request_module 源码分析

一般情况下都在user space通过insmod的方式来装载ko，所以一般insmod ko
的行为时user space 主动的，但是如果想在kernel space insmod ko是否可以呢？
答案是肯定的。
如下面这个例子就在在kernel space 主动要求安装rtc-ds1685
http://lxr.free-electrons.com/source/arch/mips/sgi-ip32/ip32-reset.c#L50
request_module("rtc-ds1685");
下面我们看看request_module 是如何实现主动安装ko的
request_module源码如下：
http://lxr.free-electrons.com/source/include/linux/kmod.h#L37
 38 #define request_module_nowait(mod...) __request_module(false, mod)
 继续看__request_module的实现
 http://lxr.free-electrons.com/source/kernel/kmod.c#L124
 124 int __request_module(bool wait, const char *fmt, ...)
125 {
126         va_list args;
127         char module_name[MODULE_NAME_LEN];
128         unsigned int max_modprobes;
129         int ret;
130         static atomic_t kmod_concurrent = ATOMIC_INIT(0);
131 #define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
132         static int kmod_loop_msg;
133 
134         /*
135          * We don't allow synchronous module loading from async.  Module
136          * init may invoke async_synchronize_full() which will end up
137          * waiting for this task which already is waiting for the module
138          * loading to complete, leading to a deadlock.
139          */
140         WARN_ON_ONCE(wait && current_is_async());
141 
142         if (!modprobe_path[0])
143                 return 0;
144 
145         va_start(args, fmt);
146         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
147         va_end(args);
148         if (ret >= MODULE_NAME_LEN)
149                 return -ENAMETOOLONG;
150 
151         ret = security_kernel_module_request(module_name);
152         if (ret)
153                 return ret;
154 
155         /* If modprobe needs a service that is in a module, we get a recursive
156          * loop.  Limit the number of running kmod threads to max_threads/2 or
157          * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
158          * would be to run the parents of this process, counting how many times
159          * kmod was invoked.  That would mean accessing the internals of the
160          * process tables to get the command line, proc_pid_cmdline is static
161          * and it is not worth changing the proc code just to handle this case. 
162          * KAO.
163          *
164          * "trace the ppid" is simple, but will fail if someone's
165          * parent exits.  I think this is as good as it gets. --RR
166          */
167         max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
168         atomic_inc(&kmod_concurrent);
169         if (atomic_read(&kmod_concurrent) > max_modprobes) {
170                 /* We may be blaming an innocent here, but unlikely */
171                 if (kmod_loop_msg < 5) {
172                         printk(KERN_ERR
173                                "request_module: runaway loop modprobe %s\n",
174                                module_name);
175                         kmod_loop_msg++;
176                 }
177                 atomic_dec(&kmod_concurrent);
178                 return -ENOMEM;
179         }
180 
181         trace_module_request(module_name, wait, _RET_IP_);
182 
183         ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
184 
185         atomic_dec(&kmod_concurrent);
186         return ret;
187 }
188 EXPORT_SYMBOL(__request_module);


在146行通过vsnprintf来格式化字符串，例如本例就是rtc-ds1685，注意ko 名字的长度不能超过MODULE_NAME_LEN
151 kernel 做security检查，在android系统总就对应的selinux
167~179行计算当前有多少个ko是kernel 主动要求insmod的，不能超过thread 个数的一般。
最重要的是183行调用call_modprobe来让user space insmod这个ko
 69 static int call_modprobe(char *module_name, int wait)
 70 {
 71         struct subprocess_info *info;
 72         static char *envp[] = {
 73                 "HOME=/",
 74                 "TERM=linux",
 75                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 76                 NULL
 77         };
 78 
 79         char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
 80         if (!argv)
 81                 goto out;
 82 
 83         module_name = kstrdup(module_name, GFP_KERNEL);
 84         if (!module_name)
 85                 goto free_argv;
 86 
 87         argv[0] = modprobe_path;
 88         argv[1] = "-q";
 89         argv[2] = "--";
 90         argv[3] = module_name;  /* check free_modprobe_argv() */
 91         argv[4] = NULL;
 92 
 93         info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
 94                                          NULL, free_modprobe_argv, NULL);
 95         if (!info)
 96                 goto free_module_name;
 97 
 98         return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
 99 
100 free_module_name:
101         kfree(module_name);
102 free_argv:
103         kfree(argv);
104 out:
105         return -ENOMEM;
106 }


87~91建立env注意其中的 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
93行调用call_usermodehelper_setup 来初始化一个sub_info->work，如530行所示
 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
520                 char **envp, gfp_t gfp_mask,
521                 int (*init)(struct subprocess_info *info, struct cred *new),
522                 void (*cleanup)(struct subprocess_info *info),
523                 void *data)
524 {
525         struct subprocess_info *sub_info;
526         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
527         if (!sub_info)
528                 goto out;
529 
530         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
531         sub_info->path = path;
532         sub_info->argv = argv;
533         sub_info->envp = envp;
534 
535         sub_info->cleanup = cleanup;
536         sub_info->init = init;
537         sub_info->data = data;
538   out:
539         return sub_info;
540 }
541 EXPORT_SYMBOL(call_usermodehelper_setup);
98调用 call_usermodehelper_exec 来让user space 通过modprobe这个命令来insmod这个ko。
我们这个例子中user space要执行的命令为 modprobe rtc-ds1685.ko
我们看call_usermodehelper_exec 主要就是通过system_unbound_wq执行call_usermodehelper_exec 
中建立的work。及call_usermodehelper_exec_work
555 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
556 {
557         DECLARE_COMPLETION_ONSTACK(done);
558         int retval = 0;
559 
560         if (!sub_info->path) {
561                 call_usermodehelper_freeinfo(sub_info);
562                 return -EINVAL;
563         }
564         helper_lock();
565         if (usermodehelper_disabled) {
566                 retval = -EBUSY;
567                 goto out;
568         }
569         /*
570          * Set the completion pointer only if there is a waiter.
571          * This makes it possible to use umh_complete to free
572          * the data structure in case of UMH_NO_WAIT.
573          */
574         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
575         sub_info->wait = wait;
576 
577         queue_work(system_unbound_wq, &sub_info->work);
}
所以我们重点看看call_usermodehelper_exec_work的实现


321 static void call_usermodehelper_exec_work(struct work_struct *work)
322 {
323         struct subprocess_info *sub_info =
324                 container_of(work, struct subprocess_info, work);
325 
326         if (sub_info->wait & UMH_WAIT_PROC) {
327                 call_usermodehelper_exec_sync(sub_info);
328         } else {
329                 pid_t pid;
330                 /*
331                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
332                  * want to pollute current->children, and we need a parent
333                  * that always ignores SIGCHLD to ensure auto-reaping.
334                  */
335                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
336                                     CLONE_PARENT | SIGCHLD);
337                 if (pid < 0) {
338                         sub_info->retval = pid;
339                         umh_complete(sub_info);
340                 }
341         }
342 }
这个函数分两种case，最终都会调用call_usermodehelper_exec_async
269 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
270 {
271         pid_t pid;
272 
273         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
274         kernel_sigaction(SIGCHLD, SIG_DFL);
275         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
276         if (pid < 0) {
277                 sub_info->retval = pid;
278         } else {
279                 int ret = -ECHILD;
280                 /*
281                  * Normally it is bogus to call wait4() from in-kernel because
282                  * wait4() wants to write the exit code to a userspace address.
283                  * But call_usermodehelper_exec_sync() always runs as kernel
284                  * thread (workqueue) and put_user() to a kernel address works
285                  * OK for kernel threads, due to their having an mm_segment_t
286                  * which spans the entire address space.
287                  *
288                  * Thus the __user pointer cast is valid here.
289                  */
290                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
291 
292                 /*
293                  * If ret is 0, either call_usermodehelper_exec_async failed and
294                  * the real error code is already in sub_info->retval or
295                  * sub_info->retval is 0 anyway, so don't mess with it then.
296                  */
297                 if (ret)
298                         sub_info->retval = ret;
299         }
300 
301         /* Restore default kernel sig handler */
302         kernel_sigaction(SIGCHLD, SIG_IGN);
303 
304         umh_complete(sub_info);
305 }
主要是建立一个 call_usermodehelper_exec_async thread
215 static int call_usermodehelper_exec_async(void *data)
216 {
217         struct subprocess_info *sub_info = data;
218         struct cred *new;
219         int retval;
220 
221         spin_lock_irq(&current->sighand->siglock);
222         flush_signal_handlers(current, 1);
223         spin_unlock_irq(&current->sighand->siglock);
224 
225         /*
226          * Our parent (unbound workqueue) runs with elevated scheduling
227          * priority. Avoid propagating that into the userspace child.
228          */
229         set_user_nice(current, 0);
230 
231         retval = -ENOMEM;
232         new = prepare_kernel_cred(current);
233         if (!new)
234                 goto out;
235 
236         spin_lock(&umh_sysctl_lock);
237         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
238         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
239                                              new->cap_inheritable);
240         spin_unlock(&umh_sysctl_lock);
241 
242         if (sub_info->init) {
243                 retval = sub_info->init(sub_info, new);
244                 if (retval) {
245                         abort_creds(new);
246                         goto out;
247                 }
248         }
249 
250         commit_creds(new);
251 
252         retval = do_execve(getname_kernel(sub_info->path),
253                            (const char __user *const __user *)sub_info->argv,
254                            (const char __user *const __user *)sub_info->envp);
255 out:
256         sub_info->retval = retval;
257         /*
258          * call_usermodehelper_exec_sync() will call umh_complete
259          * if UHM_WAIT_PROC.
260          */
261         if (!(sub_info->wait & UMH_WAIT_PROC))
262                 umh_complete(sub_info);
263         if (!retval)
264                 return 0;
265         do_exit(0);
266 }
看到这里就明白了，原来也是通过do_execve来让user space执行insmod命令.