]> git.saurik.com Git - apple/xnu.git/blob - bsd/nfs/nfs_vnops.c
projects / apple / xnu.git / blob
? search:


caa5533f53bbf8e9db612e93f442eb7ed7d20cc3
[apple/xnu.git] / bsd / nfs / nfs_vnops.c
1 /*
2 * Copyright (c) 2000-2020 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
29 /*
30 * Copyright (c) 1989, 1993
31 * The Regents of the University of California. All rights reserved.
32 *
33 * This code is derived from software contributed to Berkeley by
34 * Rick Macklem at The University of Guelph.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 * 3. All advertising materials mentioning features or use of this software
45 * must display the following acknowledgement:
46 * This product includes software developed by the University of
47 * California, Berkeley and its contributors.
48 * 4. Neither the name of the University nor the names of its contributors
49 * may be used to endorse or promote products derived from this software
50 * without specific prior written permission.
51 *
52 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * SUCH DAMAGE.
63 *
64 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
65 * FreeBSD-Id: nfs_vnops.c,v 1.72 1997/11/07 09:20:48 phk Exp $
66 */
67
68 #include <nfs/nfs_conf.h>
69 #if CONFIG_NFS_CLIENT
70
71 /*
72 * vnode op calls for Sun NFS version 2 and 3
73 */
74 #include <sys/param.h>
75 #include <sys/kernel.h>
76 #include <sys/systm.h>
77 #include <sys/resourcevar.h>
78 #include <sys/proc_internal.h>
79 #include <sys/kauth.h>
80 #include <sys/mount_internal.h>
81 #include <sys/malloc.h>
82 #include <sys/kpi_mbuf.h>
83 #include <sys/conf.h>
84 #include <sys/vnode_internal.h>
85 #include <sys/dirent.h>
86 #include <sys/fcntl.h>
87 #include <sys/lockf.h>
88 #include <sys/ubc_internal.h>
89 #include <sys/attr.h>
90 #include <sys/signalvar.h>
91 #include <sys/uio_internal.h>
92 #include <sys/xattr.h>
93
94 #include <vfs/vfs_support.h>
95
96 #include <sys/vm.h>
97
98 #include <sys/time.h>
99 #include <kern/clock.h>
100 #include <libkern/OSAtomic.h>
101
102 #include <miscfs/fifofs/fifo.h>
103 #include <miscfs/specfs/specdev.h>
104
105 #include <nfs/rpcv2.h>
106 #include <nfs/nfsproto.h>
107 #include <nfs/nfs.h>
108 #include <nfs/nfsnode.h>
109 #include <nfs/nfs_gss.h>
110 #include <nfs/nfsmount.h>
111 #include <nfs/nfs_lock.h>
112 #include <nfs/xdr_subs.h>
113 #include <nfs/nfsm_subs.h>
114
115 #include <net/if.h>
116 #include <netinet/in.h>
117 #include <netinet/in_var.h>
118
119 #include <vm/vm_kern.h>
120 #include <vm/vm_pageout.h>
121
122 #include <kern/task.h>
123 #include <kern/sched_prim.h>
124
125 #define NFS_VNOP_DBG(...) NFS_DBG(NFS_FAC_VNOP, 7, ## __VA_ARGS__)
126 #define DEFAULT_READLINK_NOCACHE 0
127
128 /*
129 * NFS vnode ops
130 */
131 int nfs_vnop_lookup(struct vnop_lookup_args *);
132 int nfsspec_vnop_read(struct vnop_read_args *);
133 int nfsspec_vnop_write(struct vnop_write_args *);
134 int nfsspec_vnop_close(struct vnop_close_args *);
135 #if FIFO
136 int nfsfifo_vnop_read(struct vnop_read_args *);
137 int nfsfifo_vnop_write(struct vnop_write_args *);
138 int nfsfifo_vnop_close(struct vnop_close_args *);
139 #endif
140 int nfs_vnop_ioctl(struct vnop_ioctl_args *);
141 int nfs_vnop_select(struct vnop_select_args *);
142 int nfs_vnop_setattr(struct vnop_setattr_args *);
143 int nfs_vnop_fsync(struct vnop_fsync_args *);
144 int nfs_vnop_rename(struct vnop_rename_args *);
145 int nfs_vnop_readdir(struct vnop_readdir_args *);
146 int nfs_vnop_readlink(struct vnop_readlink_args *);
147 int nfs_vnop_pathconf(struct vnop_pathconf_args *);
148 int nfs_vnop_pagein(struct vnop_pagein_args *);
149 int nfs_vnop_pageout(struct vnop_pageout_args *);
150 int nfs_vnop_blktooff(struct vnop_blktooff_args *);
151 int nfs_vnop_offtoblk(struct vnop_offtoblk_args *);
152 int nfs_vnop_blockmap(struct vnop_blockmap_args *);
153 int nfs_vnop_monitor(struct vnop_monitor_args *);
154
155 int nfs3_vnop_create(struct vnop_create_args *);
156 int nfs3_vnop_mknod(struct vnop_mknod_args *);
157 int nfs3_vnop_getattr(struct vnop_getattr_args *);
158 int nfs3_vnop_link(struct vnop_link_args *);
159 int nfs3_vnop_mkdir(struct vnop_mkdir_args *);
160 int nfs3_vnop_rmdir(struct vnop_rmdir_args *);
161 int nfs3_vnop_symlink(struct vnop_symlink_args *);
162
163
164 vnop_t **nfsv2_vnodeop_p;
165 static const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
166 { .opve_op = &vnop_default_desc, .opve_impl = (vnop_t *)vn_default_error },
167 { .opve_op = &vnop_lookup_desc, .opve_impl = (vnop_t *)nfs_vnop_lookup }, /* lookup */
168 { .opve_op = &vnop_create_desc, .opve_impl = (vnop_t *)nfs3_vnop_create }, /* create */
169 { .opve_op = &vnop_mknod_desc, .opve_impl = (vnop_t *)nfs3_vnop_mknod }, /* mknod */
170 { .opve_op = &vnop_open_desc, .opve_impl = (vnop_t *)nfs_vnop_open }, /* open */
171 { .opve_op = &vnop_close_desc, .opve_impl = (vnop_t *)nfs_vnop_close }, /* close */
172 { .opve_op = &vnop_access_desc, .opve_impl = (vnop_t *)nfs_vnop_access }, /* access */
173 { .opve_op = &vnop_getattr_desc, .opve_impl = (vnop_t *)nfs3_vnop_getattr }, /* getattr */
174 { .opve_op = &vnop_setattr_desc, .opve_impl = (vnop_t *)nfs_vnop_setattr }, /* setattr */
175 { .opve_op = &vnop_read_desc, .opve_impl = (vnop_t *)nfs_vnop_read }, /* read */
176 { .opve_op = &vnop_write_desc, .opve_impl = (vnop_t *)nfs_vnop_write }, /* write */
177 { .opve_op = &vnop_ioctl_desc, .opve_impl = (vnop_t *)nfs_vnop_ioctl }, /* ioctl */
178 { .opve_op = &vnop_select_desc, .opve_impl = (vnop_t *)nfs_vnop_select }, /* select */
179 { .opve_op = &vnop_revoke_desc, .opve_impl = (vnop_t *)nfs_vnop_revoke }, /* revoke */
180 { .opve_op = &vnop_mmap_desc, .opve_impl = (vnop_t *)nfs_vnop_mmap }, /* mmap */
181 { .opve_op = &vnop_mmap_check_desc, .opve_impl = (vnop_t *)nfs_vnop_mmap_check }, /* mmap_check */
182 { .opve_op = &vnop_mnomap_desc, .opve_impl = (vnop_t *)nfs_vnop_mnomap }, /* mnomap */
183 { .opve_op = &vnop_fsync_desc, .opve_impl = (vnop_t *)nfs_vnop_fsync }, /* fsync */
184 { .opve_op = &vnop_remove_desc, .opve_impl = (vnop_t *)nfs_vnop_remove }, /* remove */
185 { .opve_op = &vnop_link_desc, .opve_impl = (vnop_t *)nfs3_vnop_link }, /* link */
186 { .opve_op = &vnop_rename_desc, .opve_impl = (vnop_t *)nfs_vnop_rename }, /* rename */
187 { .opve_op = &vnop_mkdir_desc, .opve_impl = (vnop_t *)nfs3_vnop_mkdir }, /* mkdir */
188 { .opve_op = &vnop_rmdir_desc, .opve_impl = (vnop_t *)nfs3_vnop_rmdir }, /* rmdir */
189 { .opve_op = &vnop_symlink_desc, .opve_impl = (vnop_t *)nfs3_vnop_symlink }, /* symlink */
190 { .opve_op = &vnop_readdir_desc, .opve_impl = (vnop_t *)nfs_vnop_readdir }, /* readdir */
191 { .opve_op = &vnop_readlink_desc, .opve_impl = (vnop_t *)nfs_vnop_readlink }, /* readlink */
192 { .opve_op = &vnop_inactive_desc, .opve_impl = (vnop_t *)nfs_vnop_inactive }, /* inactive */
193 { .opve_op = &vnop_reclaim_desc, .opve_impl = (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
194 { .opve_op = &vnop_strategy_desc, .opve_impl = (vnop_t *)err_strategy }, /* strategy */
195 { .opve_op = &vnop_pathconf_desc, .opve_impl = (vnop_t *)nfs_vnop_pathconf }, /* pathconf */
196 { .opve_op = &vnop_advlock_desc, .opve_impl = (vnop_t *)nfs_vnop_advlock }, /* advlock */
197 { .opve_op = &vnop_bwrite_desc, .opve_impl = (vnop_t *)err_bwrite }, /* bwrite */
198 { .opve_op = &vnop_pagein_desc, .opve_impl = (vnop_t *)nfs_vnop_pagein }, /* Pagein */
199 { .opve_op = &vnop_pageout_desc, .opve_impl = (vnop_t *)nfs_vnop_pageout }, /* Pageout */
200 { .opve_op = &vnop_copyfile_desc, .opve_impl = (vnop_t *)err_copyfile }, /* Copyfile */
201 { .opve_op = &vnop_blktooff_desc, .opve_impl = (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
202 { .opve_op = &vnop_offtoblk_desc, .opve_impl = (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
203 { .opve_op = &vnop_blockmap_desc, .opve_impl = (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
204 { .opve_op = &vnop_monitor_desc, .opve_impl = (vnop_t *)nfs_vnop_monitor }, /* monitor */
205 { .opve_op = NULL, .opve_impl = NULL }
206 };
207 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
208 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
209
210
211 #if CONFIG_NFS4
212 vnop_t **nfsv4_vnodeop_p;
213 static const struct vnodeopv_entry_desc nfsv4_vnodeop_entries[] = {
214 { &vnop_default_desc, (vnop_t *)vn_default_error },
215 { &vnop_lookup_desc, (vnop_t *)nfs_vnop_lookup }, /* lookup */
216 { &vnop_create_desc, (vnop_t *)nfs4_vnop_create }, /* create */
217 { &vnop_mknod_desc, (vnop_t *)nfs4_vnop_mknod }, /* mknod */
218 { &vnop_open_desc, (vnop_t *)nfs_vnop_open }, /* open */
219 { &vnop_close_desc, (vnop_t *)nfs_vnop_close }, /* close */
220 { &vnop_access_desc, (vnop_t *)nfs_vnop_access }, /* access */
221 { &vnop_getattr_desc, (vnop_t *)nfs4_vnop_getattr }, /* getattr */
222 { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
223 { &vnop_read_desc, (vnop_t *)nfs_vnop_read }, /* read */
224 { &vnop_write_desc, (vnop_t *)nfs_vnop_write }, /* write */
225 { &vnop_ioctl_desc, (vnop_t *)nfs_vnop_ioctl }, /* ioctl */
226 { &vnop_select_desc, (vnop_t *)nfs_vnop_select }, /* select */
227 { &vnop_revoke_desc, (vnop_t *)nfs_vnop_revoke }, /* revoke */
228 { &vnop_mmap_desc, (vnop_t *)nfs_vnop_mmap }, /* mmap */
229 { &vnop_mmap_check_desc, (vnop_t *)nfs_vnop_mmap_check }, /* mmap_check */
230 { &vnop_mnomap_desc, (vnop_t *)nfs_vnop_mnomap }, /* mnomap */
231 { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
232 { &vnop_remove_desc, (vnop_t *)nfs_vnop_remove }, /* remove */
233 { &vnop_link_desc, (vnop_t *)nfs4_vnop_link }, /* link */
234 { &vnop_rename_desc, (vnop_t *)nfs_vnop_rename }, /* rename */
235 { &vnop_mkdir_desc, (vnop_t *)nfs4_vnop_mkdir }, /* mkdir */
236 { &vnop_rmdir_desc, (vnop_t *)nfs4_vnop_rmdir }, /* rmdir */
237 { &vnop_symlink_desc, (vnop_t *)nfs4_vnop_symlink }, /* symlink */
238 { &vnop_readdir_desc, (vnop_t *)nfs_vnop_readdir }, /* readdir */
239 { &vnop_readlink_desc, (vnop_t *)nfs_vnop_readlink }, /* readlink */
240 { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
241 { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
242 { &vnop_strategy_desc, (vnop_t *)err_strategy }, /* strategy */
243 { &vnop_pathconf_desc, (vnop_t *)nfs_vnop_pathconf }, /* pathconf */
244 { &vnop_advlock_desc, (vnop_t *)nfs_vnop_advlock }, /* advlock */
245 { &vnop_bwrite_desc, (vnop_t *)err_bwrite }, /* bwrite */
246 { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
247 { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
248 { &vnop_copyfile_desc, (vnop_t *)err_copyfile }, /* Copyfile */
249 { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
250 { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
251 { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
252 { &vnop_getxattr_desc, (vnop_t *)nfs4_vnop_getxattr }, /* getxattr */
253 { &vnop_setxattr_desc, (vnop_t *)nfs4_vnop_setxattr }, /* setxattr */
254 { &vnop_removexattr_desc, (vnop_t *)nfs4_vnop_removexattr }, /* removexattr */
255 { &vnop_listxattr_desc, (vnop_t *)nfs4_vnop_listxattr }, /* listxattr */
256 #if NAMEDSTREAMS
257 { &vnop_getnamedstream_desc, (vnop_t *)nfs4_vnop_getnamedstream }, /* getnamedstream */
258 { &vnop_makenamedstream_desc, (vnop_t *)nfs4_vnop_makenamedstream }, /* makenamedstream */
259 { &vnop_removenamedstream_desc, (vnop_t *)nfs4_vnop_removenamedstream }, /* removenamedstream */
260 #endif
261 { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
262 { NULL, NULL }
263 };
264 const struct vnodeopv_desc nfsv4_vnodeop_opv_desc =
265 { &nfsv4_vnodeop_p, nfsv4_vnodeop_entries };
266 #endif
267
268 /*
269 * Special device vnode ops
270 */
271 vnop_t **spec_nfsv2nodeop_p;
272 static const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
273 { &vnop_default_desc, (vnop_t *)vn_default_error },
274 { &vnop_lookup_desc, (vnop_t *)spec_lookup }, /* lookup */
275 { &vnop_create_desc, (vnop_t *)spec_create }, /* create */
276 { &vnop_mknod_desc, (vnop_t *)spec_mknod }, /* mknod */
277 { &vnop_open_desc, (vnop_t *)spec_open }, /* open */
278 { &vnop_close_desc, (vnop_t *)nfsspec_vnop_close }, /* close */
279 { &vnop_getattr_desc, (vnop_t *)nfs3_vnop_getattr }, /* getattr */
280 { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
281 { &vnop_read_desc, (vnop_t *)nfsspec_vnop_read }, /* read */
282 { &vnop_write_desc, (vnop_t *)nfsspec_vnop_write }, /* write */
283 { &vnop_ioctl_desc, (vnop_t *)spec_ioctl }, /* ioctl */
284 { &vnop_select_desc, (vnop_t *)spec_select }, /* select */
285 { &vnop_revoke_desc, (vnop_t *)spec_revoke }, /* revoke */
286 { &vnop_mmap_desc, (vnop_t *)spec_mmap }, /* mmap */
287 { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
288 { &vnop_remove_desc, (vnop_t *)spec_remove }, /* remove */
289 { &vnop_link_desc, (vnop_t *)spec_link }, /* link */
290 { &vnop_rename_desc, (vnop_t *)spec_rename }, /* rename */
291 { &vnop_mkdir_desc, (vnop_t *)spec_mkdir }, /* mkdir */
292 { &vnop_rmdir_desc, (vnop_t *)spec_rmdir }, /* rmdir */
293 { &vnop_symlink_desc, (vnop_t *)spec_symlink }, /* symlink */
294 { &vnop_readdir_desc, (vnop_t *)spec_readdir }, /* readdir */
295 { &vnop_readlink_desc, (vnop_t *)spec_readlink }, /* readlink */
296 { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
297 { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
298 { &vnop_strategy_desc, (vnop_t *)spec_strategy }, /* strategy */
299 { &vnop_pathconf_desc, (vnop_t *)spec_pathconf }, /* pathconf */
300 { &vnop_advlock_desc, (vnop_t *)spec_advlock }, /* advlock */
301 { &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
302 { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
303 { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
304 { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
305 { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
306 { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
307 { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
308 { NULL, NULL }
309 };
310 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
311 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
312 #if CONFIG_NFS4
313 vnop_t **spec_nfsv4nodeop_p;
314 static const struct vnodeopv_entry_desc spec_nfsv4nodeop_entries[] = {
315 { &vnop_default_desc, (vnop_t *)vn_default_error },
316 { &vnop_lookup_desc, (vnop_t *)spec_lookup }, /* lookup */
317 { &vnop_create_desc, (vnop_t *)spec_create }, /* create */
318 { &vnop_mknod_desc, (vnop_t *)spec_mknod }, /* mknod */
319 { &vnop_open_desc, (vnop_t *)spec_open }, /* open */
320 { &vnop_close_desc, (vnop_t *)nfsspec_vnop_close }, /* close */
321 { &vnop_getattr_desc, (vnop_t *)nfs4_vnop_getattr }, /* getattr */
322 { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
323 { &vnop_read_desc, (vnop_t *)nfsspec_vnop_read }, /* read */
324 { &vnop_write_desc, (vnop_t *)nfsspec_vnop_write }, /* write */
325 { &vnop_ioctl_desc, (vnop_t *)spec_ioctl }, /* ioctl */
326 { &vnop_select_desc, (vnop_t *)spec_select }, /* select */
327 { &vnop_revoke_desc, (vnop_t *)spec_revoke }, /* revoke */
328 { &vnop_mmap_desc, (vnop_t *)spec_mmap }, /* mmap */
329 { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
330 { &vnop_remove_desc, (vnop_t *)spec_remove }, /* remove */
331 { &vnop_link_desc, (vnop_t *)spec_link }, /* link */
332 { &vnop_rename_desc, (vnop_t *)spec_rename }, /* rename */
333 { &vnop_mkdir_desc, (vnop_t *)spec_mkdir }, /* mkdir */
334 { &vnop_rmdir_desc, (vnop_t *)spec_rmdir }, /* rmdir */
335 { &vnop_symlink_desc, (vnop_t *)spec_symlink }, /* symlink */
336 { &vnop_readdir_desc, (vnop_t *)spec_readdir }, /* readdir */
337 { &vnop_readlink_desc, (vnop_t *)spec_readlink }, /* readlink */
338 { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
339 { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
340 { &vnop_strategy_desc, (vnop_t *)spec_strategy }, /* strategy */
341 { &vnop_pathconf_desc, (vnop_t *)spec_pathconf }, /* pathconf */
342 { &vnop_advlock_desc, (vnop_t *)spec_advlock }, /* advlock */
343 { &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
344 { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
345 { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
346 { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
347 { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
348 { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
349 { &vnop_getxattr_desc, (vnop_t *)nfs4_vnop_getxattr }, /* getxattr */
350 { &vnop_setxattr_desc, (vnop_t *)nfs4_vnop_setxattr }, /* setxattr */
351 { &vnop_removexattr_desc, (vnop_t *)nfs4_vnop_removexattr },/* removexattr */
352 { &vnop_listxattr_desc, (vnop_t *)nfs4_vnop_listxattr },/* listxattr */
353 #if NAMEDSTREAMS
354 { &vnop_getnamedstream_desc, (vnop_t *)nfs4_vnop_getnamedstream }, /* getnamedstream */
355 { &vnop_makenamedstream_desc, (vnop_t *)nfs4_vnop_makenamedstream }, /* makenamedstream */
356 { &vnop_removenamedstream_desc, (vnop_t *)nfs4_vnop_removenamedstream },/* removenamedstream */
357 #endif
358 { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
359 { NULL, NULL }
360 };
361 const struct vnodeopv_desc spec_nfsv4nodeop_opv_desc =
362 { &spec_nfsv4nodeop_p, spec_nfsv4nodeop_entries };
363 #endif /* CONFIG_NFS4 */
364
365 #if FIFO
366 vnop_t **fifo_nfsv2nodeop_p;
367 static const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
368 { &vnop_default_desc, (vnop_t *)vn_default_error },
369 { &vnop_lookup_desc, (vnop_t *)fifo_lookup }, /* lookup */
370 { &vnop_create_desc, (vnop_t *)fifo_create }, /* create */
371 { &vnop_mknod_desc, (vnop_t *)fifo_mknod }, /* mknod */
372 { &vnop_open_desc, (vnop_t *)fifo_open }, /* open */
373 { &vnop_close_desc, (vnop_t *)nfsfifo_vnop_close }, /* close */
374 { &vnop_getattr_desc, (vnop_t *)nfs3_vnop_getattr }, /* getattr */
375 { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
376 { &vnop_read_desc, (vnop_t *)nfsfifo_vnop_read }, /* read */
377 { &vnop_write_desc, (vnop_t *)nfsfifo_vnop_write }, /* write */
378 { &vnop_ioctl_desc, (vnop_t *)fifo_ioctl }, /* ioctl */
379 { &vnop_select_desc, (vnop_t *)fifo_select }, /* select */
380 { &vnop_revoke_desc, (vnop_t *)fifo_revoke }, /* revoke */
381 { &vnop_mmap_desc, (vnop_t *)fifo_mmap }, /* mmap */
382 { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
383 { &vnop_remove_desc, (vnop_t *)fifo_remove }, /* remove */
384 { &vnop_link_desc, (vnop_t *)fifo_link }, /* link */
385 { &vnop_rename_desc, (vnop_t *)fifo_rename }, /* rename */
386 { &vnop_mkdir_desc, (vnop_t *)fifo_mkdir }, /* mkdir */
387 { &vnop_rmdir_desc, (vnop_t *)fifo_rmdir }, /* rmdir */
388 { &vnop_symlink_desc, (vnop_t *)fifo_symlink }, /* symlink */
389 { &vnop_readdir_desc, (vnop_t *)fifo_readdir }, /* readdir */
390 { &vnop_readlink_desc, (vnop_t *)fifo_readlink }, /* readlink */
391 { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
392 { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
393 { &vnop_strategy_desc, (vnop_t *)fifo_strategy }, /* strategy */
394 { &vnop_pathconf_desc, (vnop_t *)fifo_pathconf }, /* pathconf */
395 { &vnop_advlock_desc, (vnop_t *)fifo_advlock }, /* advlock */
396 { &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
397 { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
398 { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
399 { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
400 { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
401 { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
402 { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
403 { NULL, NULL }
404 };
405 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
406 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
407 #endif
408
409 #if CONFIG_NFS4
410 #if FIFO
411 vnop_t **fifo_nfsv4nodeop_p;
412 static const struct vnodeopv_entry_desc fifo_nfsv4nodeop_entries[] = {
413 { &vnop_default_desc, (vnop_t *)vn_default_error },
414 { &vnop_lookup_desc, (vnop_t *)fifo_lookup }, /* lookup */
415 { &vnop_create_desc, (vnop_t *)fifo_create }, /* create */
416 { &vnop_mknod_desc, (vnop_t *)fifo_mknod }, /* mknod */
417 { &vnop_open_desc, (vnop_t *)fifo_open }, /* open */
418 { &vnop_close_desc, (vnop_t *)nfsfifo_vnop_close }, /* close */
419 { &vnop_getattr_desc, (vnop_t *)nfs4_vnop_getattr }, /* getattr */
420 { &vnop_setattr_desc, (vnop_t *)nfs_vnop_setattr }, /* setattr */
421 { &vnop_read_desc, (vnop_t *)nfsfifo_vnop_read }, /* read */
422 { &vnop_write_desc, (vnop_t *)nfsfifo_vnop_write }, /* write */
423 { &vnop_ioctl_desc, (vnop_t *)fifo_ioctl }, /* ioctl */
424 { &vnop_select_desc, (vnop_t *)fifo_select }, /* select */
425 { &vnop_revoke_desc, (vnop_t *)fifo_revoke }, /* revoke */
426 { &vnop_mmap_desc, (vnop_t *)fifo_mmap }, /* mmap */
427 { &vnop_fsync_desc, (vnop_t *)nfs_vnop_fsync }, /* fsync */
428 { &vnop_remove_desc, (vnop_t *)fifo_remove }, /* remove */
429 { &vnop_link_desc, (vnop_t *)fifo_link }, /* link */
430 { &vnop_rename_desc, (vnop_t *)fifo_rename }, /* rename */
431 { &vnop_mkdir_desc, (vnop_t *)fifo_mkdir }, /* mkdir */
432 { &vnop_rmdir_desc, (vnop_t *)fifo_rmdir }, /* rmdir */
433 { &vnop_symlink_desc, (vnop_t *)fifo_symlink }, /* symlink */
434 { &vnop_readdir_desc, (vnop_t *)fifo_readdir }, /* readdir */
435 { &vnop_readlink_desc, (vnop_t *)fifo_readlink }, /* readlink */
436 { &vnop_inactive_desc, (vnop_t *)nfs_vnop_inactive }, /* inactive */
437 { &vnop_reclaim_desc, (vnop_t *)nfs_vnop_reclaim }, /* reclaim */
438 { &vnop_strategy_desc, (vnop_t *)fifo_strategy }, /* strategy */
439 { &vnop_pathconf_desc, (vnop_t *)fifo_pathconf }, /* pathconf */
440 { &vnop_advlock_desc, (vnop_t *)fifo_advlock }, /* advlock */
441 { &vnop_bwrite_desc, (vnop_t *)vn_bwrite }, /* bwrite */
442 { &vnop_pagein_desc, (vnop_t *)nfs_vnop_pagein }, /* Pagein */
443 { &vnop_pageout_desc, (vnop_t *)nfs_vnop_pageout }, /* Pageout */
444 { &vnop_blktooff_desc, (vnop_t *)nfs_vnop_blktooff }, /* blktooff */
445 { &vnop_offtoblk_desc, (vnop_t *)nfs_vnop_offtoblk }, /* offtoblk */
446 { &vnop_blockmap_desc, (vnop_t *)nfs_vnop_blockmap }, /* blockmap */
447 { &vnop_getxattr_desc, (vnop_t *)nfs4_vnop_getxattr }, /* getxattr */
448 { &vnop_setxattr_desc, (vnop_t *)nfs4_vnop_setxattr }, /* setxattr */
449 { &vnop_removexattr_desc, (vnop_t *)nfs4_vnop_removexattr },/* removexattr */
450 { &vnop_listxattr_desc, (vnop_t *)nfs4_vnop_listxattr },/* listxattr */
451 #if NAMEDSTREAMS
452 { &vnop_getnamedstream_desc, (vnop_t *)nfs4_vnop_getnamedstream }, /* getnamedstream */
453 { &vnop_makenamedstream_desc, (vnop_t *)nfs4_vnop_makenamedstream }, /* makenamedstream */
454 { &vnop_removenamedstream_desc, (vnop_t *)nfs4_vnop_removenamedstream },/* removenamedstream */
455 #endif
456 { &vnop_monitor_desc, (vnop_t *)nfs_vnop_monitor }, /* monitor */
457 { NULL, NULL }
458 };
459 const struct vnodeopv_desc fifo_nfsv4nodeop_opv_desc =
460 { &fifo_nfsv4nodeop_p, fifo_nfsv4nodeop_entries };
461 #endif /* FIFO */
462 #endif /* CONFIG_NFS4 */
463
464 int nfs_sillyrename(nfsnode_t, nfsnode_t, struct componentname *, vfs_context_t);
465 int nfs_getattr_internal(nfsnode_t, struct nfs_vattr *, vfs_context_t, int);
466 int nfs_refresh_fh(nfsnode_t, vfs_context_t);
467
468
469 ZONE_VIEW_DEFINE(ZV_NFSDIROFF, "NFSV3 diroff",
470 KHEAP_ID_DATA_BUFFERS, sizeof(struct nfsdmap));
471
472 static void
473 nfs_dir_buf_cache_lookup_boundaries(struct nfsbuf *bp, int *sof, int *eof)
474 {
475 if (bp) {
476 struct nfs_dir_buf_header *ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
477 if (sof && bp->nb_lblkno == 0) {
478 *sof = 1;
479 }
480 if (eof && ISSET(ndbhp->ndbh_flags, NDB_EOF)) {
481 *eof = 1;
482 }
483 }
484 }
485
486 /*
487 * Update nfsnode attributes to avoid extra getattr calls for each direntry.
488 * This function should be called only if RDIRPLUS flag is enabled.
489 */
490 void
491 nfs_rdirplus_update_node_attrs(nfsnode_t dnp, struct direntry *dp, fhandle_t *fhp, struct nfs_vattr *nvattrp, uint64_t *savedxidp)
492 {
493 nfsnode_t np;
494 struct componentname cn;
495 int isdot = (dp->d_namlen == 1) && (dp->d_name[0] == '.');
496 int isdotdot = (dp->d_namlen == 2) && (dp->d_name[0] == '.') && (dp->d_name[1] == '.');
497 int should_update_fileid = nvattrp->nva_flags & NFS_FFLAG_FILEID_CONTAINS_XID;
498 uint64_t xid = 0;
499
500 if (isdot || isdotdot) {
501 return;
502 }
503
504 np = NULL;
505 bzero(&cn, sizeof(cn));
506 cn.cn_nameptr = dp->d_name;
507 cn.cn_namelen = dp->d_namlen;
508 cn.cn_nameiop = LOOKUP;
509
510 /* xid might be stashed in nva_fileid is rdirplus is enabled */
511 if (should_update_fileid) {
512 xid = nvattrp->nva_fileid;
513 nvattrp->nva_fileid = dp->d_fileno;
514 }
515 nfs_nget(NFSTOMP(dnp), dnp, &cn, fhp->fh_data, fhp->fh_len, nvattrp, savedxidp, RPCAUTH_UNKNOWN, NG_NOCREATE, &np);
516 if (should_update_fileid) {
517 nvattrp->nva_fileid = xid;
518 }
519 if (np) {
520 nfs_node_unlock(np);
521 vnode_put(NFSTOV(np));
522 }
523 }
524
525 /*
526 * Find the slot in the access cache for this UID.
527 * If adding and no existing slot is found, reuse slots in FIFO order.
528 * The index of the next slot to use is kept in the last entry of the n_access array.
529 */
530 int
531 nfs_node_access_slot(nfsnode_t np, uid_t uid, int add)
532 {
533 int slot;
534
535 for (slot = 0; slot < NFS_ACCESS_CACHE_SIZE; slot++) {
536 if (np->n_accessuid[slot] == uid) {
537 break;
538 }
539 }
540 if (slot == NFS_ACCESS_CACHE_SIZE) {
541 if (!add) {
542 return -1;
543 }
544 slot = np->n_access[NFS_ACCESS_CACHE_SIZE];
545 np->n_access[NFS_ACCESS_CACHE_SIZE] = (slot + 1) % NFS_ACCESS_CACHE_SIZE;
546 }
547 return slot;
548 }
549
550 int
551 nfs3_access_rpc(nfsnode_t np, u_int32_t *access, int rpcflags, vfs_context_t ctx)
552 {
553 int error = 0, lockerror = ENOENT, status = 0, slot;
554 uint32_t access_result = 0;
555 u_int64_t xid;
556 struct nfsm_chain nmreq, nmrep;
557 struct nfsmount *nmp;
558 struct timeval now;
559 uid_t uid;
560
561 nfsm_chain_null(&nmreq);
562 nfsm_chain_null(&nmrep);
563
564 nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(NFS_VER3) + NFSX_UNSIGNED);
565 nfsm_chain_add_fh(error, &nmreq, NFS_VER3, np->n_fhp, np->n_fhsize);
566 nfsm_chain_add_32(error, &nmreq, *access);
567 nfsm_chain_build_done(error, &nmreq);
568 nfsmout_if(error);
569 error = nfs_request2(np, NULL, &nmreq, NFSPROC_ACCESS,
570 vfs_context_thread(ctx), vfs_context_ucred(ctx),
571 NULL, rpcflags, &nmrep, &xid, &status);
572 if ((lockerror = nfs_node_lock(np))) {
573 error = lockerror;
574 }
575 nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
576 if (!error) {
577 error = status;
578 }
579 nfsm_chain_get_32(error, &nmrep, access_result);
580 nfsmout_if(error);
581
582 /* XXXab do we really need mount here, also why are we doing access cache management here? */
583 nmp = NFSTONMP(np);
584 if (nfs_mount_gone(nmp)) {
585 error = ENXIO;
586 }
587 nfsmout_if(error);
588
589 #if CONFIG_NFS_GSS
590 if (auth_is_kerberized(np->n_auth) || auth_is_kerberized(nmp->nm_auth)) {
591 uid = nfs_cred_getasid2uid(vfs_context_ucred(ctx));
592 } else {
593 uid = kauth_cred_getuid(vfs_context_ucred(ctx));
594 }
595 #else
596 uid = kauth_cred_getuid(vfs_context_ucred(ctx));
597 #endif /* CONFIG_NFS_GSS */
598 slot = nfs_node_access_slot(np, uid, 1);
599 np->n_accessuid[slot] = uid;
600 microuptime(&now);
601 np->n_accessstamp[slot] = now.tv_sec;
602 np->n_access[slot] = access_result;
603
604 /*
605 * If we asked for DELETE but didn't get it, the server
606 * may simply not support returning that bit (possible
607 * on UNIX systems). So, we'll assume that it is OK,
608 * and just let any subsequent delete action fail if it
609 * really isn't deletable.
610 */
611 if ((*access & NFS_ACCESS_DELETE) &&
612 !(np->n_access[slot] & NFS_ACCESS_DELETE)) {
613 np->n_access[slot] |= NFS_ACCESS_DELETE;
614 }
615 /* ".zfs" subdirectories may erroneously give a denied answer for add/remove */
616 if (nfs_access_dotzfs && (np->n_flag & NISDOTZFSCHILD)) {
617 np->n_access[slot] |= (NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND | NFS_ACCESS_DELETE);
618 }
619 /* pass back the access returned with this request */
620 *access = np->n_access[slot];
621 nfsmout:
622 if (!lockerror) {
623 nfs_node_unlock(np);
624 }
625 nfsm_chain_cleanup(&nmreq);
626 nfsm_chain_cleanup(&nmrep);
627 return error;
628 }
629
630
631 /*
632 * NFS access vnode op.
633 * For NFS version 2, just return ok. File accesses may fail later.
634 * For NFS version 3+, use the access RPC to check accessibility. If file
635 * permissions are changed on the server, accesses might still fail later.
636 */
637 int
638 nfs_vnop_access(
639 struct vnop_access_args /* {
640 * struct vnodeop_desc *a_desc;
641 * vnode_t a_vp;
642 * int a_action;
643 * vfs_context_t a_context;
644 * } */*ap)
645 {
646 vfs_context_t ctx = ap->a_context;
647 vnode_t vp = ap->a_vp;
648 int error = 0, slot, dorpc, rpcflags = 0;
649 u_int32_t access, waccess;
650 nfsnode_t np = VTONFS(vp);
651 struct nfsmount *nmp;
652 int nfsvers;
653 struct timeval now;
654 uid_t uid;
655
656 nmp = VTONMP(vp);
657 if (nfs_mount_gone(nmp)) {
658 return ENXIO;
659 }
660 nfsvers = nmp->nm_vers;
661
662
663 if (nfsvers == NFS_VER2 || NMFLAG(nmp, NOOPAQUE_AUTH)) {
664 if ((ap->a_action & KAUTH_VNODE_WRITE_RIGHTS) &&
665 vfs_isrdonly(vnode_mount(vp))) {
666 return EROFS;
667 }
668 return 0;
669 }
670
671 /*
672 * For NFS v3, do an access rpc, otherwise you are stuck emulating
673 * ufs_access() locally using the vattr. This may not be correct,
674 * since the server may apply other access criteria such as
675 * client uid-->server uid mapping that we do not know about, but
676 * this is better than just returning anything that is lying about
677 * in the cache.
678 */
679
680 /*
681 * Convert KAUTH primitives to NFS access rights.
682 */
683 access = 0;
684 if (vnode_isdir(vp)) {
685 /* directory */
686 if (ap->a_action &
687 (KAUTH_VNODE_LIST_DIRECTORY |
688 KAUTH_VNODE_READ_EXTATTRIBUTES)) {
689 access |= NFS_ACCESS_READ;
690 }
691 if (ap->a_action & KAUTH_VNODE_SEARCH) {
692 access |= NFS_ACCESS_LOOKUP;
693 }
694 if (ap->a_action &
695 (KAUTH_VNODE_ADD_FILE |
696 KAUTH_VNODE_ADD_SUBDIRECTORY)) {
697 access |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
698 }
699 if (ap->a_action & KAUTH_VNODE_DELETE_CHILD) {
700 access |= NFS_ACCESS_MODIFY;
701 }
702 } else {
703 /* file */
704 if (ap->a_action &
705 (KAUTH_VNODE_READ_DATA |
706 KAUTH_VNODE_READ_EXTATTRIBUTES)) {
707 access |= NFS_ACCESS_READ;
708 }
709 if (ap->a_action & KAUTH_VNODE_WRITE_DATA) {
710 access |= NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
711 }
712 if (ap->a_action & KAUTH_VNODE_APPEND_DATA) {
713 access |= NFS_ACCESS_EXTEND;
714 }
715 if (ap->a_action & KAUTH_VNODE_EXECUTE) {
716 access |= NFS_ACCESS_EXECUTE;
717 }
718 }
719 /* common */
720 if (ap->a_action & KAUTH_VNODE_DELETE) {
721 access |= NFS_ACCESS_DELETE;
722 }
723 if (ap->a_action &
724 (KAUTH_VNODE_WRITE_ATTRIBUTES |
725 KAUTH_VNODE_WRITE_EXTATTRIBUTES |
726 KAUTH_VNODE_WRITE_SECURITY)) {
727 access |= NFS_ACCESS_MODIFY;
728 }
729 /* XXX this is pretty dubious */
730 if (ap->a_action & KAUTH_VNODE_CHANGE_OWNER) {
731 access |= NFS_ACCESS_MODIFY;
732 }
733
734 /* if caching, always ask for every right */
735 if (nfs_access_cache_timeout > 0) {
736 waccess = NFS_ACCESS_READ | NFS_ACCESS_MODIFY |
737 NFS_ACCESS_EXTEND | NFS_ACCESS_EXECUTE |
738 NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
739 } else {
740 waccess = access;
741 }
742
743 if ((error = nfs_node_lock(np))) {
744 return error;
745 }
746
747 /*
748 * Does our cached result allow us to give a definite yes to
749 * this request?
750 */
751 #if CONFIG_NFS_GSS
752 if (auth_is_kerberized(np->n_auth) || auth_is_kerberized(nmp->nm_auth)) {
753 uid = nfs_cred_getasid2uid(vfs_context_ucred(ctx));
754 } else {
755 uid = kauth_cred_getuid(vfs_context_ucred(ctx));
756 }
757 #else
758 uid = kauth_cred_getuid(vfs_context_ucred(ctx));
759 #endif /* CONFIG_NFS_GSS */
760 slot = nfs_node_access_slot(np, uid, 0);
761 dorpc = 1;
762 if (access == 0) {
763 /* not asking for any rights understood by NFS, so don't bother doing an RPC */
764 /* OSAddAtomic(1, &nfsstats.accesscache_hits); */
765 dorpc = 0;
766 waccess = 0;
767 } else if (NACCESSVALID(np, slot)) {
768 microuptime(&now);
769 if (((now.tv_sec < (np->n_accessstamp[slot] + nfs_access_cache_timeout)) &&
770 ((np->n_access[slot] & access) == access)) || nfs_use_cache(nmp)) {
771 /* OSAddAtomic(1, &nfsstats.accesscache_hits); */
772 dorpc = 0;
773 waccess = np->n_access[slot];
774 }
775 }
776 nfs_node_unlock(np);
777 if (dorpc) {
778 /* Either a no, or a don't know. Go to the wire. */
779 /* OSAddAtomic(1, &nfsstats.accesscache_misses); */
780
781 /*
782 * Allow an access call to timeout if we have it cached
783 * so we won't hang if the server isn't responding.
784 */
785 if (NACCESSVALID(np, slot)) {
786 rpcflags |= R_SOFT;
787 }
788
789 error = nmp->nm_funcs->nf_access_rpc(np, &waccess, rpcflags, ctx);
790
791 /*
792 * If the server didn't respond return the cached access.
793 */
794 if ((error == ETIMEDOUT) && (rpcflags & R_SOFT)) {
795 error = 0;
796 waccess = np->n_access[slot];
797 }
798 }
799 if (!error && ((waccess & access) != access)) {
800 error = EACCES;
801 }
802
803 return error;
804 }
805
806
807 /*
808 * NFS open vnode op
809 *
810 * Perform various update/invalidation checks and then add the
811 * open to the node. Regular files will have an open file structure
812 * on the node and, for NFSv4, perform an OPEN request on the server.
813 */
814 int
815 nfs_vnop_open(
816 struct vnop_open_args /* {
817 * struct vnodeop_desc *a_desc;
818 * vnode_t a_vp;
819 * int a_mode;
820 * vfs_context_t a_context;
821 * } */*ap)
822 {
823 vfs_context_t ctx = ap->a_context;
824 vnode_t vp = ap->a_vp;
825 nfsnode_t np = VTONFS(vp);
826 struct nfsmount *nmp;
827 int error, accessMode, denyMode, opened = 0;
828 struct nfs_open_owner *noop = NULL;
829 struct nfs_open_file *nofp = NULL;
830 enum vtype vtype;
831
832 if (!(ap->a_mode & (FREAD | FWRITE))) {
833 return EINVAL;
834 }
835
836 nmp = VTONMP(vp);
837 if (nfs_mount_gone(nmp)) {
838 return ENXIO;
839 }
840 if (np->n_flag & NREVOKE) {
841 return EIO;
842 }
843
844 vtype = vnode_vtype(vp);
845 if ((vtype != VREG) && (vtype != VDIR) && (vtype != VLNK)) {
846 return EACCES;
847 }
848
849 /* First, check if we need to update/invalidate */
850 if (ISSET(np->n_flag, NUPDATESIZE)) {
851 nfs_data_update_size(np, 0);
852 }
853 if ((error = nfs_node_lock(np))) {
854 return error;
855 }
856 if (np->n_flag & NNEEDINVALIDATE) {
857 np->n_flag &= ~NNEEDINVALIDATE;
858 if (vtype == VDIR) {
859 nfs_invaldir(np);
860 }
861 nfs_node_unlock(np);
862 nfs_vinvalbuf(vp, V_SAVE | V_IGNORE_WRITEERR, ctx, 1);
863 if ((error = nfs_node_lock(np))) {
864 return error;
865 }
866 }
867 if (vtype == VREG) {
868 np->n_lastrahead = -1;
869 }
870 if (np->n_flag & NMODIFIED) {
871 if (vtype == VDIR) {
872 nfs_invaldir(np);
873 }
874 nfs_node_unlock(np);
875 if ((error = nfs_vinvalbuf(vp, V_SAVE | V_IGNORE_WRITEERR, ctx, 1))) {
876 return error;
877 }
878 } else {
879 nfs_node_unlock(np);
880 }
881
882 /* nfs_getattr() will check changed and purge caches */
883 if ((error = nfs_getattr(np, NULL, ctx, NGA_UNCACHED))) {
884 return error;
885 }
886
887 if (vtype != VREG) {
888 /* Just mark that it was opened */
889 lck_mtx_lock(&np->n_openlock);
890 np->n_openrefcnt++;
891 lck_mtx_unlock(&np->n_openlock);
892 return 0;
893 }
894
895 /* mode contains some combination of: FREAD, FWRITE, O_SHLOCK, O_EXLOCK */
896 accessMode = 0;
897 if (ap->a_mode & FREAD) {
898 accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
899 }
900 if (ap->a_mode & FWRITE) {
901 accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
902 }
903 if (ap->a_mode & O_EXLOCK) {
904 denyMode = NFS_OPEN_SHARE_DENY_BOTH;
905 } else if (ap->a_mode & O_SHLOCK) {
906 denyMode = NFS_OPEN_SHARE_DENY_WRITE;
907 } else {
908 denyMode = NFS_OPEN_SHARE_DENY_NONE;
909 }
910 // XXX don't do deny modes just yet (and never do it for !v4)
911 denyMode = NFS_OPEN_SHARE_DENY_NONE;
912
913 noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 1);
914 if (!noop) {
915 return ENOMEM;
916 }
917
918 restart:
919 error = nfs_mount_state_in_use_start(nmp, vfs_context_thread(ctx));
920 if (error) {
921 nfs_open_owner_rele(noop);
922 return error;
923 }
924 if (np->n_flag & NREVOKE) {
925 error = EIO;
926 nfs_mount_state_in_use_end(nmp, 0);
927 nfs_open_owner_rele(noop);
928 return error;
929 }
930
931 error = nfs_open_file_find(np, noop, &nofp, accessMode, denyMode, 1);
932 if (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
933 NP(np, "nfs_vnop_open: LOST %d", kauth_cred_getuid(nofp->nof_owner->noo_cred));
934 error = EIO;
935 }
936 #if CONFIG_NFS4
937 if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
938 error = nfs4_reopen(nofp, vfs_context_thread(ctx));
939 nofp = NULL;
940 if (!error) {
941 nfs_mount_state_in_use_end(nmp, 0);
942 goto restart;
943 }
944 }
945 #endif
946 if (!error) {
947 error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
948 }
949 if (error) {
950 nofp = NULL;
951 goto out;
952 }
953
954 if (nmp->nm_vers < NFS_VER4) {
955 /*
956 * NFS v2/v3 opens are always allowed - so just add it.
957 */
958 nfs_open_file_add_open(nofp, accessMode, denyMode, 0);
959 goto out;
960 }
961
962 /*
963 * If we just created the file and the modes match, then we simply use
964 * the open performed in the create. Otherwise, send the request.
965 */
966 if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
967 (nofp->nof_creator == current_thread()) &&
968 (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) &&
969 (denyMode == NFS_OPEN_SHARE_DENY_NONE)) {
970 nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
971 nofp->nof_creator = NULL;
972 } else {
973 #if CONFIG_NFS4
974 if (!opened) {
975 error = nfs4_open(np, nofp, accessMode, denyMode, ctx);
976 }
977 #endif
978 if ((error == EACCES) && (nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
979 (nofp->nof_creator == current_thread())) {
980 /*
981 * Ugh. This can happen if we just created the file with read-only
982 * perms and we're trying to open it for real with different modes
983 * (e.g. write-only or with a deny mode) and the server decides to
984 * not allow the second open because of the read-only perms.
985 * The best we can do is to just use the create's open.
986 * We may have access we don't need or we may not have a requested
987 * deny mode. We may log complaints later, but we'll try to avoid it.
988 */
989 if (denyMode != NFS_OPEN_SHARE_DENY_NONE) {
990 NP(np, "nfs_vnop_open: deny mode foregone on create, %d", kauth_cred_getuid(nofp->nof_owner->noo_cred));
991 }
992 nofp->nof_creator = NULL;
993 error = 0;
994 }
995 if (error) {
996 goto out;
997 }
998 opened = 1;
999 /*
1000 * If we had just created the file, we already had it open.
1001 * If the actual open mode is less than what we grabbed at
1002 * create time, then we'll downgrade the open here.
1003 */
1004 if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
1005 (nofp->nof_creator == current_thread())) {
1006 error = nfs_close(np, nofp, NFS_OPEN_SHARE_ACCESS_BOTH, NFS_OPEN_SHARE_DENY_NONE, ctx);
1007 if (error) {
1008 NP(np, "nfs_vnop_open: create close error %d, %d", error, kauth_cred_getuid(nofp->nof_owner->noo_cred));
1009 }
1010 if (!nfs_mount_state_error_should_restart(error)) {
1011 error = 0;
1012 nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
1013 }
1014 }
1015 }
1016
1017 out:
1018 if (nofp) {
1019 nfs_open_file_clear_busy(nofp);
1020 }
1021 if (nfs_mount_state_in_use_end(nmp, error)) {
1022 nofp = NULL;
1023 goto restart;
1024 }
1025 if (error) {
1026 NP(np, "nfs_vnop_open: error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
1027 }
1028 if (noop) {
1029 nfs_open_owner_rele(noop);
1030 }
1031 if (!error && vtype == VREG && (ap->a_mode & FWRITE)) {
1032 lck_mtx_lock(&nmp->nm_lock);
1033 nmp->nm_state &= ~NFSSTA_SQUISHY;
1034 nmp->nm_curdeadtimeout = nmp->nm_deadtimeout;
1035 if (nmp->nm_curdeadtimeout <= 0) {
1036 nmp->nm_deadto_start = 0;
1037 }
1038 nmp->nm_writers++;
1039 lck_mtx_unlock(&nmp->nm_lock);
1040 }
1041
1042 return error;
1043 }
1044
1045 static uint32_t
1046 nfs_no_of_open_file_writers(nfsnode_t np)
1047 {
1048 uint32_t writers = 0;
1049 struct nfs_open_file *nofp;
1050
1051 TAILQ_FOREACH(nofp, &np->n_opens, nof_link) {
1052 writers += nofp->nof_w + nofp->nof_rw + nofp->nof_w_dw + nofp->nof_rw_dw +
1053 nofp->nof_w_drw + nofp->nof_rw_drw + nofp->nof_d_w_dw +
1054 nofp->nof_d_rw_dw + nofp->nof_d_w_drw + nofp->nof_d_rw_drw +
1055 nofp->nof_d_w + nofp->nof_d_rw;
1056 }
1057
1058 return writers;
1059 }
1060
1061 /*
1062 * NFS close vnode op
1063 *
1064 * What an NFS client should do upon close after writing is a debatable issue.
1065 * Most NFS clients push delayed writes to the server upon close, basically for
1066 * two reasons:
1067 * 1 - So that any write errors may be reported back to the client process
1068 * doing the close system call. By far the two most likely errors are
1069 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
1070 * 2 - To put a worst case upper bound on cache inconsistency between
1071 * multiple clients for the file.
1072 * There is also a consistency problem for Version 2 of the protocol w.r.t.
1073 * not being able to tell if other clients are writing a file concurrently,
1074 * since there is no way of knowing if the changed modify time in the reply
1075 * is only due to the write for this client.
1076 * (NFS Version 3 provides weak cache consistency data in the reply that
1077 * should be sufficient to detect and handle this case.)
1078 *
1079 * The current code does the following:
1080 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
1081 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate them.
1082 * for NFS Version 4 - basically the same as NFSv3
1083 */
1084 int
1085 nfs_vnop_close(
1086 struct vnop_close_args /* {
1087 * struct vnodeop_desc *a_desc;
1088 * vnode_t a_vp;
1089 * int a_fflag;
1090 * vfs_context_t a_context;
1091 * } */*ap)
1092 {
1093 vfs_context_t ctx = ap->a_context;
1094 vnode_t vp = ap->a_vp;
1095 nfsnode_t np = VTONFS(vp);
1096 struct nfsmount *nmp;
1097 int error = 0, error1, nfsvers;
1098 int fflag = ap->a_fflag;
1099 enum vtype vtype;
1100 int accessMode, denyMode;
1101 struct nfs_open_owner *noop = NULL;
1102 struct nfs_open_file *nofp = NULL;
1103
1104 nmp = VTONMP(vp);
1105 if (!nmp) {
1106 return ENXIO;
1107 }
1108 nfsvers = nmp->nm_vers;
1109 vtype = vnode_vtype(vp);
1110
1111 /* First, check if we need to update/flush/invalidate */
1112 if (ISSET(np->n_flag, NUPDATESIZE)) {
1113 nfs_data_update_size(np, 0);
1114 }
1115 nfs_node_lock_force(np);
1116 if (np->n_flag & NNEEDINVALIDATE) {
1117 np->n_flag &= ~NNEEDINVALIDATE;
1118 nfs_node_unlock(np);
1119 nfs_vinvalbuf(vp, V_SAVE | V_IGNORE_WRITEERR, ctx, 1);
1120 nfs_node_lock_force(np);
1121 }
1122 if ((vtype == VREG) && (np->n_flag & NMODIFIED) && (fflag & FWRITE)) {
1123 /* we're closing an open for write and the file is modified, so flush it */
1124 nfs_node_unlock(np);
1125 if (nfsvers != NFS_VER2) {
1126 error = nfs_flush(np, MNT_WAIT, vfs_context_thread(ctx), 0);
1127 } else {
1128 error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
1129 }
1130 nfs_node_lock_force(np);
1131 NATTRINVALIDATE(np);
1132 }
1133 if (np->n_flag & NWRITEERR) {
1134 np->n_flag &= ~NWRITEERR;
1135 error = np->n_error;
1136 }
1137 nfs_node_unlock(np);
1138
1139 if (vtype != VREG) {
1140 /* Just mark that it was closed */
1141 lck_mtx_lock(&np->n_openlock);
1142 if (np->n_openrefcnt == 0) {
1143 if (fflag & (FREAD | FWRITE)) {
1144 NP(np, "nfs_vnop_close: open reference underrun");
1145 error = EINVAL;
1146 }
1147 } else if (fflag & (FREAD | FWRITE)) {
1148 np->n_openrefcnt--;
1149 } else {
1150 /* No FREAD/FWRITE set - probably the final close */
1151 np->n_openrefcnt = 0;
1152 }
1153 lck_mtx_unlock(&np->n_openlock);
1154 return error;
1155 }
1156 error1 = error;
1157
1158 /* fflag should contain some combination of: FREAD, FWRITE */
1159 accessMode = 0;
1160 if (fflag & FREAD) {
1161 accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
1162 }
1163 if (fflag & FWRITE) {
1164 accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
1165 }
1166 // XXX It would be nice if we still had the O_EXLOCK/O_SHLOCK flags that were on the open
1167 // if (fflag & O_EXLOCK)
1168 // denyMode = NFS_OPEN_SHARE_DENY_BOTH;
1169 // else if (fflag & O_SHLOCK)
1170 // denyMode = NFS_OPEN_SHARE_DENY_WRITE;
1171 // else
1172 // denyMode = NFS_OPEN_SHARE_DENY_NONE;
1173 // XXX don't do deny modes just yet (and never do it for !v4)
1174 denyMode = NFS_OPEN_SHARE_DENY_NONE;
1175
1176 if (!accessMode) {
1177 /*
1178 * No mode given to close?
1179 * Guess this is the final close.
1180 * We should unlock all locks and close all opens.
1181 */
1182 uint32_t writers;
1183 mount_t mp = vnode_mount(vp);
1184 int force = (!mp || vfs_isforce(mp));
1185
1186 writers = nfs_no_of_open_file_writers(np);
1187 nfs_release_open_state_for_node(np, force);
1188 if (writers) {
1189 lck_mtx_lock(&nmp->nm_lock);
1190 if (writers > nmp->nm_writers) {
1191 NP(np, "nfs_vnop_close: number of write opens for mount underrun. Node has %d"
1192 " opens for write. Mount has total of %d opens for write\n",
1193 writers, nmp->nm_writers);
1194 nmp->nm_writers = 0;
1195 } else {
1196 nmp->nm_writers -= writers;
1197 }
1198 lck_mtx_unlock(&nmp->nm_lock);
1199 }
1200
1201 return error;
1202 } else if (fflag & FWRITE) {
1203 lck_mtx_lock(&nmp->nm_lock);
1204 if (nmp->nm_writers == 0) {
1205 NP(np, "nfs_vnop_close: removing open writer from mount, but mount has no files open for writing");
1206 } else {
1207 nmp->nm_writers--;
1208 }
1209 lck_mtx_unlock(&nmp->nm_lock);
1210 }
1211
1212
1213 noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 0);
1214 if (!noop) {
1215 // printf("nfs_vnop_close: can't get open owner!\n");
1216 return EIO;
1217 }
1218
1219 restart:
1220 error = nfs_mount_state_in_use_start(nmp, NULL);
1221 if (error) {
1222 nfs_open_owner_rele(noop);
1223 return error;
1224 }
1225
1226 error = nfs_open_file_find(np, noop, &nofp, 0, 0, 0);
1227 #if CONFIG_NFS4
1228 if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
1229 error = nfs4_reopen(nofp, NULL);
1230 nofp = NULL;
1231 if (!error) {
1232 nfs_mount_state_in_use_end(nmp, 0);
1233 goto restart;
1234 }
1235 }
1236 #endif
1237 if (error) {
1238 NP(np, "nfs_vnop_close: no open file for owner, error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
1239 error = EBADF;
1240 goto out;
1241 }
1242 error = nfs_open_file_set_busy(nofp, NULL);
1243 if (error) {
1244 nofp = NULL;
1245 goto out;
1246 }
1247
1248 error = nfs_close(np, nofp, accessMode, denyMode, ctx);
1249 if (error) {
1250 NP(np, "nfs_vnop_close: close error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
1251 }
1252
1253 out:
1254 if (nofp) {
1255 nfs_open_file_clear_busy(nofp);
1256 }
1257 if (nfs_mount_state_in_use_end(nmp, error)) {
1258 nofp = NULL;
1259 goto restart;
1260 }
1261 if (!error) {
1262 error = error1;
1263 }
1264 if (error) {
1265 NP(np, "nfs_vnop_close: error %d, %d", error, kauth_cred_getuid(noop->noo_cred));
1266 }
1267 if (noop) {
1268 nfs_open_owner_rele(noop);
1269 }
1270 return error;
1271 }
1272
1273 /*
1274 * nfs_close(): common function that does all the heavy lifting of file closure
1275 *
1276 * Takes an open file structure and a set of access/deny modes and figures out how
1277 * to update the open file structure (and the state on the server) appropriately.
1278 */
1279 int
1280 nfs_close(
1281 nfsnode_t np,
1282 struct nfs_open_file *nofp,
1283 uint32_t accessMode,
1284 uint32_t denyMode,
1285 __unused vfs_context_t ctx)
1286 {
1287 #if CONFIG_NFS4
1288 struct nfs_lock_owner *nlop;
1289 #endif
1290 int error = 0, changed = 0, delegated = 0, closed = 0, downgrade = 0;
1291 uint8_t newAccessMode, newDenyMode;
1292
1293 /* warn if modes don't match current state */
1294 if (((accessMode & nofp->nof_access) != accessMode) || ((denyMode & nofp->nof_deny) != denyMode)) {
1295 NP(np, "nfs_close: mode mismatch %d %d, current %d %d, %d",
1296 accessMode, denyMode, nofp->nof_access, nofp->nof_deny,
1297 kauth_cred_getuid(nofp->nof_owner->noo_cred));
1298 }
1299
1300 /*
1301 * If we're closing a write-only open, we may not have a write-only count
1302 * if we also grabbed read access. So, check the read-write count.
1303 */
1304 if (denyMode == NFS_OPEN_SHARE_DENY_NONE) {
1305 if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
1306 (nofp->nof_w == 0) && (nofp->nof_d_w == 0) &&
1307 (nofp->nof_rw || nofp->nof_d_rw)) {
1308 accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
1309 }
1310 } else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE) {
1311 if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
1312 (nofp->nof_w_dw == 0) && (nofp->nof_d_w_dw == 0) &&
1313 (nofp->nof_rw_dw || nofp->nof_d_rw_dw)) {
1314 accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
1315 }
1316 } else { /* NFS_OPEN_SHARE_DENY_BOTH */
1317 if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
1318 (nofp->nof_w_drw == 0) && (nofp->nof_d_w_drw == 0) &&
1319 (nofp->nof_rw_drw || nofp->nof_d_rw_drw)) {
1320 accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
1321 }
1322 }
1323
1324 nfs_open_file_remove_open_find(nofp, accessMode, denyMode, &newAccessMode, &newDenyMode, &delegated);
1325 if ((newAccessMode != nofp->nof_access) || (newDenyMode != nofp->nof_deny)) {
1326 changed = 1;
1327 } else {
1328 changed = 0;
1329 }
1330
1331 if (NFSTONMP(np)->nm_vers < NFS_VER4) {
1332 /* NFS v2/v3 closes simply need to remove the open. */
1333 goto v3close;
1334 }
1335 #if CONFIG_NFS4
1336 if ((newAccessMode == 0) || (nofp->nof_opencnt == 1)) {
1337 /*
1338 * No more access after this close, so clean up and close it.
1339 * Don't send a close RPC if we're closing a delegated open.
1340 */
1341 nfs_wait_bufs(np);
1342 closed = 1;
1343 if (!delegated && !(nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
1344 error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0);
1345 }
1346 if (error == NFSERR_LOCKS_HELD) {
1347 /*
1348 * Hmm... the server says we have locks we need to release first
1349 * Find the lock owner and try to unlock everything.
1350 */
1351 nlop = nfs_lock_owner_find(np, vfs_context_proc(ctx), 0);
1352 if (nlop) {
1353 nfs4_unlock_rpc(np, nlop, F_WRLCK, 0, UINT64_MAX,
1354 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
1355 nfs_lock_owner_rele(nlop);
1356 }
1357 error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0);
1358 }
1359 } else if (changed) {
1360 /*
1361 * File is still open but with less access, so downgrade the open.
1362 * Don't send a downgrade RPC if we're closing a delegated open.
1363 */
1364 if (!delegated && !(nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
1365 downgrade = 1;
1366 /*
1367 * If we have delegated opens, we should probably claim them before sending
1368 * the downgrade because the server may not know the open we are downgrading to.
1369 */
1370 if (nofp->nof_d_rw_drw || nofp->nof_d_w_drw || nofp->nof_d_r_drw ||
1371 nofp->nof_d_rw_dw || nofp->nof_d_w_dw || nofp->nof_d_r_dw ||
1372 nofp->nof_d_rw || nofp->nof_d_w || nofp->nof_d_r) {
1373 nfs4_claim_delegated_state_for_open_file(nofp, 0);
1374 }
1375 /* need to remove the open before sending the downgrade */
1376 nfs_open_file_remove_open(nofp, accessMode, denyMode);
1377 error = nfs4_open_downgrade_rpc(np, nofp, ctx);
1378 if (error) { /* Hmm.. that didn't work. Add the open back in. */
1379 nfs_open_file_add_open(nofp, accessMode, denyMode, delegated);
1380 }
1381 }
1382 }
1383 #endif
1384 v3close:
1385 if (error) {
1386 NP(np, "nfs_close: error %d, %d", error, kauth_cred_getuid(nofp->nof_owner->noo_cred));
1387 return error;
1388 }
1389
1390 if (!downgrade) {
1391 nfs_open_file_remove_open(nofp, accessMode, denyMode);
1392 }
1393
1394 if (closed) {
1395 lck_mtx_lock(&nofp->nof_lock);
1396 if (nofp->nof_r || nofp->nof_d_r || nofp->nof_w || nofp->nof_d_w || nofp->nof_d_rw ||
1397 (nofp->nof_rw && !((nofp->nof_flags & NFS_OPEN_FILE_CREATE) && !nofp->nof_creator && (nofp->nof_rw == 1))) ||
1398 nofp->nof_r_dw || nofp->nof_d_r_dw || nofp->nof_w_dw || nofp->nof_d_w_dw ||
1399 nofp->nof_rw_dw || nofp->nof_d_rw_dw || nofp->nof_r_drw || nofp->nof_d_r_drw ||
1400 nofp->nof_w_drw || nofp->nof_d_w_drw || nofp->nof_rw_drw || nofp->nof_d_rw_drw) {
1401 NP(np, "nfs_close: unexpected count: %u.%u %u.%u %u.%u dw %u.%u %u.%u %u.%u drw %u.%u %u.%u %u.%u flags 0x%x, %d",
1402 nofp->nof_r, nofp->nof_d_r, nofp->nof_w, nofp->nof_d_w,
1403 nofp->nof_rw, nofp->nof_d_rw, nofp->nof_r_dw, nofp->nof_d_r_dw,
1404 nofp->nof_w_dw, nofp->nof_d_w_dw, nofp->nof_rw_dw, nofp->nof_d_rw_dw,
1405 nofp->nof_r_drw, nofp->nof_d_r_drw, nofp->nof_w_drw, nofp->nof_d_w_drw,
1406 nofp->nof_rw_drw, nofp->nof_d_rw_drw, nofp->nof_flags,
1407 kauth_cred_getuid(nofp->nof_owner->noo_cred));
1408 }
1409 /* clear out all open info, just to be safe */
1410 nofp->nof_access = nofp->nof_deny = 0;
1411 nofp->nof_mmap_access = nofp->nof_mmap_deny = 0;
1412 nofp->nof_r = nofp->nof_d_r = 0;
1413 nofp->nof_w = nofp->nof_d_w = 0;
1414 nofp->nof_rw = nofp->nof_d_rw = 0;
1415 nofp->nof_r_dw = nofp->nof_d_r_dw = 0;
1416 nofp->nof_w_dw = nofp->nof_d_w_dw = 0;
1417 nofp->nof_rw_dw = nofp->nof_d_rw_dw = 0;
1418 nofp->nof_r_drw = nofp->nof_d_r_drw = 0;
1419 nofp->nof_w_drw = nofp->nof_d_w_drw = 0;
1420 nofp->nof_rw_drw = nofp->nof_d_rw_drw = 0;
1421 nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
1422 lck_mtx_unlock(&nofp->nof_lock);
1423 /* XXX we may potentially want to clean up idle/unused open file structures */
1424 }
1425 if (nofp->nof_flags & NFS_OPEN_FILE_LOST) {
1426 error = EIO;
1427 NP(np, "nfs_close: LOST%s, %d", !nofp->nof_opencnt ? " (last)" : "",
1428 kauth_cred_getuid(nofp->nof_owner->noo_cred));
1429 }
1430
1431 return error;
1432 }
1433
1434
1435 int
1436 nfs3_getattr_rpc(
1437 nfsnode_t np,
1438 mount_t mp,
1439 u_char *fhp,
1440 size_t fhsize,
1441 int flags,
1442 vfs_context_t ctx,
1443 struct nfs_vattr *nvap,
1444 u_int64_t *xidp)
1445 {
1446 struct nfsmount *nmp = mp ? VFSTONFS(mp) : NFSTONMP(np);
1447 int error = 0, status = 0, nfsvers, rpcflags = 0;
1448 struct nfsm_chain nmreq, nmrep;
1449
1450 if (nfs_mount_gone(nmp)) {
1451 return ENXIO;
1452 }
1453 nfsvers = nmp->nm_vers;
1454
1455 if (flags & NGA_MONITOR) { /* vnode monitor requests should be soft */
1456 rpcflags = R_RECOVER;
1457 }
1458
1459 if (flags & NGA_SOFT) { /* Return ETIMEDOUT if server not responding */
1460 rpcflags |= R_SOFT;
1461 }
1462
1463 nfsm_chain_null(&nmreq);
1464 nfsm_chain_null(&nmrep);
1465
1466 nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(nfsvers));
1467 if (nfsvers != NFS_VER2) {
1468 nfsm_chain_add_32(error, &nmreq, fhsize);
1469 }
1470 nfsm_chain_add_opaque(error, &nmreq, fhp, fhsize);
1471 nfsm_chain_build_done(error, &nmreq);
1472 nfsmout_if(error);
1473 error = nfs_request2(np, mp, &nmreq, NFSPROC_GETATTR,
1474 vfs_context_thread(ctx), vfs_context_ucred(ctx),
1475 NULL, rpcflags, &nmrep, xidp, &status);
1476 if (!error) {
1477 error = status;
1478 }
1479 nfsmout_if(error);
1480 error = nfs_parsefattr(nmp, &nmrep, nfsvers, nvap);
1481 nfsmout:
1482 nfsm_chain_cleanup(&nmreq);
1483 nfsm_chain_cleanup(&nmrep);
1484 return error;
1485 }
1486
1487 /*
1488 * nfs_refresh_fh will attempt to update the file handle for the node.
1489 *
1490 * It only does this for symbolic links and regular files that are not currently opened.
1491 *
1492 * On Success returns 0 and the nodes file handle is updated, or ESTALE on failure.
1493 */
1494 int
1495 nfs_refresh_fh(nfsnode_t np, vfs_context_t ctx)
1496 {
1497 vnode_t dvp, vp = NFSTOV(np);
1498 nfsnode_t dnp;
1499 const char *v_name = vnode_getname(vp);
1500 char *name;
1501 int namelen, refreshed;
1502 uint32_t fhsize;
1503 int error, wanted = 0;
1504 uint8_t *fhp;
1505 struct timespec ts = {.tv_sec = 2, .tv_nsec = 0};
1506
1507 NFS_VNOP_DBG("vnode is %d\n", vnode_vtype(vp));
1508
1509 dvp = vnode_parent(vp);
1510 if ((vnode_vtype(vp) != VREG && vnode_vtype(vp) != VLNK) ||
1511 v_name == NULL || *v_name == '\0' || dvp == NULL) {
1512 if (v_name != NULL) {
1513 vnode_putname(v_name);
1514 }
1515 return ESTALE;
1516 }
1517 dnp = VTONFS(dvp);
1518
1519 namelen = NFS_STRLEN_INT(v_name);
1520 MALLOC(name, char *, namelen + 1, M_TEMP, M_WAITOK);
1521 if (name == NULL) {
1522 vnode_putname(v_name);
1523 return ESTALE;
1524 }
1525 bcopy(v_name, name, namelen + 1);
1526 NFS_VNOP_DBG("Trying to refresh %s : %s\n", v_name, name);
1527 vnode_putname(v_name);
1528
1529 /* Allocate the maximum size file handle */
1530 MALLOC(fhp, uint8_t *, NFS4_FHSIZE, M_FHANDLE, M_WAITOK);
1531 if (fhp == NULL) {
1532 FREE(name, M_TEMP);
1533 return ESTALE;
1534 }
1535
1536 if ((error = nfs_node_lock(np))) {
1537 FREE(name, M_TEMP);
1538 FREE(fhp, M_FHANDLE);
1539 return ESTALE;
1540 }
1541
1542 fhsize = np->n_fhsize;
1543 bcopy(np->n_fhp, fhp, fhsize);
1544 while (ISSET(np->n_flag, NREFRESH)) {
1545 SET(np->n_flag, NREFRESHWANT);
1546 NFS_VNOP_DBG("Waiting for refresh of %s\n", name);
1547 msleep(np, &np->n_lock, PZERO - 1, "nfsrefreshwant", &ts);
1548 if ((error = nfs_sigintr(NFSTONMP(np), NULL, vfs_context_thread(ctx), 0))) {
1549 break;
1550 }
1551 }
1552 refreshed = error ? 0 : !NFS_CMPFH(np, fhp, fhsize);
1553 SET(np->n_flag, NREFRESH);
1554 nfs_node_unlock(np);
1555
1556 NFS_VNOP_DBG("error = %d, refreshed = %d\n", error, refreshed);
1557 if (error || refreshed) {
1558 goto nfsmout;
1559 }
1560
1561 /* Check that there are no open references for this file */
1562 lck_mtx_lock(&np->n_openlock);
1563 if (np->n_openrefcnt || !TAILQ_EMPTY(&np->n_opens) || !TAILQ_EMPTY(&np->n_lock_owners)) {
1564 int cnt = 0;
1565 struct nfs_open_file *ofp;
1566
1567 TAILQ_FOREACH(ofp, &np->n_opens, nof_link) {
1568 cnt += ofp->nof_opencnt;
1569 }
1570 if (cnt) {
1571 lck_mtx_unlock(&np->n_openlock);
1572 NFS_VNOP_DBG("Can not refresh file handle for %s with open state\n", name);
1573 NFS_VNOP_DBG("\topenrefcnt = %d, opens = %d lock_owners = %d\n",
1574 np->n_openrefcnt, cnt, !TAILQ_EMPTY(&np->n_lock_owners));
1575 error = ESTALE;
1576 goto nfsmout;
1577 }
1578 }
1579 lck_mtx_unlock(&np->n_openlock);
1580 /*
1581 * Since the FH is currently stale we should not be able to
1582 * establish any open state until the FH is refreshed.
1583 */
1584
1585 error = nfs_node_lock(np);
1586 nfsmout_if(error);
1587 /*
1588 * Symlinks should never need invalidations and are holding
1589 * the one and only nfsbuf in an uncached acquired state
1590 * trying to do a readlink. So we will hang if we invalidate
1591 * in that case. Only in in the VREG case do we need to
1592 * invalidate.
1593 */
1594 if (vnode_vtype(vp) == VREG) {
1595 np->n_flag &= ~NNEEDINVALIDATE;
1596 nfs_node_unlock(np);
1597 error = nfs_vinvalbuf(vp, V_IGNORE_WRITEERR, ctx, 1);
1598 if (error) {
1599 NFS_VNOP_DBG("nfs_vinvalbuf returned %d\n", error);
1600 }
1601 nfsmout_if(error);
1602 } else {
1603 nfs_node_unlock(np);
1604 }
1605
1606 NFS_VNOP_DBG("Looking up %s\n", name);
1607 error = nfs_lookitup(dnp, name, namelen, ctx, &np);
1608 if (error) {
1609 NFS_VNOP_DBG("nfs_lookitup returned %d\n", error);
1610 }
1611
1612 nfsmout:
1613 nfs_node_lock_force(np);
1614 wanted = ISSET(np->n_flag, NREFRESHWANT);
1615 CLR(np->n_flag, NREFRESH | NREFRESHWANT);
1616 nfs_node_unlock(np);
1617 if (wanted) {
1618 wakeup(np);
1619 }
1620
1621 if (error == 0) {
1622 NFS_VNOP_DBG("%s refreshed file handle\n", name);
1623 }
1624
1625 FREE(name, M_TEMP);
1626 FREE(fhp, M_FHANDLE);
1627
1628 return error ? ESTALE : 0;
1629 }
1630
1631 int
1632 nfs_getattr(nfsnode_t np, struct nfs_vattr *nvap, vfs_context_t ctx, int flags)
1633 {
1634 int error;
1635
1636 retry:
1637 error = nfs_getattr_internal(np, nvap, ctx, flags);
1638 if (error == ESTALE) {
1639 error = nfs_refresh_fh(np, ctx);
1640 if (!error) {
1641 goto retry;
1642 }
1643 }
1644 return error;
1645 }
1646
1647 int
1648 nfs_getattr_internal(nfsnode_t np, struct nfs_vattr *nvap, vfs_context_t ctx, int flags)
1649 {
1650 struct nfsmount *nmp;
1651 int error = 0, nfsvers, inprogset = 0, wanted = 0, avoidfloods = 0;
1652 struct nfs_vattr *nvattr = NULL;
1653 struct timespec ts = { .tv_sec = 2, .tv_nsec = 0 };
1654 u_int64_t xid = 0;
1655
1656 FSDBG_TOP(513, np->n_size, np, np->n_vattr.nva_size, np->n_flag);
1657
1658 nmp = NFSTONMP(np);
1659
1660 if (nfs_mount_gone(nmp)) {
1661 return ENXIO;
1662 }
1663 nfsvers = nmp->nm_vers;
1664
1665 if (!nvap) {
1666 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
1667 nvap = nvattr;
1668 }
1669 NVATTR_INIT(nvap);
1670
1671 /* Update local times for special files. */
1672 if (np->n_flag & (NACC | NUPD)) {
1673 nfs_node_lock_force(np);
1674 np->n_flag |= NCHG;
1675 nfs_node_unlock(np);
1676 }
1677 /* Update size, if necessary */
1678 if (ISSET(np->n_flag, NUPDATESIZE)) {
1679 nfs_data_update_size(np, 0);
1680 }
1681
1682 error = nfs_node_lock(np);
1683 nfsmout_if(error);
1684 if (!(flags & (NGA_UNCACHED | NGA_MONITOR)) || ((nfsvers >= NFS_VER4) && (np->n_openflags & N_DELEG_MASK))) {
1685 /*
1686 * Use the cache or wait for any getattr in progress if:
1687 * - it's a cached request, or
1688 * - we have a delegation, or
1689 * - the server isn't responding
1690 */
1691 while (1) {
1692 error = nfs_getattrcache(np, nvap, flags);
1693 if (!error || (error != ENOENT)) {
1694 nfs_node_unlock(np);
1695 goto nfsmout;
1696 }
1697 error = 0;
1698 if (!ISSET(np->n_flag, NGETATTRINPROG)) {
1699 break;
1700 }
1701 if (flags & NGA_MONITOR) {
1702 /* no need to wait if a request is pending */
1703 error = EINPROGRESS;
1704 nfs_node_unlock(np);
1705 goto nfsmout;
1706 }
1707 SET(np->n_flag, NGETATTRWANT);
1708 msleep(np, &np->n_lock, PZERO - 1, "nfsgetattrwant", &ts);
1709 if ((error = nfs_sigintr(NFSTONMP(np), NULL, vfs_context_thread(ctx), 0))) {
1710 nfs_node_unlock(np);
1711 goto nfsmout;
1712 }
1713 }
1714 SET(np->n_flag, NGETATTRINPROG);
1715 inprogset = 1;
1716 } else if (!ISSET(np->n_flag, NGETATTRINPROG)) {
1717 SET(np->n_flag, NGETATTRINPROG);
1718 inprogset = 1;
1719 } else if (flags & NGA_MONITOR) {
1720 /* no need to make a request if one is pending */
1721 error = EINPROGRESS;
1722 }
1723 nfs_node_unlock(np);
1724
1725 nmp = NFSTONMP(np);
1726 if (nfs_mount_gone(nmp)) {
1727 error = ENXIO;
1728 }
1729 if (error) {
1730 goto nfsmout;
1731 }
1732
1733 /*
1734 * Return cached attributes if they are valid,
1735 * if the server doesn't respond, and this is
1736 * some softened up style of mount.
1737 */
1738 if (NATTRVALID(np) && nfs_use_cache(nmp)) {
1739 flags |= NGA_SOFT;
1740 }
1741
1742 /*
1743 * We might want to try to get both the attributes and access info by
1744 * making an ACCESS call and seeing if it returns updated attributes.
1745 * But don't bother if we aren't caching access info or if the
1746 * attributes returned wouldn't be cached.
1747 */
1748 if (!(flags & NGA_ACL) && (nfsvers != NFS_VER2) && nfs_access_for_getattr && (nfs_access_cache_timeout > 0)) {
1749 if (nfs_attrcachetimeout(np) > 0) {
1750 /* OSAddAtomic(1, &nfsstats.accesscache_misses); */
1751 u_int32_t access = NFS_ACCESS_ALL;
1752 int rpcflags = 0;
1753
1754 /* Return cached attrs if server doesn't respond */
1755 if (flags & NGA_SOFT) {
1756 rpcflags |= R_SOFT;
1757 }
1758
1759 error = nmp->nm_funcs->nf_access_rpc(np, &access, rpcflags, ctx);
1760
1761 if (error == ETIMEDOUT) {
1762 goto returncached;
1763 }
1764
1765 if (error) {
1766 goto nfsmout;
1767 }
1768 nfs_node_lock_force(np);
1769 error = nfs_getattrcache(np, nvap, flags);
1770 nfs_node_unlock(np);
1771 if (!error || (error != ENOENT)) {
1772 goto nfsmout;
1773 }
1774 /* Well, that didn't work... just do a getattr... */
1775 error = 0;
1776 }
1777 }
1778
1779 avoidfloods = 0;
1780
1781 tryagain:
1782 error = nmp->nm_funcs->nf_getattr_rpc(np, NULL, np->n_fhp, np->n_fhsize, flags, ctx, nvap, &xid);
1783 if (!error) {
1784 nfs_node_lock_force(np);
1785 error = nfs_loadattrcache(np, nvap, &xid, 0);
1786 nfs_node_unlock(np);
1787 }
1788
1789 /*
1790 * If the server didn't respond, return cached attributes.
1791 */
1792 returncached:
1793 if ((flags & NGA_SOFT) && (error == ETIMEDOUT)) {
1794 nfs_node_lock_force(np);
1795 error = nfs_getattrcache(np, nvap, flags);
1796 if (!error || (error != ENOENT)) {
1797 nfs_node_unlock(np);
1798 goto nfsmout;
1799 }
1800 nfs_node_unlock(np);
1801 }
1802 nfsmout_if(error);
1803
1804 if (!xid) { /* out-of-order rpc - attributes were dropped */
1805 FSDBG(513, -1, np, np->n_xid >> 32, np->n_xid);
1806 if (avoidfloods++ < 20) {
1807 goto tryagain;
1808 }
1809 /* avoidfloods>1 is bizarre. at 20 pull the plug */
1810 /* just return the last attributes we got */
1811 }
1812 nfsmout:
1813 nfs_node_lock_force(np);
1814 if (inprogset) {
1815 wanted = ISSET(np->n_flag, NGETATTRWANT);
1816 CLR(np->n_flag, (NGETATTRINPROG | NGETATTRWANT));
1817 }
1818 if (!error) {
1819 /* check if the node changed on us */
1820 vnode_t vp = NFSTOV(np);
1821 enum vtype vtype = vnode_vtype(vp);
1822 if ((vtype == VDIR) && NFS_CHANGED_NC(nfsvers, np, nvap)) {
1823 FSDBG(513, -1, np, 0, np);
1824 np->n_flag &= ~NNEGNCENTRIES;
1825 cache_purge(vp);
1826 np->n_ncgen++;
1827 NFS_CHANGED_UPDATE_NC(nfsvers, np, nvap);
1828 NFS_VNOP_DBG("Purge directory 0x%llx\n",
1829 (uint64_t)VM_KERNEL_ADDRPERM(vp));
1830 }
1831 if (NFS_CHANGED(nfsvers, np, nvap)) {
1832 FSDBG(513, -1, np, -1, np);
1833 if (vtype == VDIR) {
1834 NFS_VNOP_DBG("Invalidate directory 0x%llx\n",
1835 (uint64_t)VM_KERNEL_ADDRPERM(vp));
1836 nfs_invaldir(np);
1837 }
1838 nfs_node_unlock(np);
1839 if (wanted) {
1840 wakeup(np);
1841 }
1842 error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
1843 FSDBG(513, -1, np, -2, error);
1844 if (!error) {
1845 nfs_node_lock_force(np);
1846 NFS_CHANGED_UPDATE(nfsvers, np, nvap);
1847 nfs_node_unlock(np);
1848 }
1849 } else {
1850 nfs_node_unlock(np);
1851 if (wanted) {
1852 wakeup(np);
1853 }
1854 }
1855 } else {
1856 nfs_node_unlock(np);
1857 if (wanted) {
1858 wakeup(np);
1859 }
1860 }
1861
1862 if (nvattr != NULL) {
1863 NVATTR_CLEANUP(nvap);
1864 FREE(nvattr, M_TEMP);
1865 } else if (!(flags & NGA_ACL)) {
1866 /* make sure we don't return an ACL if it wasn't asked for */
1867 NFS_BITMAP_CLR(nvap->nva_bitmap, NFS_FATTR_ACL);
1868 if (nvap->nva_acl) {
1869 kauth_acl_free(nvap->nva_acl);
1870 nvap->nva_acl = NULL;
1871 }
1872 }
1873 FSDBG_BOT(513, np->n_size, error, np->n_vattr.nva_size, np->n_flag);
1874 return error;
1875 }
1876
1877
1878 /*
1879 * NFS getattr call from vfs.
1880 */
1881
1882 /*
1883 * The attributes we support over the wire.
1884 * We also get fsid but the vfs layer gets it out of the mount
1885 * structure after this calling us so there's no need to return it,
1886 * and Finder expects to call getattrlist just looking for the FSID
1887 * with out hanging on a non responsive server.
1888 */
1889 #define NFS3_SUPPORTED_VATTRS \
1890 (VNODE_ATTR_va_rdev | \
1891 VNODE_ATTR_va_nlink | \
1892 VNODE_ATTR_va_data_size | \
1893 VNODE_ATTR_va_data_alloc | \
1894 VNODE_ATTR_va_uid | \
1895 VNODE_ATTR_va_gid | \
1896 VNODE_ATTR_va_mode | \
1897 VNODE_ATTR_va_modify_time | \
1898 VNODE_ATTR_va_change_time | \
1899 VNODE_ATTR_va_access_time | \
1900 VNODE_ATTR_va_fileid | \
1901 VNODE_ATTR_va_type)
1902
1903
1904 int
1905 nfs3_vnop_getattr(
1906 struct vnop_getattr_args /* {
1907 * struct vnodeop_desc *a_desc;
1908 * vnode_t a_vp;
1909 * struct vnode_attr *a_vap;
1910 * vfs_context_t a_context;
1911 * } */*ap)
1912 {
1913 int error;
1914 nfsnode_t np;
1915 uint64_t supported_attrs;
1916 struct nfs_vattr *nva;
1917 struct vnode_attr *vap = ap->a_vap;
1918 struct nfsmount *nmp;
1919 dev_t rdev;
1920
1921 nmp = VTONMP(ap->a_vp);
1922
1923 /*
1924 * Lets don't go over the wire if we don't support any of the attributes.
1925 * Just fall through at the VFS layer and let it cons up what it needs.
1926 */
1927 /* Return the io size no matter what, since we don't go over the wire for this */
1928 VATTR_RETURN(vap, va_iosize, nfs_iosize);
1929
1930 supported_attrs = NFS3_SUPPORTED_VATTRS;
1931
1932 if ((vap->va_active & supported_attrs) == 0) {
1933 return 0;
1934 }
1935
1936 if (VATTR_IS_ACTIVE(ap->a_vap, va_name)) {
1937 NFS_VNOP_DBG("Getting attrs for 0x%llx, vname is %s\n",
1938 (uint64_t)VM_KERNEL_ADDRPERM(ap->a_vp),
1939 ap->a_vp->v_name ? ap->a_vp->v_name : "empty");
1940 }
1941
1942 /*
1943 * We should not go over the wire if only fileid was requested and has ever been populated.
1944 */
1945 if ((vap->va_active & supported_attrs) == VNODE_ATTR_va_fileid) {
1946 np = VTONFS(ap->a_vp);
1947 if (np->n_attrstamp) {
1948 VATTR_RETURN(vap, va_fileid, np->n_vattr.nva_fileid);
1949 return 0;
1950 }
1951 }
1952
1953 MALLOC(nva, struct nfs_vattr *, sizeof(*nva), M_TEMP, M_WAITOK);
1954 error = nfs_getattr(VTONFS(ap->a_vp), nva, ap->a_context, NGA_CACHED);
1955 if (error) {
1956 goto out;
1957 }
1958
1959 /* copy nva to *a_vap */
1960 VATTR_RETURN(vap, va_type, nva->nva_type);
1961 VATTR_RETURN(vap, va_mode, nva->nva_mode);
1962 rdev = makedev(nva->nva_rawdev.specdata1, nva->nva_rawdev.specdata2);
1963 VATTR_RETURN(vap, va_rdev, rdev);
1964 VATTR_RETURN(vap, va_uid, nva->nva_uid);
1965 VATTR_RETURN(vap, va_gid, nva->nva_gid);
1966 VATTR_RETURN(vap, va_nlink, nva->nva_nlink);
1967 VATTR_RETURN(vap, va_fileid, nva->nva_fileid);
1968 VATTR_RETURN(vap, va_data_size, nva->nva_size);
1969 VATTR_RETURN(vap, va_data_alloc, nva->nva_bytes);
1970 vap->va_access_time.tv_sec = nva->nva_timesec[NFSTIME_ACCESS];
1971 vap->va_access_time.tv_nsec = nva->nva_timensec[NFSTIME_ACCESS];
1972 VATTR_SET_SUPPORTED(vap, va_access_time);
1973 vap->va_modify_time.tv_sec = nva->nva_timesec[NFSTIME_MODIFY];
1974 vap->va_modify_time.tv_nsec = nva->nva_timensec[NFSTIME_MODIFY];
1975 VATTR_SET_SUPPORTED(vap, va_modify_time);
1976 vap->va_change_time.tv_sec = nva->nva_timesec[NFSTIME_CHANGE];
1977 vap->va_change_time.tv_nsec = nva->nva_timensec[NFSTIME_CHANGE];
1978 VATTR_SET_SUPPORTED(vap, va_change_time);
1979
1980
1981 // VATTR_RETURN(vap, va_encoding, 0xffff /* kTextEncodingUnknown */);
1982 out:
1983 FREE(nva, M_TEMP);
1984 return error;
1985 }
1986
1987 /*
1988 * NFS setattr call.
1989 */
1990 int
1991 nfs_vnop_setattr(
1992 struct vnop_setattr_args /* {
1993 * struct vnodeop_desc *a_desc;
1994 * vnode_t a_vp;
1995 * struct vnode_attr *a_vap;
1996 * vfs_context_t a_context;
1997 * } */*ap)
1998 {
1999 vfs_context_t ctx = ap->a_context;
2000 vnode_t vp = ap->a_vp;
2001 nfsnode_t np = VTONFS(vp);
2002 struct nfsmount *nmp;
2003 struct vnode_attr *vap = ap->a_vap;
2004 int error = 0;
2005 int biosize, nfsvers, namedattrs;
2006 u_quad_t origsize, vapsize;
2007 struct nfs_dulookup *dul;
2008 nfsnode_t dnp = NULL;
2009 int dul_in_progress = 0;
2010 vnode_t dvp = NULL;
2011 const char *vname = NULL;
2012 #if CONFIG_NFS4
2013 struct nfs_open_owner *noop = NULL;
2014 struct nfs_open_file *nofp = NULL;
2015 #endif
2016 nmp = VTONMP(vp);
2017 if (nfs_mount_gone(nmp)) {
2018 return ENXIO;
2019 }
2020 nfsvers = nmp->nm_vers;
2021 namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
2022 biosize = nmp->nm_biosize;
2023
2024 /* Disallow write attempts if the filesystem is mounted read-only. */
2025 if (vnode_vfsisrdonly(vp)) {
2026 return EROFS;
2027 }
2028
2029 origsize = np->n_size;
2030 if (VATTR_IS_ACTIVE(vap, va_data_size)) {
2031 switch (vnode_vtype(vp)) {
2032 case VDIR:
2033 return EISDIR;
2034 case VCHR:
2035 case VBLK:
2036 case VSOCK:
2037 case VFIFO:
2038 if (!VATTR_IS_ACTIVE(vap, va_modify_time) &&
2039 !VATTR_IS_ACTIVE(vap, va_access_time) &&
2040 !VATTR_IS_ACTIVE(vap, va_mode) &&
2041 !VATTR_IS_ACTIVE(vap, va_uid) &&
2042 !VATTR_IS_ACTIVE(vap, va_gid)) {
2043 return 0;
2044 }
2045 VATTR_CLEAR_ACTIVE(vap, va_data_size);
2046 break;
2047 default:
2048 /*
2049 * Disallow write attempts if the filesystem is
2050 * mounted read-only.
2051 */
2052 if (vnode_vfsisrdonly(vp)) {
2053 return EROFS;
2054 }
2055 FSDBG_TOP(512, np->n_size, vap->va_data_size,
2056 np->n_vattr.nva_size, np->n_flag);
2057 /* clear NNEEDINVALIDATE, if set */
2058 if ((error = nfs_node_lock(np))) {
2059 return error;
2060 }
2061 if (np->n_flag & NNEEDINVALIDATE) {
2062 np->n_flag &= ~NNEEDINVALIDATE;
2063 }
2064 nfs_node_unlock(np);
2065 /* flush everything */
2066 error = nfs_vinvalbuf(vp, (vap->va_data_size ? V_SAVE : 0), ctx, 1);
2067 if (error) {
2068 NP(np, "nfs_setattr: nfs_vinvalbuf %d", error);
2069 FSDBG_BOT(512, np->n_size, vap->va_data_size, np->n_vattr.nva_size, -1);
2070 return error;
2071 }
2072 #if CONFIG_NFS4
2073 if (nfsvers >= NFS_VER4) {
2074 /* setting file size requires having the file open for write access */
2075 if (np->n_flag & NREVOKE) {
2076 return EIO;
2077 }
2078 noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 1);
2079 if (!noop) {
2080 return ENOMEM;
2081 }
2082 restart:
2083 error = nfs_mount_state_in_use_start(nmp, vfs_context_thread(ctx));
2084 if (error) {
2085 return error;
2086 }
2087 if (np->n_flag & NREVOKE) {
2088 nfs_mount_state_in_use_end(nmp, 0);
2089 return EIO;
2090 }
2091 error = nfs_open_file_find(np, noop, &nofp, 0, 0, 1);
2092 if (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
2093 error = EIO;
2094 }
2095 if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
2096 error = nfs4_reopen(nofp, vfs_context_thread(ctx));
2097 nofp = NULL;
2098 if (!error) {
2099 nfs_mount_state_in_use_end(nmp, 0);
2100 goto restart;
2101 }
2102 }
2103 if (!error) {
2104 error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
2105 }
2106 if (error) {
2107 nfs_mount_state_in_use_end(nmp, 0);
2108 nfs_open_owner_rele(noop);
2109 return error;
2110 }
2111 if (!(nofp->nof_access & NFS_OPEN_SHARE_ACCESS_WRITE)) {
2112 /* we don't have the file open for write access, so open it */
2113 error = nfs4_open(np, nofp, NFS_OPEN_SHARE_ACCESS_WRITE, NFS_OPEN_SHARE_DENY_NONE, ctx);
2114 if (!error) {
2115 nofp->nof_flags |= NFS_OPEN_FILE_SETATTR;
2116 }
2117 if (nfs_mount_state_error_should_restart(error)) {
2118 nfs_open_file_clear_busy(nofp);
2119 nofp = NULL;
2120 nfs_mount_state_in_use_end(nmp, error);
2121 goto restart;
2122 }
2123 }
2124 }
2125 #endif
2126 nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
2127 if (np->n_size > vap->va_data_size) { /* shrinking? */
2128 daddr64_t obn, bn;
2129 int mustwrite;
2130 off_t neweofoff;
2131 struct nfsbuf *bp;
2132 nfsbufpgs pagemask;
2133
2134 obn = (np->n_size - 1) / biosize;
2135 bn = vap->va_data_size / biosize;
2136 for (; obn >= bn; obn--) {
2137 if (!nfs_buf_is_incore(np, obn)) {
2138 continue;
2139 }
2140 error = nfs_buf_get(np, obn, biosize, NULL, NBLK_READ, &bp);
2141 if (error) {
2142 continue;
2143 }
2144 if (obn != bn) {
2145 FSDBG(512, bp, bp->nb_flags, 0, obn);
2146 SET(bp->nb_flags, NB_INVAL);
2147 nfs_buf_release(bp, 1);
2148 continue;
2149 }
2150 mustwrite = 0;
2151 neweofoff = vap->va_data_size - NBOFF(bp);
2152 /* check for any dirty data before the new EOF */
2153 if ((bp->nb_dirtyend > 0) && (bp->nb_dirtyoff < neweofoff)) {
2154 /* clip dirty range to EOF */
2155 if (bp->nb_dirtyend > neweofoff) {
2156 bp->nb_dirtyend = neweofoff;
2157 if (bp->nb_dirtyoff >= bp->nb_dirtyend) {
2158 bp->nb_dirtyoff = bp->nb_dirtyend = 0;
2159 }
2160 }
2161 if ((bp->nb_dirtyend > 0) && (bp->nb_dirtyoff < neweofoff)) {
2162 mustwrite++;
2163 }
2164 }
2165 nfs_buf_pgs_get_page_mask(&pagemask, round_page_64(neweofoff) / PAGE_SIZE);
2166 nfs_buf_pgs_bit_and(&bp->nb_dirty, &pagemask, &bp->nb_dirty);
2167 if (nfs_buf_pgs_is_set(&bp->nb_dirty)) {
2168 mustwrite++;
2169 }
2170 if (!mustwrite) {
2171 FSDBG(512, bp, bp->nb_flags, 0, obn);
2172 SET(bp->nb_flags, NB_INVAL);
2173 nfs_buf_release(bp, 1);
2174 continue;
2175 }
2176 /* gotta write out dirty data before invalidating */
2177 /* (NB_STABLE indicates that data writes should be FILESYNC) */
2178 /* (NB_NOCACHE indicates buffer should be discarded) */
2179 CLR(bp->nb_flags, (NB_DONE | NB_ERROR | NB_INVAL | NB_ASYNC | NB_READ));
2180 SET(bp->nb_flags, NB_STABLE | NB_NOCACHE);
2181 if (!IS_VALID_CRED(bp->nb_wcred)) {
2182 kauth_cred_t cred = vfs_context_ucred(ctx);
2183 kauth_cred_ref(cred);
2184 bp->nb_wcred = cred;
2185 }
2186 error = nfs_buf_write(bp);
2187 // Note: bp has been released
2188 if (error) {
2189 FSDBG(512, bp, 0xd00dee, 0xbad, error);
2190 nfs_node_lock_force(np);
2191 np->n_error = error;
2192 np->n_flag |= NWRITEERR;
2193 /*
2194 * There was a write error and we need to
2195 * invalidate attrs and flush buffers in
2196 * order to sync up with the server.
2197 * (if this write was extending the file,
2198 * we may no longer know the correct size)
2199 */
2200 NATTRINVALIDATE(np);
2201 nfs_node_unlock(np);
2202 nfs_data_unlock(np);
2203 nfs_vinvalbuf(vp, V_SAVE | V_IGNORE_WRITEERR, ctx, 1);
2204 nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
2205 error = 0;
2206 }
2207 }
2208 }
2209 if (vap->va_data_size != np->n_size) {
2210 ubc_setsize(vp, (off_t)vap->va_data_size); /* XXX error? */
2211 }
2212 origsize = np->n_size;
2213 np->n_size = np->n_vattr.nva_size = vap->va_data_size;
2214 nfs_node_lock_force(np);
2215 CLR(np->n_flag, NUPDATESIZE);
2216 nfs_node_unlock(np);
2217 FSDBG(512, np, np->n_size, np->n_vattr.nva_size, 0xf00d0001);
2218 }
2219 } else if (VATTR_IS_ACTIVE(vap, va_modify_time) ||
2220 VATTR_IS_ACTIVE(vap, va_access_time) ||
2221 (vap->va_vaflags & VA_UTIMES_NULL)) {
2222 if ((error = nfs_node_lock(np))) {
2223 #if CONFIG_NFS4
2224 if (nfsvers >= NFS_VER4) {
2225 nfs_mount_state_in_use_end(nmp, 0);
2226 }
2227 #endif
2228 return error;
2229 }
2230 if ((np->n_flag & NMODIFIED) && (vnode_vtype(vp) == VREG)) {
2231 nfs_node_unlock(np);
2232 error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
2233 if (error == EINTR) {
2234 #if CONFIG_NFS4
2235 if (nfsvers >= NFS_VER4) {
2236 nfs_mount_state_in_use_end(nmp, 0);
2237 }
2238 #endif
2239 return error;
2240 }
2241 } else {
2242 nfs_node_unlock(np);
2243 }
2244 }
2245
2246 MALLOC(dul, struct nfs_dulookup *, sizeof(*dul), M_TEMP, M_WAITOK);
2247
2248 if ((VATTR_IS_ACTIVE(vap, va_mode) || VATTR_IS_ACTIVE(vap, va_uid) || VATTR_IS_ACTIVE(vap, va_gid) ||
2249 VATTR_IS_ACTIVE(vap, va_acl) || VATTR_IS_ACTIVE(vap, va_uuuid) || VATTR_IS_ACTIVE(vap, va_guuid)) &&
2250 !(error = nfs_node_lock(np))) {
2251 NACCESSINVALIDATE(np);
2252 nfs_node_unlock(np);
2253 if (!namedattrs) {
2254 dvp = vnode_getparent(vp);
2255 vname = vnode_getname(vp);
2256 dnp = (dvp && vname) ? VTONFS(dvp) : NULL;
2257 if (dnp) {
2258 if (nfs_node_set_busy(dnp, vfs_context_thread(ctx))) {
2259 vnode_put(dvp);
2260 vnode_putname(vname);
2261 } else {
2262 nfs_dulookup_init(dul, dnp, vname, NFS_STRLEN_INT(vname), ctx);
2263 nfs_dulookup_start(dul, dnp, ctx);
2264 dul_in_progress = 1;
2265 }
2266 } else {
2267 if (dvp) {
2268 vnode_put(dvp);
2269 }
2270 if (vname) {
2271 vnode_putname(vname);
2272 }
2273 }
2274 }
2275 }
2276
2277 if (!error) {
2278 error = nmp->nm_funcs->nf_setattr_rpc(np, vap, ctx);
2279 }
2280
2281 if (dul_in_progress) {
2282 nfs_dulookup_finish(dul, dnp, ctx);
2283 nfs_node_clear_busy(dnp);
2284 vnode_put(dvp);
2285 vnode_putname(vname);
2286 }
2287
2288 FREE(dul, M_TEMP);
2289 FSDBG_BOT(512, np->n_size, vap->va_data_size, np->n_vattr.nva_size, error);
2290 if (VATTR_IS_ACTIVE(vap, va_data_size)) {
2291 if (error && (origsize != np->n_size) &&
2292 ((nfsvers < NFS_VER4) || !nfs_mount_state_error_should_restart(error))) {
2293 /* make every effort to resync file size w/ server... */
2294 /* (don't bother if we'll be restarting the operation) */
2295 int err; /* preserve "error" for return */
2296 np->n_size = np->n_vattr.nva_size = origsize;
2297 nfs_node_lock_force(np);
2298 CLR(np->n_flag, NUPDATESIZE);
2299 nfs_node_unlock(np);
2300 FSDBG(512, np, np->n_size, np->n_vattr.nva_size, 0xf00d0002);
2301 ubc_setsize(vp, (off_t)np->n_size); /* XXX check error */
2302 vapsize = vap->va_data_size;
2303 vap->va_data_size = origsize;
2304 err = nmp->nm_funcs->nf_setattr_rpc(np, vap, ctx);
2305 if (err) {
2306 NP(np, "nfs_vnop_setattr: nfs%d_setattr_rpc %d %d", nfsvers, error, err);
2307 }
2308 vap->va_data_size = vapsize;
2309 }
2310 nfs_node_lock_force(np);
2311 /*
2312 * The size was just set. If the size is already marked for update, don't
2313 * trust the newsize (it may have been set while the setattr was in progress).
2314 * Clear the update flag and make sure we fetch new attributes so we are sure
2315 * we have the latest size.
2316 */
2317 if (ISSET(np->n_flag, NUPDATESIZE)) {
2318 CLR(np->n_flag, NUPDATESIZE);
2319 NATTRINVALIDATE(np);
2320 nfs_node_unlock(np);
2321 nfs_getattr(np, NULL, ctx, NGA_UNCACHED);
2322 } else {
2323 nfs_node_unlock(np);
2324 }
2325 nfs_data_unlock(np);
2326 #if CONFIG_NFS4
2327 if (nfsvers >= NFS_VER4) {
2328 if (nofp) {
2329 /* don't close our setattr open if we'll be restarting... */
2330 if (!nfs_mount_state_error_should_restart(error) &&
2331 (nofp->nof_flags & NFS_OPEN_FILE_SETATTR)) {
2332 int err = nfs_close(np, nofp, NFS_OPEN_SHARE_ACCESS_WRITE, NFS_OPEN_SHARE_DENY_NONE, ctx);
2333 if (err) {
2334 NP(np, "nfs_vnop_setattr: close error: %d", err);
2335 }
2336 nofp->nof_flags &= ~NFS_OPEN_FILE_SETATTR;
2337 }
2338 nfs_open_file_clear_busy(nofp);
2339 nofp = NULL;
2340 }
2341 if (nfs_mount_state_in_use_end(nmp, error)) {
2342 goto restart;
2343 }
2344 nfs_open_owner_rele(noop);
2345 }
2346 #endif
2347 }
2348 return error;
2349 }
2350
2351 /*
2352 * Do an NFS setattr RPC.
2353 */
2354 int
2355 nfs3_setattr_rpc(
2356 nfsnode_t np,
2357 struct vnode_attr *vap,
2358 vfs_context_t ctx)
2359 {
2360 struct nfsmount *nmp = NFSTONMP(np);
2361 int error = 0, lockerror = ENOENT, status = 0, wccpostattr = 0, nfsvers;
2362 u_int64_t xid, nextxid;
2363 struct nfsm_chain nmreq, nmrep;
2364
2365 if (nfs_mount_gone(nmp)) {
2366 return ENXIO;
2367 }
2368 nfsvers = nmp->nm_vers;
2369
2370 VATTR_SET_SUPPORTED(vap, va_mode);
2371 VATTR_SET_SUPPORTED(vap, va_uid);
2372 VATTR_SET_SUPPORTED(vap, va_gid);
2373 VATTR_SET_SUPPORTED(vap, va_data_size);
2374 VATTR_SET_SUPPORTED(vap, va_access_time);
2375 VATTR_SET_SUPPORTED(vap, va_modify_time);
2376
2377
2378 if (VATTR_IS_ACTIVE(vap, va_flags)
2379 ) {
2380 if (vap->va_flags) { /* we don't support setting flags */
2381 if (vap->va_active & ~VNODE_ATTR_va_flags) {
2382 return EINVAL; /* return EINVAL if other attributes also set */
2383 } else {
2384 return ENOTSUP; /* return ENOTSUP for chflags(2) */
2385 }
2386 }
2387 /* no flags set, so we'll just ignore it */
2388 if (!(vap->va_active & ~VNODE_ATTR_va_flags)) {
2389 return 0; /* no (other) attributes to set, so nothing to do */
2390 }
2391 }
2392
2393 nfsm_chain_null(&nmreq);
2394 nfsm_chain_null(&nmrep);
2395
2396 nfsm_chain_build_alloc_init(error, &nmreq,
2397 NFSX_FH(nfsvers) + NFSX_SATTR(nfsvers));
2398 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
2399 if (nfsvers == NFS_VER3) {
2400 if (VATTR_IS_ACTIVE(vap, va_mode)) {
2401 nfsm_chain_add_32(error, &nmreq, TRUE);
2402 nfsm_chain_add_32(error, &nmreq, vap->va_mode);
2403 } else {
2404 nfsm_chain_add_32(error, &nmreq, FALSE);
2405 }
2406 if (VATTR_IS_ACTIVE(vap, va_uid)) {
2407 nfsm_chain_add_32(error, &nmreq, TRUE);
2408 nfsm_chain_add_32(error, &nmreq, vap->va_uid);
2409 } else {
2410 nfsm_chain_add_32(error, &nmreq, FALSE);
2411 }
2412 if (VATTR_IS_ACTIVE(vap, va_gid)) {
2413 nfsm_chain_add_32(error, &nmreq, TRUE);
2414 nfsm_chain_add_32(error, &nmreq, vap->va_gid);
2415 } else {
2416 nfsm_chain_add_32(error, &nmreq, FALSE);
2417 }
2418 if (VATTR_IS_ACTIVE(vap, va_data_size)) {
2419 nfsm_chain_add_32(error, &nmreq, TRUE);
2420 nfsm_chain_add_64(error, &nmreq, vap->va_data_size);
2421 } else {
2422 nfsm_chain_add_32(error, &nmreq, FALSE);
2423 }
2424 if (vap->va_vaflags & VA_UTIMES_NULL) {
2425 nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_SERVER);
2426 nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_SERVER);
2427 } else {
2428 if (VATTR_IS_ACTIVE(vap, va_access_time)) {
2429 nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_CLIENT);
2430 nfsm_chain_add_32(error, &nmreq, vap->va_access_time.tv_sec);
2431 nfsm_chain_add_32(error, &nmreq, vap->va_access_time.tv_nsec);
2432 } else {
2433 nfsm_chain_add_32(error, &nmreq, NFS_TIME_DONT_CHANGE);
2434 }
2435 if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
2436 nfsm_chain_add_32(error, &nmreq, NFS_TIME_SET_TO_CLIENT);
2437 nfsm_chain_add_32(error, &nmreq, vap->va_modify_time.tv_sec);
2438 nfsm_chain_add_32(error, &nmreq, vap->va_modify_time.tv_nsec);
2439 } else {
2440 nfsm_chain_add_32(error, &nmreq, NFS_TIME_DONT_CHANGE);
2441 }
2442 }
2443 nfsm_chain_add_32(error, &nmreq, FALSE);
2444 } else {
2445 nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_mode) ?
2446 vtonfsv2_mode(vnode_vtype(NFSTOV(np)), vap->va_mode) : -1);
2447 nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_uid) ?
2448 vap->va_uid : (uint32_t)-1);
2449 nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_gid) ?
2450 vap->va_gid : (uint32_t)-1);
2451 nfsm_chain_add_32(error, &nmreq, VATTR_IS_ACTIVE(vap, va_data_size) ?
2452 vap->va_data_size : (uint32_t)-1);
2453 if (VATTR_IS_ACTIVE(vap, va_access_time)) {
2454 nfsm_chain_add_32(error, &nmreq, vap->va_access_time.tv_sec);
2455 nfsm_chain_add_32(error, &nmreq, (vap->va_access_time.tv_nsec != -1) ?
2456 ((uint32_t)vap->va_access_time.tv_nsec / 1000) : 0xffffffff);
2457 } else {
2458 nfsm_chain_add_32(error, &nmreq, -1);
2459 nfsm_chain_add_32(error, &nmreq, -1);
2460 }
2461 if (VATTR_IS_ACTIVE(vap, va_modify_time)) {
2462 nfsm_chain_add_32(error, &nmreq, vap->va_modify_time.tv_sec);
2463 nfsm_chain_add_32(error, &nmreq, (vap->va_modify_time.tv_nsec != -1) ?
2464 ((uint32_t)vap->va_modify_time.tv_nsec / 1000) : 0xffffffff);
2465 } else {
2466 nfsm_chain_add_32(error, &nmreq, -1);
2467 nfsm_chain_add_32(error, &nmreq, -1);
2468 }
2469 }
2470 nfsm_chain_build_done(error, &nmreq);
2471 nfsmout_if(error);
2472 error = nfs_request(np, NULL, &nmreq, NFSPROC_SETATTR, ctx, NULL, &nmrep, &xid, &status);
2473 if ((lockerror = nfs_node_lock(np))) {
2474 error = lockerror;
2475 }
2476 if (nfsvers == NFS_VER3) {
2477 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
2478 nfsm_chain_get_wcc_data(error, &nmrep, np, &premtime, &wccpostattr, &xid);
2479 nfsmout_if(error);
2480 /* if file hadn't changed, update cached mtime */
2481 if (nfstimespeccmp(&np->n_mtime, &premtime, ==)) {
2482 NFS_CHANGED_UPDATE(nfsvers, np, &np->n_vattr);
2483 }
2484 /* if directory hadn't changed, update namecache mtime */
2485 if ((vnode_vtype(NFSTOV(np)) == VDIR) &&
2486 nfstimespeccmp(&np->n_ncmtime, &premtime, ==)) {
2487 NFS_CHANGED_UPDATE_NC(nfsvers, np, &np->n_vattr);
2488 }
2489 if (!wccpostattr) {
2490 NATTRINVALIDATE(np);
2491 }
2492 error = status;
2493 } else {
2494 if (!error) {
2495 error = status;
2496 }
2497 nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid);
2498 }
2499 /*
2500 * We just changed the attributes and we want to make sure that we
2501 * see the latest attributes. Get the next XID. If it's not the
2502 * next XID after the SETATTR XID, then it's possible that another
2503 * RPC was in flight at the same time and it might put stale attributes
2504 * in the cache. In that case, we invalidate the attributes and set
2505 * the attribute cache XID to guarantee that newer attributes will
2506 * get loaded next.
2507 */
2508 nextxid = 0;
2509 nfs_get_xid(&nextxid);
2510 if (nextxid != (xid + 1)) {
2511 np->n_xid = nextxid;
2512 NATTRINVALIDATE(np);
2513 }
2514 nfsmout:
2515 if (!lockerror) {
2516 nfs_node_unlock(np);
2517 }
2518 nfsm_chain_cleanup(&nmreq);
2519 nfsm_chain_cleanup(&nmrep);
2520 return error;
2521 }
2522
2523 /*
2524 * NFS lookup call, one step at a time...
2525 * First look in cache
2526 * If not found, unlock the directory nfsnode and do the RPC
2527 */
2528 int
2529 nfs_vnop_lookup(
2530 struct vnop_lookup_args /* {
2531 * struct vnodeop_desc *a_desc;
2532 * vnode_t a_dvp;
2533 * vnode_t *a_vpp;
2534 * struct componentname *a_cnp;
2535 * vfs_context_t a_context;
2536 * } */*ap)
2537 {
2538 vfs_context_t ctx = ap->a_context;
2539 struct componentname *cnp = ap->a_cnp;
2540 vnode_t dvp = ap->a_dvp;
2541 vnode_t *vpp = ap->a_vpp;
2542 int flags = cnp->cn_flags;
2543 vnode_t newvp;
2544 nfsnode_t dnp, np;
2545 struct nfsmount *nmp;
2546 mount_t mp;
2547 int nfsvers, error, busyerror = ENOENT, isdot, isdotdot, negnamecache;
2548 u_int64_t xid = 0;
2549 struct nfs_vattr *nvattr;
2550 int ngflags, skipdu = 0;
2551 struct vnop_access_args naa;
2552 fhandle_t *fh;
2553 struct nfsreq *req;
2554
2555 *vpp = NULLVP;
2556
2557 dnp = VTONFS(dvp);
2558
2559 fh = zalloc(nfs_fhandle_zone);
2560 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
2561 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
2562 NVATTR_INIT(nvattr);
2563
2564 mp = vnode_mount(dvp);
2565 nmp = VFSTONFS(mp);
2566 if (nfs_mount_gone(nmp)) {
2567 error = ENXIO;
2568 goto error_return;
2569 }
2570 nfsvers = nmp->nm_vers;
2571 negnamecache = !NMFLAG(nmp, NONEGNAMECACHE);
2572
2573 if ((error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx)))) {
2574 goto error_return;
2575 }
2576 /* nfs_getattr() will check changed and purge caches */
2577 if ((error = nfs_getattr(dnp, NULL, ctx, NGA_CACHED))) {
2578 goto error_return;
2579 }
2580
2581 error = cache_lookup(dvp, vpp, cnp);
2582 switch (error) {
2583 case ENOENT:
2584 /* negative cache entry */
2585 goto error_return;
2586 case 0:
2587 /* cache miss */
2588 if ((nfsvers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) {
2589 /* if rdirplus, try dir buf cache lookup */
2590 error = nfs_dir_buf_cache_lookup(dnp, &np, cnp, ctx, 0, &skipdu);
2591 if (!error && np) {
2592 /* dir buf cache hit */
2593 *vpp = NFSTOV(np);
2594 error = -1;
2595 } else if (skipdu) {
2596 /* Skip lookup for du files */
2597 error = ENOENT;
2598 goto error_return;
2599 }
2600 }
2601 if (error != -1) { /* cache miss */
2602 break;
2603 }
2604 OS_FALLTHROUGH;
2605 case -1:
2606 /* cache hit, not really an error */
2607 OSAddAtomic64(1, &nfsstats.lookupcache_hits);
2608
2609 nfs_node_clear_busy(dnp);
2610 busyerror = ENOENT;
2611
2612 /* check for directory access */
2613 naa.a_desc = &vnop_access_desc;
2614 naa.a_vp = dvp;
2615 naa.a_action = KAUTH_VNODE_SEARCH;
2616 naa.a_context = ctx;
2617
2618 /* compute actual success/failure based on accessibility */
2619 error = nfs_vnop_access(&naa);
2620 OS_FALLTHROUGH;
2621 default:
2622 /* unexpected error from cache_lookup */
2623 goto error_return;
2624 }
2625
2626 /* skip lookup, if we know who we are: "." or ".." */
2627 isdot = isdotdot = 0;
2628 if (cnp->cn_nameptr[0] == '.') {
2629 if (cnp->cn_namelen == 1) {
2630 isdot = 1;
2631 }
2632 if ((cnp->cn_namelen == 2) && (cnp->cn_nameptr[1] == '.')) {
2633 isdotdot = 1;
2634 }
2635 }
2636 if (isdotdot || isdot) {
2637 fh->fh_len = 0;
2638 goto found;
2639 }
2640 #if CONFIG_NFS4
2641 if ((nfsvers >= NFS_VER4) && (dnp->n_vattr.nva_flags & NFS_FFLAG_TRIGGER)) {
2642 /* we should never be looking things up in a trigger directory, return nothing */
2643 error = ENOENT;
2644 goto error_return;
2645 }
2646 #endif
2647
2648 /* do we know this name is too long? */
2649 nmp = VTONMP(dvp);
2650 if (nfs_mount_gone(nmp)) {
2651 error = ENXIO;
2652 goto error_return;
2653 }
2654 if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME) &&
2655 (cnp->cn_namelen > nmp->nm_fsattr.nfsa_maxname)) {
2656 error = ENAMETOOLONG;
2657 goto error_return;
2658 }
2659
2660 error = 0;
2661 newvp = NULLVP;
2662
2663 OSAddAtomic64(1, &nfsstats.lookupcache_misses);
2664
2665 error = nmp->nm_funcs->nf_lookup_rpc_async(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &req);
2666 nfsmout_if(error);
2667 error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, req, &xid, fh, nvattr);
2668 nfsmout_if(error);
2669
2670 /* is the file handle the same as this directory's file handle? */
2671 isdot = NFS_CMPFH(dnp, fh->fh_data, fh->fh_len);
2672
2673 found:
2674 if (flags & ISLASTCN) {
2675 switch (cnp->cn_nameiop) {
2676 case DELETE:
2677 cnp->cn_flags &= ~MAKEENTRY;
2678 break;
2679 case RENAME:
2680 cnp->cn_flags &= ~MAKEENTRY;
2681 if (isdot) {
2682 error = EISDIR;
2683 goto error_return;
2684 }
2685 break;
2686 }
2687 }
2688
2689 if (isdotdot) {
2690 newvp = vnode_getparent(dvp);
2691 if (!newvp) {
2692 error = ENOENT;
2693 goto error_return;
2694 }
2695 } else if (isdot) {
2696 error = vnode_get(dvp);
2697 if (error) {
2698 goto error_return;
2699 }
2700 newvp = dvp;
2701 nfs_node_lock_force(dnp);
2702 if (fh->fh_len && (dnp->n_xid <= xid)) {
2703 nfs_loadattrcache(dnp, nvattr, &xid, 0);
2704 }
2705 nfs_node_unlock(dnp);
2706 } else {
2707 ngflags = (cnp->cn_flags & MAKEENTRY) ? NG_MAKEENTRY : 0;
2708 error = nfs_nget(mp, dnp, cnp, fh->fh_data, fh->fh_len, nvattr, &xid, req->r_auth, ngflags, &np);
2709 if (error) {
2710 goto error_return;
2711 }
2712 newvp = NFSTOV(np);
2713 nfs_node_unlock(np);
2714 }
2715 *vpp = newvp;
2716
2717 nfsmout:
2718 if (error) {
2719 if (((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME)) &&
2720 (flags & ISLASTCN) && (error == ENOENT)) {
2721 if (vnode_mount(dvp) && vnode_vfsisrdonly(dvp)) {
2722 error = EROFS;
2723 } else {
2724 error = EJUSTRETURN;
2725 }
2726 }
2727 }
2728 if ((error == ENOENT) && (cnp->cn_flags & MAKEENTRY) &&
2729 (cnp->cn_nameiop != CREATE) && negnamecache) {
2730 /* add a negative entry in the name cache */
2731 nfs_node_lock_force(dnp);
2732 cache_enter(dvp, NULL, cnp);
2733 dnp->n_flag |= NNEGNCENTRIES;
2734 nfs_node_unlock(dnp);
2735 }
2736 error_return:
2737 NVATTR_CLEANUP(nvattr);
2738 NFS_ZFREE(nfs_fhandle_zone, fh);
2739 NFS_ZFREE(nfs_req_zone, req);
2740 FREE(nvattr, M_TEMP);
2741 if (!busyerror) {
2742 nfs_node_clear_busy(dnp);
2743 }
2744 if (error && *vpp) {
2745 vnode_put(*vpp);
2746 *vpp = NULLVP;
2747 }
2748 return error;
2749 }
2750
2751 int nfs_readlink_nocache = DEFAULT_READLINK_NOCACHE;
2752
2753 /*
2754 * NFS readlink call
2755 */
2756 int
2757 nfs_vnop_readlink(
2758 struct vnop_readlink_args /* {
2759 * struct vnodeop_desc *a_desc;
2760 * vnode_t a_vp;
2761 * struct uio *a_uio;
2762 * vfs_context_t a_context;
2763 * } */*ap)
2764 {
2765 vfs_context_t ctx = ap->a_context;
2766 nfsnode_t np = VTONFS(ap->a_vp);
2767 struct nfsmount *nmp;
2768 int error = 0, nfsvers;
2769 size_t buflen;
2770 uio_t uio = ap->a_uio;
2771 struct nfsbuf *bp = NULL;
2772 struct timespec ts = { .tv_sec = 0, .tv_nsec = 0 };
2773 long timeo = 0;
2774
2775 if (vnode_vtype(ap->a_vp) != VLNK) {
2776 return EPERM;
2777 }
2778
2779 if (uio_resid(uio) == 0) {
2780 return 0;
2781 }
2782 if (uio_offset(uio) < 0) {
2783 return EINVAL;
2784 }
2785
2786 nmp = VTONMP(ap->a_vp);
2787 if (nfs_mount_gone(nmp)) {
2788 return ENXIO;
2789 }
2790 nfsvers = nmp->nm_vers;
2791
2792
2793 /* nfs_getattr() will check changed and purge caches */
2794 if ((error = nfs_getattr(np, NULL, ctx, nfs_readlink_nocache ? NGA_UNCACHED : NGA_CACHED))) {
2795 FSDBG(531, np, 0xd1e0001, 0, error);
2796 return error;
2797 }
2798
2799 if (nfs_readlink_nocache) {
2800 timeo = nfs_attrcachetimeout(np);
2801 nanouptime(&ts);
2802 }
2803
2804 retry:
2805 OSAddAtomic64(1, &nfsstats.biocache_readlinks);
2806 error = nfs_buf_get(np, 0, NFS_MAXPATHLEN, vfs_context_thread(ctx), NBLK_META, &bp);
2807 if (error) {
2808 FSDBG(531, np, 0xd1e0002, 0, error);
2809 return error;
2810 }
2811
2812 if (nfs_readlink_nocache) {
2813 NFS_VNOP_DBG("timeo = %ld ts.tv_sec = %ld need refresh = %d cached = %d\n", timeo, ts.tv_sec,
2814 (np->n_rltim.tv_sec + timeo) < ts.tv_sec || nfs_readlink_nocache > 1,
2815 ISSET(bp->nb_flags, NB_CACHE) == NB_CACHE);
2816 /* n_rltim is synchronized by the associated nfs buf */
2817 if (ISSET(bp->nb_flags, NB_CACHE) && ((nfs_readlink_nocache > 1) || ((np->n_rltim.tv_sec + timeo) < ts.tv_sec))) {
2818 SET(bp->nb_flags, NB_INVAL);
2819 nfs_buf_release(bp, 0);
2820 goto retry;
2821 }
2822 }
2823 if (!ISSET(bp->nb_flags, NB_CACHE)) {
2824 readagain:
2825 OSAddAtomic64(1, &nfsstats.readlink_bios);
2826 buflen = bp->nb_bufsize;
2827 error = nmp->nm_funcs->nf_readlink_rpc(np, bp->nb_data, &buflen, ctx);
2828 if (error) {
2829 if (error == ESTALE) {
2830 NFS_VNOP_DBG("Stale FH from readlink rpc\n");
2831 error = nfs_refresh_fh(np, ctx);
2832 if (error == 0) {
2833 goto readagain;
2834 }
2835 }
2836 SET(bp->nb_flags, NB_ERROR);
2837 bp->nb_error = error;
2838 NFS_VNOP_DBG("readlink failed %d\n", error);
2839 } else {
2840 bp->nb_validoff = 0;
2841 bp->nb_validend = buflen;
2842 np->n_rltim = ts;
2843 NFS_VNOP_DBG("readlink of %.*s\n", (int32_t)bp->nb_validend, (char *)bp->nb_data);
2844 }
2845 } else {
2846 NFS_VNOP_DBG("got cached link of %.*s\n", (int32_t)bp->nb_validend, (char *)bp->nb_data);
2847 }
2848
2849 if (!error && (bp->nb_validend > 0)) {
2850 int validend32 = bp->nb_validend > INT_MAX ? INT_MAX : (int)bp->nb_validend;
2851 error = uiomove(bp->nb_data, validend32, uio);
2852 if (!error && bp->nb_validend > validend32) {
2853 error = uiomove(bp->nb_data + validend32, (int)(bp->nb_validend - validend32), uio);
2854 }
2855 }
2856 FSDBG(531, np, bp->nb_validend, 0, error);
2857 nfs_buf_release(bp, 1);
2858 return error;
2859 }
2860
2861 /*
2862 * Do a readlink RPC.
2863 */
2864 int
2865 nfs3_readlink_rpc(nfsnode_t np, char *buf, size_t *buflenp, vfs_context_t ctx)
2866 {
2867 struct nfsmount *nmp;
2868 int error = 0, lockerror = ENOENT, nfsvers, status;
2869 size_t len;
2870 u_int64_t xid;
2871 struct nfsm_chain nmreq, nmrep;
2872
2873 nmp = NFSTONMP(np);
2874 if (nfs_mount_gone(nmp)) {
2875 return ENXIO;
2876 }
2877 nfsvers = nmp->nm_vers;
2878 nfsm_chain_null(&nmreq);
2879 nfsm_chain_null(&nmrep);
2880
2881 nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(nfsvers));
2882 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
2883 nfsm_chain_build_done(error, &nmreq);
2884 nfsmout_if(error);
2885 error = nfs_request(np, NULL, &nmreq, NFSPROC_READLINK, ctx, NULL, &nmrep, &xid, &status);
2886 if ((lockerror = nfs_node_lock(np))) {
2887 error = lockerror;
2888 }
2889 if (nfsvers == NFS_VER3) {
2890 nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
2891 }
2892 if (!error) {
2893 error = status;
2894 }
2895 nfsm_chain_get_32(error, &nmrep, len);
2896 nfsmout_if(error);
2897 if ((nfsvers == NFS_VER2) && (len > *buflenp)) {
2898 error = EBADRPC;
2899 goto nfsmout;
2900 }
2901 if (len >= *buflenp) {
2902 if (np->n_size && (np->n_size < *buflenp)) {
2903 len = (size_t)np->n_size;
2904 } else {
2905 len = *buflenp - 1;
2906 }
2907 }
2908 nfsm_chain_get_opaque(error, &nmrep, len, buf);
2909 if (!error) {
2910 *buflenp = len;
2911 }
2912 nfsmout:
2913 if (!lockerror) {
2914 nfs_node_unlock(np);
2915 }
2916 nfsm_chain_cleanup(&nmreq);
2917 nfsm_chain_cleanup(&nmrep);
2918 return error;
2919 }
2920
2921 /*
2922 * NFS read RPC call
2923 * Ditto above
2924 */
2925 int
2926 nfs_read_rpc(nfsnode_t np, uio_t uio, vfs_context_t ctx)
2927 {
2928 struct nfsmount *nmp;
2929 int error = 0, nfsvers, eof = 0;
2930 size_t nmrsize, len, retlen;
2931 user_ssize_t tsiz;
2932 off_t txoffset;
2933 struct nfsreq *req;
2934 #if CONFIG_NFS4
2935 uint32_t stategenid = 0, restart = 0;
2936 #endif
2937 FSDBG_TOP(536, np, uio_offset(uio), uio_resid(uio), 0);
2938 nmp = NFSTONMP(np);
2939 if (nfs_mount_gone(nmp)) {
2940 return ENXIO;
2941 }
2942 nfsvers = nmp->nm_vers;
2943 nmrsize = nmp->nm_rsize;
2944
2945 txoffset = uio_offset(uio);
2946 tsiz = uio_resid(uio);
2947 if ((nfsvers == NFS_VER2) && ((uint64_t)(txoffset + tsiz) > 0xffffffffULL)) {
2948 FSDBG_BOT(536, np, uio_offset(uio), uio_resid(uio), EFBIG);
2949 return EFBIG;
2950 }
2951
2952 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
2953 while (tsiz > 0) {
2954 len = retlen = (tsiz > (user_ssize_t)nmrsize) ? nmrsize : (size_t)tsiz;
2955 FSDBG(536, np, txoffset, len, 0);
2956 if (np->n_flag & NREVOKE) {
2957 error = EIO;
2958 break;
2959 }
2960 #if CONFIG_NFS4
2961 if (nmp->nm_vers >= NFS_VER4) {
2962 stategenid = nmp->nm_stategenid;
2963 }
2964 #endif
2965 error = nmp->nm_funcs->nf_read_rpc_async(np, txoffset, len,
2966 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, &req);
2967 if (!error) {
2968 error = nmp->nm_funcs->nf_read_rpc_async_finish(np, req, uio, &retlen, &eof);
2969 }
2970 #if CONFIG_NFS4
2971 if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error) &&
2972 (++restart <= nfs_mount_state_max_restarts(nmp))) { /* guard against no progress */
2973 lck_mtx_lock(&nmp->nm_lock);
2974 if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
2975 NP(np, "nfs_read_rpc: error %d, initiating recovery", error);
2976 nfs_need_recover(nmp, error);
2977 }
2978 lck_mtx_unlock(&nmp->nm_lock);
2979 if (np->n_flag & NREVOKE) {
2980 error = EIO;
2981 } else {
2982 if (error == NFSERR_GRACE) {
2983 tsleep(&nmp->nm_state, (PZERO - 1), "nfsgrace", 2 * hz);
2984 }
2985 if (!(error = nfs_mount_state_wait_for_recovery(nmp))) {
2986 continue;
2987 }
2988 }
2989 }
2990 #endif
2991 if (error) {
2992 break;
2993 }
2994 txoffset += retlen;
2995 tsiz -= retlen;
2996 if (nfsvers != NFS_VER2) {
2997 if (eof || (retlen == 0)) {
2998 tsiz = 0;
2999 }
3000 } else if (retlen < len) {
3001 tsiz = 0;
3002 }
3003 }
3004
3005 NFS_ZFREE(nfs_req_zone, req);
3006 FSDBG_BOT(536, np, eof, uio_resid(uio), error);
3007 return error;
3008 }
3009
3010 int
3011 nfs3_read_rpc_async(
3012 nfsnode_t np,
3013 off_t offset,
3014 size_t len,
3015 thread_t thd,
3016 kauth_cred_t cred,
3017 struct nfsreq_cbinfo *cb,
3018 struct nfsreq **reqp)
3019 {
3020 struct nfsmount *nmp;
3021 int error = 0, nfsvers;
3022 struct nfsm_chain nmreq;
3023
3024 nmp = NFSTONMP(np);
3025 if (nfs_mount_gone(nmp)) {
3026 return ENXIO;
3027 }
3028 nfsvers = nmp->nm_vers;
3029
3030 nfsm_chain_null(&nmreq);
3031 nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(nfsvers) + 3 * NFSX_UNSIGNED);
3032 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
3033 if (nfsvers == NFS_VER3) {
3034 nfsm_chain_add_64(error, &nmreq, offset);
3035 nfsm_chain_add_32(error, &nmreq, len);
3036 } else {
3037 nfsm_chain_add_32(error, &nmreq, offset);
3038 nfsm_chain_add_32(error, &nmreq, len);
3039 nfsm_chain_add_32(error, &nmreq, 0);
3040 }
3041 nfsm_chain_build_done(error, &nmreq);
3042 nfsmout_if(error);
3043 error = nfs_request_async(np, NULL, &nmreq, NFSPROC_READ, thd, cred, NULL, 0, cb, reqp);
3044 nfsmout:
3045 nfsm_chain_cleanup(&nmreq);
3046 return error;
3047 }
3048
3049 int
3050 nfs3_read_rpc_async_finish(
3051 nfsnode_t np,
3052 struct nfsreq *req,
3053 uio_t uio,
3054 size_t *lenp,
3055 int *eofp)
3056 {
3057 int error = 0, lockerror, nfsvers, status = 0, eof = 0;
3058 uint32_t retlen = 0;
3059 uint64_t xid;
3060 struct nfsmount *nmp;
3061 struct nfsm_chain nmrep;
3062
3063 nmp = NFSTONMP(np);
3064 if (nfs_mount_gone(nmp)) {
3065 nfs_request_async_cancel(req);
3066 return ENXIO;
3067 }
3068 nfsvers = nmp->nm_vers;
3069
3070 nfsm_chain_null(&nmrep);
3071
3072 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
3073 if (error == EINPROGRESS) { /* async request restarted */
3074 return error;
3075 }
3076
3077 if ((lockerror = nfs_node_lock(np))) {
3078 error = lockerror;
3079 }
3080 if (nfsvers == NFS_VER3) {
3081 nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
3082 }
3083 if (!error) {
3084 error = status;
3085 }
3086 if (nfsvers == NFS_VER3) {
3087 nfsm_chain_adv(error, &nmrep, NFSX_UNSIGNED);
3088 nfsm_chain_get_32(error, &nmrep, eof);
3089 } else {
3090 nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid);
3091 }
3092 if (!lockerror) {
3093 nfs_node_unlock(np);
3094 }
3095 nfsm_chain_get_32(error, &nmrep, retlen);
3096 if ((nfsvers == NFS_VER2) && (retlen > *lenp)) {
3097 error = EBADRPC;
3098 }
3099 nfsmout_if(error);
3100 error = nfsm_chain_get_uio(&nmrep, MIN(retlen, *lenp), uio);
3101 if (eofp) {
3102 if (nfsvers == NFS_VER3) {
3103 if (!eof && !retlen) {
3104 eof = 1;
3105 }
3106 } else if (retlen < *lenp) {
3107 eof = 1;
3108 }
3109 *eofp = eof;
3110 }
3111 *lenp = MIN(retlen, *lenp);
3112 nfsmout:
3113 nfsm_chain_cleanup(&nmrep);
3114 return error;
3115 }
3116
3117 /*
3118 * NFS write call
3119 */
3120 int
3121 nfs_vnop_write(
3122 struct vnop_write_args /* {
3123 * struct vnodeop_desc *a_desc;
3124 * vnode_t a_vp;
3125 * struct uio *a_uio;
3126 * int a_ioflag;
3127 * vfs_context_t a_context;
3128 * } */*ap)
3129 {
3130 vfs_context_t ctx = ap->a_context;
3131 uio_t uio = ap->a_uio;
3132 vnode_t vp = ap->a_vp;
3133 nfsnode_t np = VTONFS(vp);
3134 int ioflag = ap->a_ioflag;
3135 struct nfsbuf *bp;
3136 struct nfsmount *nmp = VTONMP(vp);
3137 daddr64_t lbn;
3138 uint32_t biosize;
3139 int error = 0;
3140 off_t n, on;
3141 int n32;
3142 off_t boff, start, end;
3143 uio_t auio;
3144 char auio_buf[UIO_SIZEOF(1)];
3145 thread_t thd;
3146 kauth_cred_t cred;
3147
3148 FSDBG_TOP(515, np, uio_offset(uio), uio_resid(uio), ioflag);
3149
3150 if (vnode_vtype(vp) != VREG) {
3151 FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), EIO);
3152 return EIO;
3153 }
3154
3155 thd = vfs_context_thread(ctx);
3156 cred = vfs_context_ucred(ctx);
3157
3158 nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
3159
3160 if ((error = nfs_node_lock(np))) {
3161 nfs_data_unlock(np);
3162 FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), error);
3163 return error;
3164 }
3165 np->n_wrbusy++;
3166
3167 if (np->n_flag & NWRITEERR) {
3168 error = np->n_error;
3169 np->n_flag &= ~NWRITEERR;
3170 }
3171 if (np->n_flag & NNEEDINVALIDATE) {
3172 np->n_flag &= ~NNEEDINVALIDATE;
3173 nfs_node_unlock(np);
3174 nfs_data_unlock(np);
3175 nfs_vinvalbuf(vp, V_SAVE | V_IGNORE_WRITEERR, ctx, 1);
3176 nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
3177 } else {
3178 nfs_node_unlock(np);
3179 }
3180 if (error) {
3181 goto out;
3182 }
3183
3184 biosize = nmp->nm_biosize;
3185
3186 if (ioflag & (IO_APPEND | IO_SYNC)) {
3187 nfs_node_lock_force(np);
3188 if (np->n_flag & NMODIFIED) {
3189 NATTRINVALIDATE(np);
3190 nfs_node_unlock(np);
3191 nfs_data_unlock(np);
3192 error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
3193 nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
3194 if (error) {
3195 FSDBG(515, np, uio_offset(uio), 0x10bad01, error);
3196 goto out;
3197 }
3198 } else {
3199 nfs_node_unlock(np);
3200 }
3201 if (ioflag & IO_APPEND) {
3202 nfs_data_unlock(np);
3203 /* nfs_getattr() will check changed and purge caches */
3204 error = nfs_getattr(np, NULL, ctx, NGA_UNCACHED);
3205 /* we'll be extending the file, so take the data lock exclusive */
3206 nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
3207 if (error) {
3208 FSDBG(515, np, uio_offset(uio), 0x10bad02, error);
3209 goto out;
3210 }
3211 uio_setoffset(uio, np->n_size);
3212 }
3213 }
3214 if (uio_offset(uio) < 0) {
3215 error = EINVAL;
3216 FSDBG_BOT(515, np, uio_offset(uio), 0xbad0ff, error);
3217 goto out;
3218 }
3219 if (uio_resid(uio) == 0) {
3220 goto out;
3221 }
3222
3223 if (((uio_offset(uio) + uio_resid(uio)) > (off_t)np->n_size) && !(ioflag & IO_APPEND)) {
3224 /*
3225 * It looks like we'll be extending the file, so take the data lock exclusive.
3226 */
3227 nfs_data_unlock(np);
3228 nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
3229
3230 /*
3231 * Also, if the write begins after the previous EOF buffer, make sure to zero
3232 * and validate the new bytes in that buffer.
3233 */
3234 struct nfsbuf *eofbp = NULL;
3235 daddr64_t eofbn = np->n_size / biosize;
3236 uint32_t eofoff = np->n_size % biosize;
3237 lbn = uio_offset(uio) / biosize;
3238
3239 if (eofoff && (eofbn < lbn)) {
3240 if ((error = nfs_buf_get(np, eofbn, biosize, thd, NBLK_WRITE | NBLK_ONLYVALID, &eofbp))) {
3241 goto out;
3242 }
3243 np->n_size += (biosize - eofoff);
3244 nfs_node_lock_force(np);
3245 CLR(np->n_flag, NUPDATESIZE);
3246 np->n_flag |= NMODIFIED;
3247 nfs_node_unlock(np);
3248 FSDBG(516, np, np->n_size, np->n_vattr.nva_size, 0xf00d0001);
3249 ubc_setsize(vp, (off_t)np->n_size); /* XXX errors */
3250 if (eofbp) {
3251 /*
3252 * For the old last page, don't zero bytes if there
3253 * are invalid bytes in that page (i.e. the page isn't
3254 * currently valid).
3255 * For pages after the old last page, zero them and
3256 * mark them as valid.
3257 */
3258 char *d;
3259 int i;
3260 if (ioflag & IO_NOCACHE) {
3261 SET(eofbp->nb_flags, NB_NOCACHE);
3262 }
3263 NFS_BUF_MAP(eofbp);
3264 FSDBG(516, eofbp, eofoff, biosize - eofoff, 0xe0fff01e);
3265 d = eofbp->nb_data;
3266 i = eofoff / PAGE_SIZE;
3267 while (eofoff < biosize) {
3268 int poff = eofoff & PAGE_MASK;
3269 if (!poff || NBPGVALID(eofbp, i)) {
3270 bzero(d + eofoff, PAGE_SIZE - poff);
3271 NBPGVALID_SET(eofbp, i);
3272 }
3273 eofoff += PAGE_SIZE - poff;
3274 i++;
3275 }
3276 nfs_buf_release(eofbp, 1);
3277 }
3278 }
3279 }
3280
3281 do {
3282 OSAddAtomic64(1, &nfsstats.biocache_writes);
3283 lbn = uio_offset(uio) / biosize;
3284 on = uio_offset(uio) % biosize;
3285 n = biosize - on;
3286 if (uio_resid(uio) < n) {
3287 n = uio_resid(uio);
3288 }
3289 again:
3290 /*
3291 * Get a cache block for writing. The range to be written is
3292 * (off..off+n) within the block. We ensure that the block
3293 * either has no dirty region or that the given range is
3294 * contiguous with the existing dirty region.
3295 */
3296 error = nfs_buf_get(np, lbn, biosize, thd, NBLK_WRITE, &bp);
3297 if (error) {
3298 goto out;
3299 }
3300 /* map the block because we know we're going to write to it */
3301 NFS_BUF_MAP(bp);
3302
3303 if (ioflag & IO_NOCACHE) {
3304 SET(bp->nb_flags, NB_NOCACHE);
3305 }
3306
3307 if (!IS_VALID_CRED(bp->nb_wcred)) {
3308 kauth_cred_ref(cred);
3309 bp->nb_wcred = cred;
3310 }
3311
3312 /*
3313 * If there's already a dirty range AND dirty pages in this block we
3314 * need to send a commit AND write the dirty pages before continuing.
3315 *
3316 * If there's already a dirty range OR dirty pages in this block
3317 * and the new write range is not contiguous with the existing range,
3318 * then force the buffer to be written out now.
3319 * (We used to just extend the dirty range to cover the valid,
3320 * but unwritten, data in between also. But writing ranges
3321 * of data that weren't actually written by an application
3322 * risks overwriting some other client's data with stale data
3323 * that's just masquerading as new written data.)
3324 */
3325 if (bp->nb_dirtyend > 0) {
3326 if (on > bp->nb_dirtyend || (on + n) < bp->nb_dirtyoff || nfs_buf_pgs_is_set(&bp->nb_dirty)) {
3327 FSDBG(515, np, uio_offset(uio), bp, 0xd15c001);
3328 /* write/commit buffer "synchronously" */
3329 /* (NB_STABLE indicates that data writes should be FILESYNC) */
3330 CLR(bp->nb_flags, (NB_DONE | NB_ERROR | NB_INVAL));
3331 SET(bp->nb_flags, (NB_ASYNC | NB_STABLE));
3332 error = nfs_buf_write(bp);
3333 if (error) {
3334 goto out;
3335 }
3336 goto again;
3337 }
3338 } else if (nfs_buf_pgs_is_set(&bp->nb_dirty)) {
3339 off_t firstpg = 0, lastpg = 0;
3340 nfsbufpgs pagemask, pagemaskand;
3341 /* calculate write range pagemask */
3342 if (n > 0) {
3343 firstpg = on / PAGE_SIZE;
3344 lastpg = (on + n - 1) / PAGE_SIZE;
3345 nfs_buf_pgs_set_pages_between(&pagemask, firstpg, lastpg + 1);
3346 } else {
3347 NBPGS_ERASE(&pagemask);
3348 }
3349 /* check if there are dirty pages outside the write range */
3350 nfs_buf_pgs_bit_not(&pagemask);
3351 nfs_buf_pgs_bit_and(&bp->nb_dirty, &pagemask, &pagemaskand);
3352 if (nfs_buf_pgs_is_set(&pagemaskand)) {
3353 FSDBG(515, np, uio_offset(uio), bp, 0xd15c002);
3354 /* write/commit buffer "synchronously" */
3355 /* (NB_STABLE indicates that data writes should be FILESYNC) */
3356 CLR(bp->nb_flags, (NB_DONE | NB_ERROR | NB_INVAL));
3357 SET(bp->nb_flags, (NB_ASYNC | NB_STABLE));
3358 error = nfs_buf_write(bp);
3359 if (error) {
3360 goto out;
3361 }
3362 goto again;
3363 }
3364 /* if the first or last pages are already dirty */
3365 /* make sure that the dirty range encompasses those pages */
3366 if (NBPGDIRTY(bp, firstpg) || NBPGDIRTY(bp, lastpg)) {
3367 FSDBG(515, np, uio_offset(uio), bp, 0xd15c003);
3368 bp->nb_dirtyoff = MIN(on, firstpg * PAGE_SIZE);
3369 if (NBPGDIRTY(bp, lastpg)) {
3370 bp->nb_dirtyend = (lastpg + 1) * PAGE_SIZE;
3371 /* clip to EOF */
3372 if (NBOFF(bp) + bp->nb_dirtyend > (off_t)np->n_size) {
3373 bp->nb_dirtyend = np->n_size - NBOFF(bp);
3374 if (bp->nb_dirtyoff >= bp->nb_dirtyend) {
3375 bp->nb_dirtyoff = bp->nb_dirtyend = 0;
3376 }
3377 }
3378 } else {
3379 bp->nb_dirtyend = on + n;
3380 }
3381 }
3382 }
3383
3384 /*
3385 * Are we extending the size of the file with this write?
3386 * If so, update file size now that we have the block.
3387 * If there was a partial buf at the old eof, validate
3388 * and zero the new bytes.
3389 */
3390 if ((uio_offset(uio) + n) > (off_t)np->n_size) {
3391 daddr64_t eofbn = np->n_size / biosize;
3392 int neweofoff = (uio_offset(uio) + n) % biosize;
3393
3394 FSDBG(515, 0xb1ffa000, uio_offset(uio) + n, eofoff, neweofoff);
3395
3396 /* if we're extending within the same last block */
3397 /* and the block is flagged as being cached... */
3398 if ((lbn == eofbn) && ISSET(bp->nb_flags, NB_CACHE)) {
3399 /* ...check that all pages in buffer are valid */
3400 int endpg = ((neweofoff ? neweofoff : biosize) - 1) / PAGE_SIZE;
3401 nfsbufpgs pagemask, pagemaskand;
3402 /* pagemask only has to extend to last page being written to */
3403 nfs_buf_pgs_get_page_mask(&pagemask, endpg + 1);
3404 FSDBG(515, 0xb1ffa001, bp->nb_valid, pagemask, 0);
3405 nfs_buf_pgs_bit_and(&bp->nb_valid, &pagemask, &pagemaskand);
3406 if (!NBPGS_IS_EQUAL(&pagemaskand, &pagemask)) {
3407 /* zerofill any hole */
3408 if (on > bp->nb_validend) {
3409 for (off_t i = bp->nb_validend / PAGE_SIZE; i <= (on - 1) / PAGE_SIZE; i++) {
3410 NBPGVALID_SET(bp, i);
3411 }
3412 NFS_BUF_MAP(bp);
3413 FSDBG(516, bp, bp->nb_validend, on - bp->nb_validend, 0xf01e);
3414 NFS_BZERO((char *)bp->nb_data + bp->nb_validend, on - bp->nb_validend);
3415 }
3416 /* zerofill any trailing data in the last page */
3417 if (neweofoff) {
3418 NFS_BUF_MAP(bp);
3419 FSDBG(516, bp, neweofoff, PAGE_SIZE - (neweofoff & PAGE_MASK), 0xe0f);
3420 bzero((char *)bp->nb_data + neweofoff,
3421 PAGE_SIZE - (neweofoff & PAGE_MASK));
3422 }
3423 }
3424 }
3425 np->n_size = uio_offset(uio) + n;
3426 nfs_node_lock_force(np);
3427 CLR(np->n_flag, NUPDATESIZE);
3428 np->n_flag |= NMODIFIED;
3429 nfs_node_unlock(np);
3430 FSDBG(516, np, np->n_size, np->n_vattr.nva_size, 0xf00d0001);
3431 ubc_setsize(vp, (off_t)np->n_size); /* XXX errors */
3432 }
3433 /*
3434 * If dirtyend exceeds file size, chop it down. This should
3435 * not occur unless there is a race.
3436 */
3437 if (NBOFF(bp) + bp->nb_dirtyend > (off_t)np->n_size) {
3438 bp->nb_dirtyend = np->n_size - NBOFF(bp);
3439 if (bp->nb_dirtyoff >= bp->nb_dirtyend) {
3440 bp->nb_dirtyoff = bp->nb_dirtyend = 0;
3441 }
3442 }
3443 /*
3444 * UBC doesn't handle partial pages, so we need to make sure
3445 * that any pages left in the page cache are completely valid.
3446 *
3447 * Writes that are smaller than a block are delayed if they
3448 * don't extend to the end of the block.
3449 *
3450 * If the block isn't (completely) cached, we may need to read
3451 * in some parts of pages that aren't covered by the write.
3452 * If the write offset (on) isn't page aligned, we'll need to
3453 * read the start of the first page being written to. Likewise,
3454 * if the offset of the end of the write (on+n) isn't page aligned,
3455 * we'll need to read the end of the last page being written to.
3456 *
3457 * Notes:
3458 * We don't want to read anything we're just going to write over.
3459 * We don't want to read anything we're just going drop when the
3460 * I/O is complete (i.e. don't do reads for NOCACHE requests).
3461 * We don't want to issue multiple I/Os if we don't have to
3462 * (because they're synchronous rpcs).
3463 * We don't want to read anything we already have modified in the
3464 * page cache.
3465 */
3466 if (!ISSET(bp->nb_flags, NB_CACHE) && (n < biosize)) {
3467 off_t firstpgoff, lastpgoff, firstpg, lastpg, dirtypg;
3468 start = end = -1;
3469 firstpg = on / PAGE_SIZE;
3470 firstpgoff = on & PAGE_MASK;
3471 lastpg = (on + n - 1) / PAGE_SIZE;
3472 lastpgoff = (on + n) & PAGE_MASK;
3473 if (firstpgoff && !NBPGVALID(bp, firstpg)) {
3474 /* need to read start of first page */
3475 start = firstpg * PAGE_SIZE;
3476 end = start + firstpgoff;
3477 }
3478 if (lastpgoff && !NBPGVALID(bp, lastpg)) {
3479 /* need to read end of last page */
3480 if (start < 0) {
3481 start = (lastpg * PAGE_SIZE) + lastpgoff;
3482 }
3483 end = (lastpg + 1) * PAGE_SIZE;
3484 }
3485 if (ISSET(bp->nb_flags, NB_NOCACHE)) {
3486 /*
3487 * For nocache writes, if there is any partial page at the
3488 * start or end of the write range, then we do the write
3489 * synchronously to make sure that we can drop the data
3490 * from the cache as soon as the WRITE finishes. Normally,
3491 * we would do an unstable write and not drop the data until
3492 * it was committed. But doing that here would risk allowing
3493 * invalid data to be read from the cache between the WRITE
3494 * and the COMMIT.
3495 * (NB_STABLE indicates that data writes should be FILESYNC)
3496 */
3497 if (end > start) {
3498 SET(bp->nb_flags, NB_STABLE);
3499 }
3500 goto skipread;
3501 }
3502 if (end > start) {
3503 /* need to read the data in range: start...end-1 */
3504
3505 /* first, check for dirty pages in between */
3506 /* if there are, we'll have to do two reads because */
3507 /* we don't want to overwrite the dirty pages. */
3508 for (dirtypg = start / PAGE_SIZE; dirtypg <= (end - 1) / PAGE_SIZE; dirtypg++) {
3509 if (NBPGDIRTY(bp, dirtypg)) {
3510 break;
3511 }
3512 }
3513
3514 /* if start is at beginning of page, try */
3515 /* to get any preceeding pages as well. */
3516 if (!(start & PAGE_MASK)) {
3517 /* stop at next dirty/valid page or start of block */
3518 for (; start > 0; start -= PAGE_SIZE) {
3519 if (NBPGVALID(bp, ((start - 1) / PAGE_SIZE))) {
3520 break;
3521 }
3522 }
3523 }
3524
3525 NFS_BUF_MAP(bp);
3526 /* setup uio for read(s) */
3527 boff = NBOFF(bp);
3528 auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ,
3529 &auio_buf, sizeof(auio_buf));
3530
3531 if (dirtypg <= (end - 1) / PAGE_SIZE) {
3532 /* there's a dirty page in the way, so just do two reads */
3533 /* we'll read the preceding data here */
3534 uio_reset(auio, boff + start, UIO_SYSSPACE, UIO_READ);
3535 NFS_UIO_ADDIOV(auio, CAST_USER_ADDR_T(bp->nb_data + start), on - start);
3536 error = nfs_read_rpc(np, auio, ctx);
3537 if (error) {
3538 /* couldn't read the data, so treat buffer as synchronous NOCACHE */
3539 SET(bp->nb_flags, (NB_NOCACHE | NB_STABLE));
3540 goto skipread;
3541 }
3542 if (uio_resid(auio) > 0) {
3543 FSDBG(516, bp, (caddr_t)uio_curriovbase(auio) - bp->nb_data, uio_resid(auio), 0xd00dee01);
3544 bzero(CAST_DOWN(caddr_t, uio_curriovbase(auio)), uio_resid(auio));
3545 }
3546 if (!error) {
3547 /* update validoff/validend if necessary */
3548 if ((bp->nb_validoff < 0) || (bp->nb_validoff > start)) {
3549 bp->nb_validoff = start;
3550 }
3551 if ((bp->nb_validend < 0) || (bp->nb_validend < on)) {
3552 bp->nb_validend = on;
3553 }
3554 if ((off_t)np->n_size > boff + bp->nb_validend) {
3555 bp->nb_validend = MIN(np->n_size - (boff + start), biosize);
3556 }
3557 /* validate any pages before the write offset */
3558 for (; start < on / PAGE_SIZE; start += PAGE_SIZE) {
3559 NBPGVALID_SET(bp, start / PAGE_SIZE);
3560 }
3561 }
3562 /* adjust start to read any trailing data */
3563 start = on + n;
3564 }
3565
3566 /* if end is at end of page, try to */
3567 /* get any following pages as well. */
3568 if (!(end & PAGE_MASK)) {
3569 /* stop at next valid page or end of block */
3570 for (; end < biosize; end += PAGE_SIZE) {
3571 if (NBPGVALID(bp, end / PAGE_SIZE)) {
3572 break;
3573 }
3574 }
3575 }
3576
3577 if (((boff + start) >= (off_t)np->n_size) ||
3578 ((start >= on) && ((boff + on + n) >= (off_t)np->n_size))) {
3579 /*
3580 * Either this entire read is beyond the current EOF
3581 * or the range that we won't be modifying (on+n...end)
3582 * is all beyond the current EOF.
3583 * No need to make a trip across the network to
3584 * read nothing. So, just zero the buffer instead.
3585 */
3586 FSDBG(516, bp, start, end - start, 0xd00dee00);
3587 NFS_BZERO(bp->nb_data + start, end - start);
3588 error = 0;
3589 } else {
3590 /* now we'll read the (rest of the) data */
3591 uio_reset(auio, boff + start, UIO_SYSSPACE, UIO_READ);
3592 NFS_UIO_ADDIOV(auio, CAST_USER_ADDR_T(bp->nb_data + start), end - start);
3593 error = nfs_read_rpc(np, auio, ctx);
3594 if (error) {
3595 /* couldn't read the data, so treat buffer as synchronous NOCACHE */
3596 SET(bp->nb_flags, (NB_NOCACHE | NB_STABLE));
3597 goto skipread;
3598 }
3599 if (uio_resid(auio) > 0) {
3600 FSDBG(516, bp, (caddr_t)uio_curriovbase(auio) - bp->nb_data, uio_resid(auio), 0xd00dee02);
3601 bzero(CAST_DOWN(caddr_t, uio_curriovbase(auio)), uio_resid(auio));
3602 }
3603 }
3604 if (!error) {
3605 /* update validoff/validend if necessary */
3606 if ((bp->nb_validoff < 0) || (bp->nb_validoff > start)) {
3607 bp->nb_validoff = start;
3608 }
3609 if ((bp->nb_validend < 0) || (bp->nb_validend < end)) {
3610 bp->nb_validend = end;
3611 }
3612 if ((off_t)np->n_size > boff + bp->nb_validend) {
3613 bp->nb_validend = MIN(np->n_size - (boff + start), biosize);
3614 }
3615 /* validate any pages before the write offset's page */
3616 for (; start < (off_t)trunc_page_64(on); start += PAGE_SIZE) {
3617 NBPGVALID_SET(bp, start / PAGE_SIZE);
3618 }
3619 /* validate any pages after the range of pages being written to */
3620 for (; (end - 1) > (off_t)round_page_64(on + n - 1); end -= PAGE_SIZE) {
3621 NBPGVALID_SET(bp, (end - 1) / PAGE_SIZE);
3622 }
3623 }
3624 /* Note: pages being written to will be validated when written */
3625 }
3626 }
3627 skipread:
3628
3629 if (ISSET(bp->nb_flags, NB_ERROR)) {
3630 error = bp->nb_error;
3631 nfs_buf_release(bp, 1);
3632 goto out;
3633 }
3634
3635 nfs_node_lock_force(np);
3636 np->n_flag |= NMODIFIED;
3637 nfs_node_unlock(np);
3638
3639 NFS_BUF_MAP(bp);
3640 if (n < 0) {
3641 error = EINVAL;
3642 } else {
3643 n32 = n > INT_MAX ? INT_MAX : (int)n;
3644 error = uiomove(bp->nb_data + on, n32, uio);
3645 if (!error && n > n32) {
3646 error = uiomove(bp->nb_data + on + n32, (int)(n - n32), uio);
3647 }
3648 }
3649 if (error) {
3650 SET(bp->nb_flags, NB_ERROR);
3651 nfs_buf_release(bp, 1);
3652 goto out;
3653 }
3654
3655 /* validate any pages written to */
3656 start = on & ~PAGE_MASK;
3657 for (; start < on + n; start += PAGE_SIZE) {
3658 NBPGVALID_SET(bp, start / PAGE_SIZE);
3659 /*
3660 * This may seem a little weird, but we don't actually set the
3661 * dirty bits for writes. This is because we keep the dirty range
3662 * in the nb_dirtyoff/nb_dirtyend fields. Also, particularly for
3663 * delayed writes, when we give the pages back to the VM we don't
3664 * want to keep them marked dirty, because when we later write the
3665 * buffer we won't be able to tell which pages were written dirty
3666 * and which pages were mmapped and dirtied.
3667 */
3668 }
3669 if (bp->nb_dirtyend > 0) {
3670 bp->nb_dirtyoff = MIN(on, bp->nb_dirtyoff);
3671 bp->nb_dirtyend = MAX((on + n), bp->nb_dirtyend);
3672 } else {
3673 bp->nb_dirtyoff = on;
3674 bp->nb_dirtyend = on + n;
3675 }
3676 if (bp->nb_validend <= 0 || bp->nb_validend < bp->nb_dirtyoff ||
3677 bp->nb_validoff > bp->nb_dirtyend) {
3678 bp->nb_validoff = bp->nb_dirtyoff;
3679 bp->nb_validend = bp->nb_dirtyend;
3680 } else {
3681 bp->nb_validoff = MIN(bp->nb_validoff, bp->nb_dirtyoff);
3682 bp->nb_validend = MAX(bp->nb_validend, bp->nb_dirtyend);
3683 }
3684 if (!ISSET(bp->nb_flags, NB_CACHE)) {
3685 nfs_buf_normalize_valid_range(np, bp);
3686 }
3687
3688 /*
3689 * Since this block is being modified, it must be written
3690 * again and not just committed.
3691 */
3692 if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
3693 nfs_node_lock_force(np);
3694 if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
3695 np->n_needcommitcnt--;
3696 CHECK_NEEDCOMMITCNT(np);
3697 }
3698 CLR(bp->nb_flags, NB_NEEDCOMMIT);
3699 nfs_node_unlock(np);
3700 }
3701
3702 if (ioflag & IO_SYNC) {
3703 error = nfs_buf_write(bp);
3704 if (error) {
3705 goto out;
3706 }
3707 if (np->n_needcommitcnt >= NFS_A_LOT_OF_NEEDCOMMITS) {
3708 nfs_flushcommits(np, 1);
3709 }
3710 } else if (((n + on) == biosize) || (ioflag & IO_APPEND) ||
3711 (ioflag & IO_NOCACHE) || ISSET(bp->nb_flags, NB_NOCACHE)) {
3712 SET(bp->nb_flags, NB_ASYNC);
3713 error = nfs_buf_write(bp);
3714 if (error) {
3715 goto out;
3716 }
3717 } else {
3718 /* If the block wasn't already delayed: charge for the write */
3719 if (!ISSET(bp->nb_flags, NB_DELWRI)) {
3720 proc_t p = vfs_context_proc(ctx);
3721 if (p && p->p_stats) {
3722 OSIncrementAtomicLong(&p->p_stats->p_ru.ru_oublock);
3723 }
3724 }
3725 nfs_buf_write_delayed(bp);
3726 }
3727
3728 } while (uio_resid(uio) > 0 && n > 0);
3729
3730 out:
3731 nfs_node_lock_force(np);
3732 np->n_wrbusy--;
3733 if ((ioflag & IO_SYNC) && !np->n_wrbusy && !np->n_numoutput) {
3734 np->n_flag &= ~NMODIFIED;
3735 }
3736 nfs_node_unlock(np);
3737 nfs_data_unlock(np);
3738 FSDBG_BOT(515, np, uio_offset(uio), uio_resid(uio), error);
3739 return error;
3740 }
3741
3742
3743 /*
3744 * NFS write call
3745 */
3746 int
3747 nfs_write_rpc(
3748 nfsnode_t np,
3749 uio_t uio,
3750 vfs_context_t ctx,
3751 int *iomodep,
3752 uint64_t *wverfp)
3753 {
3754 return nfs_write_rpc2(np, uio, vfs_context_thread(ctx), vfs_context_ucred(ctx), iomodep, wverfp);
3755 }
3756
3757 int
3758 nfs_write_rpc2(
3759 nfsnode_t np,
3760 uio_t uio,
3761 thread_t thd,
3762 kauth_cred_t cred,
3763 int *iomodep,
3764 uint64_t *wverfp)
3765 {
3766 struct nfsmount *nmp;
3767 int error = 0, nfsvers;
3768 int wverfset, commit = 0, committed;
3769 uint64_t wverf = 0, wverf2 = 0;
3770 size_t nmwsize, totalsize, tsiz, len, rlen = 0;
3771 struct nfsreq *req;
3772 #if CONFIG_NFS4
3773 uint32_t stategenid = 0, restart = 0;
3774 #endif
3775 uint32_t vrestart = 0;
3776 uio_t uio_save = NULL;
3777
3778 #if DIAGNOSTIC
3779 /* XXX limitation based on need to back up uio on short write */
3780 if (uio_iovcnt(uio) != 1) {
3781 panic("nfs3_write_rpc: iovcnt > 1");
3782 }
3783 #endif
3784 FSDBG_TOP(537, np, uio_offset(uio), uio_resid(uio), *iomodep);
3785 nmp = NFSTONMP(np);
3786 if (nfs_mount_gone(nmp)) {
3787 return ENXIO;
3788 }
3789 nfsvers = nmp->nm_vers;
3790 nmwsize = nmp->nm_wsize;
3791
3792 wverfset = 0;
3793 committed = NFS_WRITE_FILESYNC;
3794
3795 totalsize = tsiz = uio_resid(uio);
3796 if ((nfsvers == NFS_VER2) && ((uint64_t)(uio_offset(uio) + tsiz) > 0xffffffffULL)) {
3797 FSDBG_BOT(537, np, uio_offset(uio), uio_resid(uio), EFBIG);
3798 return EFBIG;
3799 }
3800
3801 uio_save = uio_duplicate(uio);
3802 if (uio_save == NULL) {
3803 return EIO;
3804 }
3805
3806 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
3807 while (tsiz > 0) {
3808 len = (tsiz > nmwsize) ? nmwsize : tsiz;
3809 FSDBG(537, np, uio_offset(uio), len, 0);
3810 if (np->n_flag & NREVOKE) {
3811 error = EIO;
3812 break;
3813 }
3814 #if CONFIG_NFS4
3815 if (nmp->nm_vers >= NFS_VER4) {
3816 stategenid = nmp->nm_stategenid;
3817 }
3818 #endif
3819 error = nmp->nm_funcs->nf_write_rpc_async(np, uio, len, thd, cred, *iomodep, NULL, &req);
3820 if (!error) {
3821 error = nmp->nm_funcs->nf_write_rpc_async_finish(np, req, &commit, &rlen, &wverf2);
3822 }
3823 nmp = NFSTONMP(np);
3824 if (nfs_mount_gone(nmp)) {
3825 error = ENXIO;
3826 }
3827 #if CONFIG_NFS4
3828 if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error) &&
3829 (++restart <= nfs_mount_state_max_restarts(nmp))) { /* guard against no progress */
3830 lck_mtx_lock(&nmp->nm_lock);
3831 if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
3832 NP(np, "nfs_write_rpc: error %d, initiating recovery", error);
3833 nfs_need_recover(nmp, error);
3834 }
3835 lck_mtx_unlock(&nmp->nm_lock);
3836 if (np->n_flag & NREVOKE) {
3837 error = EIO;
3838 } else {
3839 if (error == NFSERR_GRACE) {
3840 tsleep(&nmp->nm_state, (PZERO - 1), "nfsgrace", 2 * hz);
3841 }
3842 if (!(error = nfs_mount_state_wait_for_recovery(nmp))) {
3843 continue;
3844 }
3845 }
3846 }
3847 #endif
3848 if (error) {
3849 break;
3850 }
3851 if (nfsvers == NFS_VER2) {
3852 tsiz -= len;
3853 continue;
3854 }
3855
3856 /* check for a short write */
3857 if (rlen < len) {
3858 /* Reset the uio to reflect the actual transfer */
3859 *uio = *uio_save;
3860 uio_update(uio, totalsize - (tsiz - rlen));
3861 len = rlen;
3862 }
3863
3864 /* return lowest commit level returned */
3865 if (commit < committed) {
3866 committed = commit;
3867 }
3868
3869 tsiz -= len;
3870
3871 /* check write verifier */
3872 if (!wverfset) {
3873 wverf = wverf2;
3874 wverfset = 1;
3875 } else if (wverf != wverf2) {
3876 /* verifier changed, so we need to restart all the writes */
3877 if (++vrestart > 100) {
3878 /* give up after too many restarts */
3879 error = EIO;
3880 break;
3881 }
3882 *uio = *uio_save; // Reset the uio back to the start
3883 committed = NFS_WRITE_FILESYNC;
3884 wverfset = 0;
3885 tsiz = totalsize;
3886 }
3887 }
3888 if (uio_save) {
3889 uio_free(uio_save);
3890 }
3891 if (wverfset && wverfp) {
3892 *wverfp = wverf;
3893 }
3894 *iomodep = committed;
3895 if (error) {
3896 uio_setresid(uio, tsiz);
3897 }
3898 NFS_ZFREE(nfs_req_zone, req);
3899 FSDBG_BOT(537, np, committed, uio_resid(uio), error);
3900 return error;
3901 }
3902
3903 int
3904 nfs3_write_rpc_async(
3905 nfsnode_t np,
3906 uio_t uio,
3907 size_t len,
3908 thread_t thd,
3909 kauth_cred_t cred,
3910 int iomode,
3911 struct nfsreq_cbinfo *cb,
3912 struct nfsreq **reqp)
3913 {
3914 struct nfsmount *nmp;
3915 mount_t mp;
3916 int error = 0, nfsvers;
3917 struct nfsm_chain nmreq;
3918
3919 nmp = NFSTONMP(np);
3920 if (nfs_mount_gone(nmp)) {
3921 return ENXIO;
3922 }
3923 nfsvers = nmp->nm_vers;
3924
3925 /* for async mounts, don't bother sending sync write requests */
3926 if ((iomode != NFS_WRITE_UNSTABLE) && nfs_allow_async &&
3927 ((mp = NFSTOMP(np))) && (vfs_flags(mp) & MNT_ASYNC)) {
3928 iomode = NFS_WRITE_UNSTABLE;
3929 }
3930
3931 nfsm_chain_null(&nmreq);
3932 nfsm_chain_build_alloc_init(error, &nmreq,
3933 NFSX_FH(nfsvers) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
3934 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
3935 if (nfsvers == NFS_VER3) {
3936 nfsm_chain_add_64(error, &nmreq, uio_offset(uio));
3937 nfsm_chain_add_32(error, &nmreq, len);
3938 nfsm_chain_add_32(error, &nmreq, iomode);
3939 } else {
3940 nfsm_chain_add_32(error, &nmreq, 0);
3941 nfsm_chain_add_32(error, &nmreq, uio_offset(uio));
3942 nfsm_chain_add_32(error, &nmreq, 0);
3943 }
3944 nfsm_chain_add_32(error, &nmreq, len);
3945 nfsmout_if(error);
3946 error = nfsm_chain_add_uio(&nmreq, uio, len);
3947 nfsm_chain_build_done(error, &nmreq);
3948 nfsmout_if(error);
3949 error = nfs_request_async(np, NULL, &nmreq, NFSPROC_WRITE, thd, cred, NULL, 0, cb, reqp);
3950 nfsmout:
3951 nfsm_chain_cleanup(&nmreq);
3952 return error;
3953 }
3954
3955 int
3956 nfs3_write_rpc_async_finish(
3957 nfsnode_t np,
3958 struct nfsreq *req,
3959 int *iomodep,
3960 size_t *rlenp,
3961 uint64_t *wverfp)
3962 {
3963 struct nfsmount *nmp;
3964 int error = 0, lockerror = ENOENT, nfsvers, status;
3965 int updatemtime = 0, wccpostattr = 0, rlen, committed = NFS_WRITE_FILESYNC;
3966 u_int64_t xid, wverf;
3967 mount_t mp;
3968 struct nfsm_chain nmrep;
3969
3970 nmp = NFSTONMP(np);
3971 if (nfs_mount_gone(nmp)) {
3972 nfs_request_async_cancel(req);
3973 return ENXIO;
3974 }
3975 nfsvers = nmp->nm_vers;
3976
3977 nfsm_chain_null(&nmrep);
3978
3979 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
3980 if (error == EINPROGRESS) { /* async request restarted */
3981 return error;
3982 }
3983 nmp = NFSTONMP(np);
3984 if (nfs_mount_gone(nmp)) {
3985 error = ENXIO;
3986 }
3987 if (!error && (lockerror = nfs_node_lock(np))) {
3988 error = lockerror;
3989 }
3990 if (nfsvers == NFS_VER3) {
3991 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
3992 nfsm_chain_get_wcc_data(error, &nmrep, np, &premtime, &wccpostattr, &xid);
3993 if (nfstimespeccmp(&np->n_mtime, &premtime, ==)) {
3994 updatemtime = 1;
3995 }
3996 if (!error) {
3997 error = status;
3998 }
3999 nfsm_chain_get_32(error, &nmrep, rlen);
4000 nfsmout_if(error);
4001 *rlenp = rlen;
4002 if (rlen <= 0) {
4003 error = NFSERR_IO;
4004 }
4005 nfsm_chain_get_32(error, &nmrep, committed);
4006 nfsm_chain_get_64(error, &nmrep, wverf);
4007 nfsmout_if(error);
4008 if (wverfp) {
4009 *wverfp = wverf;
4010 }
4011 lck_mtx_lock(&nmp->nm_lock);
4012 if (!(nmp->nm_state & NFSSTA_HASWRITEVERF)) {
4013 nmp->nm_verf = wverf;
4014 nmp->nm_state |= NFSSTA_HASWRITEVERF;
4015 } else if (nmp->nm_verf != wverf) {
4016 nmp->nm_verf = wverf;
4017 }
4018 lck_mtx_unlock(&nmp->nm_lock);
4019 } else {
4020 if (!error) {
4021 error = status;
4022 }
4023 nfsm_chain_loadattr(error, &nmrep, np, nfsvers, &xid);
4024 nfsmout_if(error);
4025 }
4026 if (updatemtime) {
4027 NFS_CHANGED_UPDATE(nfsvers, np, &np->n_vattr);
4028 }
4029 nfsmout:
4030 if (!lockerror) {
4031 nfs_node_unlock(np);
4032 }
4033 nfsm_chain_cleanup(&nmrep);
4034 if ((committed != NFS_WRITE_FILESYNC) && nfs_allow_async &&
4035 ((mp = NFSTOMP(np))) && (vfs_flags(mp) & MNT_ASYNC)) {
4036 committed = NFS_WRITE_FILESYNC;
4037 }
4038 *iomodep = committed;
4039 return error;
4040 }
4041
4042 /*
4043 * NFS mknod vnode op
4044 *
4045 * For NFS v2 this is a kludge. Use a create RPC but with the IFMT bits of the
4046 * mode set to specify the file type and the size field for rdev.
4047 */
4048 int
4049 nfs3_vnop_mknod(
4050 struct vnop_mknod_args /* {
4051 * struct vnodeop_desc *a_desc;
4052 * vnode_t a_dvp;
4053 * vnode_t *a_vpp;
4054 * struct componentname *a_cnp;
4055 * struct vnode_attr *a_vap;
4056 * vfs_context_t a_context;
4057 * } */*ap)
4058 {
4059 vnode_t dvp = ap->a_dvp;
4060 vnode_t *vpp = ap->a_vpp;
4061 struct componentname *cnp = ap->a_cnp;
4062 struct vnode_attr *vap = ap->a_vap;
4063 vfs_context_t ctx = ap->a_context;
4064 vnode_t newvp = NULL;
4065 nfsnode_t np = NULL;
4066 struct nfsmount *nmp;
4067 nfsnode_t dnp = VTONFS(dvp);
4068 struct nfs_vattr *nvattr;
4069 fhandle_t *fh;
4070 int error = 0, lockerror = ENOENT, busyerror = ENOENT, status = 0, wccpostattr = 0;
4071 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
4072 u_int32_t rdev;
4073 u_int64_t xid = 0, dxid;
4074 int nfsvers, gotuid, gotgid;
4075 struct nfsm_chain nmreq, nmrep;
4076 struct nfsreq *req;
4077
4078 nmp = VTONMP(dvp);
4079 if (nfs_mount_gone(nmp)) {
4080 return ENXIO;
4081 }
4082 nfsvers = nmp->nm_vers;
4083
4084 if (!VATTR_IS_ACTIVE(vap, va_type)) {
4085 return EINVAL;
4086 }
4087 if (vap->va_type == VCHR || vap->va_type == VBLK) {
4088 if (!VATTR_IS_ACTIVE(vap, va_rdev)) {
4089 return EINVAL;
4090 }
4091 rdev = vap->va_rdev;
4092 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
4093 rdev = 0xffffffff;
4094 } else {
4095 return ENOTSUP;
4096 }
4097 if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) {
4098 return ENAMETOOLONG;
4099 }
4100
4101 nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
4102
4103 VATTR_SET_SUPPORTED(vap, va_mode);
4104 VATTR_SET_SUPPORTED(vap, va_uid);
4105 VATTR_SET_SUPPORTED(vap, va_gid);
4106 VATTR_SET_SUPPORTED(vap, va_data_size);
4107 VATTR_SET_SUPPORTED(vap, va_access_time);
4108 VATTR_SET_SUPPORTED(vap, va_modify_time);
4109 gotuid = VATTR_IS_ACTIVE(vap, va_uid);
4110 gotgid = VATTR_IS_ACTIVE(vap, va_gid);
4111
4112 nfsm_chain_null(&nmreq);
4113 nfsm_chain_null(&nmrep);
4114
4115 fh = zalloc(nfs_fhandle_zone);
4116 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
4117 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
4118
4119 nfsm_chain_build_alloc_init(error, &nmreq,
4120 NFSX_FH(nfsvers) + 4 * NFSX_UNSIGNED +
4121 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(nfsvers));
4122 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
4123 nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
4124 if (nfsvers == NFS_VER3) {
4125 nfsm_chain_add_32(error, &nmreq, vtonfs_type(vap->va_type, nfsvers));
4126 nfsm_chain_add_v3sattr(nmp, error, &nmreq, vap);
4127 if (vap->va_type == VCHR || vap->va_type == VBLK) {
4128 nfsm_chain_add_32(error, &nmreq, major(vap->va_rdev));
4129 nfsm_chain_add_32(error, &nmreq, minor(vap->va_rdev));
4130 }
4131 } else {
4132 nfsm_chain_add_v2sattr(error, &nmreq, vap, rdev);
4133 }
4134 nfsm_chain_build_done(error, &nmreq);
4135 if (!error) {
4136 error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
4137 }
4138 nfsmout_if(error);
4139
4140 error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_MKNOD,
4141 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
4142 if (!error) {
4143 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
4144 }
4145
4146 if ((lockerror = nfs_node_lock(dnp))) {
4147 error = lockerror;
4148 }
4149 /* XXX no EEXIST kludge here? */
4150 dxid = xid;
4151 if (!error && !status) {
4152 if (dnp->n_flag & NNEGNCENTRIES) {
4153 dnp->n_flag &= ~NNEGNCENTRIES;
4154 cache_purge_negatives(dvp);
4155 }
4156 error = nfsm_chain_get_fh_attr(nmp, &nmrep, dnp, ctx, nfsvers, &xid, fh, nvattr);
4157 }
4158 if (nfsvers == NFS_VER3) {
4159 nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
4160 }
4161 if (!error) {
4162 error = status;
4163 }
4164 nfsmout:
4165 nfsm_chain_cleanup(&nmreq);
4166 nfsm_chain_cleanup(&nmrep);
4167
4168 if (!lockerror) {
4169 dnp->n_flag |= NMODIFIED;
4170 /* if directory hadn't changed, update namecache mtime */
4171 if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) {
4172 NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
4173 }
4174 nfs_node_unlock(dnp);
4175 /* nfs_getattr() will check changed and purge caches */
4176 nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
4177 }
4178
4179 if (!error && fh->fh_len) {
4180 error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh->fh_data, fh->fh_len, nvattr, &xid, req->r_auth, NG_MAKEENTRY, &np);
4181 }
4182 if (!error && !np) {
4183 error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
4184 }
4185 if (!error && np) {
4186 newvp = NFSTOV(np);
4187 }
4188 if (!busyerror) {
4189 nfs_node_clear_busy(dnp);
4190 }
4191
4192 if (!error && (gotuid || gotgid) &&
4193 (!newvp || nfs_getattrcache(np, nvattr, 0) ||
4194 (gotuid && (nvattr->nva_uid != vap->va_uid)) ||
4195 (gotgid && (nvattr->nva_gid != vap->va_gid)))) {
4196 /* clear ID bits if server didn't use them (or we can't tell) */
4197 VATTR_CLEAR_SUPPORTED(vap, va_uid);
4198 VATTR_CLEAR_SUPPORTED(vap, va_gid);
4199 }
4200 if (error) {
4201 if (newvp) {
4202 nfs_node_unlock(np);
4203 vnode_put(newvp);
4204 }
4205 } else {
4206 *vpp = newvp;
4207 nfs_node_unlock(np);
4208 }
4209 NFS_ZFREE(nfs_fhandle_zone, fh);
4210 NFS_ZFREE(nfs_req_zone, req);
4211 FREE(nvattr, M_TEMP);
4212 return error;
4213 }
4214
4215 static uint32_t create_verf;
4216 /*
4217 * NFS file create call
4218 */
4219 int
4220 nfs3_vnop_create(
4221 struct vnop_create_args /* {
4222 * struct vnodeop_desc *a_desc;
4223 * vnode_t a_dvp;
4224 * vnode_t *a_vpp;
4225 * struct componentname *a_cnp;
4226 * struct vnode_attr *a_vap;
4227 * vfs_context_t a_context;
4228 * } */*ap)
4229 {
4230 vfs_context_t ctx = ap->a_context;
4231 vnode_t dvp = ap->a_dvp;
4232 struct vnode_attr *vap = ap->a_vap;
4233 struct componentname *cnp = ap->a_cnp;
4234 struct nfs_vattr *nvattr;
4235 fhandle_t *fh;
4236 nfsnode_t np = NULL;
4237 struct nfsmount *nmp;
4238 nfsnode_t dnp = VTONFS(dvp);
4239 vnode_t newvp = NULL;
4240 int error = 0, lockerror = ENOENT, busyerror = ENOENT, status = 0, wccpostattr = 0, fmode = 0;
4241 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
4242 int nfsvers, gotuid, gotgid;
4243 u_int64_t xid = 0, dxid;
4244 uint32_t val;
4245 struct nfsm_chain nmreq, nmrep;
4246 struct nfsreq *req;
4247 struct nfs_dulookup *dul;
4248 int dul_in_progress = 0;
4249 int namedattrs;
4250
4251 nmp = VTONMP(dvp);
4252 if (nfs_mount_gone(nmp)) {
4253 return ENXIO;
4254 }
4255 nfsvers = nmp->nm_vers;
4256 namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
4257
4258 if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) {
4259 return ENAMETOOLONG;
4260 }
4261
4262 nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
4263
4264 VATTR_SET_SUPPORTED(vap, va_mode);
4265 VATTR_SET_SUPPORTED(vap, va_uid);
4266 VATTR_SET_SUPPORTED(vap, va_gid);
4267 VATTR_SET_SUPPORTED(vap, va_data_size);
4268 VATTR_SET_SUPPORTED(vap, va_access_time);
4269 VATTR_SET_SUPPORTED(vap, va_modify_time);
4270 gotuid = VATTR_IS_ACTIVE(vap, va_uid);
4271 gotgid = VATTR_IS_ACTIVE(vap, va_gid);
4272
4273 if ((vap->va_vaflags & VA_EXCLUSIVE)
4274 ) {
4275 fmode |= O_EXCL;
4276 if (!VATTR_IS_ACTIVE(vap, va_access_time) || !VATTR_IS_ACTIVE(vap, va_modify_time)) {
4277 vap->va_vaflags |= VA_UTIMES_NULL;
4278 }
4279 }
4280
4281 fh = zalloc(nfs_fhandle_zone);
4282 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
4283 MALLOC(dul, struct nfs_dulookup *, sizeof(*dul), M_TEMP, M_WAITOK);
4284 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
4285
4286 again:
4287 error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
4288 if (!namedattrs) {
4289 nfs_dulookup_init(dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
4290 }
4291
4292 nfsm_chain_null(&nmreq);
4293 nfsm_chain_null(&nmrep);
4294
4295 nfsm_chain_build_alloc_init(error, &nmreq,
4296 NFSX_FH(nfsvers) + 2 * NFSX_UNSIGNED +
4297 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(nfsvers));
4298 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
4299 nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
4300 if (nfsvers == NFS_VER3) {
4301 if (fmode & O_EXCL) {
4302 nfsm_chain_add_32(error, &nmreq, NFS_CREATE_EXCLUSIVE);
4303 lck_rw_lock_shared(in_ifaddr_rwlock);
4304 if (!TAILQ_EMPTY(&in_ifaddrhead)) {
4305 val = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
4306 } else {
4307 val = create_verf;
4308 }
4309 lck_rw_done(in_ifaddr_rwlock);
4310 nfsm_chain_add_32(error, &nmreq, val);
4311 ++create_verf;
4312 nfsm_chain_add_32(error, &nmreq, create_verf);
4313 } else {
4314 nfsm_chain_add_32(error, &nmreq, NFS_CREATE_UNCHECKED);
4315 nfsm_chain_add_v3sattr(nmp, error, &nmreq, vap);
4316 }
4317 } else {
4318 nfsm_chain_add_v2sattr(error, &nmreq, vap, 0);
4319 }
4320 nfsm_chain_build_done(error, &nmreq);
4321 nfsmout_if(error);
4322
4323 error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_CREATE,
4324 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
4325 if (!error) {
4326 if (!namedattrs) {
4327 nfs_dulookup_start(dul, dnp, ctx);
4328 dul_in_progress = 1;
4329 }
4330 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
4331 }
4332
4333 if ((lockerror = nfs_node_lock(dnp))) {
4334 error = lockerror;
4335 }
4336 dxid = xid;
4337 if (!error && !status) {
4338 if (dnp->n_flag & NNEGNCENTRIES) {
4339 dnp->n_flag &= ~NNEGNCENTRIES;
4340 cache_purge_negatives(dvp);
4341 }
4342 error = nfsm_chain_get_fh_attr(nmp, &nmrep, dnp, ctx, nfsvers, &xid, fh, nvattr);
4343 }
4344 if (nfsvers == NFS_VER3) {
4345 nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
4346 }
4347 if (!error) {
4348 error = status;
4349 }
4350 nfsmout:
4351 nfsm_chain_cleanup(&nmreq);
4352 nfsm_chain_cleanup(&nmrep);
4353
4354 if (!lockerror) {
4355 dnp->n_flag |= NMODIFIED;
4356 /* if directory hadn't changed, update namecache mtime */
4357 if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) {
4358 NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
4359 }
4360 nfs_node_unlock(dnp);
4361 /* nfs_getattr() will check changed and purge caches */
4362 nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
4363 }
4364
4365 if (!error && fh->fh_len) {
4366 error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh->fh_data, fh->fh_len, nvattr, &xid, req->r_auth, NG_MAKEENTRY, &np);
4367 }
4368 if (!error && !np) {
4369 error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
4370 }
4371 if (!error && np) {
4372 newvp = NFSTOV(np);
4373 }
4374
4375 if (dul_in_progress) {
4376 nfs_dulookup_finish(dul, dnp, ctx);
4377 }
4378 if (!busyerror) {
4379 nfs_node_clear_busy(dnp);
4380 }
4381
4382 if (error) {
4383 if ((nfsvers == NFS_VER3) && (fmode & O_EXCL) && (error == NFSERR_NOTSUPP)) {
4384 fmode &= ~O_EXCL;
4385 goto again;
4386 }
4387 if (newvp) {
4388 nfs_node_unlock(np);
4389 vnode_put(newvp);
4390 }
4391 } else if ((nfsvers == NFS_VER3) && (fmode & O_EXCL)) {
4392 nfs_node_unlock(np);
4393 error = nfs3_setattr_rpc(np, vap, ctx);
4394 if (error && (gotuid || gotgid)) {
4395 /* it's possible the server didn't like our attempt to set IDs. */
4396 /* so, let's try it again without those */
4397 VATTR_CLEAR_ACTIVE(vap, va_uid);
4398 VATTR_CLEAR_ACTIVE(vap, va_gid);
4399 error = nfs3_setattr_rpc(np, vap, ctx);
4400 }
4401 if (error) {
4402 vnode_put(newvp);
4403 } else {
4404 nfs_node_lock_force(np);
4405 }
4406 }
4407 if (!error) {
4408 *ap->a_vpp = newvp;
4409 }
4410 if (!error && (gotuid || gotgid) &&
4411 (!newvp || nfs_getattrcache(np, nvattr, 0) ||
4412 (gotuid && (nvattr->nva_uid != vap->va_uid)) ||
4413 (gotgid && (nvattr->nva_gid != vap->va_gid)))) {
4414 /* clear ID bits if server didn't use them (or we can't tell) */
4415 VATTR_CLEAR_SUPPORTED(vap, va_uid);
4416 VATTR_CLEAR_SUPPORTED(vap, va_gid);
4417 }
4418 if (!error) {
4419 nfs_node_unlock(np);
4420 }
4421 NFS_ZFREE(nfs_fhandle_zone, fh);
4422 NFS_ZFREE(nfs_req_zone, req);
4423 FREE(dul, M_TEMP);
4424 FREE(nvattr, M_TEMP);
4425 return error;
4426 }
4427
4428 /*
4429 * NFS file remove call
4430 * To try and make NFS semantics closer to UFS semantics, a file that has
4431 * other processes using the vnode is renamed instead of removed and then
4432 * removed later on the last close.
4433 * - If vnode_isinuse()
4434 * If a rename is not already in the works
4435 * call nfs_sillyrename() to set it up
4436 * else
4437 * do the remove RPC
4438 */
4439 int
4440 nfs_vnop_remove(
4441 struct vnop_remove_args /* {
4442 * struct vnodeop_desc *a_desc;
4443 * vnode_t a_dvp;
4444 * vnode_t a_vp;
4445 * struct componentname *a_cnp;
4446 * int a_flags;
4447 * vfs_context_t a_context;
4448 * } */*ap)
4449 {
4450 vfs_context_t ctx = ap->a_context;
4451 vnode_t vp = ap->a_vp;
4452 vnode_t dvp = ap->a_dvp;
4453 struct componentname *cnp = ap->a_cnp;
4454 nfsnode_t dnp = VTONFS(dvp);
4455 nfsnode_t np = VTONFS(vp);
4456 int error = 0, nfsvers, namedattrs, inuse, gotattr = 0, flushed = 0, setsize = 0;
4457 struct nfs_vattr *nvattr;
4458 struct nfsmount *nmp;
4459 struct nfs_dulookup *dul;
4460
4461 /* XXX prevent removing a sillyrenamed file? */
4462
4463 nmp = NFSTONMP(dnp);
4464 if (nfs_mount_gone(nmp)) {
4465 return ENXIO;
4466 }
4467
4468 if (vnode_isdir(vp)) {
4469 return EPERM;
4470 }
4471
4472 nfsvers = nmp->nm_vers;
4473 namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
4474 MALLOC(dul, struct nfs_dulookup *, sizeof(*dul), M_TEMP, M_WAITOK);
4475 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
4476
4477 again_relock:
4478 error = nfs_node_set_busy2(dnp, np, vfs_context_thread(ctx));
4479 if (error) {
4480 goto out_free;
4481 }
4482
4483 /* lock the node while we remove the file */
4484 lck_mtx_lock(&nfs_node_hash_mutex);
4485 while (np->n_hflag & NHLOCKED) {
4486 np->n_hflag |= NHLOCKWANT;
4487 msleep(np, &nfs_node_hash_mutex, PINOD, "nfs_remove", NULL);
4488 }
4489 np->n_hflag |= NHLOCKED;
4490 lck_mtx_unlock(&nfs_node_hash_mutex);
4491
4492 if (!namedattrs) {
4493 nfs_dulookup_init(dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
4494 }
4495
4496 again:
4497 inuse = vnode_isinuse(vp, 0);
4498 if ((ap->a_flags & VNODE_REMOVE_NODELETEBUSY) && inuse) {
4499 /* Caller requested Carbon delete semantics, but file is busy */
4500 error = EBUSY;
4501 goto out;
4502 }
4503 if (inuse && !gotattr) {
4504 if (nfs_getattr(np, nvattr, ctx, NGA_CACHED)) {
4505 nvattr->nva_nlink = 1;
4506 }
4507 gotattr = 1;
4508 goto again;
4509 }
4510 if (!inuse || (np->n_sillyrename && (nvattr->nva_nlink > 1))) {
4511 if (!inuse && !flushed) { /* flush all the buffers first */
4512 /* unlock the node */
4513 lck_mtx_lock(&nfs_node_hash_mutex);
4514 np->n_hflag &= ~NHLOCKED;
4515 if (np->n_hflag & NHLOCKWANT) {
4516 np->n_hflag &= ~NHLOCKWANT;
4517 wakeup(np);
4518 }
4519 lck_mtx_unlock(&nfs_node_hash_mutex);
4520 nfs_node_clear_busy2(dnp, np);
4521 error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1);
4522 FSDBG(260, np, np->n_size, np->n_vattr.nva_size, 0xf00d0011);
4523 flushed = 1;
4524 if (error == EINTR) {
4525 nfs_node_lock_force(np);
4526 NATTRINVALIDATE(np);
4527 nfs_node_unlock(np);
4528 goto out_free;
4529 }
4530 if (!namedattrs) {
4531 nfs_dulookup_finish(dul, dnp, ctx);
4532 }
4533 goto again_relock;
4534 }
4535 #if CONFIG_NFS4
4536 if ((nmp->nm_vers >= NFS_VER4) && (np->n_openflags & N_DELEG_MASK)) {
4537 nfs4_delegation_return(np, 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
4538 }
4539 #endif
4540 /*
4541 * Purge the name cache so that the chance of a lookup for
4542 * the name succeeding while the remove is in progress is
4543 * minimized.
4544 */
4545 nfs_name_cache_purge(dnp, np, cnp, ctx);
4546
4547 if (!namedattrs) {
4548 nfs_dulookup_start(dul, dnp, ctx);
4549 }
4550
4551 /* Do the rpc */
4552 error = nmp->nm_funcs->nf_remove_rpc(dnp, cnp->cn_nameptr, cnp->cn_namelen,
4553 vfs_context_thread(ctx), vfs_context_ucred(ctx));
4554
4555 /*
4556 * Kludge City: If the first reply to the remove rpc is lost..
4557 * the reply to the retransmitted request will be ENOENT
4558 * since the file was in fact removed
4559 * Therefore, we cheat and return success.
4560 */
4561 if (error == ENOENT) {
4562 error = 0;
4563 }
4564
4565 if (!error && !inuse && !np->n_sillyrename) {
4566 /*
4567 * removal succeeded, it's not in use, and not silly renamed so
4568 * remove nfsnode from hash now so we can't accidentally find it
4569 * again if another object gets created with the same filehandle
4570 * before this vnode gets reclaimed
4571 */
4572 lck_mtx_lock(&nfs_node_hash_mutex);
4573 if (np->n_hflag & NHHASHED) {
4574 LIST_REMOVE(np, n_hash);
4575 np->n_hflag &= ~NHHASHED;
4576 FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
4577 }
4578 lck_mtx_unlock(&nfs_node_hash_mutex);
4579 /* clear flags now: won't get nfs_vnop_inactive for recycled vnode */
4580 /* clear all flags other than these */
4581 nfs_node_lock_force(np);
4582 np->n_flag &= (NMODIFIED);
4583 NATTRINVALIDATE(np);
4584 nfs_node_unlock(np);
4585 vnode_recycle(vp);
4586 setsize = 1;
4587 } else {
4588 nfs_node_lock_force(np);
4589 NATTRINVALIDATE(np);
4590 nfs_node_unlock(np);
4591 }
4592 } else if (!np->n_sillyrename) {
4593 if (!namedattrs) {
4594 nfs_dulookup_start(dul, dnp, ctx);
4595 }
4596 error = nfs_sillyrename(dnp, np, cnp, ctx);
4597 nfs_node_lock_force(np);
4598 NATTRINVALIDATE(np);
4599 nfs_node_unlock(np);
4600 } else {
4601 nfs_node_lock_force(np);
4602 NATTRINVALIDATE(np);
4603 nfs_node_unlock(np);
4604 if (!namedattrs) {
4605 nfs_dulookup_start(dul, dnp, ctx);
4606 }
4607 }
4608
4609 /* nfs_getattr() will check changed and purge caches */
4610 nfs_getattr(dnp, NULL, ctx, NGA_CACHED);
4611 if (!namedattrs) {
4612 nfs_dulookup_finish(dul, dnp, ctx);
4613 }
4614 out:
4615 /* unlock the node */
4616 lck_mtx_lock(&nfs_node_hash_mutex);
4617 np->n_hflag &= ~NHLOCKED;
4618 if (np->n_hflag & NHLOCKWANT) {
4619 np->n_hflag &= ~NHLOCKWANT;
4620 wakeup(np);
4621 }
4622 lck_mtx_unlock(&nfs_node_hash_mutex);
4623 nfs_node_clear_busy2(dnp, np);
4624 if (setsize) {
4625 ubc_setsize(vp, 0);
4626 }
4627 out_free:
4628 FREE(dul, M_TEMP);
4629 FREE(nvattr, M_TEMP);
4630 return error;
4631 }
4632
4633 /*
4634 * NFS silly-renamed file removal function called from nfs_vnop_inactive
4635 */
4636 int
4637 nfs_removeit(struct nfs_sillyrename *nsp)
4638 {
4639 struct nfsmount *nmp = NFSTONMP(nsp->nsr_dnp);
4640 if (nfs_mount_gone(nmp)) {
4641 return ENXIO;
4642 }
4643 return nmp->nm_funcs->nf_remove_rpc(nsp->nsr_dnp, nsp->nsr_name, nsp->nsr_namlen, NULL, nsp->nsr_cred);
4644 }
4645
4646 /*
4647 * NFS remove rpc, called from nfs_remove() and nfs_removeit().
4648 */
4649 int
4650 nfs3_remove_rpc(
4651 nfsnode_t dnp,
4652 char *name,
4653 int namelen,
4654 thread_t thd,
4655 kauth_cred_t cred)
4656 {
4657 int error = 0, lockerror = ENOENT, status = 0, wccpostattr = 0;
4658 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
4659 struct nfsmount *nmp;
4660 int nfsvers;
4661 u_int64_t xid;
4662 struct nfsm_chain nmreq, nmrep;
4663
4664 nmp = NFSTONMP(dnp);
4665 if (nfs_mount_gone(nmp)) {
4666 return ENXIO;
4667 }
4668 nfsvers = nmp->nm_vers;
4669 if ((nfsvers == NFS_VER2) && (namelen > NFS_MAXNAMLEN)) {
4670 return ENAMETOOLONG;
4671 }
4672
4673 nfsm_chain_null(&nmreq);
4674 nfsm_chain_null(&nmrep);
4675
4676 nfsm_chain_build_alloc_init(error, &nmreq,
4677 NFSX_FH(nfsvers) + NFSX_UNSIGNED + nfsm_rndup(namelen));
4678 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
4679 nfsm_chain_add_name(error, &nmreq, name, namelen, nmp);
4680 nfsm_chain_build_done(error, &nmreq);
4681 nfsmout_if(error);
4682
4683 error = nfs_request2(dnp, NULL, &nmreq, NFSPROC_REMOVE, thd, cred, NULL, 0, &nmrep, &xid, &status);
4684
4685 if ((lockerror = nfs_node_lock(dnp))) {
4686 error = lockerror;
4687 }
4688 if (nfsvers == NFS_VER3) {
4689 nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &xid);
4690 }
4691 nfsmout_if(error);
4692 dnp->n_flag |= NMODIFIED;
4693 /* if directory hadn't changed, update namecache mtime */
4694 if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) {
4695 NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
4696 }
4697 if (!wccpostattr) {
4698 NATTRINVALIDATE(dnp);
4699 }
4700 if (!error) {
4701 error = status;
4702 }
4703 nfsmout:
4704 if (!lockerror) {
4705 nfs_node_unlock(dnp);
4706 }
4707 nfsm_chain_cleanup(&nmreq);
4708 nfsm_chain_cleanup(&nmrep);
4709 return error;
4710 }
4711
4712 /*
4713 * NFS file rename call
4714 */
4715 int
4716 nfs_vnop_rename(
4717 struct vnop_rename_args /* {
4718 * struct vnodeop_desc *a_desc;
4719 * vnode_t a_fdvp;
4720 * vnode_t a_fvp;
4721 * struct componentname *a_fcnp;
4722 * vnode_t a_tdvp;
4723 * vnode_t a_tvp;
4724 * struct componentname *a_tcnp;
4725 * vfs_context_t a_context;
4726 * } */*ap)
4727 {
4728 vfs_context_t ctx = ap->a_context;
4729 vnode_t fdvp = ap->a_fdvp;
4730 vnode_t fvp = ap->a_fvp;
4731 vnode_t tdvp = ap->a_tdvp;
4732 vnode_t tvp = ap->a_tvp;
4733 nfsnode_t fdnp, fnp, tdnp, tnp;
4734 struct componentname *tcnp = ap->a_tcnp;
4735 struct componentname *fcnp = ap->a_fcnp;
4736 int error, nfsvers, inuse = 0, tvprecycle = 0, locked = 0;
4737 mount_t fmp, tdmp, tmp;
4738 struct nfs_vattr *nvattr;
4739 struct nfsmount *nmp;
4740
4741 fdnp = VTONFS(fdvp);
4742 fnp = VTONFS(fvp);
4743 tdnp = VTONFS(tdvp);
4744 tnp = tvp ? VTONFS(tvp) : NULL;
4745
4746 nmp = NFSTONMP(fdnp);
4747 if (nfs_mount_gone(nmp)) {
4748 return ENXIO;
4749 }
4750 nfsvers = nmp->nm_vers;
4751
4752 error = nfs_node_set_busy4(fdnp, fnp, tdnp, tnp, vfs_context_thread(ctx));
4753 if (error) {
4754 return error;
4755 }
4756
4757 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
4758
4759 if (tvp && (tvp != fvp)) {
4760 /* lock the node while we rename over the existing file */
4761 lck_mtx_lock(&nfs_node_hash_mutex);
4762 while (tnp->n_hflag & NHLOCKED) {
4763 tnp->n_hflag |= NHLOCKWANT;
4764 msleep(tnp, &nfs_node_hash_mutex, PINOD, "nfs_rename", NULL);
4765 }
4766 tnp->n_hflag |= NHLOCKED;
4767 lck_mtx_unlock(&nfs_node_hash_mutex);
4768 locked = 1;
4769 }
4770
4771 /* Check for cross-device rename */
4772 fmp = vnode_mount(fvp);
4773 tmp = tvp ? vnode_mount(tvp) : NULL;
4774 tdmp = vnode_mount(tdvp);
4775 if ((fmp != tdmp) || (tvp && (fmp != tmp))) {
4776 error = EXDEV;
4777 goto out;
4778 }
4779
4780 /* XXX prevent renaming from/over a sillyrenamed file? */
4781
4782 /*
4783 * If the tvp exists and is in use, sillyrename it before doing the
4784 * rename of the new file over it.
4785 * XXX Can't sillyrename a directory.
4786 * Don't sillyrename if source and target are same vnode (hard
4787 * links or case-variants)
4788 */
4789 if (tvp && (tvp != fvp)) {
4790 inuse = vnode_isinuse(tvp, 0);
4791 }
4792 if (inuse && !tnp->n_sillyrename && (vnode_vtype(tvp) != VDIR)) {
4793 error = nfs_sillyrename(tdnp, tnp, tcnp, ctx);
4794 if (error) {
4795 /* sillyrename failed. Instead of pressing on, return error */
4796 goto out; /* should not be ENOENT. */
4797 } else {
4798 /* sillyrename succeeded.*/
4799 tvp = NULL;
4800 }
4801 }
4802 #if CONFIG_NFS4
4803 else if (tvp && (nmp->nm_vers >= NFS_VER4) && (tnp->n_openflags & N_DELEG_MASK)) {
4804 nfs4_delegation_return(tnp, 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
4805 }
4806 #endif
4807 error = nmp->nm_funcs->nf_rename_rpc(fdnp, fcnp->cn_nameptr, fcnp->cn_namelen,
4808 tdnp, tcnp->cn_nameptr, tcnp->cn_namelen, ctx);
4809
4810 /*
4811 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
4812 */
4813 if (error == ENOENT) {
4814 error = 0;
4815 }
4816
4817 if (tvp && (tvp != fvp) && !tnp->n_sillyrename) {
4818 nfs_node_lock_force(tnp);
4819 tvprecycle = (!error && !vnode_isinuse(tvp, 0) &&
4820 (nfs_getattrcache(tnp, nvattr, 0) || (nvattr->nva_nlink == 1)));
4821 nfs_node_unlock(tnp);
4822 lck_mtx_lock(&nfs_node_hash_mutex);
4823 if (tvprecycle && (tnp->n_hflag & NHHASHED)) {
4824 /*
4825 * remove nfsnode from hash now so we can't accidentally find it
4826 * again if another object gets created with the same filehandle
4827 * before this vnode gets reclaimed
4828 */
4829 LIST_REMOVE(tnp, n_hash);
4830 tnp->n_hflag &= ~NHHASHED;
4831 FSDBG(266, 0, tnp, tnp->n_flag, 0xb1eb1e);
4832 }
4833 lck_mtx_unlock(&nfs_node_hash_mutex);
4834 }
4835
4836 /* purge the old name cache entries and enter the new one */
4837 nfs_name_cache_purge(fdnp, fnp, fcnp, ctx);
4838 if (tvp) {
4839 nfs_name_cache_purge(tdnp, tnp, tcnp, ctx);
4840 if (tvprecycle) {
4841 /* clear flags now: won't get nfs_vnop_inactive for recycled vnode */
4842 /* clear all flags other than these */
4843 nfs_node_lock_force(tnp);
4844 tnp->n_flag &= (NMODIFIED);
4845 nfs_node_unlock(tnp);
4846 vnode_recycle(tvp);
4847 }
4848 }
4849 if (!error) {
4850 nfs_node_lock_force(tdnp);
4851 if (tdnp->n_flag & NNEGNCENTRIES) {
4852 tdnp->n_flag &= ~NNEGNCENTRIES;
4853 cache_purge_negatives(tdvp);
4854 }
4855 nfs_node_unlock(tdnp);
4856 nfs_node_lock_force(fnp);
4857 cache_enter(tdvp, fvp, tcnp);
4858 if (tdvp != fdvp) { /* update parent pointer */
4859 if (fnp->n_parent && !vnode_get(fnp->n_parent)) {
4860 /* remove ref from old parent */
4861 vnode_rele(fnp->n_parent);
4862 vnode_put(fnp->n_parent);
4863 }
4864 fnp->n_parent = tdvp;
4865 if (tdvp && !vnode_get(tdvp)) {
4866 /* add ref to new parent */
4867 vnode_ref(tdvp);
4868 vnode_put(tdvp);
4869 } else {
4870 fnp->n_parent = NULL;
4871 }
4872 }
4873 nfs_node_unlock(fnp);
4874 }
4875 out:
4876 /* nfs_getattr() will check changed and purge caches */
4877 nfs_getattr(fdnp, NULL, ctx, NGA_CACHED);
4878 nfs_getattr(tdnp, NULL, ctx, NGA_CACHED);
4879 if (locked) {
4880 /* unlock node */
4881 lck_mtx_lock(&nfs_node_hash_mutex);
4882 tnp->n_hflag &= ~NHLOCKED;
4883 if (tnp->n_hflag & NHLOCKWANT) {
4884 tnp->n_hflag &= ~NHLOCKWANT;
4885 wakeup(tnp);
4886 }
4887 lck_mtx_unlock(&nfs_node_hash_mutex);
4888 }
4889 nfs_node_clear_busy4(fdnp, fnp, tdnp, tnp);
4890 FREE(nvattr, M_TEMP);
4891 return error;
4892 }
4893
4894 /*
4895 * Do an NFS rename rpc. Called from nfs_vnop_rename() and nfs_sillyrename().
4896 */
4897 int
4898 nfs3_rename_rpc(
4899 nfsnode_t fdnp,
4900 char *fnameptr,
4901 int fnamelen,
4902 nfsnode_t tdnp,
4903 char *tnameptr,
4904 int tnamelen,
4905 vfs_context_t ctx)
4906 {
4907 int error = 0, lockerror = ENOENT, status = 0, fwccpostattr = 0, twccpostattr = 0;
4908 struct timespec fpremtime = { .tv_sec = 0, .tv_nsec = 0 }, tpremtime = { .tv_sec = 0, .tv_nsec = 0 };
4909 struct nfsmount *nmp;
4910 int nfsvers;
4911 u_int64_t xid, txid;
4912 struct nfsm_chain nmreq, nmrep;
4913
4914 nmp = NFSTONMP(fdnp);
4915 if (nfs_mount_gone(nmp)) {
4916 return ENXIO;
4917 }
4918 nfsvers = nmp->nm_vers;
4919 if ((nfsvers == NFS_VER2) &&
4920 ((fnamelen > NFS_MAXNAMLEN) || (tnamelen > NFS_MAXNAMLEN))) {
4921 return ENAMETOOLONG;
4922 }
4923
4924 nfsm_chain_null(&nmreq);
4925 nfsm_chain_null(&nmrep);
4926
4927 nfsm_chain_build_alloc_init(error, &nmreq,
4928 (NFSX_FH(nfsvers) + NFSX_UNSIGNED) * 2 +
4929 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
4930 nfsm_chain_add_fh(error, &nmreq, nfsvers, fdnp->n_fhp, fdnp->n_fhsize);
4931 nfsm_chain_add_name(error, &nmreq, fnameptr, fnamelen, nmp);
4932 nfsm_chain_add_fh(error, &nmreq, nfsvers, tdnp->n_fhp, tdnp->n_fhsize);
4933 nfsm_chain_add_name(error, &nmreq, tnameptr, tnamelen, nmp);
4934 nfsm_chain_build_done(error, &nmreq);
4935 nfsmout_if(error);
4936
4937 error = nfs_request(fdnp, NULL, &nmreq, NFSPROC_RENAME, ctx, NULL, &nmrep, &xid, &status);
4938
4939 if ((lockerror = nfs_node_lock2(fdnp, tdnp))) {
4940 error = lockerror;
4941 }
4942 if (nfsvers == NFS_VER3) {
4943 txid = xid;
4944 nfsm_chain_get_wcc_data(error, &nmrep, fdnp, &fpremtime, &fwccpostattr, &xid);
4945 nfsm_chain_get_wcc_data(error, &nmrep, tdnp, &tpremtime, &twccpostattr, &txid);
4946 }
4947 if (!error) {
4948 error = status;
4949 }
4950 nfsmout:
4951 nfsm_chain_cleanup(&nmreq);
4952 nfsm_chain_cleanup(&nmrep);
4953 if (!lockerror) {
4954 fdnp->n_flag |= NMODIFIED;
4955 /* if directory hadn't changed, update namecache mtime */
4956 if (nfstimespeccmp(&fdnp->n_ncmtime, &fpremtime, ==)) {
4957 NFS_CHANGED_UPDATE_NC(nfsvers, fdnp, &fdnp->n_vattr);
4958 }
4959 if (!fwccpostattr) {
4960 NATTRINVALIDATE(fdnp);
4961 }
4962 tdnp->n_flag |= NMODIFIED;
4963 /* if directory hadn't changed, update namecache mtime */
4964 if (nfstimespeccmp(&tdnp->n_ncmtime, &tpremtime, ==)) {
4965 NFS_CHANGED_UPDATE_NC(nfsvers, tdnp, &tdnp->n_vattr);
4966 }
4967 if (!twccpostattr) {
4968 NATTRINVALIDATE(tdnp);
4969 }
4970 nfs_node_unlock2(fdnp, tdnp);
4971 }
4972 return error;
4973 }
4974
4975 /*
4976 * NFS hard link create call
4977 */
4978 int
4979 nfs3_vnop_link(
4980 struct vnop_link_args /* {
4981 * struct vnodeop_desc *a_desc;
4982 * vnode_t a_vp;
4983 * vnode_t a_tdvp;
4984 * struct componentname *a_cnp;
4985 * vfs_context_t a_context;
4986 * } */*ap)
4987 {
4988 vfs_context_t ctx = ap->a_context;
4989 vnode_t vp = ap->a_vp;
4990 vnode_t tdvp = ap->a_tdvp;
4991 struct componentname *cnp = ap->a_cnp;
4992 int error = 0, lockerror = ENOENT, status = 0, wccpostattr = 0, attrflag = 0;
4993 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
4994 struct nfsmount *nmp;
4995 nfsnode_t np = VTONFS(vp);
4996 nfsnode_t tdnp = VTONFS(tdvp);
4997 int nfsvers;
4998 u_int64_t xid, txid;
4999 struct nfsm_chain nmreq, nmrep;
5000
5001 if (vnode_mount(vp) != vnode_mount(tdvp)) {
5002 return EXDEV;
5003 }
5004
5005 nmp = VTONMP(vp);
5006 if (nfs_mount_gone(nmp)) {
5007 return ENXIO;
5008 }
5009 nfsvers = nmp->nm_vers;
5010 if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) {
5011 return ENAMETOOLONG;
5012 }
5013
5014 /*
5015 * Push all writes to the server, so that the attribute cache
5016 * doesn't get "out of sync" with the server.
5017 * XXX There should be a better way!
5018 */
5019 nfs_flush(np, MNT_WAIT, vfs_context_thread(ctx), V_IGNORE_WRITEERR);
5020
5021 error = nfs_node_set_busy2(tdnp, np, vfs_context_thread(ctx));
5022 if (error) {
5023 return error;
5024 }
5025
5026 nfsm_chain_null(&nmreq);
5027 nfsm_chain_null(&nmrep);
5028
5029 nfsm_chain_build_alloc_init(error, &nmreq,
5030 NFSX_FH(nfsvers) * 2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
5031 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
5032 nfsm_chain_add_fh(error, &nmreq, nfsvers, tdnp->n_fhp, tdnp->n_fhsize);
5033 nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
5034 nfsm_chain_build_done(error, &nmreq);
5035 nfsmout_if(error);
5036 error = nfs_request(np, NULL, &nmreq, NFSPROC_LINK, ctx, NULL, &nmrep, &xid, &status);
5037
5038 if ((lockerror = nfs_node_lock2(tdnp, np))) {
5039 error = lockerror;
5040 goto nfsmout;
5041 }
5042 if (nfsvers == NFS_VER3) {
5043 txid = xid;
5044 nfsm_chain_postop_attr_update_flag(error, &nmrep, np, attrflag, &xid);
5045 nfsm_chain_get_wcc_data(error, &nmrep, tdnp, &premtime, &wccpostattr, &txid);
5046 }
5047 if (!error) {
5048 error = status;
5049 }
5050 nfsmout:
5051 nfsm_chain_cleanup(&nmreq);
5052 nfsm_chain_cleanup(&nmrep);
5053 if (!lockerror) {
5054 if (!attrflag) {
5055 NATTRINVALIDATE(np);
5056 }
5057 tdnp->n_flag |= NMODIFIED;
5058 /* if directory hadn't changed, update namecache mtime */
5059 if (nfstimespeccmp(&tdnp->n_ncmtime, &premtime, ==)) {
5060 NFS_CHANGED_UPDATE_NC(nfsvers, tdnp, &tdnp->n_vattr);
5061 }
5062 if (!wccpostattr) {
5063 NATTRINVALIDATE(tdnp);
5064 }
5065 if (!error && (tdnp->n_flag & NNEGNCENTRIES)) {
5066 tdnp->n_flag &= ~NNEGNCENTRIES;
5067 cache_purge_negatives(tdvp);
5068 }
5069 nfs_node_unlock2(tdnp, np);
5070 }
5071 nfs_node_clear_busy2(tdnp, np);
5072 /*
5073 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
5074 */
5075 if (error == EEXIST) {
5076 error = 0;
5077 }
5078 return error;
5079 }
5080
5081 /*
5082 * NFS symbolic link create call
5083 */
5084 int
5085 nfs3_vnop_symlink(
5086 struct vnop_symlink_args /* {
5087 * struct vnodeop_desc *a_desc;
5088 * vnode_t a_dvp;
5089 * vnode_t *a_vpp;
5090 * struct componentname *a_cnp;
5091 * struct vnode_attr *a_vap;
5092 * char *a_target;
5093 * vfs_context_t a_context;
5094 * } */*ap)
5095 {
5096 vfs_context_t ctx = ap->a_context;
5097 vnode_t dvp = ap->a_dvp;
5098 struct vnode_attr *vap = ap->a_vap;
5099 struct componentname *cnp = ap->a_cnp;
5100 struct nfs_vattr *nvattr;
5101 fhandle_t *fh;
5102 int error = 0, lockerror = ENOENT, busyerror = ENOENT, status = 0, wccpostattr = 0;
5103 size_t slen;
5104 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
5105 vnode_t newvp = NULL;
5106 int nfsvers, gotuid, gotgid;
5107 u_int64_t xid = 0, dxid;
5108 nfsnode_t np = NULL;
5109 nfsnode_t dnp = VTONFS(dvp);
5110 struct nfsmount *nmp;
5111 struct nfsm_chain nmreq, nmrep;
5112 struct nfsreq *req;
5113 struct nfs_dulookup *dul;
5114 int namedattrs;
5115 int dul_in_progress = 0;
5116
5117 nmp = VTONMP(dvp);
5118 if (nfs_mount_gone(nmp)) {
5119 return ENXIO;
5120 }
5121 nfsvers = nmp->nm_vers;
5122 namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
5123
5124 slen = strlen(ap->a_target);
5125 if ((nfsvers == NFS_VER2) &&
5126 ((cnp->cn_namelen > NFS_MAXNAMLEN) || (slen > NFS_MAXPATHLEN))) {
5127 return ENAMETOOLONG;
5128 }
5129
5130 nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
5131
5132 VATTR_SET_SUPPORTED(vap, va_mode);
5133 VATTR_SET_SUPPORTED(vap, va_uid);
5134 VATTR_SET_SUPPORTED(vap, va_gid);
5135 VATTR_SET_SUPPORTED(vap, va_data_size);
5136 VATTR_SET_SUPPORTED(vap, va_access_time);
5137 VATTR_SET_SUPPORTED(vap, va_modify_time);
5138 gotuid = VATTR_IS_ACTIVE(vap, va_uid);
5139 gotgid = VATTR_IS_ACTIVE(vap, va_gid);
5140
5141 fh = zalloc(nfs_fhandle_zone);
5142 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
5143 MALLOC(dul, struct nfs_dulookup *, sizeof(*dul), M_TEMP, M_WAITOK);
5144 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
5145
5146 error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
5147 if (!namedattrs) {
5148 nfs_dulookup_init(dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
5149 }
5150
5151 nfsm_chain_null(&nmreq);
5152 nfsm_chain_null(&nmrep);
5153
5154 nfsm_chain_build_alloc_init(error, &nmreq,
5155 NFSX_FH(nfsvers) + 2 * NFSX_UNSIGNED +
5156 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(nfsvers));
5157 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
5158 nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
5159 if (nfsvers == NFS_VER3) {
5160 nfsm_chain_add_v3sattr(nmp, error, &nmreq, vap);
5161 }
5162 nfsm_chain_add_name(error, &nmreq, ap->a_target, slen, nmp);
5163 if (nfsvers == NFS_VER2) {
5164 nfsm_chain_add_v2sattr(error, &nmreq, vap, -1);
5165 }
5166 nfsm_chain_build_done(error, &nmreq);
5167 nfsmout_if(error);
5168
5169 error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_SYMLINK,
5170 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
5171 if (!error) {
5172 if (!namedattrs) {
5173 nfs_dulookup_start(dul, dnp, ctx);
5174 dul_in_progress = 1;
5175 }
5176 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
5177 }
5178
5179 if ((lockerror = nfs_node_lock(dnp))) {
5180 error = lockerror;
5181 }
5182 dxid = xid;
5183 if (!error && !status) {
5184 if (dnp->n_flag & NNEGNCENTRIES) {
5185 dnp->n_flag &= ~NNEGNCENTRIES;
5186 cache_purge_negatives(dvp);
5187 }
5188 if (nfsvers == NFS_VER3) {
5189 error = nfsm_chain_get_fh_attr(nmp, &nmrep, dnp, ctx, nfsvers, &xid, fh, nvattr);
5190 } else {
5191 fh->fh_len = 0;
5192 }
5193 }
5194 if (nfsvers == NFS_VER3) {
5195 nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
5196 }
5197 if (!error) {
5198 error = status;
5199 }
5200 nfsmout:
5201 nfsm_chain_cleanup(&nmreq);
5202 nfsm_chain_cleanup(&nmrep);
5203
5204 if (!lockerror) {
5205 dnp->n_flag |= NMODIFIED;
5206 /* if directory hadn't changed, update namecache mtime */
5207 if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) {
5208 NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
5209 }
5210 nfs_node_unlock(dnp);
5211 /* nfs_getattr() will check changed and purge caches */
5212 nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
5213 }
5214
5215 if (!error && fh->fh_len) {
5216 error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh->fh_data, fh->fh_len, nvattr, &xid, req->r_auth, NG_MAKEENTRY, &np);
5217 }
5218 if (!error && np) {
5219 newvp = NFSTOV(np);
5220 }
5221
5222 if (dul_in_progress) {
5223 nfs_dulookup_finish(dul, dnp, ctx);
5224 }
5225
5226 /*
5227 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
5228 * if we can succeed in looking up the symlink.
5229 */
5230 if ((error == EEXIST) || (!error && !newvp)) {
5231 if (newvp) {
5232 nfs_node_unlock(np);
5233 vnode_put(newvp);
5234 newvp = NULL;
5235 }
5236 error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
5237 if (!error) {
5238 newvp = NFSTOV(np);
5239 if (vnode_vtype(newvp) != VLNK) {
5240 error = EEXIST;
5241 }
5242 }
5243 }
5244 if (!busyerror) {
5245 nfs_node_clear_busy(dnp);
5246 }
5247 if (!error && (gotuid || gotgid) &&
5248 (!newvp || nfs_getattrcache(np, nvattr, 0) ||
5249 (gotuid && (nvattr->nva_uid != vap->va_uid)) ||
5250 (gotgid && (nvattr->nva_gid != vap->va_gid)))) {
5251 /* clear ID bits if server didn't use them (or we can't tell) */
5252 VATTR_CLEAR_SUPPORTED(vap, va_uid);
5253 VATTR_CLEAR_SUPPORTED(vap, va_gid);
5254 }
5255 if (error) {
5256 if (newvp) {
5257 nfs_node_unlock(np);
5258 vnode_put(newvp);
5259 }
5260 } else {
5261 nfs_node_unlock(np);
5262 *ap->a_vpp = newvp;
5263 }
5264 NFS_ZFREE(nfs_fhandle_zone, fh);
5265 NFS_ZFREE(nfs_req_zone, req);
5266 FREE(dul, M_TEMP);
5267 FREE(nvattr, M_TEMP);
5268 return error;
5269 }
5270
5271 /*
5272 * NFS make dir call
5273 */
5274 int
5275 nfs3_vnop_mkdir(
5276 struct vnop_mkdir_args /* {
5277 * struct vnodeop_desc *a_desc;
5278 * vnode_t a_dvp;
5279 * vnode_t *a_vpp;
5280 * struct componentname *a_cnp;
5281 * struct vnode_attr *a_vap;
5282 * vfs_context_t a_context;
5283 * } */*ap)
5284 {
5285 vfs_context_t ctx = ap->a_context;
5286 vnode_t dvp = ap->a_dvp;
5287 struct vnode_attr *vap = ap->a_vap;
5288 struct componentname *cnp = ap->a_cnp;
5289 struct nfs_vattr *nvattr;
5290 nfsnode_t np = NULL;
5291 struct nfsmount *nmp;
5292 nfsnode_t dnp = VTONFS(dvp);
5293 vnode_t newvp = NULL;
5294 int error = 0, lockerror = ENOENT, busyerror = ENOENT, status = 0, wccpostattr = 0;
5295 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
5296 int nfsvers, gotuid, gotgid;
5297 u_int64_t xid = 0, dxid;
5298 fhandle_t *fh;
5299 struct nfsm_chain nmreq, nmrep;
5300 struct nfsreq *req;
5301 struct nfs_dulookup *dul;
5302 int namedattrs;
5303 int dul_in_progress = 0;
5304
5305 nmp = VTONMP(dvp);
5306 if (nfs_mount_gone(nmp)) {
5307 return ENXIO;
5308 }
5309 nfsvers = nmp->nm_vers;
5310 namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
5311
5312 if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) {
5313 return ENAMETOOLONG;
5314 }
5315
5316 nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
5317
5318 VATTR_SET_SUPPORTED(vap, va_mode);
5319 VATTR_SET_SUPPORTED(vap, va_uid);
5320 VATTR_SET_SUPPORTED(vap, va_gid);
5321 VATTR_SET_SUPPORTED(vap, va_data_size);
5322 VATTR_SET_SUPPORTED(vap, va_access_time);
5323 VATTR_SET_SUPPORTED(vap, va_modify_time);
5324 gotuid = VATTR_IS_ACTIVE(vap, va_uid);
5325 gotgid = VATTR_IS_ACTIVE(vap, va_gid);
5326
5327 fh = zalloc(nfs_fhandle_zone);
5328 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
5329 MALLOC(dul, struct nfs_dulookup *, sizeof(*dul), M_TEMP, M_WAITOK);
5330 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
5331
5332 error = busyerror = nfs_node_set_busy(dnp, vfs_context_thread(ctx));
5333 if (!namedattrs) {
5334 nfs_dulookup_init(dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
5335 }
5336
5337 nfsm_chain_null(&nmreq);
5338 nfsm_chain_null(&nmrep);
5339
5340 nfsm_chain_build_alloc_init(error, &nmreq,
5341 NFSX_FH(nfsvers) + NFSX_UNSIGNED +
5342 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(nfsvers));
5343 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
5344 nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
5345 if (nfsvers == NFS_VER3) {
5346 nfsm_chain_add_v3sattr(nmp, error, &nmreq, vap);
5347 } else {
5348 nfsm_chain_add_v2sattr(error, &nmreq, vap, -1);
5349 }
5350 nfsm_chain_build_done(error, &nmreq);
5351 nfsmout_if(error);
5352
5353 error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_MKDIR,
5354 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
5355 if (!error) {
5356 if (!namedattrs) {
5357 nfs_dulookup_start(dul, dnp, ctx);
5358 dul_in_progress = 1;
5359 }
5360 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
5361 }
5362
5363 if ((lockerror = nfs_node_lock(dnp))) {
5364 error = lockerror;
5365 }
5366 dxid = xid;
5367 if (!error && !status) {
5368 if (dnp->n_flag & NNEGNCENTRIES) {
5369 dnp->n_flag &= ~NNEGNCENTRIES;
5370 cache_purge_negatives(dvp);
5371 }
5372 error = nfsm_chain_get_fh_attr(nmp, &nmrep, dnp, ctx, nfsvers, &xid, fh, nvattr);
5373 }
5374 if (nfsvers == NFS_VER3) {
5375 nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &dxid);
5376 }
5377 if (!error) {
5378 error = status;
5379 }
5380 nfsmout:
5381 nfsm_chain_cleanup(&nmreq);
5382 nfsm_chain_cleanup(&nmrep);
5383
5384 if (!lockerror) {
5385 dnp->n_flag |= NMODIFIED;
5386 /* if directory hadn't changed, update namecache mtime */
5387 if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) {
5388 NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
5389 }
5390 nfs_node_unlock(dnp);
5391 /* nfs_getattr() will check changed and purge caches */
5392 nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
5393 }
5394
5395 if (!error && fh->fh_len) {
5396 error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh->fh_data, fh->fh_len, nvattr, &xid, req->r_auth, NG_MAKEENTRY, &np);
5397 }
5398 if (!error && np) {
5399 newvp = NFSTOV(np);
5400 }
5401
5402 if (dul_in_progress) {
5403 nfs_dulookup_finish(dul, dnp, ctx);
5404 }
5405
5406 /*
5407 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
5408 * if we can succeed in looking up the directory.
5409 */
5410 if ((error == EEXIST) || (!error && !newvp)) {
5411 if (newvp) {
5412 nfs_node_unlock(np);
5413 vnode_put(newvp);
5414 newvp = NULL;
5415 }
5416 error = nfs_lookitup(dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx, &np);
5417 if (!error) {
5418 newvp = NFSTOV(np);
5419 if (vnode_vtype(newvp) != VDIR) {
5420 error = EEXIST;
5421 }
5422 }
5423 }
5424 if (!busyerror) {
5425 nfs_node_clear_busy(dnp);
5426 }
5427 if (!error && (gotuid || gotgid) &&
5428 (!newvp || nfs_getattrcache(np, nvattr, 0) ||
5429 (gotuid && (nvattr->nva_uid != vap->va_uid)) ||
5430 (gotgid && (nvattr->nva_gid != vap->va_gid)))) {
5431 /* clear ID bits if server didn't use them (or we can't tell) */
5432 VATTR_CLEAR_SUPPORTED(vap, va_uid);
5433 VATTR_CLEAR_SUPPORTED(vap, va_gid);
5434 }
5435 if (error) {
5436 if (newvp) {
5437 nfs_node_unlock(np);
5438 vnode_put(newvp);
5439 }
5440 } else {
5441 nfs_node_unlock(np);
5442 *ap->a_vpp = newvp;
5443 }
5444 NFS_ZFREE(nfs_fhandle_zone, fh);
5445 NFS_ZFREE(nfs_req_zone, req);
5446 FREE(dul, M_TEMP);
5447 FREE(nvattr, M_TEMP);
5448 return error;
5449 }
5450
5451 /*
5452 * NFS remove directory call
5453 */
5454 int
5455 nfs3_vnop_rmdir(
5456 struct vnop_rmdir_args /* {
5457 * struct vnodeop_desc *a_desc;
5458 * vnode_t a_dvp;
5459 * vnode_t a_vp;
5460 * struct componentname *a_cnp;
5461 * vfs_context_t a_context;
5462 * } */*ap)
5463 {
5464 vfs_context_t ctx = ap->a_context;
5465 vnode_t vp = ap->a_vp;
5466 vnode_t dvp = ap->a_dvp;
5467 struct componentname *cnp = ap->a_cnp;
5468 int error = 0, lockerror = ENOENT, status = 0, wccpostattr = 0;
5469 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
5470 struct nfsmount *nmp;
5471 nfsnode_t np = VTONFS(vp);
5472 nfsnode_t dnp = VTONFS(dvp);
5473 int nfsvers;
5474 u_int64_t xid;
5475 struct nfsm_chain nmreq, nmrep;
5476 struct nfsreq *req;
5477 struct nfs_dulookup *dul;
5478 int namedattrs;
5479 int dul_in_progress = 0;
5480
5481 nmp = VTONMP(vp);
5482 if (nfs_mount_gone(nmp)) {
5483 return ENXIO;
5484 }
5485 nfsvers = nmp->nm_vers;
5486 namedattrs = (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR);
5487
5488 if ((nfsvers == NFS_VER2) && (cnp->cn_namelen > NFS_MAXNAMLEN)) {
5489 return ENAMETOOLONG;
5490 }
5491
5492 if ((error = nfs_node_set_busy2(dnp, np, vfs_context_thread(ctx)))) {
5493 return error;
5494 }
5495
5496 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
5497 MALLOC(dul, struct nfs_dulookup *, sizeof(*dul), M_TEMP, M_WAITOK);
5498
5499 if (!namedattrs) {
5500 nfs_dulookup_init(dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
5501 }
5502
5503 nfsm_chain_null(&nmreq);
5504 nfsm_chain_null(&nmrep);
5505
5506 nfsm_chain_build_alloc_init(error, &nmreq,
5507 NFSX_FH(nfsvers) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
5508 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
5509 nfsm_chain_add_name(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen, nmp);
5510 nfsm_chain_build_done(error, &nmreq);
5511 nfsmout_if(error);
5512
5513 error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_RMDIR,
5514 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, &req);
5515 if (!error) {
5516 if (!namedattrs) {
5517 nfs_dulookup_start(dul, dnp, ctx);
5518 dul_in_progress = 1;
5519 }
5520 error = nfs_request_async_finish(req, &nmrep, &xid, &status);
5521 }
5522
5523 if ((lockerror = nfs_node_lock(dnp))) {
5524 error = lockerror;
5525 }
5526 if (nfsvers == NFS_VER3) {
5527 nfsm_chain_get_wcc_data(error, &nmrep, dnp, &premtime, &wccpostattr, &xid);
5528 }
5529 if (!error) {
5530 error = status;
5531 }
5532 nfsmout:
5533 nfsm_chain_cleanup(&nmreq);
5534 nfsm_chain_cleanup(&nmrep);
5535
5536 if (!lockerror) {
5537 dnp->n_flag |= NMODIFIED;
5538 /* if directory hadn't changed, update namecache mtime */
5539 if (nfstimespeccmp(&dnp->n_ncmtime, &premtime, ==)) {
5540 NFS_CHANGED_UPDATE_NC(nfsvers, dnp, &dnp->n_vattr);
5541 }
5542 nfs_node_unlock(dnp);
5543 nfs_name_cache_purge(dnp, np, cnp, ctx);
5544 /* nfs_getattr() will check changed and purge caches */
5545 nfs_getattr(dnp, NULL, ctx, wccpostattr ? NGA_CACHED : NGA_UNCACHED);
5546 }
5547 if (dul_in_progress) {
5548 nfs_dulookup_finish(dul, dnp, ctx);
5549 }
5550 nfs_node_clear_busy2(dnp, np);
5551
5552 /*
5553 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
5554 */
5555 if (error == ENOENT) {
5556 error = 0;
5557 }
5558 if (!error) {
5559 /*
5560 * remove nfsnode from hash now so we can't accidentally find it
5561 * again if another object gets created with the same filehandle
5562 * before this vnode gets reclaimed
5563 */
5564 lck_mtx_lock(&nfs_node_hash_mutex);
5565 if (np->n_hflag & NHHASHED) {
5566 LIST_REMOVE(np, n_hash);
5567 np->n_hflag &= ~NHHASHED;
5568 FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
5569 }
5570 lck_mtx_unlock(&nfs_node_hash_mutex);
5571 }
5572 NFS_ZFREE(nfs_req_zone, req);
5573 FREE(dul, M_TEMP);
5574 return error;
5575 }
5576
5577 /*
5578 * NFS readdir call
5579 *
5580 * The incoming "offset" is a directory cookie indicating where in the
5581 * directory entries should be read from. A zero cookie means start at
5582 * the beginning of the directory. Any other cookie will be a cookie
5583 * returned from the server.
5584 *
5585 * Using that cookie, determine which buffer (and where in that buffer)
5586 * to start returning entries from. Buffer logical block numbers are
5587 * the cookies they start at. If a buffer is found that is not full,
5588 * call into the bio/RPC code to fill it. The RPC code will probably
5589 * fill several buffers (dropping the first, requiring a re-get).
5590 *
5591 * When done copying entries to the buffer, set the offset to the current
5592 * entry's cookie and enter that cookie in the cookie cache.
5593 *
5594 * Note: because the getdirentries(2) API returns a long-typed offset,
5595 * the incoming offset is a potentially truncated cookie (ptc).
5596 * The cookie matching code is aware of this and will fall back to
5597 * matching only 32 bits of the cookie.
5598 */
5599 int
5600 nfs_vnop_readdir(
5601 struct vnop_readdir_args /* {
5602 * struct vnodeop_desc *a_desc;
5603 * vnode_t a_vp;
5604 * struct uio *a_uio;
5605 * int a_flags;
5606 * int *a_eofflag;
5607 * int *a_numdirent;
5608 * vfs_context_t a_context;
5609 * } */*ap)
5610 {
5611 vfs_context_t ctx = ap->a_context;
5612 vnode_t dvp = ap->a_vp;
5613 nfsnode_t dnp = VTONFS(dvp);
5614 struct nfsmount *nmp;
5615 uio_t uio = ap->a_uio;
5616 int error, nfsvers, extended, numdirent, bigcookies, ptc, done;
5617 long attrcachetimeout;
5618 uint16_t i, iptc, rlen, nlen;
5619 uint64_t cookie, nextcookie, lbn = 0;
5620 struct nfsbuf *bp = NULL;
5621 struct nfs_dir_buf_header *ndbhp;
5622 struct direntry *dp, *dpptc;
5623 struct dirent dent;
5624 char *cp = NULL;
5625 struct timeval now;
5626 thread_t thd;
5627
5628 nmp = VTONMP(dvp);
5629 if (nfs_mount_gone(nmp)) {
5630 return ENXIO;
5631 }
5632 nfsvers = nmp->nm_vers;
5633 bigcookies = (nmp->nm_state & NFSSTA_BIGCOOKIES);
5634 extended = (ap->a_flags & VNODE_READDIR_EXTENDED);
5635
5636 if (vnode_vtype(dvp) != VDIR) {
5637 return EPERM;
5638 }
5639
5640 if (ap->a_eofflag) {
5641 *ap->a_eofflag = 0;
5642 }
5643
5644 if (uio_resid(uio) == 0) {
5645 return 0;
5646 }
5647 #if CONFIG_NFS4
5648 if ((nfsvers >= NFS_VER4) && (dnp->n_vattr.nva_flags & NFS_FFLAG_TRIGGER)) {
5649 /* trigger directories should never be read, return nothing */
5650 return 0;
5651 }
5652 #endif
5653 thd = vfs_context_thread(ctx);
5654 numdirent = done = 0;
5655 nextcookie = uio_offset(uio);
5656 ptc = bigcookies && NFS_DIR_COOKIE_POTENTIALLY_TRUNCATED(nextcookie);
5657
5658 if ((error = nfs_node_lock(dnp))) {
5659 goto out;
5660 }
5661
5662 if (dnp->n_flag & NNEEDINVALIDATE) {
5663 dnp->n_flag &= ~NNEEDINVALIDATE;
5664 nfs_invaldir(dnp);
5665 nfs_node_unlock(dnp);
5666 error = nfs_vinvalbuf(dvp, 0, ctx, 1);
5667 if (!error) {
5668 error = nfs_node_lock(dnp);
5669 }
5670 if (error) {
5671 goto out;
5672 }
5673 }
5674
5675 if (dnp->n_rdirplusstamp_eof && dnp->n_rdirplusstamp_sof) {
5676 attrcachetimeout = nfs_attrcachetimeout(dnp);
5677 microuptime(&now);
5678 if (attrcachetimeout && (now.tv_sec - dnp->n_rdirplusstamp_sof > attrcachetimeout - 1)) {
5679 dnp->n_rdirplusstamp_eof = dnp->n_rdirplusstamp_sof = 0;
5680 nfs_invaldir(dnp);
5681 nfs_node_unlock(dnp);
5682 error = nfs_vinvalbuf(dvp, 0, ctx, 1);
5683 if (!error) {
5684 error = nfs_node_lock(dnp);
5685 }
5686 if (error) {
5687 goto out;
5688 }
5689 }
5690 }
5691
5692 /*
5693 * check for need to invalidate when (re)starting at beginning
5694 */
5695 if (!nextcookie) {
5696 if (dnp->n_flag & NMODIFIED) {
5697 nfs_invaldir(dnp);
5698 nfs_node_unlock(dnp);
5699 if ((error = nfs_vinvalbuf(dvp, 0, ctx, 1))) {
5700 goto out;
5701 }
5702 } else {
5703 nfs_node_unlock(dnp);
5704 }
5705 /* nfs_getattr() will check changed and purge caches */
5706 if ((error = nfs_getattr(dnp, NULL, ctx, NGA_UNCACHED))) {
5707 goto out;
5708 }
5709 } else {
5710 nfs_node_unlock(dnp);
5711 }
5712
5713 error = nfs_dir_cookie_to_lbn(dnp, nextcookie, &ptc, &lbn);
5714 if (error) {
5715 if (error < 0) { /* just hit EOF cookie */
5716 done = 1;
5717 error = 0;
5718 }
5719 if (ap->a_eofflag) {
5720 *ap->a_eofflag = 1;
5721 }
5722 }
5723
5724 while (!error && !done) {
5725 OSAddAtomic64(1, &nfsstats.biocache_readdirs);
5726 cookie = nextcookie;
5727 getbuffer:
5728 error = nfs_buf_get(dnp, lbn, NFS_DIRBLKSIZ, thd, NBLK_READ, &bp);
5729 if (error) {
5730 goto out;
5731 }
5732 ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
5733 if (!ISSET(bp->nb_flags, NB_CACHE) || !ISSET(ndbhp->ndbh_flags, NDB_FULL)) {
5734 if (!ISSET(bp->nb_flags, NB_CACHE)) { /* initialize the buffer */
5735 ndbhp->ndbh_flags = 0;
5736 ndbhp->ndbh_count = 0;
5737 ndbhp->ndbh_entry_end = sizeof(*ndbhp);
5738 ndbhp->ndbh_ncgen = dnp->n_ncgen;
5739 }
5740 error = nfs_buf_readdir(bp, ctx);
5741 if (error == NFSERR_DIRBUFDROPPED) {
5742 goto getbuffer;
5743 }
5744 if (error) {
5745 nfs_buf_release(bp, 1);
5746 }
5747 if (error && (error != ENXIO) && (error != ETIMEDOUT) && (error != EINTR) && (error != ERESTART)) {
5748 if (!nfs_node_lock(dnp)) {
5749 nfs_invaldir(dnp);
5750 nfs_node_unlock(dnp);
5751 }
5752 nfs_vinvalbuf(dvp, 0, ctx, 1);
5753 if (error == NFSERR_BAD_COOKIE) {
5754 error = ENOENT;
5755 }
5756 }
5757 if (error) {
5758 goto out;
5759 }
5760 }
5761
5762 /* find next entry to return */
5763 dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
5764 i = 0;
5765 if ((lbn != cookie) && !(ptc && NFS_DIR_COOKIE_SAME32(lbn, cookie))) {
5766 dpptc = NULL;
5767 iptc = 0;
5768 for (; (i < ndbhp->ndbh_count) && (cookie != dp->d_seekoff); i++) {
5769 if (ptc && !dpptc && NFS_DIR_COOKIE_SAME32(cookie, dp->d_seekoff)) {
5770 iptc = i;
5771 dpptc = dp;
5772 }
5773 nextcookie = dp->d_seekoff;
5774 dp = NFS_DIRENTRY_NEXT(dp);
5775 }
5776 if ((i == ndbhp->ndbh_count) && dpptc) {
5777 i = iptc;
5778 dp = dpptc;
5779 }
5780 if (i < ndbhp->ndbh_count) {
5781 nextcookie = dp->d_seekoff;
5782 dp = NFS_DIRENTRY_NEXT(dp);
5783 i++;
5784 }
5785 }
5786 ptc = 0; /* only have to deal with ptc on first cookie */
5787
5788 /* return as many entries as we can */
5789 for (; i < ndbhp->ndbh_count; i++) {
5790 if (extended) {
5791 rlen = dp->d_reclen;
5792 cp = (char*)dp;
5793 } else {
5794 if (!cp) {
5795 cp = (char*)&dent;
5796 bzero(cp, sizeof(dent));
5797 }
5798 if (dp->d_namlen > (sizeof(dent.d_name) - 1)) {
5799 nlen = sizeof(dent.d_name) - 1;
5800 } else {
5801 nlen = dp->d_namlen;
5802 }
5803 rlen = NFS_DIRENT_LEN(nlen);
5804 dent.d_reclen = rlen;
5805 dent.d_ino = (ino_t)dp->d_ino;
5806 dent.d_type = dp->d_type;
5807 dent.d_namlen = (uint8_t)nlen;
5808 strlcpy(dent.d_name, dp->d_name, nlen + 1);
5809 }
5810 /* check that the record fits */
5811 if (rlen > uio_resid(uio)) {
5812 done = 1;
5813 break;
5814 }
5815 if ((error = uiomove(cp, rlen, uio))) {
5816 break;
5817 }
5818 numdirent++;
5819 nextcookie = dp->d_seekoff;
5820 dp = NFS_DIRENTRY_NEXT(dp);
5821 }
5822
5823 if (i == ndbhp->ndbh_count) {
5824 /* hit end of buffer, move to next buffer */
5825 lbn = nextcookie;
5826 /* if we also hit EOF, we're done */
5827 if (ISSET(ndbhp->ndbh_flags, NDB_EOF)) {
5828 done = 1;
5829 if (ap->a_eofflag) {
5830 *ap->a_eofflag = 1;
5831 }
5832 }
5833 }
5834 if (!error) {
5835 uio_setoffset(uio, nextcookie);
5836 }
5837 if (!error && !done && (nextcookie == cookie)) {
5838 printf("nfs readdir cookie didn't change 0x%llx, %d/%d\n", cookie, i, ndbhp->ndbh_count);
5839 error = EIO;
5840 }
5841 nfs_buf_release(bp, 1);
5842 }
5843
5844 if (!error) {
5845 nfs_dir_cookie_cache(dnp, nextcookie, lbn);
5846 }
5847
5848 if (ap->a_numdirent) {
5849 *ap->a_numdirent = numdirent;
5850 }
5851 out:
5852 return error;
5853 }
5854
5855
5856 /*
5857 * Invalidate cached directory information, except for the actual directory
5858 * blocks (which are invalidated separately).
5859 */
5860 static void
5861 nfs_invaldir_cookies(nfsnode_t dnp)
5862 {
5863 if (vnode_vtype(NFSTOV(dnp)) != VDIR) {
5864 return;
5865 }
5866 dnp->n_eofcookie = 0;
5867 dnp->n_cookieverf = 0;
5868 if (!dnp->n_cookiecache) {
5869 return;
5870 }
5871 dnp->n_cookiecache->free = 0;
5872 dnp->n_cookiecache->mru = -1;
5873 memset(dnp->n_cookiecache->next, -1, NFSNUMCOOKIES);
5874 }
5875
5876 void
5877 nfs_invaldir(nfsnode_t dnp)
5878 {
5879
5880 nfs_invaldir_cookies(dnp);
5881 }
5882
5883 /*
5884 * calculate how much space is available for additional directory entries.
5885 */
5886 uint64_t
5887 nfs_dir_buf_freespace(struct nfsbuf *bp, int rdirplus)
5888 {
5889 struct nfs_dir_buf_header *ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
5890 uint64_t space;
5891
5892 if (!ndbhp) {
5893 return 0;
5894 }
5895 space = bp->nb_bufsize - ndbhp->ndbh_entry_end;
5896 if (rdirplus) {
5897 space -= ndbhp->ndbh_count * sizeof(struct nfs_vattr);
5898 }
5899 return space;
5900 }
5901
5902 /*
5903 * add/update a cookie->lbn entry in the directory cookie cache
5904 */
5905 void
5906 nfs_dir_cookie_cache(nfsnode_t dnp, uint64_t cookie, uint64_t lbn)
5907 {
5908 struct nfsdmap *ndcc;
5909 int8_t i, prev;
5910
5911 if (!cookie) {
5912 return;
5913 }
5914
5915 if (nfs_node_lock(dnp)) {
5916 return;
5917 }
5918
5919 if (cookie == dnp->n_eofcookie) { /* EOF cookie */
5920 nfs_node_unlock(dnp);
5921 return;
5922 }
5923
5924 ndcc = dnp->n_cookiecache;
5925 if (!ndcc) {
5926 /* allocate the cookie cache structure */
5927 ndcc = dnp->n_cookiecache = zalloc(ZV_NFSDIROFF);
5928 ndcc->free = 0;
5929 ndcc->mru = -1;
5930 memset(ndcc->next, -1, NFSNUMCOOKIES);
5931 }
5932
5933 /*
5934 * Search the list for this cookie.
5935 * Keep track of previous and last entries.
5936 */
5937 prev = -1;
5938 i = ndcc->mru;
5939 while ((i != -1) && (cookie != ndcc->cookies[i].key)) {
5940 if (ndcc->next[i] == -1) { /* stop on last entry so we can reuse */
5941 break;
5942 }
5943 prev = i;
5944 i = ndcc->next[i];
5945 }
5946 if ((i != -1) && (cookie == ndcc->cookies[i].key)) {
5947 /* found it, remove from list */
5948 if (prev != -1) {
5949 ndcc->next[prev] = ndcc->next[i];
5950 } else {
5951 ndcc->mru = ndcc->next[i];
5952 }
5953 } else {
5954 /* not found, use next free entry or reuse last entry */
5955 if (ndcc->free != NFSNUMCOOKIES) {
5956 i = ndcc->free++;
5957 } else {
5958 ndcc->next[prev] = -1;
5959 }
5960 ndcc->cookies[i].key = cookie;
5961 ndcc->cookies[i].lbn = lbn;
5962 }
5963 /* insert cookie at head of MRU list */
5964 ndcc->next[i] = ndcc->mru;
5965 ndcc->mru = i;
5966 nfs_node_unlock(dnp);
5967 }
5968
5969 /*
5970 * Try to map the given directory cookie to a directory buffer (return lbn).
5971 * If we have a possibly truncated cookie (ptc), check for 32-bit matches too.
5972 */
5973 int
5974 nfs_dir_cookie_to_lbn(nfsnode_t dnp, uint64_t cookie, int *ptc, uint64_t *lbnp)
5975 {
5976 struct nfsdmap *ndcc = dnp->n_cookiecache;
5977 int8_t eofptc, found;
5978 int i, iptc;
5979 struct nfsmount *nmp;
5980 struct nfsbuf *bp, *lastbp;
5981 struct nfsbuflists blist;
5982 struct direntry *dp, *dpptc;
5983 struct nfs_dir_buf_header *ndbhp;
5984
5985 if (!cookie) { /* initial cookie */
5986 *lbnp = 0;
5987 *ptc = 0;
5988 return 0;
5989 }
5990
5991 if (nfs_node_lock(dnp)) {
5992 return ENOENT;
5993 }
5994
5995 if (cookie == dnp->n_eofcookie) { /* EOF cookie */
5996 nfs_node_unlock(dnp);
5997 OSAddAtomic64(1, &nfsstats.direofcache_hits);
5998 *ptc = 0;
5999 return -1;
6000 }
6001 /* note if cookie is a 32-bit match with the EOF cookie */
6002 eofptc = *ptc ? NFS_DIR_COOKIE_SAME32(cookie, dnp->n_eofcookie) : 0;
6003 iptc = -1;
6004
6005 /* search the list for the cookie */
6006 for (i = ndcc ? ndcc->mru : -1; i >= 0; i = ndcc->next[i]) {
6007 if (ndcc->cookies[i].key == cookie) {
6008 /* found a match for this cookie */
6009 *lbnp = ndcc->cookies[i].lbn;
6010 nfs_node_unlock(dnp);
6011 OSAddAtomic64(1, &nfsstats.direofcache_hits);
6012 *ptc = 0;
6013 return 0;
6014 }
6015 /* check for 32-bit match */
6016 if (*ptc && (iptc == -1) && NFS_DIR_COOKIE_SAME32(ndcc->cookies[i].key, cookie)) {
6017 iptc = i;
6018 }
6019 }
6020 /* exact match not found */
6021 if (eofptc) {
6022 /* but 32-bit match hit the EOF cookie */
6023 nfs_node_unlock(dnp);
6024 OSAddAtomic64(1, &nfsstats.direofcache_hits);
6025 return -1;
6026 }
6027 if (iptc >= 0) {
6028 /* but 32-bit match got a hit */
6029 *lbnp = ndcc->cookies[iptc].lbn;
6030 nfs_node_unlock(dnp);
6031 OSAddAtomic64(1, &nfsstats.direofcache_hits);
6032 return 0;
6033 }
6034 nfs_node_unlock(dnp);
6035
6036 /*
6037 * No match found in the cookie cache... hmm...
6038 * Let's search the directory's buffers for the cookie.
6039 */
6040 nmp = NFSTONMP(dnp);
6041 if (nfs_mount_gone(nmp)) {
6042 return ENXIO;
6043 }
6044 dpptc = NULL;
6045 found = 0;
6046
6047 lck_mtx_lock(&nfs_buf_mutex);
6048 /*
6049 * Scan the list of buffers, keeping them in order.
6050 * Note that itercomplete inserts each of the remaining buffers
6051 * into the head of list (thus reversing the elements). So, we
6052 * make sure to iterate through all buffers, inserting them after
6053 * each other, to keep them in order.
6054 * Also note: the LIST_INSERT_AFTER(lastbp) is only safe because
6055 * we don't drop nfs_buf_mutex.
6056 */
6057 if (!nfs_buf_iterprepare(dnp, &blist, NBI_CLEAN)) {
6058 lastbp = NULL;
6059 while ((bp = LIST_FIRST(&blist))) {
6060 LIST_REMOVE(bp, nb_vnbufs);
6061 if (!lastbp) {
6062 LIST_INSERT_HEAD(&dnp->n_cleanblkhd, bp, nb_vnbufs);
6063 } else {
6064 LIST_INSERT_AFTER(lastbp, bp, nb_vnbufs);
6065 }
6066 lastbp = bp;
6067 if (found) {
6068 continue;
6069 }
6070 nfs_buf_refget(bp);
6071 if (nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0)) {
6072 /* just skip this buffer */
6073 nfs_buf_refrele(bp);
6074 continue;
6075 }
6076 nfs_buf_refrele(bp);
6077
6078 /* scan the buffer for the cookie */
6079 ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
6080 dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
6081 dpptc = NULL;
6082 for (i = 0; (i < ndbhp->ndbh_count) && (cookie != dp->d_seekoff); i++) {
6083 if (*ptc && !dpptc && NFS_DIR_COOKIE_SAME32(cookie, dp->d_seekoff)) {
6084 dpptc = dp;
6085 iptc = i;
6086 }
6087 dp = NFS_DIRENTRY_NEXT(dp);
6088 }
6089 if ((i == ndbhp->ndbh_count) && dpptc) {
6090 /* found only a PTC match */
6091 dp = dpptc;
6092 i = iptc;
6093 } else if (i < ndbhp->ndbh_count) {
6094 *ptc = 0;
6095 }
6096 if (i < (ndbhp->ndbh_count - 1)) {
6097 /* next entry is *in* this buffer: return this block */
6098 *lbnp = bp->nb_lblkno;
6099 found = 1;
6100 } else if (i == (ndbhp->ndbh_count - 1)) {
6101 /* next entry refers to *next* buffer: return next block */
6102 *lbnp = dp->d_seekoff;
6103 found = 1;
6104 }
6105 nfs_buf_drop(bp);
6106 }
6107 nfs_buf_itercomplete(dnp, &blist, NBI_CLEAN);
6108 }
6109 lck_mtx_unlock(&nfs_buf_mutex);
6110 if (found) {
6111 OSAddAtomic64(1, &nfsstats.direofcache_hits);
6112 return 0;
6113 }
6114
6115 /* still not found... oh well, just start a new block */
6116 *lbnp = cookie;
6117 OSAddAtomic64(1, &nfsstats.direofcache_misses);
6118 return 0;
6119 }
6120
6121 /*
6122 * scan a directory buffer for the given name
6123 * Returns: ESRCH if not found, ENOENT if found invalid, 0 if found
6124 * Note: should only be called with RDIRPLUS directory buffers
6125 */
6126
6127 #define NDBS_PURGE 1
6128 #define NDBS_UPDATE 2
6129
6130 int
6131 nfs_dir_buf_search(
6132 struct nfsbuf *bp,
6133 struct componentname *cnp,
6134 fhandle_t *fhp,
6135 struct nfs_vattr *nvap,
6136 uint64_t *xidp,
6137 time_t *attrstampp,
6138 daddr64_t *nextlbnp,
6139 int flags)
6140 {
6141 struct direntry *dp;
6142 struct nfs_dir_buf_header *ndbhp;
6143 struct nfs_vattr *nvattrp;
6144 daddr64_t nextlbn = 0;
6145 int i, error = ESRCH;
6146 uint32_t fhlen;
6147
6148 /* scan the buffer for the name */
6149 ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
6150 dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
6151 for (i = 0; i < ndbhp->ndbh_count; i++) {
6152 nextlbn = dp->d_seekoff;
6153 if ((cnp->cn_namelen == dp->d_namlen) && !strcmp(cnp->cn_nameptr, dp->d_name)) {
6154 fhlen = (uint8_t)dp->d_name[dp->d_namlen + 1];
6155 nvattrp = NFS_DIR_BUF_NVATTR(bp, i);
6156 if ((ndbhp->ndbh_ncgen != bp->nb_np->n_ncgen) || (fhlen == 0) ||
6157 (nvattrp->nva_type == VNON) || (nvattrp->nva_fileid == 0)) {
6158 /* entry is not valid */
6159 error = ENOENT;
6160 break;
6161 }
6162 if (flags == NDBS_PURGE) {
6163 dp->d_fileno = 0;
6164 bzero(nvattrp, sizeof(*nvattrp));
6165 error = ENOENT;
6166 break;
6167 }
6168 if (flags == NDBS_UPDATE) {
6169 /* update direntry's attrs if fh matches */
6170 if ((fhp->fh_len == fhlen) && !bcmp(&dp->d_name[dp->d_namlen + 2], fhp->fh_data, fhlen)) {
6171 bcopy(nvap, nvattrp, sizeof(*nvap));
6172 dp->d_fileno = nvattrp->nva_fileid;
6173 nvattrp->nva_fileid = *xidp;
6174 nvap->nva_flags |= NFS_FFLAG_FILEID_CONTAINS_XID;
6175 *(time_t*)(&dp->d_name[dp->d_namlen + 2 + fhp->fh_len]) = *attrstampp;
6176 }
6177 error = 0;
6178 break;
6179 }
6180 /* copy out fh, attrs, attrstamp, and xid */
6181 fhp->fh_len = fhlen;
6182 bcopy(&dp->d_name[dp->d_namlen + 2], fhp->fh_data, MAX(fhp->fh_len, (int)sizeof(fhp->fh_data)));
6183 *attrstampp = *(time_t*)(&dp->d_name[dp->d_namlen + 2 + fhp->fh_len]);
6184 bcopy(nvattrp, nvap, sizeof(*nvap));
6185 *xidp = nvap->nva_fileid;
6186 nvap->nva_fileid = dp->d_fileno;
6187 nvap->nva_flags &= ~NFS_FFLAG_FILEID_CONTAINS_XID;
6188 error = 0;
6189 break;
6190 }
6191 dp = NFS_DIRENTRY_NEXT(dp);
6192 }
6193 if (nextlbnp) {
6194 *nextlbnp = nextlbn;
6195 }
6196 return error;
6197 }
6198
6199 /*
6200 * Look up a name in a directory's buffers.
6201 * Note: should only be called with RDIRPLUS directory buffers
6202 */
6203 int
6204 nfs_dir_buf_cache_lookup(nfsnode_t dnp, nfsnode_t *npp, struct componentname *cnp, vfs_context_t ctx, int purge, int *skipdu)
6205 {
6206 nfsnode_t newnp;
6207 struct nfsmount *nmp;
6208 int error = 0, i, found = 0, count = 0;
6209 u_int64_t xid;
6210 struct nfs_vattr *nvattr;
6211 fhandle_t *fh;
6212 time_t attrstamp = 0;
6213 thread_t thd = vfs_context_thread(ctx);
6214 struct nfsbuf *bp, *lastbp, *foundbp;
6215 struct nfsbuflists blist;
6216 daddr64_t lbn, nextlbn;
6217 int dotunder = (cnp->cn_namelen > 2) && (cnp->cn_nameptr[0] == '.') && (cnp->cn_nameptr[1] == '_');
6218 int isdot = (cnp->cn_namelen == 1) && (cnp->cn_nameptr[0] == '.');
6219 int isdotdot = (cnp->cn_namelen == 2) && (cnp->cn_nameptr[0] == '.') && (cnp->cn_nameptr[1] == '.');
6220 int eof = 0, sof = 0, skipped = 0;
6221
6222 nmp = NFSTONMP(dnp);
6223 if (nfs_mount_gone(nmp)) {
6224 return ENXIO;
6225 }
6226 if (!purge) {
6227 *npp = NULL;
6228 }
6229
6230 if (isdot || isdotdot) {
6231 return 0;
6232 }
6233
6234 fh = zalloc(nfs_fhandle_zone);
6235 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
6236
6237 /* first check most recent buffer (and next one too) */
6238 lbn = dnp->n_lastdbl;
6239 for (i = 0; i < 2; i++) {
6240 if ((error = nfs_buf_get(dnp, lbn, NFS_DIRBLKSIZ, thd, NBLK_READ | NBLK_ONLYVALID, &bp))) {
6241 goto out;
6242 }
6243 if (!bp) {
6244 skipped = 1;
6245 break;
6246 }
6247 count++;
6248 nfs_dir_buf_cache_lookup_boundaries(bp, &sof, &eof);
6249 error = nfs_dir_buf_search(bp, cnp, fh, nvattr, &xid, &attrstamp, &nextlbn, purge ? NDBS_PURGE : 0);
6250 nfs_buf_release(bp, 0);
6251 if (error == ESRCH) {
6252 error = 0;
6253 } else {
6254 found = 1;
6255 break;
6256 }
6257 lbn = nextlbn;
6258 }
6259
6260 lck_mtx_lock(&nfs_buf_mutex);
6261 if (found) {
6262 dnp->n_lastdbl = lbn;
6263 goto done;
6264 }
6265
6266 /* If we detect that we fetched full directory listing we should avoid sending lookups for ._ files */
6267 if (dotunder && !found && !error && eof && sof && !skipped && skipdu) {
6268 *skipdu = 1;
6269 }
6270
6271 /*
6272 * Scan the list of buffers, keeping them in order.
6273 * Note that itercomplete inserts each of the remaining buffers
6274 * into the head of list (thus reversing the elements). So, we
6275 * make sure to iterate through all buffers, inserting them after
6276 * each other, to keep them in order.
6277 * Also note: the LIST_INSERT_AFTER(lastbp) is only safe because
6278 * we don't drop nfs_buf_mutex.
6279 */
6280 eof = sof = skipped = 0;
6281 if (!nfs_buf_iterprepare(dnp, &blist, NBI_CLEAN)) {
6282 lastbp = foundbp = NULL;
6283 while ((bp = LIST_FIRST(&blist))) {
6284 LIST_REMOVE(bp, nb_vnbufs);
6285 if (!lastbp) {
6286 LIST_INSERT_HEAD(&dnp->n_cleanblkhd, bp, nb_vnbufs);
6287 } else {
6288 LIST_INSERT_AFTER(lastbp, bp, nb_vnbufs);
6289 }
6290 lastbp = bp;
6291 if (error || found) {
6292 skipped = 1;
6293 continue;
6294 }
6295 if (!purge && dotunder && (count > 100)) { /* don't waste too much time looking for ._ files */
6296 skipped = 1;
6297 continue;
6298 }
6299 nfs_buf_refget(bp);
6300 lbn = bp->nb_lblkno;
6301 if (nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0)) {
6302 /* just skip this buffer */
6303 nfs_buf_refrele(bp);
6304 skipped = 1;
6305 continue;
6306 }
6307 nfs_buf_refrele(bp);
6308 count++;
6309 nfs_dir_buf_cache_lookup_boundaries(bp, &sof, &eof);
6310 error = nfs_dir_buf_search(bp, cnp, fh, nvattr, &xid, &attrstamp, NULL, purge ? NDBS_PURGE : 0);
6311 if (error == ESRCH) {
6312 error = 0;
6313 } else {
6314 found = 1;
6315 foundbp = bp;
6316 }
6317 nfs_buf_drop(bp);
6318 }
6319 if (found) {
6320 LIST_REMOVE(foundbp, nb_vnbufs);
6321 LIST_INSERT_HEAD(&dnp->n_cleanblkhd, foundbp, nb_vnbufs);
6322 dnp->n_lastdbl = foundbp->nb_lblkno;
6323 }
6324 nfs_buf_itercomplete(dnp, &blist, NBI_CLEAN);
6325 }
6326
6327 /* If we detect that we fetched full directory listing we should avoid sending lookups for ._ files */
6328 if (dotunder && !found && !error && eof && sof && !skipped && skipdu) {
6329 *skipdu = 1;
6330 }
6331
6332 done:
6333 lck_mtx_unlock(&nfs_buf_mutex);
6334
6335 if (!error && found && !purge) {
6336 error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh->fh_data,
6337 fh->fh_len, nvattr, &xid, dnp->n_auth, NG_MAKEENTRY,
6338 &newnp);
6339 if (error) {
6340 goto out;
6341 }
6342 newnp->n_attrstamp = attrstamp;
6343 *npp = newnp;
6344 nfs_node_unlock(newnp);
6345 /* check if the dir buffer's attrs are out of date */
6346 if (!nfs_getattr(newnp, nvattr, ctx, NGA_CACHED) &&
6347 (newnp->n_attrstamp != attrstamp)) {
6348 /* they are, so update them */
6349 error = nfs_buf_get(dnp, lbn, NFS_DIRBLKSIZ, thd, NBLK_READ | NBLK_ONLYVALID, &bp);
6350 if (!error && bp) {
6351 attrstamp = newnp->n_attrstamp;
6352 xid = newnp->n_xid;
6353 nfs_dir_buf_search(bp, cnp, fh, nvattr, &xid, &attrstamp, NULL, NDBS_UPDATE);
6354 nfs_buf_release(bp, 0);
6355 }
6356 error = 0;
6357 }
6358 }
6359
6360 out:
6361 NFS_ZFREE(nfs_fhandle_zone, fh);
6362 FREE(nvattr, M_TEMP);
6363 return error;
6364 }
6365
6366 /*
6367 * Purge name cache entries for the given node.
6368 * For RDIRPLUS, also invalidate the entry in the directory's buffers.
6369 */
6370 void
6371 nfs_name_cache_purge(nfsnode_t dnp, nfsnode_t np, struct componentname *cnp, vfs_context_t ctx)
6372 {
6373 struct nfsmount *nmp = NFSTONMP(dnp);
6374
6375 cache_purge(NFSTOV(np));
6376 if (nmp && (nmp->nm_vers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) {
6377 nfs_dir_buf_cache_lookup(dnp, NULL, cnp, ctx, 1, NULL);
6378 }
6379 }
6380
6381 /*
6382 * NFS V3 readdir (plus) RPC.
6383 */
6384 int
6385 nfs3_readdir_rpc(nfsnode_t dnp, struct nfsbuf *bp, vfs_context_t ctx)
6386 {
6387 struct nfsmount *nmp;
6388 int error = 0, lockerror, nfsvers, rdirplus, bigcookies;
6389 int i, status = 0, attrflag, fhflag, more_entries = 1, eof, bp_dropped = 0;
6390 uint32_t nmreaddirsize, nmrsize;
6391 uint32_t namlen, skiplen, fhlen, xlen, attrlen;
6392 uint64_t cookie, lastcookie, xid, savedxid, fileno, space_free, space_needed;
6393 struct nfsm_chain nmreq, nmrep, nmrepsave;
6394 fhandle_t *fh;
6395 struct nfs_vattr *nvattrp;
6396 struct nfs_dir_buf_header *ndbhp;
6397 struct direntry *dp;
6398 char *padstart;
6399 struct timeval now;
6400 uint16_t reclen;
6401 size_t padlen;
6402
6403 nmp = NFSTONMP(dnp);
6404 if (nfs_mount_gone(nmp)) {
6405 return ENXIO;
6406 }
6407 nfsvers = nmp->nm_vers;
6408 nmreaddirsize = nmp->nm_readdirsize;
6409 nmrsize = nmp->nm_rsize;
6410 bigcookies = nmp->nm_state & NFSSTA_BIGCOOKIES;
6411 fh = zalloc(nfs_fhandle_zone);
6412 resend:
6413 rdirplus = ((nfsvers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) ? 1 : 0;
6414
6415 if ((lockerror = nfs_node_lock(dnp))) {
6416 NFS_ZFREE(nfs_fhandle_zone, fh);
6417 return lockerror;
6418 }
6419
6420 /* determine cookie to use, and move dp to the right offset */
6421 ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
6422 dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
6423 if (ndbhp->ndbh_count) {
6424 for (i = 0; i < ndbhp->ndbh_count - 1; i++) {
6425 dp = NFS_DIRENTRY_NEXT(dp);
6426 }
6427 cookie = dp->d_seekoff;
6428 dp = NFS_DIRENTRY_NEXT(dp);
6429 } else {
6430 cookie = bp->nb_lblkno;
6431 /* increment with every buffer read */
6432 OSAddAtomic64(1, &nfsstats.readdir_bios);
6433 }
6434 lastcookie = cookie;
6435
6436 /*
6437 * Loop around doing readdir(plus) RPCs of size nm_readdirsize until
6438 * the buffer is full (or we hit EOF). Then put the remainder of the
6439 * results in the next buffer(s).
6440 */
6441 nfsm_chain_null(&nmreq);
6442 nfsm_chain_null(&nmrep);
6443 while (nfs_dir_buf_freespace(bp, rdirplus) && !(ndbhp->ndbh_flags & NDB_FULL)) {
6444 nfsm_chain_build_alloc_init(error, &nmreq,
6445 NFSX_FH(nfsvers) + NFSX_READDIR(nfsvers) + NFSX_UNSIGNED);
6446 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
6447 if (nfsvers == NFS_VER3) {
6448 /* opaque values don't need swapping, but as long */
6449 /* as we are consistent about it, it should be ok */
6450 nfsm_chain_add_64(error, &nmreq, cookie);
6451 nfsm_chain_add_64(error, &nmreq, dnp->n_cookieverf);
6452 } else {
6453 nfsm_chain_add_32(error, &nmreq, cookie);
6454 }
6455 nfsm_chain_add_32(error, &nmreq, nmreaddirsize);
6456 if (rdirplus) {
6457 nfsm_chain_add_32(error, &nmreq, nmrsize);
6458 }
6459 nfsm_chain_build_done(error, &nmreq);
6460 nfs_node_unlock(dnp);
6461 lockerror = ENOENT;
6462 nfsmout_if(error);
6463
6464 error = nfs_request(dnp, NULL, &nmreq,
6465 rdirplus ? NFSPROC_READDIRPLUS : NFSPROC_READDIR,
6466 ctx, NULL, &nmrep, &xid, &status);
6467
6468 if ((lockerror = nfs_node_lock(dnp))) {
6469 error = lockerror;
6470 }
6471
6472 savedxid = xid;
6473 if (nfsvers == NFS_VER3) {
6474 nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid);
6475 }
6476 if (!error) {
6477 error = status;
6478 }
6479 if (nfsvers == NFS_VER3) {
6480 nfsm_chain_get_64(error, &nmrep, dnp->n_cookieverf);
6481 }
6482 nfsm_chain_get_32(error, &nmrep, more_entries);
6483
6484 if (!lockerror) {
6485 nfs_node_unlock(dnp);
6486 lockerror = ENOENT;
6487 }
6488 if (error == NFSERR_NOTSUPP) {
6489 /* oops... it doesn't look like readdirplus is supported */
6490 lck_mtx_lock(&nmp->nm_lock);
6491 NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_RDIRPLUS);
6492 lck_mtx_unlock(&nmp->nm_lock);
6493 nfsm_chain_cleanup(&nmreq);
6494 nfsm_chain_cleanup(&nmrep);
6495 goto resend;
6496 }
6497 nfsmout_if(error);
6498
6499 if (rdirplus) {
6500 microuptime(&now);
6501 if (lastcookie == 0) {
6502 dnp->n_rdirplusstamp_sof = now.tv_sec;
6503 dnp->n_rdirplusstamp_eof = 0;
6504 }
6505 }
6506
6507 /* loop through the entries packing them into the buffer */
6508 while (more_entries) {
6509 if (nfsvers == NFS_VER3) {
6510 nfsm_chain_get_64(error, &nmrep, fileno);
6511 } else {
6512 nfsm_chain_get_32(error, &nmrep, fileno);
6513 }
6514 nfsm_chain_get_32(error, &nmrep, namlen);
6515 nfsmout_if(error);
6516 /* just truncate names that don't fit in direntry.d_name */
6517 if (namlen <= 0) {
6518 error = EBADRPC;
6519 goto nfsmout;
6520 }
6521 if (namlen > (sizeof(dp->d_name) - 1)) {
6522 skiplen = namlen - sizeof(dp->d_name) + 1;
6523 namlen = sizeof(dp->d_name) - 1;
6524 } else {
6525 skiplen = 0;
6526 }
6527 /* guess that fh size will be same as parent */
6528 fhlen = rdirplus ? (1 + dnp->n_fhsize) : 0;
6529 xlen = rdirplus ? (fhlen + sizeof(time_t)) : 0;
6530 attrlen = rdirplus ? sizeof(struct nfs_vattr) : 0;
6531 reclen = NFS_DIRENTRY_LEN_16(namlen + xlen);
6532 space_needed = reclen + attrlen;
6533 space_free = nfs_dir_buf_freespace(bp, rdirplus);
6534 if (space_needed > space_free) {
6535 /*
6536 * We still have entries to pack, but we've
6537 * run out of room in the current buffer.
6538 * So we need to move to the next buffer.
6539 * The block# for the next buffer is the
6540 * last cookie in the current buffer.
6541 */
6542 nextbuffer:
6543 ndbhp->ndbh_flags |= NDB_FULL;
6544 nfs_buf_release(bp, 0);
6545 bp_dropped = 1;
6546 bp = NULL;
6547 error = nfs_buf_get(dnp, lastcookie, NFS_DIRBLKSIZ, vfs_context_thread(ctx), NBLK_READ, &bp);
6548 nfsmout_if(error);
6549 /* initialize buffer */
6550 ndbhp = (struct nfs_dir_buf_header*)bp->nb_data;
6551 ndbhp->ndbh_flags = 0;
6552 ndbhp->ndbh_count = 0;
6553 ndbhp->ndbh_entry_end = sizeof(*ndbhp);
6554 ndbhp->ndbh_ncgen = dnp->n_ncgen;
6555 space_free = nfs_dir_buf_freespace(bp, rdirplus);
6556 dp = NFS_DIR_BUF_FIRST_DIRENTRY(bp);
6557 /* increment with every buffer read */
6558 OSAddAtomic64(1, &nfsstats.readdir_bios);
6559 }
6560 nmrepsave = nmrep;
6561 dp->d_fileno = fileno;
6562 dp->d_namlen = (uint16_t)namlen;
6563 dp->d_reclen = reclen;
6564 dp->d_type = DT_UNKNOWN;
6565 nfsm_chain_get_opaque(error, &nmrep, namlen, dp->d_name);
6566 nfsmout_if(error);
6567 dp->d_name[namlen] = '\0';
6568 if (skiplen) {
6569 nfsm_chain_adv(error, &nmrep,
6570 nfsm_rndup(namlen + skiplen) - nfsm_rndup(namlen));
6571 }
6572 if (nfsvers == NFS_VER3) {
6573 nfsm_chain_get_64(error, &nmrep, cookie);
6574 } else {
6575 nfsm_chain_get_32(error, &nmrep, cookie);
6576 }
6577 nfsmout_if(error);
6578 dp->d_seekoff = cookie;
6579 if (!bigcookies && (cookie >> 32) && (nmp == NFSTONMP(dnp))) {
6580 /* we've got a big cookie, make sure flag is set */
6581 lck_mtx_lock(&nmp->nm_lock);
6582 nmp->nm_state |= NFSSTA_BIGCOOKIES;
6583 lck_mtx_unlock(&nmp->nm_lock);
6584 bigcookies = 1;
6585 }
6586 if (rdirplus) {
6587 nvattrp = NFS_DIR_BUF_NVATTR(bp, ndbhp->ndbh_count);
6588 /* check for attributes */
6589 nfsm_chain_get_32(error, &nmrep, attrflag);
6590 nfsmout_if(error);
6591 if (attrflag) {
6592 /* grab attributes */
6593 error = nfs_parsefattr(nmp, &nmrep, NFS_VER3, nvattrp);
6594 nfsmout_if(error);
6595 dp->d_type = IFTODT(VTTOIF(nvattrp->nva_type));
6596 /* fileid is already in d_fileno, so stash xid in attrs */
6597 nvattrp->nva_fileid = savedxid;
6598 nvattrp->nva_flags |= NFS_FFLAG_FILEID_CONTAINS_XID;
6599 } else {
6600 /* mark the attributes invalid */
6601 bzero(nvattrp, sizeof(struct nfs_vattr));
6602 }
6603 /* check for file handle */
6604 nfsm_chain_get_32(error, &nmrep, fhflag);
6605 nfsmout_if(error);
6606 if (fhflag) {
6607 nfsm_chain_get_fh(error, &nmrep, NFS_VER3, fh);
6608 nfsmout_if(error);
6609 fhlen = fh->fh_len + 1;
6610 xlen = fhlen + sizeof(time_t);
6611 reclen = NFS_DIRENTRY_LEN_16(namlen + xlen);
6612 space_needed = reclen + attrlen;
6613 if (space_needed > space_free) {
6614 /* didn't actually have the room... move on to next buffer */
6615 nmrep = nmrepsave;
6616 goto nextbuffer;
6617 }
6618 /* pack the file handle into the record */
6619 dp->d_name[dp->d_namlen + 1] = (unsigned char)fh->fh_len; /* No truncation because fh_len's value is checked during nfsm_chain_get_fh() */
6620 bcopy(fh->fh_data, &dp->d_name[dp->d_namlen + 2], fh->fh_len);
6621 } else {
6622 /* mark the file handle invalid */
6623 fh->fh_len = 0;
6624 fhlen = fh->fh_len + 1;
6625 xlen = fhlen + sizeof(time_t);
6626 reclen = NFS_DIRENTRY_LEN_16(namlen + xlen);
6627 bzero(&dp->d_name[dp->d_namlen + 1], fhlen);
6628 }
6629 *(time_t*)(&dp->d_name[dp->d_namlen + 1 + fhlen]) = now.tv_sec;
6630 dp->d_reclen = reclen;
6631 nfs_rdirplus_update_node_attrs(dnp, dp, fh, nvattrp, &savedxid);
6632 }
6633 padstart = dp->d_name + dp->d_namlen + 1 + xlen;
6634 ndbhp->ndbh_count++;
6635 lastcookie = cookie;
6636 /* advance to next direntry in buffer */
6637 dp = NFS_DIRENTRY_NEXT(dp);
6638 ndbhp->ndbh_entry_end = (char*)dp - bp->nb_data;
6639 /* zero out the pad bytes */
6640 padlen = (char*)dp - padstart;
6641 if (padlen > 0) {
6642 bzero(padstart, padlen);
6643 }
6644 /* check for more entries */
6645 nfsm_chain_get_32(error, &nmrep, more_entries);
6646 nfsmout_if(error);
6647 }
6648 /* Finally, get the eof boolean */
6649 nfsm_chain_get_32(error, &nmrep, eof);
6650 nfsmout_if(error);
6651 if (eof) {
6652 ndbhp->ndbh_flags |= (NDB_FULL | NDB_EOF);
6653 nfs_node_lock_force(dnp);
6654 dnp->n_eofcookie = lastcookie;
6655 if (rdirplus) {
6656 dnp->n_rdirplusstamp_eof = now.tv_sec;
6657 }
6658 nfs_node_unlock(dnp);
6659 } else {
6660 more_entries = 1;
6661 }
6662 if (bp_dropped) {
6663 nfs_buf_release(bp, 0);
6664 bp = NULL;
6665 break;
6666 }
6667 if ((lockerror = nfs_node_lock(dnp))) {
6668 error = lockerror;
6669 }
6670 nfsmout_if(error);
6671 nfsm_chain_cleanup(&nmrep);
6672 nfsm_chain_null(&nmreq);
6673 }
6674 nfsmout:
6675 if (bp_dropped && bp) {
6676 nfs_buf_release(bp, 0);
6677 }
6678 if (!lockerror) {
6679 nfs_node_unlock(dnp);
6680 }
6681 nfsm_chain_cleanup(&nmreq);
6682 nfsm_chain_cleanup(&nmrep);
6683 NFS_ZFREE(nfs_fhandle_zone, fh);
6684 return bp_dropped ? NFSERR_DIRBUFDROPPED : error;
6685 }
6686
6687 /*
6688 * Silly rename. To make the NFS filesystem that is stateless look a little
6689 * more like the "ufs" a remove of an active vnode is translated to a rename
6690 * to a funny looking filename that is removed by nfs_vnop_inactive on the
6691 * nfsnode. There is the potential for another process on a different client
6692 * to create the same funny name between when the lookitup() fails and the
6693 * rename() completes, but...
6694 */
6695
6696 /* format of "random" silly names - includes a number and pid */
6697 /* (note: shouldn't exceed size of nfs_sillyrename.nsr_name) */
6698 #define NFS_SILLYNAME_FORMAT ".nfs.%08x.%04x"
6699 /* starting from zero isn't silly enough */
6700 static uint32_t nfs_sillyrename_number = 0x20051025;
6701
6702 int
6703 nfs_sillyrename(
6704 nfsnode_t dnp,
6705 nfsnode_t np,
6706 struct componentname *cnp,
6707 vfs_context_t ctx)
6708 {
6709 struct nfs_sillyrename *nsp;
6710 int error;
6711 pid_t pid;
6712 kauth_cred_t cred;
6713 uint32_t num;
6714 struct nfsmount *nmp;
6715
6716 nmp = NFSTONMP(dnp);
6717 if (nfs_mount_gone(nmp)) {
6718 return ENXIO;
6719 }
6720
6721 nfs_name_cache_purge(dnp, np, cnp, ctx);
6722
6723 MALLOC(nsp, struct nfs_sillyrename *,
6724 sizeof(struct nfs_sillyrename), M_TEMP, M_WAITOK);
6725 if (!nsp) {
6726 return ENOMEM;
6727 }
6728 cred = vfs_context_ucred(ctx);
6729 kauth_cred_ref(cred);
6730 nsp->nsr_cred = cred;
6731 nsp->nsr_dnp = dnp;
6732 error = vnode_ref(NFSTOV(dnp));
6733 if (error) {
6734 goto bad_norele;
6735 }
6736
6737 /* Fudge together a funny name */
6738 pid = vfs_context_pid(ctx);
6739 num = OSAddAtomic(1, &nfs_sillyrename_number);
6740 nsp->nsr_namlen = snprintf(nsp->nsr_name, sizeof(nsp->nsr_name),
6741 NFS_SILLYNAME_FORMAT, num, (pid & 0xffff));
6742 if (nsp->nsr_namlen >= (int)sizeof(nsp->nsr_name)) {
6743 nsp->nsr_namlen = sizeof(nsp->nsr_name) - 1;
6744 }
6745
6746 /* Try lookitups until we get one that isn't there */
6747 while (nfs_lookitup(dnp, nsp->nsr_name, nsp->nsr_namlen, ctx, NULL) == 0) {
6748 num = OSAddAtomic(1, &nfs_sillyrename_number);
6749 nsp->nsr_namlen = snprintf(nsp->nsr_name, sizeof(nsp->nsr_name),
6750 NFS_SILLYNAME_FORMAT, num, (pid & 0xffff));
6751 if (nsp->nsr_namlen >= (int)sizeof(nsp->nsr_name)) {
6752 nsp->nsr_namlen = sizeof(nsp->nsr_name) - 1;
6753 }
6754 }
6755
6756 /* now, do the rename */
6757 error = nmp->nm_funcs->nf_rename_rpc(dnp, cnp->cn_nameptr, cnp->cn_namelen,
6758 dnp, nsp->nsr_name, nsp->nsr_namlen, ctx);
6759
6760 /* Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. */
6761 if (error == ENOENT) {
6762 error = 0;
6763 }
6764 if (!error) {
6765 nfs_node_lock_force(dnp);
6766 if (dnp->n_flag & NNEGNCENTRIES) {
6767 dnp->n_flag &= ~NNEGNCENTRIES;
6768 cache_purge_negatives(NFSTOV(dnp));
6769 }
6770 nfs_node_unlock(dnp);
6771 }
6772 FSDBG(267, dnp, np, num, error);
6773 if (error) {
6774 goto bad;
6775 }
6776 error = nfs_lookitup(dnp, nsp->nsr_name, nsp->nsr_namlen, ctx, &np);
6777 nfs_node_lock_force(np);
6778 np->n_sillyrename = nsp;
6779 nfs_node_unlock(np);
6780 return 0;
6781 bad:
6782 vnode_rele(NFSTOV(dnp));
6783 bad_norele:
6784 nsp->nsr_cred = NOCRED;
6785 kauth_cred_unref(&cred);
6786 FREE(nsp, M_TEMP);
6787 return error;
6788 }
6789
6790 int
6791 nfs3_lookup_rpc_async(
6792 nfsnode_t dnp,
6793 char *name,
6794 int namelen,
6795 vfs_context_t ctx,
6796 struct nfsreq **reqp)
6797 {
6798 struct nfsmount *nmp;
6799 struct nfsm_chain nmreq;
6800 int error = 0, nfsvers;
6801
6802 nmp = NFSTONMP(dnp);
6803 if (nfs_mount_gone(nmp)) {
6804 return ENXIO;
6805 }
6806 nfsvers = nmp->nm_vers;
6807
6808 nfsm_chain_null(&nmreq);
6809
6810 nfsm_chain_build_alloc_init(error, &nmreq,
6811 NFSX_FH(nfsvers) + NFSX_UNSIGNED + nfsm_rndup(namelen));
6812 nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
6813 nfsm_chain_add_name(error, &nmreq, name, namelen, nmp);
6814 nfsm_chain_build_done(error, &nmreq);
6815 nfsmout_if(error);
6816 error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC_LOOKUP,
6817 vfs_context_thread(ctx), vfs_context_ucred(ctx), NULL, 0, NULL, reqp);
6818 nfsmout:
6819 nfsm_chain_cleanup(&nmreq);
6820 return error;
6821 }
6822
6823 int
6824 nfs3_lookup_rpc_async_finish(
6825 nfsnode_t dnp,
6826 __unused char *name,
6827 __unused int namelen,
6828 vfs_context_t ctx,
6829 struct nfsreq *req,
6830 u_int64_t *xidp,
6831 fhandle_t *fhp,
6832 struct nfs_vattr *nvap)
6833 {
6834 int error = 0, lockerror = ENOENT, status = 0, nfsvers, attrflag;
6835 u_int64_t xid;
6836 struct nfsmount *nmp;
6837 struct nfsm_chain nmrep;
6838
6839 nmp = NFSTONMP(dnp);
6840 if (nmp == NULL) {
6841 return ENXIO;
6842 }
6843 nfsvers = nmp->nm_vers;
6844
6845 nfsm_chain_null(&nmrep);
6846
6847 error = nfs_request_async_finish(req, &nmrep, xidp, &status);
6848
6849 if ((lockerror = nfs_node_lock(dnp))) {
6850 error = lockerror;
6851 }
6852 xid = *xidp;
6853 if (error || status) {
6854 if (nfsvers == NFS_VER3) {
6855 nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid);
6856 }
6857 if (!error) {
6858 error = status;
6859 }
6860 goto nfsmout;
6861 }
6862
6863 nfsmout_if(error || !fhp || !nvap);
6864
6865 /* get the file handle */
6866 nfsm_chain_get_fh(error, &nmrep, nfsvers, fhp);
6867
6868 /* get the attributes */
6869 if (nfsvers == NFS_VER3) {
6870 nfsm_chain_postop_attr_get(nmp, error, &nmrep, attrflag, nvap);
6871 nfsm_chain_postop_attr_update(error, &nmrep, dnp, &xid);
6872 if (!error && !attrflag) {
6873 error = nfs3_getattr_rpc(NULL, NFSTOMP(dnp), fhp->fh_data, fhp->fh_len, 0, ctx, nvap, xidp);
6874 }
6875 } else {
6876 error = nfs_parsefattr(nmp, &nmrep, nfsvers, nvap);
6877 }
6878 nfsmout:
6879 if (!lockerror) {
6880 nfs_node_unlock(dnp);
6881 }
6882 nfsm_chain_cleanup(&nmrep);
6883 return error;
6884 }
6885
6886 /*
6887 * Look up a file name and optionally either update the file handle or
6888 * allocate an nfsnode, depending on the value of npp.
6889 * npp == NULL --> just do the lookup
6890 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
6891 * handled too
6892 * *npp != NULL --> update the file handle in the vnode
6893 */
6894 int
6895 nfs_lookitup(
6896 nfsnode_t dnp,
6897 char *name,
6898 int namelen,
6899 vfs_context_t ctx,
6900 nfsnode_t *npp)
6901 {
6902 int error = 0;
6903 nfsnode_t np, newnp = NULL;
6904 u_int64_t xid;
6905 fhandle_t *fh;
6906 struct nfsmount *nmp;
6907 struct nfs_vattr *nvattr;
6908 struct nfsreq *req;
6909
6910 nmp = NFSTONMP(dnp);
6911 if (nfs_mount_gone(nmp)) {
6912 return ENXIO;
6913 }
6914
6915 if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME) &&
6916 (namelen > nmp->nm_fsattr.nfsa_maxname)) {
6917 return ENAMETOOLONG;
6918 }
6919
6920 fh = zalloc(nfs_fhandle_zone);
6921 req = zalloc_flags(nfs_req_zone, Z_WAITOK);
6922 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
6923 NVATTR_INIT(nvattr);
6924
6925 /* check for lookup of "." */
6926 if ((name[0] == '.') && (namelen == 1)) {
6927 /* skip lookup, we know who we are */
6928 fh->fh_len = 0;
6929 newnp = dnp;
6930 goto nfsmout;
6931 }
6932
6933 error = nmp->nm_funcs->nf_lookup_rpc_async(dnp, name, namelen, ctx, &req);
6934 nfsmout_if(error);
6935 error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, name, namelen, ctx, req, &xid, fh, nvattr);
6936 nfsmout_if(!npp || error);
6937
6938 if (*npp) {
6939 np = *npp;
6940 if (fh->fh_len != np->n_fhsize) {
6941 u_char *oldbuf = (np->n_fhsize > NFS_SMALLFH) ? np->n_fhp : NULL;
6942 if (fh->fh_len > NFS_SMALLFH) {
6943 MALLOC(np->n_fhp, u_char *, fh->fh_len, M_NFSBIGFH, M_WAITOK);
6944 if (!np->n_fhp) {
6945 np->n_fhp = oldbuf;
6946 error = ENOMEM;
6947 goto nfsmout;
6948 }
6949 } else {
6950 np->n_fhp = &np->n_fh[0];
6951 }
6952 if (oldbuf) {
6953 FREE(oldbuf, M_NFSBIGFH);
6954 }
6955 }
6956 bcopy(fh->fh_data, np->n_fhp, fh->fh_len);
6957 np->n_fhsize = fh->fh_len;
6958 nfs_node_lock_force(np);
6959 error = nfs_loadattrcache(np, nvattr, &xid, 0);
6960 nfs_node_unlock(np);
6961 nfsmout_if(error);
6962 newnp = np;
6963 } else if (NFS_CMPFH(dnp, fh->fh_data, fh->fh_len)) {
6964 nfs_node_lock_force(dnp);
6965 if (dnp->n_xid <= xid) {
6966 error = nfs_loadattrcache(dnp, nvattr, &xid, 0);
6967 }
6968 nfs_node_unlock(dnp);
6969 nfsmout_if(error);
6970 newnp = dnp;
6971 } else {
6972 struct componentname cn, *cnp = &cn;
6973 bzero(cnp, sizeof(*cnp));
6974 cnp->cn_nameptr = name;
6975 cnp->cn_namelen = namelen;
6976 error = nfs_nget(NFSTOMP(dnp), dnp, cnp, fh->fh_data, fh->fh_len,
6977 nvattr, &xid, req->r_auth, NG_MAKEENTRY, &np);
6978 nfsmout_if(error);
6979 newnp = np;
6980 }
6981
6982 nfsmout:
6983 if (npp && !*npp && !error) {
6984 *npp = newnp;
6985 }
6986 NVATTR_CLEANUP(nvattr);
6987 NFS_ZFREE(nfs_fhandle_zone, fh);
6988 NFS_ZFREE(nfs_req_zone, req);
6989 FREE(nvattr, M_TEMP);
6990 return error;
6991 }
6992
6993 /*
6994 * set up and initialize a "._" file lookup structure used for
6995 * performing async lookups.
6996 */
6997 void
6998 nfs_dulookup_init(struct nfs_dulookup *dulp, nfsnode_t dnp, const char *name, int namelen, vfs_context_t ctx)
6999 {
7000 int error, du_namelen;
7001 vnode_t du_vp;
7002 struct nfsmount *nmp = NFSTONMP(dnp);
7003
7004 /* check for ._ file in name cache */
7005 dulp->du_flags = 0;
7006 bzero(&dulp->du_cn, sizeof(dulp->du_cn));
7007 du_namelen = namelen + 2;
7008 if (!nmp || NMFLAG(nmp, NONEGNAMECACHE)) {
7009 return;
7010 }
7011 if ((namelen >= 2) && (name[0] == '.') && (name[1] == '_')) {
7012 return;
7013 }
7014 if (du_namelen >= (int)sizeof(dulp->du_smallname)) {
7015 MALLOC(dulp->du_cn.cn_nameptr, char *, du_namelen + 1, M_TEMP, M_WAITOK);
7016 } else {
7017 dulp->du_cn.cn_nameptr = dulp->du_smallname;
7018 }
7019 if (!dulp->du_cn.cn_nameptr) {
7020 return;
7021 }
7022 dulp->du_cn.cn_namelen = du_namelen;
7023 snprintf(dulp->du_cn.cn_nameptr, du_namelen + 1, "._%s", name);
7024 dulp->du_cn.cn_nameptr[du_namelen] = '\0';
7025 dulp->du_cn.cn_nameiop = LOOKUP;
7026 dulp->du_cn.cn_flags = MAKEENTRY;
7027
7028 error = cache_lookup(NFSTOV(dnp), &du_vp, &dulp->du_cn);
7029 if (error == -1) {
7030 vnode_put(du_vp);
7031 } else if (!error) {
7032 nmp = NFSTONMP(dnp);
7033 if (nmp && (nmp->nm_vers > NFS_VER2) && NMFLAG(nmp, RDIRPLUS)) {
7034 /* if rdirplus, try dir buf cache lookup */
7035 nfsnode_t du_np = NULL;
7036 if (!nfs_dir_buf_cache_lookup(dnp, &du_np, &dulp->du_cn, ctx, 0, NULL) && du_np) {
7037 /* dir buf cache hit */
7038 du_vp = NFSTOV(du_np);
7039 vnode_put(du_vp);
7040 error = -1;
7041 }
7042 }
7043 if (!error) {
7044 dulp->du_flags |= NFS_DULOOKUP_DOIT;
7045 }
7046 }
7047 }
7048
7049 /*
7050 * start an async "._" file lookup request
7051 */
7052 void
7053 nfs_dulookup_start(struct nfs_dulookup *dulp, nfsnode_t dnp, vfs_context_t ctx)
7054 {
7055 struct nfsmount *nmp = NFSTONMP(dnp);
7056 struct nfsreq *req = &dulp->du_req;
7057
7058 if (!nmp || !(dulp->du_flags & NFS_DULOOKUP_DOIT) || (dulp->du_flags & NFS_DULOOKUP_INPROG)) {
7059 return;
7060 }
7061 if (!nmp->nm_funcs->nf_lookup_rpc_async(dnp, dulp->du_cn.cn_nameptr,
7062 dulp->du_cn.cn_namelen, ctx, &req)) {
7063 dulp->du_flags |= NFS_DULOOKUP_INPROG;
7064 }
7065 }
7066
7067 /*
7068 * finish an async "._" file lookup request and clean up the structure
7069 */
7070 void
7071 nfs_dulookup_finish(struct nfs_dulookup *dulp, nfsnode_t dnp, vfs_context_t ctx)
7072 {
7073 struct nfsmount *nmp = NFSTONMP(dnp);
7074 int error;
7075 nfsnode_t du_np;
7076 u_int64_t xid;
7077 fhandle_t *fh;
7078 struct nfs_vattr *nvattr;
7079
7080 if (!nmp || !(dulp->du_flags & NFS_DULOOKUP_INPROG)) {
7081 goto out;
7082 }
7083
7084 fh = zalloc(nfs_fhandle_zone);
7085 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
7086 NVATTR_INIT(nvattr);
7087 error = nmp->nm_funcs->nf_lookup_rpc_async_finish(dnp, dulp->du_cn.cn_nameptr,
7088 dulp->du_cn.cn_namelen, ctx, &dulp->du_req, &xid, fh, nvattr);
7089 dulp->du_flags &= ~NFS_DULOOKUP_INPROG;
7090 if (error == ENOENT) {
7091 /* add a negative entry in the name cache */
7092 nfs_node_lock_force(dnp);
7093 cache_enter(NFSTOV(dnp), NULL, &dulp->du_cn);
7094 dnp->n_flag |= NNEGNCENTRIES;
7095 nfs_node_unlock(dnp);
7096 } else if (!error) {
7097 error = nfs_nget(NFSTOMP(dnp), dnp, &dulp->du_cn, fh->fh_data, fh->fh_len,
7098 nvattr, &xid, dulp->du_req.r_auth, NG_MAKEENTRY, &du_np);
7099 if (!error) {
7100 nfs_node_unlock(du_np);
7101 vnode_put(NFSTOV(du_np));
7102 }
7103 }
7104 NVATTR_CLEANUP(nvattr);
7105 NFS_ZFREE(nfs_fhandle_zone, fh);
7106 FREE(nvattr, M_TEMP);
7107 out:
7108 if (dulp->du_flags & NFS_DULOOKUP_INPROG) {
7109 nfs_request_async_cancel(&dulp->du_req);
7110 }
7111 if (dulp->du_cn.cn_nameptr && (dulp->du_cn.cn_nameptr != dulp->du_smallname)) {
7112 FREE(dulp->du_cn.cn_nameptr, M_TEMP);
7113 }
7114 }
7115
7116
7117 /*
7118 * NFS Version 3 commit RPC
7119 */
7120 int
7121 nfs3_commit_rpc(
7122 nfsnode_t np,
7123 uint64_t offset,
7124 uint64_t count,
7125 kauth_cred_t cred,
7126 uint64_t wverf)
7127 {
7128 struct nfsmount *nmp;
7129 int error = 0, lockerror, status = 0, wccpostattr = 0, nfsvers;
7130 struct timespec premtime = { .tv_sec = 0, .tv_nsec = 0 };
7131 u_int64_t xid, newwverf;
7132 uint32_t count32;
7133 struct nfsm_chain nmreq, nmrep;
7134
7135 nmp = NFSTONMP(np);
7136 FSDBG(521, np, offset, count, nmp ? nmp->nm_state : 0);
7137 if (nfs_mount_gone(nmp)) {
7138 return ENXIO;
7139 }
7140 if (!(nmp->nm_state & NFSSTA_HASWRITEVERF)) {
7141 return 0;
7142 }
7143 nfsvers = nmp->nm_vers;
7144 count32 = count > UINT32_MAX ? 0 : (uint32_t)count;
7145
7146 nfsm_chain_null(&nmreq);
7147 nfsm_chain_null(&nmrep);
7148
7149 nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(NFS_VER3));
7150 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
7151 nfsm_chain_add_64(error, &nmreq, offset);
7152 nfsm_chain_add_32(error, &nmreq, count32);
7153 nfsm_chain_build_done(error, &nmreq);
7154 nfsmout_if(error);
7155 error = nfs_request2(np, NULL, &nmreq, NFSPROC_COMMIT,
7156 current_thread(), cred, NULL, 0, &nmrep, &xid, &status);
7157 if ((lockerror = nfs_node_lock(np))) {
7158 error = lockerror;
7159 }
7160 /* can we do anything useful with the wcc info? */
7161 nfsm_chain_get_wcc_data(error, &nmrep, np, &premtime, &wccpostattr, &xid);
7162 if (!lockerror) {
7163 nfs_node_unlock(np);
7164 }
7165 if (!error) {
7166 error = status;
7167 }
7168 nfsm_chain_get_64(error, &nmrep, newwverf);
7169 nfsmout_if(error);
7170 lck_mtx_lock(&nmp->nm_lock);
7171 if (nmp->nm_verf != newwverf) {
7172 nmp->nm_verf = newwverf;
7173 }
7174 if (wverf != newwverf) {
7175 error = NFSERR_STALEWRITEVERF;
7176 }
7177 lck_mtx_unlock(&nmp->nm_lock);
7178 nfsmout:
7179 nfsm_chain_cleanup(&nmreq);
7180 nfsm_chain_cleanup(&nmrep);
7181 return error;
7182 }
7183
7184
7185 int
7186 nfs_vnop_blockmap(
7187 __unused struct vnop_blockmap_args /* {
7188 * struct vnodeop_desc *a_desc;
7189 * vnode_t a_vp;
7190 * off_t a_foffset;
7191 * size_t a_size;
7192 * daddr64_t *a_bpn;
7193 * size_t *a_run;
7194 * void *a_poff;
7195 * int a_flags;
7196 * } */*ap)
7197 {
7198 return ENOTSUP;
7199 }
7200
7201
7202 /*
7203 * fsync vnode op. Just call nfs_flush().
7204 */
7205 /* ARGSUSED */
7206 int
7207 nfs_vnop_fsync(
7208 struct vnop_fsync_args /* {
7209 * struct vnodeop_desc *a_desc;
7210 * vnode_t a_vp;
7211 * int a_waitfor;
7212 * vfs_context_t a_context;
7213 * } */*ap)
7214 {
7215 return nfs_flush(VTONFS(ap->a_vp), ap->a_waitfor, vfs_context_thread(ap->a_context), 0);
7216 }
7217
7218
7219 /*
7220 * Do an NFS pathconf RPC.
7221 */
7222 int
7223 nfs3_pathconf_rpc(
7224 nfsnode_t np,
7225 struct nfs_fsattr *nfsap,
7226 vfs_context_t ctx)
7227 {
7228 u_int64_t xid;
7229 int error = 0, lockerror, status = 0, nfsvers;
7230 struct nfsm_chain nmreq, nmrep;
7231 struct nfsmount *nmp = NFSTONMP(np);
7232 uint32_t val = 0;
7233
7234 if (nfs_mount_gone(nmp)) {
7235 return ENXIO;
7236 }
7237 nfsvers = nmp->nm_vers;
7238
7239 nfsm_chain_null(&nmreq);
7240 nfsm_chain_null(&nmrep);
7241
7242 /* fetch pathconf info from server */
7243 nfsm_chain_build_alloc_init(error, &nmreq, NFSX_FH(NFS_VER3));
7244 nfsm_chain_add_fh(error, &nmreq, nfsvers, np->n_fhp, np->n_fhsize);
7245 nfsm_chain_build_done(error, &nmreq);
7246 nfsmout_if(error);
7247 error = nfs_request(np, NULL, &nmreq, NFSPROC_PATHCONF, ctx, NULL, &nmrep, &xid, &status);
7248 if ((lockerror = nfs_node_lock(np))) {
7249 error = lockerror;
7250 }
7251 nfsm_chain_postop_attr_update(error, &nmrep, np, &xid);
7252 if (!lockerror) {
7253 nfs_node_unlock(np);
7254 }
7255 if (!error) {
7256 error = status;
7257 }
7258 nfsm_chain_get_32(error, &nmrep, nfsap->nfsa_maxlink);
7259 nfsm_chain_get_32(error, &nmrep, nfsap->nfsa_maxname);
7260 nfsap->nfsa_flags &= ~(NFS_FSFLAG_NO_TRUNC | NFS_FSFLAG_CHOWN_RESTRICTED | NFS_FSFLAG_CASE_INSENSITIVE | NFS_FSFLAG_CASE_PRESERVING);
7261 nfsm_chain_get_32(error, &nmrep, val);
7262 if (val) {
7263 nfsap->nfsa_flags |= NFS_FSFLAG_NO_TRUNC;
7264 }
7265 nfsm_chain_get_32(error, &nmrep, val);
7266 if (val) {
7267 nfsap->nfsa_flags |= NFS_FSFLAG_CHOWN_RESTRICTED;
7268 }
7269 nfsm_chain_get_32(error, &nmrep, val);
7270 if (val) {
7271 nfsap->nfsa_flags |= NFS_FSFLAG_CASE_INSENSITIVE;
7272 }
7273 nfsm_chain_get_32(error, &nmrep, val);
7274 if (val) {
7275 nfsap->nfsa_flags |= NFS_FSFLAG_CASE_PRESERVING;
7276 }
7277 NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_MAXLINK);
7278 NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_MAXNAME);
7279 NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_NO_TRUNC);
7280 NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED);
7281 NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE);
7282 NFS_BITMAP_SET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_PRESERVING);
7283 nfsmout:
7284 nfsm_chain_cleanup(&nmreq);
7285 nfsm_chain_cleanup(&nmrep);
7286 return error;
7287 }
7288
7289 /* save pathconf info for NFSv3 mount */
7290 void
7291 nfs3_pathconf_cache(struct nfsmount *nmp, struct nfs_fsattr *nfsap)
7292 {
7293 nmp->nm_fsattr.nfsa_maxlink = nfsap->nfsa_maxlink;
7294 nmp->nm_fsattr.nfsa_maxname = nfsap->nfsa_maxname;
7295 nmp->nm_fsattr.nfsa_flags &= ~(NFS_FSFLAG_NO_TRUNC | NFS_FSFLAG_CHOWN_RESTRICTED | NFS_FSFLAG_CASE_INSENSITIVE | NFS_FSFLAG_CASE_PRESERVING);
7296 nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_NO_TRUNC;
7297 nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_CHOWN_RESTRICTED;
7298 nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE;
7299 nmp->nm_fsattr.nfsa_flags |= nfsap->nfsa_flags & NFS_FSFLAG_CASE_PRESERVING;
7300 NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXLINK);
7301 NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME);
7302 NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_NO_TRUNC);
7303 NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED);
7304 NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE);
7305 NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_CASE_PRESERVING);
7306 nmp->nm_state |= NFSSTA_GOTPATHCONF;
7307 }
7308
7309 /*
7310 * Return POSIX pathconf information applicable to nfs.
7311 *
7312 * The NFS V2 protocol doesn't support this, so just return EINVAL
7313 * for V2.
7314 */
7315 /* ARGSUSED */
7316 int
7317 nfs_vnop_pathconf(
7318 struct vnop_pathconf_args /* {
7319 * struct vnodeop_desc *a_desc;
7320 * vnode_t a_vp;
7321 * int a_name;
7322 * int32_t *a_retval;
7323 * vfs_context_t a_context;
7324 * } */*ap)
7325 {
7326 vnode_t vp = ap->a_vp;
7327 nfsnode_t np = VTONFS(vp);
7328 struct nfsmount *nmp;
7329 struct nfs_fsattr nfsa, *nfsap;
7330 int error = 0;
7331 uint64_t maxFileSize;
7332 uint nbits;
7333
7334 nmp = VTONMP(vp);
7335 if (nfs_mount_gone(nmp)) {
7336 return ENXIO;
7337 }
7338
7339 switch (ap->a_name) {
7340 case _PC_LINK_MAX:
7341 case _PC_NAME_MAX:
7342 case _PC_CHOWN_RESTRICTED:
7343 case _PC_NO_TRUNC:
7344 case _PC_CASE_SENSITIVE:
7345 case _PC_CASE_PRESERVING:
7346 break;
7347 case _PC_FILESIZEBITS:
7348 if (nmp->nm_vers == NFS_VER2) {
7349 *ap->a_retval = 32;
7350 return 0;
7351 }
7352 break;
7353 case _PC_XATTR_SIZE_BITS:
7354 /* Do we support xattrs natively? */
7355 if (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_NAMED_ATTR) {
7356 break; /* Yes */
7357 }
7358 /* No... so just return an error */
7359 return EINVAL;
7360 default:
7361 /* don't bother contacting the server if we know the answer */
7362 return EINVAL;
7363 }
7364
7365 if (nmp->nm_vers == NFS_VER2) {
7366 return EINVAL;
7367 }
7368
7369 lck_mtx_lock(&nmp->nm_lock);
7370 if (nmp->nm_vers == NFS_VER3) {
7371 if (!(nmp->nm_state & NFSSTA_GOTPATHCONF) || (!(nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS) && nmp->nm_dnp != np)) {
7372 /* no pathconf info cached OR we were asked for non-root pathconf and filesystem does not support FSF_HOMOGENEOUS */
7373 lck_mtx_unlock(&nmp->nm_lock);
7374 NFS_CLEAR_ATTRIBUTES(nfsa.nfsa_bitmap);
7375 error = nfs3_pathconf_rpc(np, &nfsa, ap->a_context);
7376 if (error) {
7377 return error;
7378 }
7379 nmp = VTONMP(vp);
7380 if (nfs_mount_gone(nmp)) {
7381 return ENXIO;
7382 }
7383 lck_mtx_lock(&nmp->nm_lock);
7384 if (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS) {
7385 /* all files have the same pathconf info, */
7386 /* so cache a copy of the results */
7387 nfs3_pathconf_cache(nmp, &nfsa);
7388 }
7389 nfsap = &nfsa;
7390 } else {
7391 nfsap = &nmp->nm_fsattr;
7392 }
7393 }
7394 #if CONFIG_NFS4
7395 else if (!(nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS)) {
7396 /* no pathconf info cached */
7397 lck_mtx_unlock(&nmp->nm_lock);
7398 NFS_CLEAR_ATTRIBUTES(nfsa.nfsa_bitmap);
7399 error = nfs4_pathconf_rpc(np, &nfsa, ap->a_context);
7400 if (error) {
7401 return error;
7402 }
7403 nmp = VTONMP(vp);
7404 if (nfs_mount_gone(nmp)) {
7405 return ENXIO;
7406 }
7407 lck_mtx_lock(&nmp->nm_lock);
7408 nfsap = &nfsa;
7409 }
7410 #endif
7411 else {
7412 nfsap = &nmp->nm_fsattr;
7413 }
7414 switch (ap->a_name) {
7415 case _PC_LINK_MAX:
7416 if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXLINK)) {
7417 *ap->a_retval = nfsap->nfsa_maxlink;
7418 #if CONFIG_NFS4
7419 } else if ((nmp->nm_vers == NFS_VER4) && NFS_BITMAP_ISSET(np->n_vattr.nva_bitmap, NFS_FATTR_MAXLINK)) {
7420 *ap->a_retval = np->n_vattr.nva_maxlink;
7421 #endif
7422 } else {
7423 error = EINVAL;
7424 }
7425 break;
7426 case _PC_NAME_MAX:
7427 if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXNAME)) {
7428 *ap->a_retval = nfsap->nfsa_maxname;
7429 } else {
7430 error = EINVAL;
7431 }
7432 break;
7433 case _PC_CHOWN_RESTRICTED:
7434 if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CHOWN_RESTRICTED)) {
7435 *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CHOWN_RESTRICTED) ? 200112 /* _POSIX_CHOWN_RESTRICTED */ : 0;
7436 } else {
7437 error = EINVAL;
7438 }
7439 break;
7440 case _PC_NO_TRUNC:
7441 if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_NO_TRUNC)) {
7442 *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_NO_TRUNC) ? 200112 /* _POSIX_NO_TRUNC */ : 0;
7443 } else {
7444 error = EINVAL;
7445 }
7446 break;
7447 case _PC_CASE_SENSITIVE:
7448 if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_INSENSITIVE)) {
7449 *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE) ? 0 : 1;
7450 } else {
7451 error = EINVAL;
7452 }
7453 break;
7454 case _PC_CASE_PRESERVING:
7455 if (NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_CASE_PRESERVING)) {
7456 *ap->a_retval = (nfsap->nfsa_flags & NFS_FSFLAG_CASE_PRESERVING) ? 1 : 0;
7457 } else {
7458 error = EINVAL;
7459 }
7460 break;
7461 case _PC_XATTR_SIZE_BITS: /* same as file size bits if named attrs supported */
7462 case _PC_FILESIZEBITS:
7463 if (!NFS_BITMAP_ISSET(nfsap->nfsa_bitmap, NFS_FATTR_MAXFILESIZE)) {
7464 *ap->a_retval = 64;
7465 error = 0;
7466 break;
7467 }
7468 maxFileSize = nfsap->nfsa_maxfilesize;
7469 nbits = 1;
7470 if (maxFileSize & 0xffffffff00000000ULL) {
7471 nbits += 32;
7472 maxFileSize >>= 32;
7473 }
7474 if (maxFileSize & 0xffff0000) {
7475 nbits += 16;
7476 maxFileSize >>= 16;
7477 }
7478 if (maxFileSize & 0xff00) {
7479 nbits += 8;
7480 maxFileSize >>= 8;
7481 }
7482 if (maxFileSize & 0xf0) {
7483 nbits += 4;
7484 maxFileSize >>= 4;
7485 }
7486 if (maxFileSize & 0xc) {
7487 nbits += 2;
7488 maxFileSize >>= 2;
7489 }
7490 if (maxFileSize & 0x2) {
7491 nbits += 1;
7492 }
7493 *ap->a_retval = nbits;
7494 break;
7495 default:
7496 error = EINVAL;
7497 }
7498
7499 lck_mtx_unlock(&nmp->nm_lock);
7500
7501 return error;
7502 }
7503
7504 /*
7505 * Read wrapper for special devices.
7506 */
7507 int
7508 nfsspec_vnop_read(
7509 struct vnop_read_args /* {
7510 * struct vnodeop_desc *a_desc;
7511 * vnode_t a_vp;
7512 * struct uio *a_uio;
7513 * int a_ioflag;
7514 * vfs_context_t a_context;
7515 * } */*ap)
7516 {
7517 nfsnode_t np = VTONFS(ap->a_vp);
7518 struct timespec now;
7519 int error;
7520
7521 /*
7522 * Set access flag.
7523 */
7524 if ((error = nfs_node_lock(np))) {
7525 return error;
7526 }
7527 np->n_flag |= NACC;
7528 nanotime(&now);
7529 np->n_atim.tv_sec = now.tv_sec;
7530 np->n_atim.tv_nsec = now.tv_nsec;
7531 nfs_node_unlock(np);
7532 return VOCALL(spec_vnodeop_p, VOFFSET(vnop_read), ap);
7533 }
7534
7535 /*
7536 * Write wrapper for special devices.
7537 */
7538 int
7539 nfsspec_vnop_write(
7540 struct vnop_write_args /* {
7541 * struct vnodeop_desc *a_desc;
7542 * vnode_t a_vp;
7543 * struct uio *a_uio;
7544 * int a_ioflag;
7545 * vfs_context_t a_context;
7546 * } */*ap)
7547 {
7548 nfsnode_t np = VTONFS(ap->a_vp);
7549 struct timespec now;
7550 int error;
7551
7552 /*
7553 * Set update flag.
7554 */
7555 if ((error = nfs_node_lock(np))) {
7556 return error;
7557 }
7558 np->n_flag |= NUPD;
7559 nanotime(&now);
7560 np->n_mtim.tv_sec = now.tv_sec;
7561 np->n_mtim.tv_nsec = now.tv_nsec;
7562 nfs_node_unlock(np);
7563 return VOCALL(spec_vnodeop_p, VOFFSET(vnop_write), ap);
7564 }
7565
7566 /*
7567 * Close wrapper for special devices.
7568 *
7569 * Update the times on the nfsnode then do device close.
7570 */
7571 int
7572 nfsspec_vnop_close(
7573 struct vnop_close_args /* {
7574 * struct vnodeop_desc *a_desc;
7575 * vnode_t a_vp;
7576 * int a_fflag;
7577 * vfs_context_t a_context;
7578 * } */*ap)
7579 {
7580 vnode_t vp = ap->a_vp;
7581 nfsnode_t np = VTONFS(vp);
7582 struct vnode_attr vattr;
7583 mount_t mp;
7584 int error;
7585
7586 if ((error = nfs_node_lock(np))) {
7587 return error;
7588 }
7589 if (np->n_flag & (NACC | NUPD)) {
7590 np->n_flag |= NCHG;
7591 if (!vnode_isinuse(vp, 0) && (mp = vnode_mount(vp)) && !vfs_isrdonly(mp)) {
7592 VATTR_INIT(&vattr);
7593 if (np->n_flag & NACC) {
7594 vattr.va_access_time = np->n_atim;
7595 VATTR_SET_ACTIVE(&vattr, va_access_time);
7596 }
7597 if (np->n_flag & NUPD) {
7598 vattr.va_modify_time = np->n_mtim;
7599 VATTR_SET_ACTIVE(&vattr, va_modify_time);
7600 }
7601 nfs_node_unlock(np);
7602 vnode_setattr(vp, &vattr, ap->a_context);
7603 } else {
7604 nfs_node_unlock(np);
7605 }
7606 } else {
7607 nfs_node_unlock(np);
7608 }
7609 return VOCALL(spec_vnodeop_p, VOFFSET(vnop_close), ap);
7610 }
7611
7612 #if FIFO
7613 extern vnop_t **fifo_vnodeop_p;
7614
7615 /*
7616 * Read wrapper for fifos.
7617 */
7618 int
7619 nfsfifo_vnop_read(
7620 struct vnop_read_args /* {
7621 * struct vnodeop_desc *a_desc;
7622 * vnode_t a_vp;
7623 * struct uio *a_uio;
7624 * int a_ioflag;
7625 * vfs_context_t a_context;
7626 * } */*ap)
7627 {
7628 nfsnode_t np = VTONFS(ap->a_vp);
7629 struct timespec now;
7630 int error;
7631
7632 /*
7633 * Set access flag.
7634 */
7635 if ((error = nfs_node_lock(np))) {
7636 return error;
7637 }
7638 np->n_flag |= NACC;
7639 nanotime(&now);
7640 np->n_atim.tv_sec = now.tv_sec;
7641 np->n_atim.tv_nsec = now.tv_nsec;
7642 nfs_node_unlock(np);
7643 return VOCALL(fifo_vnodeop_p, VOFFSET(vnop_read), ap);
7644 }
7645
7646 /*
7647 * Write wrapper for fifos.
7648 */
7649 int
7650 nfsfifo_vnop_write(
7651 struct vnop_write_args /* {
7652 * struct vnodeop_desc *a_desc;
7653 * vnode_t a_vp;
7654 * struct uio *a_uio;
7655 * int a_ioflag;
7656 * vfs_context_t a_context;
7657 * } */*ap)
7658 {
7659 nfsnode_t np = VTONFS(ap->a_vp);
7660 struct timespec now;
7661 int error;
7662
7663 /*
7664 * Set update flag.
7665 */
7666 if ((error = nfs_node_lock(np))) {
7667 return error;
7668 }
7669 np->n_flag |= NUPD;
7670 nanotime(&now);
7671 np->n_mtim.tv_sec = now.tv_sec;
7672 np->n_mtim.tv_nsec = now.tv_nsec;
7673 nfs_node_unlock(np);
7674 return VOCALL(fifo_vnodeop_p, VOFFSET(vnop_write), ap);
7675 }
7676
7677 /*
7678 * Close wrapper for fifos.
7679 *
7680 * Update the times on the nfsnode then do fifo close.
7681 */
7682 int
7683 nfsfifo_vnop_close(
7684 struct vnop_close_args /* {
7685 * struct vnodeop_desc *a_desc;
7686 * vnode_t a_vp;
7687 * int a_fflag;
7688 * vfs_context_t a_context;
7689 * } */*ap)
7690 {
7691 vnode_t vp = ap->a_vp;
7692 nfsnode_t np = VTONFS(vp);
7693 struct vnode_attr vattr;
7694 struct timespec now;
7695 mount_t mp;
7696 int error;
7697
7698 if ((error = nfs_node_lock(np))) {
7699 return error;
7700 }
7701 if (np->n_flag & (NACC | NUPD)) {
7702 nanotime(&now);
7703 if (np->n_flag & NACC) {
7704 np->n_atim.tv_sec = now.tv_sec;
7705 np->n_atim.tv_nsec = now.tv_nsec;
7706 }
7707 if (np->n_flag & NUPD) {
7708 np->n_mtim.tv_sec = now.tv_sec;
7709 np->n_mtim.tv_nsec = now.tv_nsec;
7710 }
7711 np->n_flag |= NCHG;
7712 if (!vnode_isinuse(vp, 1) && (mp = vnode_mount(vp)) && !vfs_isrdonly(mp)) {
7713 VATTR_INIT(&vattr);
7714 if (np->n_flag & NACC) {
7715 vattr.va_access_time = np->n_atim;
7716 VATTR_SET_ACTIVE(&vattr, va_access_time);
7717 }
7718 if (np->n_flag & NUPD) {
7719 vattr.va_modify_time = np->n_mtim;
7720 VATTR_SET_ACTIVE(&vattr, va_modify_time);
7721 }
7722 nfs_node_unlock(np);
7723 vnode_setattr(vp, &vattr, ap->a_context);
7724 } else {
7725 nfs_node_unlock(np);
7726 }
7727 } else {
7728 nfs_node_unlock(np);
7729 }
7730 return VOCALL(fifo_vnodeop_p, VOFFSET(vnop_close), ap);
7731 }
7732 #endif /* FIFO */
7733
7734 /*ARGSUSED*/
7735 int
7736 nfs_vnop_ioctl(
7737 struct vnop_ioctl_args /* {
7738 * struct vnodeop_desc *a_desc;
7739 * vnode_t a_vp;
7740 * u_int32_t a_command;
7741 * caddr_t a_data;
7742 * int a_fflag;
7743 * vfs_context_t a_context;
7744 * } */*ap)
7745 {
7746 vfs_context_t ctx = ap->a_context;
7747 vnode_t vp = ap->a_vp;
7748 struct nfsmount *mp = VTONMP(vp);
7749 int error = ENOTTY;
7750 #if CONFIG_NFS_GSS
7751 struct user_nfs_gss_principal gprinc = {};
7752 size_t len;
7753 #endif
7754
7755 if (mp == NULL) {
7756 return ENXIO;
7757 }
7758 switch (ap->a_command) {
7759 case F_FULLFSYNC:
7760 if (vnode_vfsisrdonly(vp)) {
7761 return EROFS;
7762 }
7763 error = nfs_flush(VTONFS(vp), MNT_WAIT, vfs_context_thread(ctx), 0);
7764 break;
7765 #if CONFIG_NFS_GSS
7766 case NFS_IOC_DESTROY_CRED:
7767 if (!auth_is_kerberized(mp->nm_auth)) {
7768 return ENOTSUP;
7769 }
7770 if ((error = nfs_gss_clnt_ctx_remove(mp, vfs_context_ucred(ctx))) == ENOENT) {
7771 error = 0;
7772 }
7773 break;
7774 case NFS_IOC_SET_CRED:
7775 case NFS_IOC_SET_CRED64:
7776 if (!auth_is_kerberized(mp->nm_auth)) {
7777 return ENOTSUP;
7778 }
7779 if ((ap->a_command == NFS_IOC_SET_CRED && vfs_context_is64bit(ctx)) ||
7780 (ap->a_command == NFS_IOC_SET_CRED64 && !vfs_context_is64bit(ctx))) {
7781 return EINVAL;
7782 }
7783 if (vfs_context_is64bit(ctx)) {
7784 gprinc = *(struct user_nfs_gss_principal *)ap->a_data;
7785 } else {
7786 struct nfs_gss_principal *tp;
7787 tp = (struct nfs_gss_principal *)ap->a_data;
7788 gprinc.princlen = tp->princlen;
7789 gprinc.nametype = tp->nametype;
7790 gprinc.principal = CAST_USER_ADDR_T(tp->principal);
7791 }
7792 NFS_DBG(NFS_FAC_GSS, 7, "Enter NFS_FSCTL_SET_CRED (64-bit=%d): principal length %zu name type %d usr pointer 0x%llx\n", vfs_context_is64bit(ctx), gprinc.princlen, gprinc.nametype, gprinc.principal);
7793 if (gprinc.princlen > MAXPATHLEN) {
7794 return EINVAL;
7795 }
7796 uint8_t *p;
7797 MALLOC(p, uint8_t *, gprinc.princlen + 1, M_TEMP, M_WAITOK | M_ZERO);
7798 if (p == NULL) {
7799 return ENOMEM;
7800 }
7801 assert((user_addr_t)gprinc.principal == gprinc.principal);
7802 error = copyin((user_addr_t)gprinc.principal, p, gprinc.princlen);
7803 if (error) {
7804 NFS_DBG(NFS_FAC_GSS, 7, "NFS_FSCTL_SET_CRED could not copy in princiapl data of len %zu: %d\n",
7805 gprinc.princlen, error);
7806 FREE(p, M_TEMP);
7807 return error;
7808 }
7809 NFS_DBG(NFS_FAC_GSS, 7, "Seting credential to principal %s\n", p);
7810 error = nfs_gss_clnt_ctx_set_principal(mp, ctx, p, gprinc.princlen, gprinc.nametype);
7811 NFS_DBG(NFS_FAC_GSS, 7, "Seting credential to principal %s returned %d\n", p, error);
7812 FREE(p, M_TEMP);
7813 break;
7814 case NFS_IOC_GET_CRED:
7815 case NFS_IOC_GET_CRED64:
7816 if (!auth_is_kerberized(mp->nm_auth)) {
7817 return ENOTSUP;
7818 }
7819 if ((ap->a_command == NFS_IOC_GET_CRED && vfs_context_is64bit(ctx)) ||
7820 (ap->a_command == NFS_IOC_GET_CRED64 && !vfs_context_is64bit(ctx))) {
7821 return EINVAL;
7822 }
7823 error = nfs_gss_clnt_ctx_get_principal(mp, ctx, &gprinc);
7824 if (error) {
7825 break;
7826 }
7827 if (vfs_context_is64bit(ctx)) {
7828 struct user_nfs_gss_principal *upp = (struct user_nfs_gss_principal *)ap->a_data;
7829 len = upp->princlen;
7830 if (gprinc.princlen < len) {
7831 len = gprinc.princlen;
7832 }
7833 upp->princlen = gprinc.princlen;
7834 upp->nametype = gprinc.nametype;
7835 upp->flags = gprinc.flags;
7836 if (gprinc.principal) {
7837 assert((user_addr_t)upp->principal == upp->principal);
7838 error = copyout((void *)gprinc.principal, (user_addr_t)upp->principal, len);
7839 } else {
7840 upp->principal = USER_ADDR_NULL;
7841 }
7842 } else {
7843 struct nfs_gss_principal *u32pp = (struct nfs_gss_principal *)ap->a_data;
7844 len = u32pp->princlen;
7845 if (gprinc.princlen < len) {
7846 len = gprinc.princlen;
7847 }
7848 u32pp->princlen = gprinc.princlen;
7849 u32pp->nametype = gprinc.nametype;
7850 u32pp->flags = gprinc.flags;
7851 if (gprinc.principal) {
7852 error = copyout((void *)gprinc.principal, u32pp->principal, len);
7853 } else {
7854 u32pp->principal = (user32_addr_t)0;
7855 }
7856 }
7857 if (error) {
7858 NFS_DBG(NFS_FAC_GSS, 7, "NFS_FSCTL_GET_CRED could not copy out princiapl data of len %zu: %d\n",
7859 gprinc.princlen, error);
7860 }
7861 if (gprinc.principal) {
7862 void *ptr = (void *)gprinc.principal;
7863 gprinc.principal = 0;
7864 FREE(ptr, M_TEMP);
7865 }
7866 #endif /* CONFIG_NFS_GSS */
7867 }
7868
7869 return error;
7870 }
7871
7872 /*ARGSUSED*/
7873 int
7874 nfs_vnop_select(
7875 __unused struct vnop_select_args /* {
7876 * struct vnodeop_desc *a_desc;
7877 * vnode_t a_vp;
7878 * int a_which;
7879 * int a_fflags;
7880 * void *a_wql;
7881 * vfs_context_t a_context;
7882 * } */*ap)
7883 {
7884 /*
7885 * We were once bogusly seltrue() which returns 1. Is this right?
7886 */
7887 return 1;
7888 }
7889
7890 /*
7891 * vnode OP for pagein using UPL
7892 *
7893 * No buffer I/O, just RPCs straight into the mapped pages.
7894 */
7895 int
7896 nfs_vnop_pagein(
7897 struct vnop_pagein_args /* {
7898 * struct vnodeop_desc *a_desc;
7899 * vnode_t a_vp;
7900 * upl_t a_pl;
7901 * vm_offset_t a_pl_offset;
7902 * off_t a_f_offset;
7903 * size_t a_size;
7904 * int a_flags;
7905 * vfs_context_t a_context;
7906 * } */*ap)
7907 {
7908 vnode_t vp = ap->a_vp;
7909 upl_t pl = ap->a_pl;
7910 upl_size_t size = (upl_size_t)ap->a_size;
7911 off_t f_offset = ap->a_f_offset;
7912 upl_offset_t pl_offset = ap->a_pl_offset;
7913 int flags = ap->a_flags;
7914 thread_t thd;
7915 kauth_cred_t cred;
7916 nfsnode_t np = VTONFS(vp);
7917 size_t nmrsize, iosize, txsize, rxsize, retsize;
7918 off_t txoffset;
7919 struct nfsmount *nmp;
7920 int error = 0;
7921 vm_offset_t ioaddr, rxaddr;
7922 uio_t uio;
7923 char uio_buf[UIO_SIZEOF(1)];
7924 int nofreeupl = flags & UPL_NOCOMMIT;
7925 upl_page_info_t *plinfo;
7926 #define MAXPAGINGREQS 16 /* max outstanding RPCs for pagein/pageout */
7927 struct nfsreq *req[MAXPAGINGREQS];
7928 int nextsend, nextwait;
7929 #if CONFIG_NFS4
7930 uint32_t stategenid = 0;
7931 #endif
7932 uint32_t restart = 0;
7933 kern_return_t kret;
7934
7935 FSDBG(322, np, f_offset, size, flags);
7936 if (pl == (upl_t)NULL) {
7937 panic("nfs_pagein: no upl");
7938 }
7939
7940 if (size <= 0) {
7941 printf("nfs_pagein: invalid size %u", size);
7942 if (!nofreeupl) {
7943 (void) ubc_upl_abort_range(pl, pl_offset, size, 0);
7944 }
7945 return EINVAL;
7946 }
7947 if (f_offset < 0 || f_offset >= (off_t)np->n_size || (f_offset & PAGE_MASK_64)) {
7948 if (!nofreeupl) {
7949 ubc_upl_abort_range(pl, pl_offset, size,
7950 UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
7951 }
7952 return EINVAL;
7953 }
7954
7955 thd = vfs_context_thread(ap->a_context);
7956 cred = ubc_getcred(vp);
7957 if (!IS_VALID_CRED(cred)) {
7958 cred = vfs_context_ucred(ap->a_context);
7959 }
7960
7961 uio = uio_createwithbuffer(1, f_offset, UIO_SYSSPACE, UIO_READ,
7962 &uio_buf, sizeof(uio_buf));
7963
7964 nmp = VTONMP(vp);
7965 if (nfs_mount_gone(nmp)) {
7966 if (!nofreeupl) {
7967 ubc_upl_abort_range(pl, pl_offset, size,
7968 UPL_ABORT_ERROR | UPL_ABORT_FREE_ON_EMPTY);
7969 }
7970 return ENXIO;
7971 }
7972 nmrsize = nmp->nm_rsize;
7973
7974 plinfo = ubc_upl_pageinfo(pl);
7975 kret = ubc_upl_map(pl, &ioaddr);
7976 if (kret != KERN_SUCCESS) {
7977 panic("nfs_vnop_pagein: ubc_upl_map() failed with (%d)", kret);
7978 }
7979 ioaddr += pl_offset;
7980
7981 tryagain:
7982 #if CONFIG_NFS4
7983 if (nmp->nm_vers >= NFS_VER4) {
7984 stategenid = nmp->nm_stategenid;
7985 }
7986 #endif
7987 txsize = rxsize = size;
7988 txoffset = f_offset;
7989 rxaddr = ioaddr;
7990
7991 bzero(req, sizeof(req));
7992 nextsend = nextwait = 0;
7993 do {
7994 if (np->n_flag & NREVOKE) {
7995 error = EIO;
7996 break;
7997 }
7998 /* send requests while we need to and have available slots */
7999 while ((txsize > 0) && (req[nextsend] == NULL)) {
8000 iosize = MIN(nmrsize, txsize);
8001 if ((error = nmp->nm_funcs->nf_read_rpc_async(np, txoffset, iosize, thd, cred, NULL, &req[nextsend]))) {
8002 req[nextsend] = NULL;
8003 break;
8004 }
8005 txoffset += iosize;
8006 txsize -= iosize;
8007 nextsend = (nextsend + 1) % MAXPAGINGREQS;
8008 }
8009 /* wait while we need to and break out if more requests to send */
8010 while ((rxsize > 0) && req[nextwait]) {
8011 iosize = retsize = MIN(nmrsize, rxsize);
8012 uio_reset(uio, uio_offset(uio), UIO_SYSSPACE, UIO_READ);
8013 uio_addiov(uio, CAST_USER_ADDR_T(rxaddr), iosize);
8014 FSDBG(322, uio_offset(uio), uio_resid(uio), rxaddr, rxsize);
8015 #if UPL_DEBUG
8016 upl_ubc_alias_set(pl, (uintptr_t) current_thread(), (uintptr_t) 2);
8017 #endif /* UPL_DEBUG */
8018 OSAddAtomic64(1, &nfsstats.pageins);
8019 error = nmp->nm_funcs->nf_read_rpc_async_finish(np, req[nextwait], uio, &retsize, NULL);
8020 req[nextwait] = NULL;
8021 nextwait = (nextwait + 1) % MAXPAGINGREQS;
8022 #if CONFIG_NFS4
8023 if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error)) {
8024 lck_mtx_lock(&nmp->nm_lock);
8025 if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
8026 NP(np, "nfs_vnop_pagein: error %d, initiating recovery", error);
8027 nfs_need_recover(nmp, error);
8028 }
8029 lck_mtx_unlock(&nmp->nm_lock);
8030 restart++;
8031 goto cancel;
8032 }
8033 #endif
8034 if (error) {
8035 FSDBG(322, uio_offset(uio), uio_resid(uio), error, -1);
8036 break;
8037 }
8038 if (retsize < iosize) {
8039 /* Just zero fill the rest of the valid area. */
8040 size_t zcnt = iosize - retsize;
8041 bzero((char *)rxaddr + retsize, zcnt);
8042 FSDBG(324, uio_offset(uio), retsize, zcnt, rxaddr);
8043 uio_update(uio, zcnt);
8044 }
8045 rxaddr += iosize;
8046 rxsize -= iosize;
8047 if (txsize) {
8048 break;
8049 }
8050 }
8051 } while (!error && (txsize || rxsize));
8052
8053 restart = 0;
8054
8055 if (error) {
8056 #if CONFIG_NFS4
8057 cancel:
8058 #endif
8059 /* cancel any outstanding requests */
8060 while (req[nextwait]) {
8061 nfs_request_async_cancel(req[nextwait]);
8062 req[nextwait] = NULL;
8063 nextwait = (nextwait + 1) % MAXPAGINGREQS;
8064 }
8065 if (np->n_flag & NREVOKE) {
8066 error = EIO;
8067 } else if (restart) {
8068 if (restart <= nfs_mount_state_max_restarts(nmp)) { /* guard against no progress */
8069 if (error == NFSERR_GRACE) {
8070 tsleep(&nmp->nm_state, (PZERO - 1), "nfsgrace", 2 * hz);
8071 }
8072 if (!(error = nfs_mount_state_wait_for_recovery(nmp))) {
8073 goto tryagain;
8074 }
8075 } else {
8076 NP(np, "nfs_pagein: too many restarts, aborting");
8077 }
8078 }
8079 }
8080
8081 ubc_upl_unmap(pl);
8082
8083 if (!nofreeupl) {
8084 if (error) {
8085 /*
8086 * See comment in vnode_pagein() on handling EAGAIN, even though UPL_NOCOMMIT flag
8087 * is not set, we will not abort this upl, since VM subsystem will handle it.
8088 */
8089 if (error != EAGAIN && error != EPERM) {
8090 ubc_upl_abort_range(pl, pl_offset, size,
8091 UPL_ABORT_ERROR |
8092 UPL_ABORT_FREE_ON_EMPTY);
8093 }
8094 } else {
8095 ubc_upl_commit_range(pl, pl_offset, size,
8096 UPL_COMMIT_CLEAR_DIRTY |
8097 UPL_COMMIT_FREE_ON_EMPTY);
8098 }
8099 }
8100 return error;
8101 }
8102
8103
8104 /*
8105 * the following are needed only by nfs_pageout to know how to handle errors
8106 * see nfs_pageout comments on explanation of actions.
8107 * the errors here are copied from errno.h and errors returned by servers
8108 * are expected to match the same numbers here. If not, our actions maybe
8109 * erroneous.
8110 */
8111 char nfs_pageouterrorhandler(int);
8112 enum actiontype {NOACTION, DUMP, DUMPANDLOG, RETRY, SEVER};
8113 #define NFS_ELAST 88
8114 static u_char errorcount[NFS_ELAST + 1]; /* better be zeros when initialized */
8115 static const char errortooutcome[NFS_ELAST + 1] = {
8116 NOACTION,
8117 DUMP, /* EPERM 1 Operation not permitted */
8118 DUMP, /* ENOENT 2 No such file or directory */
8119 DUMPANDLOG, /* ESRCH 3 No such process */
8120 RETRY, /* EINTR 4 Interrupted system call */
8121 DUMP, /* EIO 5 Input/output error */
8122 DUMP, /* ENXIO 6 Device not configured */
8123 DUMPANDLOG, /* E2BIG 7 Argument list too long */
8124 DUMPANDLOG, /* ENOEXEC 8 Exec format error */
8125 DUMPANDLOG, /* EBADF 9 Bad file descriptor */
8126 DUMPANDLOG, /* ECHILD 10 No child processes */
8127 DUMPANDLOG, /* EDEADLK 11 Resource deadlock avoided - was EAGAIN */
8128 RETRY, /* ENOMEM 12 Cannot allocate memory */
8129 DUMP, /* EACCES 13 Permission denied */
8130 DUMPANDLOG, /* EFAULT 14 Bad address */
8131 DUMPANDLOG, /* ENOTBLK 15 POSIX - Block device required */
8132 RETRY, /* EBUSY 16 Device busy */
8133 DUMP, /* EEXIST 17 File exists */
8134 DUMP, /* EXDEV 18 Cross-device link */
8135 DUMP, /* ENODEV 19 Operation not supported by device */
8136 DUMP, /* ENOTDIR 20 Not a directory */
8137 DUMP, /* EISDIR 21 Is a directory */
8138 DUMP, /* EINVAL 22 Invalid argument */
8139 DUMPANDLOG, /* ENFILE 23 Too many open files in system */
8140 DUMPANDLOG, /* EMFILE 24 Too many open files */
8141 DUMPANDLOG, /* ENOTTY 25 Inappropriate ioctl for device */
8142 DUMPANDLOG, /* ETXTBSY 26 Text file busy - POSIX */
8143 DUMP, /* EFBIG 27 File too large */
8144 DUMP, /* ENOSPC 28 No space left on device */
8145 DUMPANDLOG, /* ESPIPE 29 Illegal seek */
8146 DUMP, /* EROFS 30 Read-only file system */
8147 DUMP, /* EMLINK 31 Too many links */
8148 RETRY, /* EPIPE 32 Broken pipe */
8149 /* math software */
8150 DUMPANDLOG, /* EDOM 33 Numerical argument out of domain */
8151 DUMPANDLOG, /* ERANGE 34 Result too large */
8152 RETRY, /* EAGAIN/EWOULDBLOCK 35 Resource temporarily unavailable */
8153 DUMPANDLOG, /* EINPROGRESS 36 Operation now in progress */
8154 DUMPANDLOG, /* EALREADY 37 Operation already in progress */
8155 /* ipc/network software -- argument errors */
8156 DUMPANDLOG, /* ENOTSOC 38 Socket operation on non-socket */
8157 DUMPANDLOG, /* EDESTADDRREQ 39 Destination address required */
8158 DUMPANDLOG, /* EMSGSIZE 40 Message too long */
8159 DUMPANDLOG, /* EPROTOTYPE 41 Protocol wrong type for socket */
8160 DUMPANDLOG, /* ENOPROTOOPT 42 Protocol not available */
8161 DUMPANDLOG, /* EPROTONOSUPPORT 43 Protocol not supported */
8162 DUMPANDLOG, /* ESOCKTNOSUPPORT 44 Socket type not supported */
8163 DUMPANDLOG, /* ENOTSUP 45 Operation not supported */
8164 DUMPANDLOG, /* EPFNOSUPPORT 46 Protocol family not supported */
8165 DUMPANDLOG, /* EAFNOSUPPORT 47 Address family not supported by protocol family */
8166 DUMPANDLOG, /* EADDRINUSE 48 Address already in use */
8167 DUMPANDLOG, /* EADDRNOTAVAIL 49 Can't assign requested address */
8168 /* ipc/network software -- operational errors */
8169 RETRY, /* ENETDOWN 50 Network is down */
8170 RETRY, /* ENETUNREACH 51 Network is unreachable */
8171 RETRY, /* ENETRESET 52 Network dropped connection on reset */
8172 RETRY, /* ECONNABORTED 53 Software caused connection abort */
8173 RETRY, /* ECONNRESET 54 Connection reset by peer */
8174 RETRY, /* ENOBUFS 55 No buffer space available */
8175 RETRY, /* EISCONN 56 Socket is already connected */
8176 RETRY, /* ENOTCONN 57 Socket is not connected */
8177 RETRY, /* ESHUTDOWN 58 Can't send after socket shutdown */
8178 RETRY, /* ETOOMANYREFS 59 Too many references: can't splice */
8179 RETRY, /* ETIMEDOUT 60 Operation timed out */
8180 RETRY, /* ECONNREFUSED 61 Connection refused */
8181
8182 DUMPANDLOG, /* ELOOP 62 Too many levels of symbolic links */
8183 DUMP, /* ENAMETOOLONG 63 File name too long */
8184 RETRY, /* EHOSTDOWN 64 Host is down */
8185 RETRY, /* EHOSTUNREACH 65 No route to host */
8186 DUMP, /* ENOTEMPTY 66 Directory not empty */
8187 /* quotas & mush */
8188 DUMPANDLOG, /* PROCLIM 67 Too many processes */
8189 DUMPANDLOG, /* EUSERS 68 Too many users */
8190 DUMPANDLOG, /* EDQUOT 69 Disc quota exceeded */
8191 /* Network File System */
8192 DUMP, /* ESTALE 70 Stale NFS file handle */
8193 DUMP, /* EREMOTE 71 Too many levels of remote in path */
8194 DUMPANDLOG, /* EBADRPC 72 RPC struct is bad */
8195 DUMPANDLOG, /* ERPCMISMATCH 73 RPC version wrong */
8196 DUMPANDLOG, /* EPROGUNAVAIL 74 RPC prog. not avail */
8197 DUMPANDLOG, /* EPROGMISMATCH 75 Program version wrong */
8198 DUMPANDLOG, /* EPROCUNAVAIL 76 Bad procedure for program */
8199
8200 DUMPANDLOG, /* ENOLCK 77 No locks available */
8201 DUMPANDLOG, /* ENOSYS 78 Function not implemented */
8202 DUMPANDLOG, /* EFTYPE 79 Inappropriate file type or format */
8203 DUMPANDLOG, /* EAUTH 80 Authentication error */
8204 DUMPANDLOG, /* ENEEDAUTH 81 Need authenticator */
8205 /* Intelligent device errors */
8206 DUMPANDLOG, /* EPWROFF 82 Device power is off */
8207 DUMPANDLOG, /* EDEVERR 83 Device error, e.g. paper out */
8208 DUMPANDLOG, /* EOVERFLOW 84 Value too large to be stored in data type */
8209 /* Program loading errors */
8210 DUMPANDLOG, /* EBADEXEC 85 Bad executable */
8211 DUMPANDLOG, /* EBADARCH 86 Bad CPU type in executable */
8212 DUMPANDLOG, /* ESHLIBVERS 87 Shared library version mismatch */
8213 DUMPANDLOG, /* EBADMACHO 88 Malformed Macho file */
8214 };
8215
8216 char
8217 nfs_pageouterrorhandler(int error)
8218 {
8219 if (error > NFS_ELAST) {
8220 return DUMP;
8221 } else {
8222 return errortooutcome[error];
8223 }
8224 }
8225
8226
8227 /*
8228 * vnode OP for pageout using UPL
8229 *
8230 * No buffer I/O, just RPCs straight from the mapped pages.
8231 * File size changes are not permitted in pageout.
8232 */
8233 int
8234 nfs_vnop_pageout(
8235 struct vnop_pageout_args /* {
8236 * struct vnodeop_desc *a_desc;
8237 * vnode_t a_vp;
8238 * upl_t a_pl;
8239 * vm_offset_t a_pl_offset;
8240 * off_t a_f_offset;
8241 * size_t a_size;
8242 * int a_flags;
8243 * vfs_context_t a_context;
8244 * } */*ap)
8245 {
8246 vnode_t vp = ap->a_vp;
8247 upl_t pl = ap->a_pl;
8248 upl_size_t size = (upl_size_t)ap->a_size;
8249 off_t f_offset = ap->a_f_offset;
8250 upl_offset_t pl_offset = ap->a_pl_offset;
8251 upl_offset_t pgsize;
8252 int flags = ap->a_flags;
8253 nfsnode_t np = VTONFS(vp);
8254 thread_t thd;
8255 kauth_cred_t cred;
8256 struct nfsbuf *bp;
8257 struct nfsmount *nmp = VTONMP(vp);
8258 daddr64_t lbn;
8259 int error = 0, iomode;
8260 off_t off, txoffset, rxoffset;
8261 vm_offset_t ioaddr, txaddr, rxaddr;
8262 uio_t auio;
8263 char uio_buf[UIO_SIZEOF(1)];
8264 int nofreeupl = flags & UPL_NOCOMMIT;
8265 size_t nmwsize, biosize, iosize, remsize;
8266 struct nfsreq *req[MAXPAGINGREQS];
8267 int nextsend, nextwait, wverfset, commit;
8268 uint64_t wverf, wverf2, xsize, txsize, rxsize;
8269 #if CONFIG_NFS4
8270 uint32_t stategenid = 0;
8271 #endif
8272 uint32_t vrestart = 0, restart = 0, vrestarts = 0, restarts = 0;
8273 kern_return_t kret;
8274
8275 FSDBG(323, f_offset, size, pl, pl_offset);
8276
8277 if (pl == (upl_t)NULL) {
8278 panic("nfs_pageout: no upl");
8279 }
8280
8281 if (size <= 0) {
8282 printf("nfs_pageout: invalid size %u", size);
8283 if (!nofreeupl) {
8284 ubc_upl_abort_range(pl, pl_offset, size, 0);
8285 }
8286 return EINVAL;
8287 }
8288
8289 if (!nmp) {
8290 if (!nofreeupl) {
8291 ubc_upl_abort(pl, UPL_ABORT_DUMP_PAGES | UPL_ABORT_FREE_ON_EMPTY);
8292 }
8293 return ENXIO;
8294 }
8295 biosize = nmp->nm_biosize;
8296 nmwsize = nmp->nm_wsize;
8297
8298 nfs_data_lock_noupdate(np, NFS_DATA_LOCK_SHARED);
8299
8300 /*
8301 * Check to see whether the buffer is incore.
8302 * If incore and not busy, invalidate it from the cache.
8303 */
8304 for (iosize = 0; iosize < size; iosize += xsize) {
8305 off = f_offset + iosize;
8306 /* need make sure we do things on block boundaries */
8307 xsize = biosize - (off % biosize);
8308 if (off + (off_t)xsize > f_offset + (off_t)size) {
8309 xsize = f_offset + size - off;
8310 }
8311 lbn = (daddr64_t)(off / biosize);
8312 lck_mtx_lock(&nfs_buf_mutex);
8313 if ((bp = nfs_buf_incore(np, lbn))) {
8314 FSDBG(323, off, bp, bp->nb_lflags, bp->nb_flags);
8315 if (nfs_buf_acquire(bp, NBAC_NOWAIT, 0, 0)) {
8316 lck_mtx_unlock(&nfs_buf_mutex);
8317 nfs_data_unlock_noupdate(np);
8318 /* no panic. just tell vm we are busy */
8319 if (!nofreeupl) {
8320 ubc_upl_abort_range(pl, pl_offset, size, 0);
8321 }
8322 return EBUSY;
8323 }
8324 if (bp->nb_dirtyend > 0) {
8325 /*
8326 * if there's a dirty range in the buffer, check
8327 * to see if it extends beyond the pageout region
8328 *
8329 * if the dirty region lies completely within the
8330 * pageout region, we just invalidate the buffer
8331 * because it's all being written out now anyway.
8332 *
8333 * if any of the dirty region lies outside the
8334 * pageout region, we'll try to clip the dirty
8335 * region to eliminate the portion that's being
8336 * paged out. If that's not possible, because
8337 * the dirty region extends before and after the
8338 * pageout region, then we'll just return EBUSY.
8339 */
8340 off_t boff, start, end;
8341 boff = NBOFF(bp);
8342 start = off;
8343 end = off + xsize;
8344 /* clip end to EOF */
8345 if (end > (off_t)np->n_size) {
8346 end = np->n_size;
8347 }
8348 start -= boff;
8349 end -= boff;
8350 if ((bp->nb_dirtyoff < start) &&
8351 (bp->nb_dirtyend > end)) {
8352 /*
8353 * not gonna be able to clip the dirty region
8354 *
8355 * But before returning the bad news, move the
8356 * buffer to the start of the delwri list and
8357 * give the list a push to try to flush the
8358 * buffer out.
8359 */
8360 FSDBG(323, np, bp, 0xd00deebc, EBUSY);
8361 nfs_buf_remfree(bp);
8362 TAILQ_INSERT_HEAD(&nfsbufdelwri, bp, nb_free);
8363 nfsbufdelwricnt++;
8364 nfs_buf_drop(bp);
8365 nfs_buf_delwri_push(1);
8366 lck_mtx_unlock(&nfs_buf_mutex);
8367 nfs_data_unlock_noupdate(np);
8368 if (!nofreeupl) {
8369 ubc_upl_abort_range(pl, pl_offset, size, 0);
8370 }
8371 return EBUSY;
8372 }
8373 if ((bp->nb_dirtyoff < start) ||
8374 (bp->nb_dirtyend > end)) {
8375 /* clip dirty region, if necessary */
8376 if (bp->nb_dirtyoff < start) {
8377 bp->nb_dirtyend = MIN(bp->nb_dirtyend, start);
8378 }
8379 if (bp->nb_dirtyend > end) {
8380 bp->nb_dirtyoff = MAX(bp->nb_dirtyoff, end);
8381 }
8382 FSDBG(323, bp, bp->nb_dirtyoff, bp->nb_dirtyend, 0xd00dee00);
8383 /* we're leaving this block dirty */
8384 nfs_buf_drop(bp);
8385 lck_mtx_unlock(&nfs_buf_mutex);
8386 continue;
8387 }
8388 }
8389 nfs_buf_remfree(bp);
8390 lck_mtx_unlock(&nfs_buf_mutex);
8391 SET(bp->nb_flags, NB_INVAL);
8392 nfs_node_lock_force(np);
8393 if (ISSET(bp->nb_flags, NB_NEEDCOMMIT)) {
8394 CLR(bp->nb_flags, NB_NEEDCOMMIT);
8395 np->n_needcommitcnt--;
8396 CHECK_NEEDCOMMITCNT(np);
8397 }
8398 nfs_node_unlock(np);
8399 nfs_buf_release(bp, 1);
8400 } else {
8401 lck_mtx_unlock(&nfs_buf_mutex);
8402 }
8403 }
8404
8405 thd = vfs_context_thread(ap->a_context);
8406 cred = ubc_getcred(vp);
8407 if (!IS_VALID_CRED(cred)) {
8408 cred = vfs_context_ucred(ap->a_context);
8409 }
8410
8411 nfs_node_lock_force(np);
8412 if (np->n_flag & NWRITEERR) {
8413 error = np->n_error;
8414 nfs_node_unlock(np);
8415 nfs_data_unlock_noupdate(np);
8416 if (!nofreeupl) {
8417 ubc_upl_abort_range(pl, pl_offset, size,
8418 UPL_ABORT_FREE_ON_EMPTY);
8419 }
8420 return error;
8421 }
8422 nfs_node_unlock(np);
8423
8424 if (f_offset < 0 || f_offset >= (off_t)np->n_size ||
8425 f_offset & PAGE_MASK_64 || size & PAGE_MASK_64) {
8426 nfs_data_unlock_noupdate(np);
8427 if (!nofreeupl) {
8428 ubc_upl_abort_range(pl, pl_offset, size,
8429 UPL_ABORT_FREE_ON_EMPTY);
8430 }
8431 return EINVAL;
8432 }
8433
8434 kret = ubc_upl_map(pl, &ioaddr);
8435 if (kret != KERN_SUCCESS) {
8436 panic("nfs_vnop_pageout: ubc_upl_map() failed with (%d)", kret);
8437 }
8438 ioaddr += pl_offset;
8439
8440 if ((u_quad_t)f_offset + size > np->n_size) {
8441 xsize = np->n_size - f_offset;
8442 } else {
8443 xsize = size;
8444 }
8445
8446 pgsize = (upl_offset_t)round_page_64(xsize);
8447 if ((size > pgsize) && !nofreeupl) {
8448 ubc_upl_abort_range(pl, pl_offset + pgsize, size - pgsize,
8449 UPL_ABORT_FREE_ON_EMPTY);
8450 }
8451
8452 /*
8453 * check for partial page and clear the
8454 * contents past end of the file before
8455 * releasing it in the VM page cache
8456 */
8457 if ((u_quad_t)f_offset < np->n_size && (u_quad_t)f_offset + size > np->n_size) {
8458 uint64_t io = np->n_size - f_offset;
8459 NFS_BZERO((caddr_t)(ioaddr + io), size - io);
8460 FSDBG(321, np->n_size, f_offset, f_offset + io, size - io);
8461 }
8462 nfs_data_unlock_noupdate(np);
8463
8464 auio = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_WRITE,
8465 &uio_buf, sizeof(uio_buf));
8466
8467 tryagain:
8468 #if CONFIG_NFS4
8469 if (nmp->nm_vers >= NFS_VER4) {
8470 stategenid = nmp->nm_stategenid;
8471 }
8472 #endif
8473 wverf = wverf2 = wverfset = 0;
8474 txsize = rxsize = xsize;
8475 txoffset = rxoffset = f_offset;
8476 txaddr = rxaddr = ioaddr;
8477 commit = NFS_WRITE_FILESYNC;
8478
8479 bzero(req, sizeof(req));
8480 nextsend = nextwait = 0;
8481 do {
8482 if (np->n_flag & NREVOKE) {
8483 error = EIO;
8484 break;
8485 }
8486 /* send requests while we need to and have available slots */
8487 while ((txsize > 0) && (req[nextsend] == NULL)) {
8488 iosize = (size_t)MIN(nmwsize, txsize);
8489 uio_reset(auio, txoffset, UIO_SYSSPACE, UIO_WRITE);
8490 uio_addiov(auio, CAST_USER_ADDR_T(txaddr), iosize);
8491 FSDBG(323, uio_offset(auio), iosize, txaddr, txsize);
8492 OSAddAtomic64(1, &nfsstats.pageouts);
8493 nfs_node_lock_force(np);
8494 np->n_numoutput++;
8495 nfs_node_unlock(np);
8496 vnode_startwrite(vp);
8497 iomode = NFS_WRITE_UNSTABLE;
8498 if ((error = nmp->nm_funcs->nf_write_rpc_async(np, auio, iosize, thd, cred, iomode, NULL, &req[nextsend]))) {
8499 req[nextsend] = NULL;
8500 vnode_writedone(vp);
8501 nfs_node_lock_force(np);
8502 np->n_numoutput--;
8503 nfs_node_unlock(np);
8504 break;
8505 }
8506 txaddr += iosize;
8507 txoffset += iosize;
8508 txsize -= iosize;
8509 nextsend = (nextsend + 1) % MAXPAGINGREQS;
8510 }
8511 /* wait while we need to and break out if more requests to send */
8512 while ((rxsize > 0) && req[nextwait]) {
8513 iosize = remsize = (size_t)MIN(nmwsize, rxsize);
8514 error = nmp->nm_funcs->nf_write_rpc_async_finish(np, req[nextwait], &iomode, &iosize, &wverf2);
8515 req[nextwait] = NULL;
8516 nextwait = (nextwait + 1) % MAXPAGINGREQS;
8517 vnode_writedone(vp);
8518 nfs_node_lock_force(np);
8519 np->n_numoutput--;
8520 nfs_node_unlock(np);
8521 #if CONFIG_NFS4
8522 if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error)) {
8523 lck_mtx_lock(&nmp->nm_lock);
8524 if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
8525 NP(np, "nfs_vnop_pageout: error %d, initiating recovery", error);
8526 nfs_need_recover(nmp, error);
8527 }
8528 lck_mtx_unlock(&nmp->nm_lock);
8529 restart = 1;
8530 goto cancel;
8531 }
8532 #endif
8533 if (error) {
8534 FSDBG(323, rxoffset, rxsize, error, -1);
8535 break;
8536 }
8537 if (!wverfset) {
8538 wverf = wverf2;
8539 wverfset = 1;
8540 } else if (wverf != wverf2) {
8541 /* verifier changed, so we need to restart all the writes */
8542 vrestart = 1;
8543 goto cancel;
8544 }
8545 /* Retain the lowest commitment level returned. */
8546 if (iomode < commit) {
8547 commit = iomode;
8548 }
8549 rxaddr += iosize;
8550 rxoffset += iosize;
8551 rxsize -= iosize;
8552 remsize -= iosize;
8553 if (remsize > 0) {
8554 /* need to try sending the remainder */
8555 iosize = remsize;
8556 uio_reset(auio, rxoffset, UIO_SYSSPACE, UIO_WRITE);
8557 uio_addiov(auio, CAST_USER_ADDR_T(rxaddr), remsize);
8558 iomode = NFS_WRITE_UNSTABLE;
8559 error = nfs_write_rpc2(np, auio, thd, cred, &iomode, &wverf2);
8560 #if CONFIG_NFS4
8561 if ((nmp->nm_vers >= NFS_VER4) && nfs_mount_state_error_should_restart(error)) {
8562 NP(np, "nfs_vnop_pageout: restart: error %d", error);
8563 lck_mtx_lock(&nmp->nm_lock);
8564 if ((error != NFSERR_GRACE) && (stategenid == nmp->nm_stategenid)) {
8565 NP(np, "nfs_vnop_pageout: error %d, initiating recovery", error);
8566 nfs_need_recover(nmp, error);
8567 }
8568 lck_mtx_unlock(&nmp->nm_lock);
8569 restart = 1;
8570 goto cancel;
8571 }
8572 #endif
8573 if (error) {
8574 FSDBG(323, rxoffset, rxsize, error, -1);
8575 break;
8576 }
8577 if (wverf != wverf2) {
8578 /* verifier changed, so we need to restart all the writes */
8579 vrestart = 1;
8580 goto cancel;
8581 }
8582 if (iomode < commit) {
8583 commit = iomode;
8584 }
8585 rxaddr += iosize;
8586 rxoffset += iosize;
8587 rxsize -= iosize;
8588 }
8589 if (txsize) {
8590 break;
8591 }
8592 }
8593 } while (!error && (txsize || rxsize));
8594
8595 vrestart = 0;
8596
8597 if (!error && (commit != NFS_WRITE_FILESYNC)) {
8598 error = nmp->nm_funcs->nf_commit_rpc(np, f_offset, xsize, cred, wverf);
8599 if (error == NFSERR_STALEWRITEVERF) {
8600 vrestart = 1;
8601 error = EIO;
8602 }
8603 }
8604
8605 if (error) {
8606 cancel:
8607 /* cancel any outstanding requests */
8608 while (req[nextwait]) {
8609 nfs_request_async_cancel(req[nextwait]);
8610 req[nextwait] = NULL;
8611 nextwait = (nextwait + 1) % MAXPAGINGREQS;
8612 vnode_writedone(vp);
8613 nfs_node_lock_force(np);
8614 np->n_numoutput--;
8615 nfs_node_unlock(np);
8616 }
8617 if (np->n_flag & NREVOKE) {
8618 error = EIO;
8619 } else {
8620 if (vrestart) {
8621 if (++vrestarts <= 100) { /* guard against no progress */
8622 goto tryagain;
8623 }
8624 NP(np, "nfs_pageout: too many restarts, aborting");
8625 FSDBG(323, f_offset, xsize, ERESTART, -1);
8626 }
8627 if (restart) {
8628 if (restarts <= nfs_mount_state_max_restarts(nmp)) { /* guard against no progress */
8629 if (error == NFSERR_GRACE) {
8630 tsleep(&nmp->nm_state, (PZERO - 1), "nfsgrace", 2 * hz);
8631 }
8632 if (!(error = nfs_mount_state_wait_for_recovery(nmp))) {
8633 goto tryagain;
8634 }
8635 } else {
8636 NP(np, "nfs_pageout: too many restarts, aborting");
8637 FSDBG(323, f_offset, xsize, ERESTART, -1);
8638 }
8639 }
8640 }
8641 }
8642
8643 ubc_upl_unmap(pl);
8644
8645 /*
8646 * We've had several different solutions on what to do when the pageout
8647 * gets an error. If we don't handle it, and return an error to the
8648 * caller, vm, it will retry . This can end in endless looping
8649 * between vm and here doing retries of the same page. Doing a dump
8650 * back to vm, will get it out of vm's knowledge and we lose whatever
8651 * data existed. This is risky, but in some cases necessary. For
8652 * example, the initial fix here was to do that for ESTALE. In that case
8653 * the server is telling us that the file is no longer the same. We
8654 * would not want to keep paging out to that. We also saw some 151
8655 * errors from Auspex server and NFSv3 can return errors higher than
8656 * ELAST. Those along with NFS known server errors we will "dump" from
8657 * vm. Errors we don't expect to occur, we dump and log for further
8658 * analysis. Errors that could be transient, networking ones,
8659 * we let vm "retry". Lastly, errors that we retry, but may have potential
8660 * to storm the network, we "retrywithsleep". "sever" will be used in
8661 * in the future to dump all pages of object for cases like ESTALE.
8662 * All this is the basis for the states returned and first guesses on
8663 * error handling. Tweaking expected as more statistics are gathered.
8664 * Note, in the long run we may need another more robust solution to
8665 * have some kind of persistant store when the vm cannot dump nor keep
8666 * retrying as a solution, but this would be a file architectural change
8667 */
8668 if (!nofreeupl) { /* otherwise stacked file system has to handle this */
8669 if (error) {
8670 int abortflags = 0;
8671 char action = nfs_pageouterrorhandler(error);
8672
8673 switch (action) {
8674 case DUMP:
8675 abortflags = UPL_ABORT_DUMP_PAGES | UPL_ABORT_FREE_ON_EMPTY;
8676 break;
8677 case DUMPANDLOG:
8678 abortflags = UPL_ABORT_DUMP_PAGES | UPL_ABORT_FREE_ON_EMPTY;
8679 if (error <= NFS_ELAST) {
8680 if ((errorcount[error] % 100) == 0) {
8681 NP(np, "nfs_pageout: unexpected error %d. dumping vm page", error);
8682 }
8683 errorcount[error]++;
8684 }
8685 break;
8686 case RETRY:
8687 abortflags = UPL_ABORT_FREE_ON_EMPTY;
8688 break;
8689 case SEVER: /* not implemented */
8690 default:
8691 NP(np, "nfs_pageout: action %d not expected", action);
8692 break;
8693 }
8694
8695 ubc_upl_abort_range(pl, pl_offset, pgsize, abortflags);
8696 /* return error in all cases above */
8697 } else {
8698 ubc_upl_commit_range(pl, pl_offset, pgsize,
8699 UPL_COMMIT_CLEAR_DIRTY |
8700 UPL_COMMIT_FREE_ON_EMPTY);
8701 }
8702 }
8703 return error;
8704 }
8705
8706 /* Blktooff derives file offset given a logical block number */
8707 int
8708 nfs_vnop_blktooff(
8709 struct vnop_blktooff_args /* {
8710 * struct vnodeop_desc *a_desc;
8711 * vnode_t a_vp;
8712 * daddr64_t a_lblkno;
8713 * off_t *a_offset;
8714 * } */*ap)
8715 {
8716 int biosize;
8717 vnode_t vp = ap->a_vp;
8718 struct nfsmount *nmp = VTONMP(vp);
8719
8720 if (nfs_mount_gone(nmp)) {
8721 return ENXIO;
8722 }
8723 biosize = nmp->nm_biosize;
8724
8725 *ap->a_offset = (off_t)(ap->a_lblkno * biosize);
8726
8727 return 0;
8728 }
8729
8730 int
8731 nfs_vnop_offtoblk(
8732 struct vnop_offtoblk_args /* {
8733 * struct vnodeop_desc *a_desc;
8734 * vnode_t a_vp;
8735 * off_t a_offset;
8736 * daddr64_t *a_lblkno;
8737 * } */*ap)
8738 {
8739 int biosize;
8740 vnode_t vp = ap->a_vp;
8741 struct nfsmount *nmp = VTONMP(vp);
8742
8743 if (nfs_mount_gone(nmp)) {
8744 return ENXIO;
8745 }
8746 biosize = nmp->nm_biosize;
8747
8748 *ap->a_lblkno = (daddr64_t)(ap->a_offset / biosize);
8749
8750 return 0;
8751 }
8752
8753 /*
8754 * vnode change monitoring
8755 */
8756 int
8757 nfs_vnop_monitor(
8758 struct vnop_monitor_args /* {
8759 * struct vnodeop_desc *a_desc;
8760 * vnode_t a_vp;
8761 * uint32_t a_events;
8762 * uint32_t a_flags;
8763 * void *a_handle;
8764 * vfs_context_t a_context;
8765 * } */*ap)
8766 {
8767 nfsnode_t np = VTONFS(ap->a_vp);
8768 struct nfsmount *nmp = VTONMP(ap->a_vp);
8769 int error = 0;
8770
8771 if (nfs_mount_gone(nmp)) {
8772 return ENXIO;
8773 }
8774
8775 /* make sure that the vnode's monitoring status is up to date */
8776 lck_mtx_lock(&nmp->nm_lock);
8777 if (vnode_ismonitored(ap->a_vp)) {
8778 /* This vnode is currently being monitored, make sure we're tracking it. */
8779 if (np->n_monlink.le_next == NFSNOLIST) {
8780 LIST_INSERT_HEAD(&nmp->nm_monlist, np, n_monlink);
8781 nfs_mount_sock_thread_wake(nmp);
8782 }
8783 } else {
8784 /* This vnode is no longer being monitored, make sure we're not tracking it. */
8785 /* Wait for any in-progress getattr to complete first. */
8786 while (np->n_mflag & NMMONSCANINPROG) {
8787 struct timespec ts = { .tv_sec = 1, .tv_nsec = 0 };
8788 np->n_mflag |= NMMONSCANWANT;
8789 msleep(&np->n_mflag, &nmp->nm_lock, PZERO - 1, "nfswaitmonscan", &ts);
8790 }
8791 if (np->n_monlink.le_next != NFSNOLIST) {
8792 LIST_REMOVE(np, n_monlink);
8793 np->n_monlink.le_next = NFSNOLIST;
8794 }
8795 }
8796 lck_mtx_unlock(&nmp->nm_lock);
8797
8798 return error;
8799 }
8800
8801 /*
8802 * Send a vnode notification for the given events.
8803 */
8804 void
8805 nfs_vnode_notify(nfsnode_t np, uint32_t events)
8806 {
8807 struct nfsmount *nmp = NFSTONMP(np);
8808 struct nfs_vattr *nvattr;
8809 struct vnode_attr vattr, *vap = NULL;
8810 struct timeval now;
8811
8812 microuptime(&now);
8813 if ((np->n_evtstamp == now.tv_sec) || !nmp) {
8814 /* delay sending this notify */
8815 np->n_events |= events;
8816 return;
8817 }
8818 events |= np->n_events;
8819 np->n_events = 0;
8820 np->n_evtstamp = now.tv_sec;
8821 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
8822
8823 vfs_get_notify_attributes(&vattr);
8824 if (!nfs_getattrcache(np, nvattr, 0)) {
8825 vap = &vattr;
8826 VATTR_INIT(vap);
8827
8828 VATTR_RETURN(vap, va_fsid, vfs_statfs(nmp->nm_mountp)->f_fsid.val[0]);
8829 VATTR_RETURN(vap, va_fileid, nvattr->nva_fileid);
8830 VATTR_RETURN(vap, va_mode, nvattr->nva_mode);
8831 VATTR_RETURN(vap, va_uid, nvattr->nva_uid);
8832 VATTR_RETURN(vap, va_gid, nvattr->nva_gid);
8833 VATTR_RETURN(vap, va_nlink, nvattr->nva_nlink);
8834 }
8835 vnode_notify(NFSTOV(np), events, vap);
8836 FREE(nvattr, M_TEMP);
8837 }
8838
8839 #endif /* CONFIG_NFS_CLIENT */
mirror of XNU