linux_dsm_epyc7002/fs/orangefs/xattr.c
Jann Horn 78fee0b684 orangefs: fix namespace handling
In orangefs_inode_getxattr(), an fsuid is written to dmesg. The kuid is
converted to a userspace uid via from_kuid(current_user_ns(), [...]), but
since dmesg is global, init_user_ns should be used here instead.

In copy_attributes_from_inode(), op_alloc() and fill_default_sys_attrs(),
upcall structures are populated with uids/gids that have been mapped into
the caller's namespace. However, those upcall structures are read by
another process (the userspace filesystem driver), and that process might
be running in another namespace. This effectively lets any user spoof its
uid and gid as seen by the userspace filesystem driver.

To fix the second issue, I just construct the opcall structures with
init_user_ns uids/gids and require the filesystem server to run in the
init namespace. Since orangefs is full of global state anyway (as the error
message in DUMP_DEVICE_ERROR explains, there can only be one userspace
orangefs filesystem driver at once), that shouldn't be a problem.

[
Why does orangefs even exist in the kernel if everything does upcalls into
userspace? What does orangefs do that couldn't be done with the FUSE
interface? If there is no good answer to those questions, I'd prefer to see
orangefs kicked out of the kernel. Can that be done for something that
shipped in a release?

According to commit f7ab093f74 ("Orangefs: kernel client part 1"), they
even already have a FUSE daemon, and the only rational reason (apart from
"but most of our users report preferring to use our kernel module instead")
given for not wanting to use FUSE is one "in-the-works" feature that could
probably be integated into FUSE instead.
]

This patch has been compile-tested.

Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
2016-07-05 15:47:43 -04:00

464 lines
12 KiB
C

/*
* (C) 2001 Clemson University and The University of Chicago
*
* See COPYING in top-level directory.
*/
/*
* Linux VFS extended attribute operations.
*/
#include "protocol.h"
#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"
#include <linux/posix_acl_xattr.h>
#include <linux/xattr.h>
#define SYSTEM_ORANGEFS_KEY "system.pvfs2."
#define SYSTEM_ORANGEFS_KEY_LEN 13
/*
* this function returns
* 0 if the key corresponding to name is not meant to be printed as part
* of a listxattr.
* 1 if the key corresponding to name is meant to be returned as part of
* a listxattr.
* The ones that start SYSTEM_ORANGEFS_KEY are the ones to avoid printing.
*/
static int is_reserved_key(const char *key, size_t size)
{
if (size < SYSTEM_ORANGEFS_KEY_LEN)
return 1;
return strncmp(key, SYSTEM_ORANGEFS_KEY, SYSTEM_ORANGEFS_KEY_LEN) ? 1 : 0;
}
static inline int convert_to_internal_xattr_flags(int setxattr_flags)
{
int internal_flag = 0;
if (setxattr_flags & XATTR_REPLACE) {
/* Attribute must exist! */
internal_flag = ORANGEFS_XATTR_REPLACE;
} else if (setxattr_flags & XATTR_CREATE) {
/* Attribute must not exist */
internal_flag = ORANGEFS_XATTR_CREATE;
}
return internal_flag;
}
/*
* Tries to get a specified key's attributes of a given
* file into a user-specified buffer. Note that the getxattr
* interface allows for the users to probe the size of an
* extended attribute by passing in a value of 0 to size.
* Thus our return value is always the size of the attribute
* unless the key does not exist for the file and/or if
* there were errors in fetching the attribute value.
*/
ssize_t orangefs_inode_getxattr(struct inode *inode, const char *name,
void *buffer, size_t size)
{
struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
struct orangefs_kernel_op_s *new_op = NULL;
ssize_t ret = -ENOMEM;
ssize_t length = 0;
int fsuid;
int fsgid;
gossip_debug(GOSSIP_XATTR_DEBUG,
"%s: name %s, buffer_size %zd\n",
__func__, name, size);
if (strlen(name) >= ORANGEFS_MAX_XATTR_NAMELEN) {
gossip_err("Invalid key length (%d)\n",
(int)strlen(name));
return -EINVAL;
}
fsuid = from_kuid(&init_user_ns, current_fsuid());
fsgid = from_kgid(&init_user_ns, current_fsgid());
gossip_debug(GOSSIP_XATTR_DEBUG,
"getxattr on inode %pU, name %s "
"(uid %o, gid %o)\n",
get_khandle_from_ino(inode),
name,
fsuid,
fsgid);
down_read(&orangefs_inode->xattr_sem);
new_op = op_alloc(ORANGEFS_VFS_OP_GETXATTR);
if (!new_op)
goto out_unlock;
new_op->upcall.req.getxattr.refn = orangefs_inode->refn;
strcpy(new_op->upcall.req.getxattr.key, name);
/*
* NOTE: Although keys are meant to be NULL terminated textual
* strings, I am going to explicitly pass the length just in case
* we change this later on...
*/
new_op->upcall.req.getxattr.key_sz = strlen(name) + 1;
ret = service_operation(new_op, "orangefs_inode_getxattr",
get_interruptible_flag(inode));
if (ret != 0) {
if (ret == -ENOENT) {
ret = -ENODATA;
gossip_debug(GOSSIP_XATTR_DEBUG,
"orangefs_inode_getxattr: inode %pU key %s"
" does not exist!\n",
get_khandle_from_ino(inode),
(char *)new_op->upcall.req.getxattr.key);
}
goto out_release_op;
}
/*
* Length returned includes null terminator.
*/
length = new_op->downcall.resp.getxattr.val_sz;
/*
* Just return the length of the queried attribute.
*/
if (size == 0) {
ret = length;
goto out_release_op;
}
/*
* Check to see if key length is > provided buffer size.
*/
if (length > size) {
ret = -ERANGE;
goto out_release_op;
}
memcpy(buffer, new_op->downcall.resp.getxattr.val, length);
memset(buffer + length, 0, size - length);
gossip_debug(GOSSIP_XATTR_DEBUG,
"orangefs_inode_getxattr: inode %pU "
"key %s key_sz %d, val_len %d\n",
get_khandle_from_ino(inode),
(char *)new_op->
upcall.req.getxattr.key,
(int)new_op->
upcall.req.getxattr.key_sz,
(int)ret);
ret = length;
out_release_op:
op_release(new_op);
out_unlock:
up_read(&orangefs_inode->xattr_sem);
return ret;
}
static int orangefs_inode_removexattr(struct inode *inode, const char *name,
int flags)
{
struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
struct orangefs_kernel_op_s *new_op = NULL;
int ret = -ENOMEM;
down_write(&orangefs_inode->xattr_sem);
new_op = op_alloc(ORANGEFS_VFS_OP_REMOVEXATTR);
if (!new_op)
goto out_unlock;
new_op->upcall.req.removexattr.refn = orangefs_inode->refn;
/*
* NOTE: Although keys are meant to be NULL terminated
* textual strings, I am going to explicitly pass the
* length just in case we change this later on...
*/
strcpy(new_op->upcall.req.removexattr.key, name);
new_op->upcall.req.removexattr.key_sz = strlen(name) + 1;
gossip_debug(GOSSIP_XATTR_DEBUG,
"orangefs_inode_removexattr: key %s, key_sz %d\n",
(char *)new_op->upcall.req.removexattr.key,
(int)new_op->upcall.req.removexattr.key_sz);
ret = service_operation(new_op,
"orangefs_inode_removexattr",
get_interruptible_flag(inode));
if (ret == -ENOENT) {
/*
* Request to replace a non-existent attribute is an error.
*/
if (flags & XATTR_REPLACE)
ret = -ENODATA;
else
ret = 0;
}
gossip_debug(GOSSIP_XATTR_DEBUG,
"orangefs_inode_removexattr: returning %d\n", ret);
op_release(new_op);
out_unlock:
up_write(&orangefs_inode->xattr_sem);
return ret;
}
/*
* Tries to set an attribute for a given key on a file.
*
* Returns a -ve number on error and 0 on success. Key is text, but value
* can be binary!
*/
int orangefs_inode_setxattr(struct inode *inode, const char *name,
const void *value, size_t size, int flags)
{
struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
struct orangefs_kernel_op_s *new_op;
int internal_flag = 0;
int ret = -ENOMEM;
gossip_debug(GOSSIP_XATTR_DEBUG,
"%s: name %s, buffer_size %zd\n",
__func__, name, size);
if (size >= ORANGEFS_MAX_XATTR_VALUELEN ||
flags < 0) {
gossip_err("orangefs_inode_setxattr: bogus values of size(%d), flags(%d)\n",
(int)size,
flags);
return -EINVAL;
}
internal_flag = convert_to_internal_xattr_flags(flags);
if (strlen(name) >= ORANGEFS_MAX_XATTR_NAMELEN) {
gossip_err
("orangefs_inode_setxattr: bogus key size (%d)\n",
(int)(strlen(name)));
return -EINVAL;
}
/* This is equivalent to a removexattr */
if (size == 0 && value == NULL) {
gossip_debug(GOSSIP_XATTR_DEBUG,
"removing xattr (%s)\n",
name);
return orangefs_inode_removexattr(inode, name, flags);
}
gossip_debug(GOSSIP_XATTR_DEBUG,
"setxattr on inode %pU, name %s\n",
get_khandle_from_ino(inode),
name);
down_write(&orangefs_inode->xattr_sem);
new_op = op_alloc(ORANGEFS_VFS_OP_SETXATTR);
if (!new_op)
goto out_unlock;
new_op->upcall.req.setxattr.refn = orangefs_inode->refn;
new_op->upcall.req.setxattr.flags = internal_flag;
/*
* NOTE: Although keys are meant to be NULL terminated textual
* strings, I am going to explicitly pass the length just in
* case we change this later on...
*/
strcpy(new_op->upcall.req.setxattr.keyval.key, name);
new_op->upcall.req.setxattr.keyval.key_sz = strlen(name) + 1;
memcpy(new_op->upcall.req.setxattr.keyval.val, value, size);
new_op->upcall.req.setxattr.keyval.val_sz = size;
gossip_debug(GOSSIP_XATTR_DEBUG,
"orangefs_inode_setxattr: key %s, key_sz %d "
" value size %zd\n",
(char *)new_op->upcall.req.setxattr.keyval.key,
(int)new_op->upcall.req.setxattr.keyval.key_sz,
size);
ret = service_operation(new_op,
"orangefs_inode_setxattr",
get_interruptible_flag(inode));
gossip_debug(GOSSIP_XATTR_DEBUG,
"orangefs_inode_setxattr: returning %d\n",
ret);
/* when request is serviced properly, free req op struct */
op_release(new_op);
out_unlock:
up_write(&orangefs_inode->xattr_sem);
return ret;
}
/*
* Tries to get a specified object's keys into a user-specified buffer of a
* given size. Note that like the previous instances of xattr routines, this
* also allows you to pass in a NULL pointer and 0 size to probe the size for
* subsequent memory allocations. Thus our return value is always the size of
* all the keys unless there were errors in fetching the keys!
*/
ssize_t orangefs_listxattr(struct dentry *dentry, char *buffer, size_t size)
{
struct inode *inode = dentry->d_inode;
struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
struct orangefs_kernel_op_s *new_op;
__u64 token = ORANGEFS_ITERATE_START;
ssize_t ret = -ENOMEM;
ssize_t total = 0;
int count_keys = 0;
int key_size;
int i = 0;
int returned_count = 0;
if (size > 0 && buffer == NULL) {
gossip_err("%s: bogus NULL pointers\n", __func__);
return -EINVAL;
}
down_read(&orangefs_inode->xattr_sem);
new_op = op_alloc(ORANGEFS_VFS_OP_LISTXATTR);
if (!new_op)
goto out_unlock;
if (buffer && size > 0)
memset(buffer, 0, size);
try_again:
key_size = 0;
new_op->upcall.req.listxattr.refn = orangefs_inode->refn;
new_op->upcall.req.listxattr.token = token;
new_op->upcall.req.listxattr.requested_count =
(size == 0) ? 0 : ORANGEFS_MAX_XATTR_LISTLEN;
ret = service_operation(new_op, __func__,
get_interruptible_flag(inode));
if (ret != 0)
goto done;
if (size == 0) {
/*
* This is a bit of a big upper limit, but I did not want to
* spend too much time getting this correct, since users end
* up allocating memory rather than us...
*/
total = new_op->downcall.resp.listxattr.returned_count *
ORANGEFS_MAX_XATTR_NAMELEN;
goto done;
}
returned_count = new_op->downcall.resp.listxattr.returned_count;
if (returned_count < 0 ||
returned_count >= ORANGEFS_MAX_XATTR_LISTLEN) {
gossip_err("%s: impossible value for returned_count:%d:\n",
__func__,
returned_count);
ret = -EIO;
goto done;
}
/*
* Check to see how much can be fit in the buffer. Fit only whole keys.
*/
for (i = 0; i < returned_count; i++) {
if (new_op->downcall.resp.listxattr.lengths[i] < 0 ||
new_op->downcall.resp.listxattr.lengths[i] >
ORANGEFS_MAX_XATTR_NAMELEN) {
gossip_err("%s: impossible value for lengths[%d]\n",
__func__,
new_op->downcall.resp.listxattr.lengths[i]);
ret = -EIO;
goto done;
}
if (total + new_op->downcall.resp.listxattr.lengths[i] > size)
goto done;
/*
* Since many dumb programs try to setxattr() on our reserved
* xattrs this is a feeble attempt at defeating those by not
* listing them in the output of listxattr.. sigh
*/
if (is_reserved_key(new_op->downcall.resp.listxattr.key +
key_size,
new_op->downcall.resp.
listxattr.lengths[i])) {
gossip_debug(GOSSIP_XATTR_DEBUG, "Copying key %d -> %s\n",
i, new_op->downcall.resp.listxattr.key +
key_size);
memcpy(buffer + total,
new_op->downcall.resp.listxattr.key + key_size,
new_op->downcall.resp.listxattr.lengths[i]);
total += new_op->downcall.resp.listxattr.lengths[i];
count_keys++;
} else {
gossip_debug(GOSSIP_XATTR_DEBUG, "[RESERVED] key %d -> %s\n",
i, new_op->downcall.resp.listxattr.key +
key_size);
}
key_size += new_op->downcall.resp.listxattr.lengths[i];
}
/*
* Since the buffer was large enough, we might have to continue
* fetching more keys!
*/
token = new_op->downcall.resp.listxattr.token;
if (token != ORANGEFS_ITERATE_END)
goto try_again;
done:
gossip_debug(GOSSIP_XATTR_DEBUG, "%s: returning %d"
" [size of buffer %ld] (filled in %d keys)\n",
__func__,
ret ? (int)ret : (int)total,
(long)size,
count_keys);
op_release(new_op);
if (ret == 0)
ret = total;
out_unlock:
up_read(&orangefs_inode->xattr_sem);
return ret;
}
static int orangefs_xattr_set_default(const struct xattr_handler *handler,
struct dentry *unused,
struct inode *inode,
const char *name,
const void *buffer,
size_t size,
int flags)
{
return orangefs_inode_setxattr(inode, name, buffer, size, flags);
}
static int orangefs_xattr_get_default(const struct xattr_handler *handler,
struct dentry *unused,
struct inode *inode,
const char *name,
void *buffer,
size_t size)
{
return orangefs_inode_getxattr(inode, name, buffer, size);
}
static struct xattr_handler orangefs_xattr_default_handler = {
.prefix = "", /* match any name => handlers called with full name */
.get = orangefs_xattr_get_default,
.set = orangefs_xattr_set_default,
};
const struct xattr_handler *orangefs_xattr_handlers[] = {
&posix_acl_access_xattr_handler,
&posix_acl_default_xattr_handler,
&orangefs_xattr_default_handler,
NULL
};