Deleted Added
full compact
tmpfs_subr.c (194766) tmpfs_subr.c (197953)
1/* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
2
3/*-
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33/*
34 * Efficient memory file system supporting functions.
35 */
36#include <sys/cdefs.h>
1/* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
2
3/*-
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * 2005 program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33/*
34 * Efficient memory file system supporting functions.
35 */
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: head/sys/fs/tmpfs/tmpfs_subr.c 194766 2009-06-23 20:45:22Z kib $");
37__FBSDID("$FreeBSD: head/sys/fs/tmpfs/tmpfs_subr.c 197953 2009-10-11 07:03:56Z delphij $");
38
39#include <sys/param.h>
40#include <sys/namei.h>
41#include <sys/priv.h>
42#include <sys/proc.h>
43#include <sys/stat.h>
44#include <sys/systm.h>
45#include <sys/vnode.h>
46#include <sys/vmmeter.h>
47
48#include <vm/vm.h>
49#include <vm/vm_object.h>
50#include <vm/vm_page.h>
51#include <vm/vm_pager.h>
52#include <vm/vm_extern.h>
53
54#include <fs/tmpfs/tmpfs.h>
55#include <fs/tmpfs/tmpfs_fifoops.h>
56#include <fs/tmpfs/tmpfs_vnops.h>
57
58/* --------------------------------------------------------------------- */
59
60/*
61 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
62 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
63 * using the credentials of the process 'p'.
64 *
65 * If the node type is set to 'VDIR', then the parent parameter must point
66 * to the parent directory of the node being created. It may only be NULL
67 * while allocating the root node.
68 *
69 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
70 * specifies the device the node represents.
71 *
72 * If the node type is set to 'VLNK', then the parameter target specifies
73 * the file name of the target file for the symbolic link that is being
74 * created.
75 *
76 * Note that new nodes are retrieved from the available list if it has
77 * items or, if it is empty, from the node pool as long as there is enough
78 * space to create them.
79 *
80 * Returns zero on success or an appropriate error code on failure.
81 */
82int
83tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
84 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
85 char *target, dev_t rdev, struct tmpfs_node **node)
86{
87 struct tmpfs_node *nnode;
88
89 /* If the root directory of the 'tmp' file system is not yet
90 * allocated, this must be the request to do it. */
91 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
92
93 MPASS(IFF(type == VLNK, target != NULL));
94 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
95
96 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max)
97 return (ENOSPC);
98
99 nnode = (struct tmpfs_node *)uma_zalloc_arg(
100 tmp->tm_node_pool, tmp, M_WAITOK);
101
102 /* Generic initialization. */
103 nnode->tn_type = type;
104 vfs_timestamp(&nnode->tn_atime);
105 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
106 nnode->tn_atime;
107 nnode->tn_uid = uid;
108 nnode->tn_gid = gid;
109 nnode->tn_mode = mode;
110 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
111
112 /* Type-specific initialization. */
113 switch (nnode->tn_type) {
114 case VBLK:
115 case VCHR:
116 nnode->tn_rdev = rdev;
117 break;
118
119 case VDIR:
120 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
121 MPASS(parent != nnode);
122 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
123 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
124 nnode->tn_dir.tn_readdir_lastn = 0;
125 nnode->tn_dir.tn_readdir_lastp = NULL;
126 nnode->tn_links++;
38
39#include <sys/param.h>
40#include <sys/namei.h>
41#include <sys/priv.h>
42#include <sys/proc.h>
43#include <sys/stat.h>
44#include <sys/systm.h>
45#include <sys/vnode.h>
46#include <sys/vmmeter.h>
47
48#include <vm/vm.h>
49#include <vm/vm_object.h>
50#include <vm/vm_page.h>
51#include <vm/vm_pager.h>
52#include <vm/vm_extern.h>
53
54#include <fs/tmpfs/tmpfs.h>
55#include <fs/tmpfs/tmpfs_fifoops.h>
56#include <fs/tmpfs/tmpfs_vnops.h>
57
58/* --------------------------------------------------------------------- */
59
60/*
61 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
62 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
63 * using the credentials of the process 'p'.
64 *
65 * If the node type is set to 'VDIR', then the parent parameter must point
66 * to the parent directory of the node being created. It may only be NULL
67 * while allocating the root node.
68 *
69 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
70 * specifies the device the node represents.
71 *
72 * If the node type is set to 'VLNK', then the parameter target specifies
73 * the file name of the target file for the symbolic link that is being
74 * created.
75 *
76 * Note that new nodes are retrieved from the available list if it has
77 * items or, if it is empty, from the node pool as long as there is enough
78 * space to create them.
79 *
80 * Returns zero on success or an appropriate error code on failure.
81 */
82int
83tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
84 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
85 char *target, dev_t rdev, struct tmpfs_node **node)
86{
87 struct tmpfs_node *nnode;
88
89 /* If the root directory of the 'tmp' file system is not yet
90 * allocated, this must be the request to do it. */
91 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
92
93 MPASS(IFF(type == VLNK, target != NULL));
94 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
95
96 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max)
97 return (ENOSPC);
98
99 nnode = (struct tmpfs_node *)uma_zalloc_arg(
100 tmp->tm_node_pool, tmp, M_WAITOK);
101
102 /* Generic initialization. */
103 nnode->tn_type = type;
104 vfs_timestamp(&nnode->tn_atime);
105 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
106 nnode->tn_atime;
107 nnode->tn_uid = uid;
108 nnode->tn_gid = gid;
109 nnode->tn_mode = mode;
110 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
111
112 /* Type-specific initialization. */
113 switch (nnode->tn_type) {
114 case VBLK:
115 case VCHR:
116 nnode->tn_rdev = rdev;
117 break;
118
119 case VDIR:
120 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
121 MPASS(parent != nnode);
122 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
123 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
124 nnode->tn_dir.tn_readdir_lastn = 0;
125 nnode->tn_dir.tn_readdir_lastp = NULL;
126 nnode->tn_links++;
127 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
127 nnode->tn_dir.tn_parent->tn_links++;
128 nnode->tn_dir.tn_parent->tn_links++;
129 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
128 break;
129
130 case VFIFO:
131 /* FALLTHROUGH */
132 case VSOCK:
133 break;
134
135 case VLNK:
136 MPASS(strlen(target) < MAXPATHLEN);
137 nnode->tn_size = strlen(target);
138 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
139 M_WAITOK);
140 memcpy(nnode->tn_link, target, nnode->tn_size);
141 break;
142
143 case VREG:
144 nnode->tn_reg.tn_aobj =
145 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
146 NULL /* XXXKIB - tmpfs needs swap reservation */);
147 nnode->tn_reg.tn_aobj_pages = 0;
148 break;
149
150 default:
151 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
152 }
153
154 TMPFS_LOCK(tmp);
155 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
156 tmp->tm_nodes_inuse++;
157 TMPFS_UNLOCK(tmp);
158
159 *node = nnode;
160 return 0;
161}
162
163/* --------------------------------------------------------------------- */
164
165/*
166 * Destroys the node pointed to by node from the file system 'tmp'.
167 * If the node does not belong to the given mount point, the results are
168 * unpredicted.
169 *
170 * If the node references a directory; no entries are allowed because
171 * their removal could need a recursive algorithm, something forbidden in
172 * kernel space. Furthermore, there is not need to provide such
173 * functionality (recursive removal) because the only primitives offered
174 * to the user are the removal of empty directories and the deletion of
175 * individual files.
176 *
177 * Note that nodes are not really deleted; in fact, when a node has been
178 * allocated, it cannot be deleted during the whole life of the file
179 * system. Instead, they are moved to the available list and remain there
180 * until reused.
181 */
182void
183tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
184{
185 size_t pages = 0;
186
187#ifdef INVARIANTS
188 TMPFS_NODE_LOCK(node);
189 MPASS(node->tn_vnode == NULL);
130 break;
131
132 case VFIFO:
133 /* FALLTHROUGH */
134 case VSOCK:
135 break;
136
137 case VLNK:
138 MPASS(strlen(target) < MAXPATHLEN);
139 nnode->tn_size = strlen(target);
140 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
141 M_WAITOK);
142 memcpy(nnode->tn_link, target, nnode->tn_size);
143 break;
144
145 case VREG:
146 nnode->tn_reg.tn_aobj =
147 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
148 NULL /* XXXKIB - tmpfs needs swap reservation */);
149 nnode->tn_reg.tn_aobj_pages = 0;
150 break;
151
152 default:
153 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
154 }
155
156 TMPFS_LOCK(tmp);
157 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
158 tmp->tm_nodes_inuse++;
159 TMPFS_UNLOCK(tmp);
160
161 *node = nnode;
162 return 0;
163}
164
165/* --------------------------------------------------------------------- */
166
167/*
168 * Destroys the node pointed to by node from the file system 'tmp'.
169 * If the node does not belong to the given mount point, the results are
170 * unpredicted.
171 *
172 * If the node references a directory; no entries are allowed because
173 * their removal could need a recursive algorithm, something forbidden in
174 * kernel space. Furthermore, there is not need to provide such
175 * functionality (recursive removal) because the only primitives offered
176 * to the user are the removal of empty directories and the deletion of
177 * individual files.
178 *
179 * Note that nodes are not really deleted; in fact, when a node has been
180 * allocated, it cannot be deleted during the whole life of the file
181 * system. Instead, they are moved to the available list and remain there
182 * until reused.
183 */
184void
185tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
186{
187 size_t pages = 0;
188
189#ifdef INVARIANTS
190 TMPFS_NODE_LOCK(node);
191 MPASS(node->tn_vnode == NULL);
192 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
190 TMPFS_NODE_UNLOCK(node);
191#endif
192
193 TMPFS_LOCK(tmp);
194 LIST_REMOVE(node, tn_entries);
195 tmp->tm_nodes_inuse--;
196 TMPFS_UNLOCK(tmp);
197
198 switch (node->tn_type) {
199 case VNON:
200 /* Do not do anything. VNON is provided to let the
201 * allocation routine clean itself easily by avoiding
202 * duplicating code in it. */
203 /* FALLTHROUGH */
204 case VBLK:
205 /* FALLTHROUGH */
206 case VCHR:
207 /* FALLTHROUGH */
208 case VDIR:
209 /* FALLTHROUGH */
210 case VFIFO:
211 /* FALLTHROUGH */
212 case VSOCK:
213 break;
214
215 case VLNK:
216 free(node->tn_link, M_TMPFSNAME);
217 break;
218
219 case VREG:
220 if (node->tn_reg.tn_aobj != NULL)
221 vm_object_deallocate(node->tn_reg.tn_aobj);
222 pages = node->tn_reg.tn_aobj_pages;
223 break;
224
225 default:
226 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
227 }
228
229 free_unr(tmp->tm_ino_unr, node->tn_id);
230 uma_zfree(tmp->tm_node_pool, node);
231
232 TMPFS_LOCK(tmp);
233 tmp->tm_pages_used -= pages;
234 TMPFS_UNLOCK(tmp);
235}
236
237/* --------------------------------------------------------------------- */
238
239/*
240 * Allocates a new directory entry for the node node with a name of name.
241 * The new directory entry is returned in *de.
242 *
243 * The link count of node is increased by one to reflect the new object
244 * referencing it.
245 *
246 * Returns zero on success or an appropriate error code on failure.
247 */
248int
249tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
250 const char *name, uint16_t len, struct tmpfs_dirent **de)
251{
252 struct tmpfs_dirent *nde;
253
254 nde = (struct tmpfs_dirent *)uma_zalloc(
255 tmp->tm_dirent_pool, M_WAITOK);
256 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
257 nde->td_namelen = len;
258 memcpy(nde->td_name, name, len);
259
260 nde->td_node = node;
261 node->tn_links++;
262
263 *de = nde;
264
265 return 0;
266}
267
268/* --------------------------------------------------------------------- */
269
270/*
271 * Frees a directory entry. It is the caller's responsibility to destroy
272 * the node referenced by it if needed.
273 *
274 * The link count of node is decreased by one to reflect the removal of an
275 * object that referenced it. This only happens if 'node_exists' is true;
276 * otherwise the function will not access the node referred to by the
277 * directory entry, as it may already have been released from the outside.
278 */
279void
280tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
281 boolean_t node_exists)
282{
283 if (node_exists) {
284 struct tmpfs_node *node;
285
286 node = de->td_node;
287
288 MPASS(node->tn_links > 0);
289 node->tn_links--;
290 }
291
292 free(de->td_name, M_TMPFSNAME);
293 uma_zfree(tmp->tm_dirent_pool, de);
294}
295
296/* --------------------------------------------------------------------- */
297
298/*
299 * Allocates a new vnode for the node node or returns a new reference to
300 * an existing one if the node had already a vnode referencing it. The
301 * resulting locked vnode is returned in *vpp.
302 *
303 * Returns zero on success or an appropriate error code on failure.
304 */
305int
306tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
307 struct vnode **vpp)
308{
309 int error = 0;
310 struct vnode *vp;
311
312loop:
313 TMPFS_NODE_LOCK(node);
314 if ((vp = node->tn_vnode) != NULL) {
193 TMPFS_NODE_UNLOCK(node);
194#endif
195
196 TMPFS_LOCK(tmp);
197 LIST_REMOVE(node, tn_entries);
198 tmp->tm_nodes_inuse--;
199 TMPFS_UNLOCK(tmp);
200
201 switch (node->tn_type) {
202 case VNON:
203 /* Do not do anything. VNON is provided to let the
204 * allocation routine clean itself easily by avoiding
205 * duplicating code in it. */
206 /* FALLTHROUGH */
207 case VBLK:
208 /* FALLTHROUGH */
209 case VCHR:
210 /* FALLTHROUGH */
211 case VDIR:
212 /* FALLTHROUGH */
213 case VFIFO:
214 /* FALLTHROUGH */
215 case VSOCK:
216 break;
217
218 case VLNK:
219 free(node->tn_link, M_TMPFSNAME);
220 break;
221
222 case VREG:
223 if (node->tn_reg.tn_aobj != NULL)
224 vm_object_deallocate(node->tn_reg.tn_aobj);
225 pages = node->tn_reg.tn_aobj_pages;
226 break;
227
228 default:
229 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
230 }
231
232 free_unr(tmp->tm_ino_unr, node->tn_id);
233 uma_zfree(tmp->tm_node_pool, node);
234
235 TMPFS_LOCK(tmp);
236 tmp->tm_pages_used -= pages;
237 TMPFS_UNLOCK(tmp);
238}
239
240/* --------------------------------------------------------------------- */
241
242/*
243 * Allocates a new directory entry for the node node with a name of name.
244 * The new directory entry is returned in *de.
245 *
246 * The link count of node is increased by one to reflect the new object
247 * referencing it.
248 *
249 * Returns zero on success or an appropriate error code on failure.
250 */
251int
252tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
253 const char *name, uint16_t len, struct tmpfs_dirent **de)
254{
255 struct tmpfs_dirent *nde;
256
257 nde = (struct tmpfs_dirent *)uma_zalloc(
258 tmp->tm_dirent_pool, M_WAITOK);
259 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
260 nde->td_namelen = len;
261 memcpy(nde->td_name, name, len);
262
263 nde->td_node = node;
264 node->tn_links++;
265
266 *de = nde;
267
268 return 0;
269}
270
271/* --------------------------------------------------------------------- */
272
273/*
274 * Frees a directory entry. It is the caller's responsibility to destroy
275 * the node referenced by it if needed.
276 *
277 * The link count of node is decreased by one to reflect the removal of an
278 * object that referenced it. This only happens if 'node_exists' is true;
279 * otherwise the function will not access the node referred to by the
280 * directory entry, as it may already have been released from the outside.
281 */
282void
283tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
284 boolean_t node_exists)
285{
286 if (node_exists) {
287 struct tmpfs_node *node;
288
289 node = de->td_node;
290
291 MPASS(node->tn_links > 0);
292 node->tn_links--;
293 }
294
295 free(de->td_name, M_TMPFSNAME);
296 uma_zfree(tmp->tm_dirent_pool, de);
297}
298
299/* --------------------------------------------------------------------- */
300
301/*
302 * Allocates a new vnode for the node node or returns a new reference to
303 * an existing one if the node had already a vnode referencing it. The
304 * resulting locked vnode is returned in *vpp.
305 *
306 * Returns zero on success or an appropriate error code on failure.
307 */
308int
309tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
310 struct vnode **vpp)
311{
312 int error = 0;
313 struct vnode *vp;
314
315loop:
316 TMPFS_NODE_LOCK(node);
317 if ((vp = node->tn_vnode) != NULL) {
318 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
315 VI_LOCK(vp);
316 TMPFS_NODE_UNLOCK(node);
317 vholdl(vp);
318 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, curthread);
319 vdrop(vp);
320
321 /*
322 * Make sure the vnode is still there after
323 * getting the interlock to avoid racing a free.
324 */
325 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
326 vput(vp);
327 goto loop;
328 }
329
330 goto out;
331 }
332
319 VI_LOCK(vp);
320 TMPFS_NODE_UNLOCK(node);
321 vholdl(vp);
322 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, curthread);
323 vdrop(vp);
324
325 /*
326 * Make sure the vnode is still there after
327 * getting the interlock to avoid racing a free.
328 */
329 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
330 vput(vp);
331 goto loop;
332 }
333
334 goto out;
335 }
336
337 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
338 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
339 TMPFS_NODE_UNLOCK(node);
340 error = ENOENT;
341 vp = NULL;
342 goto out;
343 }
344
333 /*
334 * otherwise lock the vp list while we call getnewvnode
335 * since that can block.
336 */
337 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
338 node->tn_vpstate |= TMPFS_VNODE_WANT;
339 error = msleep((caddr_t) &node->tn_vpstate,
340 TMPFS_NODE_MTX(node), PDROP | PCATCH,
341 "tmpfs_alloc_vp", 0);
342 if (error)
343 return error;
344
345 goto loop;
346 } else
347 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
348
349 TMPFS_NODE_UNLOCK(node);
350
351 /* Get a new vnode and associate it with our node. */
352 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
353 if (error != 0)
354 goto unlock;
355 MPASS(vp != NULL);
356
357 (void) vn_lock(vp, lkflag | LK_RETRY);
358
359 vp->v_data = node;
360 vp->v_type = node->tn_type;
361
362 /* Type-specific initialization. */
363 switch (node->tn_type) {
364 case VBLK:
365 /* FALLTHROUGH */
366 case VCHR:
367 /* FALLTHROUGH */
368 case VLNK:
369 /* FALLTHROUGH */
370 case VREG:
371 /* FALLTHROUGH */
372 case VSOCK:
373 break;
374 case VFIFO:
375 vp->v_op = &tmpfs_fifoop_entries;
376 break;
377 case VDIR:
345 /*
346 * otherwise lock the vp list while we call getnewvnode
347 * since that can block.
348 */
349 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
350 node->tn_vpstate |= TMPFS_VNODE_WANT;
351 error = msleep((caddr_t) &node->tn_vpstate,
352 TMPFS_NODE_MTX(node), PDROP | PCATCH,
353 "tmpfs_alloc_vp", 0);
354 if (error)
355 return error;
356
357 goto loop;
358 } else
359 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
360
361 TMPFS_NODE_UNLOCK(node);
362
363 /* Get a new vnode and associate it with our node. */
364 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
365 if (error != 0)
366 goto unlock;
367 MPASS(vp != NULL);
368
369 (void) vn_lock(vp, lkflag | LK_RETRY);
370
371 vp->v_data = node;
372 vp->v_type = node->tn_type;
373
374 /* Type-specific initialization. */
375 switch (node->tn_type) {
376 case VBLK:
377 /* FALLTHROUGH */
378 case VCHR:
379 /* FALLTHROUGH */
380 case VLNK:
381 /* FALLTHROUGH */
382 case VREG:
383 /* FALLTHROUGH */
384 case VSOCK:
385 break;
386 case VFIFO:
387 vp->v_op = &tmpfs_fifoop_entries;
388 break;
389 case VDIR:
390 MPASS(node->tn_dir.tn_parent != NULL);
378 if (node->tn_dir.tn_parent == node)
379 vp->v_vflag |= VV_ROOT;
380 break;
381
382 default:
383 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
384 }
385
386 vnode_pager_setsize(vp, node->tn_size);
387 error = insmntque(vp, mp);
388 if (error)
389 vp = NULL;
390
391unlock:
392 TMPFS_NODE_LOCK(node);
393
394 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
395 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
396 node->tn_vnode = vp;
397
398 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
399 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
400 TMPFS_NODE_UNLOCK(node);
401 wakeup((caddr_t) &node->tn_vpstate);
402 } else
403 TMPFS_NODE_UNLOCK(node);
404
405out:
406 *vpp = vp;
407
408 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
409#ifdef INVARIANTS
410 TMPFS_NODE_LOCK(node);
411 MPASS(*vpp == node->tn_vnode);
412 TMPFS_NODE_UNLOCK(node);
413#endif
414
415 return error;
416}
417
418/* --------------------------------------------------------------------- */
419
420/*
421 * Destroys the association between the vnode vp and the node it
422 * references.
423 */
424void
425tmpfs_free_vp(struct vnode *vp)
426{
427 struct tmpfs_node *node;
428
429 node = VP_TO_TMPFS_NODE(vp);
430
391 if (node->tn_dir.tn_parent == node)
392 vp->v_vflag |= VV_ROOT;
393 break;
394
395 default:
396 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
397 }
398
399 vnode_pager_setsize(vp, node->tn_size);
400 error = insmntque(vp, mp);
401 if (error)
402 vp = NULL;
403
404unlock:
405 TMPFS_NODE_LOCK(node);
406
407 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
408 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
409 node->tn_vnode = vp;
410
411 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
412 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
413 TMPFS_NODE_UNLOCK(node);
414 wakeup((caddr_t) &node->tn_vpstate);
415 } else
416 TMPFS_NODE_UNLOCK(node);
417
418out:
419 *vpp = vp;
420
421 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
422#ifdef INVARIANTS
423 TMPFS_NODE_LOCK(node);
424 MPASS(*vpp == node->tn_vnode);
425 TMPFS_NODE_UNLOCK(node);
426#endif
427
428 return error;
429}
430
431/* --------------------------------------------------------------------- */
432
433/*
434 * Destroys the association between the vnode vp and the node it
435 * references.
436 */
437void
438tmpfs_free_vp(struct vnode *vp)
439{
440 struct tmpfs_node *node;
441
442 node = VP_TO_TMPFS_NODE(vp);
443
431 TMPFS_NODE_LOCK(node);
444 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
432 node->tn_vnode = NULL;
433 vp->v_data = NULL;
445 node->tn_vnode = NULL;
446 vp->v_data = NULL;
434 TMPFS_NODE_UNLOCK(node);
435}
436
437/* --------------------------------------------------------------------- */
438
439/*
440 * Allocates a new file of type 'type' and adds it to the parent directory
441 * 'dvp'; this addition is done using the component name given in 'cnp'.
442 * The ownership of the new file is automatically assigned based on the
443 * credentials of the caller (through 'cnp'), the group is set based on
444 * the parent directory and the mode is determined from the 'vap' argument.
445 * If successful, *vpp holds a vnode to the newly created file and zero
446 * is returned. Otherwise *vpp is NULL and the function returns an
447 * appropriate error code.
448 */
449int
450tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
451 struct componentname *cnp, char *target)
452{
453 int error;
454 struct tmpfs_dirent *de;
455 struct tmpfs_mount *tmp;
456 struct tmpfs_node *dnode;
457 struct tmpfs_node *node;
458 struct tmpfs_node *parent;
459
460 MPASS(VOP_ISLOCKED(dvp));
461 MPASS(cnp->cn_flags & HASBUF);
462
463 tmp = VFS_TO_TMPFS(dvp->v_mount);
464 dnode = VP_TO_TMPFS_DIR(dvp);
465 *vpp = NULL;
466
467 /* If the entry we are creating is a directory, we cannot overflow
468 * the number of links of its parent, because it will get a new
469 * link. */
470 if (vap->va_type == VDIR) {
471 /* Ensure that we do not overflow the maximum number of links
472 * imposed by the system. */
473 MPASS(dnode->tn_links <= LINK_MAX);
474 if (dnode->tn_links == LINK_MAX) {
475 error = EMLINK;
476 goto out;
477 }
478
479 parent = dnode;
480 MPASS(parent != NULL);
481 } else
482 parent = NULL;
483
484 /* Allocate a node that represents the new file. */
485 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
486 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
487 if (error != 0)
488 goto out;
489
490 /* Allocate a directory entry that points to the new file. */
491 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
492 &de);
493 if (error != 0) {
494 tmpfs_free_node(tmp, node);
495 goto out;
496 }
497
498 /* Allocate a vnode for the new file. */
499 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
500 if (error != 0) {
501 tmpfs_free_dirent(tmp, de, TRUE);
502 tmpfs_free_node(tmp, node);
503 goto out;
504 }
505
506 /* Now that all required items are allocated, we can proceed to
507 * insert the new node into the directory, an operation that
508 * cannot fail. */
509 tmpfs_dir_attach(dvp, de);
510
511out:
512
513 return error;
514}
515
516/* --------------------------------------------------------------------- */
517
518/*
519 * Attaches the directory entry de to the directory represented by vp.
520 * Note that this does not change the link count of the node pointed by
521 * the directory entry, as this is done by tmpfs_alloc_dirent.
522 */
523void
524tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
525{
526 struct tmpfs_node *dnode;
527
528 ASSERT_VOP_ELOCKED(vp, __func__);
529 dnode = VP_TO_TMPFS_DIR(vp);
530 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
531 dnode->tn_size += sizeof(struct tmpfs_dirent);
532 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
533 TMPFS_NODE_MODIFIED;
534}
535
536/* --------------------------------------------------------------------- */
537
538/*
539 * Detaches the directory entry de from the directory represented by vp.
540 * Note that this does not change the link count of the node pointed by
541 * the directory entry, as this is done by tmpfs_free_dirent.
542 */
543void
544tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
545{
546 struct tmpfs_node *dnode;
547
548 ASSERT_VOP_ELOCKED(vp, __func__);
549 dnode = VP_TO_TMPFS_DIR(vp);
550
551 if (dnode->tn_dir.tn_readdir_lastp == de) {
552 dnode->tn_dir.tn_readdir_lastn = 0;
553 dnode->tn_dir.tn_readdir_lastp = NULL;
554 }
555
556 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
557 dnode->tn_size -= sizeof(struct tmpfs_dirent);
558 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
559 TMPFS_NODE_MODIFIED;
560}
561
562/* --------------------------------------------------------------------- */
563
564/*
565 * Looks for a directory entry in the directory represented by node.
566 * 'cnp' describes the name of the entry to look for. Note that the .
567 * and .. components are not allowed as they do not physically exist
568 * within directories.
569 *
570 * Returns a pointer to the entry when found, otherwise NULL.
571 */
572struct tmpfs_dirent *
573tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
574 struct componentname *cnp)
575{
576 boolean_t found;
577 struct tmpfs_dirent *de;
578
579 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
580 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
581 cnp->cn_nameptr[1] == '.')));
582 TMPFS_VALIDATE_DIR(node);
583
584 found = 0;
585 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
586 if (f != NULL && de->td_node != f)
587 continue;
588 MPASS(cnp->cn_namelen < 0xffff);
589 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
590 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
591 found = 1;
592 break;
593 }
594 }
595 node->tn_status |= TMPFS_NODE_ACCESSED;
596
597 return found ? de : NULL;
598}
599
600/* --------------------------------------------------------------------- */
601
602/*
603 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
604 * directory and returns it in the uio space. The function returns 0
605 * on success, -1 if there was not enough space in the uio structure to
606 * hold the directory entry or an appropriate error code if another
607 * error happens.
608 */
609int
610tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
611{
612 int error;
613 struct dirent dent;
614
615 TMPFS_VALIDATE_DIR(node);
616 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
617
618 dent.d_fileno = node->tn_id;
619 dent.d_type = DT_DIR;
620 dent.d_namlen = 1;
621 dent.d_name[0] = '.';
622 dent.d_name[1] = '\0';
623 dent.d_reclen = GENERIC_DIRSIZ(&dent);
624
625 if (dent.d_reclen > uio->uio_resid)
626 error = -1;
627 else {
628 error = uiomove(&dent, dent.d_reclen, uio);
629 if (error == 0)
630 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
631 }
632
633 node->tn_status |= TMPFS_NODE_ACCESSED;
634
635 return error;
636}
637
638/* --------------------------------------------------------------------- */
639
640/*
641 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
642 * directory and returns it in the uio space. The function returns 0
643 * on success, -1 if there was not enough space in the uio structure to
644 * hold the directory entry or an appropriate error code if another
645 * error happens.
646 */
647int
648tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
649{
650 int error;
651 struct dirent dent;
652
653 TMPFS_VALIDATE_DIR(node);
654 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
655
447}
448
449/* --------------------------------------------------------------------- */
450
451/*
452 * Allocates a new file of type 'type' and adds it to the parent directory
453 * 'dvp'; this addition is done using the component name given in 'cnp'.
454 * The ownership of the new file is automatically assigned based on the
455 * credentials of the caller (through 'cnp'), the group is set based on
456 * the parent directory and the mode is determined from the 'vap' argument.
457 * If successful, *vpp holds a vnode to the newly created file and zero
458 * is returned. Otherwise *vpp is NULL and the function returns an
459 * appropriate error code.
460 */
461int
462tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
463 struct componentname *cnp, char *target)
464{
465 int error;
466 struct tmpfs_dirent *de;
467 struct tmpfs_mount *tmp;
468 struct tmpfs_node *dnode;
469 struct tmpfs_node *node;
470 struct tmpfs_node *parent;
471
472 MPASS(VOP_ISLOCKED(dvp));
473 MPASS(cnp->cn_flags & HASBUF);
474
475 tmp = VFS_TO_TMPFS(dvp->v_mount);
476 dnode = VP_TO_TMPFS_DIR(dvp);
477 *vpp = NULL;
478
479 /* If the entry we are creating is a directory, we cannot overflow
480 * the number of links of its parent, because it will get a new
481 * link. */
482 if (vap->va_type == VDIR) {
483 /* Ensure that we do not overflow the maximum number of links
484 * imposed by the system. */
485 MPASS(dnode->tn_links <= LINK_MAX);
486 if (dnode->tn_links == LINK_MAX) {
487 error = EMLINK;
488 goto out;
489 }
490
491 parent = dnode;
492 MPASS(parent != NULL);
493 } else
494 parent = NULL;
495
496 /* Allocate a node that represents the new file. */
497 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
498 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
499 if (error != 0)
500 goto out;
501
502 /* Allocate a directory entry that points to the new file. */
503 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
504 &de);
505 if (error != 0) {
506 tmpfs_free_node(tmp, node);
507 goto out;
508 }
509
510 /* Allocate a vnode for the new file. */
511 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
512 if (error != 0) {
513 tmpfs_free_dirent(tmp, de, TRUE);
514 tmpfs_free_node(tmp, node);
515 goto out;
516 }
517
518 /* Now that all required items are allocated, we can proceed to
519 * insert the new node into the directory, an operation that
520 * cannot fail. */
521 tmpfs_dir_attach(dvp, de);
522
523out:
524
525 return error;
526}
527
528/* --------------------------------------------------------------------- */
529
530/*
531 * Attaches the directory entry de to the directory represented by vp.
532 * Note that this does not change the link count of the node pointed by
533 * the directory entry, as this is done by tmpfs_alloc_dirent.
534 */
535void
536tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
537{
538 struct tmpfs_node *dnode;
539
540 ASSERT_VOP_ELOCKED(vp, __func__);
541 dnode = VP_TO_TMPFS_DIR(vp);
542 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
543 dnode->tn_size += sizeof(struct tmpfs_dirent);
544 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
545 TMPFS_NODE_MODIFIED;
546}
547
548/* --------------------------------------------------------------------- */
549
550/*
551 * Detaches the directory entry de from the directory represented by vp.
552 * Note that this does not change the link count of the node pointed by
553 * the directory entry, as this is done by tmpfs_free_dirent.
554 */
555void
556tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
557{
558 struct tmpfs_node *dnode;
559
560 ASSERT_VOP_ELOCKED(vp, __func__);
561 dnode = VP_TO_TMPFS_DIR(vp);
562
563 if (dnode->tn_dir.tn_readdir_lastp == de) {
564 dnode->tn_dir.tn_readdir_lastn = 0;
565 dnode->tn_dir.tn_readdir_lastp = NULL;
566 }
567
568 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
569 dnode->tn_size -= sizeof(struct tmpfs_dirent);
570 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
571 TMPFS_NODE_MODIFIED;
572}
573
574/* --------------------------------------------------------------------- */
575
576/*
577 * Looks for a directory entry in the directory represented by node.
578 * 'cnp' describes the name of the entry to look for. Note that the .
579 * and .. components are not allowed as they do not physically exist
580 * within directories.
581 *
582 * Returns a pointer to the entry when found, otherwise NULL.
583 */
584struct tmpfs_dirent *
585tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
586 struct componentname *cnp)
587{
588 boolean_t found;
589 struct tmpfs_dirent *de;
590
591 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
592 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
593 cnp->cn_nameptr[1] == '.')));
594 TMPFS_VALIDATE_DIR(node);
595
596 found = 0;
597 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
598 if (f != NULL && de->td_node != f)
599 continue;
600 MPASS(cnp->cn_namelen < 0xffff);
601 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
602 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
603 found = 1;
604 break;
605 }
606 }
607 node->tn_status |= TMPFS_NODE_ACCESSED;
608
609 return found ? de : NULL;
610}
611
612/* --------------------------------------------------------------------- */
613
614/*
615 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
616 * directory and returns it in the uio space. The function returns 0
617 * on success, -1 if there was not enough space in the uio structure to
618 * hold the directory entry or an appropriate error code if another
619 * error happens.
620 */
621int
622tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
623{
624 int error;
625 struct dirent dent;
626
627 TMPFS_VALIDATE_DIR(node);
628 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
629
630 dent.d_fileno = node->tn_id;
631 dent.d_type = DT_DIR;
632 dent.d_namlen = 1;
633 dent.d_name[0] = '.';
634 dent.d_name[1] = '\0';
635 dent.d_reclen = GENERIC_DIRSIZ(&dent);
636
637 if (dent.d_reclen > uio->uio_resid)
638 error = -1;
639 else {
640 error = uiomove(&dent, dent.d_reclen, uio);
641 if (error == 0)
642 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
643 }
644
645 node->tn_status |= TMPFS_NODE_ACCESSED;
646
647 return error;
648}
649
650/* --------------------------------------------------------------------- */
651
652/*
653 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
654 * directory and returns it in the uio space. The function returns 0
655 * on success, -1 if there was not enough space in the uio structure to
656 * hold the directory entry or an appropriate error code if another
657 * error happens.
658 */
659int
660tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
661{
662 int error;
663 struct dirent dent;
664
665 TMPFS_VALIDATE_DIR(node);
666 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
667
668 /*
669 * Return ENOENT if the current node is already removed.
670 */
671 TMPFS_ASSERT_LOCKED(node);
672 if (node->tn_dir.tn_parent == NULL) {
673 return (ENOENT);
674 }
675
676 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
656 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
677 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
678 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
679
657 dent.d_type = DT_DIR;
658 dent.d_namlen = 2;
659 dent.d_name[0] = '.';
660 dent.d_name[1] = '.';
661 dent.d_name[2] = '\0';
662 dent.d_reclen = GENERIC_DIRSIZ(&dent);
663
664 if (dent.d_reclen > uio->uio_resid)
665 error = -1;
666 else {
667 error = uiomove(&dent, dent.d_reclen, uio);
668 if (error == 0) {
669 struct tmpfs_dirent *de;
670
671 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
672 if (de == NULL)
673 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
674 else
675 uio->uio_offset = tmpfs_dircookie(de);
676 }
677 }
678
679 node->tn_status |= TMPFS_NODE_ACCESSED;
680
681 return error;
682}
683
684/* --------------------------------------------------------------------- */
685
686/*
687 * Lookup a directory entry by its associated cookie.
688 */
689struct tmpfs_dirent *
690tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
691{
692 struct tmpfs_dirent *de;
693
694 if (cookie == node->tn_dir.tn_readdir_lastn &&
695 node->tn_dir.tn_readdir_lastp != NULL) {
696 return node->tn_dir.tn_readdir_lastp;
697 }
698
699 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
700 if (tmpfs_dircookie(de) == cookie) {
701 break;
702 }
703 }
704
705 return de;
706}
707
708/* --------------------------------------------------------------------- */
709
710/*
711 * Helper function for tmpfs_readdir. Returns as much directory entries
712 * as can fit in the uio space. The read starts at uio->uio_offset.
713 * The function returns 0 on success, -1 if there was not enough space
714 * in the uio structure to hold the directory entry or an appropriate
715 * error code if another error happens.
716 */
717int
718tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
719{
720 int error;
721 off_t startcookie;
722 struct tmpfs_dirent *de;
723
724 TMPFS_VALIDATE_DIR(node);
725
726 /* Locate the first directory entry we have to return. We have cached
727 * the last readdir in the node, so use those values if appropriate.
728 * Otherwise do a linear scan to find the requested entry. */
729 startcookie = uio->uio_offset;
730 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
731 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
732 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
733 return 0;
734 } else {
735 de = tmpfs_dir_lookupbycookie(node, startcookie);
736 }
737 if (de == NULL) {
738 return EINVAL;
739 }
740
741 /* Read as much entries as possible; i.e., until we reach the end of
742 * the directory or we exhaust uio space. */
743 do {
744 struct dirent d;
745
746 /* Create a dirent structure representing the current
747 * tmpfs_node and fill it. */
748 d.d_fileno = de->td_node->tn_id;
749 switch (de->td_node->tn_type) {
750 case VBLK:
751 d.d_type = DT_BLK;
752 break;
753
754 case VCHR:
755 d.d_type = DT_CHR;
756 break;
757
758 case VDIR:
759 d.d_type = DT_DIR;
760 break;
761
762 case VFIFO:
763 d.d_type = DT_FIFO;
764 break;
765
766 case VLNK:
767 d.d_type = DT_LNK;
768 break;
769
770 case VREG:
771 d.d_type = DT_REG;
772 break;
773
774 case VSOCK:
775 d.d_type = DT_SOCK;
776 break;
777
778 default:
779 panic("tmpfs_dir_getdents: type %p %d",
780 de->td_node, (int)de->td_node->tn_type);
781 }
782 d.d_namlen = de->td_namelen;
783 MPASS(de->td_namelen < sizeof(d.d_name));
784 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
785 d.d_name[de->td_namelen] = '\0';
786 d.d_reclen = GENERIC_DIRSIZ(&d);
787
788 /* Stop reading if the directory entry we are treating is
789 * bigger than the amount of data that can be returned. */
790 if (d.d_reclen > uio->uio_resid) {
791 error = -1;
792 break;
793 }
794
795 /* Copy the new dirent structure into the output buffer and
796 * advance pointers. */
797 error = uiomove(&d, d.d_reclen, uio);
798
799 (*cntp)++;
800 de = TAILQ_NEXT(de, td_entries);
801 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
802
803 /* Update the offset and cache. */
804 if (de == NULL) {
805 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
806 node->tn_dir.tn_readdir_lastn = 0;
807 node->tn_dir.tn_readdir_lastp = NULL;
808 } else {
809 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
810 node->tn_dir.tn_readdir_lastp = de;
811 }
812
813 node->tn_status |= TMPFS_NODE_ACCESSED;
814 return error;
815}
816
817/* --------------------------------------------------------------------- */
818
819/*
820 * Resizes the aobj associated to the regular file pointed to by vp to
821 * the size newsize. 'vp' must point to a vnode that represents a regular
822 * file. 'newsize' must be positive.
823 *
824 * Returns zero on success or an appropriate error code on failure.
825 */
826int
827tmpfs_reg_resize(struct vnode *vp, off_t newsize)
828{
829 int error;
830 size_t newpages, oldpages;
831 struct tmpfs_mount *tmp;
832 struct tmpfs_node *node;
833 off_t oldsize;
834
835 MPASS(vp->v_type == VREG);
836 MPASS(newsize >= 0);
837
838 node = VP_TO_TMPFS_NODE(vp);
839 tmp = VFS_TO_TMPFS(vp->v_mount);
840
841 /* Convert the old and new sizes to the number of pages needed to
842 * store them. It may happen that we do not need to do anything
843 * because the last allocated page can accommodate the change on
844 * its own. */
845 oldsize = node->tn_size;
846 oldpages = round_page(oldsize) / PAGE_SIZE;
847 MPASS(oldpages == node->tn_reg.tn_aobj_pages);
848 newpages = round_page(newsize) / PAGE_SIZE;
849
850 if (newpages > oldpages &&
851 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
852 error = ENOSPC;
853 goto out;
854 }
855
856 node->tn_reg.tn_aobj_pages = newpages;
857
858 TMPFS_LOCK(tmp);
859 tmp->tm_pages_used += (newpages - oldpages);
860 TMPFS_UNLOCK(tmp);
861
862 node->tn_size = newsize;
863 vnode_pager_setsize(vp, newsize);
864 if (newsize < oldsize) {
865 size_t zerolen = round_page(newsize) - newsize;
866 vm_object_t uobj = node->tn_reg.tn_aobj;
867 vm_page_t m;
868
869 /*
870 * free "backing store"
871 */
872 VM_OBJECT_LOCK(uobj);
873 if (newpages < oldpages) {
874 swap_pager_freespace(uobj,
875 newpages, oldpages - newpages);
876 vm_object_page_remove(uobj,
877 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE);
878 }
879
880 /*
881 * zero out the truncated part of the last page.
882 */
883
884 if (zerolen > 0) {
885 m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
886 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
887 pmap_zero_page_area(m, PAGE_SIZE - zerolen,
888 zerolen);
889 vm_page_wakeup(m);
890 }
891 VM_OBJECT_UNLOCK(uobj);
892
893 }
894
895 error = 0;
896
897out:
898 return error;
899}
900
901/* --------------------------------------------------------------------- */
902
903/*
904 * Change flags of the given vnode.
905 * Caller should execute tmpfs_update on vp after a successful execution.
906 * The vnode must be locked on entry and remain locked on exit.
907 */
908int
909tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
910{
911 int error;
912 struct tmpfs_node *node;
913
914 MPASS(VOP_ISLOCKED(vp));
915
916 node = VP_TO_TMPFS_NODE(vp);
917
918 /* Disallow this operation if the file system is mounted read-only. */
919 if (vp->v_mount->mnt_flag & MNT_RDONLY)
920 return EROFS;
921
922 /*
923 * Callers may only modify the file flags on objects they
924 * have VADMIN rights for.
925 */
926 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
927 return (error);
928 /*
929 * Unprivileged processes are not permitted to unset system
930 * flags, or modify flags if any system flags are set.
931 */
932 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
933 if (node->tn_flags
934 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
935 error = securelevel_gt(cred, 0);
936 if (error)
937 return (error);
938 }
939 /* Snapshot flag cannot be set or cleared */
940 if (((flags & SF_SNAPSHOT) != 0 &&
941 (node->tn_flags & SF_SNAPSHOT) == 0) ||
942 ((flags & SF_SNAPSHOT) == 0 &&
943 (node->tn_flags & SF_SNAPSHOT) != 0))
944 return (EPERM);
945 node->tn_flags = flags;
946 } else {
947 if (node->tn_flags
948 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
949 (flags & UF_SETTABLE) != flags)
950 return (EPERM);
951 node->tn_flags &= SF_SETTABLE;
952 node->tn_flags |= (flags & UF_SETTABLE);
953 }
954 node->tn_status |= TMPFS_NODE_CHANGED;
955
956 MPASS(VOP_ISLOCKED(vp));
957
958 return 0;
959}
960
961/* --------------------------------------------------------------------- */
962
963/*
964 * Change access mode on the given vnode.
965 * Caller should execute tmpfs_update on vp after a successful execution.
966 * The vnode must be locked on entry and remain locked on exit.
967 */
968int
969tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
970{
971 int error;
972 struct tmpfs_node *node;
973
974 MPASS(VOP_ISLOCKED(vp));
975
976 node = VP_TO_TMPFS_NODE(vp);
977
978 /* Disallow this operation if the file system is mounted read-only. */
979 if (vp->v_mount->mnt_flag & MNT_RDONLY)
980 return EROFS;
981
982 /* Immutable or append-only files cannot be modified, either. */
983 if (node->tn_flags & (IMMUTABLE | APPEND))
984 return EPERM;
985
986 /*
987 * To modify the permissions on a file, must possess VADMIN
988 * for that file.
989 */
990 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
991 return (error);
992
993 /*
994 * Privileged processes may set the sticky bit on non-directories,
995 * as well as set the setgid bit on a file with a group that the
996 * process is not a member of.
997 */
998 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
999 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1000 return (EFTYPE);
1001 }
1002 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1003 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1004 if (error)
1005 return (error);
1006 }
1007
1008
1009 node->tn_mode &= ~ALLPERMS;
1010 node->tn_mode |= mode & ALLPERMS;
1011
1012 node->tn_status |= TMPFS_NODE_CHANGED;
1013
1014 MPASS(VOP_ISLOCKED(vp));
1015
1016 return 0;
1017}
1018
1019/* --------------------------------------------------------------------- */
1020
1021/*
1022 * Change ownership of the given vnode. At least one of uid or gid must
1023 * be different than VNOVAL. If one is set to that value, the attribute
1024 * is unchanged.
1025 * Caller should execute tmpfs_update on vp after a successful execution.
1026 * The vnode must be locked on entry and remain locked on exit.
1027 */
1028int
1029tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1030 struct thread *p)
1031{
1032 int error;
1033 struct tmpfs_node *node;
1034 uid_t ouid;
1035 gid_t ogid;
1036
1037 MPASS(VOP_ISLOCKED(vp));
1038
1039 node = VP_TO_TMPFS_NODE(vp);
1040
1041 /* Assign default values if they are unknown. */
1042 MPASS(uid != VNOVAL || gid != VNOVAL);
1043 if (uid == VNOVAL)
1044 uid = node->tn_uid;
1045 if (gid == VNOVAL)
1046 gid = node->tn_gid;
1047 MPASS(uid != VNOVAL && gid != VNOVAL);
1048
1049 /* Disallow this operation if the file system is mounted read-only. */
1050 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1051 return EROFS;
1052
1053 /* Immutable or append-only files cannot be modified, either. */
1054 if (node->tn_flags & (IMMUTABLE | APPEND))
1055 return EPERM;
1056
1057 /*
1058 * To modify the ownership of a file, must possess VADMIN for that
1059 * file.
1060 */
1061 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1062 return (error);
1063
1064 /*
1065 * To change the owner of a file, or change the group of a file to a
1066 * group of which we are not a member, the caller must have
1067 * privilege.
1068 */
1069 if ((uid != node->tn_uid ||
1070 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1071 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1072 return (error);
1073
1074 ogid = node->tn_gid;
1075 ouid = node->tn_uid;
1076
1077 node->tn_uid = uid;
1078 node->tn_gid = gid;
1079
1080 node->tn_status |= TMPFS_NODE_CHANGED;
1081
1082 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1083 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1084 node->tn_mode &= ~(S_ISUID | S_ISGID);
1085 }
1086
1087 MPASS(VOP_ISLOCKED(vp));
1088
1089 return 0;
1090}
1091
1092/* --------------------------------------------------------------------- */
1093
1094/*
1095 * Change size of the given vnode.
1096 * Caller should execute tmpfs_update on vp after a successful execution.
1097 * The vnode must be locked on entry and remain locked on exit.
1098 */
1099int
1100tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1101 struct thread *p)
1102{
1103 int error;
1104 struct tmpfs_node *node;
1105
1106 MPASS(VOP_ISLOCKED(vp));
1107
1108 node = VP_TO_TMPFS_NODE(vp);
1109
1110 /* Decide whether this is a valid operation based on the file type. */
1111 error = 0;
1112 switch (vp->v_type) {
1113 case VDIR:
1114 return EISDIR;
1115
1116 case VREG:
1117 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1118 return EROFS;
1119 break;
1120
1121 case VBLK:
1122 /* FALLTHROUGH */
1123 case VCHR:
1124 /* FALLTHROUGH */
1125 case VFIFO:
1126 /* Allow modifications of special files even if in the file
1127 * system is mounted read-only (we are not modifying the
1128 * files themselves, but the objects they represent). */
1129 return 0;
1130
1131 default:
1132 /* Anything else is unsupported. */
1133 return EOPNOTSUPP;
1134 }
1135
1136 /* Immutable or append-only files cannot be modified, either. */
1137 if (node->tn_flags & (IMMUTABLE | APPEND))
1138 return EPERM;
1139
1140 error = tmpfs_truncate(vp, size);
1141 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1142 * for us, as will update tn_status; no need to do that here. */
1143
1144 MPASS(VOP_ISLOCKED(vp));
1145
1146 return error;
1147}
1148
1149/* --------------------------------------------------------------------- */
1150
1151/*
1152 * Change access and modification times of the given vnode.
1153 * Caller should execute tmpfs_update on vp after a successful execution.
1154 * The vnode must be locked on entry and remain locked on exit.
1155 */
1156int
1157tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1158 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1159{
1160 int error;
1161 struct tmpfs_node *node;
1162
1163 MPASS(VOP_ISLOCKED(vp));
1164
1165 node = VP_TO_TMPFS_NODE(vp);
1166
1167 /* Disallow this operation if the file system is mounted read-only. */
1168 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1169 return EROFS;
1170
1171 /* Immutable or append-only files cannot be modified, either. */
1172 if (node->tn_flags & (IMMUTABLE | APPEND))
1173 return EPERM;
1174
1175 /* Determine if the user have proper privilege to update time. */
1176 if (vaflags & VA_UTIMES_NULL) {
1177 error = VOP_ACCESS(vp, VADMIN, cred, l);
1178 if (error)
1179 error = VOP_ACCESS(vp, VWRITE, cred, l);
1180 } else
1181 error = VOP_ACCESS(vp, VADMIN, cred, l);
1182 if (error)
1183 return (error);
1184
1185 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1186 node->tn_status |= TMPFS_NODE_ACCESSED;
1187
1188 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1189 node->tn_status |= TMPFS_NODE_MODIFIED;
1190
1191 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1192 node->tn_status |= TMPFS_NODE_MODIFIED;
1193
1194 tmpfs_itimes(vp, atime, mtime);
1195
1196 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1197 node->tn_birthtime = *birthtime;
1198 MPASS(VOP_ISLOCKED(vp));
1199
1200 return 0;
1201}
1202
1203/* --------------------------------------------------------------------- */
1204/* Sync timestamps */
1205void
1206tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1207 const struct timespec *mod)
1208{
1209 struct tmpfs_node *node;
1210 struct timespec now;
1211
1212 node = VP_TO_TMPFS_NODE(vp);
1213
1214 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1215 TMPFS_NODE_CHANGED)) == 0)
1216 return;
1217
1218 vfs_timestamp(&now);
1219 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1220 if (acc == NULL)
1221 acc = &now;
1222 node->tn_atime = *acc;
1223 }
1224 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1225 if (mod == NULL)
1226 mod = &now;
1227 node->tn_mtime = *mod;
1228 }
1229 if (node->tn_status & TMPFS_NODE_CHANGED) {
1230 node->tn_ctime = now;
1231 }
1232 node->tn_status &=
1233 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1234}
1235
1236/* --------------------------------------------------------------------- */
1237
1238void
1239tmpfs_update(struct vnode *vp)
1240{
1241
1242 tmpfs_itimes(vp, NULL, NULL);
1243}
1244
1245/* --------------------------------------------------------------------- */
1246
1247int
1248tmpfs_truncate(struct vnode *vp, off_t length)
1249{
1250 int error;
1251 struct tmpfs_node *node;
1252
1253 node = VP_TO_TMPFS_NODE(vp);
1254
1255 if (length < 0) {
1256 error = EINVAL;
1257 goto out;
1258 }
1259
1260 if (node->tn_size == length) {
1261 error = 0;
1262 goto out;
1263 }
1264
1265 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1266 return (EFBIG);
1267
1268 error = tmpfs_reg_resize(vp, length);
1269 if (error == 0) {
1270 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1271 }
1272
1273out:
1274 tmpfs_update(vp);
1275
1276 return error;
1277}
680 dent.d_type = DT_DIR;
681 dent.d_namlen = 2;
682 dent.d_name[0] = '.';
683 dent.d_name[1] = '.';
684 dent.d_name[2] = '\0';
685 dent.d_reclen = GENERIC_DIRSIZ(&dent);
686
687 if (dent.d_reclen > uio->uio_resid)
688 error = -1;
689 else {
690 error = uiomove(&dent, dent.d_reclen, uio);
691 if (error == 0) {
692 struct tmpfs_dirent *de;
693
694 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
695 if (de == NULL)
696 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
697 else
698 uio->uio_offset = tmpfs_dircookie(de);
699 }
700 }
701
702 node->tn_status |= TMPFS_NODE_ACCESSED;
703
704 return error;
705}
706
707/* --------------------------------------------------------------------- */
708
709/*
710 * Lookup a directory entry by its associated cookie.
711 */
712struct tmpfs_dirent *
713tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
714{
715 struct tmpfs_dirent *de;
716
717 if (cookie == node->tn_dir.tn_readdir_lastn &&
718 node->tn_dir.tn_readdir_lastp != NULL) {
719 return node->tn_dir.tn_readdir_lastp;
720 }
721
722 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
723 if (tmpfs_dircookie(de) == cookie) {
724 break;
725 }
726 }
727
728 return de;
729}
730
731/* --------------------------------------------------------------------- */
732
733/*
734 * Helper function for tmpfs_readdir. Returns as much directory entries
735 * as can fit in the uio space. The read starts at uio->uio_offset.
736 * The function returns 0 on success, -1 if there was not enough space
737 * in the uio structure to hold the directory entry or an appropriate
738 * error code if another error happens.
739 */
740int
741tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
742{
743 int error;
744 off_t startcookie;
745 struct tmpfs_dirent *de;
746
747 TMPFS_VALIDATE_DIR(node);
748
749 /* Locate the first directory entry we have to return. We have cached
750 * the last readdir in the node, so use those values if appropriate.
751 * Otherwise do a linear scan to find the requested entry. */
752 startcookie = uio->uio_offset;
753 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
754 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
755 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
756 return 0;
757 } else {
758 de = tmpfs_dir_lookupbycookie(node, startcookie);
759 }
760 if (de == NULL) {
761 return EINVAL;
762 }
763
764 /* Read as much entries as possible; i.e., until we reach the end of
765 * the directory or we exhaust uio space. */
766 do {
767 struct dirent d;
768
769 /* Create a dirent structure representing the current
770 * tmpfs_node and fill it. */
771 d.d_fileno = de->td_node->tn_id;
772 switch (de->td_node->tn_type) {
773 case VBLK:
774 d.d_type = DT_BLK;
775 break;
776
777 case VCHR:
778 d.d_type = DT_CHR;
779 break;
780
781 case VDIR:
782 d.d_type = DT_DIR;
783 break;
784
785 case VFIFO:
786 d.d_type = DT_FIFO;
787 break;
788
789 case VLNK:
790 d.d_type = DT_LNK;
791 break;
792
793 case VREG:
794 d.d_type = DT_REG;
795 break;
796
797 case VSOCK:
798 d.d_type = DT_SOCK;
799 break;
800
801 default:
802 panic("tmpfs_dir_getdents: type %p %d",
803 de->td_node, (int)de->td_node->tn_type);
804 }
805 d.d_namlen = de->td_namelen;
806 MPASS(de->td_namelen < sizeof(d.d_name));
807 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
808 d.d_name[de->td_namelen] = '\0';
809 d.d_reclen = GENERIC_DIRSIZ(&d);
810
811 /* Stop reading if the directory entry we are treating is
812 * bigger than the amount of data that can be returned. */
813 if (d.d_reclen > uio->uio_resid) {
814 error = -1;
815 break;
816 }
817
818 /* Copy the new dirent structure into the output buffer and
819 * advance pointers. */
820 error = uiomove(&d, d.d_reclen, uio);
821
822 (*cntp)++;
823 de = TAILQ_NEXT(de, td_entries);
824 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
825
826 /* Update the offset and cache. */
827 if (de == NULL) {
828 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
829 node->tn_dir.tn_readdir_lastn = 0;
830 node->tn_dir.tn_readdir_lastp = NULL;
831 } else {
832 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
833 node->tn_dir.tn_readdir_lastp = de;
834 }
835
836 node->tn_status |= TMPFS_NODE_ACCESSED;
837 return error;
838}
839
840/* --------------------------------------------------------------------- */
841
842/*
843 * Resizes the aobj associated to the regular file pointed to by vp to
844 * the size newsize. 'vp' must point to a vnode that represents a regular
845 * file. 'newsize' must be positive.
846 *
847 * Returns zero on success or an appropriate error code on failure.
848 */
849int
850tmpfs_reg_resize(struct vnode *vp, off_t newsize)
851{
852 int error;
853 size_t newpages, oldpages;
854 struct tmpfs_mount *tmp;
855 struct tmpfs_node *node;
856 off_t oldsize;
857
858 MPASS(vp->v_type == VREG);
859 MPASS(newsize >= 0);
860
861 node = VP_TO_TMPFS_NODE(vp);
862 tmp = VFS_TO_TMPFS(vp->v_mount);
863
864 /* Convert the old and new sizes to the number of pages needed to
865 * store them. It may happen that we do not need to do anything
866 * because the last allocated page can accommodate the change on
867 * its own. */
868 oldsize = node->tn_size;
869 oldpages = round_page(oldsize) / PAGE_SIZE;
870 MPASS(oldpages == node->tn_reg.tn_aobj_pages);
871 newpages = round_page(newsize) / PAGE_SIZE;
872
873 if (newpages > oldpages &&
874 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
875 error = ENOSPC;
876 goto out;
877 }
878
879 node->tn_reg.tn_aobj_pages = newpages;
880
881 TMPFS_LOCK(tmp);
882 tmp->tm_pages_used += (newpages - oldpages);
883 TMPFS_UNLOCK(tmp);
884
885 node->tn_size = newsize;
886 vnode_pager_setsize(vp, newsize);
887 if (newsize < oldsize) {
888 size_t zerolen = round_page(newsize) - newsize;
889 vm_object_t uobj = node->tn_reg.tn_aobj;
890 vm_page_t m;
891
892 /*
893 * free "backing store"
894 */
895 VM_OBJECT_LOCK(uobj);
896 if (newpages < oldpages) {
897 swap_pager_freespace(uobj,
898 newpages, oldpages - newpages);
899 vm_object_page_remove(uobj,
900 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE);
901 }
902
903 /*
904 * zero out the truncated part of the last page.
905 */
906
907 if (zerolen > 0) {
908 m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
909 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
910 pmap_zero_page_area(m, PAGE_SIZE - zerolen,
911 zerolen);
912 vm_page_wakeup(m);
913 }
914 VM_OBJECT_UNLOCK(uobj);
915
916 }
917
918 error = 0;
919
920out:
921 return error;
922}
923
924/* --------------------------------------------------------------------- */
925
926/*
927 * Change flags of the given vnode.
928 * Caller should execute tmpfs_update on vp after a successful execution.
929 * The vnode must be locked on entry and remain locked on exit.
930 */
931int
932tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
933{
934 int error;
935 struct tmpfs_node *node;
936
937 MPASS(VOP_ISLOCKED(vp));
938
939 node = VP_TO_TMPFS_NODE(vp);
940
941 /* Disallow this operation if the file system is mounted read-only. */
942 if (vp->v_mount->mnt_flag & MNT_RDONLY)
943 return EROFS;
944
945 /*
946 * Callers may only modify the file flags on objects they
947 * have VADMIN rights for.
948 */
949 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
950 return (error);
951 /*
952 * Unprivileged processes are not permitted to unset system
953 * flags, or modify flags if any system flags are set.
954 */
955 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
956 if (node->tn_flags
957 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
958 error = securelevel_gt(cred, 0);
959 if (error)
960 return (error);
961 }
962 /* Snapshot flag cannot be set or cleared */
963 if (((flags & SF_SNAPSHOT) != 0 &&
964 (node->tn_flags & SF_SNAPSHOT) == 0) ||
965 ((flags & SF_SNAPSHOT) == 0 &&
966 (node->tn_flags & SF_SNAPSHOT) != 0))
967 return (EPERM);
968 node->tn_flags = flags;
969 } else {
970 if (node->tn_flags
971 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
972 (flags & UF_SETTABLE) != flags)
973 return (EPERM);
974 node->tn_flags &= SF_SETTABLE;
975 node->tn_flags |= (flags & UF_SETTABLE);
976 }
977 node->tn_status |= TMPFS_NODE_CHANGED;
978
979 MPASS(VOP_ISLOCKED(vp));
980
981 return 0;
982}
983
984/* --------------------------------------------------------------------- */
985
986/*
987 * Change access mode on the given vnode.
988 * Caller should execute tmpfs_update on vp after a successful execution.
989 * The vnode must be locked on entry and remain locked on exit.
990 */
991int
992tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
993{
994 int error;
995 struct tmpfs_node *node;
996
997 MPASS(VOP_ISLOCKED(vp));
998
999 node = VP_TO_TMPFS_NODE(vp);
1000
1001 /* Disallow this operation if the file system is mounted read-only. */
1002 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1003 return EROFS;
1004
1005 /* Immutable or append-only files cannot be modified, either. */
1006 if (node->tn_flags & (IMMUTABLE | APPEND))
1007 return EPERM;
1008
1009 /*
1010 * To modify the permissions on a file, must possess VADMIN
1011 * for that file.
1012 */
1013 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1014 return (error);
1015
1016 /*
1017 * Privileged processes may set the sticky bit on non-directories,
1018 * as well as set the setgid bit on a file with a group that the
1019 * process is not a member of.
1020 */
1021 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1022 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1023 return (EFTYPE);
1024 }
1025 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1026 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1027 if (error)
1028 return (error);
1029 }
1030
1031
1032 node->tn_mode &= ~ALLPERMS;
1033 node->tn_mode |= mode & ALLPERMS;
1034
1035 node->tn_status |= TMPFS_NODE_CHANGED;
1036
1037 MPASS(VOP_ISLOCKED(vp));
1038
1039 return 0;
1040}
1041
1042/* --------------------------------------------------------------------- */
1043
1044/*
1045 * Change ownership of the given vnode. At least one of uid or gid must
1046 * be different than VNOVAL. If one is set to that value, the attribute
1047 * is unchanged.
1048 * Caller should execute tmpfs_update on vp after a successful execution.
1049 * The vnode must be locked on entry and remain locked on exit.
1050 */
1051int
1052tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1053 struct thread *p)
1054{
1055 int error;
1056 struct tmpfs_node *node;
1057 uid_t ouid;
1058 gid_t ogid;
1059
1060 MPASS(VOP_ISLOCKED(vp));
1061
1062 node = VP_TO_TMPFS_NODE(vp);
1063
1064 /* Assign default values if they are unknown. */
1065 MPASS(uid != VNOVAL || gid != VNOVAL);
1066 if (uid == VNOVAL)
1067 uid = node->tn_uid;
1068 if (gid == VNOVAL)
1069 gid = node->tn_gid;
1070 MPASS(uid != VNOVAL && gid != VNOVAL);
1071
1072 /* Disallow this operation if the file system is mounted read-only. */
1073 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1074 return EROFS;
1075
1076 /* Immutable or append-only files cannot be modified, either. */
1077 if (node->tn_flags & (IMMUTABLE | APPEND))
1078 return EPERM;
1079
1080 /*
1081 * To modify the ownership of a file, must possess VADMIN for that
1082 * file.
1083 */
1084 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1085 return (error);
1086
1087 /*
1088 * To change the owner of a file, or change the group of a file to a
1089 * group of which we are not a member, the caller must have
1090 * privilege.
1091 */
1092 if ((uid != node->tn_uid ||
1093 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1094 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1095 return (error);
1096
1097 ogid = node->tn_gid;
1098 ouid = node->tn_uid;
1099
1100 node->tn_uid = uid;
1101 node->tn_gid = gid;
1102
1103 node->tn_status |= TMPFS_NODE_CHANGED;
1104
1105 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1106 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1107 node->tn_mode &= ~(S_ISUID | S_ISGID);
1108 }
1109
1110 MPASS(VOP_ISLOCKED(vp));
1111
1112 return 0;
1113}
1114
1115/* --------------------------------------------------------------------- */
1116
1117/*
1118 * Change size of the given vnode.
1119 * Caller should execute tmpfs_update on vp after a successful execution.
1120 * The vnode must be locked on entry and remain locked on exit.
1121 */
1122int
1123tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1124 struct thread *p)
1125{
1126 int error;
1127 struct tmpfs_node *node;
1128
1129 MPASS(VOP_ISLOCKED(vp));
1130
1131 node = VP_TO_TMPFS_NODE(vp);
1132
1133 /* Decide whether this is a valid operation based on the file type. */
1134 error = 0;
1135 switch (vp->v_type) {
1136 case VDIR:
1137 return EISDIR;
1138
1139 case VREG:
1140 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1141 return EROFS;
1142 break;
1143
1144 case VBLK:
1145 /* FALLTHROUGH */
1146 case VCHR:
1147 /* FALLTHROUGH */
1148 case VFIFO:
1149 /* Allow modifications of special files even if in the file
1150 * system is mounted read-only (we are not modifying the
1151 * files themselves, but the objects they represent). */
1152 return 0;
1153
1154 default:
1155 /* Anything else is unsupported. */
1156 return EOPNOTSUPP;
1157 }
1158
1159 /* Immutable or append-only files cannot be modified, either. */
1160 if (node->tn_flags & (IMMUTABLE | APPEND))
1161 return EPERM;
1162
1163 error = tmpfs_truncate(vp, size);
1164 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1165 * for us, as will update tn_status; no need to do that here. */
1166
1167 MPASS(VOP_ISLOCKED(vp));
1168
1169 return error;
1170}
1171
1172/* --------------------------------------------------------------------- */
1173
1174/*
1175 * Change access and modification times of the given vnode.
1176 * Caller should execute tmpfs_update on vp after a successful execution.
1177 * The vnode must be locked on entry and remain locked on exit.
1178 */
1179int
1180tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1181 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1182{
1183 int error;
1184 struct tmpfs_node *node;
1185
1186 MPASS(VOP_ISLOCKED(vp));
1187
1188 node = VP_TO_TMPFS_NODE(vp);
1189
1190 /* Disallow this operation if the file system is mounted read-only. */
1191 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1192 return EROFS;
1193
1194 /* Immutable or append-only files cannot be modified, either. */
1195 if (node->tn_flags & (IMMUTABLE | APPEND))
1196 return EPERM;
1197
1198 /* Determine if the user have proper privilege to update time. */
1199 if (vaflags & VA_UTIMES_NULL) {
1200 error = VOP_ACCESS(vp, VADMIN, cred, l);
1201 if (error)
1202 error = VOP_ACCESS(vp, VWRITE, cred, l);
1203 } else
1204 error = VOP_ACCESS(vp, VADMIN, cred, l);
1205 if (error)
1206 return (error);
1207
1208 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1209 node->tn_status |= TMPFS_NODE_ACCESSED;
1210
1211 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1212 node->tn_status |= TMPFS_NODE_MODIFIED;
1213
1214 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1215 node->tn_status |= TMPFS_NODE_MODIFIED;
1216
1217 tmpfs_itimes(vp, atime, mtime);
1218
1219 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1220 node->tn_birthtime = *birthtime;
1221 MPASS(VOP_ISLOCKED(vp));
1222
1223 return 0;
1224}
1225
1226/* --------------------------------------------------------------------- */
1227/* Sync timestamps */
1228void
1229tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1230 const struct timespec *mod)
1231{
1232 struct tmpfs_node *node;
1233 struct timespec now;
1234
1235 node = VP_TO_TMPFS_NODE(vp);
1236
1237 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1238 TMPFS_NODE_CHANGED)) == 0)
1239 return;
1240
1241 vfs_timestamp(&now);
1242 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1243 if (acc == NULL)
1244 acc = &now;
1245 node->tn_atime = *acc;
1246 }
1247 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1248 if (mod == NULL)
1249 mod = &now;
1250 node->tn_mtime = *mod;
1251 }
1252 if (node->tn_status & TMPFS_NODE_CHANGED) {
1253 node->tn_ctime = now;
1254 }
1255 node->tn_status &=
1256 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1257}
1258
1259/* --------------------------------------------------------------------- */
1260
1261void
1262tmpfs_update(struct vnode *vp)
1263{
1264
1265 tmpfs_itimes(vp, NULL, NULL);
1266}
1267
1268/* --------------------------------------------------------------------- */
1269
1270int
1271tmpfs_truncate(struct vnode *vp, off_t length)
1272{
1273 int error;
1274 struct tmpfs_node *node;
1275
1276 node = VP_TO_TMPFS_NODE(vp);
1277
1278 if (length < 0) {
1279 error = EINVAL;
1280 goto out;
1281 }
1282
1283 if (node->tn_size == length) {
1284 error = 0;
1285 goto out;
1286 }
1287
1288 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1289 return (EFBIG);
1290
1291 error = tmpfs_reg_resize(vp, length);
1292 if (error == 0) {
1293 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1294 }
1295
1296out:
1297 tmpfs_update(vp);
1298
1299 return error;
1300}