diff --git a/drivers/misc/Kconfig b/drivers/misc/Kconfig index 1b27bf88870c..ba9d624472a2 100644 --- a/drivers/misc/Kconfig +++ b/drivers/misc/Kconfig @@ -515,9 +515,18 @@ config MISC_RTSX config UID_SYS_STATS bool "Per-UID statistics" - depends on PROFILING + depends on PROFILING && TASK_IO_ACCOUNTING help Per UID based cpu time statistics exported to /proc/uid_cputime + Per UID based io statistics exported to /proc/uid_io + Per UID based procstat control in /proc/uid_procstat + +config UID_SYS_STATS_DEBUG + bool "Per-TASK statistics" + depends on UID_SYS_STATS + default n + help + Per TASK based io statistics exported to /proc/uid_io source "drivers/misc/c2port/Kconfig" source "drivers/misc/eeprom/Kconfig" diff --git a/drivers/misc/uid_sys_stats.c b/drivers/misc/uid_sys_stats.c index e1a9ef29468f..ea23f5d0af43 100644 --- a/drivers/misc/uid_sys_stats.c +++ b/drivers/misc/uid_sys_stats.c @@ -19,6 +19,7 @@ #include #include #include +#include #include #include #include @@ -32,7 +33,35 @@ DECLARE_HASHTABLE(hash_table, UID_HASH_BITS); static DEFINE_RT_MUTEX(uid_lock); -static struct proc_dir_entry *parent; +static struct proc_dir_entry *cpu_parent; +static struct proc_dir_entry *io_parent; +static struct proc_dir_entry *proc_parent; + +struct io_stats { + u64 read_bytes; + u64 write_bytes; + u64 rchar; + u64 wchar; + u64 fsync; +}; + +#define UID_STATE_FOREGROUND 0 +#define UID_STATE_BACKGROUND 1 +#define UID_STATE_BUCKET_SIZE 2 + +#define UID_STATE_TOTAL_CURR 2 +#define UID_STATE_TOTAL_LAST 3 +#define UID_STATE_DEAD_TASKS 4 +#define UID_STATE_SIZE 5 + +#define MAX_TASK_COMM_LEN 256 + +struct task_entry { + char comm[MAX_TASK_COMM_LEN]; + pid_t pid; + struct io_stats io[UID_STATE_SIZE]; + struct hlist_node hash; +}; struct uid_entry { uid_t uid; @@ -40,9 +69,233 @@ struct uid_entry { u64 stime; u64 active_utime; u64 active_stime; + int state; + struct io_stats io[UID_STATE_SIZE]; struct hlist_node hash; +#ifdef CONFIG_UID_SYS_STATS_DEBUG + DECLARE_HASHTABLE(task_entries, UID_HASH_BITS); +#endif }; +static u64 compute_write_bytes(struct task_struct *task) +{ + if (task->ioac.write_bytes <= task->ioac.cancelled_write_bytes) + return 0; + + return task->ioac.write_bytes - task->ioac.cancelled_write_bytes; +} + +static void compute_io_bucket_stats(struct io_stats *io_bucket, + struct io_stats *io_curr, + struct io_stats *io_last, + struct io_stats *io_dead) +{ + /* tasks could switch to another uid group, but its io_last in the + * previous uid group could still be positive. + * therefore before each update, do an overflow check first + */ + int64_t delta; + + delta = io_curr->read_bytes + io_dead->read_bytes - + io_last->read_bytes; + io_bucket->read_bytes += delta > 0 ? delta : 0; + delta = io_curr->write_bytes + io_dead->write_bytes - + io_last->write_bytes; + io_bucket->write_bytes += delta > 0 ? delta : 0; + delta = io_curr->rchar + io_dead->rchar - io_last->rchar; + io_bucket->rchar += delta > 0 ? delta : 0; + delta = io_curr->wchar + io_dead->wchar - io_last->wchar; + io_bucket->wchar += delta > 0 ? delta : 0; + delta = io_curr->fsync + io_dead->fsync - io_last->fsync; + io_bucket->fsync += delta > 0 ? delta : 0; + + io_last->read_bytes = io_curr->read_bytes; + io_last->write_bytes = io_curr->write_bytes; + io_last->rchar = io_curr->rchar; + io_last->wchar = io_curr->wchar; + io_last->fsync = io_curr->fsync; + + memset(io_dead, 0, sizeof(struct io_stats)); +} + +#ifdef CONFIG_UID_SYS_STATS_DEBUG +static void get_full_task_comm(struct task_entry *task_entry, + struct task_struct *task) +{ + int i = 0, offset = 0, len = 0; + /* save one byte for terminating null character */ + int unused_len = MAX_TASK_COMM_LEN - TASK_COMM_LEN - 1; + char buf[unused_len]; + struct mm_struct *mm = task->mm; + + /* fill the first TASK_COMM_LEN bytes with thread name */ + __get_task_comm(task_entry->comm, TASK_COMM_LEN, task); + i = strlen(task_entry->comm); + while (i < TASK_COMM_LEN) + task_entry->comm[i++] = ' '; + + /* next the executable file name */ + if (mm) { + down_read(&mm->mmap_sem); + if (mm->exe_file) { + char *pathname = d_path(&mm->exe_file->f_path, buf, + unused_len); + + if (!IS_ERR(pathname)) { + len = strlcpy(task_entry->comm + i, pathname, + unused_len); + i += len; + task_entry->comm[i++] = ' '; + unused_len--; + } + } + up_read(&mm->mmap_sem); + } + unused_len -= len; + + /* fill the rest with command line argument + * replace each null or new line character + * between args in argv with whitespace */ + len = get_cmdline(task, buf, unused_len); + while (offset < len) { + if (buf[offset] != '\0' && buf[offset] != '\n') + task_entry->comm[i++] = buf[offset]; + else + task_entry->comm[i++] = ' '; + offset++; + } + + /* get rid of trailing whitespaces in case when arg is memset to + * zero before being reset in userspace + */ + while (task_entry->comm[i-1] == ' ') + i--; + task_entry->comm[i] = '\0'; +} + +static struct task_entry *find_task_entry(struct uid_entry *uid_entry, + struct task_struct *task) +{ + struct task_entry *task_entry; + + hash_for_each_possible(uid_entry->task_entries, task_entry, hash, + task->pid) { + if (task->pid == task_entry->pid) { + /* if thread name changed, update the entire command */ + int len = strnchr(task_entry->comm, ' ', TASK_COMM_LEN) + - task_entry->comm; + + if (strncmp(task_entry->comm, task->comm, len)) + get_full_task_comm(task_entry, task); + return task_entry; + } + } + return NULL; +} + +static struct task_entry *find_or_register_task(struct uid_entry *uid_entry, + struct task_struct *task) +{ + struct task_entry *task_entry; + pid_t pid = task->pid; + + task_entry = find_task_entry(uid_entry, task); + if (task_entry) + return task_entry; + + task_entry = kzalloc(sizeof(struct task_entry), GFP_ATOMIC); + if (!task_entry) + return NULL; + + get_full_task_comm(task_entry, task); + + task_entry->pid = pid; + hash_add(uid_entry->task_entries, &task_entry->hash, (unsigned int)pid); + + return task_entry; +} + +static void remove_uid_tasks(struct uid_entry *uid_entry) +{ + struct task_entry *task_entry; + unsigned long bkt_task; + struct hlist_node *tmp_task; + + hash_for_each_safe(uid_entry->task_entries, bkt_task, + tmp_task, task_entry, hash) { + hash_del(&task_entry->hash); + kfree(task_entry); + } +} + +static void set_io_uid_tasks_zero(struct uid_entry *uid_entry) +{ + struct task_entry *task_entry; + unsigned long bkt_task; + + hash_for_each(uid_entry->task_entries, bkt_task, task_entry, hash) { + memset(&task_entry->io[UID_STATE_TOTAL_CURR], 0, + sizeof(struct io_stats)); + } +} + +static void add_uid_tasks_io_stats(struct uid_entry *uid_entry, + struct task_struct *task, int slot) +{ + struct task_entry *task_entry = find_or_register_task(uid_entry, task); + struct io_stats *task_io_slot = &task_entry->io[slot]; + + task_io_slot->read_bytes += task->ioac.read_bytes; + task_io_slot->write_bytes += compute_write_bytes(task); + task_io_slot->rchar += task->ioac.rchar; + task_io_slot->wchar += task->ioac.wchar; +} + +static void compute_io_uid_tasks(struct uid_entry *uid_entry) +{ + struct task_entry *task_entry; + unsigned long bkt_task; + + hash_for_each(uid_entry->task_entries, bkt_task, task_entry, hash) { + compute_io_bucket_stats(&task_entry->io[uid_entry->state], + &task_entry->io[UID_STATE_TOTAL_CURR], + &task_entry->io[UID_STATE_TOTAL_LAST], + &task_entry->io[UID_STATE_DEAD_TASKS]); + } +} + +static void show_io_uid_tasks(struct seq_file *m, struct uid_entry *uid_entry) +{ + struct task_entry *task_entry; + unsigned long bkt_task; + + hash_for_each(uid_entry->task_entries, bkt_task, task_entry, hash) { + /* Separated by comma because space exists in task comm */ + seq_printf(m, "task,%s,%lu,%llu,%llu,%llu,%llu,%llu,%llu,%llu,%llu,%llu,%llu\n", + task_entry->comm, + (unsigned long)task_entry->pid, + task_entry->io[UID_STATE_FOREGROUND].rchar, + task_entry->io[UID_STATE_FOREGROUND].wchar, + task_entry->io[UID_STATE_FOREGROUND].read_bytes, + task_entry->io[UID_STATE_FOREGROUND].write_bytes, + task_entry->io[UID_STATE_BACKGROUND].rchar, + task_entry->io[UID_STATE_BACKGROUND].wchar, + task_entry->io[UID_STATE_BACKGROUND].read_bytes, + task_entry->io[UID_STATE_BACKGROUND].write_bytes, + task_entry->io[UID_STATE_FOREGROUND].fsync, + task_entry->io[UID_STATE_BACKGROUND].fsync); + } +} +#else +static void remove_uid_tasks(struct uid_entry *uid_entry) {}; +static void set_io_uid_tasks_zero(struct uid_entry *uid_entry) {}; +static void add_uid_tasks_io_stats(struct uid_entry *uid_entry, + struct task_struct *task, int slot) {}; +static void compute_io_uid_tasks(struct uid_entry *uid_entry) {}; +static void show_io_uid_tasks(struct seq_file *m, + struct uid_entry *uid_entry) {} +#endif + static struct uid_entry *find_uid_entry(uid_t uid) { struct uid_entry *uid_entry; @@ -66,13 +319,15 @@ static struct uid_entry *find_or_register_uid(uid_t uid) return NULL; uid_entry->uid = uid; - +#ifdef CONFIG_UID_SYS_STATS_DEBUG + hash_init(uid_entry->task_entries); +#endif hash_add(hash_table, &uid_entry->hash, uid); return uid_entry; } -static int uid_stat_show(struct seq_file *m, void *v) +static int uid_cputime_show(struct seq_file *m, void *v) { struct uid_entry *uid_entry = NULL; struct task_struct *task, *temp; @@ -120,13 +375,13 @@ static int uid_stat_show(struct seq_file *m, void *v) return 0; } -static int uid_stat_open(struct inode *inode, struct file *file) +static int uid_cputime_open(struct inode *inode, struct file *file) { - return single_open(file, uid_stat_show, PDE_DATA(inode)); + return single_open(file, uid_cputime_show, PDE_DATA(inode)); } -static const struct file_operations uid_stat_fops = { - .open = uid_stat_open, +static const struct file_operations uid_cputime_fops = { + .open = uid_cputime_open, .read = seq_read, .llseek = seq_lseek, .release = single_release, @@ -163,12 +418,14 @@ static ssize_t uid_remove_write(struct file *file, kstrtol(end_uid, 10, &uid_end) != 0) { return -EINVAL; } + rt_mutex_lock(&uid_lock); for (; uid_start <= uid_end; uid_start++) { hash_for_each_possible_safe(hash_table, uid_entry, tmp, hash, (uid_t)uid_start) { if (uid_start == uid_entry->uid) { + remove_uid_tasks(uid_entry); hash_del(&uid_entry->hash); kfree(uid_entry); } @@ -185,6 +442,177 @@ static const struct file_operations uid_remove_fops = { .write = uid_remove_write, }; + +static void add_uid_io_stats(struct uid_entry *uid_entry, + struct task_struct *task, int slot) +{ + struct io_stats *io_slot = &uid_entry->io[slot]; + + io_slot->read_bytes += task->ioac.read_bytes; + io_slot->write_bytes += compute_write_bytes(task); + io_slot->rchar += task->ioac.rchar; + io_slot->wchar += task->ioac.wchar; + + add_uid_tasks_io_stats(uid_entry, task, slot); +} + +static void update_io_stats_all_locked(void) +{ + struct uid_entry *uid_entry = NULL; + struct task_struct *task, *temp; + struct user_namespace *user_ns = current_user_ns(); + unsigned long bkt; + uid_t uid; + + hash_for_each(hash_table, bkt, uid_entry, hash) { + memset(&uid_entry->io[UID_STATE_TOTAL_CURR], 0, + sizeof(struct io_stats)); + set_io_uid_tasks_zero(uid_entry); + } + + rcu_read_lock(); + do_each_thread(temp, task) { + uid = from_kuid_munged(user_ns, task_uid(task)); + if (!uid_entry || uid_entry->uid != uid) + uid_entry = find_or_register_uid(uid); + if (!uid_entry) + continue; + add_uid_io_stats(uid_entry, task, UID_STATE_TOTAL_CURR); + } while_each_thread(temp, task); + rcu_read_unlock(); + + hash_for_each(hash_table, bkt, uid_entry, hash) { + compute_io_bucket_stats(&uid_entry->io[uid_entry->state], + &uid_entry->io[UID_STATE_TOTAL_CURR], + &uid_entry->io[UID_STATE_TOTAL_LAST], + &uid_entry->io[UID_STATE_DEAD_TASKS]); + compute_io_uid_tasks(uid_entry); + } +} + +static void update_io_stats_uid_locked(struct uid_entry *uid_entry) +{ + struct task_struct *task, *temp; + struct user_namespace *user_ns = current_user_ns(); + + memset(&uid_entry->io[UID_STATE_TOTAL_CURR], 0, + sizeof(struct io_stats)); + set_io_uid_tasks_zero(uid_entry); + + rcu_read_lock(); + do_each_thread(temp, task) { + if (from_kuid_munged(user_ns, task_uid(task)) != uid_entry->uid) + continue; + add_uid_io_stats(uid_entry, task, UID_STATE_TOTAL_CURR); + } while_each_thread(temp, task); + rcu_read_unlock(); + + compute_io_bucket_stats(&uid_entry->io[uid_entry->state], + &uid_entry->io[UID_STATE_TOTAL_CURR], + &uid_entry->io[UID_STATE_TOTAL_LAST], + &uid_entry->io[UID_STATE_DEAD_TASKS]); + compute_io_uid_tasks(uid_entry); +} + + +static int uid_io_show(struct seq_file *m, void *v) +{ + struct uid_entry *uid_entry; + unsigned long bkt; + + rt_mutex_lock(&uid_lock); + + update_io_stats_all_locked(); + + hash_for_each(hash_table, bkt, uid_entry, hash) { + seq_printf(m, "%d %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu\n", + uid_entry->uid, + uid_entry->io[UID_STATE_FOREGROUND].rchar, + uid_entry->io[UID_STATE_FOREGROUND].wchar, + uid_entry->io[UID_STATE_FOREGROUND].read_bytes, + uid_entry->io[UID_STATE_FOREGROUND].write_bytes, + uid_entry->io[UID_STATE_BACKGROUND].rchar, + uid_entry->io[UID_STATE_BACKGROUND].wchar, + uid_entry->io[UID_STATE_BACKGROUND].read_bytes, + uid_entry->io[UID_STATE_BACKGROUND].write_bytes, + uid_entry->io[UID_STATE_FOREGROUND].fsync, + uid_entry->io[UID_STATE_BACKGROUND].fsync); + + show_io_uid_tasks(m, uid_entry); + } + + rt_mutex_unlock(&uid_lock); + return 0; +} + +static int uid_io_open(struct inode *inode, struct file *file) +{ + return single_open(file, uid_io_show, PDE_DATA(inode)); +} + +static const struct file_operations uid_io_fops = { + .open = uid_io_open, + .read = seq_read, + .llseek = seq_lseek, + .release = single_release, +}; + +static int uid_procstat_open(struct inode *inode, struct file *file) +{ + return single_open(file, NULL, NULL); +} + +static ssize_t uid_procstat_write(struct file *file, + const char __user *buffer, size_t count, loff_t *ppos) +{ + struct uid_entry *uid_entry; + uid_t uid; + int argc, state; + char input[128]; + + if (count >= sizeof(input)) + return -EINVAL; + + if (copy_from_user(input, buffer, count)) + return -EFAULT; + + input[count] = '\0'; + + argc = sscanf(input, "%u %d", &uid, &state); + if (argc != 2) + return -EINVAL; + + if (state != UID_STATE_BACKGROUND && state != UID_STATE_FOREGROUND) + return -EINVAL; + + rt_mutex_lock(&uid_lock); + + uid_entry = find_or_register_uid(uid); + if (!uid_entry) { + rt_mutex_unlock(&uid_lock); + return -EINVAL; + } + + if (uid_entry->state == state) { + rt_mutex_unlock(&uid_lock); + return count; + } + + update_io_stats_uid_locked(uid_entry); + + uid_entry->state = state; + + rt_mutex_unlock(&uid_lock); + + return count; +} + +static const struct file_operations uid_procstat_fops = { + .open = uid_procstat_open, + .release = single_release, + .write = uid_procstat_write, +}; + static int process_notifier(struct notifier_block *self, unsigned long cmd, void *v) { @@ -208,6 +636,8 @@ static int process_notifier(struct notifier_block *self, uid_entry->utime += utime; uid_entry->stime += stime; + add_uid_io_stats(uid_entry, task, UID_STATE_DEAD_TASKS); + exit: rt_mutex_unlock(&uid_lock); return NOTIFY_OK; @@ -217,25 +647,51 @@ static struct notifier_block process_notifier_block = { .notifier_call = process_notifier, }; -static int __init proc_uid_cputime_init(void) +static int __init proc_uid_sys_stats_init(void) { hash_init(hash_table); - parent = proc_mkdir("uid_cputime", NULL); - if (!parent) { - pr_err("%s: failed to create proc entry\n", __func__); - return -ENOMEM; + cpu_parent = proc_mkdir("uid_cputime", NULL); + if (!cpu_parent) { + pr_err("%s: failed to create uid_cputime proc entry\n", + __func__); + goto err; } - proc_create_data("remove_uid_range", S_IWUGO, parent, &uid_remove_fops, - NULL); + proc_create_data("remove_uid_range", 0222, cpu_parent, + &uid_remove_fops, NULL); + proc_create_data("show_uid_stat", 0444, cpu_parent, + &uid_cputime_fops, NULL); - proc_create_data("show_uid_stat", S_IRUGO, parent, &uid_stat_fops, - NULL); + io_parent = proc_mkdir("uid_io", NULL); + if (!io_parent) { + pr_err("%s: failed to create uid_io proc entry\n", + __func__); + goto err; + } + + proc_create_data("stats", 0444, io_parent, + &uid_io_fops, NULL); + + proc_parent = proc_mkdir("uid_procstat", NULL); + if (!proc_parent) { + pr_err("%s: failed to create uid_procstat proc entry\n", + __func__); + goto err; + } + + proc_create_data("set", 0222, proc_parent, + &uid_procstat_fops, NULL); profile_event_register(PROFILE_TASK_EXIT, &process_notifier_block); return 0; + +err: + remove_proc_subtree("uid_cputime", NULL); + remove_proc_subtree("uid_io", NULL); + remove_proc_subtree("uid_procstat", NULL); + return -ENOMEM; } -early_initcall(proc_uid_cputime_init); +early_initcall(proc_uid_sys_stats_init);