xref: /openbmc/linux/fs/notify/fanotify/fanotify.c (revision 9fb29c73)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/fanotify.h>
3 #include <linux/fdtable.h>
4 #include <linux/fsnotify_backend.h>
5 #include <linux/init.h>
6 #include <linux/jiffies.h>
7 #include <linux/kernel.h> /* UINT_MAX */
8 #include <linux/mount.h>
9 #include <linux/sched.h>
10 #include <linux/sched/user.h>
11 #include <linux/sched/signal.h>
12 #include <linux/types.h>
13 #include <linux/wait.h>
14 #include <linux/audit.h>
15 #include <linux/sched/mm.h>
16 
17 #include "fanotify.h"
18 
19 static bool should_merge(struct fsnotify_event *old_fsn,
20 			 struct fsnotify_event *new_fsn)
21 {
22 	struct fanotify_event_info *old, *new;
23 
24 	pr_debug("%s: old=%p new=%p\n", __func__, old_fsn, new_fsn);
25 	old = FANOTIFY_E(old_fsn);
26 	new = FANOTIFY_E(new_fsn);
27 
28 	if (old_fsn->inode == new_fsn->inode && old->pid == new->pid &&
29 	    old->path.mnt == new->path.mnt &&
30 	    old->path.dentry == new->path.dentry)
31 		return true;
32 	return false;
33 }
34 
35 /* and the list better be locked by something too! */
36 static int fanotify_merge(struct list_head *list, struct fsnotify_event *event)
37 {
38 	struct fsnotify_event *test_event;
39 
40 	pr_debug("%s: list=%p event=%p\n", __func__, list, event);
41 
42 	/*
43 	 * Don't merge a permission event with any other event so that we know
44 	 * the event structure we have created in fanotify_handle_event() is the
45 	 * one we should check for permission response.
46 	 */
47 	if (fanotify_is_perm_event(event->mask))
48 		return 0;
49 
50 	list_for_each_entry_reverse(test_event, list, list) {
51 		if (should_merge(test_event, event)) {
52 			test_event->mask |= event->mask;
53 			return 1;
54 		}
55 	}
56 
57 	return 0;
58 }
59 
60 static int fanotify_get_response(struct fsnotify_group *group,
61 				 struct fanotify_perm_event_info *event,
62 				 struct fsnotify_iter_info *iter_info)
63 {
64 	int ret;
65 
66 	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
67 
68 	wait_event(group->fanotify_data.access_waitq, event->response);
69 
70 	/* userspace responded, convert to something usable */
71 	switch (event->response & ~FAN_AUDIT) {
72 	case FAN_ALLOW:
73 		ret = 0;
74 		break;
75 	case FAN_DENY:
76 	default:
77 		ret = -EPERM;
78 	}
79 
80 	/* Check if the response should be audited */
81 	if (event->response & FAN_AUDIT)
82 		audit_fanotify(event->response & ~FAN_AUDIT);
83 
84 	event->response = 0;
85 
86 	pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
87 		 group, event, ret);
88 
89 	return ret;
90 }
91 
92 /*
93  * This function returns a mask for an event that only contains the flags
94  * that have been specifically requested by the user. Flags that may have
95  * been included within the event mask, but have not been explicitly
96  * requested by the user, will not be present in the returned mask.
97  */
98 static u32 fanotify_group_event_mask(struct fsnotify_iter_info *iter_info,
99 				       u32 event_mask, const void *data,
100 				       int data_type)
101 {
102 	__u32 marks_mask = 0, marks_ignored_mask = 0;
103 	const struct path *path = data;
104 	struct fsnotify_mark *mark;
105 	int type;
106 
107 	pr_debug("%s: report_mask=%x mask=%x data=%p data_type=%d\n",
108 		 __func__, iter_info->report_mask, event_mask, data, data_type);
109 
110 	/* If we don't have enough info to send an event to userspace say no */
111 	if (data_type != FSNOTIFY_EVENT_PATH)
112 		return 0;
113 
114 	/* Sorry, fanotify only gives a damn about files and dirs */
115 	if (!d_is_reg(path->dentry) &&
116 	    !d_can_lookup(path->dentry))
117 		return 0;
118 
119 	fsnotify_foreach_obj_type(type) {
120 		if (!fsnotify_iter_should_report_type(iter_info, type))
121 			continue;
122 		mark = iter_info->marks[type];
123 		/*
124 		 * If the event is for a child and this mark doesn't care about
125 		 * events on a child, don't send it!
126 		 */
127 		if (event_mask & FS_EVENT_ON_CHILD &&
128 		    (type != FSNOTIFY_OBJ_TYPE_INODE ||
129 		     !(mark->mask & FS_EVENT_ON_CHILD)))
130 			continue;
131 
132 		marks_mask |= mark->mask;
133 		marks_ignored_mask |= mark->ignored_mask;
134 	}
135 
136 	if (d_is_dir(path->dentry) &&
137 	    !(marks_mask & FS_ISDIR & ~marks_ignored_mask))
138 		return 0;
139 
140 	return event_mask & FANOTIFY_OUTGOING_EVENTS & marks_mask &
141 		~marks_ignored_mask;
142 }
143 
144 struct fanotify_event_info *fanotify_alloc_event(struct fsnotify_group *group,
145 						 struct inode *inode, u32 mask,
146 						 const struct path *path)
147 {
148 	struct fanotify_event_info *event = NULL;
149 	gfp_t gfp = GFP_KERNEL_ACCOUNT;
150 
151 	/*
152 	 * For queues with unlimited length lost events are not expected and
153 	 * can possibly have security implications. Avoid losing events when
154 	 * memory is short.
155 	 */
156 	if (group->max_events == UINT_MAX)
157 		gfp |= __GFP_NOFAIL;
158 
159 	/* Whoever is interested in the event, pays for the allocation. */
160 	memalloc_use_memcg(group->memcg);
161 
162 	if (fanotify_is_perm_event(mask)) {
163 		struct fanotify_perm_event_info *pevent;
164 
165 		pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
166 		if (!pevent)
167 			goto out;
168 		event = &pevent->fae;
169 		pevent->response = 0;
170 		goto init;
171 	}
172 	event = kmem_cache_alloc(fanotify_event_cachep, gfp);
173 	if (!event)
174 		goto out;
175 init: __maybe_unused
176 	fsnotify_init_event(&event->fse, inode, mask);
177 	if (FAN_GROUP_FLAG(group, FAN_REPORT_TID))
178 		event->pid = get_pid(task_pid(current));
179 	else
180 		event->pid = get_pid(task_tgid(current));
181 	if (path) {
182 		event->path = *path;
183 		path_get(&event->path);
184 	} else {
185 		event->path.mnt = NULL;
186 		event->path.dentry = NULL;
187 	}
188 out:
189 	memalloc_unuse_memcg();
190 	return event;
191 }
192 
193 static int fanotify_handle_event(struct fsnotify_group *group,
194 				 struct inode *inode,
195 				 u32 mask, const void *data, int data_type,
196 				 const unsigned char *file_name, u32 cookie,
197 				 struct fsnotify_iter_info *iter_info)
198 {
199 	int ret = 0;
200 	struct fanotify_event_info *event;
201 	struct fsnotify_event *fsn_event;
202 
203 	BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
204 	BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
205 	BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
206 	BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
207 	BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
208 	BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
209 	BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
210 	BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
211 	BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
212 	BUILD_BUG_ON(FAN_ONDIR != FS_ISDIR);
213 	BUILD_BUG_ON(FAN_OPEN_EXEC != FS_OPEN_EXEC);
214 	BUILD_BUG_ON(FAN_OPEN_EXEC_PERM != FS_OPEN_EXEC_PERM);
215 
216 	BUILD_BUG_ON(HWEIGHT32(ALL_FANOTIFY_EVENT_BITS) != 12);
217 
218 	mask = fanotify_group_event_mask(iter_info, mask, data, data_type);
219 	if (!mask)
220 		return 0;
221 
222 	pr_debug("%s: group=%p inode=%p mask=%x\n", __func__, group, inode,
223 		 mask);
224 
225 	if (fanotify_is_perm_event(mask)) {
226 		/*
227 		 * fsnotify_prepare_user_wait() fails if we race with mark
228 		 * deletion.  Just let the operation pass in that case.
229 		 */
230 		if (!fsnotify_prepare_user_wait(iter_info))
231 			return 0;
232 	}
233 
234 	event = fanotify_alloc_event(group, inode, mask, data);
235 	ret = -ENOMEM;
236 	if (unlikely(!event)) {
237 		/*
238 		 * We don't queue overflow events for permission events as
239 		 * there the access is denied and so no event is in fact lost.
240 		 */
241 		if (!fanotify_is_perm_event(mask))
242 			fsnotify_queue_overflow(group);
243 		goto finish;
244 	}
245 
246 	fsn_event = &event->fse;
247 	ret = fsnotify_add_event(group, fsn_event, fanotify_merge);
248 	if (ret) {
249 		/* Permission events shouldn't be merged */
250 		BUG_ON(ret == 1 && mask & FANOTIFY_PERM_EVENTS);
251 		/* Our event wasn't used in the end. Free it. */
252 		fsnotify_destroy_event(group, fsn_event);
253 
254 		ret = 0;
255 	} else if (fanotify_is_perm_event(mask)) {
256 		ret = fanotify_get_response(group, FANOTIFY_PE(fsn_event),
257 					    iter_info);
258 		fsnotify_destroy_event(group, fsn_event);
259 	}
260 finish:
261 	if (fanotify_is_perm_event(mask))
262 		fsnotify_finish_user_wait(iter_info);
263 
264 	return ret;
265 }
266 
267 static void fanotify_free_group_priv(struct fsnotify_group *group)
268 {
269 	struct user_struct *user;
270 
271 	user = group->fanotify_data.user;
272 	atomic_dec(&user->fanotify_listeners);
273 	free_uid(user);
274 }
275 
276 static void fanotify_free_event(struct fsnotify_event *fsn_event)
277 {
278 	struct fanotify_event_info *event;
279 
280 	event = FANOTIFY_E(fsn_event);
281 	path_put(&event->path);
282 	put_pid(event->pid);
283 	if (fanotify_is_perm_event(fsn_event->mask)) {
284 		kmem_cache_free(fanotify_perm_event_cachep,
285 				FANOTIFY_PE(fsn_event));
286 		return;
287 	}
288 	kmem_cache_free(fanotify_event_cachep, event);
289 }
290 
291 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
292 {
293 	kmem_cache_free(fanotify_mark_cache, fsn_mark);
294 }
295 
296 const struct fsnotify_ops fanotify_fsnotify_ops = {
297 	.handle_event = fanotify_handle_event,
298 	.free_group_priv = fanotify_free_group_priv,
299 	.free_event = fanotify_free_event,
300 	.free_mark = fanotify_free_mark,
301 };
302