cgroups: revamp subsys array
[linux-2.6.git] / kernel / printk.c
index 41fe609..75077ad 100644 (file)
@@ -33,6 +33,9 @@
 #include <linux/bootmem.h>
 #include <linux/syscalls.h>
 #include <linux/kexec.h>
+#include <linux/ratelimit.h>
+#include <linux/kmsg_dump.h>
+#include <linux/syslog.h>
 
 #include <asm/uaccess.h>
 
@@ -141,6 +144,7 @@ static char __log_buf[__LOG_BUF_LEN];
 static char *log_buf = __log_buf;
 static int log_buf_len = __LOG_BUF_LEN;
 static unsigned logged_chars; /* Number of chars produced since last read+clear operation */
+static int saved_console_loglevel = -1;
 
 #ifdef CONFIG_KEXEC
 /*
@@ -204,12 +208,11 @@ __setup("log_buf_len=", log_buf_len_setup);
 #ifdef CONFIG_BOOT_PRINTK_DELAY
 
 static unsigned int boot_delay; /* msecs delay after each printk during bootup */
-static unsigned long long printk_delay_msec; /* per msec, based on boot_delay */
+static unsigned long long loops_per_msec;      /* based on boot_delay */
 
 static int __init boot_delay_setup(char *str)
 {
        unsigned long lpj;
-       unsigned long long loops_per_msec;
 
        lpj = preset_lpj ? preset_lpj : 1000000;        /* some guess */
        loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
@@ -218,10 +221,9 @@ static int __init boot_delay_setup(char *str)
        if (boot_delay > 10 * 1000)
                boot_delay = 0;
 
-       printk_delay_msec = loops_per_msec;
-       printk(KERN_DEBUG "boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
-               "HZ: %d, printk_delay_msec: %llu\n",
-               boot_delay, preset_lpj, lpj, HZ, printk_delay_msec);
+       pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
+               "HZ: %d, loops_per_msec: %llu\n",
+               boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
        return 1;
 }
 __setup("boot_delay=", boot_delay_setup);
@@ -234,7 +236,7 @@ static void boot_delay_msec(void)
        if (boot_delay == 0 || system_state != SYSTEM_BOOTING)
                return;
 
-       k = (unsigned long long)printk_delay_msec * boot_delay;
+       k = (unsigned long long)loops_per_msec * boot_delay;
 
        timeout = jiffies + msecs_to_jiffies(boot_delay);
        while (k) {
@@ -256,38 +258,23 @@ static inline void boot_delay_msec(void)
 }
 #endif
 
-/*
- * Commands to do_syslog:
- *
- *     0 -- Close the log.  Currently a NOP.
- *     1 -- Open the log. Currently a NOP.
- *     2 -- Read from the log.
- *     3 -- Read all messages remaining in the ring buffer.
- *     4 -- Read and clear all messages remaining in the ring buffer
- *     5 -- Clear ring buffer.
- *     6 -- Disable printk's to console
- *     7 -- Enable printk's to console
- *     8 -- Set level of messages printed to console
- *     9 -- Return number of unread characters in the log buffer
- *     10 -- Return size of the log buffer
- */
-int do_syslog(int type, char __user *buf, int len)
+int do_syslog(int type, char __user *buf, int len, bool from_file)
 {
        unsigned i, j, limit, count;
        int do_clear = 0;
        char c;
        int error = 0;
 
-       error = security_syslog(type);
+       error = security_syslog(type, from_file);
        if (error)
                return error;
 
        switch (type) {
-       case 0:         /* Close log */
+       case SYSLOG_ACTION_CLOSE:       /* Close log */
                break;
-       case 1:         /* Open log */
+       case SYSLOG_ACTION_OPEN:        /* Open log */
                break;
-       case 2:         /* Read from log */
+       case SYSLOG_ACTION_READ:        /* Read from log */
                error = -EINVAL;
                if (!buf || len < 0)
                        goto out;
@@ -318,10 +305,12 @@ int do_syslog(int type, char __user *buf, int len)
                if (!error)
                        error = i;
                break;
-       case 4:         /* Read/clear last kernel messages */
+       /* Read/clear last kernel messages */
+       case SYSLOG_ACTION_READ_CLEAR:
                do_clear = 1;
                /* FALL THRU */
-       case 3:         /* Read last kernel messages */
+       /* Read last kernel messages */
+       case SYSLOG_ACTION_READ_ALL:
                error = -EINVAL;
                if (!buf || len < 0)
                        goto out;
@@ -374,28 +363,41 @@ int do_syslog(int type, char __user *buf, int len)
                        }
                }
                break;
-       case 5:         /* Clear ring buffer */
+       /* Clear ring buffer */
+       case SYSLOG_ACTION_CLEAR:
                logged_chars = 0;
                break;
-       case 6:         /* Disable logging to console */
+       /* Disable logging to console */
+       case SYSLOG_ACTION_CONSOLE_OFF:
+               if (saved_console_loglevel == -1)
+                       saved_console_loglevel = console_loglevel;
                console_loglevel = minimum_console_loglevel;
                break;
-       case 7:         /* Enable logging to console */
-               console_loglevel = default_console_loglevel;
+       /* Enable logging to console */
+       case SYSLOG_ACTION_CONSOLE_ON:
+               if (saved_console_loglevel != -1) {
+                       console_loglevel = saved_console_loglevel;
+                       saved_console_loglevel = -1;
+               }
                break;
-       case 8:         /* Set level of messages printed to console */
+       /* Set level of messages printed to console */
+       case SYSLOG_ACTION_CONSOLE_LEVEL:
                error = -EINVAL;
                if (len < 1 || len > 8)
                        goto out;
                if (len < minimum_console_loglevel)
                        len = minimum_console_loglevel;
                console_loglevel = len;
+               /* Implicitly re-enable logging to console */
+               saved_console_loglevel = -1;
                error = 0;
                break;
-       case 9:         /* Number of chars in the log buffer */
+       /* Number of chars in the log buffer */
+       case SYSLOG_ACTION_SIZE_UNREAD:
                error = log_end - log_start;
                break;
-       case 10:        /* Size of the log buffer */
+       /* Size of the log buffer */
+       case SYSLOG_ACTION_SIZE_BUFFER:
                error = log_buf_len;
                break;
        default:
@@ -408,7 +410,7 @@ out:
 
 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
 {
-       return do_syslog(type, buf, len);
+       return do_syslog(type, buf, len, SYSLOG_FROM_CALL);
 }
 
 /*
@@ -646,6 +648,20 @@ static int recursion_bug;
 static int new_text_line = 1;
 static char printk_buf[1024];
 
+int printk_delay_msec __read_mostly;
+
+static inline void printk_delay(void)
+{
+       if (unlikely(printk_delay_msec)) {
+               int m = printk_delay_msec;
+
+               while (m--) {
+                       mdelay(1);
+                       touch_nmi_watchdog();
+               }
+       }
+}
+
 asmlinkage int vprintk(const char *fmt, va_list args)
 {
        int printed_len = 0;
@@ -655,6 +671,7 @@ asmlinkage int vprintk(const char *fmt, va_list args)
        char *p;
 
        boot_delay_msec();
+       printk_delay();
 
        preempt_disable();
        /* This stops the holder of console_sem just where we want him */
@@ -1066,12 +1083,6 @@ void __sched console_conditional_schedule(void)
 }
 EXPORT_SYMBOL(console_conditional_schedule);
 
-void console_print(const char *s)
-{
-       printk(KERN_EMERG "%s", s);
-}
-EXPORT_SYMBOL(console_print);
-
 void console_unblank(void)
 {
        struct console *c;
@@ -1240,22 +1251,14 @@ void register_console(struct console *newcon)
        if (!(newcon->flags & CON_ENABLED))
                return;
 
-       if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
-               /* we need to iterate through twice, to make sure we print
-                * everything out, before we unregister the console(s)
-                */
-               printk(KERN_INFO "console handover:");
-               for_each_console(bcon)
-                       printk("boot [%s%d] ", bcon->name, bcon->index);
-               printk(" -> real [%s%d]\n", newcon->name, newcon->index);
-               for_each_console(bcon)
-                       unregister_console(bcon);
+       /*
+        * If we have a bootconsole, and are switching to a real console,
+        * don't print everything out again, since when the boot console, and
+        * the real console are the same physical device, it's annoying to
+        * see the beginning boot messages twice
+        */
+       if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
                newcon->flags &= ~CON_PRINTBUFFER;
-       } else {
-               printk(KERN_INFO "%sconsole [%s%d] enabled\n",
-                       (newcon->flags & CON_BOOT) ? "boot" : "" ,
-                       newcon->name, newcon->index);
-       }
 
        /*
         *      Put this console in the list - keep the
@@ -1281,6 +1284,28 @@ void register_console(struct console *newcon)
                spin_unlock_irqrestore(&logbuf_lock, flags);
        }
        release_console_sem();
+
+       /*
+        * By unregistering the bootconsoles after we enable the real console
+        * we get the "console xxx enabled" message on all the consoles -
+        * boot consoles, real consoles, etc - this is to ensure that end
+        * users know there might be something in the kernel's log buffer that
+        * went to the bootconsole (that they do not see on the real console)
+        */
+       if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
+               /* we need to iterate through twice, to make sure we print
+                * everything out, before we unregister the console(s)
+                */
+               printk(KERN_INFO "console [%s%d] enabled, bootconsole disabled\n",
+                       newcon->name, newcon->index);
+               for_each_console(bcon)
+                       if (bcon->flags & CON_BOOT)
+                               unregister_console(bcon);
+       } else {
+               printk(KERN_INFO "%sconsole [%s%d] enabled\n",
+                       (newcon->flags & CON_BOOT) ? "boot" : "" ,
+                       newcon->name, newcon->index);
+       }
 }
 EXPORT_SYMBOL(register_console);
 
@@ -1329,7 +1354,7 @@ static int __init disable_boot_consoles(void)
                if (con->flags & CON_BOOT) {
                        printk(KERN_INFO "turn off boot console %s%d\n",
                                con->name, con->index);
-                       return unregister_console(con);
+                       unregister_console(con);
                }
        }
        return 0;
@@ -1346,11 +1371,11 @@ late_initcall(disable_boot_consoles);
  */
 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
 
-int printk_ratelimit(void)
+int __printk_ratelimit(const char *func)
 {
-       return __ratelimit(&printk_ratelimit_state);
+       return ___ratelimit(&printk_ratelimit_state, func);
 }
-EXPORT_SYMBOL(printk_ratelimit);
+EXPORT_SYMBOL(__printk_ratelimit);
 
 /**
  * printk_timed_ratelimit - caller-controlled printk ratelimiting
@@ -1374,4 +1399,123 @@ bool printk_timed_ratelimit(unsigned long *caller_jiffies,
        return false;
 }
 EXPORT_SYMBOL(printk_timed_ratelimit);
+
+static DEFINE_SPINLOCK(dump_list_lock);
+static LIST_HEAD(dump_list);
+
+/**
+ * kmsg_dump_register - register a kernel log dumper.
+ * @dumper: pointer to the kmsg_dumper structure
+ *
+ * Adds a kernel log dumper to the system. The dump callback in the
+ * structure will be called when the kernel oopses or panics and must be
+ * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
+ */
+int kmsg_dump_register(struct kmsg_dumper *dumper)
+{
+       unsigned long flags;
+       int err = -EBUSY;
+
+       /* The dump callback needs to be set */
+       if (!dumper->dump)
+               return -EINVAL;
+
+       spin_lock_irqsave(&dump_list_lock, flags);
+       /* Don't allow registering multiple times */
+       if (!dumper->registered) {
+               dumper->registered = 1;
+               list_add_tail(&dumper->list, &dump_list);
+               err = 0;
+       }
+       spin_unlock_irqrestore(&dump_list_lock, flags);
+
+       return err;
+}
+EXPORT_SYMBOL_GPL(kmsg_dump_register);
+
+/**
+ * kmsg_dump_unregister - unregister a kmsg dumper.
+ * @dumper: pointer to the kmsg_dumper structure
+ *
+ * Removes a dump device from the system. Returns zero on success and
+ * %-EINVAL otherwise.
+ */
+int kmsg_dump_unregister(struct kmsg_dumper *dumper)
+{
+       unsigned long flags;
+       int err = -EINVAL;
+
+       spin_lock_irqsave(&dump_list_lock, flags);
+       if (dumper->registered) {
+               dumper->registered = 0;
+               list_del(&dumper->list);
+               err = 0;
+       }
+       spin_unlock_irqrestore(&dump_list_lock, flags);
+
+       return err;
+}
+EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
+
+static const char const *kmsg_reasons[] = {
+       [KMSG_DUMP_OOPS]        = "oops",
+       [KMSG_DUMP_PANIC]       = "panic",
+       [KMSG_DUMP_KEXEC]       = "kexec",
+};
+
+static const char *kmsg_to_str(enum kmsg_dump_reason reason)
+{
+       if (reason >= ARRAY_SIZE(kmsg_reasons) || reason < 0)
+               return "unknown";
+
+       return kmsg_reasons[reason];
+}
+
+/**
+ * kmsg_dump - dump kernel log to kernel message dumpers.
+ * @reason: the reason (oops, panic etc) for dumping
+ *
+ * Iterate through each of the dump devices and call the oops/panic
+ * callbacks with the log buffer.
+ */
+void kmsg_dump(enum kmsg_dump_reason reason)
+{
+       unsigned long end;
+       unsigned chars;
+       struct kmsg_dumper *dumper;
+       const char *s1, *s2;
+       unsigned long l1, l2;
+       unsigned long flags;
+
+       /* Theoretically, the log could move on after we do this, but
+          there's not a lot we can do about that. The new messages
+          will overwrite the start of what we dump. */
+       spin_lock_irqsave(&logbuf_lock, flags);
+       end = log_end & LOG_BUF_MASK;
+       chars = logged_chars;
+       spin_unlock_irqrestore(&logbuf_lock, flags);
+
+       if (logged_chars > end) {
+               s1 = log_buf + log_buf_len - logged_chars + end;
+               l1 = logged_chars - end;
+
+               s2 = log_buf;
+               l2 = end;
+       } else {
+               s1 = "";
+               l1 = 0;
+
+               s2 = log_buf + end - logged_chars;
+               l2 = logged_chars;
+       }
+
+       if (!spin_trylock_irqsave(&dump_list_lock, flags)) {
+               printk(KERN_ERR "dump_kmsg: dump list lock is held during %s, skipping dump\n",
+                               kmsg_to_str(reason));
+               return;
+       }
+       list_for_each_entry(dumper, &dump_list, list)
+               dumper->dump(dumper, reason, s1, l1, s2, l2);
+       spin_unlock_irqrestore(&dump_list_lock, flags);
+}
 #endif