随便写写

本文记录本人学习LDD3的过程，零碎记录，只为自己留个笔记。观赏者请勿因其零碎职责我，呵呵。

本文的扩展代码的相关分析，除特别指明外，将基于linux 2.6.34, archs=arm。

从最简单的概念开始吧，设备被linux系统分为三种基本类型（字符模块，块模块和网络模块），每个模块通常实现为其中某一类。注意这里讲的是通常，不是必须的。程序员可以构造一个大的模块，在其中实现不同类型的设备驱动程序。当然，通常也就是说明了最好的方式，所以不通常的行为不被指责，但也不被提倡。就像脚踩几条船的行为，不能予以法律追究，仅仅是道德上鄙视一下。

世间万物，皆因看事物的眼睛角度不同而不同，当然也存在另外一种划分驱动程序模块类型的方法。一般而言，某些驱动程序类型同内核用来支持某种给定类型设备的附加层一起工作。比如，通用串行总线(usb)模块、串行模块、SCSI模块，等等。每个USB设备由一个USB模块驱动，而该USB模块和USB子系统一同工作，但设备本身在系统中表现为一个字符设备（USB串口）、一个块设备（USB存储卡读取器），或者一个网络设备（USB以太网接口）。

记得方世玉的老妈的暧昧师兄一直说的口头禅吗？“安全第一，安全第一”。做事之前还是要先来看看几个安全问题的原则性概念。

1.系统中的所有安全检查都是由内核代码进行的，如果内核有安全漏洞，则整个系统就有安全漏洞，所以只能允许授权用户才能装载模块。

2.驱动程序中应当尽量避免在代码中实现安全策略。

3.任何从内核中得到的内存，都必须在提供给用户进程或者设备之前清零或者以其他方式初始化，否则就可能发生信息泄露。

4.由于内核中并发的存在，linux内核代码（包括驱动程序代码）必须是可以重入的。

这里要说一下，书中关于ldd3中代码的下载位置是不对的，应该是ftp://ar.linux.it/pub/ldd3，不用感谢我，你也很容易知道为什么不对。

驱动程序可以使用内核的部分函数，比如printk，为什么呢？首先因为内核运行不依赖于C库，故当然不可以用printf函数，而模块能够调用内核的printk是因为在insmod装入模块后，模块就连接到了内核，因而可以访问内核的公用符号（包含函数和变量）。

在linux内核编程中会遇到一个current变量，多数人知道这是一个全局变量，是一个指向struct task_struct的指针。实际上，2.6中的current不再是一个全局变量，而是将其隐藏在内核栈中。内核程序（当然也包括内核空间下运行的设备驱动程序）只要包含<linux/sched.h>头文件，即可使用current引用当前进程。

例如，下面的语句通过访问struct task_struct的某些成员来打印当前进程的进程ID和命令名。

printk(KERN_INFO "The process is /"%s'" (pid %i)/n:, current->comm, current->pid);

current->comm成员存储当前进程所执行的程序文件的基本名称（base name）， 如果必要，会裁剪到15个字符以内。

current指针指向的结构体struct task_struct定义在linux/include/linux/sched.h中。

1170 struct task_struct { 1171 volatile long state; /* -1 unrunnable, 0 runnable, >0 stopped */ 1172 void *stack; 1173 atomic_t usage; 1174 unsigned int flags; /* per process flags, defined below */ 1175 unsigned int ptrace; 1176 1177 int lock_depth; /* BKL lock depth */ 1178 1179 #ifdef CONFIG_SMP 1180 #ifdef __ARCH_WANT_UNLOCKED_CTXSW 1181 int oncpu; 1182 #endif 1183 #endif 1184 1185 int prio, static_prio, normal_prio; 1186 unsigned int rt_priority; 1187 const struct sched_class *sched_class; 1188 struct sched_entity se; 1189 struct sched_rt_entity rt; 1190 1191 #ifdef CONFIG_PREEMPT_NOTIFIERS 1192 /* list of struct preempt_notifier: */ 1193 struct hlist_head preempt_notifiers; 1194 #endif 1195 1196 /* 1197 * fpu_counter contains the number of consecutive context switches 1198 * that the FPU is used. If this is over a threshold, the lazy fpu 1199 * saving becomes unlazy to save the trap. This is an unsigned char 1200 * so that after 256 times the counter wraps and the behavior turns 1201 * lazy again; this to deal with bursty apps that only use FPU for 1202 * a short time 1203 */ 1204 unsigned char fpu_counter; 1205 #ifdef CONFIG_BLK_DEV_IO_TRACE 1206 unsigned int btrace_seq; 1207 #endif 1208 1209 unsigned int policy; 1210 cpumask_t cpus_allowed; 1211 1212 #ifdef CONFIG_TREE_PREEMPT_RCU 1213 int rcu_read_lock_nesting; 1214 char rcu_read_unlock_special; 1215 struct rcu_node *rcu_blocked_node; 1216 struct list_head rcu_node_entry; 1217 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ 1218 1219 #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) 1220 struct sched_info sched_info; 1221 #endif 1222 1223 struct list_head tasks; 1224 struct plist_node pushable_tasks; 1225 1226 struct mm_struct *mm, *active_mm; 1227 #if defined(SPLIT_RSS_COUNTING) 1228 struct task_rss_stat rss_stat; 1229 #endif 1230 /* task state */ 1231 int exit_state; 1232 int exit_code, exit_signal; 1233 int pdeath_signal; /* The signal sent when the parent dies */ 1234 /* ??? */ 1235 unsigned int personality; 1236 unsigned did_exec:1; 1237 unsigned in_execve:1; /* Tell the LSMs that the process is doing an 1238 * execve */ 1239 unsigned in_iowait:1; 1240 1241 1242 /* Revert to default priority/policy when forking */ 1243 unsigned sched_reset_on_fork:1; 1244 1245 pid_t pid; 1246 pid_t tgid; 1247 1248 #ifdef CONFIG_CC_STACKPROTECTOR 1249 /* Canary value for the -fstack-protector gcc feature */ 1250 unsigned long stack_canary; 1251 #endif 1252 1253 /* 1254 * pointers to (original) parent process, youngest child, younger sibling, 1255 * older sibling, respectively. (p->father can be replaced with 1256 * p->real_parent->pid) 1257 */ 1258 struct task_struct *real_parent; /* real parent process */ 1259 struct task_struct *parent; /* recipient of SIGCHLD, wait4() reports */ 1260 /* 1261 * children/sibling forms the list of my natural children 1262 */ 1263 struct list_head children; /* list of my children */ 1264 struct list_head sibling; /* linkage in my parent's children list */ 1265 struct task_struct *group_leader; /* threadgroup leader */ 1266 1267 /* 1268 * ptraced is the list of tasks this task is using ptrace on. 1269 * This includes both natural children and PTRACE_ATTACH targets. 1270 * p->ptrace_entry is p's link on the p->parent->ptraced list. 1271 */ 1272 struct list_head ptraced; 1273 struct list_head ptrace_entry; 1274 1275 /* 1276 * This is the tracer handle for the ptrace BTS extension. 1277 * This field actually belongs to the ptracer task. 1278 */ 1279 struct bts_context *bts; 1280 1281 /* PID/PID hash table linkage. */ 1282 struct pid_link pids[PIDTYPE_MAX]; 1283 struct list_head thread_group; 1284 1285 struct completion *vfork_done; /* for vfork() */ 1286 int __user *set_child_tid; /* CLONE_CHILD_SETTID */ 1287 int __user *clear_child_tid; /* CLONE_CHILD_CLEARTID */ 1288 1289 cputime_t utime, stime, utimescaled, stimescaled; 1290 cputime_t gtime; 1291 #ifndef CONFIG_VIRT_CPU_ACCOUNTING 1292 cputime_t prev_utime, prev_stime; 1293 #endif 1294 unsigned long nvcsw, nivcsw; /* context switch counts */ 1295 struct timespec start_time; /* monotonic time */ 1296 struct timespec real_start_time; /* boot based time */ 1297 /* mm fault and swap info: this can arguably be seen as either mm-specific or thread-specific */ 1298 unsigned long min_flt, maj_flt; 1299 1300 struct task_cputime cputime_expires; 1301 struct list_head cpu_timers[3]; 1302 1303 /* process credentials */ 1304 const struct cred *real_cred; /* objective and real subjective task 1305 * credentials (COW) */ 1306 const struct cred *cred; /* effective (overridable) subjective task 1307 * credentials (COW) */ 1308 struct mutex cred_guard_mutex; /* guard against foreign influences on 1309 * credential calculations 1310 * (notably. ptrace) */ 1311 struct cred *replacement_session_keyring; /* for KEYCTL_SESSION_TO_PARENT */ 1312 1313 char comm[TASK_COMM_LEN]; /* executable name excluding path 1314 - access with [gs]et_task_comm (which lock 1315 it with task_lock()) 1316 - initialized normally by setup_new_exec */ 1317 /* file system info */ 1318 int link_count, total_link_count; 1319 #ifdef CONFIG_SYSVIPC 1320 /* ipc stuff */ 1321 struct sysv_sem sysvsem; 1322 #endif 1323 #ifdef CONFIG_DETECT_HUNG_TASK 1324 /* hung task detection */ 1325 unsigned long last_switch_count; 1326 #endif 1327 /* CPU-specific state of this task */ 1328 struct thread_struct thread; 1329 /* filesystem information */ 1330 struct fs_struct *fs; 1331 /* open file information */ 1332 struct files_struct *files; 1333 /* namespaces */ 1334 struct nsproxy *nsproxy; 1335 /* signal handlers */ 1336 struct signal_struct *signal; 1337 struct sighand_struct *sighand; 1338 1339 sigset_t blocked, real_blocked; 1340 sigset_t saved_sigmask; /* restored if set_restore_sigmask() was used */ 1341 struct sigpending pending; 1342 1343 unsigned long sas_ss_sp; 1344 size_t sas_ss_size; 1345 int (*notifier)(void *priv); 1346 void *notifier_data; 1347 sigset_t *notifier_mask; 1348 struct audit_context *audit_context; 1349 #ifdef CONFIG_AUDITSYSCALL 1350 uid_t loginuid; 1351 unsigned int sessionid; 1352 #endif 1353 seccomp_t seccomp; 1354 1355 /* Thread group tracking */ 1356 u32 parent_exec_id; 1357 u32 self_exec_id; 1358 /* Protection of (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, 1359 * mempolicy */ 1360 spinlock_t alloc_lock; 1361 1362 #ifdef CONFIG_GENERIC_HARDIRQS 1363 /* IRQ handler threads */ 1364 struct irqaction *irqaction; 1365 #endif 1366 1367 /* Protection of the PI data structures: */ 1368 raw_spinlock_t pi_lock; 1369 1370 #ifdef CONFIG_RT_MUTEXES 1371 /* PI waiters blocked on a rt_mutex held by this task */ 1372 struct plist_head pi_waiters; 1373 /* Deadlock detection and priority inheritance handling */ 1374 struct rt_mutex_waiter *pi_blocked_on; 1375 #endif 1376 1377 #ifdef CONFIG_DEBUG_MUTEXES 1378 /* mutex deadlock detection */ 1379 struct mutex_waiter *blocked_on; 1380 #endif 1381 #ifdef CONFIG_TRACE_IRQFLAGS 1382 unsigned int irq_events; 1383 unsigned long hardirq_enable_ip; 1384 unsigned long hardirq_disable_ip; 1385 unsigned int hardirq_enable_event; 1386 unsigned int hardirq_disable_event; 1387 int hardirqs_enabled; 1388 int hardirq_context; 1389 unsigned long softirq_disable_ip; 1390 unsigned long softirq_enable_ip; 1391 unsigned int softirq_disable_event; 1392 unsigned int softirq_enable_event; 1393 int softirqs_enabled; 1394 int softirq_context; 1395 #endif 1396 #ifdef CONFIG_LOCKDEP 1397 # define MAX_LOCK_DEPTH 48UL 1398 u64 curr_chain_key; 1399 int lockdep_depth; 1400 unsigned int lockdep_recursion; 1401 struct held_lock held_locks[MAX_LOCK_DEPTH]; 1402 gfp_t lockdep_reclaim_gfp; 1403 #endif 1404 1405 /* journalling filesystem info */ 1406 void *journal_info; 1407 1408 /* stacked block device info */ 1409 struct bio_list *bio_list; 1410 1411 /* VM state */ 1412 struct reclaim_state *reclaim_state; 1413 1414 struct backing_dev_info *backing_dev_info; 1415 1416 struct io_context *io_context; 1417 1418 unsigned long ptrace_message; 1419 siginfo_t *last_siginfo; /* For ptrace use. */ 1420 struct task_io_accounting ioac; 1421 #if defined(CONFIG_TASK_XACCT) 1422 u64 acct_rss_mem1; /* accumulated rss usage */ 1423 u64 acct_vm_mem1; /* accumulated virtual memory usage */ 1424 cputime_t acct_timexpd; /* stime + utime since last update */ 1425 #endif 1426 #ifdef CONFIG_CPUSETS 1427 nodemask_t mems_allowed; /* Protected by alloc_lock */ 1428 int cpuset_mem_spread_rotor; 1429 #endif 1430 #ifdef CONFIG_CGROUPS 1431 /* Control Group info protected by css_set_lock */ 1432 struct css_set *cgroups; 1433 /* cg_list protected by css_set_lock and tsk->alloc_lock */ 1434 struct list_head cg_list; 1435 #endif 1436 #ifdef CONFIG_FUTEX 1437 struct robust_list_head __user *robust_list; 1438 #ifdef CONFIG_COMPAT 1439 struct compat_robust_list_head __user *compat_robust_list; 1440 #endif 1441 struct list_head pi_state_list; 1442 struct futex_pi_state *pi_state_cache; 1443 #endif 1444 #ifdef CONFIG_PERF_EVENTS 1445 struct perf_event_context *perf_event_ctxp; 1446 struct mutex perf_event_mutex; 1447 struct list_head perf_event_list; 1448 #endif 1449 #ifdef CONFIG_NUMA 1450 struct mempolicy *mempolicy; /* Protected by alloc_lock */ 1451 short il_next; 1452 #endif 1453 atomic_t fs_excl; /* holding fs exclusive resources */ 1454 struct rcu_head rcu; 1455 1456 /* 1457 * cache last used pipe for splice 1458 */ 1459 struct pipe_inode_info *splice_pipe; 1460 #ifdef CONFIG_TASK_DELAY_ACCT 1461 struct task_delay_info *delays; 1462 #endif 1463 #ifdef CONFIG_FAULT_INJECTION 1464 int make_it_fail; 1465 #endif 1466 struct prop_local_single dirties; 1467 #ifdef CONFIG_LATENCYTOP 1468 int latency_record_count; 1469 struct latency_record latency_record[LT_SAVECOUNT]; 1470 #endif 1471 /* 1472 * time slack values; these are used to round up poll() and 1473 * select() etc timeout values. These are in nanoseconds. 1474 */ 1475 unsigned long timer_slack_ns; 1476 unsigned long default_timer_slack_ns; 1477 1478 struct list_head *scm_work_list; 1479 #ifdef CONFIG_FUNCTION_GRAPH_TRACER 1480 /* Index of current stored address in ret_stack */ 1481 int curr_ret_stack; 1482 /* Stack of return addresses for return function tracing */ 1483 struct ftrace_ret_stack *ret_stack; 1484 /* time stamp for last schedule */ 1485 unsigned long long ftrace_timestamp; 1486 /* 1487 * Number of functions that haven't been traced 1488 * because of depth overrun. 1489 */ 1490 atomic_t trace_overrun; 1491 /* Pause for the tracing */ 1492 atomic_t tracing_graph_pause; 1493 #endif 1494 #ifdef CONFIG_TRACING 1495 /* state flags for use by tracers */ 1496 unsigned long trace; 1497 /* bitmask of trace recursion */ 1498 unsigned long trace_recursion; 1499 #endif /* CONFIG_TRACING */ 1500 #ifdef CONFIG_CGROUP_MEM_RES_CTLR /* memcg uses this to do batch job */ 1501 struct memcg_batch_info { 1502 int do_batch; /* incremented when batch uncharge started */ 1503 struct mem_cgroup *memcg; /* target memcg of uncharge */ 1504 unsigned long bytes; /* uncharged usage */ 1505 unsigned long memsw_bytes; /* uncharged mem+swap usage */ 1506 } memcg_batch; 1507 #endif 1508 };

内核代码有很多方面区别于用户空间的编程，这里的问题比较多，先点下基本的。

1.内核栈很小，驱动程序和内核共同享用，所以，声明大的自动变量并不是好主意，如果需要大的结构，则应该在调用时动态分配该结构。（想一想：动态分配的占用什么内存空间呢？）

2.关于具有两个下划线前缀（__）的函数，具有这种名字的函数通常是接口的底层组件，应谨慎使用。

3.内核代码不能实现浮点数运算。

下面来到编译模块，linux的make语法很奇怪，看下最简单的对“hello world”程序的makefile：

obj-m := hello.o

（想一想：":="的意思是什么？）

上门的赋值语句（它利用了GNU make的扩展语法）说明了有一个模块需要从目标文件hello.o中构造，这个模块名称为hello.ko。

如果源文件有两个，则可以参考下面的例子：

obj-m := module.o

module-objs := file1.o file2.o

由于这种makefile不是通常的makefile文件，想要其能正常工作，必须在大的内核构造系统环境中调用它。这里就必须建立内核源码树（比如~/kernel-2.6）。则可在该makefile的当前目录下执行：

make -C ~/kernel-2.6 M='pwd' modules

上述命令首先该买目录到-C选项指定的目录，其中保存有内核的顶层makefile文件。M=选项让该makefile在构造modules目标之前返回到模块源代码目录（当前目录），然后modules目标指向obj-m变量中设定的模块（module.o）。

如果你不喜欢上面的做法，看看linux内核开发者提供的另外一种makefile的写法（更是难解，这里不解释了）。