nfs_vnops.c revision 1.313 1 /* $NetBSD: nfs_vnops.c,v 1.313 2020/02/23 15:46:41 ad 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 1.313 2020/02/23 15:46:41 ad 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_fallocate_desc, genfs_eopnotsupp }, /* fallocate */
111 { &vop_fdiscard_desc, genfs_eopnotsupp }, /* fdiscard */
112 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
113 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
114 { &vop_poll_desc, nfs_poll }, /* poll */
115 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
116 { &vop_revoke_desc, nfs_revoke }, /* revoke */
117 { &vop_mmap_desc, nfs_mmap }, /* mmap */
118 { &vop_fsync_desc, nfs_fsync }, /* fsync */
119 { &vop_seek_desc, nfs_seek }, /* seek */
120 { &vop_remove_desc, nfs_remove }, /* remove */
121 { &vop_link_desc, nfs_link }, /* link */
122 { &vop_rename_desc, nfs_rename }, /* rename */
123 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
124 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
125 { &vop_symlink_desc, nfs_symlink }, /* symlink */
126 { &vop_readdir_desc, nfs_readdir }, /* readdir */
127 { &vop_readlink_desc, nfs_readlink }, /* readlink */
128 { &vop_abortop_desc, nfs_abortop }, /* abortop */
129 { &vop_inactive_desc, nfs_inactive }, /* inactive */
130 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
131 { &vop_lock_desc, nfs_lock }, /* lock */
132 { &vop_unlock_desc, nfs_unlock }, /* unlock */
133 { &vop_bmap_desc, nfs_bmap }, /* bmap */
134 { &vop_strategy_desc, nfs_strategy }, /* strategy */
135 { &vop_print_desc, nfs_print }, /* print */
136 { &vop_islocked_desc, nfs_islocked }, /* islocked */
137 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
138 { &vop_advlock_desc, nfs_advlock }, /* advlock */
139 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
140 { &vop_getpages_desc, nfs_getpages }, /* getpages */
141 { &vop_putpages_desc, genfs_putpages }, /* putpages */
142 { NULL, NULL }
143 };
144 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
145 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
146
147 /*
148 * Special device vnode ops
149 */
150 int (**spec_nfsv2nodeop_p)(void *);
151 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
152 { &vop_default_desc, vn_default_error },
153 { &vop_lookup_desc, spec_lookup }, /* lookup */
154 { &vop_create_desc, spec_create }, /* create */
155 { &vop_mknod_desc, spec_mknod }, /* mknod */
156 { &vop_open_desc, spec_open }, /* open */
157 { &vop_close_desc, nfsspec_close }, /* close */
158 { &vop_access_desc, nfsspec_access }, /* access */
159 { &vop_getattr_desc, nfs_getattr }, /* getattr */
160 { &vop_setattr_desc, nfs_setattr }, /* setattr */
161 { &vop_read_desc, nfsspec_read }, /* read */
162 { &vop_write_desc, nfsspec_write }, /* write */
163 { &vop_fallocate_desc, spec_fallocate }, /* fallocate */
164 { &vop_fdiscard_desc, spec_fdiscard }, /* fdiscard */
165 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
166 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
167 { &vop_poll_desc, spec_poll }, /* poll */
168 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
169 { &vop_revoke_desc, spec_revoke }, /* revoke */
170 { &vop_mmap_desc, spec_mmap }, /* mmap */
171 { &vop_fsync_desc, spec_fsync }, /* fsync */
172 { &vop_seek_desc, spec_seek }, /* seek */
173 { &vop_remove_desc, spec_remove }, /* remove */
174 { &vop_link_desc, spec_link }, /* link */
175 { &vop_rename_desc, spec_rename }, /* rename */
176 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
177 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
178 { &vop_symlink_desc, spec_symlink }, /* symlink */
179 { &vop_readdir_desc, spec_readdir }, /* readdir */
180 { &vop_readlink_desc, spec_readlink }, /* readlink */
181 { &vop_abortop_desc, spec_abortop }, /* abortop */
182 { &vop_inactive_desc, nfs_inactive }, /* inactive */
183 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
184 { &vop_lock_desc, nfs_lock }, /* lock */
185 { &vop_unlock_desc, nfs_unlock }, /* unlock */
186 { &vop_bmap_desc, spec_bmap }, /* bmap */
187 { &vop_strategy_desc, spec_strategy }, /* strategy */
188 { &vop_print_desc, nfs_print }, /* print */
189 { &vop_islocked_desc, nfs_islocked }, /* islocked */
190 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
191 { &vop_advlock_desc, spec_advlock }, /* advlock */
192 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */
193 { &vop_getpages_desc, spec_getpages }, /* getpages */
194 { &vop_putpages_desc, spec_putpages }, /* putpages */
195 { NULL, NULL }
196 };
197 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
198 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
199
200 int (**fifo_nfsv2nodeop_p)(void *);
201 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
202 { &vop_default_desc, vn_default_error },
203 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */
204 { &vop_create_desc, vn_fifo_bypass }, /* create */
205 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */
206 { &vop_open_desc, vn_fifo_bypass }, /* open */
207 { &vop_close_desc, nfsfifo_close }, /* close */
208 { &vop_access_desc, nfsspec_access }, /* access */
209 { &vop_getattr_desc, nfs_getattr }, /* getattr */
210 { &vop_setattr_desc, nfs_setattr }, /* setattr */
211 { &vop_read_desc, nfsfifo_read }, /* read */
212 { &vop_write_desc, nfsfifo_write }, /* write */
213 { &vop_fallocate_desc, vn_fifo_bypass }, /* fallocate */
214 { &vop_fdiscard_desc, vn_fifo_bypass }, /* fdiscard */
215 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
216 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
217 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
218 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
219 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */
220 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */
221 { &vop_fsync_desc, nfs_fsync }, /* fsync */
222 { &vop_seek_desc, vn_fifo_bypass }, /* seek */
223 { &vop_remove_desc, vn_fifo_bypass }, /* remove */
224 { &vop_link_desc, vn_fifo_bypass }, /* link */
225 { &vop_rename_desc, vn_fifo_bypass }, /* rename */
226 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */
227 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */
228 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */
229 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */
230 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */
231 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */
232 { &vop_inactive_desc, nfs_inactive }, /* inactive */
233 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
234 { &vop_lock_desc, nfs_lock }, /* lock */
235 { &vop_unlock_desc, nfs_unlock }, /* unlock */
236 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */
237 { &vop_strategy_desc, genfs_badop }, /* strategy */
238 { &vop_print_desc, nfs_print }, /* print */
239 { &vop_islocked_desc, nfs_islocked }, /* islocked */
240 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
241 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */
242 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
243 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */
244 { NULL, NULL }
245 };
246 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
247 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
248
249 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
250 size_t, kauth_cred_t, struct lwp *);
251 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *);
252
253 /*
254 * Global variables
255 */
256 extern u_int32_t nfs_true, nfs_false;
257 extern u_int32_t nfs_xdrneg1;
258 extern const nfstype nfsv3_type[9];
259
260 int nfs_numasync = 0;
261 #define DIRHDSIZ _DIRENT_NAMEOFF(dp)
262 #define UIO_ADVANCE(uio, siz) \
263 (void)((uio)->uio_resid -= (siz), \
264 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
265 (uio)->uio_iov->iov_len -= (siz))
266
267 static void nfs_cache_enter(struct vnode *, struct vnode *,
268 struct componentname *);
269
270 static void
271 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
272 struct componentname *cnp)
273 {
274 struct nfsnode *dnp = VTONFS(dvp);
275
276 if ((cnp->cn_flags & MAKEENTRY) == 0) {
277 return;
278 }
279 if (vp != NULL) {
280 struct nfsnode *np = VTONFS(vp);
281
282 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
283 }
284
285 if (!timespecisset(&dnp->n_nctime))
286 dnp->n_nctime = dnp->n_vattr->va_mtime;
287
288 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
289 }
290
291 /*
292 * nfs null call from vfs.
293 */
294 int
295 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l)
296 {
297 char *bpos, *dpos;
298 int error = 0;
299 struct mbuf *mreq, *mrep, *md, *mb __unused;
300 struct nfsnode *np = VTONFS(vp);
301
302 nfsm_reqhead(np, NFSPROC_NULL, 0);
303 nfsm_request(np, NFSPROC_NULL, l, cred);
304 nfsm_reqdone;
305 return (error);
306 }
307
308 /*
309 * nfs access vnode op.
310 * For nfs version 2, just return ok. File accesses may fail later.
311 * For nfs version 3, use the access rpc to check accessibility. If file modes
312 * are changed on the server, accesses might still fail later.
313 */
314 int
315 nfs_access(void *v)
316 {
317 struct vop_access_args /* {
318 struct vnode *a_vp;
319 int a_mode;
320 kauth_cred_t a_cred;
321 } */ *ap = v;
322 struct vnode *vp = ap->a_vp;
323 #ifndef NFS_V2_ONLY
324 u_int32_t *tl;
325 char *cp;
326 int32_t t1, t2;
327 char *bpos, *dpos, *cp2;
328 int error = 0, attrflag;
329 struct mbuf *mreq, *mrep, *md, *mb;
330 u_int32_t mode, rmode;
331 const int v3 = NFS_ISV3(vp);
332 #endif
333 int cachevalid;
334 struct nfsnode *np = VTONFS(vp);
335 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
336
337 cachevalid = (np->n_accstamp != -1 &&
338 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
339 np->n_accuid == kauth_cred_geteuid(ap->a_cred));
340
341 /*
342 * Check access cache first. If this request has been made for this
343 * uid shortly before, use the cached result.
344 */
345 if (cachevalid) {
346 if (!np->n_accerror) {
347 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
348 return np->n_accerror;
349 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
350 return np->n_accerror;
351 }
352
353 #ifndef NFS_V2_ONLY
354 /*
355 * For nfs v3, do an access rpc, otherwise you are stuck emulating
356 * ufs_access() locally using the vattr. This may not be correct,
357 * since the server may apply other access criteria such as
358 * client uid-->server uid mapping that we do not know about, but
359 * this is better than just returning anything that is lying about
360 * in the cache.
361 */
362 if (v3) {
363 nfsstats.rpccnt[NFSPROC_ACCESS]++;
364 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
365 nfsm_fhtom(np, v3);
366 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
367 if (ap->a_mode & VREAD)
368 mode = NFSV3ACCESS_READ;
369 else
370 mode = 0;
371 if (vp->v_type != VDIR) {
372 if (ap->a_mode & VWRITE)
373 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
374 if (ap->a_mode & VEXEC)
375 mode |= NFSV3ACCESS_EXECUTE;
376 } else {
377 if (ap->a_mode & VWRITE)
378 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
379 NFSV3ACCESS_DELETE);
380 if (ap->a_mode & VEXEC)
381 mode |= NFSV3ACCESS_LOOKUP;
382 }
383 *tl = txdr_unsigned(mode);
384 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
385 nfsm_postop_attr(vp, attrflag, 0);
386 if (!error) {
387 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
388 rmode = fxdr_unsigned(u_int32_t, *tl);
389 /*
390 * The NFS V3 spec does not clarify whether or not
391 * the returned access bits can be a superset of
392 * the ones requested, so...
393 */
394 if ((rmode & mode) != mode)
395 error = EACCES;
396 }
397 nfsm_reqdone;
398 } else
399 #endif
400 return (nfsspec_access(ap));
401 #ifndef NFS_V2_ONLY
402 /*
403 * Disallow write attempts on filesystems mounted read-only;
404 * unless the file is a socket, fifo, or a block or character
405 * device resident on the filesystem.
406 */
407 if (!error && (ap->a_mode & VWRITE) &&
408 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
409 switch (vp->v_type) {
410 case VREG:
411 case VDIR:
412 case VLNK:
413 error = EROFS;
414 default:
415 break;
416 }
417 }
418
419 if (!error || error == EACCES) {
420 /*
421 * If we got the same result as for a previous,
422 * different request, OR it in. Don't update
423 * the timestamp in that case.
424 */
425 if (cachevalid && np->n_accstamp != -1 &&
426 error == np->n_accerror) {
427 if (!error)
428 np->n_accmode |= ap->a_mode;
429 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
430 np->n_accmode = ap->a_mode;
431 } else {
432 np->n_accstamp = time_uptime;
433 np->n_accuid = kauth_cred_geteuid(ap->a_cred);
434 np->n_accmode = ap->a_mode;
435 np->n_accerror = error;
436 }
437 }
438
439 return (error);
440 #endif
441 }
442
443 /*
444 * nfs open vnode op
445 * Check to see if the type is ok
446 * and that deletion is not in progress.
447 * For paged in text files, you will need to flush the page cache
448 * if consistency is lost.
449 */
450 /* ARGSUSED */
451 int
452 nfs_open(void *v)
453 {
454 struct vop_open_args /* {
455 struct vnode *a_vp;
456 int a_mode;
457 kauth_cred_t a_cred;
458 } */ *ap = v;
459 struct vnode *vp = ap->a_vp;
460 struct nfsnode *np = VTONFS(vp);
461 int error;
462
463 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
464 return (EACCES);
465 }
466
467 if (ap->a_mode & FREAD) {
468 if (np->n_rcred != NULL)
469 kauth_cred_free(np->n_rcred);
470 np->n_rcred = ap->a_cred;
471 kauth_cred_hold(np->n_rcred);
472 }
473 if (ap->a_mode & FWRITE) {
474 if (np->n_wcred != NULL)
475 kauth_cred_free(np->n_wcred);
476 np->n_wcred = ap->a_cred;
477 kauth_cred_hold(np->n_wcred);
478 }
479
480 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
481 if (error)
482 return error;
483
484 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
485
486 return (0);
487 }
488
489 /*
490 * nfs close vnode op
491 * What an NFS client should do upon close after writing is a debatable issue.
492 * Most NFS clients push delayed writes to the server upon close, basically for
493 * two reasons:
494 * 1 - So that any write errors may be reported back to the client process
495 * doing the close system call. By far the two most likely errors are
496 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
497 * 2 - To put a worst case upper bound on cache inconsistency between
498 * multiple clients for the file.
499 * There is also a consistency problem for Version 2 of the protocol w.r.t.
500 * not being able to tell if other clients are writing a file concurrently,
501 * since there is no way of knowing if the changed modify time in the reply
502 * is only due to the write for this client.
503 * (NFS Version 3 provides weak cache consistency data in the reply that
504 * should be sufficient to detect and handle this case.)
505 *
506 * The current code does the following:
507 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
508 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
509 * or commit them (this satisfies 1 and 2 except for the
510 * case where the server crashes after this close but
511 * before the commit RPC, which is felt to be "good
512 * enough". Changing the last argument to nfs_flush() to
513 * a 1 would force a commit operation, if it is felt a
514 * commit is necessary now.
515 */
516 /* ARGSUSED */
517 int
518 nfs_close(void *v)
519 {
520 struct vop_close_args /* {
521 struct vnodeop_desc *a_desc;
522 struct vnode *a_vp;
523 int a_fflag;
524 kauth_cred_t a_cred;
525 } */ *ap = v;
526 struct vnode *vp = ap->a_vp;
527 struct nfsnode *np = VTONFS(vp);
528 int error = 0;
529 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
530
531 if (vp->v_type == VREG) {
532 if (np->n_flag & NMODIFIED) {
533 #ifndef NFS_V2_ONLY
534 if (NFS_ISV3(vp)) {
535 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
536 np->n_flag &= ~NMODIFIED;
537 } else
538 #endif
539 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
540 NFS_INVALIDATE_ATTRCACHE(np);
541 }
542 if (np->n_flag & NWRITEERR) {
543 np->n_flag &= ~NWRITEERR;
544 error = np->n_error;
545 }
546 }
547 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
548 return (error);
549 }
550
551 /*
552 * nfs getattr call from vfs.
553 */
554 int
555 nfs_getattr(void *v)
556 {
557 struct vop_getattr_args /* {
558 struct vnode *a_vp;
559 struct vattr *a_vap;
560 kauth_cred_t a_cred;
561 } */ *ap = v;
562 struct vnode *vp = ap->a_vp;
563 struct nfsnode *np = VTONFS(vp);
564 char *cp;
565 u_int32_t *tl;
566 int32_t t1, t2;
567 char *bpos, *dpos;
568 int error = 0;
569 struct mbuf *mreq, *mrep, *md, *mb;
570 const int v3 = NFS_ISV3(vp);
571
572 /*
573 * Update local times for special files.
574 */
575 if (np->n_flag & (NACC | NUPD))
576 np->n_flag |= NCHG;
577
578 /*
579 * if we have delayed truncation, do it now.
580 */
581 nfs_delayedtruncate(vp);
582
583 /*
584 * First look in the cache.
585 */
586 if (nfs_getattrcache(vp, ap->a_vap) == 0)
587 return (0);
588 nfsstats.rpccnt[NFSPROC_GETATTR]++;
589 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
590 nfsm_fhtom(np, v3);
591 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
592 if (!error) {
593 nfsm_loadattr(vp, ap->a_vap, 0);
594 if (vp->v_type == VDIR &&
595 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
596 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
597 }
598 nfsm_reqdone;
599 return (error);
600 }
601
602 /*
603 * nfs setattr call.
604 */
605 int
606 nfs_setattr(void *v)
607 {
608 struct vop_setattr_args /* {
609 struct vnodeop_desc *a_desc;
610 struct vnode *a_vp;
611 struct vattr *a_vap;
612 kauth_cred_t a_cred;
613 } */ *ap = v;
614 struct vnode *vp = ap->a_vp;
615 struct nfsnode *np = VTONFS(vp);
616 struct vattr *vap = ap->a_vap;
617 int error = 0;
618 u_quad_t tsize = 0;
619
620 /*
621 * Setting of flags is not supported.
622 */
623 if (vap->va_flags != VNOVAL)
624 return (EOPNOTSUPP);
625
626 /*
627 * Disallow write attempts if the filesystem is mounted read-only.
628 */
629 if ((vap->va_uid != (uid_t)VNOVAL ||
630 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
631 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
632 (vp->v_mount->mnt_flag & MNT_RDONLY))
633 return (EROFS);
634 if (vap->va_size != VNOVAL) {
635 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
636 return EFBIG;
637 }
638 switch (vp->v_type) {
639 case VDIR:
640 return (EISDIR);
641 case VCHR:
642 case VBLK:
643 case VSOCK:
644 case VFIFO:
645 if (vap->va_mtime.tv_sec == VNOVAL &&
646 vap->va_atime.tv_sec == VNOVAL &&
647 vap->va_mode == (mode_t)VNOVAL &&
648 vap->va_uid == (uid_t)VNOVAL &&
649 vap->va_gid == (gid_t)VNOVAL)
650 return (0);
651 vap->va_size = VNOVAL;
652 break;
653 default:
654 /*
655 * Disallow write attempts if the filesystem is
656 * mounted read-only.
657 */
658 if (vp->v_mount->mnt_flag & MNT_RDONLY)
659 return (EROFS);
660 genfs_node_wrlock(vp);
661 uvm_vnp_setsize(vp, vap->va_size);
662 tsize = np->n_size;
663 np->n_size = vap->va_size;
664 if (vap->va_size == 0)
665 error = nfs_vinvalbuf(vp, 0,
666 ap->a_cred, curlwp, 1);
667 else
668 error = nfs_vinvalbuf(vp, V_SAVE,
669 ap->a_cred, curlwp, 1);
670 if (error) {
671 uvm_vnp_setsize(vp, tsize);
672 genfs_node_unlock(vp);
673 return (error);
674 }
675 np->n_vattr->va_size = vap->va_size;
676 }
677 } else {
678 /*
679 * flush files before setattr because a later write of
680 * cached data might change timestamps or reset sugid bits
681 */
682 if ((vap->va_mtime.tv_sec != VNOVAL ||
683 vap->va_atime.tv_sec != VNOVAL ||
684 vap->va_mode != VNOVAL) &&
685 vp->v_type == VREG &&
686 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
687 curlwp, 1)) == EINTR)
688 return (error);
689 }
690 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
691 if (vap->va_size != VNOVAL) {
692 if (error) {
693 np->n_size = np->n_vattr->va_size = tsize;
694 uvm_vnp_setsize(vp, np->n_size);
695 }
696 genfs_node_unlock(vp);
697 }
698 VN_KNOTE(vp, NOTE_ATTRIB);
699 return (error);
700 }
701
702 /*
703 * Do an nfs setattr rpc.
704 */
705 int
706 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l)
707 {
708 struct nfsv2_sattr *sp;
709 char *cp;
710 int32_t t1, t2;
711 char *bpos, *dpos;
712 u_int32_t *tl;
713 int error = 0;
714 struct mbuf *mreq, *mrep, *md, *mb;
715 const int v3 = NFS_ISV3(vp);
716 struct nfsnode *np = VTONFS(vp);
717 #ifndef NFS_V2_ONLY
718 int wccflag = NFSV3_WCCRATTR;
719 char *cp2;
720 #endif
721
722 nfsstats.rpccnt[NFSPROC_SETATTR]++;
723 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
724 nfsm_fhtom(np, v3);
725 #ifndef NFS_V2_ONLY
726 if (v3) {
727 nfsm_v3attrbuild(vap, true);
728 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
729 *tl = nfs_false;
730 } else {
731 #endif
732 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
733 if (vap->va_mode == (mode_t)VNOVAL)
734 sp->sa_mode = nfs_xdrneg1;
735 else
736 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
737 if (vap->va_uid == (uid_t)VNOVAL)
738 sp->sa_uid = nfs_xdrneg1;
739 else
740 sp->sa_uid = txdr_unsigned(vap->va_uid);
741 if (vap->va_gid == (gid_t)VNOVAL)
742 sp->sa_gid = nfs_xdrneg1;
743 else
744 sp->sa_gid = txdr_unsigned(vap->va_gid);
745 sp->sa_size = txdr_unsigned(vap->va_size);
746 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
747 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
748 #ifndef NFS_V2_ONLY
749 }
750 #endif
751 nfsm_request(np, NFSPROC_SETATTR, l, cred);
752 #ifndef NFS_V2_ONLY
753 if (v3) {
754 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
755 } else
756 #endif
757 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
758 nfsm_reqdone;
759 return (error);
760 }
761
762 /*
763 * nfs lookup call, one step at a time...
764 * First look in cache
765 * If not found, do the rpc.
766 */
767 int
768 nfs_lookup(void *v)
769 {
770 struct vop_lookup_v2_args /* {
771 struct vnodeop_desc *a_desc;
772 struct vnode *a_dvp;
773 struct vnode **a_vpp;
774 struct componentname *a_cnp;
775 } */ *ap = v;
776 struct componentname *cnp = ap->a_cnp;
777 struct vnode *dvp = ap->a_dvp;
778 struct vnode **vpp = ap->a_vpp;
779 int flags;
780 struct vnode *newvp;
781 u_int32_t *tl;
782 char *cp;
783 int32_t t1, t2;
784 char *bpos, *dpos, *cp2;
785 struct mbuf *mreq, *mrep, *md, *mb;
786 long len;
787 nfsfh_t *fhp;
788 struct nfsnode *np;
789 int cachefound;
790 int error = 0, attrflag, fhsize;
791 const int v3 = NFS_ISV3(dvp);
792
793 flags = cnp->cn_flags;
794
795 *vpp = NULLVP;
796 newvp = NULLVP;
797 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
798 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
799 return (EROFS);
800 if (dvp->v_type != VDIR)
801 return (ENOTDIR);
802
803 /*
804 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
805 */
806 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
807 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
808 if (error)
809 return error;
810 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
811 return EISDIR;
812 vref(dvp);
813 *vpp = dvp;
814 return 0;
815 }
816
817 np = VTONFS(dvp);
818
819 /*
820 * Before performing an RPC, check the name cache to see if
821 * the directory/name pair we are looking for is known already.
822 * If the directory/name pair is found in the name cache,
823 * we have to ensure the directory has not changed from
824 * the time the cache entry has been created. If it has,
825 * the cache entry has to be ignored.
826 */
827 cachefound = cache_lookup_raw(dvp, cnp->cn_nameptr, cnp->cn_namelen,
828 cnp->cn_flags, NULL, vpp);
829 KASSERT(dvp != *vpp);
830 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0);
831 if (cachefound) {
832 struct vattr vattr;
833
834 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
835 if (error != 0) {
836 if (*vpp != NULLVP)
837 vrele(*vpp);
838 *vpp = NULLVP;
839 return error;
840 }
841
842 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
843 || timespeccmp(&vattr.va_mtime,
844 &VTONFS(dvp)->n_nctime, !=)) {
845 if (*vpp != NULLVP) {
846 vrele(*vpp);
847 *vpp = NULLVP;
848 }
849 cache_purge1(dvp, NULL, 0, PURGE_CHILDREN);
850 timespecclear(&np->n_nctime);
851 goto dorpc;
852 }
853
854 if (*vpp == NULLVP) {
855 /* namecache gave us a negative result */
856 error = ENOENT;
857 goto noentry;
858 }
859
860 /*
861 * investigate the vnode returned by cache_lookup_raw.
862 * if it isn't appropriate, do an rpc.
863 */
864 newvp = *vpp;
865 if ((flags & ISDOTDOT) != 0) {
866 VOP_UNLOCK(dvp);
867 }
868 error = vn_lock(newvp, LK_SHARED);
869 if ((flags & ISDOTDOT) != 0) {
870 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
871 }
872 if (error != 0) {
873 /* newvp has been reclaimed. */
874 vrele(newvp);
875 *vpp = NULLVP;
876 goto dorpc;
877 }
878 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
879 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
880 nfsstats.lookupcache_hits++;
881 KASSERT(newvp->v_type != VNON);
882 VOP_UNLOCK(newvp);
883 return (0);
884 }
885 cache_purge1(newvp, NULL, 0, PURGE_PARENTS);
886 vput(newvp);
887 *vpp = NULLVP;
888 }
889 dorpc:
890 #if 0
891 /*
892 * because nfsv3 has the same CREATE semantics as ours,
893 * we don't have to perform LOOKUPs beforehand.
894 *
895 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
896 * XXX although we have no way to know if O_EXCL is requested or not.
897 */
898
899 if (v3 && cnp->cn_nameiop == CREATE &&
900 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
901 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
902 return (EJUSTRETURN);
903 }
904 #endif /* 0 */
905
906 error = 0;
907 newvp = NULLVP;
908 nfsstats.lookupcache_misses++;
909 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
910 len = cnp->cn_namelen;
911 nfsm_reqhead(np, NFSPROC_LOOKUP,
912 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
913 nfsm_fhtom(np, v3);
914 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
915 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
916 if (error) {
917 nfsm_postop_attr(dvp, attrflag, 0);
918 m_freem(mrep);
919 goto nfsmout;
920 }
921 nfsm_getfh(fhp, fhsize, v3);
922
923 /*
924 * Handle RENAME case...
925 */
926 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
927 if (NFS_CMPFH(np, fhp, fhsize)) {
928 m_freem(mrep);
929 return (EISDIR);
930 }
931 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
932 if (error) {
933 m_freem(mrep);
934 return error;
935 }
936 newvp = NFSTOV(np);
937 #ifndef NFS_V2_ONLY
938 if (v3) {
939 nfsm_postop_attr(newvp, attrflag, 0);
940 nfsm_postop_attr(dvp, attrflag, 0);
941 } else
942 #endif
943 nfsm_loadattr(newvp, (struct vattr *)0, 0);
944 *vpp = newvp;
945 m_freem(mrep);
946 goto validate;
947 }
948
949 /*
950 * The postop attr handling is duplicated for each if case,
951 * because it should be done while dvp is locked (unlocking
952 * dvp is different for each case).
953 */
954
955 if (NFS_CMPFH(np, fhp, fhsize)) {
956 /*
957 * As we handle "." lookup locally, this is
958 * a broken server.
959 */
960 m_freem(mrep);
961 return EBADRPC;
962 } else if (flags & ISDOTDOT) {
963 /*
964 * ".." lookup
965 */
966 VOP_UNLOCK(dvp);
967 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
968 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
969 if (error) {
970 m_freem(mrep);
971 return error;
972 }
973 newvp = NFSTOV(np);
974
975 #ifndef NFS_V2_ONLY
976 if (v3) {
977 nfsm_postop_attr(newvp, attrflag, 0);
978 nfsm_postop_attr(dvp, attrflag, 0);
979 } else
980 #endif
981 nfsm_loadattr(newvp, (struct vattr *)0, 0);
982 } else {
983 /*
984 * Other lookups.
985 */
986 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
987 if (error) {
988 m_freem(mrep);
989 return error;
990 }
991 newvp = NFSTOV(np);
992 #ifndef NFS_V2_ONLY
993 if (v3) {
994 nfsm_postop_attr(newvp, attrflag, 0);
995 nfsm_postop_attr(dvp, attrflag, 0);
996 } else
997 #endif
998 nfsm_loadattr(newvp, (struct vattr *)0, 0);
999 }
1000 if (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_nameiop != CREATE) {
1013 nfs_cache_enter(dvp, NULL, cnp);
1014 }
1015 if (newvp != NULLVP) {
1016 if (newvp == dvp) {
1017 vrele(newvp);
1018 } else {
1019 vput(newvp);
1020 }
1021 }
1022 noentry:
1023 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1024 (flags & ISLASTCN) && error == ENOENT) {
1025 if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1026 error = EROFS;
1027 } else {
1028 error = EJUSTRETURN;
1029 }
1030 }
1031 *vpp = NULL;
1032 return error;
1033 }
1034
1035 validate:
1036 /*
1037 * make sure we have valid type and size.
1038 */
1039
1040 newvp = *vpp;
1041 if (newvp->v_type == VNON) {
1042 struct vattr vattr; /* dummy */
1043
1044 KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1045 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1046 if (error) {
1047 vput(newvp);
1048 *vpp = NULL;
1049 }
1050 }
1051 if (error)
1052 return error;
1053 if (newvp != dvp)
1054 VOP_UNLOCK(newvp);
1055 return 0;
1056 }
1057
1058 /*
1059 * nfs read call.
1060 * Just call nfs_bioread() to do the work.
1061 */
1062 int
1063 nfs_read(void *v)
1064 {
1065 struct vop_read_args /* {
1066 struct vnode *a_vp;
1067 struct uio *a_uio;
1068 int a_ioflag;
1069 kauth_cred_t a_cred;
1070 } */ *ap = v;
1071 struct vnode *vp = ap->a_vp;
1072
1073 if (vp->v_type != VREG)
1074 return EISDIR;
1075 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1076 }
1077
1078 /*
1079 * nfs readlink call
1080 */
1081 int
1082 nfs_readlink(void *v)
1083 {
1084 struct vop_readlink_args /* {
1085 struct vnode *a_vp;
1086 struct uio *a_uio;
1087 kauth_cred_t a_cred;
1088 } */ *ap = v;
1089 struct vnode *vp = ap->a_vp;
1090 struct nfsnode *np = VTONFS(vp);
1091
1092 if (vp->v_type != VLNK)
1093 return (EPERM);
1094
1095 if (np->n_rcred != NULL) {
1096 kauth_cred_free(np->n_rcred);
1097 }
1098 np->n_rcred = ap->a_cred;
1099 kauth_cred_hold(np->n_rcred);
1100
1101 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1102 }
1103
1104 /*
1105 * Do a readlink rpc.
1106 * Called by nfs_doio() from below the buffer cache.
1107 */
1108 int
1109 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
1110 {
1111 u_int32_t *tl;
1112 char *cp;
1113 int32_t t1, t2;
1114 char *bpos, *dpos, *cp2;
1115 int error = 0;
1116 uint32_t len;
1117 struct mbuf *mreq, *mrep, *md, *mb;
1118 const int v3 = NFS_ISV3(vp);
1119 struct nfsnode *np = VTONFS(vp);
1120 #ifndef NFS_V2_ONLY
1121 int attrflag;
1122 #endif
1123
1124 nfsstats.rpccnt[NFSPROC_READLINK]++;
1125 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1126 nfsm_fhtom(np, v3);
1127 nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1128 #ifndef NFS_V2_ONLY
1129 if (v3)
1130 nfsm_postop_attr(vp, attrflag, 0);
1131 #endif
1132 if (!error) {
1133 #ifndef NFS_V2_ONLY
1134 if (v3) {
1135 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1136 len = fxdr_unsigned(uint32_t, *tl);
1137 if (len > NFS_MAXPATHLEN) {
1138 /*
1139 * this pathname is too long for us.
1140 */
1141 m_freem(mrep);
1142 /* Solaris returns EINVAL. should we follow? */
1143 error = ENAMETOOLONG;
1144 goto nfsmout;
1145 }
1146 } else
1147 #endif
1148 {
1149 nfsm_strsiz(len, NFS_MAXPATHLEN);
1150 }
1151 nfsm_mtouio(uiop, len);
1152 }
1153 nfsm_reqdone;
1154 return (error);
1155 }
1156
1157 /*
1158 * nfs read rpc call
1159 * Ditto above
1160 */
1161 int
1162 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1163 {
1164 u_int32_t *tl;
1165 char *cp;
1166 int32_t t1, t2;
1167 char *bpos, *dpos, *cp2;
1168 struct mbuf *mreq, *mrep, *md, *mb;
1169 struct nfsmount *nmp;
1170 int error = 0, len, retlen, tsiz, eof __unused, byte_count;
1171 const int v3 = NFS_ISV3(vp);
1172 struct nfsnode *np = VTONFS(vp);
1173 #ifndef NFS_V2_ONLY
1174 int attrflag;
1175 #endif
1176
1177 #ifndef nolint
1178 eof = 0;
1179 #endif
1180 nmp = VFSTONFS(vp->v_mount);
1181 tsiz = uiop->uio_resid;
1182 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1183 return (EFBIG);
1184 iostat_busy(nmp->nm_stats);
1185 byte_count = 0; /* count bytes actually transferred */
1186 while (tsiz > 0) {
1187 nfsstats.rpccnt[NFSPROC_READ]++;
1188 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1189 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1190 nfsm_fhtom(np, v3);
1191 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1192 #ifndef NFS_V2_ONLY
1193 if (v3) {
1194 txdr_hyper(uiop->uio_offset, tl);
1195 *(tl + 2) = txdr_unsigned(len);
1196 } else
1197 #endif
1198 {
1199 *tl++ = txdr_unsigned(uiop->uio_offset);
1200 *tl++ = txdr_unsigned(len);
1201 *tl = 0;
1202 }
1203 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1204 #ifndef NFS_V2_ONLY
1205 if (v3) {
1206 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1207 if (error) {
1208 m_freem(mrep);
1209 goto nfsmout;
1210 }
1211 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1212 eof = fxdr_unsigned(int, *(tl + 1));
1213 } else
1214 #endif
1215 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1216 nfsm_strsiz(retlen, nmp->nm_rsize);
1217 nfsm_mtouio(uiop, retlen);
1218 m_freem(mrep);
1219 tsiz -= retlen;
1220 byte_count += retlen;
1221 #ifndef NFS_V2_ONLY
1222 if (v3) {
1223 if (eof || retlen == 0)
1224 tsiz = 0;
1225 } else
1226 #endif
1227 if (retlen < len)
1228 tsiz = 0;
1229 }
1230 nfsmout:
1231 iostat_unbusy(nmp->nm_stats, byte_count, 1);
1232 return (error);
1233 }
1234
1235 struct nfs_writerpc_context {
1236 kmutex_t nwc_lock;
1237 kcondvar_t nwc_cv;
1238 int nwc_mbufcount;
1239 };
1240
1241 /*
1242 * free mbuf used to refer protected pages while write rpc call.
1243 * called at splvm.
1244 */
1245 static void
1246 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1247 {
1248 struct nfs_writerpc_context *ctx = arg;
1249
1250 KASSERT(m != NULL);
1251 KASSERT(ctx != NULL);
1252 pool_cache_put(mb_cache, m);
1253 mutex_enter(&ctx->nwc_lock);
1254 if (--ctx->nwc_mbufcount == 0) {
1255 cv_signal(&ctx->nwc_cv);
1256 }
1257 mutex_exit(&ctx->nwc_lock);
1258 }
1259
1260 /*
1261 * nfs write call
1262 */
1263 int
1264 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp)
1265 {
1266 u_int32_t *tl;
1267 char *cp;
1268 int32_t t1, t2;
1269 char *bpos, *dpos;
1270 struct mbuf *mreq, *mrep, *md, *mb;
1271 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1272 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1273 const int v3 = NFS_ISV3(vp);
1274 int committed = NFSV3WRITE_FILESYNC;
1275 struct nfsnode *np = VTONFS(vp);
1276 struct nfs_writerpc_context ctx;
1277 int byte_count;
1278 size_t origresid;
1279 #ifndef NFS_V2_ONLY
1280 char *cp2;
1281 int rlen, commit;
1282 #endif
1283
1284 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1285 panic("writerpc readonly vp %p", vp);
1286 }
1287
1288 #ifdef DIAGNOSTIC
1289 if (uiop->uio_iovcnt != 1)
1290 panic("nfs: writerpc iovcnt > 1");
1291 #endif
1292 tsiz = uiop->uio_resid;
1293 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1294 return EFBIG;
1295
1296 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1297 cv_init(&ctx.nwc_cv, "nfsmblk");
1298 ctx.nwc_mbufcount = 1;
1299
1300 retry:
1301 origresid = uiop->uio_resid;
1302 KASSERT(origresid == uiop->uio_iov->iov_len);
1303 iostat_busy(nmp->nm_stats);
1304 byte_count = 0; /* count of bytes actually written */
1305 while (tsiz > 0) {
1306 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1307 size_t backup;
1308 bool stalewriteverf = false;
1309
1310 nfsstats.rpccnt[NFSPROC_WRITE]++;
1311 len = uimin(tsiz, nmp->nm_wsize);
1312 datalen = pageprotected ? 0 : nfsm_rndup(len);
1313 nfsm_reqhead(np, NFSPROC_WRITE,
1314 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1315 nfsm_fhtom(np, v3);
1316 #ifndef NFS_V2_ONLY
1317 if (v3) {
1318 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1319 txdr_hyper(uiop->uio_offset, tl);
1320 tl += 2;
1321 *tl++ = txdr_unsigned(len);
1322 *tl++ = txdr_unsigned(*iomode);
1323 *tl = txdr_unsigned(len);
1324 } else
1325 #endif
1326 {
1327 u_int32_t x;
1328
1329 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1330 /* Set both "begin" and "current" to non-garbage. */
1331 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1332 *tl++ = x; /* "begin offset" */
1333 *tl++ = x; /* "current offset" */
1334 x = txdr_unsigned(len);
1335 *tl++ = x; /* total to this offset */
1336 *tl = x; /* size of this write */
1337
1338 }
1339 if (pageprotected) {
1340 /*
1341 * since we know pages can't be modified during i/o,
1342 * no need to copy them for us.
1343 */
1344 struct mbuf *m;
1345 struct iovec *iovp = uiop->uio_iov;
1346
1347 m = m_get(M_WAIT, MT_DATA);
1348 MCLAIM(m, &nfs_mowner);
1349 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1350 nfs_writerpc_extfree, &ctx);
1351 m->m_flags |= M_EXT_ROMAP;
1352 m->m_len = len;
1353 mb->m_next = m;
1354 /*
1355 * no need to maintain mb and bpos here
1356 * because no one care them later.
1357 */
1358 #if 0
1359 mb = m;
1360 bpos = mtod(void *, mb) + mb->m_len;
1361 #endif
1362 UIO_ADVANCE(uiop, len);
1363 uiop->uio_offset += len;
1364 mutex_enter(&ctx.nwc_lock);
1365 ctx.nwc_mbufcount++;
1366 mutex_exit(&ctx.nwc_lock);
1367 nfs_zeropad(mb, 0, nfsm_padlen(len));
1368 } else {
1369 nfsm_uiotom(uiop, len);
1370 }
1371 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1372 #ifndef NFS_V2_ONLY
1373 if (v3) {
1374 wccflag = NFSV3_WCCCHK;
1375 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1376 if (!error) {
1377 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1378 + NFSX_V3WRITEVERF);
1379 rlen = fxdr_unsigned(int, *tl++);
1380 if (rlen == 0) {
1381 error = NFSERR_IO;
1382 m_freem(mrep);
1383 break;
1384 } else if (rlen < len) {
1385 backup = len - rlen;
1386 UIO_ADVANCE(uiop, -backup);
1387 uiop->uio_offset -= backup;
1388 len = rlen;
1389 }
1390 commit = fxdr_unsigned(int, *tl++);
1391
1392 /*
1393 * Return the lowest committment level
1394 * obtained by any of the RPCs.
1395 */
1396 if (committed == NFSV3WRITE_FILESYNC)
1397 committed = commit;
1398 else if (committed == NFSV3WRITE_DATASYNC &&
1399 commit == NFSV3WRITE_UNSTABLE)
1400 committed = commit;
1401 mutex_enter(&nmp->nm_lock);
1402 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1403 memcpy(nmp->nm_writeverf, tl,
1404 NFSX_V3WRITEVERF);
1405 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1406 } else if ((nmp->nm_iflag &
1407 NFSMNT_STALEWRITEVERF) ||
1408 memcmp(tl, nmp->nm_writeverf,
1409 NFSX_V3WRITEVERF)) {
1410 memcpy(nmp->nm_writeverf, tl,
1411 NFSX_V3WRITEVERF);
1412 /*
1413 * note NFSMNT_STALEWRITEVERF
1414 * if we're the first thread to
1415 * notice it.
1416 */
1417 if ((nmp->nm_iflag &
1418 NFSMNT_STALEWRITEVERF) == 0) {
1419 stalewriteverf = true;
1420 nmp->nm_iflag |=
1421 NFSMNT_STALEWRITEVERF;
1422 }
1423 }
1424 mutex_exit(&nmp->nm_lock);
1425 }
1426 } else
1427 #endif
1428 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1429 if (wccflag)
1430 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1431 m_freem(mrep);
1432 if (error)
1433 break;
1434 tsiz -= len;
1435 byte_count += len;
1436 if (stalewriteverf) {
1437 *stalewriteverfp = true;
1438 stalewriteverf = false;
1439 if (committed == NFSV3WRITE_UNSTABLE &&
1440 len != origresid) {
1441 /*
1442 * if our write requests weren't atomic but
1443 * unstable, datas in previous iterations
1444 * might have already been lost now.
1445 * then, we should resend them to nfsd.
1446 */
1447 backup = origresid - tsiz;
1448 UIO_ADVANCE(uiop, -backup);
1449 uiop->uio_offset -= backup;
1450 tsiz = origresid;
1451 goto retry;
1452 }
1453 }
1454 }
1455 nfsmout:
1456 iostat_unbusy(nmp->nm_stats, byte_count, 0);
1457 if (pageprotected) {
1458 /*
1459 * wait until mbufs go away.
1460 * retransmitted mbufs can survive longer than rpc requests
1461 * themselves.
1462 */
1463 mutex_enter(&ctx.nwc_lock);
1464 ctx.nwc_mbufcount--;
1465 while (ctx.nwc_mbufcount > 0) {
1466 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1467 }
1468 mutex_exit(&ctx.nwc_lock);
1469 }
1470 mutex_destroy(&ctx.nwc_lock);
1471 cv_destroy(&ctx.nwc_cv);
1472 *iomode = committed;
1473 if (error)
1474 uiop->uio_resid = tsiz;
1475 return (error);
1476 }
1477
1478 /*
1479 * nfs mknod rpc
1480 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1481 * mode set to specify the file type and the size field for rdev.
1482 */
1483 int
1484 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap)
1485 {
1486 struct nfsv2_sattr *sp;
1487 u_int32_t *tl;
1488 char *cp;
1489 int32_t t1, t2;
1490 struct vnode *newvp = (struct vnode *)0;
1491 struct nfsnode *dnp, *np;
1492 char *cp2;
1493 char *bpos, *dpos;
1494 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1495 struct mbuf *mreq, *mrep, *md, *mb;
1496 u_int32_t rdev;
1497 const int v3 = NFS_ISV3(dvp);
1498
1499 if (vap->va_type == VCHR || vap->va_type == VBLK)
1500 rdev = txdr_unsigned(vap->va_rdev);
1501 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1502 rdev = nfs_xdrneg1;
1503 else {
1504 VOP_ABORTOP(dvp, cnp);
1505 return (EOPNOTSUPP);
1506 }
1507 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1508 dnp = VTONFS(dvp);
1509 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1510 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1511 nfsm_fhtom(dnp, v3);
1512 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1513 #ifndef NFS_V2_ONLY
1514 if (v3) {
1515 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1516 *tl++ = vtonfsv3_type(vap->va_type);
1517 nfsm_v3attrbuild(vap, false);
1518 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1519 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1520 *tl++ = txdr_unsigned(major(vap->va_rdev));
1521 *tl = txdr_unsigned(minor(vap->va_rdev));
1522 }
1523 } else
1524 #endif
1525 {
1526 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1527 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1528 sp->sa_uid = nfs_xdrneg1;
1529 sp->sa_gid = nfs_xdrneg1;
1530 sp->sa_size = rdev;
1531 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1532 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1533 }
1534 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1535 if (!error) {
1536 nfsm_mtofh(dvp, newvp, v3, gotvp);
1537 if (!gotvp) {
1538 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1539 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1540 if (!error)
1541 newvp = NFSTOV(np);
1542 }
1543 }
1544 #ifndef NFS_V2_ONLY
1545 if (v3)
1546 nfsm_wcc_data(dvp, wccflag, 0, !error);
1547 #endif
1548 nfsm_reqdone;
1549 if (error) {
1550 if (newvp)
1551 vput(newvp);
1552 } else {
1553 nfs_cache_enter(dvp, newvp, cnp);
1554 *vpp = newvp;
1555 VOP_UNLOCK(newvp);
1556 }
1557 VTONFS(dvp)->n_flag |= NMODIFIED;
1558 if (!wccflag)
1559 NFS_INVALIDATE_ATTRCACHE(VTONFS(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_v3_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->cn_nameptr, cnp->cn_namelen, 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_v3_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->cn_nameptr, cnp->cn_namelen, 0);
1712 *ap->a_vpp = newvp;
1713 VOP_UNLOCK(newvp);
1714 } else {
1715 if (newvp)
1716 vput(newvp);
1717 if (error == EEXIST)
1718 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0);
1719 }
1720 VTONFS(dvp)->n_flag |= NMODIFIED;
1721 if (!wccflag)
1722 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1723 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
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_v2_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 return (error);
1794 }
1795
1796 /*
1797 * nfs file remove rpc called from nfs_inactive
1798 */
1799 int
1800 nfs_removeit(struct sillyrename *sp)
1801 {
1802
1803 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1804 (struct lwp *)0));
1805 }
1806
1807 /*
1808 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1809 */
1810 int
1811 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l)
1812 {
1813 u_int32_t *tl;
1814 char *cp;
1815 #ifndef NFS_V2_ONLY
1816 int32_t t1;
1817 char *cp2;
1818 #endif
1819 int32_t t2;
1820 char *bpos, *dpos;
1821 int error = 0, wccflag = NFSV3_WCCRATTR;
1822 struct mbuf *mreq, *mrep, *md, *mb;
1823 const int v3 = NFS_ISV3(dvp);
1824 int rexmit = 0;
1825 struct nfsnode *dnp = VTONFS(dvp);
1826
1827 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1828 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1829 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1830 nfsm_fhtom(dnp, v3);
1831 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1832 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1833 #ifndef NFS_V2_ONLY
1834 if (v3)
1835 nfsm_wcc_data(dvp, wccflag, 0, !error);
1836 #endif
1837 nfsm_reqdone;
1838 VTONFS(dvp)->n_flag |= NMODIFIED;
1839 if (!wccflag)
1840 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1841 /*
1842 * Kludge City: If the first reply to the remove rpc is lost..
1843 * the reply to the retransmitted request will be ENOENT
1844 * since the file was in fact removed
1845 * Therefore, we cheat and return success.
1846 */
1847 if (rexmit && error == ENOENT)
1848 error = 0;
1849 return (error);
1850 }
1851
1852 /*
1853 * nfs file rename call
1854 */
1855 int
1856 nfs_rename(void *v)
1857 {
1858 struct vop_rename_args /* {
1859 struct vnode *a_fdvp;
1860 struct vnode *a_fvp;
1861 struct componentname *a_fcnp;
1862 struct vnode *a_tdvp;
1863 struct vnode *a_tvp;
1864 struct componentname *a_tcnp;
1865 } */ *ap = v;
1866 struct vnode *fvp = ap->a_fvp;
1867 struct vnode *tvp = ap->a_tvp;
1868 struct vnode *fdvp = ap->a_fdvp;
1869 struct vnode *tdvp = ap->a_tdvp;
1870 struct componentname *tcnp = ap->a_tcnp;
1871 struct componentname *fcnp = ap->a_fcnp;
1872 int error;
1873
1874 /* Check for cross-device rename */
1875 if ((fvp->v_mount != tdvp->v_mount) ||
1876 (tvp && (fvp->v_mount != tvp->v_mount))) {
1877 error = EXDEV;
1878 goto out;
1879 }
1880
1881 /*
1882 * If the tvp exists and is in use, sillyrename it before doing the
1883 * rename of the new file over it.
1884 *
1885 * Have sillyrename use link instead of rename if possible,
1886 * so that we don't lose the file if the rename fails, and so
1887 * that there's no window when the "to" file doesn't exist.
1888 */
1889 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1890 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1891 VN_KNOTE(tvp, NOTE_DELETE);
1892 vput(tvp);
1893 tvp = NULL;
1894 }
1895
1896 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1897 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1898 curlwp);
1899
1900 VN_KNOTE(fdvp, NOTE_WRITE);
1901 VN_KNOTE(tdvp, NOTE_WRITE);
1902 if (error == 0 || error == EEXIST) {
1903 if (fvp->v_type == VDIR)
1904 cache_purge(fvp);
1905 else
1906 cache_purge1(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1907 0);
1908 if (tvp != NULL && tvp->v_type == VDIR)
1909 cache_purge(tvp);
1910 else
1911 cache_purge1(tdvp, tcnp->cn_nameptr, tcnp->cn_namelen,
1912 0);
1913 }
1914 out:
1915 if (tdvp == tvp)
1916 vrele(tdvp);
1917 else
1918 vput(tdvp);
1919 if (tvp)
1920 vput(tvp);
1921 vrele(fdvp);
1922 vrele(fvp);
1923 return (error);
1924 }
1925
1926 /*
1927 * nfs file rename rpc called from nfs_remove() above
1928 */
1929 int
1930 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp)
1931 {
1932 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1933 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
1934 }
1935
1936 /*
1937 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1938 */
1939 int
1940 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)
1941 {
1942 u_int32_t *tl;
1943 char *cp;
1944 #ifndef NFS_V2_ONLY
1945 int32_t t1;
1946 char *cp2;
1947 #endif
1948 int32_t t2;
1949 char *bpos, *dpos;
1950 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1951 struct mbuf *mreq, *mrep, *md, *mb;
1952 const int v3 = NFS_ISV3(fdvp);
1953 int rexmit = 0;
1954 struct nfsnode *fdnp = VTONFS(fdvp);
1955
1956 nfsstats.rpccnt[NFSPROC_RENAME]++;
1957 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1958 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1959 nfsm_rndup(tnamelen));
1960 nfsm_fhtom(fdnp, v3);
1961 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1962 nfsm_fhtom(VTONFS(tdvp), v3);
1963 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1964 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
1965 #ifndef NFS_V2_ONLY
1966 if (v3) {
1967 nfsm_wcc_data(fdvp, fwccflag, 0, !error);
1968 nfsm_wcc_data(tdvp, twccflag, 0, !error);
1969 }
1970 #endif
1971 nfsm_reqdone;
1972 VTONFS(fdvp)->n_flag |= NMODIFIED;
1973 VTONFS(tdvp)->n_flag |= NMODIFIED;
1974 if (!fwccflag)
1975 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
1976 if (!twccflag)
1977 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
1978 /*
1979 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1980 */
1981 if (rexmit && error == ENOENT)
1982 error = 0;
1983 return (error);
1984 }
1985
1986 /*
1987 * NFS link RPC, called from nfs_link.
1988 * Assumes dvp and vp locked, and leaves them that way.
1989 */
1990
1991 static int
1992 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
1993 size_t namelen, kauth_cred_t cred, struct lwp *l)
1994 {
1995 u_int32_t *tl;
1996 char *cp;
1997 #ifndef NFS_V2_ONLY
1998 int32_t t1;
1999 char *cp2;
2000 #endif
2001 int32_t t2;
2002 char *bpos, *dpos;
2003 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2004 struct mbuf *mreq, *mrep, *md, *mb;
2005 const int v3 = NFS_ISV3(dvp);
2006 int rexmit = 0;
2007 struct nfsnode *np = VTONFS(vp);
2008
2009 nfsstats.rpccnt[NFSPROC_LINK]++;
2010 nfsm_reqhead(np, NFSPROC_LINK,
2011 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2012 nfsm_fhtom(np, v3);
2013 nfsm_fhtom(VTONFS(dvp), v3);
2014 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2015 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2016 #ifndef NFS_V2_ONLY
2017 if (v3) {
2018 nfsm_postop_attr(vp, attrflag, 0);
2019 nfsm_wcc_data(dvp, wccflag, 0, !error);
2020 }
2021 #endif
2022 nfsm_reqdone;
2023
2024 VTONFS(dvp)->n_flag |= NMODIFIED;
2025 if (!attrflag)
2026 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2027 if (!wccflag)
2028 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2029
2030 /*
2031 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2032 */
2033 if (rexmit && error == EEXIST)
2034 error = 0;
2035
2036 return error;
2037 }
2038
2039 /*
2040 * nfs hard link create call
2041 */
2042 int
2043 nfs_link(void *v)
2044 {
2045 struct vop_link_v2_args /* {
2046 struct vnode *a_dvp;
2047 struct vnode *a_vp;
2048 struct componentname *a_cnp;
2049 } */ *ap = v;
2050 struct vnode *vp = ap->a_vp;
2051 struct vnode *dvp = ap->a_dvp;
2052 struct componentname *cnp = ap->a_cnp;
2053 int error = 0;
2054
2055 error = vn_lock(vp, LK_EXCLUSIVE);
2056 if (error != 0) {
2057 VOP_ABORTOP(dvp, cnp);
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->cn_nameptr, cnp->cn_namelen, 0);
2073 }
2074 VOP_UNLOCK(vp);
2075 VN_KNOTE(vp, NOTE_LINK);
2076 VN_KNOTE(dvp, NOTE_WRITE);
2077 return (error);
2078 }
2079
2080 /*
2081 * nfs symbolic link create call
2082 */
2083 int
2084 nfs_symlink(void *v)
2085 {
2086 struct vop_symlink_v3_args /* {
2087 struct vnode *a_dvp;
2088 struct vnode **a_vpp;
2089 struct componentname *a_cnp;
2090 struct vattr *a_vap;
2091 char *a_target;
2092 } */ *ap = v;
2093 struct vnode *dvp = ap->a_dvp;
2094 struct vattr *vap = ap->a_vap;
2095 struct componentname *cnp = ap->a_cnp;
2096 struct nfsv2_sattr *sp;
2097 u_int32_t *tl;
2098 char *cp;
2099 int32_t t1, t2;
2100 char *bpos, *dpos, *cp2;
2101 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2102 struct mbuf *mreq, *mrep, *md, *mb;
2103 struct vnode *newvp = (struct vnode *)0;
2104 const int v3 = NFS_ISV3(dvp);
2105 int rexmit = 0;
2106 struct nfsnode *dnp = VTONFS(dvp);
2107
2108 *ap->a_vpp = NULL;
2109 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2110 slen = strlen(ap->a_target);
2111 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2112 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2113 nfsm_fhtom(dnp, v3);
2114 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2115 #ifndef NFS_V2_ONlY
2116 if (v3)
2117 nfsm_v3attrbuild(vap, false);
2118 #endif
2119 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2120 #ifndef NFS_V2_ONlY
2121 if (!v3) {
2122 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2123 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2124 sp->sa_uid = nfs_xdrneg1;
2125 sp->sa_gid = nfs_xdrneg1;
2126 sp->sa_size = nfs_xdrneg1;
2127 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2128 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2129 }
2130 #endif
2131 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2132 &rexmit);
2133 #ifndef NFS_V2_ONlY
2134 if (v3) {
2135 if (!error)
2136 nfsm_mtofh(dvp, newvp, v3, gotvp);
2137 nfsm_wcc_data(dvp, wccflag, 0, !error);
2138 }
2139 #endif
2140 nfsm_reqdone;
2141 /*
2142 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2143 */
2144 if (rexmit && error == EEXIST)
2145 error = 0;
2146 if (error == 0 || error == EEXIST)
2147 cache_purge1(dvp, cnp->cn_nameptr, cnp->cn_namelen, 0);
2148 if (error == 0 && newvp == NULL) {
2149 struct nfsnode *np = NULL;
2150
2151 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2152 cnp->cn_cred, curlwp, &np);
2153 if (error == 0)
2154 newvp = NFSTOV(np);
2155 }
2156 if (error) {
2157 if (newvp != NULL)
2158 vput(newvp);
2159 } else {
2160 *ap->a_vpp = newvp;
2161 VOP_UNLOCK(newvp);
2162 }
2163 VTONFS(dvp)->n_flag |= NMODIFIED;
2164 if (!wccflag)
2165 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2166 VN_KNOTE(dvp, NOTE_WRITE);
2167 return (error);
2168 }
2169
2170 /*
2171 * nfs make dir call
2172 */
2173 int
2174 nfs_mkdir(void *v)
2175 {
2176 struct vop_mkdir_v3_args /* {
2177 struct vnode *a_dvp;
2178 struct vnode **a_vpp;
2179 struct componentname *a_cnp;
2180 struct vattr *a_vap;
2181 } */ *ap = v;
2182 struct vnode *dvp = ap->a_dvp;
2183 struct vattr *vap = ap->a_vap;
2184 struct componentname *cnp = ap->a_cnp;
2185 struct nfsv2_sattr *sp;
2186 u_int32_t *tl;
2187 char *cp;
2188 int32_t t1, t2;
2189 int len;
2190 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2191 struct vnode *newvp = (struct vnode *)0;
2192 char *bpos, *dpos, *cp2;
2193 int error = 0, wccflag = NFSV3_WCCRATTR;
2194 int gotvp = 0;
2195 int rexmit = 0;
2196 struct mbuf *mreq, *mrep, *md, *mb;
2197 const int v3 = NFS_ISV3(dvp);
2198
2199 len = cnp->cn_namelen;
2200 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2201 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2202 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2203 nfsm_fhtom(dnp, v3);
2204 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2205 #ifndef NFS_V2_ONLY
2206 if (v3) {
2207 nfsm_v3attrbuild(vap, false);
2208 } else
2209 #endif
2210 {
2211 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2212 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2213 sp->sa_uid = nfs_xdrneg1;
2214 sp->sa_gid = nfs_xdrneg1;
2215 sp->sa_size = nfs_xdrneg1;
2216 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2217 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2218 }
2219 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2220 if (!error)
2221 nfsm_mtofh(dvp, newvp, v3, gotvp);
2222 if (v3)
2223 nfsm_wcc_data(dvp, wccflag, 0, !error);
2224 nfsm_reqdone;
2225 VTONFS(dvp)->n_flag |= NMODIFIED;
2226 if (!wccflag)
2227 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2228 /*
2229 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2230 * if we can succeed in looking up the directory.
2231 */
2232 if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2233 if (newvp) {
2234 vput(newvp);
2235 newvp = (struct vnode *)0;
2236 }
2237 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2238 curlwp, &np);
2239 if (!error) {
2240 newvp = NFSTOV(np);
2241 if (newvp->v_type != VDIR || newvp == dvp)
2242 error = EEXIST;
2243 }
2244 }
2245 if (error) {
2246 if (newvp) {
2247 if (dvp != newvp)
2248 vput(newvp);
2249 else
2250 vrele(newvp);
2251 }
2252 } else {
2253 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2254 nfs_cache_enter(dvp, newvp, cnp);
2255 *ap->a_vpp = newvp;
2256 VOP_UNLOCK(newvp);
2257 }
2258 return (error);
2259 }
2260
2261 /*
2262 * nfs remove directory call
2263 */
2264 int
2265 nfs_rmdir(void *v)
2266 {
2267 struct vop_rmdir_v2_args /* {
2268 struct vnode *a_dvp;
2269 struct vnode *a_vp;
2270 struct componentname *a_cnp;
2271 } */ *ap = v;
2272 struct vnode *vp = ap->a_vp;
2273 struct vnode *dvp = ap->a_dvp;
2274 struct componentname *cnp = ap->a_cnp;
2275 u_int32_t *tl;
2276 char *cp;
2277 #ifndef NFS_V2_ONLY
2278 int32_t t1;
2279 char *cp2;
2280 #endif
2281 int32_t t2;
2282 char *bpos, *dpos;
2283 int error = 0, wccflag = NFSV3_WCCRATTR;
2284 int rexmit = 0;
2285 struct mbuf *mreq, *mrep, *md, *mb;
2286 const int v3 = NFS_ISV3(dvp);
2287 struct nfsnode *dnp;
2288
2289 if (dvp == vp) {
2290 vrele(vp);
2291 return (EINVAL);
2292 }
2293 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2294 dnp = VTONFS(dvp);
2295 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2296 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2297 nfsm_fhtom(dnp, v3);
2298 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2299 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2300 #ifndef NFS_V2_ONLY
2301 if (v3)
2302 nfsm_wcc_data(dvp, wccflag, 0, !error);
2303 #endif
2304 nfsm_reqdone;
2305 VTONFS(dvp)->n_flag |= NMODIFIED;
2306 if (!wccflag)
2307 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2308 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2309 VN_KNOTE(vp, NOTE_DELETE);
2310 cache_purge(vp);
2311 vput(vp);
2312 /*
2313 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2314 */
2315 if (rexmit && error == ENOENT)
2316 error = 0;
2317 return (error);
2318 }
2319
2320 /*
2321 * nfs readdir call
2322 */
2323 int
2324 nfs_readdir(void *v)
2325 {
2326 struct vop_readdir_args /* {
2327 struct vnode *a_vp;
2328 struct uio *a_uio;
2329 kauth_cred_t a_cred;
2330 int *a_eofflag;
2331 off_t **a_cookies;
2332 int *a_ncookies;
2333 } */ *ap = v;
2334 struct vnode *vp = ap->a_vp;
2335 struct uio *uio = ap->a_uio;
2336 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2337 char *base = uio->uio_iov->iov_base;
2338 int tresid, error;
2339 size_t count, lost;
2340 struct dirent *dp;
2341 off_t *cookies = NULL;
2342 int ncookies = 0, nc;
2343
2344 if (vp->v_type != VDIR)
2345 return (EPERM);
2346
2347 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2348 count = uio->uio_resid - lost;
2349 if (count <= 0)
2350 return (EINVAL);
2351
2352 /*
2353 * Call nfs_bioread() to do the real work.
2354 */
2355 tresid = uio->uio_resid = count;
2356 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2357 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2358
2359 if (!error && ap->a_cookies) {
2360 ncookies = count / 16;
2361 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2362 *ap->a_cookies = cookies;
2363 }
2364
2365 if (!error && uio->uio_resid == tresid) {
2366 uio->uio_resid += lost;
2367 nfsstats.direofcache_misses++;
2368 if (ap->a_cookies)
2369 *ap->a_ncookies = 0;
2370 *ap->a_eofflag = 1;
2371 return (0);
2372 }
2373
2374 if (!error && ap->a_cookies) {
2375 /*
2376 * Only the NFS server and emulations use cookies, and they
2377 * load the directory block into system space, so we can
2378 * just look at it directly.
2379 */
2380 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2381 uio->uio_iovcnt != 1)
2382 panic("nfs_readdir: lost in space");
2383 for (nc = 0; ncookies-- &&
2384 base < (char *)uio->uio_iov->iov_base; nc++){
2385 dp = (struct dirent *) base;
2386 if (dp->d_reclen == 0)
2387 break;
2388 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2389 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2390 else
2391 *(cookies++) = NFS_GETCOOKIE(dp);
2392 base += dp->d_reclen;
2393 }
2394 uio->uio_resid +=
2395 ((char *)uio->uio_iov->iov_base - base);
2396 uio->uio_iov->iov_len +=
2397 ((char *)uio->uio_iov->iov_base - base);
2398 uio->uio_iov->iov_base = base;
2399 *ap->a_ncookies = nc;
2400 }
2401
2402 uio->uio_resid += lost;
2403 *ap->a_eofflag = 0;
2404 return (error);
2405 }
2406
2407 /*
2408 * Readdir rpc call.
2409 * Called from below the buffer cache by nfs_doio().
2410 */
2411 int
2412 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
2413 {
2414 int len, left;
2415 struct dirent *dp = NULL;
2416 u_int32_t *tl;
2417 char *cp;
2418 int32_t t1, t2;
2419 char *bpos, *dpos, *cp2;
2420 struct mbuf *mreq, *mrep, *md, *mb;
2421 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2422 struct nfsnode *dnp = VTONFS(vp);
2423 u_quad_t fileno;
2424 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2425 #ifndef NFS_V2_ONLY
2426 int attrflag;
2427 #endif
2428 int nrpcs = 0, reclen;
2429 const int v3 = NFS_ISV3(vp);
2430
2431 #ifdef DIAGNOSTIC
2432 /*
2433 * Should be called from buffer cache, so only amount of
2434 * NFS_DIRBLKSIZ will be requested.
2435 */
2436 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2437 panic("nfs readdirrpc bad uio");
2438 #endif
2439
2440 /*
2441 * Loop around doing readdir rpc's of size nm_readdirsize
2442 * truncated to a multiple of NFS_DIRFRAGSIZ.
2443 * The stopping criteria is EOF or buffer full.
2444 */
2445 while (more_dirs && bigenough) {
2446 /*
2447 * Heuristic: don't bother to do another RPC to further
2448 * fill up this block if there is not much room left. (< 50%
2449 * of the readdir RPC size). This wastes some buffer space
2450 * but can save up to 50% in RPC calls.
2451 */
2452 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2453 bigenough = 0;
2454 break;
2455 }
2456 nfsstats.rpccnt[NFSPROC_READDIR]++;
2457 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2458 NFSX_READDIR(v3));
2459 nfsm_fhtom(dnp, v3);
2460 #ifndef NFS_V2_ONLY
2461 if (v3) {
2462 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2463 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2464 txdr_swapcookie3(uiop->uio_offset, tl);
2465 } else {
2466 txdr_cookie3(uiop->uio_offset, tl);
2467 }
2468 tl += 2;
2469 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2470 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2471 } else
2472 #endif
2473 {
2474 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2475 *tl++ = txdr_unsigned(uiop->uio_offset);
2476 }
2477 *tl = txdr_unsigned(nmp->nm_readdirsize);
2478 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2479 nrpcs++;
2480 #ifndef NFS_V2_ONLY
2481 if (v3) {
2482 nfsm_postop_attr(vp, attrflag, 0);
2483 if (!error) {
2484 nfsm_dissect(tl, u_int32_t *,
2485 2 * NFSX_UNSIGNED);
2486 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2487 dnp->n_cookieverf.nfsuquad[1] = *tl;
2488 } else {
2489 m_freem(mrep);
2490 goto nfsmout;
2491 }
2492 }
2493 #endif
2494 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2495 more_dirs = fxdr_unsigned(int, *tl);
2496
2497 /* loop thru the dir entries, doctoring them to 4bsd form */
2498 while (more_dirs && bigenough) {
2499 #ifndef NFS_V2_ONLY
2500 if (v3) {
2501 nfsm_dissect(tl, u_int32_t *,
2502 3 * NFSX_UNSIGNED);
2503 fileno = fxdr_hyper(tl);
2504 len = fxdr_unsigned(int, *(tl + 2));
2505 } else
2506 #endif
2507 {
2508 nfsm_dissect(tl, u_int32_t *,
2509 2 * NFSX_UNSIGNED);
2510 fileno = fxdr_unsigned(u_quad_t, *tl++);
2511 len = fxdr_unsigned(int, *tl);
2512 }
2513 if (len <= 0 || len > NFS_MAXNAMLEN) {
2514 error = EBADRPC;
2515 m_freem(mrep);
2516 goto nfsmout;
2517 }
2518 /* for cookie stashing */
2519 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2520 left = NFS_DIRFRAGSIZ - blksiz;
2521 if (reclen > left) {
2522 memset(uiop->uio_iov->iov_base, 0, left);
2523 dp->d_reclen += left;
2524 UIO_ADVANCE(uiop, left);
2525 blksiz = 0;
2526 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2527 }
2528 if (reclen > uiop->uio_resid)
2529 bigenough = 0;
2530 if (bigenough) {
2531 int tlen;
2532
2533 dp = (struct dirent *)uiop->uio_iov->iov_base;
2534 dp->d_fileno = fileno;
2535 dp->d_namlen = len;
2536 dp->d_reclen = reclen;
2537 dp->d_type = DT_UNKNOWN;
2538 blksiz += reclen;
2539 if (blksiz == NFS_DIRFRAGSIZ)
2540 blksiz = 0;
2541 UIO_ADVANCE(uiop, DIRHDSIZ);
2542 nfsm_mtouio(uiop, len);
2543 tlen = reclen - (DIRHDSIZ + len);
2544 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2545 UIO_ADVANCE(uiop, tlen);
2546 } else
2547 nfsm_adv(nfsm_rndup(len));
2548 #ifndef NFS_V2_ONLY
2549 if (v3) {
2550 nfsm_dissect(tl, u_int32_t *,
2551 3 * NFSX_UNSIGNED);
2552 } else
2553 #endif
2554 {
2555 nfsm_dissect(tl, u_int32_t *,
2556 2 * NFSX_UNSIGNED);
2557 }
2558 if (bigenough) {
2559 #ifndef NFS_V2_ONLY
2560 if (v3) {
2561 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2562 uiop->uio_offset =
2563 fxdr_swapcookie3(tl);
2564 else
2565 uiop->uio_offset =
2566 fxdr_cookie3(tl);
2567 }
2568 else
2569 #endif
2570 {
2571 uiop->uio_offset =
2572 fxdr_unsigned(off_t, *tl);
2573 }
2574 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2575 }
2576 if (v3)
2577 tl += 2;
2578 else
2579 tl++;
2580 more_dirs = fxdr_unsigned(int, *tl);
2581 }
2582 /*
2583 * If at end of rpc data, get the eof boolean
2584 */
2585 if (!more_dirs) {
2586 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2587 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2588
2589 /*
2590 * kludge: if we got no entries, treat it as EOF.
2591 * some server sometimes send a reply without any
2592 * entries or EOF.
2593 * although it might mean the server has very long name,
2594 * we can't handle such entries anyway.
2595 */
2596
2597 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2598 more_dirs = 0;
2599 }
2600 m_freem(mrep);
2601 }
2602 /*
2603 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2604 * by increasing d_reclen for the last record.
2605 */
2606 if (blksiz > 0) {
2607 left = NFS_DIRFRAGSIZ - blksiz;
2608 memset(uiop->uio_iov->iov_base, 0, left);
2609 dp->d_reclen += left;
2610 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2611 UIO_ADVANCE(uiop, left);
2612 }
2613
2614 /*
2615 * We are now either at the end of the directory or have filled the
2616 * block.
2617 */
2618 if (bigenough) {
2619 dnp->n_direofoffset = uiop->uio_offset;
2620 dnp->n_flag |= NEOFVALID;
2621 }
2622 nfsmout:
2623 return (error);
2624 }
2625
2626 #ifndef NFS_V2_ONLY
2627 /*
2628 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2629 */
2630 int
2631 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
2632 {
2633 int len, left;
2634 struct dirent *dp = NULL;
2635 u_int32_t *tl;
2636 char *cp;
2637 int32_t t1, t2;
2638 struct vnode *newvp;
2639 char *bpos, *dpos, *cp2;
2640 struct mbuf *mreq, *mrep, *md, *mb;
2641 struct nameidata nami, *ndp = &nami;
2642 struct componentname *cnp = &ndp->ni_cnd;
2643 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2644 struct nfsnode *dnp = VTONFS(vp), *np;
2645 nfsfh_t *fhp;
2646 u_quad_t fileno;
2647 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2648 int attrflag, fhsize, nrpcs = 0, reclen;
2649 struct nfs_fattr fattr, *fp;
2650
2651 #ifdef DIAGNOSTIC
2652 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2653 panic("nfs readdirplusrpc bad uio");
2654 #endif
2655 ndp->ni_dvp = vp;
2656 newvp = NULLVP;
2657
2658 /*
2659 * Loop around doing readdir rpc's of size nm_readdirsize
2660 * truncated to a multiple of NFS_DIRFRAGSIZ.
2661 * The stopping criteria is EOF or buffer full.
2662 */
2663 while (more_dirs && bigenough) {
2664 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2665 bigenough = 0;
2666 break;
2667 }
2668 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2669 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2670 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2671 nfsm_fhtom(dnp, 1);
2672 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2673 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2674 txdr_swapcookie3(uiop->uio_offset, tl);
2675 } else {
2676 txdr_cookie3(uiop->uio_offset, tl);
2677 }
2678 tl += 2;
2679 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2680 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2681 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2682 *tl = txdr_unsigned(nmp->nm_rsize);
2683 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2684 nfsm_postop_attr(vp, attrflag, 0);
2685 if (error) {
2686 m_freem(mrep);
2687 goto nfsmout;
2688 }
2689 nrpcs++;
2690 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2691 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2692 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2693 more_dirs = fxdr_unsigned(int, *tl);
2694
2695 /* loop thru the dir entries, doctoring them to 4bsd form */
2696 while (more_dirs && bigenough) {
2697 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2698 fileno = fxdr_hyper(tl);
2699 len = fxdr_unsigned(int, *(tl + 2));
2700 if (len <= 0 || len > NFS_MAXNAMLEN) {
2701 error = EBADRPC;
2702 m_freem(mrep);
2703 goto nfsmout;
2704 }
2705 /* for cookie stashing */
2706 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2707 left = NFS_DIRFRAGSIZ - blksiz;
2708 if (reclen > left) {
2709 /*
2710 * DIRFRAGSIZ is aligned, no need to align
2711 * again here.
2712 */
2713 memset(uiop->uio_iov->iov_base, 0, left);
2714 dp->d_reclen += left;
2715 UIO_ADVANCE(uiop, left);
2716 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2717 blksiz = 0;
2718 }
2719 if (reclen > uiop->uio_resid)
2720 bigenough = 0;
2721 if (bigenough) {
2722 int tlen;
2723
2724 dp = (struct dirent *)uiop->uio_iov->iov_base;
2725 dp->d_fileno = fileno;
2726 dp->d_namlen = len;
2727 dp->d_reclen = reclen;
2728 dp->d_type = DT_UNKNOWN;
2729 blksiz += reclen;
2730 if (blksiz == NFS_DIRFRAGSIZ)
2731 blksiz = 0;
2732 UIO_ADVANCE(uiop, DIRHDSIZ);
2733 nfsm_mtouio(uiop, len);
2734 tlen = reclen - (DIRHDSIZ + len);
2735 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2736 UIO_ADVANCE(uiop, tlen);
2737 cnp->cn_nameptr = dp->d_name;
2738 cnp->cn_namelen = dp->d_namlen;
2739 } else
2740 nfsm_adv(nfsm_rndup(len));
2741 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2742 if (bigenough) {
2743 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2744 uiop->uio_offset =
2745 fxdr_swapcookie3(tl);
2746 else
2747 uiop->uio_offset =
2748 fxdr_cookie3(tl);
2749 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2750 }
2751 tl += 2;
2752
2753 /*
2754 * Since the attributes are before the file handle
2755 * (sigh), we must skip over the attributes and then
2756 * come back and get them.
2757 */
2758 attrflag = fxdr_unsigned(int, *tl);
2759 if (attrflag) {
2760 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2761 memcpy(&fattr, fp, NFSX_V3FATTR);
2762 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2763 doit = fxdr_unsigned(int, *tl);
2764 if (doit) {
2765 nfsm_getfh(fhp, fhsize, 1);
2766 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2767 vref(vp);
2768 newvp = vp;
2769 np = dnp;
2770 } else {
2771 error = nfs_nget1(vp->v_mount, fhp,
2772 fhsize, &np, LK_NOWAIT);
2773 if (!error)
2774 newvp = NFSTOV(np);
2775 }
2776 if (!error) {
2777 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2778 if (bigenough) {
2779 dp->d_type =
2780 IFTODT(VTTOIF(np->n_vattr->va_type));
2781 ndp->ni_vp = newvp;
2782 nfs_cache_enter(ndp->ni_dvp,
2783 ndp->ni_vp, cnp);
2784 }
2785 }
2786 error = 0;
2787 }
2788 } else {
2789 /* Just skip over the file handle */
2790 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2791 i = fxdr_unsigned(int, *tl);
2792 nfsm_adv(nfsm_rndup(i));
2793 }
2794 if (newvp != NULLVP) {
2795 if (newvp == vp)
2796 vrele(newvp);
2797 else
2798 vput(newvp);
2799 newvp = NULLVP;
2800 }
2801 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2802 more_dirs = fxdr_unsigned(int, *tl);
2803 }
2804 /*
2805 * If at end of rpc data, get the eof boolean
2806 */
2807 if (!more_dirs) {
2808 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2809 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2810
2811 /*
2812 * kludge: see a comment in nfs_readdirrpc.
2813 */
2814
2815 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2816 more_dirs = 0;
2817 }
2818 m_freem(mrep);
2819 }
2820 /*
2821 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2822 * by increasing d_reclen for the last record.
2823 */
2824 if (blksiz > 0) {
2825 left = NFS_DIRFRAGSIZ - blksiz;
2826 memset(uiop->uio_iov->iov_base, 0, left);
2827 dp->d_reclen += left;
2828 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2829 UIO_ADVANCE(uiop, left);
2830 }
2831
2832 /*
2833 * We are now either at the end of the directory or have filled the
2834 * block.
2835 */
2836 if (bigenough) {
2837 dnp->n_direofoffset = uiop->uio_offset;
2838 dnp->n_flag |= NEOFVALID;
2839 }
2840 nfsmout:
2841 if (newvp != NULLVP) {
2842 if(newvp == vp)
2843 vrele(newvp);
2844 else
2845 vput(newvp);
2846 }
2847 return (error);
2848 }
2849 #endif
2850
2851 /*
2852 * Silly rename. To make the NFS filesystem that is stateless look a little
2853 * more like the "ufs" a remove of an active vnode is translated to a rename
2854 * to a funny looking filename that is removed by nfs_inactive on the
2855 * nfsnode. There is the potential for another process on a different client
2856 * to create the same funny name between the nfs_lookitup() fails and the
2857 * nfs_rename() completes, but...
2858 */
2859 int
2860 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink)
2861 {
2862 struct sillyrename *sp;
2863 struct nfsnode *np;
2864 int error;
2865 pid_t pid;
2866
2867 cache_purge(dvp);
2868 np = VTONFS(vp);
2869 #ifndef DIAGNOSTIC
2870 if (vp->v_type == VDIR)
2871 panic("nfs: sillyrename dir");
2872 #endif
2873 sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
2874 sp->s_cred = kauth_cred_dup(cnp->cn_cred);
2875 sp->s_dvp = dvp;
2876 vref(dvp);
2877
2878 /* Fudge together a funny name */
2879 pid = curlwp->l_proc->p_pid;
2880 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2881 sp->s_namlen = 12;
2882 sp->s_name[8] = hexdigits[pid & 0xf];
2883 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
2884 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
2885 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
2886
2887 /* Try lookitups until we get one that isn't there */
2888 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2889 curlwp, (struct nfsnode **)0) == 0) {
2890 sp->s_name[4]++;
2891 if (sp->s_name[4] > 'z') {
2892 error = EINVAL;
2893 goto bad;
2894 }
2895 }
2896 if (dolink) {
2897 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
2898 sp->s_cred, curlwp);
2899 /*
2900 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
2901 */
2902 if (error == ENOTSUP) {
2903 error = nfs_renameit(dvp, cnp, sp);
2904 }
2905 } else {
2906 error = nfs_renameit(dvp, cnp, sp);
2907 }
2908 if (error)
2909 goto bad;
2910 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2911 curlwp, &np);
2912 np->n_sillyrename = sp;
2913 return (0);
2914 bad:
2915 vrele(sp->s_dvp);
2916 kauth_cred_free(sp->s_cred);
2917 kmem_free(sp, sizeof(*sp));
2918 return (error);
2919 }
2920
2921 /*
2922 * Look up a file name and optionally either update the file handle or
2923 * allocate an nfsnode, depending on the value of npp.
2924 * npp == NULL --> just do the lookup
2925 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2926 * handled too
2927 * *npp != NULL --> update the file handle in the vnode
2928 */
2929 int
2930 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp)
2931 {
2932 u_int32_t *tl;
2933 char *cp;
2934 int32_t t1, t2;
2935 struct vnode *newvp = (struct vnode *)0;
2936 struct nfsnode *np, *dnp = VTONFS(dvp);
2937 char *bpos, *dpos, *cp2;
2938 int error = 0, ofhlen, fhlen;
2939 #ifndef NFS_V2_ONLY
2940 int attrflag;
2941 #endif
2942 struct mbuf *mreq, *mrep, *md, *mb;
2943 nfsfh_t *ofhp, *nfhp;
2944 const int v3 = NFS_ISV3(dvp);
2945
2946 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2947 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2948 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2949 nfsm_fhtom(dnp, v3);
2950 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2951 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
2952 if (npp && !error) {
2953 nfsm_getfh(nfhp, fhlen, v3);
2954 if (*npp) {
2955 np = *npp;
2956 newvp = NFSTOV(np);
2957 ofhlen = np->n_fhsize;
2958 ofhp = kmem_alloc(ofhlen, KM_SLEEP);
2959 memcpy(ofhp, np->n_fhp, ofhlen);
2960 error = vcache_rekey_enter(newvp->v_mount, newvp,
2961 ofhp, ofhlen, nfhp, fhlen);
2962 if (error) {
2963 kmem_free(ofhp, ofhlen);
2964 m_freem(mrep);
2965 return error;
2966 }
2967 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2968 kmem_free(np->n_fhp, np->n_fhsize);
2969 np->n_fhp = &np->n_fh;
2970 }
2971 #if NFS_SMALLFH < NFSX_V3FHMAX
2972 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
2973 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
2974 #endif
2975 memcpy(np->n_fhp, nfhp, fhlen);
2976 np->n_fhsize = fhlen;
2977 vcache_rekey_exit(newvp->v_mount, newvp,
2978 ofhp, ofhlen, np->n_fhp, fhlen);
2979 kmem_free(ofhp, ofhlen);
2980 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2981 vref(dvp);
2982 newvp = dvp;
2983 np = dnp;
2984 } else {
2985 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2986 if (error) {
2987 m_freem(mrep);
2988 return (error);
2989 }
2990 newvp = NFSTOV(np);
2991 }
2992 #ifndef NFS_V2_ONLY
2993 if (v3) {
2994 nfsm_postop_attr(newvp, attrflag, 0);
2995 if (!attrflag && *npp == NULL) {
2996 m_freem(mrep);
2997 vput(newvp);
2998 return (ENOENT);
2999 }
3000 } else
3001 #endif
3002 nfsm_loadattr(newvp, (struct vattr *)0, 0);
3003 }
3004 nfsm_reqdone;
3005 if (npp && *npp == NULL) {
3006 if (error) {
3007 if (newvp)
3008 vput(newvp);
3009 } else
3010 *npp = np;
3011 }
3012 return (error);
3013 }
3014
3015 #ifndef NFS_V2_ONLY
3016 /*
3017 * Nfs Version 3 commit rpc
3018 */
3019 int
3020 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l)
3021 {
3022 char *cp;
3023 u_int32_t *tl;
3024 int32_t t1, t2;
3025 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3026 char *bpos, *dpos, *cp2;
3027 int error = 0, wccflag = NFSV3_WCCRATTR;
3028 struct mbuf *mreq, *mrep, *md, *mb;
3029 struct nfsnode *np;
3030
3031 KASSERT(NFS_ISV3(vp));
3032
3033 #ifdef NFS_DEBUG_COMMIT
3034 printf("commit %lu - %lu\n", (unsigned long)offset,
3035 (unsigned long)(offset + cnt));
3036 #endif
3037
3038 mutex_enter(&nmp->nm_lock);
3039 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3040 mutex_exit(&nmp->nm_lock);
3041 return (0);
3042 }
3043 mutex_exit(&nmp->nm_lock);
3044 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3045 np = VTONFS(vp);
3046 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3047 nfsm_fhtom(np, 1);
3048 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3049 txdr_hyper(offset, tl);
3050 tl += 2;
3051 *tl = txdr_unsigned(cnt);
3052 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3053 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3054 if (!error) {
3055 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3056 mutex_enter(&nmp->nm_lock);
3057 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3058 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3059 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3060 error = NFSERR_STALEWRITEVERF;
3061 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3062 }
3063 mutex_exit(&nmp->nm_lock);
3064 }
3065 nfsm_reqdone;
3066 return (error);
3067 }
3068 #endif
3069
3070 /*
3071 * Kludge City..
3072 * - make nfs_bmap() essentially a no-op that does no translation
3073 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3074 * (Maybe I could use the process's page mapping, but I was concerned that
3075 * Kernel Write might not be enabled and also figured copyout() would do
3076 * a lot more work than memcpy() and also it currently happens in the
3077 * context of the swapper process (2).
3078 */
3079 int
3080 nfs_bmap(void *v)
3081 {
3082 struct vop_bmap_args /* {
3083 struct vnode *a_vp;
3084 daddr_t a_bn;
3085 struct vnode **a_vpp;
3086 daddr_t *a_bnp;
3087 int *a_runp;
3088 } */ *ap = v;
3089 struct vnode *vp = ap->a_vp;
3090 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3091
3092 if (ap->a_vpp != NULL)
3093 *ap->a_vpp = vp;
3094 if (ap->a_bnp != NULL)
3095 *ap->a_bnp = ap->a_bn << bshift;
3096 if (ap->a_runp != NULL)
3097 *ap->a_runp = 1024 * 1024; /* XXX */
3098 return (0);
3099 }
3100
3101 /*
3102 * Strategy routine.
3103 * For async requests when nfsiod(s) are running, queue the request by
3104 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3105 * request.
3106 */
3107 int
3108 nfs_strategy(void *v)
3109 {
3110 struct vop_strategy_args *ap = v;
3111 struct buf *bp = ap->a_bp;
3112 int error = 0;
3113
3114 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3115 panic("nfs physio/async");
3116
3117 /*
3118 * If the op is asynchronous and an i/o daemon is waiting
3119 * queue the request, wake it up and wait for completion
3120 * otherwise just do it ourselves.
3121 */
3122 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3123 error = nfs_doio(bp);
3124 return (error);
3125 }
3126
3127 /*
3128 * fsync vnode op. Just call nfs_flush() with commit == 1.
3129 */
3130 /* ARGSUSED */
3131 int
3132 nfs_fsync(void *v)
3133 {
3134 struct vop_fsync_args /* {
3135 struct vnodeop_desc *a_desc;
3136 struct vnode * a_vp;
3137 kauth_cred_t a_cred;
3138 int a_flags;
3139 off_t offlo;
3140 off_t offhi;
3141 struct lwp * a_l;
3142 } */ *ap = v;
3143
3144 struct vnode *vp = ap->a_vp;
3145
3146 if (vp->v_type != VREG)
3147 return 0;
3148
3149 return (nfs_flush(vp, ap->a_cred,
3150 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3151 }
3152
3153 /*
3154 * Flush all the data associated with a vnode.
3155 */
3156 int
3157 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3158 int commit)
3159 {
3160 struct nfsnode *np = VTONFS(vp);
3161 int error;
3162 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3163 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3164
3165 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
3166 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3167 if (np->n_flag & NWRITEERR) {
3168 error = np->n_error;
3169 np->n_flag &= ~NWRITEERR;
3170 }
3171 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3172 return (error);
3173 }
3174
3175 /*
3176 * Return POSIX pathconf information applicable to nfs.
3177 *
3178 * N.B. The NFS V2 protocol doesn't support this RPC.
3179 */
3180 /* ARGSUSED */
3181 int
3182 nfs_pathconf(void *v)
3183 {
3184 struct vop_pathconf_args /* {
3185 struct vnode *a_vp;
3186 int a_name;
3187 register_t *a_retval;
3188 } */ *ap = v;
3189 struct nfsv3_pathconf *pcp;
3190 struct vnode *vp = ap->a_vp;
3191 struct mbuf *mreq, *mrep, *md, *mb;
3192 int32_t t1, t2;
3193 u_int32_t *tl;
3194 char *bpos, *dpos, *cp, *cp2;
3195 int error = 0, attrflag;
3196 #ifndef NFS_V2_ONLY
3197 struct nfsmount *nmp;
3198 unsigned int l;
3199 u_int64_t maxsize;
3200 #endif
3201 const int v3 = NFS_ISV3(vp);
3202 struct nfsnode *np = VTONFS(vp);
3203
3204 switch (ap->a_name) {
3205 /* Names that can be resolved locally. */
3206 case _PC_PIPE_BUF:
3207 *ap->a_retval = PIPE_BUF;
3208 break;
3209 case _PC_SYNC_IO:
3210 *ap->a_retval = 1;
3211 break;
3212 /* Names that cannot be resolved locally; do an RPC, if possible. */
3213 case _PC_LINK_MAX:
3214 case _PC_NAME_MAX:
3215 case _PC_CHOWN_RESTRICTED:
3216 case _PC_NO_TRUNC:
3217 if (!v3) {
3218 error = EINVAL;
3219 break;
3220 }
3221 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3222 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3223 nfsm_fhtom(np, 1);
3224 nfsm_request(np, NFSPROC_PATHCONF,
3225 curlwp, curlwp->l_cred); /* XXX */
3226 nfsm_postop_attr(vp, attrflag, 0);
3227 if (!error) {
3228 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3229 NFSX_V3PATHCONF);
3230 switch (ap->a_name) {
3231 case _PC_LINK_MAX:
3232 *ap->a_retval =
3233 fxdr_unsigned(register_t, pcp->pc_linkmax);
3234 break;
3235 case _PC_NAME_MAX:
3236 *ap->a_retval =
3237 fxdr_unsigned(register_t, pcp->pc_namemax);
3238 break;
3239 case _PC_CHOWN_RESTRICTED:
3240 *ap->a_retval =
3241 (pcp->pc_chownrestricted == nfs_true);
3242 break;
3243 case _PC_NO_TRUNC:
3244 *ap->a_retval =
3245 (pcp->pc_notrunc == nfs_true);
3246 break;
3247 }
3248 }
3249 nfsm_reqdone;
3250 break;
3251 case _PC_FILESIZEBITS:
3252 #ifndef NFS_V2_ONLY
3253 if (v3) {
3254 nmp = VFSTONFS(vp->v_mount);
3255 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3256 if ((error = nfs_fsinfo(nmp, vp,
3257 curlwp->l_cred, curlwp)) != 0) /* XXX */
3258 break;
3259 for (l = 0, maxsize = nmp->nm_maxfilesize;
3260 (maxsize >> l) > 0; l++)
3261 ;
3262 *ap->a_retval = l + 1;
3263 } else
3264 #endif
3265 {
3266 *ap->a_retval = 32; /* NFS V2 limitation */
3267 }
3268 break;
3269 default:
3270 error = EINVAL;
3271 break;
3272 }
3273
3274 return (error);
3275 }
3276
3277 /*
3278 * NFS advisory byte-level locks.
3279 */
3280 int
3281 nfs_advlock(void *v)
3282 {
3283 struct vop_advlock_args /* {
3284 struct vnode *a_vp;
3285 void *a_id;
3286 int a_op;
3287 struct flock *a_fl;
3288 int a_flags;
3289 } */ *ap = v;
3290 struct nfsnode *np = VTONFS(ap->a_vp);
3291
3292 return lf_advlock(ap, &np->n_lockf, np->n_size);
3293 }
3294
3295 /*
3296 * Print out the contents of an nfsnode.
3297 */
3298 int
3299 nfs_print(void *v)
3300 {
3301 struct vop_print_args /* {
3302 struct vnode *a_vp;
3303 } */ *ap = v;
3304 struct vnode *vp = ap->a_vp;
3305 struct nfsnode *np = VTONFS(vp);
3306
3307 printf("tag VT_NFS, fileid %lld fsid 0x%llx",
3308 (unsigned long long)np->n_vattr->va_fileid,
3309 (unsigned long long)np->n_vattr->va_fsid);
3310 if (vp->v_type == VFIFO)
3311 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
3312 printf("\n");
3313 return (0);
3314 }
3315
3316 /*
3317 * nfs unlock wrapper.
3318 */
3319 int
3320 nfs_unlock(void *v)
3321 {
3322 struct vop_unlock_args /* {
3323 struct vnode *a_vp;
3324 int a_flags;
3325 } */ *ap = v;
3326 struct vnode *vp = ap->a_vp;
3327
3328 /*
3329 * VOP_UNLOCK can be called by nfs_loadattrcache
3330 * with v_data == 0.
3331 */
3332 if (VTONFS(vp)) {
3333 nfs_delayedtruncate(vp);
3334 }
3335
3336 return genfs_unlock(v);
3337 }
3338
3339 /*
3340 * nfs special file access vnode op.
3341 * Essentially just get vattr and then imitate iaccess() since the device is
3342 * local to the client.
3343 */
3344 int
3345 nfsspec_access(void *v)
3346 {
3347 struct vop_access_args /* {
3348 struct vnode *a_vp;
3349 int a_mode;
3350 kauth_cred_t a_cred;
3351 struct lwp *a_l;
3352 } */ *ap = v;
3353 struct vattr va;
3354 struct vnode *vp = ap->a_vp;
3355 int error;
3356
3357 error = VOP_GETATTR(vp, &va, ap->a_cred);
3358 if (error)
3359 return (error);
3360
3361 /*
3362 * Disallow write attempts on filesystems mounted read-only;
3363 * unless the file is a socket, fifo, or a block or character
3364 * device resident on the filesystem.
3365 */
3366 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3367 switch (vp->v_type) {
3368 case VREG:
3369 case VDIR:
3370 case VLNK:
3371 return (EROFS);
3372 default:
3373 break;
3374 }
3375 }
3376
3377 return kauth_authorize_vnode(ap->a_cred, KAUTH_ACCESS_ACTION(ap->a_mode,
3378 va.va_type, va.va_mode), vp, NULL, genfs_can_access(va.va_type,
3379 va.va_mode, va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3380 }
3381
3382 /*
3383 * Read wrapper for special devices.
3384 */
3385 int
3386 nfsspec_read(void *v)
3387 {
3388 struct vop_read_args /* {
3389 struct vnode *a_vp;
3390 struct uio *a_uio;
3391 int a_ioflag;
3392 kauth_cred_t a_cred;
3393 } */ *ap = v;
3394 struct nfsnode *np = VTONFS(ap->a_vp);
3395
3396 /*
3397 * Set access flag.
3398 */
3399 np->n_flag |= NACC;
3400 getnanotime(&np->n_atim);
3401 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3402 }
3403
3404 /*
3405 * Write wrapper for special devices.
3406 */
3407 int
3408 nfsspec_write(void *v)
3409 {
3410 struct vop_write_args /* {
3411 struct vnode *a_vp;
3412 struct uio *a_uio;
3413 int a_ioflag;
3414 kauth_cred_t a_cred;
3415 } */ *ap = v;
3416 struct nfsnode *np = VTONFS(ap->a_vp);
3417
3418 /*
3419 * Set update flag.
3420 */
3421 np->n_flag |= NUPD;
3422 getnanotime(&np->n_mtim);
3423 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3424 }
3425
3426 /*
3427 * Close wrapper for special devices.
3428 *
3429 * Update the times on the nfsnode then do device close.
3430 */
3431 int
3432 nfsspec_close(void *v)
3433 {
3434 struct vop_close_args /* {
3435 struct vnode *a_vp;
3436 int a_fflag;
3437 kauth_cred_t a_cred;
3438 struct lwp *a_l;
3439 } */ *ap = v;
3440 struct vnode *vp = ap->a_vp;
3441 struct nfsnode *np = VTONFS(vp);
3442 struct vattr vattr;
3443
3444 if (np->n_flag & (NACC | NUPD)) {
3445 np->n_flag |= NCHG;
3446 if (vp->v_usecount == 1 &&
3447 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3448 vattr_null(&vattr);
3449 if (np->n_flag & NACC)
3450 vattr.va_atime = np->n_atim;
3451 if (np->n_flag & NUPD)
3452 vattr.va_mtime = np->n_mtim;
3453 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3454 }
3455 }
3456 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3457 }
3458
3459 /*
3460 * Read wrapper for fifos.
3461 */
3462 int
3463 nfsfifo_read(void *v)
3464 {
3465 struct vop_read_args /* {
3466 struct vnode *a_vp;
3467 struct uio *a_uio;
3468 int a_ioflag;
3469 kauth_cred_t a_cred;
3470 } */ *ap = v;
3471 struct nfsnode *np = VTONFS(ap->a_vp);
3472
3473 /*
3474 * Set access flag.
3475 */
3476 np->n_flag |= NACC;
3477 getnanotime(&np->n_atim);
3478 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3479 }
3480
3481 /*
3482 * Write wrapper for fifos.
3483 */
3484 int
3485 nfsfifo_write(void *v)
3486 {
3487 struct vop_write_args /* {
3488 struct vnode *a_vp;
3489 struct uio *a_uio;
3490 int a_ioflag;
3491 kauth_cred_t a_cred;
3492 } */ *ap = v;
3493 struct nfsnode *np = VTONFS(ap->a_vp);
3494
3495 /*
3496 * Set update flag.
3497 */
3498 np->n_flag |= NUPD;
3499 getnanotime(&np->n_mtim);
3500 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3501 }
3502
3503 /*
3504 * Close wrapper for fifos.
3505 *
3506 * Update the times on the nfsnode then do fifo close.
3507 */
3508 int
3509 nfsfifo_close(void *v)
3510 {
3511 struct vop_close_args /* {
3512 struct vnode *a_vp;
3513 int a_fflag;
3514 kauth_cred_t a_cred;
3515 struct lwp *a_l;
3516 } */ *ap = v;
3517 struct vnode *vp = ap->a_vp;
3518 struct nfsnode *np = VTONFS(vp);
3519 struct vattr vattr;
3520
3521 if (np->n_flag & (NACC | NUPD)) {
3522 struct timespec ts;
3523
3524 getnanotime(&ts);
3525 if (np->n_flag & NACC)
3526 np->n_atim = ts;
3527 if (np->n_flag & NUPD)
3528 np->n_mtim = ts;
3529 np->n_flag |= NCHG;
3530 if (vp->v_usecount == 1 &&
3531 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3532 vattr_null(&vattr);
3533 if (np->n_flag & NACC)
3534 vattr.va_atime = np->n_atim;
3535 if (np->n_flag & NUPD)
3536 vattr.va_mtime = np->n_mtim;
3537 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3538 }
3539 }
3540 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3541 }
3542