maps4: add /proc/pid/pagemap interface
[deliverable/linux.git] / fs / proc / base.c
index 39a3d7c969c5de96496e418eb397b3dec4ed789b..9004db04efa0c06b24ec75e8e9b1768efd1b92fb 100644 (file)
  *     in /proc for a task before it execs a suid executable.
  */
 
-
-/* Worst case buffer size needed for holding an integer. */
-#define PROC_NUMBUF 13
-
 struct pid_entry {
        char *name;
        int len;
@@ -199,9 +195,29 @@ static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vf
        (task == current || \
        (task->parent == current && \
        (task->ptrace & PT_PTRACED) && \
-        (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \
+        (task_is_stopped_or_traced(task)) && \
         security_ptrace(current,task) == 0))
 
+struct mm_struct *mm_for_maps(struct task_struct *task)
+{
+       struct mm_struct *mm = get_task_mm(task);
+       if (!mm)
+               return NULL;
+       down_read(&mm->mmap_sem);
+       task_lock(task);
+       if (task->mm != mm)
+               goto out;
+       if (task->mm != current->mm && __ptrace_may_attach(task) < 0)
+               goto out;
+       task_unlock(task);
+       return mm;
+out:
+       task_unlock(task);
+       up_read(&mm->mmap_sem);
+       mmput(mm);
+       return NULL;
+}
+
 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 {
        int res = 0;
@@ -290,6 +306,77 @@ static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 }
 #endif
 
+#ifdef CONFIG_LATENCYTOP
+static int lstats_show_proc(struct seq_file *m, void *v)
+{
+       int i;
+       struct task_struct *task = m->private;
+       seq_puts(m, "Latency Top version : v0.1\n");
+
+       for (i = 0; i < 32; i++) {
+               if (task->latency_record[i].backtrace[0]) {
+                       int q;
+                       seq_printf(m, "%i %li %li ",
+                               task->latency_record[i].count,
+                               task->latency_record[i].time,
+                               task->latency_record[i].max);
+                       for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
+                               char sym[KSYM_NAME_LEN];
+                               char *c;
+                               if (!task->latency_record[i].backtrace[q])
+                                       break;
+                               if (task->latency_record[i].backtrace[q] == ULONG_MAX)
+                                       break;
+                               sprint_symbol(sym, task->latency_record[i].backtrace[q]);
+                               c = strchr(sym, '+');
+                               if (c)
+                                       *c = 0;
+                               seq_printf(m, "%s ", sym);
+                       }
+                       seq_printf(m, "\n");
+               }
+
+       }
+       return 0;
+}
+
+static int lstats_open(struct inode *inode, struct file *file)
+{
+       int ret;
+       struct seq_file *m;
+       struct task_struct *task = get_proc_task(inode);
+
+       ret = single_open(file, lstats_show_proc, NULL);
+       if (!ret) {
+               m = file->private_data;
+               m->private = task;
+       }
+       return ret;
+}
+
+static ssize_t lstats_write(struct file *file, const char __user *buf,
+                           size_t count, loff_t *offs)
+{
+       struct seq_file *m;
+       struct task_struct *task;
+
+       m = file->private_data;
+       task = m->private;
+       clear_all_latency_tracing(task);
+
+       return count;
+}
+
+static const struct file_operations proc_lstats_operations = {
+       .open           = lstats_open,
+       .read           = seq_read,
+       .write          = lstats_write,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+#endif
+
 /* The badness from the OOM killer */
 unsigned long badness(struct task_struct *p, unsigned long uptime);
 static int proc_oom_score(struct task_struct *task, char *buffer)
@@ -696,7 +783,7 @@ out_no_task:
 }
 #endif
 
-static loff_t mem_lseek(struct file * file, loff_t offset, int orig)
+loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 {
        switch (orig) {
        case 0:
@@ -844,42 +931,6 @@ static const struct file_operations proc_oom_adjust_operations = {
        .write          = oom_adjust_write,
 };
 
-#ifdef CONFIG_MMU
-static ssize_t clear_refs_write(struct file *file, const char __user *buf,
-                               size_t count, loff_t *ppos)
-{
-       struct task_struct *task;
-       char buffer[PROC_NUMBUF], *end;
-       struct mm_struct *mm;
-
-       memset(buffer, 0, sizeof(buffer));
-       if (count > sizeof(buffer) - 1)
-               count = sizeof(buffer) - 1;
-       if (copy_from_user(buffer, buf, count))
-               return -EFAULT;
-       if (!simple_strtol(buffer, &end, 0))
-               return -EINVAL;
-       if (*end == '\n')
-               end++;
-       task = get_proc_task(file->f_path.dentry->d_inode);
-       if (!task)
-               return -ESRCH;
-       mm = get_task_mm(task);
-       if (mm) {
-               clear_refs_smap(mm);
-               mmput(mm);
-       }
-       put_task_struct(task);
-       if (end - buffer == 0)
-               return -EIO;
-       return end - buffer;
-}
-
-static struct file_operations proc_clear_refs_operations = {
-       .write          = clear_refs_write,
-};
-#endif
-
 #ifdef CONFIG_AUDITSYSCALL
 #define TMPBUFLEN 21
 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
@@ -893,7 +944,7 @@ static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
        if (!task)
                return -ESRCH;
        length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
-                               audit_get_loginuid(task->audit_context));
+                               audit_get_loginuid(task));
        put_task_struct(task);
        return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
 }
@@ -1000,6 +1051,7 @@ static const struct file_operations proc_fault_inject_operations = {
 };
 #endif
 
+
 #ifdef CONFIG_SCHED_DEBUG
 /*
  * Print out various scheduling related per-task fields:
@@ -2200,6 +2252,7 @@ static const struct pid_entry tgid_base_stuff[] = {
 #ifdef CONFIG_MMU
        REG("clear_refs", S_IWUSR, clear_refs),
        REG("smaps",      S_IRUGO, smaps),
+       REG("pagemap",    S_IRUSR, pagemap),
 #endif
 #ifdef CONFIG_SECURITY
        DIR("attr",       S_IRUGO|S_IXUGO, attr_dir),
@@ -2210,6 +2263,9 @@ static const struct pid_entry tgid_base_stuff[] = {
 #ifdef CONFIG_SCHEDSTATS
        INF("schedstat",  S_IRUGO, pid_schedstat),
 #endif
+#ifdef CONFIG_LATENCYTOP
+       REG("latency",  S_IRUGO, lstats),
+#endif
 #ifdef CONFIG_PROC_PID_CPUSET
        REG("cpuset",     S_IRUGO, cpuset),
 #endif
@@ -2255,27 +2311,6 @@ static const struct inode_operations proc_tgid_base_inode_operations = {
        .setattr        = proc_setattr,
 };
 
-/**
- * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
- *
- * @task: task that should be flushed.
- *
- * Looks in the dcache for
- * /proc/@pid
- * /proc/@tgid/task/@pid
- * if either directory is present flushes it and all of it'ts children
- * from the dcache.
- *
- * It is safe and reasonable to cache /proc entries for a task until
- * that task exits.  After that they just clog up the dcache with
- * useless entries, possibly causing useful dcache entries to be
- * flushed instead.  This routine is proved to flush those useless
- * dcache entries at process exit time.
- *
- * NOTE: This routine is just an optimization so it does not guarantee
- *       that no dcache entries will exist at process exit time it
- *       just makes it very unlikely that any will persist.
- */
 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
 {
        struct dentry *dentry, *leader, *dir;
@@ -2322,29 +2357,45 @@ out:
        return;
 }
 
-/*
- * when flushing dentries from proc one need to flush them from global
+/**
+ * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
+ * @task: task that should be flushed.
+ *
+ * When flushing dentries from proc, one needs to flush them from global
  * proc (proc_mnt) and from all the namespaces' procs this task was seen
- * in. this call is supposed to make all this job.
+ * in. This call is supposed to do all of this job.
+ *
+ * Looks in the dcache for
+ * /proc/@pid
+ * /proc/@tgid/task/@pid
+ * if either directory is present flushes it and all of it'ts children
+ * from the dcache.
+ *
+ * It is safe and reasonable to cache /proc entries for a task until
+ * that task exits.  After that they just clog up the dcache with
+ * useless entries, possibly causing useful dcache entries to be
+ * flushed instead.  This routine is proved to flush those useless
+ * dcache entries at process exit time.
+ *
+ * NOTE: This routine is just an optimization so it does not guarantee
+ *       that no dcache entries will exist at process exit time it
+ *       just makes it very unlikely that any will persist.
  */
 
 void proc_flush_task(struct task_struct *task)
 {
-       int i, leader;
-       struct pid *pid, *tgid;
+       int i;
+       struct pid *pid, *tgid = NULL;
        struct upid *upid;
 
-       leader = thread_group_leader(task);
-       proc_flush_task_mnt(proc_mnt, task->pid, leader ? task->tgid : 0);
        pid = task_pid(task);
-       if (pid->level == 0)
-               return;
+       if (thread_group_leader(task))
+               tgid = task_tgid(task);
 
-       tgid = task_tgid(task);
-       for (i = 1; i <= pid->level; i++) {
+       for (i = 0; i <= pid->level; i++) {
                upid = &pid->numbers[i];
                proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
-                               leader ? 0 : tgid->numbers[i].nr);
+                       tgid ? tgid->numbers[i].nr : 0);
        }
 
        upid = &pid->numbers[pid->level];
@@ -2416,19 +2467,23 @@ out:
  * Find the first task with tgid >= tgid
  *
  */
-static struct task_struct *next_tgid(unsigned int tgid,
-               struct pid_namespace *ns)
-{
+struct tgid_iter {
+       unsigned int tgid;
        struct task_struct *task;
+};
+static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
+{
        struct pid *pid;
 
+       if (iter.task)
+               put_task_struct(iter.task);
        rcu_read_lock();
 retry:
-       task = NULL;
-       pid = find_ge_pid(tgid, ns);
+       iter.task = NULL;
+       pid = find_ge_pid(iter.tgid, ns);
        if (pid) {
-               tgid = pid_nr_ns(pid, ns) + 1;
-               task = pid_task(pid, PIDTYPE_PID);
+               iter.tgid = pid_nr_ns(pid, ns);
+               iter.task = pid_task(pid, PIDTYPE_PID);
                /* What we to know is if the pid we have find is the
                 * pid of a thread_group_leader.  Testing for task
                 * being a thread_group_leader is the obvious thing
@@ -2441,23 +2496,25 @@ retry:
                 * found doesn't happen to be a thread group leader.
                 * As we don't care in the case of readdir.
                 */
-               if (!task || !has_group_leader_pid(task))
+               if (!iter.task || !has_group_leader_pid(iter.task)) {
+                       iter.tgid += 1;
                        goto retry;
-               get_task_struct(task);
+               }
+               get_task_struct(iter.task);
        }
        rcu_read_unlock();
-       return task;
+       return iter;
 }
 
 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
 
 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
-       struct task_struct *task, int tgid)
+       struct tgid_iter iter)
 {
        char name[PROC_NUMBUF];
-       int len = snprintf(name, sizeof(name), "%d", tgid);
+       int len = snprintf(name, sizeof(name), "%d", iter.tgid);
        return proc_fill_cache(filp, dirent, filldir, name, len,
-                               proc_pid_instantiate, task, NULL);
+                               proc_pid_instantiate, iter.task, NULL);
 }
 
 /* for the /proc/ directory itself, after non-process stuff has been done */
@@ -2465,8 +2522,7 @@ int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
 {
        unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
        struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
-       struct task_struct *task;
-       int tgid;
+       struct tgid_iter iter;
        struct pid_namespace *ns;
 
        if (!reaper)
@@ -2479,14 +2535,14 @@ int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
        }
 
        ns = filp->f_dentry->d_sb->s_fs_info;
-       tgid = filp->f_pos - TGID_OFFSET;
-       for (task = next_tgid(tgid, ns);
-            task;
-            put_task_struct(task), task = next_tgid(tgid + 1, ns)) {
-               tgid = task_pid_nr_ns(task, ns);
-               filp->f_pos = tgid + TGID_OFFSET;
-               if (proc_pid_fill_cache(filp, dirent, filldir, task, tgid) < 0) {
-                       put_task_struct(task);
+       iter.task = NULL;
+       iter.tgid = filp->f_pos - TGID_OFFSET;
+       for (iter = next_tgid(ns, iter);
+            iter.task;
+            iter.tgid += 1, iter = next_tgid(ns, iter)) {
+               filp->f_pos = iter.tgid + TGID_OFFSET;
+               if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
+                       put_task_struct(iter.task);
                        goto out;
                }
        }
@@ -2525,6 +2581,7 @@ static const struct pid_entry tid_base_stuff[] = {
 #ifdef CONFIG_MMU
        REG("clear_refs", S_IWUSR, clear_refs),
        REG("smaps",     S_IRUGO, smaps),
+       REG("pagemap",    S_IRUSR, pagemap),
 #endif
 #ifdef CONFIG_SECURITY
        DIR("attr",      S_IRUGO|S_IXUGO, attr_dir),
@@ -2535,6 +2592,9 @@ static const struct pid_entry tid_base_stuff[] = {
 #ifdef CONFIG_SCHEDSTATS
        INF("schedstat", S_IRUGO, pid_schedstat),
 #endif
+#ifdef CONFIG_LATENCYTOP
+       REG("latency",  S_IRUGO, lstats),
+#endif
 #ifdef CONFIG_PROC_PID_CPUSET
        REG("cpuset",    S_IRUGO, cpuset),
 #endif
This page took 0.032943 seconds and 5 git commands to generate.