32();
1643 *tl = cprng_fast32();
1644 } else {
1645 *tl = txdr_unsigned(excl_mode);
1646 nfsm_v3attrbuild(vap, false);
1647 }
1648 } else
1649 #endif
1650 {
1651 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1652 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1653 sp->sa_uid = nfs_xdrneg1;
1654 sp->sa_gid = nfs_xdrneg1;
1655 sp->sa_size = 0;
1656 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1657 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1658 }
1659 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred);
1660 if (!error) {
1661 nfsm_mtofh(dvp, newvp, v3, gotvp);
1662 if (!gotvp) {
1663 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1664 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1665 if (!error)
1666 newvp = NFSTOV(np);
1667 }
1668 }
1669 #ifndef NFS_V2_ONLY
1670 if (v3)
1671 nfsm_wcc_data(dvp, wccflag, 0, !error);
1672 #endif
1673 nfsm_reqdone;
1674 if (error) {
1675 /*
1676 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1677 */
1678 if (v3 && error == ENOTSUP) {
1679 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1680 excl_mode = NFSV3CREATE_GUARDED;
1681 goto again;
1682 } else if (excl_mode == NFSV3CREATE_GUARDED) {
1683 excl_mode = NFSV3CREATE_UNCHECKED;
1684 goto again;
1685 }
1686 }
1687 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) {
1688 struct timespec ts;
1689
1690 getnanotime(&ts);
1691
1692 /*
1693 * make sure that we'll update timestamps as
1694 * most server implementations use them to store
1695 * the create verifier.
1696 *
1697 * XXX it's better to use TOSERVER always.
1698 */
1699
1700 if (vap->va_atime.tv_sec == VNOVAL)
1701 vap->va_atime = ts;
1702 if (vap->va_mtime.tv_sec == VNOVAL)
1703 vap->va_mtime = ts;
1704
1705 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp);
1706 }
1707 if (error == 0) {
1708 if (cnp->cn_flags & MAKEENTRY)
1709 nfs_cache_enter(dvp, newvp, cnp);
1710 else
1711 cache_purge1(dvp, cnp, 0);
1712 *ap->a_vpp = newvp;
1713 } else {
1714 if (newvp)
1715 vput(newvp);
1716 if (error == EEXIST)
1717 cache_purge1(dvp, cnp, 0);
1718 }
1719 VTONFS(dvp)->n_flag |= NMODIFIED;
1720 if (!wccflag)
1721 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1722 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1723 vput(dvp);
1724 return (error);
1725 }
1726
1727 /*
1728 * nfs file remove call
1729 * To try and make nfs semantics closer to ufs semantics, a file that has
1730 * other processes using the vnode is renamed instead of removed and then
1731 * removed later on the last close.
1732 * - If v_usecount > 1
1733 * If a rename is not already in the works
1734 * call nfs_sillyrename() to set it up
1735 * else
1736 * do the remove rpc
1737 */
1738 int
1739 nfs_remove(void *v)
1740 {
1741 struct vop_remove_args /* {
1742 struct vnodeop_desc *a_desc;
1743 struct vnode * a_dvp;
1744 struct vnode * a_vp;
1745 struct componentname * a_cnp;
1746 } */ *ap = v;
1747 struct vnode *vp = ap->a_vp;
1748 struct vnode *dvp = ap->a_dvp;
1749 struct componentname *cnp = ap->a_cnp;
1750 struct nfsnode *np = VTONFS(vp);
1751 int error = 0;
1752 struct vattr vattr;
1753
1754 #ifndef DIAGNOSTIC
1755 if (vp->v_usecount < 1)
1756 panic("nfs_remove: bad v_usecount");
1757 #endif
1758 if (vp->v_type == VDIR)
1759 error = EPERM;
1760 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1761 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1762 vattr.va_nlink > 1)) {
1763 /*
1764 * Purge the name cache so that the chance of a lookup for
1765 * the name succeeding while the remove is in progress is
1766 * minimized. Without node locking it can still happen, such
1767 * that an I/O op returns ESTALE, but since you get this if
1768 * another host removes the file..
1769 */
1770 cache_purge(vp);
1771 /*
1772 * throw away biocache buffers, mainly to avoid
1773 * unnecessary delayed writes later.
1774 */
1775 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1);
1776 /* Do the rpc */
1777 if (error != EINTR)
1778 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1779 cnp->cn_namelen, cnp->cn_cred, curlwp);
1780 } else if (!np->n_sillyrename)
1781 error = nfs_sillyrename(dvp, vp, cnp, false);
1782 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1783 vattr.va_nlink == 1) {
1784 np->n_flag |= NREMOVED;
1785 }
1786 NFS_INVALIDATE_ATTRCACHE(np);
1787 VN_KNOTE(vp, NOTE_DELETE);
1788 VN_KNOTE(dvp, NOTE_WRITE);
1789 if (dvp == vp)
1790 vrele(vp);
1791 else
1792 vput(vp);
1793 vput(dvp);
1794 return (error);
1795 }
1796
1797 /*
1798 * nfs file remove rpc called from nfs_inactive
1799 */
1800 int
1801 nfs_removeit(struct sillyrename *sp)
1802 {
1803
1804 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1805 (struct lwp *)0));
1806 }
1807
1808 /*
1809 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1810 */
1811 int
1812 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l)
1813 {
1814 u_int32_t *tl;
1815 char *cp;
1816 #ifndef NFS_V2_ONLY
1817 int32_t t1;
1818 char *cp2;
1819 #endif
1820 int32_t t2;
1821 char *bpos, *dpos;
1822 int error = 0, wccflag = NFSV3_WCCRATTR;
1823 struct mbuf *mreq, *mrep, *md, *mb;
1824 const int v3 = NFS_ISV3(dvp);
1825 int rexmit = 0;
1826 struct nfsnode *dnp = VTONFS(dvp);
1827
1828 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1829 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1830 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1831 nfsm_fhtom(dnp, v3);
1832 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1833 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1834 #ifndef NFS_V2_ONLY
1835 if (v3)
1836 nfsm_wcc_data(dvp, wccflag, 0, !error);
1837 #endif
1838 nfsm_reqdone;
1839 VTONFS(dvp)->n_flag |= NMODIFIED;
1840 if (!wccflag)
1841 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1842 /*
1843 * Kludge City: If the first reply to the remove rpc is lost..
1844 * the reply to the retransmitted request will be ENOENT
1845 * since the file was in fact removed
1846 * Therefore, we cheat and return success.
1847 */
1848 if (rexmit && error == ENOENT)
1849 error = 0;
1850 return (error);
1851 }
1852
1853 /*
1854 * nfs file rename call
1855 */
1856 int
1857 nfs_rename(void *v)
1858 {
1859 struct vop_rename_args /* {
1860 struct vnode *a_fdvp;
1861 struct vnode *a_fvp;
1862 struct componentname *a_fcnp;
1863 struct vnode *a_tdvp;
1864 struct vnode *a_tvp;
1865 struct componentname *a_tcnp;
1866 } */ *ap = v;
1867 struct vnode *fvp = ap->a_fvp;
1868 struct vnode *tvp = ap->a_tvp;
1869 struct vnode *fdvp = ap->a_fdvp;
1870 struct vnode *tdvp = ap->a_tdvp;
1871 struct componentname *tcnp = ap->a_tcnp;
1872 struct componentname *fcnp = ap->a_fcnp;
1873 int error;
1874
1875 /* Check for cross-device rename */
1876 if ((fvp->v_mount != tdvp->v_mount) ||
1877 (tvp && (fvp->v_mount != tvp->v_mount))) {
1878 error = EXDEV;
1879 goto out;
1880 }
1881
1882 /*
1883 * If the tvp exists and is in use, sillyrename it before doing the
1884 * rename of the new file over it.
1885 *
1886 * Have sillyrename use link instead of rename if possible,
1887 * so that we don't lose the file if the rename fails, and so
1888 * that there's no window when the "to" file doesn't exist.
1889 */
1890 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1891 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1892 VN_KNOTE(tvp, NOTE_DELETE);
1893 vput(tvp);
1894 tvp = NULL;
1895 }
1896
1897 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1898 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1899 curlwp);
1900
1901 VN_KNOTE(fdvp, NOTE_WRITE);
1902 VN_KNOTE(tdvp, NOTE_WRITE);
1903 if (error == 0 || error == EEXIST) {
1904 if (fvp->v_type == VDIR)
1905 cache_purge(fvp);
1906 else
1907 cache_purge1(fdvp, fcnp, 0);
1908 if (tvp != NULL && tvp->v_type == VDIR)
1909 cache_purge(tvp);
1910 else
1911 cache_purge1(tdvp, tcnp, 0);
1912 }
1913 out:
1914 if (tdvp == tvp)
1915 vrele(tdvp);
1916 else
1917 vput(tdvp);
1918 if (tvp)
1919 vput(tvp);
1920 vrele(fdvp);
1921 vrele(fvp);
1922 return (error);
1923 }
1924
1925 /*
1926 * nfs file rename rpc called from nfs_remove() above
1927 */
1928 int
1929 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp)
1930 {
1931 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1932 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
1933 }
1934
1935 /*
1936 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1937 */
1938 int
1939 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, struct vnode *tdvp, const char *tnameptr, int tnamelen, kauth_cred_t cred, struct lwp *l)
1940 {
1941 u_int32_t *tl;
1942 char *cp;
1943 #ifndef NFS_V2_ONLY
1944 int32_t t1;
1945 char *cp2;
1946 #endif
1947 int32_t t2;
1948 char *bpos, *dpos;
1949 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1950 struct mbuf *mreq, *mrep, *md, *mb;
1951 const int v3 = NFS_ISV3(fdvp);
1952 int rexmit = 0;
1953 struct nfsnode *fdnp = VTONFS(fdvp);
1954
1955 nfsstats.rpccnt[NFSPROC_RENAME]++;
1956 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1957 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1958 nfsm_rndup(tnamelen));
1959 nfsm_fhtom(fdnp, v3);
1960 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1961 nfsm_fhtom(VTONFS(tdvp), v3);
1962 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1963 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
1964 #ifndef NFS_V2_ONLY
1965 if (v3) {
1966 nfsm_wcc_data(fdvp, fwccflag, 0, !error);
1967 nfsm_wcc_data(tdvp, twccflag, 0, !error);
1968 }
1969 #endif
1970 nfsm_reqdone;
1971 VTONFS(fdvp)->n_flag |= NMODIFIED;
1972 VTONFS(tdvp)->n_flag |= NMODIFIED;
1973 if (!fwccflag)
1974 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
1975 if (!twccflag)
1976 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
1977 /*
1978 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1979 */
1980 if (rexmit && error == ENOENT)
1981 error = 0;
1982 return (error);
1983 }
1984
1985 /*
1986 * NFS link RPC, called from nfs_link.
1987 * Assumes dvp and vp locked, and leaves them that way.
1988 */
1989
1990 static int
1991 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
1992 size_t namelen, kauth_cred_t cred, struct lwp *l)
1993 {
1994 u_int32_t *tl;
1995 char *cp;
1996 #ifndef NFS_V2_ONLY
1997 int32_t t1;
1998 char *cp2;
1999 #endif
2000 int32_t t2;
2001 char *bpos, *dpos;
2002 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2003 struct mbuf *mreq, *mrep, *md, *mb;
2004 const int v3 = NFS_ISV3(dvp);
2005 int rexmit = 0;
2006 struct nfsnode *np = VTONFS(vp);
2007
2008 nfsstats.rpccnt[NFSPROC_LINK]++;
2009 nfsm_reqhead(np, NFSPROC_LINK,
2010 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2011 nfsm_fhtom(np, v3);
2012 nfsm_fhtom(VTONFS(dvp), v3);
2013 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2014 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2015 #ifndef NFS_V2_ONLY
2016 if (v3) {
2017 nfsm_postop_attr(vp, attrflag, 0);
2018 nfsm_wcc_data(dvp, wccflag, 0, !error);
2019 }
2020 #endif
2021 nfsm_reqdone;
2022
2023 VTONFS(dvp)->n_flag |= NMODIFIED;
2024 if (!attrflag)
2025 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2026 if (!wccflag)
2027 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2028
2029 /*
2030 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2031 */
2032 if (rexmit && error == EEXIST)
2033 error = 0;
2034
2035 return error;
2036 }
2037
2038 /*
2039 * nfs hard link create call
2040 */
2041 int
2042 nfs_link(void *v)
2043 {
2044 struct vop_link_args /* {
2045 struct vnode *a_dvp;
2046 struct vnode *a_vp;
2047 struct componentname *a_cnp;
2048 } */ *ap = v;
2049 struct vnode *vp = ap->a_vp;
2050 struct vnode *dvp = ap->a_dvp;
2051 struct componentname *cnp = ap->a_cnp;
2052 int error = 0;
2053
2054 error = vn_lock(vp, LK_EXCLUSIVE);
2055 if (error != 0) {
2056 VOP_ABORTOP(dvp, cnp);
2057 vput(dvp);
2058 return error;
2059 }
2060
2061 /*
2062 * Push all writes to the server, so that the attribute cache
2063 * doesn't get "out of sync" with the server.
2064 * XXX There should be a better way!
2065 */
2066 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0);
2067
2068 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2069 cnp->cn_cred, curlwp);
2070
2071 if (error == 0) {
2072 cache_purge1(dvp, cnp, 0);
2073 }
2074 VOP_UNLOCK(vp);
2075 VN_KNOTE(vp, NOTE_LINK);
2076 VN_KNOTE(dvp, NOTE_WRITE);
2077 vput(dvp);
2078 return (error);
2079 }
2080
2081 /*
2082 * nfs symbolic link create call
2083 */
2084 int
2085 nfs_symlink(void *v)
2086 {
2087 struct vop_symlink_args /* {
2088 struct vnode *a_dvp;
2089 struct vnode **a_vpp;
2090 struct componentname *a_cnp;
2091 struct vattr *a_vap;
2092 char *a_target;
2093 } */ *ap = v;
2094 struct vnode *dvp = ap->a_dvp;
2095 struct vattr *vap = ap->a_vap;
2096 struct componentname *cnp = ap->a_cnp;
2097 struct nfsv2_sattr *sp;
2098 u_int32_t *tl;
2099 char *cp;
2100 int32_t t1, t2;
2101 char *bpos, *dpos, *cp2;
2102 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2103 struct mbuf *mreq, *mrep, *md, *mb;
2104 struct vnode *newvp = (struct vnode *)0;
2105 const int v3 = NFS_ISV3(dvp);
2106 int rexmit = 0;
2107 struct nfsnode *dnp = VTONFS(dvp);
2108
2109 *ap->a_vpp = NULL;
2110 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2111 slen = strlen(ap->a_target);
2112 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2113 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2114 nfsm_fhtom(dnp, v3);
2115 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2116 #ifndef NFS_V2_ONlY
2117 if (v3)
2118 nfsm_v3attrbuild(vap, false);
2119 #endif
2120 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2121 #ifndef NFS_V2_ONlY
2122 if (!v3) {
2123 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2124 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2125 sp->sa_uid = nfs_xdrneg1;
2126 sp->sa_gid = nfs_xdrneg1;
2127 sp->sa_size = nfs_xdrneg1;
2128 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2129 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2130 }
2131 #endif
2132 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2133 &rexmit);
2134 #ifndef NFS_V2_ONlY
2135 if (v3) {
2136 if (!error)
2137 nfsm_mtofh(dvp, newvp, v3, gotvp);
2138 nfsm_wcc_data(dvp, wccflag, 0, !error);
2139 }
2140 #endif
2141 nfsm_reqdone;
2142 /*
2143 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2144 */
2145 if (rexmit && error == EEXIST)
2146 error = 0;
2147 if (error == 0 || error == EEXIST)
2148 cache_purge1(dvp, cnp, 0);
2149 if (error == 0 && newvp == NULL) {
2150 struct nfsnode *np = NULL;
2151
2152 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2153 cnp->cn_cred, curlwp, &np);
2154 if (error == 0)
2155 newvp = NFSTOV(np);
2156 }
2157 if (error) {
2158 if (newvp != NULL)
2159 vput(newvp);
2160 } else {
2161 *ap->a_vpp = newvp;
2162 }
2163 VTONFS(dvp)->n_flag |= NMODIFIED;
2164 if (!wccflag)
2165 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2166 VN_KNOTE(dvp, NOTE_WRITE);
2167 vput(dvp);
2168 return (error);
2169 }
2170
2171 /*
2172 * nfs make dir call
2173 */
2174 int
2175 nfs_mkdir(void *v)
2176 {
2177 struct vop_mkdir_args /* {
2178 struct vnode *a_dvp;
2179 struct vnode **a_vpp;
2180 struct componentname *a_cnp;
2181 struct vattr *a_vap;
2182 } */ *ap = v;
2183 struct vnode *dvp = ap->a_dvp;
2184 struct vattr *vap = ap->a_vap;
2185 struct componentname *cnp = ap->a_cnp;
2186 struct nfsv2_sattr *sp;
2187 u_int32_t *tl;
2188 char *cp;
2189 int32_t t1, t2;
2190 int len;
2191 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2192 struct vnode *newvp = (struct vnode *)0;
2193 char *bpos, *dpos, *cp2;
2194 int error = 0, wccflag = NFSV3_WCCRATTR;
2195 int gotvp = 0;
2196 int rexmit = 0;
2197 struct mbuf *mreq, *mrep, *md, *mb;
2198 const int v3 = NFS_ISV3(dvp);
2199
2200 len = cnp->cn_namelen;
2201 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2202 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2203 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2204 nfsm_fhtom(dnp, v3);
2205 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2206 #ifndef NFS_V2_ONLY
2207 if (v3) {
2208 nfsm_v3attrbuild(vap, false);
2209 } else
2210 #endif
2211 {
2212 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2213 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2214 sp->sa_uid = nfs_xdrneg1;
2215 sp->sa_gid = nfs_xdrneg1;
2216 sp->sa_size = nfs_xdrneg1;
2217 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2218 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2219 }
2220 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2221 if (!error)
2222 nfsm_mtofh(dvp, newvp, v3, gotvp);
2223 if (v3)
2224 nfsm_wcc_data(dvp, wccflag, 0, !error);
2225 nfsm_reqdone;
2226 VTONFS(dvp)->n_flag |= NMODIFIED;
2227 if (!wccflag)
2228 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2229 /*
2230 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2231 * if we can succeed in looking up the directory.
2232 */
2233 if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2234 if (newvp) {
2235 vput(newvp);
2236 newvp = (struct vnode *)0;
2237 }
2238 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2239 curlwp, &np);
2240 if (!error) {
2241 newvp = NFSTOV(np);
2242 if (newvp->v_type != VDIR || newvp == dvp)
2243 error = EEXIST;
2244 }
2245 }
2246 if (error) {
2247 if (newvp) {
2248 if (dvp != newvp)
2249 vput(newvp);
2250 else
2251 vrele(newvp);
2252 }
2253 } else {
2254 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2255 if (cnp->cn_flags & MAKEENTRY)
2256 nfs_cache_enter(dvp, newvp, cnp);
2257 *ap->a_vpp = newvp;
2258 }
2259 vput(dvp);
2260 return (error);
2261 }
2262
2263 /*
2264 * nfs remove directory call
2265 */
2266 int
2267 nfs_rmdir(void *v)
2268 {
2269 struct vop_rmdir_args /* {
2270 struct vnode *a_dvp;
2271 struct vnode *a_vp;
2272 struct componentname *a_cnp;
2273 } */ *ap = v;
2274 struct vnode *vp = ap->a_vp;
2275 struct vnode *dvp = ap->a_dvp;
2276 struct componentname *cnp = ap->a_cnp;
2277 u_int32_t *tl;
2278 char *cp;
2279 #ifndef NFS_V2_ONLY
2280 int32_t t1;
2281 char *cp2;
2282 #endif
2283 int32_t t2;
2284 char *bpos, *dpos;
2285 int error = 0, wccflag = NFSV3_WCCRATTR;
2286 int rexmit = 0;
2287 struct mbuf *mreq, *mrep, *md, *mb;
2288 const int v3 = NFS_ISV3(dvp);
2289 struct nfsnode *dnp;
2290
2291 if (dvp == vp) {
2292 vrele(dvp);
2293 vput(dvp);
2294 return (EINVAL);
2295 }
2296 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2297 dnp = VTONFS(dvp);
2298 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2299 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2300 nfsm_fhtom(dnp, v3);
2301 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2302 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2303 #ifndef NFS_V2_ONLY
2304 if (v3)
2305 nfsm_wcc_data(dvp, wccflag, 0, !error);
2306 #endif
2307 nfsm_reqdone;
2308 VTONFS(dvp)->n_flag |= NMODIFIED;
2309 if (!wccflag)
2310 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2311 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2312 VN_KNOTE(vp, NOTE_DELETE);
2313 cache_purge(vp);
2314 vput(vp);
2315 vput(dvp);
2316 /*
2317 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2318 */
2319 if (rexmit && error == ENOENT)
2320 error = 0;
2321 return (error);
2322 }
2323
2324 /*
2325 * nfs readdir call
2326 */
2327 int
2328 nfs_readdir(void *v)
2329 {
2330 struct vop_readdir_args /* {
2331 struct vnode *a_vp;
2332 struct uio *a_uio;
2333 kauth_cred_t a_cred;
2334 int *a_eofflag;
2335 off_t **a_cookies;
2336 int *a_ncookies;
2337 } */ *ap = v;
2338 struct vnode *vp = ap->a_vp;
2339 struct uio *uio = ap->a_uio;
2340 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2341 char *base = uio->uio_iov->iov_base;
2342 int tresid, error;
2343 size_t count, lost;
2344 struct dirent *dp;
2345 off_t *cookies = NULL;
2346 int ncookies = 0, nc;
2347
2348 if (vp->v_type != VDIR)
2349 return (EPERM);
2350
2351 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2352 count = uio->uio_resid - lost;
2353 if (count <= 0)
2354 return (EINVAL);
2355
2356 /*
2357 * Call nfs_bioread() to do the real work.
2358 */
2359 tresid = uio->uio_resid = count;
2360 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2361 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2362
2363 if (!error && ap->a_cookies) {
2364 ncookies = count / 16;
2365 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2366 *ap->a_cookies = cookies;
2367 }
2368
2369 if (!error && uio->uio_resid == tresid) {
2370 uio->uio_resid += lost;
2371 nfsstats.direofcache_misses++;
2372 if (ap->a_cookies)
2373 *ap->a_ncookies = 0;
2374 *ap->a_eofflag = 1;
2375 return (0);
2376 }
2377
2378 if (!error && ap->a_cookies) {
2379 /*
2380 * Only the NFS server and emulations use cookies, and they
2381 * load the directory block into system space, so we can
2382 * just look at it directly.
2383 */
2384 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2385 uio->uio_iovcnt != 1)
2386 panic("nfs_readdir: lost in space");
2387 for (nc = 0; ncookies-- &&
2388 base < (char *)uio->uio_iov->iov_base; nc++){
2389 dp = (struct dirent *) base;
2390 if (dp->d_reclen == 0)
2391 break;
2392 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2393 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2394 else
2395 *(cookies++) = NFS_GETCOOKIE(dp);
2396 base += dp->d_reclen;
2397 }
2398 uio->uio_resid +=
2399 ((char *)uio->uio_iov->iov_base - base);
2400 uio->uio_iov->iov_len +=
2401 ((char *)uio->uio_iov->iov_base - base);
2402 uio->uio_iov->iov_base = base;
2403 *ap->a_ncookies = nc;
2404 }
2405
2406 uio->uio_resid += lost;
2407 *ap->a_eofflag = 0;
2408 return (error);
2409 }
2410
2411 /*
2412 * Readdir rpc call.
2413 * Called from below the buffer cache by nfs_doio().
2414 */
2415 int
2416 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
2417 {
2418 int len, left;
2419 struct dirent *dp = NULL;
2420 u_int32_t *tl;
2421 char *cp;
2422 int32_t t1, t2;
2423 char *bpos, *dpos, *cp2;
2424 struct mbuf *mreq, *mrep, *md, *mb;
2425 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2426 struct nfsnode *dnp = VTONFS(vp);
2427 u_quad_t fileno;
2428 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2429 #ifndef NFS_V2_ONLY
2430 int attrflag;
2431 #endif
2432 int nrpcs = 0, reclen;
2433 const int v3 = NFS_ISV3(vp);
2434
2435 #ifdef DIAGNOSTIC
2436 /*
2437 * Should be called from buffer cache, so only amount of
2438 * NFS_DIRBLKSIZ will be requested.
2439 */
2440 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2441 panic("nfs readdirrpc bad uio");
2442 #endif
2443
2444 /*
2445 * Loop around doing readdir rpc's of size nm_readdirsize
2446 * truncated to a multiple of NFS_DIRFRAGSIZ.
2447 * The stopping criteria is EOF or buffer full.
2448 */
2449 while (more_dirs && bigenough) {
2450 /*
2451 * Heuristic: don't bother to do another RPC to further
2452 * fill up this block if there is not much room left. (< 50%
2453 * of the readdir RPC size). This wastes some buffer space
2454 * but can save up to 50% in RPC calls.
2455 */
2456 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2457 bigenough = 0;
2458 break;
2459 }
2460 nfsstats.rpccnt[NFSPROC_READDIR]++;
2461 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2462 NFSX_READDIR(v3));
2463 nfsm_fhtom(dnp, v3);
2464 #ifndef NFS_V2_ONLY
2465 if (v3) {
2466 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2467 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2468 txdr_swapcookie3(uiop->uio_offset, tl);
2469 } else {
2470 txdr_cookie3(uiop->uio_offset, tl);
2471 }
2472 tl += 2;
2473 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2474 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2475 } else
2476 #endif
2477 {
2478 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2479 *tl++ = txdr_unsigned(uiop->uio_offset);
2480 }
2481 *tl = txdr_unsigned(nmp->nm_readdirsize);
2482 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2483 nrpcs++;
2484 #ifndef NFS_V2_ONLY
2485 if (v3) {
2486 nfsm_postop_attr(vp, attrflag, 0);
2487 if (!error) {
2488 nfsm_dissect(tl, u_int32_t *,
2489 2 * NFSX_UNSIGNED);
2490 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2491 dnp->n_cookieverf.nfsuquad[1] = *tl;
2492 } else {
2493 m_freem(mrep);
2494 goto nfsmout;
2495 }
2496 }
2497 #endif
2498 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2499 more_dirs = fxdr_unsigned(int, *tl);
2500
2501 /* loop thru the dir entries, doctoring them to 4bsd form */
2502 while (more_dirs && bigenough) {
2503 #ifndef NFS_V2_ONLY
2504 if (v3) {
2505 nfsm_dissect(tl, u_int32_t *,
2506 3 * NFSX_UNSIGNED);
2507 fileno = fxdr_hyper(tl);
2508 len = fxdr_unsigned(int, *(tl + 2));
2509 } else
2510 #endif
2511 {
2512 nfsm_dissect(tl, u_int32_t *,
2513 2 * NFSX_UNSIGNED);
2514 fileno = fxdr_unsigned(u_quad_t, *tl++);
2515 len = fxdr_unsigned(int, *tl);
2516 }
2517 if (len <= 0 || len > NFS_MAXNAMLEN) {
2518 error = EBADRPC;
2519 m_freem(mrep);
2520 goto nfsmout;
2521 }
2522 /* for cookie stashing */
2523 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2524 left = NFS_DIRFRAGSIZ - blksiz;
2525 if (reclen > left) {
2526 memset(uiop->uio_iov->iov_base, 0, left);
2527 dp->d_reclen += left;
2528 UIO_ADVANCE(uiop, left);
2529 blksiz = 0;
2530 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2531 }
2532 if (reclen > uiop->uio_resid)
2533 bigenough = 0;
2534 if (bigenough) {
2535 int tlen;
2536
2537 dp = (struct dirent *)uiop->uio_iov->iov_base;
2538 dp->d_fileno = fileno;
2539 dp->d_namlen = len;
2540 dp->d_reclen = reclen;
2541 dp->d_type = DT_UNKNOWN;
2542 blksiz += reclen;
2543 if (blksiz == NFS_DIRFRAGSIZ)
2544 blksiz = 0;
2545 UIO_ADVANCE(uiop, DIRHDSIZ);
2546 nfsm_mtouio(uiop, len);
2547 tlen = reclen - (DIRHDSIZ + len);
2548 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2549 UIO_ADVANCE(uiop, tlen);
2550 } else
2551 nfsm_adv(nfsm_rndup(len));
2552 #ifndef NFS_V2_ONLY
2553 if (v3) {
2554 nfsm_dissect(tl, u_int32_t *,
2555 3 * NFSX_UNSIGNED);
2556 } else
2557 #endif
2558 {
2559 nfsm_dissect(tl, u_int32_t *,
2560 2 * NFSX_UNSIGNED);
2561 }
2562 if (bigenough) {
2563 #ifndef NFS_V2_ONLY
2564 if (v3) {
2565 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2566 uiop->uio_offset =
2567 fxdr_swapcookie3(tl);
2568 else
2569 uiop->uio_offset =
2570 fxdr_cookie3(tl);
2571 }
2572 else
2573 #endif
2574 {
2575 uiop->uio_offset =
2576 fxdr_unsigned(off_t, *tl);
2577 }
2578 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2579 }
2580 if (v3)
2581 tl += 2;
2582 else
2583 tl++;
2584 more_dirs = fxdr_unsigned(int, *tl);
2585 }
2586 /*
2587 * If at end of rpc data, get the eof boolean
2588 */
2589 if (!more_dirs) {
2590 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2591 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2592
2593 /*
2594 * kludge: if we got no entries, treat it as EOF.
2595 * some server sometimes send a reply without any
2596 * entries or EOF.
2597 * although it might mean the server has very long name,
2598 * we can't handle such entries anyway.
2599 */
2600
2601 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2602 more_dirs = 0;
2603 }
2604 m_freem(mrep);
2605 }
2606 /*
2607 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2608 * by increasing d_reclen for the last record.
2609 */
2610 if (blksiz > 0) {
2611 left = NFS_DIRFRAGSIZ - blksiz;
2612 memset(uiop->uio_iov->iov_base, 0, left);
2613 dp->d_reclen += left;
2614 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2615 UIO_ADVANCE(uiop, left);
2616 }
2617
2618 /*
2619 * We are now either at the end of the directory or have filled the
2620 * block.
2621 */
2622 if (bigenough) {
2623 dnp->n_direofoffset = uiop->uio_offset;
2624 dnp->n_flag |= NEOFVALID;
2625 }
2626 nfsmout:
2627 return (error);
2628 }
2629
2630 #ifndef NFS_V2_ONLY
2631 /*
2632 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2633 */
2634 int
2635 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
2636 {
2637 int len, left;
2638 struct dirent *dp = NULL;
2639 u_int32_t *tl;
2640 char *cp;
2641 int32_t t1, t2;
2642 struct vnode *newvp;
2643 char *bpos, *dpos, *cp2;
2644 struct mbuf *mreq, *mrep, *md, *mb;
2645 struct nameidata nami, *ndp = &nami;
2646 struct componentname *cnp = &ndp->ni_cnd;
2647 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2648 struct nfsnode *dnp = VTONFS(vp), *np;
2649 nfsfh_t *fhp;
2650 u_quad_t fileno;
2651 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2652 int attrflag, fhsize, nrpcs = 0, reclen;
2653 struct nfs_fattr fattr, *fp;
2654
2655 #ifdef DIAGNOSTIC
2656 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2657 panic("nfs readdirplusrpc bad uio");
2658 #endif
2659 ndp->ni_dvp = vp;
2660 newvp = NULLVP;
2661
2662 /*
2663 * Loop around doing readdir rpc's of size nm_readdirsize
2664 * truncated to a multiple of NFS_DIRFRAGSIZ.
2665 * The stopping criteria is EOF or buffer full.
2666 */
2667 while (more_dirs && bigenough) {
2668 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2669 bigenough = 0;
2670 break;
2671 }
2672 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2673 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2674 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2675 nfsm_fhtom(dnp, 1);
2676 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2677 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2678 txdr_swapcookie3(uiop->uio_offset, tl);
2679 } else {
2680 txdr_cookie3(uiop->uio_offset, tl);
2681 }
2682 tl += 2;
2683 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2684 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2685 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2686 *tl = txdr_unsigned(nmp->nm_rsize);
2687 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2688 nfsm_postop_attr(vp, attrflag, 0);
2689 if (error) {
2690 m_freem(mrep);
2691 goto nfsmout;
2692 }
2693 nrpcs++;
2694 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2695 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2696 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2697 more_dirs = fxdr_unsigned(int, *tl);
2698
2699 /* loop thru the dir entries, doctoring them to 4bsd form */
2700 while (more_dirs && bigenough) {
2701 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2702 fileno = fxdr_hyper(tl);
2703 len = fxdr_unsigned(int, *(tl + 2));
2704 if (len <= 0 || len > NFS_MAXNAMLEN) {
2705 error = EBADRPC;
2706 m_freem(mrep);
2707 goto nfsmout;
2708 }
2709 /* for cookie stashing */
2710 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2711 left = NFS_DIRFRAGSIZ - blksiz;
2712 if (reclen > left) {
2713 /*
2714 * DIRFRAGSIZ is aligned, no need to align
2715 * again here.
2716 */
2717 memset(uiop->uio_iov->iov_base, 0, left);
2718 dp->d_reclen += left;
2719 UIO_ADVANCE(uiop, left);
2720 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2721 blksiz = 0;
2722 }
2723 if (reclen > uiop->uio_resid)
2724 bigenough = 0;
2725 if (bigenough) {
2726 int tlen;
2727
2728 dp = (struct dirent *)uiop->uio_iov->iov_base;
2729 dp->d_fileno = fileno;
2730 dp->d_namlen = len;
2731 dp->d_reclen = reclen;
2732 dp->d_type = DT_UNKNOWN;
2733 blksiz += reclen;
2734 if (blksiz == NFS_DIRFRAGSIZ)
2735 blksiz = 0;
2736 UIO_ADVANCE(uiop, DIRHDSIZ);
2737 nfsm_mtouio(uiop, len);
2738 tlen = reclen - (DIRHDSIZ + len);
2739 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2740 UIO_ADVANCE(uiop, tlen);
2741 cnp->cn_nameptr = dp->d_name;
2742 cnp->cn_namelen = dp->d_namlen;
2743 } else
2744 nfsm_adv(nfsm_rndup(len));
2745 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2746 if (bigenough) {
2747 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2748 uiop->uio_offset =
2749 fxdr_swapcookie3(tl);
2750 else
2751 uiop->uio_offset =
2752 fxdr_cookie3(tl);
2753 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2754 }
2755 tl += 2;
2756
2757 /*
2758 * Since the attributes are before the file handle
2759 * (sigh), we must skip over the attributes and then
2760 * come back and get them.
2761 */
2762 attrflag = fxdr_unsigned(int, *tl);
2763 if (attrflag) {
2764 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2765 memcpy(&fattr, fp, NFSX_V3FATTR);
2766 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2767 doit = fxdr_unsigned(int, *tl);
2768 if (doit) {
2769 nfsm_getfh(fhp, fhsize, 1);
2770 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2771 vref(vp);
2772 newvp = vp;
2773 np = dnp;
2774 } else {
2775 error = nfs_nget1(vp->v_mount, fhp,
2776 fhsize, &np, LK_NOWAIT);
2777 if (!error)
2778 newvp = NFSTOV(np);
2779 }
2780 if (!error) {
2781 const char *xcp;
2782
2783 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2784 if (bigenough) {
2785 dp->d_type =
2786 IFTODT(VTTOIF(np->n_vattr->va_type));
2787 if (cnp->cn_namelen <= NCHNAMLEN) {
2788 ndp->ni_vp = newvp;
2789 xcp = cnp->cn_nameptr +
2790 cnp->cn_namelen;
2791 cnp->cn_hash =
2792 namei_hash(cnp->cn_nameptr, &xcp);
2793 nfs_cache_enter(ndp->ni_dvp,
2794 ndp->ni_vp, cnp);
2795 }
2796 }
2797 }
2798 error = 0;
2799 }
2800 } else {
2801 /* Just skip over the file handle */
2802 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2803 i = fxdr_unsigned(int, *tl);
2804 nfsm_adv(nfsm_rndup(i));
2805 }
2806 if (newvp != NULLVP) {
2807 if (newvp == vp)
2808 vrele(newvp);
2809 else
2810 vput(newvp);
2811 newvp = NULLVP;
2812 }
2813 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2814 more_dirs = fxdr_unsigned(int, *tl);
2815 }
2816 /*
2817 * If at end of rpc data, get the eof boolean
2818 */
2819 if (!more_dirs) {
2820 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2821 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2822
2823 /*
2824 * kludge: see a comment in nfs_readdirrpc.
2825 */
2826
2827 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2828 more_dirs = 0;
2829 }
2830 m_freem(mrep);
2831 }
2832 /*
2833 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2834 * by increasing d_reclen for the last record.
2835 */
2836 if (blksiz > 0) {
2837 left = NFS_DIRFRAGSIZ - blksiz;
2838 memset(uiop->uio_iov->iov_base, 0, left);
2839 dp->d_reclen += left;
2840 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2841 UIO_ADVANCE(uiop, left);
2842 }
2843
2844 /*
2845 * We are now either at the end of the directory or have filled the
2846 * block.
2847 */
2848 if (bigenough) {
2849 dnp->n_direofoffset = uiop->uio_offset;
2850 dnp->n_flag |= NEOFVALID;
2851 }
2852 nfsmout:
2853 if (newvp != NULLVP) {
2854 if(newvp == vp)
2855 vrele(newvp);
2856 else
2857 vput(newvp);
2858 }
2859 return (error);
2860 }
2861 #endif
2862
2863 /*
2864 * Silly rename. To make the NFS filesystem that is stateless look a little
2865 * more like the "ufs" a remove of an active vnode is translated to a rename
2866 * to a funny looking filename that is removed by nfs_inactive on the
2867 * nfsnode. There is the potential for another process on a different client
2868 * to create the same funny name between the nfs_lookitup() fails and the
2869 * nfs_rename() completes, but...
2870 */
2871 int
2872 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink)
2873 {
2874 struct sillyrename *sp;
2875 struct nfsnode *np;
2876 int error;
2877 pid_t pid;
2878
2879 cache_purge(dvp);
2880 np = VTONFS(vp);
2881 #ifndef DIAGNOSTIC
2882 if (vp->v_type == VDIR)
2883 panic("nfs: sillyrename dir");
2884 #endif
2885 sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
2886 sp->s_cred = kauth_cred_dup(cnp->cn_cred);
2887 sp->s_dvp = dvp;
2888 vref(dvp);
2889
2890 /* Fudge together a funny name */
2891 pid = curlwp->l_proc->p_pid;
2892 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2893 sp->s_namlen = 12;
2894 sp->s_name[8] = hexdigits[pid & 0xf];
2895 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
2896 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
2897 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
2898
2899 /* Try lookitups until we get one that isn't there */
2900 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2901 curlwp, (struct nfsnode **)0) == 0) {
2902 sp->s_name[4]++;
2903 if (sp->s_name[4] > 'z') {
2904 error = EINVAL;
2905 goto bad;
2906 }
2907 }
2908 if (dolink) {
2909 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
2910 sp->s_cred, curlwp);
2911 /*
2912 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
2913 */
2914 if (error == ENOTSUP) {
2915 error = nfs_renameit(dvp, cnp, sp);
2916 }
2917 } else {
2918 error = nfs_renameit(dvp, cnp, sp);
2919 }
2920 if (error)
2921 goto bad;
2922 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2923 curlwp, &np);
2924 np->n_sillyrename = sp;
2925 return (0);
2926 bad:
2927 vrele(sp->s_dvp);
2928 kauth_cred_free(sp->s_cred);
2929 kmem_free(sp, sizeof(*sp));
2930 return (error);
2931 }
2932
2933 /*
2934 * Look up a file name and optionally either update the file handle or
2935 * allocate an nfsnode, depending on the value of npp.
2936 * npp == NULL --> just do the lookup
2937 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2938 * handled too
2939 * *npp != NULL --> update the file handle in the vnode
2940 */
2941 int
2942 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp)
2943 {
2944 u_int32_t *tl;
2945 char *cp;
2946 int32_t t1, t2;
2947 struct vnode *newvp = (struct vnode *)0;
2948 struct nfsnode *np, *dnp = VTONFS(dvp);
2949 char *bpos, *dpos, *cp2;
2950 int error = 0, fhlen;
2951 #ifndef NFS_V2_ONLY
2952 int attrflag;
2953 #endif
2954 struct mbuf *mreq, *mrep, *md, *mb;
2955 nfsfh_t *nfhp;
2956 const int v3 = NFS_ISV3(dvp);
2957
2958 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2959 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2960 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2961 nfsm_fhtom(dnp, v3);
2962 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2963 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
2964 if (npp && !error) {
2965 nfsm_getfh(nfhp, fhlen, v3);
2966 if (*npp) {
2967 np = *npp;
2968 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2969 kmem_free(np->n_fhp, np->n_fhsize);
2970 np->n_fhp = &np->n_fh;
2971 }
2972 #if NFS_SMALLFH < NFSX_V3FHMAX
2973 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
2974 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
2975 #endif
2976 memcpy(np->n_fhp, nfhp, fhlen);
2977 np->n_fhsize = fhlen;
2978 newvp = NFSTOV(np);
2979 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2980 vref(dvp);
2981 newvp = dvp;
2982 np = dnp;
2983 } else {
2984 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2985 if (error) {
2986 m_freem(mrep);
2987 return (error);
2988 }
2989 newvp = NFSTOV(np);
2990 }
2991 #ifndef NFS_V2_ONLY
2992 if (v3) {
2993 nfsm_postop_attr(newvp, attrflag, 0);
2994 if (!attrflag && *npp == NULL) {
2995 m_freem(mrep);
2996 vput(newvp);
2997 return (ENOENT);
2998 }
2999 } else
3000 #endif
3001 nfsm_loadattr(newvp, (struct vattr *)0, 0);
3002 }
3003 nfsm_reqdone;
3004 if (npp && *npp == NULL) {
3005 if (error) {
3006 if (newvp)
3007 vput(newvp);
3008 } else
3009 *npp = np;
3010 }
3011 return (error);
3012 }
3013
3014 #ifndef NFS_V2_ONLY
3015 /*
3016 * Nfs Version 3 commit rpc
3017 */
3018 int
3019 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l)
3020 {
3021 char *cp;
3022 u_int32_t *tl;
3023 int32_t t1, t2;
3024 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3025 char *bpos, *dpos, *cp2;
3026 int error = 0, wccflag = NFSV3_WCCRATTR;
3027 struct mbuf *mreq, *mrep, *md, *mb;
3028 struct nfsnode *np;
3029
3030 KASSERT(NFS_ISV3(vp));
3031
3032 #ifdef NFS_DEBUG_COMMIT
3033 printf("commit %lu - %lu\n", (unsigned long)offset,
3034 (unsigned long)(offset + cnt));
3035 #endif
3036
3037 mutex_enter(&nmp->nm_lock);
3038 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3039 mutex_exit(&nmp->nm_lock);
3040 return (0);
3041 }
3042 mutex_exit(&nmp->nm_lock);
3043 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3044 np = VTONFS(vp);
3045 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3046 nfsm_fhtom(np, 1);
3047 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3048 txdr_hyper(offset, tl);
3049 tl += 2;
3050 *tl = txdr_unsigned(cnt);
3051 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3052 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3053 if (!error) {
3054 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3055 mutex_enter(&nmp->nm_lock);
3056 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3057 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3058 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3059 error = NFSERR_STALEWRITEVERF;
3060 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3061 }
3062 mutex_exit(&nmp->nm_lock);
3063 }
3064 nfsm_reqdone;
3065 return (error);
3066 }
3067 #endif
3068
3069 /*
3070 * Kludge City..
3071 * - make nfs_bmap() essentially a no-op that does no translation
3072 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3073 * (Maybe I could use the process's page mapping, but I was concerned that
3074 * Kernel Write might not be enabled and also figured copyout() would do
3075 * a lot more work than memcpy() and also it currently happens in the
3076 * context of the swapper process (2).
3077 */
3078 int
3079 nfs_bmap(void *v)
3080 {
3081 struct vop_bmap_args /* {
3082 struct vnode *a_vp;
3083 daddr_t a_bn;
3084 struct vnode **a_vpp;
3085 daddr_t *a_bnp;
3086 int *a_runp;
3087 } */ *ap = v;
3088 struct vnode *vp = ap->a_vp;
3089 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3090
3091 if (ap->a_vpp != NULL)
3092 *ap->a_vpp = vp;
3093 if (ap->a_bnp != NULL)
3094 *ap->a_bnp = ap->a_bn << bshift;
3095 if (ap->a_runp != NULL)
3096 *ap->a_runp = 1024 * 1024; /* XXX */
3097 return (0);
3098 }
3099
3100 /*
3101 * Strategy routine.
3102 * For async requests when nfsiod(s) are running, queue the request by
3103 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3104 * request.
3105 */
3106 int
3107 nfs_strategy(void *v)
3108 {
3109 struct vop_strategy_args *ap = v;
3110 struct buf *bp = ap->a_bp;
3111 int error = 0;
3112
3113 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3114 panic("nfs physio/async");
3115
3116 /*
3117 * If the op is asynchronous and an i/o daemon is waiting
3118 * queue the request, wake it up and wait for completion
3119 * otherwise just do it ourselves.
3120 */
3121 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3122 error = nfs_doio(bp);
3123 return (error);
3124 }
3125
3126 /*
3127 * fsync vnode op. Just call nfs_flush() with commit == 1.
3128 */
3129 /* ARGSUSED */
3130 int
3131 nfs_fsync(void *v)
3132 {
3133 struct vop_fsync_args /* {
3134 struct vnodeop_desc *a_desc;
3135 struct vnode * a_vp;
3136 kauth_cred_t a_cred;
3137 int a_flags;
3138 off_t offlo;
3139 off_t offhi;
3140 struct lwp * a_l;
3141 } */ *ap = v;
3142
3143 struct vnode *vp = ap->a_vp;
3144
3145 if (vp->v_type != VREG)
3146 return 0;
3147
3148 return (nfs_flush(vp, ap->a_cred,
3149 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3150 }
3151
3152 /*
3153 * Flush all the data associated with a vnode.
3154 */
3155 int
3156 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3157 int commit)
3158 {
3159 struct nfsnode *np = VTONFS(vp);
3160 int error;
3161 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3162 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3163
3164 mutex_enter(vp->v_interlock);
3165 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3166 if (np->n_flag & NWRITEERR) {
3167 error = np->n_error;
3168 np->n_flag &= ~NWRITEERR;
3169 }
3170 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3171 return (error);
3172 }
3173
3174 /*
3175 * Return POSIX pathconf information applicable to nfs.
3176 *
3177 * N.B. The NFS V2 protocol doesn't support this RPC.
3178 */
3179 /* ARGSUSED */
3180 int
3181 nfs_pathconf(void *v)
3182 {
3183 struct vop_pathconf_args /* {
3184 struct vnode *a_vp;
3185 int a_name;
3186 register_t *a_retval;
3187 } */ *ap = v;
3188 struct nfsv3_pathconf *pcp;
3189 struct vnode *vp = ap->a_vp;
3190 struct mbuf *mreq, *mrep, *md, *mb;
3191 int32_t t1, t2;
3192 u_int32_t *tl;
3193 char *bpos, *dpos, *cp, *cp2;
3194 int error = 0, attrflag;
3195 #ifndef NFS_V2_ONLY
3196 struct nfsmount *nmp;
3197 unsigned int l;
3198 u_int64_t maxsize;
3199 #endif
3200 const int v3 = NFS_ISV3(vp);
3201 struct nfsnode *np = VTONFS(vp);
3202
3203 switch (ap->a_name) {
3204 /* Names that can be resolved locally. */
3205 case _PC_PIPE_BUF:
3206 *ap->a_retval = PIPE_BUF;
3207 break;
3208 case _PC_SYNC_IO:
3209 *ap->a_retval = 1;
3210 break;
3211 /* Names that cannot be resolved locally; do an RPC, if possible. */
3212 case _PC_LINK_MAX:
3213 case _PC_NAME_MAX:
3214 case _PC_CHOWN_RESTRICTED:
3215 case _PC_NO_TRUNC:
3216 if (!v3) {
3217 error = EINVAL;
3218 break;
3219 }
3220 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3221 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3222 nfsm_fhtom(np, 1);
3223 nfsm_request(np, NFSPROC_PATHCONF,
3224 curlwp, curlwp->l_cred); /* XXX */
3225 nfsm_postop_attr(vp, attrflag, 0);
3226 if (!error) {
3227 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3228 NFSX_V3PATHCONF);
3229 switch (ap->a_name) {
3230 case _PC_LINK_MAX:
3231 *ap->a_retval =
3232 fxdr_unsigned(register_t, pcp->pc_linkmax);
3233 break;
3234 case _PC_NAME_MAX:
3235 *ap->a_retval =
3236 fxdr_unsigned(register_t, pcp->pc_namemax);
3237 break;
3238 case _PC_CHOWN_RESTRICTED:
3239 *ap->a_retval =
3240 (pcp->pc_chownrestricted == nfs_true);
3241 break;
3242 case _PC_NO_TRUNC:
3243 *ap->a_retval =
3244 (pcp->pc_notrunc == nfs_true);
3245 break;
3246 }
3247 }
3248 nfsm_reqdone;
3249 break;
3250 case _PC_FILESIZEBITS:
3251 #ifndef NFS_V2_ONLY
3252 if (v3) {
3253 nmp = VFSTONFS(vp->v_mount);
3254 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3255 if ((error = nfs_fsinfo(nmp, vp,
3256 curlwp->l_cred, curlwp)) != 0) /* XXX */
3257 break;
3258 for (l = 0, maxsize = nmp->nm_maxfilesize;
3259 (maxsize >> l) > 0; l++)
3260 ;
3261 *ap->a_retval = l + 1;
3262 } else
3263 #endif
3264 {
3265 *ap->a_retval = 32; /* NFS V2 limitation */
3266 }
3267 break;
3268 default:
3269 error = EINVAL;
3270 break;
3271 }
3272
3273 return (error);
3274 }
3275
3276 /*
3277 * NFS advisory byte-level locks.
3278 */
3279 int
3280 nfs_advlock(void *v)
3281 {
3282 struct vop_advlock_args /* {
3283 struct vnode *a_vp;
3284 void *a_id;
3285 int a_op;
3286 struct flock *a_fl;
3287 int a_flags;
3288 } */ *ap = v;
3289 struct nfsnode *np = VTONFS(ap->a_vp);
3290
3291 return lf_advlock(ap, &np->n_lockf, np->n_size);
3292 }
3293
3294 /*
3295 * Print out the contents of an nfsnode.
3296 */
3297 int
3298 nfs_print(void *v)
3299 {
3300 struct vop_print_args /* {
3301 struct vnode *a_vp;
3302 } */ *ap = v;
3303 struct vnode *vp = ap->a_vp;
3304 struct nfsnode *np = VTONFS(vp);
3305
3306 printf("tag VT_NFS, fileid %lld fsid 0x%llx",
3307 (unsigned long long)np->n_vattr->va_fileid,
3308 (unsigned long long)np->n_vattr->va_fsid);
3309 if (vp->v_type == VFIFO)
3310 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
3311 printf("\n");
3312 return (0);
3313 }
3314
3315 /*
3316 * nfs unlock wrapper.
3317 */
3318 int
3319 nfs_unlock(void *v)
3320 {
3321 struct vop_unlock_args /* {
3322 struct vnode *a_vp;
3323 int a_flags;
3324 } */ *ap = v;
3325 struct vnode *vp = ap->a_vp;
3326
3327 /*
3328 * VOP_UNLOCK can be called by nfs_loadattrcache
3329 * with v_data == 0.
3330 */
3331 if (VTONFS(vp)) {
3332 nfs_delayedtruncate(vp);
3333 }
3334
3335 return genfs_unlock(v);
3336 }
3337
3338 /*
3339 * nfs special file access vnode op.
3340 * Essentially just get vattr and then imitate iaccess() since the device is
3341 * local to the client.
3342 */
3343 int
3344 nfsspec_access(void *v)
3345 {
3346 struct vop_access_args /* {
3347 struct vnode *a_vp;
3348 int a_mode;
3349 kauth_cred_t a_cred;
3350 struct lwp *a_l;
3351 } */ *ap = v;
3352 struct vattr va;
3353 struct vnode *vp = ap->a_vp;
3354 int error;
3355
3356 error = VOP_GETATTR(vp, &va, ap->a_cred);
3357 if (error)
3358 return (error);
3359
3360 /*
3361 * Disallow write attempts on filesystems mounted read-only;
3362 * unless the file is a socket, fifo, or a block or character
3363 * device resident on the filesystem.
3364 */
3365 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3366 switch (vp->v_type) {
3367 case VREG:
3368 case VDIR:
3369 case VLNK:
3370 return (EROFS);
3371 default:
3372 break;
3373 }
3374 }
3375
3376 return kauth_authorize_vnode(ap->a_cred, kauth_access_action(ap->a_mode,
3377 va.va_type, va.va_mode), vp, NULL, genfs_can_access(va.va_type,
3378 va.va_mode, va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3379 }
3380
3381 /*
3382 * Read wrapper for special devices.
3383 */
3384 int
3385 nfsspec_read(void *v)
3386 {
3387 struct vop_read_args /* {
3388 struct vnode *a_vp;
3389 struct uio *a_uio;
3390 int a_ioflag;
3391 kauth_cred_t a_cred;
3392 } */ *ap = v;
3393 struct nfsnode *np = VTONFS(ap->a_vp);
3394
3395 /*
3396 * Set access flag.
3397 */
3398 np->n_flag |= NACC;
3399 getnanotime(&np->n_atim);
3400 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3401 }
3402
3403 /*
3404 * Write wrapper for special devices.
3405 */
3406 int
3407 nfsspec_write(void *v)
3408 {
3409 struct vop_write_args /* {
3410 struct vnode *a_vp;
3411 struct uio *a_uio;
3412 int a_ioflag;
3413 kauth_cred_t a_cred;
3414 } */ *ap = v;
3415 struct nfsnode *np = VTONFS(ap->a_vp);
3416
3417 /*
3418 * Set update flag.
3419 */
3420 np->n_flag |= NUPD;
3421 getnanotime(&np->n_mtim);
3422 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3423 }
3424
3425 /*
3426 * Close wrapper for special devices.
3427 *
3428 * Update the times on the nfsnode then do device close.
3429 */
3430 int
3431 nfsspec_close(void *v)
3432 {
3433 struct vop_close_args /* {
3434 struct vnode *a_vp;
3435 int a_fflag;
3436 kauth_cred_t a_cred;
3437 struct lwp *a_l;
3438 } */ *ap = v;
3439 struct vnode *vp = ap->a_vp;
3440 struct nfsnode *np = VTONFS(vp);
3441 struct vattr vattr;
3442
3443 if (np->n_flag & (NACC | NUPD)) {
3444 np->n_flag |= NCHG;
3445 if (vp->v_usecount == 1 &&
3446 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3447 vattr_null(&vattr);
3448 if (np->n_flag & NACC)
3449 vattr.va_atime = np->n_atim;
3450 if (np->n_flag & NUPD)
3451 vattr.va_mtime = np->n_mtim;
3452 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3453 }
3454 }
3455 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3456 }
3457
3458 /*
3459 * Read wrapper for fifos.
3460 */
3461 int
3462 nfsfifo_read(void *v)
3463 {
3464 struct vop_read_args /* {
3465 struct vnode *a_vp;
3466 struct uio *a_uio;
3467 int a_ioflag;
3468 kauth_cred_t a_cred;
3469 } */ *ap = v;
3470 struct nfsnode *np = VTONFS(ap->a_vp);
3471
3472 /*
3473 * Set access flag.
3474 */
3475 np->n_flag |= NACC;
3476 getnanotime(&np->n_atim);
3477 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3478 }
3479
3480 /*
3481 * Write wrapper for fifos.
3482 */
3483 int
3484 nfsfifo_write(void *v)
3485 {
3486 struct vop_write_args /* {
3487 struct vnode *a_vp;
3488 struct uio *a_uio;
3489 int a_ioflag;
3490 kauth_cred_t a_cred;
3491 } */ *ap = v;
3492 struct nfsnode *np = VTONFS(ap->a_vp);
3493
3494 /*
3495 * Set update flag.
3496 */
3497 np->n_flag |= NUPD;
3498 getnanotime(&np->n_mtim);
3499 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3500 }
3501
3502 /*
3503 * Close wrapper for fifos.
3504 *
3505 * Update the times on the nfsnode then do fifo close.
3506 */
3507 int
3508 nfsfifo_close(void *v)
3509 {
3510 struct vop_close_args /* {
3511 struct vnode *a_vp;
3512 int a_fflag;
3513 kauth_cred_t a_cred;
3514 struct lwp *a_l;
3515 } */ *ap = v;
3516 struct vnode *vp = ap->a_vp;
3517 struct nfsnode *np = VTONFS(vp);
3518 struct vattr vattr;
3519
3520 if (np->n_flag & (NACC | NUPD)) {
3521 struct timespec ts;
3522
3523 getnanotime(&ts);
3524 if (np->n_flag & NACC)
3525 np->n_atim = ts;
3526 if (np->n_flag & NUPD)
3527 np->n_mtim = ts;
3528 np->n_flag |= NCHG;
3529 if (vp->v_usecount == 1 &&
3530 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3531 vattr_null(&vattr);
3532 if (np->n_flag & NACC)
3533 vattr.va_atime = np->n_atim;
3534 if (np->n_flag & NUPD)
3535 vattr.va_mtime = np->n_mtim;
3536 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3537 }
3538 }
3539 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3540 }
3541