Index: kernel/sched.c =================================================================== --- kernel/sched.c (.../tags/starting_base_kernel_2.6.12) (revision 300) +++ kernel/sched.c (.../trunk/kernel) (revision 300) @@ -49,6 +49,8 @@ #include #include +#include + #include /* @@ -1372,6 +1374,8 @@ rq->prev_mm = oldmm; } + LOGSWITCH(prev,next); + /* Here we just switch the register state and the stack. */ switch_to(prev, next, prev); Index: include/linux/logdev.h =================================================================== --- include/linux/logdev.h (.../tags/starting_base_kernel_2.6.12) (revision 0) +++ include/linux/logdev.h (.../trunk/kernel) (revision 300) @@ -0,0 +1,207 @@ +/* + * Logdevice - A device used to record debuging information in the kernel. + * It uses a large memory ring buffer consisting of individual pages + * to keep down on hogging large sections. A user may then read the device + * to get debugging information out of it. Or if configured, this can + * be dumped to the network on a system crash. + * + * Copyright - 2005 - Steven Rostedt, Kihon Technologies, (rostedt at kihontech dot com) + */ +#ifndef _LOG_DEV_H +#define _LOG_DEV_H + +#define LOGDEV_CUSTOM 0x1afb +#define LOGDEV_SWITCH_ID 0x2afc +#define LOGDEV_PKT_ID 0x42aa +#define LOGDEV_PRINT 0x4adb +#define LOGDEV_PRINT_TIME 0x4adc +#define LOGDEV_PRINT_TIME_FUNC 0x4add + +struct logdev_switch_struct { + unsigned long high; + unsigned long low; + short pid_prev; + short pid_next; + short prev_len; + short next_len; + char prev_comm[0]; + char next_comm[0]; +}; + +struct logdev_pkt { + unsigned long high; + unsigned long low; + short protocol; + short dir; + char packet[0]; +}; + +struct logdev_print { + char str[0]; +}; + +struct logdev_print_time { + unsigned long high; + unsigned long low; + char str[0]; +}; + +struct logdev_print_time_func { + const char *file; + int line; + /* need to be after line, since we use this with print_time. */ + unsigned long high; + unsigned long low; + char str[0]; +}; + +struct logdev_custom { + int id; + char data[0]; +}; + +struct logdev_header { + int id; + int size; + int cpu; +}; + +struct logdev_item { + struct logdev_header hdr; + union { + struct logdev_switch_struct sw; + struct logdev_pkt pkt; + struct logdev_print print; + struct logdev_print_time print_time; + struct logdev_print_time_func print_time_func; + struct logdev_custom custom; + char data[0]; + } u; +}; + +#ifdef __KERNEL__ +#include +#include +#include +#include + +/* + * I also use this with Ingo Molnar's RT kernel. + */ +#ifndef CONFIG_PREEMPT_RT +#undef raw_spinlock_t +#undef RAW_SPIN_LOCK_UNLOCKED +#undef raw_local_irq_save +#undef raw_local_irq_restore +#undef raw_local_irq_enable +#undef raw_local_irq_disable +#undef raw_irqs_disabled +#define raw_spinlock_t spinlock_t +#define RAW_SPIN_LOCK_UNLOCKED SPIN_LOCK_UNLOCKED +#define spin_lock_rt do {} while(0) +#define raw_local_irq_save local_irq_save +#define raw_local_irq_restore local_irq_restore +#define raw_local_irq_enable local_irq_enable +#define raw_local_irq_disable local_irq_disable +#define raw_irqs_disabled irqs_disabled +#endif + +#ifdef CONFIG_LOGDEV_HOOKS +extern int (*logdev_print_hook)(const char *fmt, ...); +extern int (*logdev_vprint_hook)(const char *fmt, va_list va); +extern int (*logdev_print_time_hook)(const char *fmt, ...); +extern int (*logdev_print_time_func_hook)(const char *fmt, ...); +extern void (*logdev_record_switch_hook)(struct task_struct *a,struct task_struct *b); +extern int (*logdev_record_hook)(int id, int size, const void *data); +extern int (*logdev_write_hook)(const char *data, int size); +extern int (*logdev_read_hook)(void *buf, int size); +extern void (*logdev_pkt_hook)(struct sk_buff *skb, int dir); +extern raw_spinlock_t logdev_add_hook_lock; +extern void (*logdev_dump_hook)(void); +extern void (*logdev_dumpnet_hook)(void); + +extern atomic_t logdev_level; +extern atomic_t logdev_switch; +extern atomic_t logdev_print_enabled; + +#define logdev_switch_on() atomic_inc(&logdev_switch) +#define logdev_switch_off() \ + do { \ + if (atomic_read(&logdev_switch)) \ + atomic_dec(&logdev_switch); \ + } while(0); +void logdev_write_msg(const char *msg, unsigned int len, int net); + +typedef void (*logdev_callback_func)(struct logdev_header *hdr, + struct logdev_custom *custom, + int net, + void *rec); + +/* + * Only exists when compiled in. + */ +#ifdef CONFIG_LOGDEV +void logdev_gettimeofday(struct timeval *tv); +#endif + +#ifndef CONFIG_LOGDEV +/* + * dumpnet is expected to be called on a bug, so we don't + * need to lock it. + */ +#define LOGDUMP() do { if (logdev_dump_hook) logdev_dump_hook(); } while(0) +#define LOGDUMPNET() do { if (logdev_dumpnet_hook) logdev_dumpnet_hook(); } while(0) + +#define LOGDEV(x,y...) do { \ + spin_lock(&logdev_add_hook_lock); \ + if (logdev_##x##_hook) \ + logdev_##x##_hook(y); \ + spin_unlock(&logdev_add_hook_lock); \ +} while (0) +#else + +void logdev_record_switch(struct task_struct *prev, struct task_struct *next); +void logdev_pkt(struct sk_buff *skb, int direction); +int logdev_print(const char *str, ...) + __attribute__ ((format (printf, 1, 2))); +int logdev_vprint(const char *str, va_list va); +int logdev_print_time(const char *str, ...) + __attribute__ ((format (printf, 1, 2))); +int logdev_print_time_func(const char *file, int line, const char *str, ...) + __attribute__ ((format (printf, 3, 4))); +int logdev_record(int id, int size, const void *data); +int logdev_record_write(const char *data, int size); +int logdev_record_read(void *data, int size); +void logdev_dumpnet(void); +void logdev_dump(void); +int logdev_register_callback(int custom_id, logdev_callback_func func); +int logdev_unregister_callback(int custom_id); + +#define LOGDUMP() logdev_dump() +#define LOGDUMPNET() logdev_dumpnet() +#define LOGDEV(x,y...) logdev_##x(y) + +#endif +#define LOGPRINT(x...) do { if (atomic_read(&logdev_print_enabled)) LOGDEV(print,x); } while(0) +#define LOGTPRINT(x...) do { if (atomic_read(&logdev_print_enabled)) LOGDEV(print_time,x); } while(0) +#define LOGTFPRINT(x...) do { if (atomic_read(&logdev_print_enabled)) LOGDEV(print_time_func,__FUNCTION__,__LINE__,x); } while(0) +#define LOGSWITCH(prev,next) LOGDEV(record_switch,prev,next) +#define LOGPKT(skb,direction) LOGDEV(pkt,skb,direction) +void logdev_time(void); + +#define lprint(x...) do { if (atomic_read(&logdev_switch)) { LOGTPRINT(x); } } while(0) +#define lfprint(x...) do { if (atomic_read(&logdev_switch)) { LOGTFPRINT(x); } } while(0) + +#else /* !LOGDEV_HOOKS */ +#define LOGDUMP() do {} while(0) +#define LOGDEV(x,y...) do {} while(0) +#define LOGSWITCH(prev,next) do {} while(0) +#define LOGPKT(skb,direction) do {} while(0) +#define logdev_switch_on() do {} while(0) +#define logdev_switch_off() do {} while(0) +#define logdev_time() do {} while(0) +#endif /* LOGDEV_HOOKS */ + +#endif + +#endif Index: lib/Kconfig.debug =================================================================== --- lib/Kconfig.debug (.../tags/starting_base_kernel_2.6.12) (revision 300) +++ lib/Kconfig.debug (.../trunk/kernel) (revision 300) @@ -159,3 +159,64 @@ If you don't debug the kernel, you can say N, but we may not be able to solve problems without frame pointers. +config LOGDEV_HOOKS + bool "Enable hooks for logdev device" + default n + help + If you plan on using the logdev device, you need to include this + into your kernel. This will add the hooks necessary for the logdev + device to be loaded. + + +config LOGDEV + tristate "Enable logdev device" + depends on LOGDEV_HOOKS + default m + help + The logdev device stores data into the kernel that can be retrieved + later through a misc device (major 10). The minor number is + dynamic and is posted through /proc/logdev_minor. Utilities + to open and read the device can be found at + http://www.kihontech.com/logdev + + This device allows for tracing lots of information in the kernel + when simply printk is too expensive. When the logdev is initialized, + it allocates a default of 1 meg of memory (in page size units). This + allows for saving data in a ring buffer without the need to allocate. + + +config LOGDEV_PAGES + int "Number of pages to allocate for logdev device" + depends on LOGDEV + default 256 + help + The Logdev device allocates a number of pages for the sole + purpose of logging data. This is the number of pages that + the Logdev device should allocate upon loading / initializing. + +config LOGDEV_MULTI_CPUS + bool "Logdev Multiple CPU buffers" + depends on LOGDEV && SMP + help + The Logdev device records to a memory buffer allocate once + on startup. With this option set, a separate buffer will be + used for each online CPU. This allows for no blocking while + needing to write to the buffer, but it adds more space since + the amount of pages allocated with the above LOGDEV_PAGES will + be allocated for each CPU. Also, if you are debugging race + conditions, then you don't want this option, since one buffer + is better to see the interleaving actions between the CPUS. + But if you don't need to see the interleaving of actions + between CPUS, than it is probably good to set this. + + If you don't need interleaving changes between CPU writes, say Y. + +config LOGDEV_PRINT_ENABLED + bool "Default Logdev prints should be enabled on startup" + depends on LOGDEV_HOOKS + default y + help + Enable this if you expect the LOGPRINT macros to be enabled + as soon as the logdev device is loaded. Otherwise you must + enable it with /proc/logdev_print + Index: arch/i386/kernel/traps.c =================================================================== --- arch/i386/kernel/traps.c (.../tags/starting_base_kernel_2.6.12) (revision 300) +++ arch/i386/kernel/traps.c (.../trunk/kernel) (revision 300) @@ -27,6 +27,7 @@ #include #include #include +#include #ifdef CONFIG_EISA #include @@ -565,6 +566,7 @@ printk(" on CPU%d, eip %08lx, registers:\n", smp_processor_id(), regs->eip); show_registers(regs); + LOGDUMP(); printk("console shuts up ...\n"); console_silent(); spin_unlock(&nmi_print_lock); Index: Makefile =================================================================== --- Makefile (.../tags/starting_base_kernel_2.6.12) (revision 300) +++ Makefile (.../trunk/kernel) (revision 300) @@ -1,7 +1,7 @@ VERSION = 2 PATCHLEVEL = 6 SUBLEVEL = 12 -EXTRAVERSION = +EXTRAVERSION = -logdev NAME=Woozy Numbat # *DOCUMENTATION* Index: drivers/char/Makefile =================================================================== --- drivers/char/Makefile (.../tags/starting_base_kernel_2.6.12) (revision 300) +++ drivers/char/Makefile (.../trunk/kernel) (revision 300) @@ -89,6 +89,9 @@ obj-$(CONFIG_PCMCIA) += pcmcia/ obj-$(CONFIG_IPMI_HANDLER) += ipmi/ +obj-$(CONFIG_LOGDEV_HOOKS) += logdev_hooks.o +obj-$(CONFIG_LOGDEV) += logdev.o + obj-$(CONFIG_HANGCHECK_TIMER) += hangcheck-timer.o obj-$(CONFIG_TCG_TPM) += tpm/ # Files generated that shall be removed upon make clean Index: drivers/char/logdev.c =================================================================== --- drivers/char/logdev.c (.../tags/starting_base_kernel_2.6.12) (revision 0) +++ drivers/char/logdev.c (.../trunk/kernel) (revision 300) @@ -0,0 +1,2006 @@ +/* + * logdev.c + * + * Copyright (C) 2004 Steven Rostedt + * + * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; version 2 of the License (not later!) + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#include + +#include + +#include +#include + +static char *logdev_version = "0.3.0"; + +static int pages = CONFIG_LOGDEV_PAGES; + +module_param(pages, int, 0644); +MODULE_PARM_DESC(pages, " number of pages to allocate for the logdev ring buffer"); +MODULE_LICENSE("GPL"); + +#define LOGDEV_PROC_DIR "logdev" +#define LOGDEV_PROC_SW "switch" +#define LOGDEV_PROC_LEVEL "level" +#define LOGDEV_PROC_PRINT "print" +#define LOGDEV_PROC_MINOR "minor" +#define LOGDEV_PROC_ENTRY "entry" + +struct proc_dir_entry *logdev_proc_dir; + +/* + * The testing markers are used to cut and paste into a userland program for + * easy algorithm debugging. + */ + +/********************* Start testing code here ***********************/ + +static LIST_HEAD(logdev_callbacks); +static raw_spinlock_t logdev_callbacks_lock = RAW_SPIN_LOCK_UNLOCKED; + +struct logdev_callback { + struct list_head list; + int id; + logdev_callback_func func; +}; + +struct logdev_entry { + unsigned int head; + unsigned int tail; + char *dat; +}; + +struct logdev_dev { + struct logdev_entry *entry; + int init; + int size; + int len; + int start; + int end; + int corrupted; + atomic_t minor; + struct proc_dir_entry *dir; + wait_queue_head_t wait; + raw_spinlock_t lock; +}; + +#ifdef CONFIG_LOGDEV_MULTI_CPUS +# define LOGDEV_CPUS NR_CPUS +# define get_idx(cpu) cpu +# define get_logdev(cpu) &logdev_dev[cpu] +# define kern_spin_lock() do {} while(0) +# define kern_spin_unlock() do {} while(0) +# define kern_spin_lock_save(flags) raw_local_irq_save(flags) +# define kern_spin_unlock_restore(flags) raw_local_irq_restore(flags) +#else + +# define LOGDEV_CPUS 1 +# define get_idx(cpu) ({ (void)cpu; 0; }) +# define get_logdev(cpu) ({ (void)cpu; &logdev_dev[0]; }) +/* + * With a single buffer, we need to have locks. + */ +# define kern_spin_lock() spin_lock(&kern_buffer_lock) +# define kern_spin_unlock() spin_unlock(&kern_buffer_lock) +# define kern_spin_lock_save(flags) spin_lock_irqsave(&kern_buffer_lock,flags) +# define kern_spin_unlock_restore(flags) spin_unlock_irqrestore(&kern_buffer_lock,flags) + static raw_spinlock_t kern_buffer_lock = RAW_SPIN_LOCK_UNLOCKED; +#endif + +static DECLARE_MUTEX(user_sem); +/* + * doesn't really need to be atomic, but helps that we don't need to + * write another proc interface function. + */ +static struct proc_dir_entry *logdev_proc_entry[LOGDEV_CPUS]; +static struct proc_dir_entry *logdev_proc_minor[LOGDEV_CPUS]; +static struct proc_dir_entry *logdev_proc_sw; +static struct proc_dir_entry *logdev_proc_print; +static struct proc_dir_entry *logdev_proc_level; + +static struct logdev_dev logdev_dev[LOGDEV_CPUS]; + +static int logdev_copy_to_dev(struct logdev_dev *dev, const void *dat, + int size); + +void logdev_switch_turnoff(void) +{ + atomic_set(&logdev_switch,0); +} + +/* + * logdev_record_switch is used to track context switches. + * + * If the logdev_switch is not set, then this doesn't record. Thus allowing + * you to just record the context switches that are needed to record. + * Just use logdev_switch_on and logdev_switch_off to turn on this function. + * + * This is called from schedule, and interrupts should already be turned off. + */ +void logdev_record_switch(struct task_struct *prev, struct task_struct *next) +{ + struct logdev_dev *dev; + struct logdev_header hdr; + struct logdev_switch_struct rs; + unsigned long long t; + int cpu = smp_processor_id(); + + if (!atomic_read(&logdev_switch)) + return; + + dev = get_logdev(cpu); + if (!dev->init) + return; + + WARN_ON(!irqs_disabled()); + + hdr.id = LOGDEV_SWITCH_ID; + hdr.cpu = smp_processor_id(); + + t = sched_clock(); + rs.high = (unsigned long)(t>>BITS_PER_LONG); + rs.low = (unsigned long)(t); + + rs.pid_prev = prev->pid; + rs.pid_next = next->pid; + rs.prev_len = strlen(prev->comm); + rs.next_len = strlen(next->comm); + + hdr.size = sizeof(hdr) + sizeof(rs) + rs.prev_len + rs.next_len; + + WARN_ON(!irqs_disabled()); + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev,&hdr,sizeof(hdr)); + logdev_copy_to_dev(dev,&rs,sizeof(rs)); + logdev_copy_to_dev(dev,prev->comm,rs.prev_len); + logdev_copy_to_dev(dev,next->comm,rs.next_len); + spin_unlock(&dev->lock); + +} + +#ifdef __KERNEL__ /* we debug in userland but not this function */ +#include +#include +/* + * This routine may be placed in the network code if you want to see + * what packets are traveling through. You may use the direction to + * indicate if they are coming or going. + * direction = 0 : saddr ==> daddr + * 1 : saddr <== daddr + */ +void logdev_pkt(struct sk_buff *skb, int direction) +{ + struct logdev_dev *dev; + struct logdev_header hdr; + struct logdev_pkt rs; + struct tcphdr *th; + struct iphdr *iph; + unsigned long long t; + unsigned long flags; + int cpu; + + t = sched_clock(); + + raw_local_irq_save(flags); + + cpu = smp_processor_id(); + + dev = get_logdev(cpu); + if (!dev->init) + goto out; + + hdr.id = LOGDEV_PKT_ID; + hdr.size = sizeof(hdr) + sizeof(rs); + hdr.cpu = smp_processor_id(); + + rs.high = (unsigned long)(t>>BITS_PER_LONG); + rs.low = (unsigned long)(t); + + iph = skb->nh.iph; + + rs.protocol = skb->protocol; + rs.dir = direction; + + if ((skb->protocol != __constant_htons(ETH_P_IP)) || + (skb->pkt_type != PACKET_HOST) || + (iph->protocol != IPPROTO_TCP)) { + /* Just copy the first 20 bytes of the packet */ + + hdr.size += 20; + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev,&hdr,sizeof(hdr)); + logdev_copy_to_dev(dev,&rs,sizeof(rs)); + logdev_copy_to_dev(dev,iph,20); + spin_unlock(&dev->lock); + + } else { + int iplen; + int tcplen; + + iplen = iph->ihl<<2; + th = (struct tcphdr*)((void*)(skb->nh.iph)+(iplen)); + tcplen = th->doff<<2; + + hdr.size += iplen + tcplen; + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev,&hdr,sizeof(hdr)); + logdev_copy_to_dev(dev,&rs,sizeof(rs)); + logdev_copy_to_dev(dev,iph,iplen+tcplen); + spin_unlock(&dev->lock); + } + + out: + raw_local_irq_restore(flags); +} +#endif /* endif __KERNEL__ */ + + +#define ENTRY_SIZE(e) ((((e)->tail - (e)->head)) & (PAGE_SIZE-1)) +#define ENTRY_FREE(e) ((PAGE_SIZE-1) - ENTRY_SIZE(e)) +#define ENTRY_ADD(e,x) ((e) = ((e) + x) & (PAGE_SIZE-1)) +#define ENTRY_INC(e) ENTRY_ADD(e,1) +#define ENTRY_MAX (PAGE_SIZE-1) + +/* + * User buffer is used to get data from userland, and this can sleep when + * copying. + */ +static char user_buffer[PAGE_SIZE]; + +/* + * Kernel buffer is used inside the kernel, and MUST NOT SLEEP. So + * we use a raw spinlock to synchronize things. This lock must be taken first + * before the logdev lock can be taken (unless of course you don't need to + * take this lock at all). + */ +static char kern_buffer[LOGDEV_CPUS][PAGE_SIZE]; + +/* + * logdev_print acts like printk but it writes to the logdev device instead + * of a console. + */ +int logdev_vprint(const char *str, va_list va) +{ + struct logdev_dev *dev; + int len=0; + struct logdev_header hdr; + struct logdev_print rs; + unsigned long flags; + int cpu; + int idx; + + kern_spin_lock_save(flags); + + cpu = smp_processor_id(); + idx = get_idx(cpu); + + dev = get_logdev(cpu); + if (!dev->init) + goto out; + + len = vsnprintf(kern_buffer[idx],PAGE_SIZE,str,va); + + if (len > PAGE_SIZE) { + kern_buffer[idx][PAGE_SIZE-1] = 0; + len = PAGE_SIZE; + } + + hdr.id = LOGDEV_PRINT; + hdr.size = sizeof(hdr) + sizeof(rs) + len; + + hdr.cpu = smp_processor_id(); + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev, &hdr, sizeof(hdr)); + if (sizeof(rs)) + logdev_copy_to_dev(dev, &rs, sizeof(rs)); + logdev_copy_to_dev(dev, kern_buffer[idx], len); + spin_unlock(&dev->lock); + + out: + kern_spin_unlock_restore(flags); + + return len; +} + +int logdev_print(const char *str, ...) +{ + va_list va; + struct logdev_dev *dev; + int len=0; + struct logdev_header hdr; + struct logdev_print rs; + unsigned long flags; + int cpu; + int idx; + + kern_spin_lock_save(flags); + + cpu = smp_processor_id(); + idx = get_idx(cpu); + + dev = get_logdev(cpu); + if (!dev->init) + goto out; + + va_start(va,str); + len = vsnprintf(kern_buffer[idx],PAGE_SIZE,str,va); + va_end(va); + + if (len > PAGE_SIZE) { + kern_buffer[idx][PAGE_SIZE-1] = 0; + len = PAGE_SIZE; + } + + hdr.id = LOGDEV_PRINT; + hdr.size = sizeof(hdr) + sizeof(rs) + len; + + hdr.cpu = smp_processor_id(); + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev, &hdr, sizeof(hdr)); + if (sizeof(rs)) + logdev_copy_to_dev(dev, &rs, sizeof(rs)); + logdev_copy_to_dev(dev, kern_buffer[idx], len); + spin_unlock(&dev->lock); + + out: + kern_spin_unlock_restore(flags); + + return len; +} + +/* + * logdev_print_time is the same as logdev_print but it attaches a timestamp to it. + * saves on doing it yourself. + */ +int logdev_print_time(const char *str, ...) +{ + struct logdev_dev *dev; + va_list va; + int len = 0; + struct logdev_header hdr; + struct logdev_print_time rs; + unsigned long long t; + unsigned long flags; + int cpu; + int idx; + + t = sched_clock(); + + hdr.id = LOGDEV_PRINT_TIME; + + kern_spin_lock_save(flags); + + cpu = smp_processor_id(); + idx = get_idx(cpu); + + dev = get_logdev(cpu); + if (!dev->init) + goto out; + + rs.high = (unsigned long)(t>>BITS_PER_LONG); + rs.low = (unsigned long)t; + + va_start(va,str); + len = vsnprintf(kern_buffer[idx],PAGE_SIZE,str,va); + va_end(va); + + if (len > PAGE_SIZE) { + kern_buffer[idx][PAGE_SIZE-1] = 0; + len = PAGE_SIZE; + } + + hdr.size = sizeof(hdr) + sizeof(rs) + len; + + hdr.cpu = cpu; + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev, &hdr, sizeof(hdr)); + logdev_copy_to_dev(dev, &rs, sizeof(rs)); + logdev_copy_to_dev(dev, kern_buffer[idx], len); + spin_unlock(&dev->lock); + + out: + kern_spin_unlock_restore(flags); + + return len; +} + +/* + * logdev_print_time_func quickly stores the time, function and line number. + * this is really only good for live runs since the function is just a pointer, + * so a user land process would need to have the System.map available. + */ +int logdev_print_time_func(const char *file, int line, const char *str, ...) +{ + struct logdev_dev *dev; + va_list va; + int len = 0; + struct logdev_header hdr; + struct logdev_print_time_func rs; + unsigned long long t; + unsigned long flags; + int cpu; + int idx; + + t = sched_clock(); + + hdr.id = LOGDEV_PRINT_TIME_FUNC; + + kern_spin_lock_save(flags); + + cpu = smp_processor_id(); + idx = get_idx(cpu); + + dev = get_logdev(cpu); + if (!dev->init) + goto out; + + rs.high = (unsigned long)(t>>BITS_PER_LONG); + rs.low = (unsigned long)t; + + rs.file = file; + rs.line = line; + + va_start(va,str); + len = vsnprintf(kern_buffer[idx],PAGE_SIZE,str,va); + va_end(va); + + if (len > PAGE_SIZE) { + kern_buffer[idx][PAGE_SIZE-1] = 0; + len = PAGE_SIZE; + } + + hdr.size = sizeof(hdr) + sizeof(rs) + len; + + hdr.cpu = cpu; + + spin_lock(&dev->lock); + logdev_copy_to_dev(dev, &hdr, sizeof(hdr)); + logdev_copy_to_dev(dev, &rs, sizeof(rs)); + logdev_copy_to_dev(dev, kern_buffer[idx], len); + spin_unlock(&dev->lock); + + out: + kern_spin_unlock_restore(flags); + + return len; +} + +int logdev_copy_from_dev(struct logdev_dev *dev, void *buf, int size) +{ + struct logdev_entry *entry = &dev->entry[dev->start]; + int ret = 0; + + if (size < 0) { + printk("logdev_copy_from_dev: size < 0 ???\n"); + return -1; + } + if (size > dev->size) + size = dev->size; + + while (size && dev->size) { + int copy = size; + int used; + + if (!ENTRY_SIZE(entry)) { + dev->start = (dev->start + 1) % dev->len; + entry = &dev->entry[dev->start]; + } + + if (copy > (used=ENTRY_SIZE(entry))) + copy = used; + if (entry->head+copy > PAGE_SIZE) + copy = PAGE_SIZE - entry->head; + memcpy(buf,entry->dat+entry->head,copy); + ENTRY_ADD(entry->head,copy); + buf += copy; + dev->size -= copy; + size -= copy; + ret += copy; + + } + + return ret; +} + +static void move_start_to_next_entry(struct logdev_dev *dev) +{ + struct logdev_entry *entry = &dev->entry[dev->start]; + int start = dev->start; + int size; + struct header { + int id; + int size; + } hdr; + + if (dev->corrupted) { + /* Don't trust headers, just skip to the next entry */ + dev->size -= ENTRY_SIZE(entry); + entry->head = entry->tail = 0; + dev->start = (dev->start+1) % dev->len; + return; + } + + while (start == dev->start) { + logdev_copy_from_dev(dev,(char*)&hdr,sizeof(hdr)); + entry = &dev->entry[dev->start]; + switch (hdr.id) { + case LOGDEV_CUSTOM: + case LOGDEV_SWITCH_ID: + case LOGDEV_PKT_ID: + case LOGDEV_PRINT: + case LOGDEV_PRINT_TIME: + case LOGDEV_PRINT_TIME_FUNC: + break; + default: + dev->corrupted = 1; + } + if (dev->corrupted) { + if (start == dev->start) + move_start_to_next_entry(dev); + return; + } + + size = sizeof(hdr); + while (size < hdr.size) { + int count = hdr.size - size; + if (count > ENTRY_SIZE(entry)) + count = ENTRY_SIZE(entry); + dev->size -= count; + size += count; + ENTRY_ADD(entry->head,count); + if (ENTRY_SIZE(entry) == 0) { + dev->start = (dev->start + 1) % dev->len; + entry->head = entry->tail = 0; + entry = &dev->entry[dev->start]; + } + } + } +} + +static int logdev_copy_to_dev(struct logdev_dev *dev, const void *dat, + int size) +{ + struct logdev_entry *entry = &dev->entry[dev->end]; + const char *buf = dat; + int ret = 0; + + while (size) { + int copy; + int free; + if (ENTRY_SIZE(entry) == ENTRY_MAX) { + dev->end = (dev->end+1) % dev->len; + entry = &dev->entry[dev->end]; + /* if we wrapped, then clear out this entire + * buffer. + */ + if (dev->end == dev->start) { + move_start_to_next_entry(dev); + } + } + copy = size; + if (copy > (free=ENTRY_FREE(entry))) + copy = free; + if (entry->tail+copy > PAGE_SIZE) + copy = PAGE_SIZE - entry->tail; + memcpy(entry->dat+entry->tail,buf,copy); + ENTRY_ADD(entry->tail,copy); + buf += copy; + size -= copy; + ret += copy; + dev->size += copy; + + } +#if 0 + /* wake up those waiting for data */ + if (waitqueue_active(&dev->wait)) + wake_up_interruptible(&dev->wait); +#endif + + return ret; + +} + + +/* + * If you feel like recording your own data, you can use logdev_record. + * just pass your own id, size and data. The size is the size of + * the data being passed and not the size actually being written to the device. + * That is already calculated. + * + * This record will be added as LOGDEV_CUSTOM and the given id will be the custom id. + */ +int logdev_record(int id, int size, const void *data) +{ + struct logdev_dev *dev; + struct logdev_header hdr; + struct logdev_custom rs; + unsigned long flags; + int cpu; + int ret = 0; + + hdr.id = LOGDEV_CUSTOM; + hdr.size = sizeof(hdr) + sizeof(rs) + size; + + rs.id = id; + + raw_local_irq_save(flags); + + cpu = smp_processor_id(); + + dev = get_logdev(cpu); + if (!dev->init) { + local_irq_restore(flags); + return 0; + } + spin_lock(&dev->lock); + hdr.cpu = smp_processor_id(); + logdev_copy_to_dev(dev, &hdr, sizeof(hdr)); + logdev_copy_to_dev(dev, &rs, sizeof(rs)); + ret = logdev_copy_to_dev(dev, data, size); + spin_unlock_irqrestore(&dev->lock,flags); + + return ret; +} + +/* + * If you just want to write into the buffer using your own methods, then this + * is perfectly fine. Just pass in your data and the size of the data being + * passed in. You can read it out later with logdev_record_read. But you wont + * have the benefits of keeping integrity when the buffer overflows. + */ +int logdev_record_write(const char *data, int size) +{ + struct logdev_dev *dev; + int ret; + unsigned long flags; + int cpu; + + raw_local_irq_save(flags); + + cpu = smp_processor_id(); + dev = get_logdev(cpu); + if (!dev->init) { + raw_local_irq_restore(flags); + return 0; + } + + spin_lock(&dev->lock); + ret = logdev_copy_to_dev(dev, data, size); + spin_unlock_irqrestore(&dev->lock,flags); + + return ret; +} + +/* + * logdev_record_read reads some data from the logdev device no matter what + * it was. + */ +int logdev_record_read(void *data, int size) +{ + struct logdev_dev *dev; + int ret; + unsigned long flags; + int cpu; + + raw_local_irq_save(flags); + + cpu = smp_processor_id(); + dev = get_logdev(cpu); + if (!dev->init) { + raw_local_irq_restore(flags); + return 0; + } + + spin_lock(&dev->lock); + ret = logdev_copy_from_dev(dev, data, size); + spin_unlock_irqrestore(&dev->lock,flags); + + return ret; +} + +/************************ Stop testing code here ******************************/ + +ssize_t logdev_write(struct file *filp, const char *buf, size_t count, loff_t *f_pos) +{ + int cpu = (int)filp->private_data; + struct logdev_dev *dev; + unsigned long flags; + + if (count > PAGE_SIZE) + count = PAGE_SIZE; + + if (down_interruptible(&user_sem)) + return -EINTR; + + dev = get_logdev(cpu); + if (!dev->init) + goto out; + + if (copy_from_user(user_buffer,buf,count)) { + count = -EFAULT; + goto out; + } + + spin_lock_irqsave(&dev->lock,flags); + count = logdev_copy_to_dev(dev,user_buffer,count); + spin_unlock_irqrestore(&dev->lock,flags); + + /* wake up those waiting for data */ + if (waitqueue_active(&dev->wait)) + wake_up_interruptible(&dev->wait); + +out: + up(&user_sem); + return count; +} + +ssize_t logdev_read(struct file *filp, char *buf, size_t count, loff_t *f_pos) +{ + int cpu = (int)filp->private_data; + struct logdev_dev *dev; + unsigned long flags; + + if (down_interruptible(&user_sem)) + return -EINTR; + + + dev = get_logdev(cpu); + + if (!dev->init) + goto out_up; + + spin_lock_irqsave(&dev->lock,flags); + + if (!dev->size) { + + /* TBD - FIXME */ +#if 1 + + count = 0; + goto out; +#endif + + if (filp->f_flags & O_NONBLOCK) { + count = -EAGAIN; + goto out; + } + + do { + DECLARE_WAITQUEUE(wait,current); + current->state = TASK_INTERRUPTIBLE; + add_wait_queue(&dev->wait,&wait); + spin_unlock_irqrestore(&dev->lock,flags); + schedule(); + spin_lock_irqsave(&dev->lock,flags); + remove_wait_queue(&dev->wait,&wait); + if (dev->size) + break; + if (signal_pending(current)) { + count = -ERESTARTSYS; + goto out; + } + } while(1); + } + + if (count > PAGE_SIZE) + count = PAGE_SIZE; + + + count = logdev_copy_from_dev(dev,user_buffer,count); + + out: + /* We can't be corrupted if we have no data */ + if (!dev->size) + dev->corrupted = 0; + spin_unlock_irqrestore(&dev->lock,flags); + + if (count > 0) { + /* Well if we fail here, we just lost the data read :-( */ + if (copy_to_user(buf,user_buffer,count)) + count = -EFAULT; + } + out_up: + up(&user_sem); + + return count; +} + +static int logdev_ioctl(struct inode *inode, struct file *filp, + unsigned int cmd, unsigned long arg) +{ + return -ENOTTY; +} + +static int logdev_close(struct inode *inode, struct file *filp) +{ +#if 0 + int cpu = (int)filp->private_data; +#endif + return 0; +} + + +static int logdev_open (struct inode *inode, struct file *filp) +{ + int minor = iminor(inode); + int i; + + for (i=0; i < LOGDEV_CPUS; i++) { + if (logdev_dev[i].init && atomic_read(&logdev_dev[i].minor) == minor) + break; + } + if (i == LOGDEV_CPUS) + return -ENODEV; + + filp->private_data = (void*)i; + + return 0; +} + +/** + * logdev_poll - poll file op for logdev + * @filp: the file + * @wait: poll table + * + * Poll implemention. + */ +static unsigned int +logdev_poll(struct file *filp, poll_table *wait) +{ + struct logdev_dev *dev = (struct logdev_dev*)filp->private_data; + unsigned int mask = 0; + + (void)dev; +#if 0 + if (filp->f_mode & FMODE_READ) { + poll_wait(filp, &app->read_wait, wait); + if (!empty_channel(dev)) + mask |= POLLIN | POLLRDNORM; + } +#endif + + return mask; +} + +static struct file_operations logdev_fops = { + .read = logdev_read, + .write = logdev_write, + .ioctl = logdev_ioctl, + .open = logdev_open, + .release = logdev_close, + .poll = logdev_poll, + .llseek = no_llseek, +}; + +/* + * The following is to register call back functions to print out + * a custom record. + */ +int logdev_register_callback(int custom_id, logdev_callback_func func) +{ + struct list_head *p; + struct logdev_callback *cb; + unsigned long flags; + int ret = 0; + + cb = kmalloc(sizeof(*cb),GFP_KERNEL); + if (!cb) { + ret = -ENOMEM; + goto out; + } + + spin_lock_irqsave(&logdev_callbacks_lock,flags); + list_for_each(p,&logdev_callbacks) { + struct logdev_callback *c = list_entry(p, struct logdev_callback, list); + if (c->id == custom_id) { + spin_unlock_irqrestore(&logdev_callbacks_lock,flags); + kfree(cb); + ret = -EBUSY; + goto out; + } + } + + cb->id = custom_id; + cb->func = func; + list_add(&cb->list,&logdev_callbacks); + spin_unlock_irqrestore(&logdev_callbacks_lock,flags); + + out: + return ret; +} + +int logdev_unregister_callback(int custom_id) +{ + struct list_head *p; + struct logdev_callback *cb; + unsigned long flags; + int ret = 0; + + spin_lock_irqsave(&logdev_callbacks_lock,flags); + list_for_each(p,&logdev_callbacks) { + cb = list_entry(p, struct logdev_callback, list); + if (cb->id == custom_id) + break; + } + if (p == &logdev_callbacks) { + ret = -ENODEV; + spin_unlock_irqrestore(&logdev_callbacks_lock,flags); + goto out; + } + + list_del(&cb->list); + spin_unlock_irqrestore(&logdev_callbacks_lock,flags); + + kfree(cb); + + out: + return ret; +} + +/* + * The following UDP sending is based off of netconsole. + */ +static int configured = 0; + +#ifdef CONFIG_NETPOLL +static char config[256]; +module_param_string(logdevnet, config, 256, 0); +MODULE_PARM_DESC(logdevnet, " logdevnet=[src-port]@[src-ip]/[dev],[tgt-port]@/[tgt-macaddr]\n"); + +static struct netpoll np = { + .name = "logdevnet", + .dev_name = "eth0", + .local_port = 6665, + .remote_port = 6666, + .remote_mac = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}, +}; + +static int option_setup(char *opt) +{ + configured = !netpoll_parse_options(&np, opt); + + return 0; +} + +__setup("logdevnet=", option_setup); +#endif + +#define MAX_PRINT_CHUNK 1000 + +#ifdef CONFIG_NETPOLL +void logdev_write_msg(const char *msg, unsigned int len, int net) +{ + int frag, left; + + if (!net) { + printk("%s",msg); + return; + } + + if (!np.dev) + return; + + for(left = len; left; ) { + frag = min(left, MAX_PRINT_CHUNK); + netpoll_send_udp(&np, msg, frag); + msg += frag; + left -= frag; + } + +} +#else /* !CONFIG_NETPOLL */ +void logdev_write_msg(const char *msg, unsigned int len, int net) +{ + printk("%s",msg); +} +#endif /* CONFIG_NETPOLL */ + +static void __logdev_dump(int net) +{ + int r; + int i; + int count; + int save_switch; + char *msg; + struct logdev_dev *dev; + struct logdev_header hdr; + static int started = 0; + int newline = 1; + int do_lock = 1; + extern int in_logdump; + unsigned long long t; + unsigned long secs; + unsigned long nanosec_rem; + const char *file = NULL; + int line = 0; + int cpu; + + if (started) + return; + + if (oops_in_progress) + do_lock = 0; + in_logdump++; + + /* + * This is for debugging, so we don't want to reintroduce more output. + */ + save_switch = atomic_read(&logdev_switch); + atomic_set(&logdev_switch,0); +#if 0 + atomic_set(&logdev_print_enabled,0); +#endif + + for (cpu = 0; cpu < LOGDEV_CPUS; cpu++) { + unsigned long flags; + int corrupt = 0; + + dev = &logdev_dev[cpu]; + if (!dev->init) + continue; + + /* This is a long time to hold a lock, but we are dumping + * debug so it's ok. This is not a normal code path. */ + spin_lock_irqsave(&dev->lock,flags); + + if (do_lock) + kern_spin_lock(); + + i = sprintf(kern_buffer[cpu],"****** Starting Logdev for CPU %d ********\n",cpu); + logdev_write_msg(kern_buffer[cpu],i,net); + + if (do_lock) + kern_spin_unlock(); + + if (dev->corrupted) { + msg = ">>>>> Warning: Logdev corrupted <<<<<\n"; + i = strlen(msg); + logdev_write_msg(msg,i,net); + corrupt = 1; + } + + while ((r = logdev_copy_from_dev(dev,&hdr,sizeof(hdr))) == sizeof(hdr)) { + switch (hdr.id) { + + case LOGDEV_PRINT_TIME_FUNC: + { + struct logdev_print_time_func rs; + int cap = sizeof(rs) - sizeof(struct logdev_print_time); + +// printk("hdr.size=%d r=%d sizeof cap=%d sizeof rs=%d\n", +// hdr.size,r,cap,sizeof(rs)); + logdev_copy_from_dev(dev,&rs,cap); + file = rs.file; + line = rs.line; + + r += cap; + /* fall through */ + } + case LOGDEV_PRINT_TIME: + { + struct logdev_print_time rs; + + // printk("sizeof(rs)=%d\n",sizeof(rs)); + logdev_copy_from_dev(dev,&rs,sizeof(rs)); + + if (newline) { + if (do_lock) + kern_spin_lock(); + t = ((unsigned long long)rs.high << BITS_PER_LONG) | + rs.low; + nanosec_rem = do_div(t, 1000000000)/1000; + secs = (unsigned long)t; + + i = snprintf(kern_buffer[cpu],100,"[%5lu.%06lu] ", + secs,nanosec_rem); + + logdev_write_msg(kern_buffer[cpu],i,net); + if (do_lock) + kern_spin_unlock(); + } + + r += sizeof(rs); + + /* fall through */ + } + + case LOGDEV_PRINT: + if (do_lock) + kern_spin_lock(); + if (newline) { + i = sprintf(kern_buffer[cpu],"cpu:%d ",hdr.cpu); + logdev_write_msg(kern_buffer[cpu],i,net); + + if (hdr.id == LOGDEV_PRINT_TIME_FUNC) { + i = sprintf(kern_buffer[cpu],"%s:%d ",file, line); + logdev_write_msg(kern_buffer[cpu],i,net); + } + } + for (i=r; i < hdr.size; i += r) { + count = hdr.size - i; + if (count > PAGE_SIZE-1) + count = PAGE_SIZE-1; + r = logdev_copy_from_dev(dev,kern_buffer[cpu],count); + if (r < 0) + break; + kern_buffer[cpu][count] = 0; + logdev_write_msg(kern_buffer[cpu],count,net); + newline = (count) && (kern_buffer[cpu][count - 1] != '\n') ? 0 : 1; + } + if (do_lock) + kern_spin_unlock(); + break; + case LOGDEV_SWITCH_ID: + { + struct logdev_switch_struct rs; + + if (do_lock) + kern_spin_lock(); + + i = sprintf(kern_buffer[cpu],">>>> IN LOGDEV SWITCH <<<< cpu:%d \n",hdr.cpu); + logdev_write_msg(kern_buffer[cpu],i,net); + + logdev_copy_from_dev(dev,&rs,sizeof(rs)); + + t = ((unsigned long long)rs.high << BITS_PER_LONG) | + (unsigned long long)rs.low; + nanosec_rem = do_div(t, 1000000000)/1000; + secs = (unsigned long)t; + + i = snprintf(kern_buffer[cpu],100,"CPU=%d [%5lu.%06lu] ", + hdr.cpu, + secs,nanosec_rem); + logdev_copy_from_dev(dev,&kern_buffer[cpu][i],rs.prev_len); + i += rs.prev_len; + + i += snprintf(&kern_buffer[cpu][i],100,":%d -->> ",rs.pid_prev); + logdev_copy_from_dev(dev,&kern_buffer[cpu][i],rs.next_len); + i += rs.next_len; + i += snprintf(&kern_buffer[cpu][i],100,":%d\n",rs.pid_next); + + logdev_write_msg(kern_buffer[cpu],i,net); + + if (do_lock) + kern_spin_unlock(); + + break; + } + + case LOGDEV_PKT_ID: + { + struct logdev_pkt rs; + struct tcphdr th; + struct iphdr iph; + int len = sizeof(hdr); + int iplen; + unsigned long end; + int tcplen; + char buf[16]; + unsigned long oaddr; + unsigned long iaddr; + unsigned short oport; + unsigned short iport; + int dir; + + /* + * Here we have "len" keeping track of how much data + * we taken out of the device. So if there's data + * left in the device for this record, after we have + * all the printable data, we know how much still needs + * to be retreived. + * + * We also use "i" to keep track of where on the kern_buffer + * we last wrote data to. + */ + + logdev_copy_from_dev(dev,&rs,sizeof(rs)); + len += sizeof(rs); + + if (do_lock) + kern_spin_lock(); + + t = ((unsigned long long)rs.high << BITS_PER_LONG) | + (unsigned long long)rs.low; + nanosec_rem = do_div(t, 1000000000)/1000; + secs = (unsigned long)t; + + i = snprintf(kern_buffer[cpu],100,"[%5lu.%06lu] cpu:%d", + secs,nanosec_rem,hdr.cpu); + + if (rs.protocol != htons(ETH_P_IP)) { + sprintf(&kern_buffer[cpu][i],"Not IP packet\n"); + goto pkt_done; + } + + + if ((hdr.size - len) < sizeof(iph)) { + i += sprintf(&kern_buffer[cpu][i]," IP Header corrupted\n"); + goto pkt_done; + } + + logdev_copy_from_dev(dev,&iph,sizeof(iph)); + len += sizeof(iph); + + iph.tot_len = ntohs(iph.tot_len); + iph.id = ntohs(iph.id); + iph.frag_off = ntohs(iph.frag_off); + iph.check = ntohs(iph.check); + iph.saddr = ntohl(iph.saddr); + iph.daddr = ntohl(iph.daddr); + + /* + * dir is the direction of the packet. + * 0 is incoming, 1 is outgoing. + */ + dir = rs.dir; + + if (dir) { + /* outgoing */ + oaddr = iph.daddr; + iaddr = iph.saddr; + } else { + /* incomming */ + oaddr = iph.saddr; + iaddr = iph.daddr; + } + + +#define IPBYTE(ip,b) (unsigned char)(((ip)>>((b)*8))&0xff) + + if (iph.protocol != 6) { + i += sprintf(&kern_buffer[cpu][i],"%d.%d.%d.%d %s " + "%d.%d.%d.%d (Not TCP packet)\n", + IPBYTE(iaddr,3), + IPBYTE(iaddr,2), + IPBYTE(iaddr,1), + IPBYTE(iaddr,0), + dir ? "==>" : "<==", + IPBYTE(oaddr,3), + IPBYTE(oaddr,2), + IPBYTE(oaddr,1), + IPBYTE(oaddr,0)); + + goto pkt_done; + } + + iplen = iph.ihl<<2; + + /* + * Skip any ip options. + */ + while (iplen > sizeof(iph)) { + int count = iplen - sizeof(iph); + if (count > 16) + count = 16; + logdev_copy_from_dev(dev,buf,count); + len += count; + iplen -= count; + } + + if ((hdr.size - len) < sizeof(th)) { + i += sprintf(&kern_buffer[cpu][i]," IP Header corrupted\n"); + goto pkt_done; + } + + logdev_copy_from_dev(dev,&th,sizeof(th)); + len += sizeof(th); + + tcplen = th.doff<<2; + + th.source = ntohs(th.source); + th.dest = ntohs(th.dest); + th.seq = ntohl(th.seq); + th.ack_seq = ntohl(th.ack_seq); + th.window = ntohs(th.window); + th.check = ntohs(th.check); + th.urg_ptr = ntohs(th.urg_ptr); + + iplen = iph.tot_len - (iplen + tcplen); + end = th.seq; + if (th.fin || th.syn) { + end++; + } + end += iplen; + + if (dir) { + /* outgoing */ + oport = th.dest; + iport = th.source; + } else { + /* incomming */ + oport = th.source; + iport = th.dest; + } + + i += sprintf(&kern_buffer[cpu][i], + "%d.%d.%d.%d:%d %s %d.%d.%d.%d:%d seq:%u ack:%u " + "(%c%c%c%c%c%c) len:%d win:%d end_seq:%lu\n", + IPBYTE(iaddr,3), + IPBYTE(iaddr,2), + IPBYTE(iaddr,1), + IPBYTE(iaddr,0), + iport, + dir ? "==>" : "<==", + IPBYTE(oaddr,3), + IPBYTE(oaddr,2), + IPBYTE(oaddr,1), + IPBYTE(oaddr,0), + oport, + th.seq, + th.ack_seq, + th.fin ? 'F' : '-', + th.syn ? 'S' : '-', + th.rst ? 'R' : '-', + th.psh ? 'P' : '-', + th.ack ? 'A' : '-', + th.urg ? 'U' : '-', + iplen, + th.window, + end + ); + + pkt_done: + + logdev_write_msg(kern_buffer[cpu],i,net); + + /* + * Clean up anything left over. + */ + while (len < hdr.size) { + int count = hdr.size - len; + logdev_copy_from_dev(dev,buf,count); + len += count; + } + + if (do_lock) + kern_spin_unlock(); + + break; + } + case LOGDEV_CUSTOM: + { + struct list_head *p; + struct logdev_custom custom; + int len; + + logdev_copy_from_dev(dev,&custom,sizeof(custom)); + if (do_lock) { + spin_lock(&logdev_callbacks_lock); + kern_spin_lock(); + } + + i = len = hdr.size - sizeof(hdr) - sizeof(custom); + if (i > PAGE_SIZE) { + i = PAGE_SIZE; + /* show that we truncated */ + hdr.size -= len - i; + } + logdev_copy_from_dev(dev,kern_buffer[cpu],i); + + list_for_each(p, &logdev_callbacks) { + struct logdev_callback *cb = list_entry(p, struct logdev_callback, list); + if (cb->id == custom.id) { + cb->func(&hdr,&custom,net,kern_buffer[cpu]); + break; + } + } + + /* No record should be bigger than a page. Ignore all else */ + while (i < len) { + int count = len - i; + if (count > PAGE_SIZE) + count = PAGE_SIZE; + logdev_copy_from_dev(dev,kern_buffer[cpu],count); + i += count; + } + + /* check if we didn't find a call back */ + if (p == &logdev_callbacks) { + i = sprintf(kern_buffer[cpu],"skipping! LOGDEV_CUSTOM id %d\n",custom.id); + logdev_write_msg(kern_buffer[cpu],i,net); + } + + if (do_lock) { + kern_spin_unlock(); + spin_unlock(&logdev_callbacks_lock); + } + break; + } + default: + if (!corrupt) { + corrupt = 1; + if (do_lock) + kern_spin_lock(); + i = snprintf(kern_buffer[cpu],100,">>>>> Unknown logdev header, may be " + "corrupted from this point on\n"); + logdev_write_msg(kern_buffer[cpu],i,net); + if (do_lock) + kern_spin_unlock(); + } + + break; + } + } + msg = ">>>>> done <<<<<\n"; + i = strlen(msg); + logdev_write_msg(msg,i,net); + + spin_unlock_irqrestore(&dev->lock,flags); + } + atomic_set(&logdev_switch,save_switch); + in_logdump--; + started = 0; +} + +void logdev_dumpnet(void) +{ + if (!configured) + return; + + __logdev_dump(1); +} + +void logdev_dump(void) +{ + __logdev_dump(0); +} + +#ifdef CONFIG_NETPOLL +int __init logdevnet_init(void) +{ + if(strlen(config)) + option_setup(config); + /* + * Don't need to check the return code on this, + * if it fails then we just don't use the network. + */ + if (configured) { + if (netpoll_setup(&np)) { + printk("logdev: warning, netpoll failed to setup\n"); + } + } + return 0; +} +#endif + +#if 1 +static int atoi(const char *p) +{ + int n = 0; + while (*p && '0' <= *p && *p <= '9') { + n *= 10; + n += *p - '0'; + p++; + } + return (n); +} + +static int proc_var_read(char *buffer, char **start, off_t offset, int count, + int *eof, void *data) +{ + int var; + int len; + + var = atomic_read((atomic_t*)data); + + len = sprintf(buffer,"%d\n",var); + + if (offset >= len) { + *start = buffer; + *eof = 1; + return 0; + } + *start = buffer + offset; + if ((len -= offset) > count) + return count; + *eof = 1; + return len; +} + +static int proc_var_write(struct file * file, const char * buffer, + unsigned long count, void *data) +{ + atomic_t *var; + int val; + char buf[10]; + + var = (atomic_t*)data; + + if (count > 9) + count = 9; + + if(copy_from_user(&buf, buffer, count)) + return -EFAULT; + + buf[count] = 0; + + val = atoi(buf); + + atomic_set(var,val); + + file->f_pos += count; + + return count; + +} +#endif + +static void *s_next(struct seq_file *m, void *v, loff_t *pos) +{ + struct logdev_dev *dev = m->private; + int i = (int)*pos; + + (*pos)++; + + if (i >= dev->len) + return NULL; + + return &dev->entry[i]; +} + +static void *s_start(struct seq_file *m, loff_t *pos) + __acquires(logdev_dev.lock) +{ + struct logdev_dev *dev = m->private; + void *p = NULL; + loff_t l = 0; + + /* + * A little strong? Perhaps, but we know that this is bad right + * from the start. Anyway this is for debugging purposes only, + * so it's OK, as well as the big latency we get by turning off + * intrerrupts. But we also never know who will be locking + * this. + */ + if (raw_irqs_disabled()) + BUG(); + + raw_local_irq_disable(); + spin_lock(&dev->lock); + for (p = (void *)1; p && l < *pos; p = s_next(m,p,&l)) + ; + + return p; +} + +static void s_stop(struct seq_file *m, void *p) + __releases(logdev_dev.lock) +{ + struct logdev_dev *dev = m->private; + spin_unlock(&dev->lock); + raw_local_irq_enable(); +} + +static int s_show(struct seq_file *m, void *v) +{ + int i = (int)(v); + struct logdev_entry *entry = v; + struct logdev_dev *dev; + + dev = m->private; + + if (i == 1) { + seq_printf(m,"Logdev:\n"); + seq_printf(m,"\tlen:\t%d\n",dev->len); + seq_printf(m,"\tsize:\t%d\n",dev->size); + seq_printf(m,"\tstart:\t%d\n",dev->start); + seq_printf(m,"\tend:\t%d\n",dev->end); + seq_printf(m,"\tcorrupted:%d\n",dev->corrupted); + seq_printf(m,"\n\tEntries:\n"); + + } else { + i = (int)((char*)entry - (char*)dev->entry) / sizeof(struct logdev_entry); + + seq_printf(m,"\t %d:\t%8u : %8u\tsize: %lu\n", i, + entry->head, entry->tail, + ENTRY_SIZE(entry)); + } + + + return 0; +} + +static struct seq_operations logdev_proc_op = { + .start = s_start, + .next = s_next, + .stop = s_stop, + .show = s_show, +}; + +static int logdev_proc_open (struct inode *inode, struct file *file) +{ + struct proc_dir_entry * de; + int ret; + + de = PDE(inode); + + ret = seq_open(file, &logdev_proc_op); + if (!ret) { + struct seq_file *m = file->private_data; + m->private = de->data; + } + + return ret; +} + + +static struct file_operations logdev_proc_operations = { + .open = logdev_proc_open, + .read = seq_read, + .llseek = seq_lseek, + .release = seq_release, +}; + + +static void logdev_proc_setup(void) +{ + struct proc_dir_entry *entry; + int i; + char buff[30]; + + logdev_proc_dir = proc_mkdir(LOGDEV_PROC_DIR, NULL); + if (!logdev_proc_dir) + return; + + for (i=0; i < LOGDEV_CPUS; i++) { + if (!logdev_dev[i].init) + continue; + + sprintf(buff,"%d",i); + logdev_dev[i].dir = proc_mkdir(buff,logdev_proc_dir); + + entry = create_proc_entry(LOGDEV_PROC_ENTRY, + S_IFREG | S_IRUGO, + logdev_dev[i].dir); + if (entry) { + logdev_proc_entry[i] = entry; + entry->proc_fops = &logdev_proc_operations; + entry->data = &logdev_dev[i]; + } + + entry = create_proc_entry(LOGDEV_PROC_MINOR, + S_IFREG | S_IRUGO, + logdev_dev[i].dir); + if (entry) { + entry->read_proc = proc_var_read; + entry->data = &logdev_dev[i].minor; + logdev_proc_minor[i] = entry; + } + } + + entry = create_proc_entry(LOGDEV_PROC_SW, + S_IFREG | S_IRUGO | S_IWUGO, + logdev_proc_dir); + if (entry) { + entry->read_proc = proc_var_read; + entry->write_proc = proc_var_write; + entry->data = &logdev_switch; + logdev_proc_sw = entry; + } + entry = create_proc_entry(LOGDEV_PROC_LEVEL, + S_IFREG | S_IRUGO | S_IWUGO, + logdev_proc_dir); + if (entry) { + entry->read_proc = proc_var_read; + entry->write_proc = proc_var_write; + entry->data = &logdev_level; + logdev_proc_level = entry; + } + + entry = create_proc_entry(LOGDEV_PROC_PRINT, + S_IFREG | S_IRUGO | S_IWUGO, + logdev_proc_dir); + if (entry) { + entry->read_proc = proc_var_read; + entry->write_proc = proc_var_write; + entry->data = &logdev_print_enabled; + logdev_proc_print = entry; + } +} + +static struct miscdevice logdev_misc_dev[LOGDEV_CPUS] = +{ [ 0 ... (LOGDEV_CPUS-1) ] = + { + .minor = MISC_DYNAMIC_MINOR, + .fops = &logdev_fops + } +}; + +static int logdev_panic_handler(struct notifier_block *this, + unsigned long event, + void *unused) +{ + LOGDUMP(); + return NOTIFY_OK; +} + +static struct notifier_block logdev_panic_notifier = { + .notifier_call = logdev_panic_handler, + .next = NULL, + .priority = 150 /* priority: INT_MAX >= x >= 0 */ +}; + +/* + * Unfortunately, the die handlers have no way to unregister, so + * we don't want to do this if we are a module. We would if we could + * find a way to clean ourselves up when unloaded. + */ +#ifndef MODULE +int logdev_die_handler(struct notifier_block *self, unsigned long val, + void *data) +{ + switch (val) { + case DIE_OOPS: + LOGDUMP(); + break; + default: + break; + } + return NOTIFY_OK; +} + +static struct notifier_block logdev_die_notifier = { + .notifier_call = logdev_die_handler, + .priority = 200 +}; +#endif /* !MODULE */ + +void __exit logdev_cleanup(void) +{ + int i; + int cpu; + char buff[30]; + + notifier_chain_unregister(&panic_notifier_list, &logdev_panic_notifier); + +#ifdef MODULE + { + unsigned long flags; + spin_lock_irqsave(&logdev_add_hook_lock,flags); + logdev_record_switch_hook = NULL; + logdev_pkt_hook = NULL; + logdev_print_hook = NULL; + logdev_vprint_hook = NULL; + logdev_print_time_hook = NULL; + logdev_print_time_func_hook = NULL; + logdev_record_hook = NULL; + logdev_write_hook = NULL; + logdev_read_hook = NULL; + spin_unlock_irqrestore(&logdev_add_hook_lock,flags); + } +#endif + + for (i=0; i < LOGDEV_CPUS; i++) { + if (logdev_dev[i].dir) { + struct proc_dir_entry *parent = logdev_dev[i].dir; + if (logdev_proc_entry[i]) { + sprintf(buff,"%s",LOGDEV_PROC_ENTRY); + remove_proc_entry(buff,parent); + } + if (logdev_proc_minor[i]) { + sprintf(buff,"%s",LOGDEV_PROC_MINOR); + remove_proc_entry(buff,parent); + } + sprintf(buff,"%s/%d",LOGDEV_PROC_DIR,i); + remove_proc_entry(buff,0); + } + } + + if (logdev_proc_sw) { + sprintf(buff,"%s/%s",LOGDEV_PROC_DIR,LOGDEV_PROC_SW); + remove_proc_entry(buff,0); + } + if (logdev_proc_level) { + sprintf(buff,"%s/%s",LOGDEV_PROC_DIR,LOGDEV_PROC_LEVEL); + remove_proc_entry(buff,0); + } + if (logdev_proc_print) { + sprintf(buff,"%s/%s",LOGDEV_PROC_DIR,LOGDEV_PROC_PRINT); + remove_proc_entry(buff,0); + } + if (logdev_proc_dir) + remove_proc_entry(LOGDEV_PROC_DIR,0); + + for_each_present_cpu(cpu) { + + if (cpu >= LOGDEV_CPUS) + break; + + if (!logdev_dev[cpu].init) + continue; + + if (logdev_misc_dev[cpu].name) + kfree(logdev_misc_dev[cpu].name); + + if (logdev_dev[cpu].entry) { + struct logdev_entry *entry; + for (i=0, entry=logdev_dev[cpu].entry; idat) { + free_page((unsigned long)entry->dat); + entry->dat = NULL; + } + logdev_dev[cpu].len = 0; + kfree(logdev_dev[cpu].entry); + logdev_dev[cpu].entry = NULL; + } + + misc_deregister(&logdev_misc_dev[cpu]); + } +} + +int __init logdev_init(void) +{ + int res = 0; + int cpu; + struct logdev_entry *entry; + + printk("Logdevice: copyright Steven Rostedt, Kihon Technologies Inc." + " (Version %s)\n", + logdev_version); + + for_each_present_cpu(cpu) { + int i; + char *name; + + if (cpu >= LOGDEV_CPUS) { + printk(KERN_WARNING "More present cpus (%d) than NR_CPUS (%d)\n", + cpu,LOGDEV_CPUS); + break; + } + + printk("Initializing logdev for cpu: %d\n",cpu); + + name = kmalloc(32,GFP_KERNEL); + if (!name) { + printk(KERN_WARNING "logdev: can't allocate misc name\n"); + continue; + } + + sprintf(name,"logdev%d",cpu); + + logdev_misc_dev[cpu].name = name; + + res = misc_register(&logdev_misc_dev[cpu]); + if (res) { + printk("Can't register misc device for Logdev CPU %d\n",cpu); + kfree(name); + logdev_misc_dev[cpu].name = NULL; + continue; + } + + atomic_set(&logdev_dev[cpu].minor,logdev_misc_dev[cpu].minor); + + logdev_dev[cpu].len = pages; + + res = -ENOMEM; + + logdev_dev[cpu].entry = kmalloc(sizeof(struct logdev_entry)*logdev_dev[cpu].len,GFP_KERNEL); + if (!logdev_dev[cpu].entry) + goto fail; + + memset(logdev_dev[cpu].entry,0,sizeof(*logdev_dev[cpu].entry)*logdev_dev[cpu].len); + + for (i=0,entry=logdev_dev[cpu].entry; idat = (void*)__get_free_pages(GFP_KERNEL,0); + if (!entry->dat) + goto fail; + } + init_waitqueue_head(&logdev_dev[cpu].wait); + spin_lock_init(&logdev_dev[cpu].lock); + + logdev_dev[cpu].init = 1; + +#ifndef CONFIG_LOGDEV_MULTI_CPUS + break; +#endif + + } + + logdev_proc_setup(); + + notifier_chain_register(&panic_notifier_list, &logdev_panic_notifier); + +#ifndef MODULE + register_die_notifier(&logdev_die_notifier); +#endif + +#undef LOGDEV_SWITCHON +#ifdef LOGDEV_SWITCHON + atomic_set(&logdev_switch,1); +#endif + + res = 0; + + +#ifdef MODULE + { + unsigned long flags; + /* Make hooks last! */ + spin_lock_irqsave(&logdev_add_hook_lock,flags); + logdev_record_switch_hook = logdev_record_switch; + logdev_pkt_hook = logdev_pkt; + logdev_print_hook = logdev_print; + logdev_vprint_hook = logdev_vprint; + logdev_print_time_hook = logdev_print_time; + logdev_record_hook = logdev_record; + logdev_write_hook = logdev_record_write; + logdev_read_hook = logdev_record_read; + logdev_dumpnet_hook = logdev_dumpnet; + logdev_dump_hook = logdev_dump; + spin_unlock_irqrestore(&logdev_add_hook_lock,flags); + } +#ifdef CONFIG_NETPOLL + logdevnet_init(); +#endif +#endif + +out: + return res; + +fail: + logdev_cleanup(); + goto out; +} + +#ifndef MODULE +EXPORT_SYMBOL(logdev_record_switch); +EXPORT_SYMBOL(logdev_switch_turnoff); +EXPORT_SYMBOL(logdev_pkt); +EXPORT_SYMBOL(logdev_print); +EXPORT_SYMBOL(logdev_print_time); +EXPORT_SYMBOL(logdev_record); +EXPORT_SYMBOL(logdev_record_write); +EXPORT_SYMBOL(logdev_record_read); +EXPORT_SYMBOL(logdev_dumpnet); +EXPORT_SYMBOL(logdev_dump); + +#ifdef CONFIG_NETPOLL +late_initcall(logdevnet_init); +#endif +#endif + +module_init(logdev_init); +module_exit(logdev_cleanup); Index: drivers/char/sysrq.c =================================================================== --- drivers/char/sysrq.c (.../tags/starting_base_kernel_2.6.12) (revision 300) +++ drivers/char/sysrq.c (.../trunk/kernel) (revision 300) @@ -31,6 +31,7 @@ #include #include #include /* for fsync_bdev() */ +#include #include #include #include @@ -241,6 +242,39 @@ .enable_mask = SYSRQ_ENABLE_SIGNAL, }; +static void sysrq_handle_dumplog(int key, struct pt_regs *pt_regs, + struct tty_struct *tty) +{ + LOGDUMP(); +} + +static struct sysrq_key_op sysrq_dumplog_op = { + .handler = sysrq_handle_dumplog, + .help_msg = "Dumplog", + .action_msg = "Dump logdev to serial", +}; + +static void sysrq_handle_togglelogswitch(int key, struct pt_regs *pt_regs, + struct tty_struct *tty) +{ + /* + * Not so atomic, but we really don't care! + */ + if (atomic_read(&logdev_switch)) { + atomic_set(&logdev_switch,0); + printk("logdev_switch now off\n"); + } else { + atomic_set(&logdev_switch,1); + printk("logdev_switch now on\n"); + } +} + +static struct sysrq_key_op sysrq_togglelogdevswitch_op = { + .handler = sysrq_handle_togglelogswitch, + .help_msg = "Togglelogswitch", + .action_msg = "Toggling logdev_switch", +}; + /* END SIGNAL SYSRQ HANDLERS BLOCK */ static void sysrq_handle_unrt(int key, struct pt_regs *pt_regs, @@ -273,11 +307,11 @@ it is handled specially on the sparc and will never arrive */ /* b */ &sysrq_reboot_op, -/* c */ NULL, +/* c */ &sysrq_moom_op, /* d */ NULL, /* e */ &sysrq_term_op, -/* f */ &sysrq_moom_op, -/* g */ NULL, +/* f */ &sysrq_togglelogdevswitch_op, +/* g */ &sysrq_dumplog_op, /* h */ NULL, /* i */ &sysrq_kill_op, /* j */ NULL, Index: drivers/char/logdev_hooks.c =================================================================== --- drivers/char/logdev_hooks.c (.../tags/starting_base_kernel_2.6.12) (revision 0) +++ drivers/char/logdev_hooks.c (.../trunk/kernel) (revision 300) @@ -0,0 +1,81 @@ +#include +#include +#include +#include +#include +#include + +/* + * Only need to add this is LOGDEV is defined as a module + */ +#ifndef CONFIG_LOGDEV +int (*logdev_print_hook)(const char *fmt, ...); +int (*logdev_vprint_hook)(const char *fmt, va_list va); +int (*logdev_print_time_hook)(const char *fmt, ...); +int (*logdev_print_time_func_hook)(const char *fmt, ...); +void (*logdev_record_switch_hook)(struct task_struct *a,struct task_struct *b); +int (*logdev_record_hook)(int id, int size, const void *data); +int (*logdev_write_hook)(const char *data, int size); +int (*logdev_read_hook)(char *buf, int size); +void (*logdev_pkt_hook)(struct sk_buff *skb, int dir); +void (*logdev_dump_hook)(void); +void (*logdev_dumpnet_hook)(void); + +EXPORT_SYMBOL(logdev_print_hook); +EXPORT_SYMBOL(logdev_vprint_hook); +EXPORT_SYMBOL(logdev_print_time_hook); +EXPORT_SYMBOL(logdev_print_time_func_hook); +EXPORT_SYMBOL(logdev_record_switch_hook); +EXPORT_SYMBOL(logdev_record_hook); +EXPORT_SYMBOL(logdev_write_hook); +EXPORT_SYMBOL(logdev_read_hook); +EXPORT_SYMBOL(logdev_pkt_hook); +EXPORT_SYMBOL(logdev_dumpnet_hook); +EXPORT_SYMBOL(logdev_dump_hook); +EXPORT_SYMBOL(logdev_add_hook_lock); + +void logdev_add_hook(void **hook, void *func) +{ + spin_lock(&logdev_add_hook_lock); + *hook = func; + spin_unlock(&logdev_add_hook_lock); +} + +void logdev_remove_hook(void **hook) +{ + spin_lock(&logdev_add_hook_lock); + *hook = NULL; + spin_unlock(&logdev_add_hook_lock); +} +#endif + +void logdev_time(void) +{ + unsigned long long t; + unsigned long nanosec_rem; + t = sched_clock(); + nanosec_rem = do_div(t, 1000000000); + LOGPRINT("[%5lu.%06lu] ", + (unsigned long)t, + nanosec_rem/1000); +} + +atomic_t logdev_level = ATOMIC_INIT(0); +atomic_t logdev_switch = ATOMIC_INIT(0); +#ifdef CONFIG_LOGDEV_PRINT_ENABLED +atomic_t logdev_print_enabled = ATOMIC_INIT(1); +#else +atomic_t logdev_print_enabled = ATOMIC_INIT(0); +#endif + +int in_logdump; + +raw_spinlock_t logdev_add_hook_lock = RAW_SPIN_LOCK_UNLOCKED; + +/* + * sched_clock isn't exported, so we export it ourselves. + */ +EXPORT_SYMBOL(logdev_switch); +EXPORT_SYMBOL(logdev_level); +EXPORT_SYMBOL(logdev_print_enabled); +EXPORT_SYMBOL(sched_clock);