Skip to contentSkip to conversation

proc: move proc mount options out of pid_namespace

PATCH In review Stephen Mell 2 participants 8 messages May 23, 2013 – May 24, 2013 Proc filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 proc: move proc mount options out of pid_namespace

From: Stephen Mell <[email protected]>

hide_pid and pid_gid are proc mount options whose values are stored in the pid_namespace struct. As a result, if one mounts proc again for the same PID namespace with different mount options, all mounts for that PID namespace will be affected. This seems undesirable. This patch creates a proc_sb_info struct that points to the applicable PID namespace, and it moves hide_pid, pid_gid, and proc_self from pid_namespace to proc_sb_info.

---

fs/proc/base.c +24 -14
index dd51e50..853eb68 100644
@@ -584,13 +584,13 @@ int proc_setattr(struct dentry *dentry, struct iattr *attr)
  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
  * or euid/egid (for hide_pid_min=2)?
  */
-static bool has_pid_permissions(struct pid_namespace *pid,
+static bool has_pid_permissions(struct proc_sb_info *fsi,
 				 struct task_struct *task,
 				 int hide_pid_min)
 {
-	if (pid->hide_pid < hide_pid_min)
+	if (fsi->hide_pid < hide_pid_min)
 		return true;
-	if (in_group_p(pid->pid_gid))
+	if (in_group_p(fsi->pid_gid))
 		return true;
 	return ptrace_may_access(task, PTRACE_MODE_READ);
 }
@@ -598,18 +598,18 @@ static bool has_pid_permissions(struct pid_namespace *pid,
 
 static int proc_pid_permission(struct inode *inode, int mask)
 {
-	struct pid_namespace *pid = inode->i_sb->s_fs_info;
+	struct proc_sb_info *fsi = inode->i_sb->s_fs_info;
 	struct task_struct *task;
 	bool has_perms;
 
 	task = get_proc_task(inode);
 	if (!task)
 		return -ESRCH;
-	has_perms = has_pid_permissions(pid, task, 1);
+	has_perms = has_pid_permissions(fsi, task, 1);
 	put_task_struct(task);
 
 	if (!has_perms) {
-		if (pid->hide_pid == 2) {
+		if (fsi->hide_pid == 2) {
 			/*
 			 * Let's make getdents(), stat(), and open()
 			 * consistent with each other.  If a process
@@ -670,12 +670,14 @@ static const struct file_operations proc_info_file_operations = {
 static int proc_single_show(struct seq_file *m, void *v)
 {
 	struct inode *inode = m->private;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 	struct pid *pid;
 	struct task_struct *task;
 	int ret;
 
-	ns = inode->i_sb->s_fs_info;
+	fsi = inode->i_sb->s_fs_info;
+	ns = fsi->ns;
 	pid = proc_pid(inode);
 	task = get_pid_task(pid, PIDTYPE_PID);
 	if (!task)
@@ -1574,7 +1576,7 @@ int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 	struct inode *inode = dentry->d_inode;
 	struct task_struct *task;
 	const struct cred *cred;
-	struct pid_namespace *pid = dentry->d_sb->s_fs_info;
+	struct proc_sb_info *fsi = dentry->d_sb->s_fs_info;
 
 	generic_fillattr(inode, stat);
 
@@ -1583,7 +1585,7 @@ int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 	stat->gid = GLOBAL_ROOT_GID;
 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
 	if (task) {
-		if (!has_pid_permissions(pid, task, 2)) {
+		if (!has_pid_permissions(fsi, task, 2)) {
 			rcu_read_unlock();
 			/*
 			 * This doesn't prevent learning whether PID exists,
@@ -2869,13 +2871,15 @@ struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsign
 	struct dentry *result = NULL;
 	struct task_struct *task;
 	unsigned tgid;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
 	tgid = name_to_int(dentry);
 	if (tgid == ~0U)
 		goto out;
 
-	ns = dentry->d_sb->s_fs_info;
+	fsi = dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	rcu_read_lock();
 	task = find_task_by_pid_ns(tgid, ns);
 	if (task)
@@ -2954,6 +2958,7 @@ static int fake_filldir(void *buf, const char *name, int namelen,
 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
 {
 	struct tgid_iter iter;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 	filldir_t __filldir;
 	loff_t pos = filp->f_pos;
@@ -2970,11 +2975,12 @@ int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
 		iter.tgid = pos - TGID_OFFSET;
 	}
 	iter.task = NULL;
-	ns = filp->f_dentry->d_sb->s_fs_info;
+	fsi = filp->f_dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	for (iter = next_tgid(ns, iter);
 	     iter.task;
 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
-		if (has_pid_permissions(ns, iter.task, 2))
+		if (has_pid_permissions(fsi, iter.task, 2))
 			__filldir = filldir;
 		else
 			__filldir = fake_filldir;
@@ -3132,6 +3138,7 @@ static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry
 	struct task_struct *task;
 	struct task_struct *leader = get_proc_task(dir);
 	unsigned tid;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
 	if (!leader)
@@ -3141,7 +3148,8 @@ static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry
 	if (tid == ~0U)
 		goto out;
 
-	ns = dentry->d_sb->s_fs_info;
+	fsi = dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	rcu_read_lock();
 	task = find_task_by_pid_ns(tid, ns);
 	if (task)
@@ -3249,6 +3257,7 @@ static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldi
 	int retval = -ENOENT;
 	ino_t ino;
 	int tid;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
 	task = get_proc_task(inode);
@@ -3283,7 +3292,8 @@ static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldi
 	/* f_version caches the tgid value that the last readdir call couldn't
 	 * return. lseek aka telldir automagically resets f_version to 0.
 	 */
-	ns = filp->f_dentry->d_sb->s_fs_info;
+	fsi = filp->f_dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	tid = (int)filp->f_version;
 	filp->f_version = 0;
 	for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
fs/proc/inode.c +5 -5
index 073aea6..148737d 100644
@@ -110,12 +110,12 @@ void __init proc_init_inodecache(void)
 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 {
 	struct super_block *sb = root->d_sb;
-	struct pid_namespace *pid = sb->s_fs_info;
+	struct proc_sb_info *fsi = sb->s_fs_info;
 
-	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
-		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
-	if (pid->hide_pid != 0)
-		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
+	if (!gid_eq(fsi->pid_gid, GLOBAL_ROOT_GID))
+		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, fsi->pid_gid));
+	if (fsi->hide_pid != 0)
+		seq_printf(seq, ",hidepid=%u", fsi->hide_pid);
 
 	return 0;
 }
fs/proc/internal.h +8 -0
index d600fb0..ff970e4 100644
@@ -29,6 +29,14 @@ struct mempolicy;
  * /proc file has a parent, but "subdir" is NULL for all
  * non-directory entries).
  */
+
+struct proc_sb_info {
+	struct pid_namespace *ns;
+	struct dentry *proc_self;
+	kgid_t pid_gid;
+	int hide_pid;
+};
+
 struct proc_dir_entry {
 	unsigned int low_ino;
 	umode_t mode;
fs/proc/root.c +36 -25

122 lines not on this page · raw message

fs/proc/self.c +6 -4
index 6b6a993..ff4a2e2 100644
@@ -10,7 +10,8 @@
 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
 			      int buflen)
 {
-	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
+	struct proc_sb_info *fsi = dentry->d_sb->s_fs_info;
+	struct pid_namespace *ns = fsi->ns;
 	pid_t tgid = task_tgid_nr_ns(current, ns);
 	char tmp[PROC_NUMBUF];
 	if (!tgid)
@@ -21,7 +22,8 @@ static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
 
 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
-	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
+	struct proc_sb_info *fsi = dentry->d_sb->s_fs_info;
+	struct pid_namespace *ns = fsi->ns;
 	pid_t tgid = task_tgid_nr_ns(current, ns);
 	char *name = ERR_PTR(-ENOENT);
 	if (tgid) {
@@ -55,7 +57,7 @@ static unsigned self_inum;
 int proc_setup_self(struct super_block *s)
 {
 	struct inode *root_inode = s->s_root->d_inode;
-	struct pid_namespace *ns = s->s_fs_info;
+	struct proc_sb_info *fsi = s->s_fs_info;
 	struct dentry *self;
 	
 	mutex_lock(&root_inode->i_mutex);
@@ -82,7 +84,7 @@ int proc_setup_self(struct super_block *s)
 		pr_err("proc_fill_super: can't allocate /proc/self\n");
 		return PTR_ERR(self);
 	}
-	ns->proc_self = self;
+	fsi->proc_self = self;
 	return 0;
 }
 
include/linux/pid_namespace.h +0 -3
index e277266..6185386 100644
@@ -31,15 +31,12 @@ struct pid_namespace {
 	struct pid_namespace *parent;
 #ifdef CONFIG_PROC_FS
 	struct vfsmount *proc_mnt;
-	struct dentry *proc_self;
 #endif
 #ifdef CONFIG_BSD_PROCESS_ACCT
 	struct bsd_acct_struct *bacct;
 #endif
 	struct user_namespace *user_ns;
 	struct work_struct proc_work;
-	kgid_t pid_gid;
-	int hide_pid;
 	int reboot;	/* group exit code if this pidns was rebooted */
 	unsigned int proc_inum;
 };
Reviewed-by on v1 1/1
#1 lore raw

Hi Stephen,
Refer to following inline comments.
Thanks,
Gu

Quoted 5 lines · Stephen Mell

From: Stephen Mell <[email protected]>

hide_pid and pid_gid are proc mount options whose values are stored in the pid_namespace struct. As a result, if one mounts proc again for the same PID namespace with different mount options, all mounts for that PID namespace will be affected. This seems undesirable. This patch creates a proc_sb_info struct that points to the applicable PID namespace, and it moves hide_pid, pid_gid, and proc_self from pid_namespace to proc_sb_info.

Quoted 262 lines

---

fs/proc/base.c +24 -14
index dd51e50..853eb68 100644
@@ -584,13 +584,13 @@ int proc_setattr(struct dentry *dentry, struct iattr *attr)
  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
  * or euid/egid (for hide_pid_min=2)?
  */
-static bool has_pid_permissions(struct pid_namespace *pid,
+static bool has_pid_permissions(struct proc_sb_info *fsi,
 				 struct task_struct *task,
 				 int hide_pid_min)
 {
-	if (pid->hide_pid < hide_pid_min)
+	if (fsi->hide_pid < hide_pid_min)
 		return true;
-	if (in_group_p(pid->pid_gid))
+	if (in_group_p(fsi->pid_gid))
 		return true;
 	return ptrace_may_access(task, PTRACE_MODE_READ);
 }
@@ -598,18 +598,18 @@ static bool has_pid_permissions(struct pid_namespace *pid,
 
 static int proc_pid_permission(struct inode *inode, int mask)
 {
-	struct pid_namespace *pid = inode->i_sb->s_fs_info;
+	struct proc_sb_info *fsi = inode->i_sb->s_fs_info;
 	struct task_struct *task;
 	bool has_perms;
 
 	task = get_proc_task(inode);
 	if (!task)
 		return -ESRCH;
-	has_perms = has_pid_permissions(pid, task, 1);
+	has_perms = has_pid_permissions(fsi, task, 1);
 	put_task_struct(task);
 
 	if (!has_perms) {
-		if (pid->hide_pid == 2) {
+		if (fsi->hide_pid == 2) {
 			/*
 			 * Let's make getdents(), stat(), and open()
 			 * consistent with each other.  If a process
@@ -670,12 +670,14 @@ static const struct file_operations proc_info_file_operations = {
 static int proc_single_show(struct seq_file *m, void *v)
 {
 	struct inode *inode = m->private;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 	struct pid *pid;
 	struct task_struct *task;
 	int ret;
 
-	ns = inode->i_sb->s_fs_info;
+	fsi = inode->i_sb->s_fs_info;
+	ns = fsi->ns;
 	pid = proc_pid(inode);
 	task = get_pid_task(pid, PIDTYPE_PID);
 	if (!task)
@@ -1574,7 +1576,7 @@ int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 	struct inode *inode = dentry->d_inode;
 	struct task_struct *task;
 	const struct cred *cred;
-	struct pid_namespace *pid = dentry->d_sb->s_fs_info;
+	struct proc_sb_info *fsi = dentry->d_sb->s_fs_info;
 
 	generic_fillattr(inode, stat);
 
@@ -1583,7 +1585,7 @@ int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 	stat->gid = GLOBAL_ROOT_GID;
 	task = pid_task(proc_pid(inode), PIDTYPE_PID);
 	if (task) {
-		if (!has_pid_permissions(pid, task, 2)) {
+		if (!has_pid_permissions(fsi, task, 2)) {
 			rcu_read_unlock();
 			/*
 			 * This doesn't prevent learning whether PID exists,
@@ -2869,13 +2871,15 @@ struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsign
 	struct dentry *result = NULL;
 	struct task_struct *task;
 	unsigned tgid;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
 	tgid = name_to_int(dentry);
 	if (tgid == ~0U)
 		goto out;
 
-	ns = dentry->d_sb->s_fs_info;
+	fsi = dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	rcu_read_lock();
 	task = find_task_by_pid_ns(tgid, ns);
 	if (task)
@@ -2954,6 +2958,7 @@ static int fake_filldir(void *buf, const char *name, int namelen,
 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
 {
 	struct tgid_iter iter;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 	filldir_t __filldir;
 	loff_t pos = filp->f_pos;
@@ -2970,11 +2975,12 @@ int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
 		iter.tgid = pos - TGID_OFFSET;
 	}
 	iter.task = NULL;
-	ns = filp->f_dentry->d_sb->s_fs_info;
+	fsi = filp->f_dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	for (iter = next_tgid(ns, iter);
 	     iter.task;
 	     iter.tgid += 1, iter = next_tgid(ns, iter)) {
-		if (has_pid_permissions(ns, iter.task, 2))
+		if (has_pid_permissions(fsi, iter.task, 2))
 			__filldir = filldir;
 		else
 			__filldir = fake_filldir;
@@ -3132,6 +3138,7 @@ static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry
 	struct task_struct *task;
 	struct task_struct *leader = get_proc_task(dir);
 	unsigned tid;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
 	if (!leader)
@@ -3141,7 +3148,8 @@ static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry
 	if (tid == ~0U)
 		goto out;
 
-	ns = dentry->d_sb->s_fs_info;
+	fsi = dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	rcu_read_lock();
 	task = find_task_by_pid_ns(tid, ns);
 	if (task)
@@ -3249,6 +3257,7 @@ static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldi
 	int retval = -ENOENT;
 	ino_t ino;
 	int tid;
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
 	task = get_proc_task(inode);
@@ -3283,7 +3292,8 @@ static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldi
 	/* f_version caches the tgid value that the last readdir call couldn't
 	 * return. lseek aka telldir automagically resets f_version to 0.
 	 */
-	ns = filp->f_dentry->d_sb->s_fs_info;
+	fsi = filp->f_dentry->d_sb->s_fs_info;
+	ns = fsi->ns;
 	tid = (int)filp->f_version;
 	filp->f_version = 0;
 	for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
fs/proc/inode.c +5 -5
index 073aea6..148737d 100644
@@ -110,12 +110,12 @@ void __init proc_init_inodecache(void)
 static int proc_show_options(struct seq_file *seq, struct dentry *root)
 {
 	struct super_block *sb = root->d_sb;
-	struct pid_namespace *pid = sb->s_fs_info;
+	struct proc_sb_info *fsi = sb->s_fs_info;
 
-	if (!gid_eq(pid->pid_gid, GLOBAL_ROOT_GID))
-		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, pid->pid_gid));
-	if (pid->hide_pid != 0)
-		seq_printf(seq, ",hidepid=%u", pid->hide_pid);
+	if (!gid_eq(fsi->pid_gid, GLOBAL_ROOT_GID))
+		seq_printf(seq, ",gid=%u", from_kgid_munged(&init_user_ns, fsi->pid_gid));
+	if (fsi->hide_pid != 0)
+		seq_printf(seq, ",hidepid=%u", fsi->hide_pid);
 
 	return 0;
 }
fs/proc/internal.h +8 -0
index d600fb0..ff970e4 100644
@@ -29,6 +29,14 @@ struct mempolicy;
  * /proc file has a parent, but "subdir" is NULL for all
  * non-directory entries).
  */
+
+struct proc_sb_info {
+	struct pid_namespace *ns;
+	struct dentry *proc_self;
+	kgid_t pid_gid;
+	int hide_pid;
+};
+
 struct proc_dir_entry {
 	unsigned int low_ino;
 	umode_t mode;
fs/proc/root.c +9 -9
index 41a6ea9..526aec6 100644
@@ -20,6 +20,7 @@
 #include <linux/mount.h>
 #include <linux/pid_namespace.h>
 #include <linux/parser.h>
+#include <linux/slab.h>
 
 #include "internal.h"
 
@@ -31,10 +32,9 @@ static int proc_test_super(struct super_block *sb, void *data)
 static int proc_set_super(struct super_block *sb, void *data)
 {
 	int err = set_anon_super(sb, NULL);
-	if (!err) {
-		struct pid_namespace *ns = (struct pid_namespace *)data;
-		sb->s_fs_info = get_pid_ns(ns);
-	}
+	if (!err)
+		sb->s_fs_info = (struct proc_sb_info *)data;
+
 	return err;
 }
 
@@ -48,7 +48,7 @@ static const match_table_t tokens = {
 	{Opt_err, NULL},
 };
 
-static int proc_parse_options(char *options, struct pid_namespace *pid)
+static int proc_parse_options(char *options, struct proc_sb_info *fsi)
 {
 	char *p;
 	substring_t args[MAX_OPT_ARGS];
@@ -68,7 +68,7 @@ static int proc_parse_options(char *options, struct pid_namespace *pid)
 		case Opt_gid:
 			if (match_int(&args[0], &option))
 				return 0;
-			pid->pid_gid = make_kgid(current_user_ns(), option);
+			fsi->pid_gid = make_kgid(current_user_ns(), option);
 			break;
 		case Opt_hidepid:
 			if (match_int(&args[0], &option))
@@ -77,7 +77,7 @@ static int proc_parse_options(char *options, struct pid_namespace *pid)
 				pr_err("proc: hidepid value must be between 0 and 2.\n");
 				return 0;
 			}
-			pid->hide_pid = option;
+			fsi->hide_pid = option;
 			break;
 		default:
 			pr_err("proc: unrecognized mount option \"%s\" "
@@ -91,36 +91,44 @@ static int proc_parse_options(char *options, struct pid_namespace *pid)
 
 int proc_remount(struct super_block *sb, int *flags, char *data)
 {
-	struct pid_namespace *pid = sb->s_fs_info;
-	return !proc_parse_options(data, pid);
+	struct proc_sb_info *fsi = sb->s_fs_info;
+	return !proc_parse_options(data, fsi);
 }
 
 static struct dentry *proc_mount(struct file_system_type *fs_type,
 	int flags, const char *dev_name, void *data)
 {
 	int err;

fs/proc/root.c static int proc_parse_options(char *options, struct pid_namespace *pid) · line 110

Quoted 10 lines
fs/proc/root.c +7 -1
+	struct proc_sb_info *fsi;
 	struct super_block *sb;
-	struct pid_namespace *ns;
 	char *options;
+	
+	fsi = kzalloc(sizeof(struct proc_sb_info), GFP_KERNEL);
+	if (!fsi)
+		return ERR_PTR(-ENOMEM);
+
+	sb = sget(fs_type, proc_test_super, proc_set_super, flags, fsi);

Here it'll create a new proc sb instance which holds the same context as the old ones each time we mount proc though in the same PID namespace, won't it?

Quoted 39 lines
fs/proc/root.c +14 -12
+	if (IS_ERR(sb))
+		return ERR_CAST(sb);
 
 	if (flags & MS_KERNMOUNT) {
-		ns = (struct pid_namespace *)data;
+		fsi->ns = (struct pid_namespace *)data;
+		get_pid_ns(fsi->ns);
 		options = NULL;
 	} else {
-		ns = task_active_pid_ns(current);
-		options = data;
-
 		if (!current_user_ns()->may_mount_proc)
 			return ERR_PTR(-EPERM);
-	}
-
-	sb = sget(fs_type, proc_test_super, proc_set_super, flags, ns);
-	if (IS_ERR(sb))
-		return ERR_CAST(sb);
+		
+		options = data;
+		if (!proc_parse_options(options, fsi)) {
+			deactivate_locked_super(sb);
+			return ERR_PTR(-EINVAL);
+		}
 
-	if (!proc_parse_options(options, ns)) {
-		deactivate_locked_super(sb);
-		return ERR_PTR(-EINVAL);
+		if (!fsi->ns) {
+			fsi->ns = task_active_pid_ns(current);
+			get_pid_ns(fsi->ns);
+		}
 	}
 
 	if (!sb->s_root) {
@@ -138,13 +146,16 @@ static struct dentry *proc_mount(struct file_system_type *fs_type,
 
 static void proc_kill_sb(struct super_block *sb)

fs/proc/root.c static struct dentry *proc_mount(struct file_system_type *fs_type, · line 154

Quoted 10 lines
fs/proc/root.c +4 -3
 {
+	struct proc_sb_info *fsi;
 	struct pid_namespace *ns;
 
-	ns = (struct pid_namespace *)sb->s_fs_info;
-	if (ns->proc_self)
-		dput(ns->proc_self);
+	fsi = (struct proc_sb_info *)sb->s_fs_info;
+	ns = (struct pid_namespace *)fsi->ns;
+	if (fsi->proc_self)

Here the pre-check seems needless.

Quoted 75 lines
fs/proc/root.c +2 -0
+		dput(fsi->proc_self);
 	kill_anon_super(sb);
 	put_pid_ns(ns);
+	kfree(fsi);
 }
 
 static struct file_system_type proc_fs_type = {
fs/proc/self.c +6 -4
index 6b6a993..ff4a2e2 100644
@@ -10,7 +10,8 @@
 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
 			      int buflen)
 {
-	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
+	struct proc_sb_info *fsi = dentry->d_sb->s_fs_info;
+	struct pid_namespace *ns = fsi->ns;
 	pid_t tgid = task_tgid_nr_ns(current, ns);
 	char tmp[PROC_NUMBUF];
 	if (!tgid)
@@ -21,7 +22,8 @@ static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
 
 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
-	struct pid_namespace *ns = dentry->d_sb->s_fs_info;
+	struct proc_sb_info *fsi = dentry->d_sb->s_fs_info;
+	struct pid_namespace *ns = fsi->ns;
 	pid_t tgid = task_tgid_nr_ns(current, ns);
 	char *name = ERR_PTR(-ENOENT);
 	if (tgid) {
@@ -55,7 +57,7 @@ static unsigned self_inum;
 int proc_setup_self(struct super_block *s)
 {
 	struct inode *root_inode = s->s_root->d_inode;
-	struct pid_namespace *ns = s->s_fs_info;
+	struct proc_sb_info *fsi = s->s_fs_info;
 	struct dentry *self;
 
 	mutex_lock(&root_inode->i_mutex);
@@ -82,7 +84,7 @@ int proc_setup_self(struct super_block *s)
 		pr_err("proc_fill_super: can't allocate /proc/self\n");
 		return PTR_ERR(self);
 	}
-	ns->proc_self = self;
+	fsi->proc_self = self;
 	return 0;
 }
 
include/linux/pid_namespace.h +0 -4
index e277266..6185386 100644
@@ -31,15 +31,12 @@ struct pid_namespace {
 	struct pid_namespace *parent;
 #ifdef CONFIG_PROC_FS
 	struct vfsmount *proc_mnt;
-	struct dentry *proc_self;
 #endif
 #ifdef CONFIG_BSD_PROCESS_ACCT
 	struct bsd_acct_struct *bacct;
 #endif
 	struct user_namespace *user_ns;
 	struct work_struct proc_work;
-	kgid_t pid_gid;
-	int hide_pid;
 	int reboot;	/* group exit code if this pidns was rebooted */
 	unsigned int proc_inum;
 };
--

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#2 lore raw
Quoted 2 lines · On Thursday, May 23, 2013 18:20:57 Gu Zheng wrote:

Here it'll create a new proc sb instance which holds the same context as the old ones each time we mount proc though in the same PID namespace, won't it?

I believe so. But this is the point, right? They won't be identical if different mount options are used, I don't think.

Quoted 1 line

Here the pre-check seems needless.

Is that new with my patch, or has it always been needless?

Thanks,
Stephen

#3 lore raw

Hi Stephen,

Quoted 5 lines · Stephen Mell
Quoted 2 lines · On Thursday, May 23, 2013 18:20:57 Gu Zheng wrote:

Here it'll create a new proc sb instance which holds the same context as the old ones each time we mount proc though in the same PID namespace, won't it?

I believe so. But this is the point, right?

Yes, but I think it's also the problem.

Quoted 1 line

They won't be identical if different mount options are used, I don't think.

If different mount options are used, we'll create different super block instance, and they have the same context, only the difference is each one holds different proc_sb_info. But I think what we really want is only one proc sb instance and create different proc_sb_info if different mount options are used.

Quoted 3 lines
Quoted 2 lines

Here the pre-check seems needless.

Is that new with my patch, or has it always been needless?

Yeah, it's always needless.

Thanks,
Gu

Quoted 4 lines

Thanks,
Stephen

#4 lore raw

Gu,

Quoted 10 lines · On Friday, May 24, 2013 11:03:31 Gu Zheng wrote:

Hi Stephen,

Quoted 5 lines · Stephen Mell
Quoted 2 lines · On Thursday, May 23, 2013 18:20:57 Gu Zheng wrote:

Here it'll create a new proc sb instance which holds the same context as the old ones each time we mount proc though in the same PID namespace, won't it?

I believe so. But this is the point, right?

Quoted 8 lines

Yes, but I think it's also the problem.

Quoted 1 line

They won't be identical if different mount options are used, I don't think.

If different mount options are used, we'll create different super block instance, and they have the same context, only the difference is each one holds different proc_sb_info. But I think what we really want is only one proc sb instance and create different proc_sb_info if different mount options are used.

Will having several different superblocks cause problems, or is it merely inefficient? I freely admit to not really knowing what I'm doing, and I thank you for your assistance.
How is this situation distinct from that of ramfs? It appears to have a superblock for each mount.
It would seem to me as though one cannot have different sb_infos with the same superblock, making storing the mount options in sb_info effectively the same as storing them in the superblock itself, for the purposes of this discussion. Where would the mount options be stored, if not in the superblock?

Quoted 12 lines
Quoted 3 lines
Quoted 2 lines

Here the pre-check seems needless.

Is that new with my patch, or has it always been needless?

Yeah, it's always needless.

Thanks,
Gu

Quoted 3 lines

Thanks,
Stephen

Thanks again,
Stephen

#5 lore raw

Hi Stephen,

Quoted 15 lines · Stephen Mell

Gu,

Quoted 12 lines · On Friday, May 24, 2013 11:03:31 Gu Zheng wrote:

Hi Stephen,

Quoted 5 lines · Stephen Mell
Quoted 2 lines · On Thursday, May 23, 2013 18:20:57 Gu Zheng wrote:

Here it'll create a new proc sb instance which holds the same context as the old ones each time we mount proc though in the same PID namespace, won't it?

I believe so. But this is the point, right?

Yes, but I think it's also the problem.

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

They won't be identical if different mount options are used, I don't think.

If different mount options are used, we'll create different super block instance, and they have the same context, only the difference is each one holds different proc_sb_info. But I think what we really want is only one proc sb instance and create different proc_sb_info if different mount options are used.

Will having several different superblocks cause problems, or is it merely inefficient?

Not only inefficient.

Quoted 2 lines

I freely admit to not really knowing what I'm doing, and I thank you for your assistance. How is this situation distinct from that of ramfs? It appears to have a superblock for each mount.

They're different, each ramfs uses a split superblock to manage it's infos, e.g. the usage, and each ramfs mount instance should not affect others. But proc is a common interface, each mount instance holds the same context, so there's no need to create a new superblock instance each time mount if we in the same PID namespace.

Quoted 2 lines

It would seem to me as though one cannot have different sb_infos with the same superblock, making storing the mount options in sb_info effectively the same as storing them in the superblock itself, for the purposes of this discussion. Where would the mount options be stored, if not in the superblock?

One fuzzy way in my mind, I'm not sure whether it's OK, but we can discuss it.
Split hide_pid, pid_gid, and proc_self from pid_namespace, and create struct proc_sb_info(maybe the name "proc_mount_info" is better). And create a new list domain in the pid_namespace to contain the proc_sb_info instances. Each time we mount proc in a new directory only create a new proc_sb_info instance, and added it to the list in pid_namespace.
But this leads to another problem, how to get the right proc_sb_info instance in proc permission checking routine, do you have any idea? what do you think of this way?

Thanks,
Gu

Quoted 21 lines
Quoted 12 lines
Quoted 3 lines
Quoted 2 lines

Here the pre-check seems needless.

Is that new with my patch, or has it always been needless?

Yeah, it's always needless.

Thanks,
Gu

Quoted 3 lines

Thanks,
Stephen

Thanks again,
Stephen
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#6 lore raw
Quoted 6 lines · On Friday, May 24, 2013 17:14:13 Gu Zheng wrote:

One fuzzy way in my mind, I'm not sure whether it's OK, but we can discuss it.
Split hide_pid, pid_gid, and proc_self from pid_namespace, and create struct proc_sb_info(maybe the name "proc_mount_info" is better). And create a new list domain in the pid_namespace to contain the proc_sb_info instances. Each time we mount proc in a new directory only create a new proc_sb_info instance, and added it to the list in pid_namespace.
But this leads to another problem, how to get the right proc_sb_info instance in proc permission checking routine, do you have any idea? what do you think of this way?

I understand now what you're getting at.
However, I'd like to make the argument that it is desirable to have each proc mount, even those that reference the same namespace, have a different superblock. One option that I intend to add is the ability to specify which pid namespace proc is mounted for, rather than forcing the current one. If this were to be added, it wouldn't work to share superblocks, as that would prevent remounts of proc from switching which namespace that the refer to (or that is my understanding, which may be incorrect). Considering the small number of proc mounts that tend to be around anyway, perhaps this outweighs the drawbacks.
Let me know what you think.

Quoted 2 lines

Thanks,
Gu

Thanks,
Stephen

#7 lore raw
Quoted 9 lines · Stephen Mell
Quoted 6 lines · On Friday, May 24, 2013 17:14:13 Gu Zheng wrote:

One fuzzy way in my mind, I'm not sure whether it's OK, but we can discuss it.
Split hide_pid, pid_gid, and proc_self from pid_namespace, and create struct proc_sb_info(maybe the name "proc_mount_info" is better). And create a new list domain in the pid_namespace to contain the proc_sb_info instances. Each time we mount proc in a new directory only create a new proc_sb_info instance, and added it to the list in pid_namespace.
But this leads to another problem, how to get the right proc_sb_info instance in proc permission checking routine, do you have any idea? what do you think of this way?

Quoted 4 lines

I understand now what you're getting at.
However, I'd like to make the argument that it is desirable to have each proc mount, even those that reference the same namespace, have a different superblock. One option that I intend to add is the ability to specify which pid namespace proc is mounted for, rather than forcing the current one. If this were to be added, it wouldn't work to share superblocks, as that would prevent remounts of proc from switching which namespace that the refer to (or that is my understanding, which may be incorrect). Considering the small number of proc mounts that tend to be around anyway, perhaps this outweighs the drawbacks.
Let me know what you think.

It doesn't matter, if we can not find a better way, the current one is best, go ahead!:)

Thanks,
Gu

Quoted 12 lines
Quoted 3 lines

Thanks,
Gu

Thanks,
Stephen
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/