]> err.no Git - linux-2.6/blobdiff - fs/proc/base.c
Merge branches 'release', 'bugzilla-9916', 'bugzilla-9982', 'bugzilla-9989', 'misc...
[linux-2.6] / fs / proc / base.c
index db7636041c107bd787ac281416c013abf0f85f53..88f8edf18258ac50681474ffc9542c75de30ecda 100644 (file)
@@ -63,6 +63,7 @@
 #include <linux/mm.h>
 #include <linux/rcupdate.h>
 #include <linux/kallsyms.h>
+#include <linux/resource.h>
 #include <linux/module.h>
 #include <linux/mount.h>
 #include <linux/security.h>
  *     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;
@@ -124,6 +121,10 @@ struct pid_entry {
        NOD(NAME, (S_IFREG|(MODE)),                     \
                NULL, &proc_info_file_operations,       \
                { .proc_read = &proc_##OTYPE } )
+#define ONE(NAME, MODE, OTYPE)                         \
+       NOD(NAME, (S_IFREG|(MODE)),                     \
+               NULL, &proc_single_file_operations,     \
+               { .proc_show = &proc_##OTYPE } )
 
 int maps_protect;
 EXPORT_SYMBOL(maps_protect);
@@ -152,7 +153,7 @@ static int get_nr_threads(struct task_struct *tsk)
        return count;
 }
 
-static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
+static int proc_cwd_link(struct inode *inode, struct path *path)
 {
        struct task_struct *task = get_proc_task(inode);
        struct fs_struct *fs = NULL;
@@ -164,8 +165,8 @@ static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfs
        }
        if (fs) {
                read_lock(&fs->lock);
-               *mnt = mntget(fs->pwdmnt);
-               *dentry = dget(fs->pwd);
+               *path = fs->pwd;
+               path_get(&fs->pwd);
                read_unlock(&fs->lock);
                result = 0;
                put_fs_struct(fs);
@@ -173,7 +174,7 @@ static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfs
        return result;
 }
 
-static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
+static int proc_root_link(struct inode *inode, struct path *path)
 {
        struct task_struct *task = get_proc_task(inode);
        struct fs_struct *fs = NULL;
@@ -185,8 +186,8 @@ static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vf
        }
        if (fs) {
                read_lock(&fs->lock);
-               *mnt = mntget(fs->rootmnt);
-               *dentry = dget(fs->root);
+               *path = fs->root;
+               path_get(&fs->root);
                read_unlock(&fs->lock);
                result = 0;
                put_fs_struct(fs);
@@ -198,9 +199,29 @@ 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;
@@ -289,6 +310,77 @@ 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)
@@ -303,6 +395,78 @@ static int proc_oom_score(struct task_struct *task, char *buffer)
        return sprintf(buffer, "%lu\n", points);
 }
 
+struct limit_names {
+       char *name;
+       char *unit;
+};
+
+static const struct limit_names lnames[RLIM_NLIMITS] = {
+       [RLIMIT_CPU] = {"Max cpu time", "ms"},
+       [RLIMIT_FSIZE] = {"Max file size", "bytes"},
+       [RLIMIT_DATA] = {"Max data size", "bytes"},
+       [RLIMIT_STACK] = {"Max stack size", "bytes"},
+       [RLIMIT_CORE] = {"Max core file size", "bytes"},
+       [RLIMIT_RSS] = {"Max resident set", "bytes"},
+       [RLIMIT_NPROC] = {"Max processes", "processes"},
+       [RLIMIT_NOFILE] = {"Max open files", "files"},
+       [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
+       [RLIMIT_AS] = {"Max address space", "bytes"},
+       [RLIMIT_LOCKS] = {"Max file locks", "locks"},
+       [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
+       [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
+       [RLIMIT_NICE] = {"Max nice priority", NULL},
+       [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
+};
+
+/* Display limits for a process */
+static int proc_pid_limits(struct task_struct *task, char *buffer)
+{
+       unsigned int i;
+       int count = 0;
+       unsigned long flags;
+       char *bufptr = buffer;
+
+       struct rlimit rlim[RLIM_NLIMITS];
+
+       rcu_read_lock();
+       if (!lock_task_sighand(task,&flags)) {
+               rcu_read_unlock();
+               return 0;
+       }
+       memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
+       unlock_task_sighand(task, &flags);
+       rcu_read_unlock();
+
+       /*
+        * print the file header
+        */
+       count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
+                       "Limit", "Soft Limit", "Hard Limit", "Units");
+
+       for (i = 0; i < RLIM_NLIMITS; i++) {
+               if (rlim[i].rlim_cur == RLIM_INFINITY)
+                       count += sprintf(&bufptr[count], "%-25s %-20s ",
+                                        lnames[i].name, "unlimited");
+               else
+                       count += sprintf(&bufptr[count], "%-25s %-20lu ",
+                                        lnames[i].name, rlim[i].rlim_cur);
+
+               if (rlim[i].rlim_max == RLIM_INFINITY)
+                       count += sprintf(&bufptr[count], "%-20s ", "unlimited");
+               else
+                       count += sprintf(&bufptr[count], "%-20lu ",
+                                        rlim[i].rlim_max);
+
+               if (lnames[i].unit)
+                       count += sprintf(&bufptr[count], "%-10s\n",
+                                        lnames[i].unit);
+               else
+                       count += sprintf(&bufptr[count], "\n");
+       }
+
+       return count;
+}
+
 /************************************************************************/
 /*                       Here the fs part begins                        */
 /************************************************************************/
@@ -342,7 +506,7 @@ static const struct inode_operations proc_def_inode_operations = {
        .setattr        = proc_setattr,
 };
 
-extern struct seq_operations mounts_op;
+extern const struct seq_operations mounts_op;
 struct proc_mounts {
        struct seq_file m;
        int event;
@@ -421,7 +585,7 @@ static const struct file_operations proc_mounts_operations = {
        .poll           = mounts_poll,
 };
 
-extern struct seq_operations mountstats_op;
+extern const struct seq_operations mountstats_op;
 static int mountstats_open(struct inode *inode, struct file *file)
 {
        int ret = seq_open(file, &mountstats_op);
@@ -498,6 +662,45 @@ static const struct file_operations proc_info_file_operations = {
        .read           = proc_info_read,
 };
 
+static int proc_single_show(struct seq_file *m, void *v)
+{
+       struct inode *inode = m->private;
+       struct pid_namespace *ns;
+       struct pid *pid;
+       struct task_struct *task;
+       int ret;
+
+       ns = inode->i_sb->s_fs_info;
+       pid = proc_pid(inode);
+       task = get_pid_task(pid, PIDTYPE_PID);
+       if (!task)
+               return -ESRCH;
+
+       ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
+
+       put_task_struct(task);
+       return ret;
+}
+
+static int proc_single_open(struct inode *inode, struct file *filp)
+{
+       int ret;
+       ret = single_open(filp, proc_single_show, NULL);
+       if (!ret) {
+               struct seq_file *m = filp->private_data;
+
+               m->private = inode;
+       }
+       return ret;
+}
+
+static const struct file_operations proc_single_file_operations = {
+       .open           = proc_single_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
 static int mem_open(struct inode* inode, struct file* file)
 {
        file->private_data = (void*)((long)current->self_exec_id);
@@ -623,7 +826,7 @@ 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:
@@ -771,42 +974,6 @@ 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,
@@ -820,7 +987,7 @@ 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);
 }
@@ -927,6 +1094,7 @@ static const struct file_operations proc_fault_inject_operations = {
 };
 #endif
 
+
 #ifdef CONFIG_SCHED_DEBUG
 /*
  * Print out various scheduling related per-task fields:
@@ -996,39 +1164,36 @@ static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
        int error = -EACCES;
 
        /* We don't need a base pointer in the /proc filesystem */
-       path_release(nd);
+       path_put(&nd->path);
 
        /* Are we allowed to snoop on the tasks file descriptors? */
        if (!proc_fd_access_allowed(inode))
                goto out;
 
-       error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt);
+       error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
        nd->last_type = LAST_BIND;
 out:
        return ERR_PTR(error);
 }
 
-static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt,
-                           char __user *buffer, int buflen)
+static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
 {
-       struct inode * inode;
        char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
-       char *path;
+       char *pathname;
        int len;
 
        if (!tmp)
                return -ENOMEM;
 
-       inode = dentry->d_inode;
-       path = d_path(dentry, mnt, tmp, PAGE_SIZE);
-       len = PTR_ERR(path);
-       if (IS_ERR(path))
+       pathname = d_path(path, tmp, PAGE_SIZE);
+       len = PTR_ERR(pathname);
+       if (IS_ERR(pathname))
                goto out;
-       len = tmp + PAGE_SIZE - 1 - path;
+       len = tmp + PAGE_SIZE - 1 - pathname;
 
        if (len > buflen)
                len = buflen;
-       if (copy_to_user(buffer, path, len))
+       if (copy_to_user(buffer, pathname, len))
                len = -EFAULT;
  out:
        free_page((unsigned long)tmp);
@@ -1039,20 +1204,18 @@ static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int b
 {
        int error = -EACCES;
        struct inode *inode = dentry->d_inode;
-       struct dentry *de;
-       struct vfsmount *mnt = NULL;
+       struct path path;
 
        /* Are we allowed to snoop on the tasks file descriptors? */
        if (!proc_fd_access_allowed(inode))
                goto out;
 
-       error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt);
+       error = PROC_I(inode)->op.proc_get_link(inode, &path);
        if (error)
                goto out;
 
-       error = do_proc_readlink(de, mnt, buffer, buflen);
-       dput(de);
-       mntput(mnt);
+       error = do_proc_readlink(&path, buffer, buflen);
+       path_put(&path);
 out:
        return error;
 }
@@ -1279,8 +1442,7 @@ out:
 
 #define PROC_FDINFO_MAX 64
 
-static int proc_fd_info(struct inode *inode, struct dentry **dentry,
-                       struct vfsmount **mnt, char *info)
+static int proc_fd_info(struct inode *inode, struct path *path, char *info)
 {
        struct task_struct *task = get_proc_task(inode);
        struct files_struct *files = NULL;
@@ -1299,10 +1461,10 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
                spin_lock(&files->file_lock);
                file = fcheck_files(files, fd);
                if (file) {
-                       if (mnt)
-                               *mnt = mntget(file->f_path.mnt);
-                       if (dentry)
-                               *dentry = dget(file->f_path.dentry);
+                       if (path) {
+                               *path = file->f_path;
+                               path_get(&file->f_path);
+                       }
                        if (info)
                                snprintf(info, PROC_FDINFO_MAX,
                                         "pos:\t%lli\n"
@@ -1319,10 +1481,9 @@ static int proc_fd_info(struct inode *inode, struct dentry **dentry,
        return -ENOENT;
 }
 
-static int proc_fd_link(struct inode *inode, struct dentry **dentry,
-                       struct vfsmount **mnt)
+static int proc_fd_link(struct inode *inode, struct path *path)
 {
-       return proc_fd_info(inode, dentry, mnt, NULL);
+       return proc_fd_info(inode, path, NULL);
 }
 
 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
@@ -1446,7 +1607,7 @@ static int proc_readfd_common(struct file * filp, void * dirent,
        struct dentry *dentry = filp->f_path.dentry;
        struct inode *inode = dentry->d_inode;
        struct task_struct *p = get_proc_task(inode);
-       unsigned int fd, tid, ino;
+       unsigned int fd, ino;
        int retval;
        struct files_struct * files;
        struct fdtable *fdt;
@@ -1455,7 +1616,6 @@ static int proc_readfd_common(struct file * filp, void * dirent,
        if (!p)
                goto out_no_task;
        retval = 0;
-       tid = p->pid;
 
        fd = filp->f_pos;
        switch (fd) {
@@ -1517,7 +1677,7 @@ static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
                                      size_t len, loff_t *ppos)
 {
        char tmp[PROC_FDINFO_MAX];
-       int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, NULL, tmp);
+       int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
        if (!err)
                err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
        return err;
@@ -1690,7 +1850,6 @@ static int proc_pident_readdir(struct file *filp,
                const struct pid_entry *ents, unsigned int nents)
 {
        int i;
-       int pid;
        struct dentry *dentry = filp->f_path.dentry;
        struct inode *inode = dentry->d_inode;
        struct task_struct *task = get_proc_task(inode);
@@ -1703,7 +1862,6 @@ static int proc_pident_readdir(struct file *filp,
                goto out_no_task;
 
        ret = 0;
-       pid = task->pid;
        i = filp->f_pos;
        switch (i) {
        case 0:
@@ -1936,15 +2094,23 @@ static const struct file_operations proc_coredump_filter_operations = {
 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
                              int buflen)
 {
+       struct pid_namespace *ns = dentry->d_sb->s_fs_info;
+       pid_t tgid = task_tgid_nr_ns(current, ns);
        char tmp[PROC_NUMBUF];
-       sprintf(tmp, "%d", task_tgid_vnr(current));
+       if (!tgid)
+               return -ENOENT;
+       sprintf(tmp, "%d", tgid);
        return vfs_readlink(dentry,buffer,buflen,tmp);
 }
 
 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
+       struct pid_namespace *ns = dentry->d_sb->s_fs_info;
+       pid_t tgid = task_tgid_nr_ns(current, ns);
        char tmp[PROC_NUMBUF];
-       sprintf(tmp, "%d", task_tgid_vnr(current));
+       if (!tgid)
+               return ERR_PTR(-ENOENT);
+       sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
        return ERR_PTR(vfs_follow_link(nd,tmp));
 }
 
@@ -2109,13 +2275,14 @@ static const struct pid_entry tgid_base_stuff[] = {
        DIR("fdinfo",     S_IRUSR|S_IXUSR, fdinfo),
        REG("environ",    S_IRUSR, environ),
        INF("auxv",       S_IRUSR, pid_auxv),
-       INF("status",     S_IRUGO, pid_status),
+       ONE("status",     S_IRUGO, pid_status),
+       INF("limits",     S_IRUSR, pid_limits),
 #ifdef CONFIG_SCHED_DEBUG
        REG("sched",      S_IRUGO|S_IWUSR, pid_sched),
 #endif
        INF("cmdline",    S_IRUGO, pid_cmdline),
-       INF("stat",       S_IRUGO, tgid_stat),
-       INF("statm",      S_IRUGO, pid_statm),
+       ONE("stat",       S_IRUGO, tgid_stat),
+       ONE("statm",      S_IRUGO, pid_statm),
        REG("maps",       S_IRUGO, maps),
 #ifdef CONFIG_NUMA
        REG("numa_maps",  S_IRUGO, numa_maps),
@@ -2126,9 +2293,10 @@ static const struct pid_entry tgid_base_stuff[] = {
        LNK("exe",        exe),
        REG("mounts",     S_IRUGO, mounts),
        REG("mountstats", S_IRUSR, mountstats),
-#ifdef CONFIG_MMU
+#ifdef CONFIG_PROC_PAGE_MONITOR
        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),
@@ -2139,6 +2307,9 @@ 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
@@ -2184,27 +2355,6 @@ 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;
@@ -2215,7 +2365,8 @@ static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
        name.len = snprintf(buf, sizeof(buf), "%d", pid);
        dentry = d_hash_and_lookup(mnt->mnt_root, &name);
        if (dentry) {
-               shrink_dcache_parent(dentry);
+               if (!(current->flags & PF_EXITING))
+                       shrink_dcache_parent(dentry);
                d_drop(dentry);
                dput(dentry);
        }
@@ -2251,29 +2402,45 @@ 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];
@@ -2345,19 +2512,23 @@ 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
@@ -2370,23 +2541,25 @@ 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 */
@@ -2394,8 +2567,7 @@ 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)
@@ -2408,14 +2580,14 @@ 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;
                }
        }
@@ -2434,13 +2606,14 @@ static const struct pid_entry tid_base_stuff[] = {
        DIR("fdinfo",    S_IRUSR|S_IXUSR, fdinfo),
        REG("environ",   S_IRUSR, environ),
        INF("auxv",      S_IRUSR, pid_auxv),
-       INF("status",    S_IRUGO, pid_status),
+       ONE("status",    S_IRUGO, pid_status),
+       INF("limits",    S_IRUSR, pid_limits),
 #ifdef CONFIG_SCHED_DEBUG
        REG("sched",     S_IRUGO|S_IWUSR, pid_sched),
 #endif
        INF("cmdline",   S_IRUGO, pid_cmdline),
-       INF("stat",      S_IRUGO, tid_stat),
-       INF("statm",     S_IRUGO, pid_statm),
+       ONE("stat",      S_IRUGO, tid_stat),
+       ONE("statm",     S_IRUGO, pid_statm),
        REG("maps",      S_IRUGO, maps),
 #ifdef CONFIG_NUMA
        REG("numa_maps", S_IRUGO, numa_maps),
@@ -2450,9 +2623,10 @@ static const struct pid_entry tid_base_stuff[] = {
        LNK("root",      root),
        LNK("exe",       exe),
        REG("mounts",    S_IRUGO, mounts),
-#ifdef CONFIG_MMU
+#ifdef CONFIG_PROC_PAGE_MONITOR
        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),
@@ -2463,6 +2637,9 @@ 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
@@ -2553,7 +2730,7 @@ static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry
        rcu_read_unlock();
        if (!task)
                goto out;
-       if (leader->tgid != task->tgid)
+       if (!same_thread_group(leader, task))
                goto out_drop_task;
 
        result = proc_task_instantiate(dir, dentry, task, NULL);