aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAl Viro <viro@zeniv.linux.org.uk>2008-07-15 08:54:06 -0400
committerAl Viro <viro@zeniv.linux.org.uk>2008-07-26 20:53:12 -0400
commit9043476f726802f4b00c96d0c4f418dde48d1304 (patch)
tree9ead0294bc75e219c12b44fc7eb8996248400f2a
parentae7edecc9b8810770a8e5cb9a466ea4bdcfa8401 (diff)
[PATCH] sanitize proc_sysctl
* keep references to ctl_table_head and ctl_table in /proc/sys inodes * grab the former during operations, use the latter for access to entry if that succeeds * have ->d_compare() check if table should be seen for one who does lookup; that allows us to avoid flipping inodes - if we have the same name resolve to different things, we'll just keep several dentries and ->d_compare() will reject the wrong ones. * have ->lookup() and ->readdir() scan the table of our inode first, then walk all ctl_table_header and scan ->attached_by for those that are attached to our directory. * implement ->getattr(). * get rid of insane amounts of tree-walking * get rid of the need to know dentry in ->permission() and of the contortions induced by that. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
-rw-r--r--fs/proc/inode.c5
-rw-r--r--fs/proc/proc_sysctl.c427
-rw-r--r--include/linux/proc_fs.h5
-rw-r--r--include/linux/sysctl.h1
-rw-r--r--kernel/sysctl.c15
5 files changed, 218 insertions, 235 deletions
diff --git a/fs/proc/inode.c b/fs/proc/inode.c
index b37f25dc45a5..8bb03f056c28 100644
--- a/fs/proc/inode.c
+++ b/fs/proc/inode.c
@@ -17,6 +17,7 @@
17#include <linux/init.h> 17#include <linux/init.h>
18#include <linux/module.h> 18#include <linux/module.h>
19#include <linux/smp_lock.h> 19#include <linux/smp_lock.h>
20#include <linux/sysctl.h>
20 21
21#include <asm/system.h> 22#include <asm/system.h>
22#include <asm/uaccess.h> 23#include <asm/uaccess.h>
@@ -65,6 +66,8 @@ static void proc_delete_inode(struct inode *inode)
65 module_put(de->owner); 66 module_put(de->owner);
66 de_put(de); 67 de_put(de);
67 } 68 }
69 if (PROC_I(inode)->sysctl)
70 sysctl_head_put(PROC_I(inode)->sysctl);
68 clear_inode(inode); 71 clear_inode(inode);
69} 72}
70 73
@@ -84,6 +87,8 @@ static struct inode *proc_alloc_inode(struct super_block *sb)
84 ei->fd = 0; 87 ei->fd = 0;
85 ei->op.proc_get_link = NULL; 88 ei->op.proc_get_link = NULL;
86 ei->pde = NULL; 89 ei->pde = NULL;
90 ei->sysctl = NULL;
91 ei->sysctl_entry = NULL;
87 inode = &ei->vfs_inode; 92 inode = &ei->vfs_inode;
88 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 93 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
89 return inode; 94 return inode;
diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
index 5acc001d49f6..fa1ec2433e44 100644
--- a/fs/proc/proc_sysctl.c
+++ b/fs/proc/proc_sysctl.c
@@ -10,149 +10,110 @@
10static struct dentry_operations proc_sys_dentry_operations; 10static struct dentry_operations proc_sys_dentry_operations;
11static const struct file_operations proc_sys_file_operations; 11static const struct file_operations proc_sys_file_operations;
12static const struct inode_operations proc_sys_inode_operations; 12static const struct inode_operations proc_sys_inode_operations;
13static const struct file_operations proc_sys_dir_file_operations;
14static const struct inode_operations proc_sys_dir_operations;
13 15
14static void proc_sys_refresh_inode(struct inode *inode, struct ctl_table *table) 16static struct inode *proc_sys_make_inode(struct super_block *sb,
15{ 17 struct ctl_table_header *head, struct ctl_table *table)
16 /* Refresh the cached information bits in the inode */
17 if (table) {
18 inode->i_uid = 0;
19 inode->i_gid = 0;
20 inode->i_mode = table->mode;
21 if (table->proc_handler) {
22 inode->i_mode |= S_IFREG;
23 inode->i_nlink = 1;
24 } else {
25 inode->i_mode |= S_IFDIR;
26 inode->i_nlink = 0; /* It is too hard to figure out */
27 }
28 }
29}
30
31static struct inode *proc_sys_make_inode(struct inode *dir, struct ctl_table *table)
32{ 18{
33 struct inode *inode; 19 struct inode *inode;
34 struct proc_inode *dir_ei, *ei; 20 struct proc_inode *ei;
35 int depth;
36 21
37 inode = new_inode(dir->i_sb); 22 inode = new_inode(sb);
38 if (!inode) 23 if (!inode)
39 goto out; 24 goto out;
40 25
41 /* A directory is always one deeper than it's parent */ 26 sysctl_head_get(head);
42 dir_ei = PROC_I(dir);
43 depth = dir_ei->fd + 1;
44
45 ei = PROC_I(inode); 27 ei = PROC_I(inode);
46 ei->fd = depth; 28 ei->sysctl = head;
29 ei->sysctl_entry = table;
30
47 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 31 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
48 inode->i_op = &proc_sys_inode_operations;
49 inode->i_fop = &proc_sys_file_operations;
50 inode->i_flags |= S_PRIVATE; /* tell selinux to ignore this inode */ 32 inode->i_flags |= S_PRIVATE; /* tell selinux to ignore this inode */
51 proc_sys_refresh_inode(inode, table); 33 inode->i_mode = table->mode;
34 if (!table->child) {
35 inode->i_mode |= S_IFREG;
36 inode->i_op = &proc_sys_inode_operations;
37 inode->i_fop = &proc_sys_file_operations;
38 } else {
39 inode->i_mode |= S_IFDIR;
40 inode->i_nlink = 0;
41 inode->i_op = &proc_sys_dir_operations;
42 inode->i_fop = &proc_sys_dir_file_operations;
43 }
52out: 44out:
53 return inode; 45 return inode;
54} 46}
55 47
56static struct dentry *proc_sys_ancestor(struct dentry *dentry, int depth) 48static struct ctl_table *find_in_table(struct ctl_table *p, struct qstr *name)
57{
58 for (;;) {
59 struct proc_inode *ei;
60
61 ei = PROC_I(dentry->d_inode);
62 if (ei->fd == depth)
63 break; /* found */
64
65 dentry = dentry->d_parent;
66 }
67 return dentry;
68}
69
70static struct ctl_table *proc_sys_lookup_table_one(struct ctl_table *table,
71 struct qstr *name)
72{ 49{
73 int len; 50 int len;
74 for ( ; table->ctl_name || table->procname; table++) { 51 for ( ; p->ctl_name || p->procname; p++) {
75 52
76 if (!table->procname) 53 if (!p->procname)
77 continue; 54 continue;
78 55
79 len = strlen(table->procname); 56 len = strlen(p->procname);
80 if (len != name->len) 57 if (len != name->len)
81 continue; 58 continue;
82 59
83 if (memcmp(table->procname, name->name, len) != 0) 60 if (memcmp(p->procname, name->name, len) != 0)
84 continue; 61 continue;
85 62
86 /* I have a match */ 63 /* I have a match */
87 return table; 64 return p;
88 } 65 }
89 return NULL; 66 return NULL;
90} 67}
91 68
92static struct ctl_table *proc_sys_lookup_table(struct dentry *dentry, 69struct ctl_table_header *grab_header(struct inode *inode)
93 struct ctl_table *table)
94{ 70{
95 struct dentry *ancestor; 71 if (PROC_I(inode)->sysctl)
96 struct proc_inode *ei; 72 return sysctl_head_grab(PROC_I(inode)->sysctl);
97 int depth, i; 73 else
74 return sysctl_head_next(NULL);
75}
98 76
99 ei = PROC_I(dentry->d_inode); 77static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
100 depth = ei->fd; 78 struct nameidata *nd)
79{
80 struct ctl_table_header *head = grab_header(dir);
81 struct ctl_table *table = PROC_I(dir)->sysctl_entry;
82 struct ctl_table_header *h = NULL;
83 struct qstr *name = &dentry->d_name;
84 struct ctl_table *p;
85 struct inode *inode;
86 struct dentry *err = ERR_PTR(-ENOENT);
101 87
102 if (depth == 0) 88 if (IS_ERR(head))
103 return table; 89 return ERR_CAST(head);
104 90
105 for (i = 1; table && (i <= depth); i++) { 91 if (table && !table->child) {
106 ancestor = proc_sys_ancestor(dentry, i); 92 WARN_ON(1);
107 table = proc_sys_lookup_table_one(table, &ancestor->d_name); 93 goto out;
108 if (table)
109 table = table->child;
110 } 94 }
111 return table;
112
113}
114static struct ctl_table *proc_sys_lookup_entry(struct dentry *dparent,
115 struct qstr *name,
116 struct ctl_table *table)
117{
118 table = proc_sys_lookup_table(dparent, table);
119 if (table)
120 table = proc_sys_lookup_table_one(table, name);
121 return table;
122}
123 95
124static struct ctl_table *do_proc_sys_lookup(struct dentry *parent, 96 table = table ? table->child : head->ctl_table;
125 struct qstr *name,
126 struct ctl_table_header **ptr)
127{
128 struct ctl_table_header *head;
129 struct ctl_table *table = NULL;
130 97
131 for (head = sysctl_head_next(NULL); head; 98 p = find_in_table(table, name);
132 head = sysctl_head_next(head)) { 99 if (!p) {
133 table = proc_sys_lookup_entry(parent, name, head->ctl_table); 100 for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
134 if (table) 101 if (h->attached_to != table)
135 break; 102 continue;
103 p = find_in_table(h->attached_by, name);
104 if (p)
105 break;
106 }
136 } 107 }
137 *ptr = head;
138 return table;
139}
140
141static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
142 struct nameidata *nd)
143{
144 struct ctl_table_header *head;
145 struct inode *inode;
146 struct dentry *err;
147 struct ctl_table *table;
148 108
149 err = ERR_PTR(-ENOENT); 109 if (!p)
150 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head);
151 if (!table)
152 goto out; 110 goto out;
153 111
154 err = ERR_PTR(-ENOMEM); 112 err = ERR_PTR(-ENOMEM);
155 inode = proc_sys_make_inode(dir, table); 113 inode = proc_sys_make_inode(dir->i_sb, h ? h : head, p);
114 if (h)
115 sysctl_head_finish(h);
116
156 if (!inode) 117 if (!inode)
157 goto out; 118 goto out;
158 119
@@ -168,22 +129,14 @@ out:
168static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf, 129static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
169 size_t count, loff_t *ppos, int write) 130 size_t count, loff_t *ppos, int write)
170{ 131{
171 struct dentry *dentry = filp->f_dentry; 132 struct inode *inode = filp->f_path.dentry->d_inode;
172 struct ctl_table_header *head; 133 struct ctl_table_header *head = grab_header(inode);
173 struct ctl_table *table; 134 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
174 ssize_t error; 135 ssize_t error;
175 size_t res; 136 size_t res;
176 137
177 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head); 138 if (IS_ERR(head))
178 /* Has the sysctl entry disappeared on us? */ 139 return PTR_ERR(head);
179 error = -ENOENT;
180 if (!table)
181 goto out;
182
183 /* Has the sysctl entry been replaced by a directory? */
184 error = -EISDIR;
185 if (!table->proc_handler)
186 goto out;
187 140
188 /* 141 /*
189 * At this point we know that the sysctl was not unregistered 142 * At this point we know that the sysctl was not unregistered
@@ -193,6 +146,11 @@ static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
193 if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ)) 146 if (sysctl_perm(head->root, table, write ? MAY_WRITE : MAY_READ))
194 goto out; 147 goto out;
195 148
149 /* if that can happen at all, it should be -EINVAL, not -EISDIR */
150 error = -EINVAL;
151 if (!table->proc_handler)
152 goto out;
153
196 /* careful: calling conventions are nasty here */ 154 /* careful: calling conventions are nasty here */
197 res = count; 155 res = count;
198 error = table->proc_handler(table, write, filp, buf, &res, ppos); 156 error = table->proc_handler(table, write, filp, buf, &res, ppos);
@@ -218,82 +176,86 @@ static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
218 176
219 177
220static int proc_sys_fill_cache(struct file *filp, void *dirent, 178static int proc_sys_fill_cache(struct file *filp, void *dirent,
221 filldir_t filldir, struct ctl_table *table) 179 filldir_t filldir,
180 struct ctl_table_header *head,
181 struct ctl_table *table)
222{ 182{
223 struct ctl_table_header *head;
224 struct ctl_table *child_table = NULL;
225 struct dentry *child, *dir = filp->f_path.dentry; 183 struct dentry *child, *dir = filp->f_path.dentry;
226 struct inode *inode; 184 struct inode *inode;
227 struct qstr qname; 185 struct qstr qname;
228 ino_t ino = 0; 186 ino_t ino = 0;
229 unsigned type = DT_UNKNOWN; 187 unsigned type = DT_UNKNOWN;
230 int ret;
231 188
232 qname.name = table->procname; 189 qname.name = table->procname;
233 qname.len = strlen(table->procname); 190 qname.len = strlen(table->procname);
234 qname.hash = full_name_hash(qname.name, qname.len); 191 qname.hash = full_name_hash(qname.name, qname.len);
235 192
236 /* Suppress duplicates.
237 * Only fill a directory entry if it is the value that
238 * an ordinary lookup of that name returns. Hide all
239 * others.
240 *
241 * If we ever cache this translation in the dcache
242 * I should do a dcache lookup first. But for now
243 * it is just simpler not to.
244 */
245 ret = 0;
246 child_table = do_proc_sys_lookup(dir, &qname, &head);
247 sysctl_head_finish(head);
248 if (child_table != table)
249 return 0;
250
251 child = d_lookup(dir, &qname); 193 child = d_lookup(dir, &qname);
252 if (!child) { 194 if (!child) {
253 struct dentry *new; 195 child = d_alloc(dir, &qname);
254 new = d_alloc(dir, &qname); 196 if (child) {
255 if (new) { 197 inode = proc_sys_make_inode(dir->d_sb, head, table);
256 inode = proc_sys_make_inode(dir->d_inode, table); 198 if (!inode) {
257 if (!inode) 199 dput(child);
258 child = ERR_PTR(-ENOMEM); 200 return -ENOMEM;
259 else { 201 } else {
260 new->d_op = &proc_sys_dentry_operations; 202 child->d_op = &proc_sys_dentry_operations;
261 d_add(new, inode); 203 d_add(child, inode);
262 } 204 }
263 if (child) 205 } else {
264 dput(new); 206 return -ENOMEM;
265 else
266 child = new;
267 } 207 }
268 } 208 }
269 if (!child || IS_ERR(child) || !child->d_inode)
270 goto end_instantiate;
271 inode = child->d_inode; 209 inode = child->d_inode;
272 if (inode) { 210 ino = inode->i_ino;
273 ino = inode->i_ino; 211 type = inode->i_mode >> 12;
274 type = inode->i_mode >> 12;
275 }
276 dput(child); 212 dput(child);
277end_instantiate: 213 return !!filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type);
278 if (!ino) 214}
279 ino= find_inode_number(dir, &qname); 215
280 if (!ino) 216static int scan(struct ctl_table_header *head, ctl_table *table,
281 ino = 1; 217 unsigned long *pos, struct file *file,
282 return filldir(dirent, qname.name, qname.len, filp->f_pos, ino, type); 218 void *dirent, filldir_t filldir)
219{
220
221 for (; table->ctl_name || table->procname; table++, (*pos)++) {
222 int res;
223
224 /* Can't do anything without a proc name */
225 if (!table->procname)
226 continue;
227
228 if (*pos < file->f_pos)
229 continue;
230
231 res = proc_sys_fill_cache(file, dirent, filldir, head, table);
232 if (res)
233 return res;
234
235 file->f_pos = *pos + 1;
236 }
237 return 0;
283} 238}
284 239
285static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir) 240static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
286{ 241{
287 struct dentry *dentry = filp->f_dentry; 242 struct dentry *dentry = filp->f_path.dentry;
288 struct inode *inode = dentry->d_inode; 243 struct inode *inode = dentry->d_inode;
289 struct ctl_table_header *head = NULL; 244 struct ctl_table_header *head = grab_header(inode);
290 struct ctl_table *table; 245 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
246 struct ctl_table_header *h = NULL;
291 unsigned long pos; 247 unsigned long pos;
292 int ret; 248 int ret = -EINVAL;
249
250 if (IS_ERR(head))
251 return PTR_ERR(head);
293 252
294 ret = -ENOTDIR; 253 if (table && !table->child) {
295 if (!S_ISDIR(inode->i_mode)) 254 WARN_ON(1);
296 goto out; 255 goto out;
256 }
257
258 table = table ? table->child : head->ctl_table;
297 259
298 ret = 0; 260 ret = 0;
299 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */ 261 /* Avoid a switch here: arm builds fail with missing __cmpdi2 */
@@ -311,30 +273,17 @@ static int proc_sys_readdir(struct file *filp, void *dirent, filldir_t filldir)
311 } 273 }
312 pos = 2; 274 pos = 2;
313 275
314 /* - Find each instance of the directory 276 ret = scan(head, table, &pos, filp, dirent, filldir);
315 * - Read all entries in each instance 277 if (ret)
316 * - Before returning an entry to user space lookup the entry 278 goto out;
317 * by name and if I find a different entry don't return
318 * this one because it means it is a buried dup.
319 * For sysctl this should only happen for directory entries.
320 */
321 for (head = sysctl_head_next(NULL); head; head = sysctl_head_next(head)) {
322 table = proc_sys_lookup_table(dentry, head->ctl_table);
323 279
324 if (!table) 280 for (h = sysctl_head_next(NULL); h; h = sysctl_head_next(h)) {
281 if (h->attached_to != table)
325 continue; 282 continue;
326 283 ret = scan(h, h->attached_by, &pos, filp, dirent, filldir);
327 for (; table->ctl_name || table->procname; table++, pos++) { 284 if (ret) {
328 /* Can't do anything without a proc name */ 285 sysctl_head_finish(h);
329 if (!table->procname) 286 break;
330 continue;
331
332 if (pos < filp->f_pos)
333 continue;
334
335 if (proc_sys_fill_cache(filp, dirent, filldir, table) < 0)
336 goto out;
337 filp->f_pos = pos + 1;
338 } 287 }
339 } 288 }
340 ret = 1; 289 ret = 1;
@@ -349,47 +298,18 @@ static int proc_sys_permission(struct inode *inode, int mask, struct nameidata *
349 * sysctl entries that are not writeable, 298 * sysctl entries that are not writeable,
350 * are _NOT_ writeable, capabilities or not. 299 * are _NOT_ writeable, capabilities or not.
351 */ 300 */
352 struct ctl_table_header *head; 301 struct ctl_table_header *head = grab_header(inode);
353 struct ctl_table *table; 302 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
354 struct dentry *dentry;
355 int mode;
356 int depth;
357 int error; 303 int error;
358 304
359 head = NULL; 305 if (IS_ERR(head))
360 depth = PROC_I(inode)->fd; 306 return PTR_ERR(head);
361
362 /* First check the cached permissions, in case we don't have
363 * enough information to lookup the sysctl table entry.
364 */
365 error = -EACCES;
366 mode = inode->i_mode;
367
368 if (current->euid == 0)
369 mode >>= 6;
370 else if (in_group_p(0))
371 mode >>= 3;
372
373 if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
374 error = 0;
375
376 /* If we can't get a sysctl table entry the permission
377 * checks on the cached mode will have to be enough.
378 */
379 if (!nd || !depth)
380 goto out;
381 307
382 dentry = nd->path.dentry; 308 if (!table) /* global root - r-xr-xr-x */
383 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head); 309 error = mask & MAY_WRITE ? -EACCES : 0;
310 else /* Use the permissions on the sysctl table entry */
311 error = sysctl_perm(head->root, table, mask);
384 312
385 /* If the entry does not exist deny permission */
386 error = -EACCES;
387 if (!table)
388 goto out;
389
390 /* Use the permissions on the sysctl table entry */
391 error = sysctl_perm(head->root, table, mask);
392out:
393 sysctl_head_finish(head); 313 sysctl_head_finish(head);
394 return error; 314 return error;
395} 315}
@@ -409,33 +329,70 @@ static int proc_sys_setattr(struct dentry *dentry, struct iattr *attr)
409 return error; 329 return error;
410} 330}
411 331
412/* I'm lazy and don't distinguish between files and directories, 332static int proc_sys_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
413 * until access time. 333{
414 */ 334 struct inode *inode = dentry->d_inode;
335 struct ctl_table_header *head = grab_header(inode);
336 struct ctl_table *table = PROC_I(inode)->sysctl_entry;
337
338 if (IS_ERR(head))
339 return PTR_ERR(head);
340
341 generic_fillattr(inode, stat);
342 if (table)
343 stat->mode = (stat->mode & S_IFMT) | table->mode;
344
345 sysctl_head_finish(head);
346 return 0;
347}
348
415static const struct file_operations proc_sys_file_operations = { 349static const struct file_operations proc_sys_file_operations = {
416 .read = proc_sys_read, 350 .read = proc_sys_read,
417 .write = proc_sys_write, 351 .write = proc_sys_write,
352};
353
354static const struct file_operations proc_sys_dir_file_operations = {
418 .readdir = proc_sys_readdir, 355 .readdir = proc_sys_readdir,
419}; 356};
420 357
421static const struct inode_operations proc_sys_inode_operations = { 358static const struct inode_operations proc_sys_inode_operations = {
359 .permission = proc_sys_permission,
360 .setattr = proc_sys_setattr,
361 .getattr = proc_sys_getattr,
362};
363
364static const struct inode_operations proc_sys_dir_operations = {
422 .lookup = proc_sys_lookup, 365 .lookup = proc_sys_lookup,
423 .permission = proc_sys_permission, 366 .permission = proc_sys_permission,
424 .setattr = proc_sys_setattr, 367 .setattr = proc_sys_setattr,
368 .getattr = proc_sys_getattr,
425}; 369};
426 370
427static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd) 371static int proc_sys_revalidate(struct dentry *dentry, struct nameidata *nd)
428{ 372{
429 struct ctl_table_header *head; 373 return !PROC_I(dentry->d_inode)->sysctl->unregistering;
430 struct ctl_table *table; 374}
431 table = do_proc_sys_lookup(dentry->d_parent, &dentry->d_name, &head); 375
432 proc_sys_refresh_inode(dentry->d_inode, table); 376static int proc_sys_delete(struct dentry *dentry)
433 sysctl_head_finish(head); 377{
434 return !!table; 378 return !!PROC_I(dentry->d_inode)->sysctl->unregistering;
379}
380
381static int proc_sys_compare(struct dentry *dir, struct qstr *qstr,
382 struct qstr *name)
383{
384 struct dentry *dentry = container_of(qstr, struct dentry, d_name);
385 if (qstr->len != name->len)
386 return 1;
387 if (memcmp(qstr->name, name->name, name->len))
388 return 1;
389 return !sysctl_is_seen(PROC_I(dentry->d_inode)->sysctl);
435} 390}
436 391
437static struct dentry_operations proc_sys_dentry_operations = { 392static struct dentry_operations proc_sys_dentry_operations = {
438 .d_revalidate = proc_sys_revalidate, 393 .d_revalidate = proc_sys_revalidate,
394 .d_delete = proc_sys_delete,
395 .d_compare = proc_sys_compare,
439}; 396};
440 397
441static struct proc_dir_entry *proc_sys_root; 398static struct proc_dir_entry *proc_sys_root;
@@ -443,8 +400,8 @@ static struct proc_dir_entry *proc_sys_root;
443int proc_sys_init(void) 400int proc_sys_init(void)
444{ 401{
445 proc_sys_root = proc_mkdir("sys", NULL); 402 proc_sys_root = proc_mkdir("sys", NULL);
446 proc_sys_root->proc_iops = &proc_sys_inode_operations; 403 proc_sys_root->proc_iops = &proc_sys_dir_operations;
447 proc_sys_root->proc_fops = &proc_sys_file_operations; 404 proc_sys_root->proc_fops = &proc_sys_dir_file_operations;
448 proc_sys_root->nlink = 0; 405 proc_sys_root->nlink = 0;
449 return 0; 406 return 0;
450} 407}
diff --git a/include/linux/proc_fs.h b/include/linux/proc_fs.h
index f560d1705afe..fb61850d1cfc 100644
--- a/include/linux/proc_fs.h
+++ b/include/linux/proc_fs.h
@@ -282,11 +282,16 @@ union proc_op {
282 struct task_struct *task); 282 struct task_struct *task);
283}; 283};
284 284
285struct ctl_table_header;
286struct ctl_table;
287
285struct proc_inode { 288struct proc_inode {
286 struct pid *pid; 289 struct pid *pid;
287 int fd; 290 int fd;
288 union proc_op op; 291 union proc_op op;
289 struct proc_dir_entry *pde; 292 struct proc_dir_entry *pde;
293 struct ctl_table_header *sysctl;
294 struct ctl_table *sysctl_entry;
290 struct inode vfs_inode; 295 struct inode vfs_inode;
291}; 296};
292 297
diff --git a/include/linux/sysctl.h b/include/linux/sysctl.h
index 3f6599aeb0db..d0437f36921f 100644
--- a/include/linux/sysctl.h
+++ b/include/linux/sysctl.h
@@ -961,6 +961,7 @@ struct ctl_table_header;
961 961
962extern void sysctl_head_get(struct ctl_table_header *); 962extern void sysctl_head_get(struct ctl_table_header *);
963extern void sysctl_head_put(struct ctl_table_header *); 963extern void sysctl_head_put(struct ctl_table_header *);
964extern int sysctl_is_seen(struct ctl_table_header *);
964extern struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *); 965extern struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *);
965extern struct ctl_table_header *sysctl_head_next(struct ctl_table_header *prev); 966extern struct ctl_table_header *sysctl_head_next(struct ctl_table_header *prev);
966extern struct ctl_table_header *__sysctl_head_next(struct nsproxy *namespaces, 967extern struct ctl_table_header *__sysctl_head_next(struct nsproxy *namespaces,
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index c9a0af887033..ff5abcca5ddf 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1932,6 +1932,21 @@ void unregister_sysctl_table(struct ctl_table_header * header)
1932 spin_unlock(&sysctl_lock); 1932 spin_unlock(&sysctl_lock);
1933} 1933}
1934 1934
1935int sysctl_is_seen(struct ctl_table_header *p)
1936{
1937 struct ctl_table_set *set = p->set;
1938 int res;
1939 spin_lock(&sysctl_lock);
1940 if (p->unregistering)
1941 res = 0;
1942 else if (!set->is_seen)
1943 res = 1;
1944 else
1945 res = set->is_seen(set);
1946 spin_unlock(&sysctl_lock);
1947 return res;
1948}
1949
1935void setup_sysctl_set(struct ctl_table_set *p, 1950void setup_sysctl_set(struct ctl_table_set *p,
1936 struct ctl_table_set *parent, 1951 struct ctl_table_set *parent,
1937 int (*is_seen)(struct ctl_table_set *)) 1952 int (*is_seen)(struct ctl_table_set *))