2009-08-11 18:26:11 -07:00
/*
FUSE : Filesystem in Userspace
Copyright ( C ) 2001 - 2007 Miklos Szeredi < miklos @ szeredi . hu >
This program can be distributed under the terms of the GNU GPL .
See the file COPYING .
*/
/* This program does the mounting and unmounting of FUSE filesystems */
# include <config.h>
# include "mount_util.h"
# include <stdio.h>
# include <stdlib.h>
# include <string.h>
# include <ctype.h>
# include <unistd.h>
# include <getopt.h>
# include <errno.h>
# include <fcntl.h>
# include <pwd.h>
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
# include <limits.h>
2009-08-11 18:26:11 -07:00
# include <mntent.h>
# include <sys/wait.h>
# include <sys/stat.h>
# include <sys/mount.h>
# include <sys/fsuid.h>
# include <sys/socket.h>
# include <sys/utsname.h>
2010-02-20 18:53:38 +00:00
# include <sched.h>
2009-08-11 18:26:11 -07:00
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
# define FUSE_DEVFD_ENV "_FUSE_DEVFD"
2009-08-11 18:26:11 -07:00
# define FUSE_COMMFD_ENV "_FUSE_COMMFD"
# define FUSE_DEV_OLD " / proc / fs / fuse / dev"
# define FUSE_DEV_NEW " / dev / fuse"
# define FUSE_VERSION_FILE_OLD " / proc / fs / fuse / version"
# define FUSE_CONF " / etc / fuse.conf"
# ifndef MS_DIRSYNC
# define MS_DIRSYNC 128
# endif
2010-02-20 18:53:38 +00:00
# ifndef MS_REC
# define MS_REC 16384
# endif
2010-04-28 01:37:39 +00:00
# ifndef MS_PRIVATE
# define MS_PRIVATE (1<<18)
2010-02-20 18:53:38 +00:00
# endif
2009-08-11 18:26:11 -07:00
static const char * progname ;
static int user_allow_other = 0 ;
static int mount_max = 1000 ;
static const char * get_user_name ( void )
{
struct passwd * pw = getpwuid ( getuid ( ) ) ;
if ( pw ! = NULL & & pw - > pw_name ! = NULL )
return pw - > pw_name ;
else {
fprintf ( stderr , " %s: could not determine username \n " , progname ) ;
return NULL ;
}
}
static uid_t oldfsuid ;
static gid_t oldfsgid ;
static void drop_privs ( void )
{
if ( getuid ( ) ! = 0 ) {
oldfsuid = setfsuid ( getuid ( ) ) ;
oldfsgid = setfsgid ( getgid ( ) ) ;
}
}
static void restore_privs ( void )
{
if ( getuid ( ) ! = 0 ) {
setfsuid ( oldfsuid ) ;
setfsgid ( oldfsgid ) ;
}
}
# ifndef IGNORE_MTAB
2010-02-20 18:53:38 +00:00
/*
* Make sure that / etc / mtab is checked and updated atomically
*/
static int lock_umount ( void )
{
const char * mtab_lock = _PATH_MOUNTED " .fuselock " ;
int mtablock ;
int res ;
struct stat mtab_stat ;
/* /etc/mtab could be a symlink to /proc/mounts */
if ( lstat ( _PATH_MOUNTED , & mtab_stat ) = = 0 & & S_ISLNK ( mtab_stat . st_mode ) )
return - 1 ;
mtablock = open ( mtab_lock , O_RDWR | O_CREAT , 0600 ) ;
if ( mtablock = = - 1 ) {
fprintf ( stderr , " %s: unable to open fuse lock file: %s \n " ,
progname , strerror ( errno ) ) ;
return - 1 ;
}
res = lockf ( mtablock , F_LOCK , 0 ) ;
if ( res < 0 ) {
fprintf ( stderr , " %s: error getting lock: %s \n " , progname ,
strerror ( errno ) ) ;
close ( mtablock ) ;
return - 1 ;
}
return mtablock ;
}
static void unlock_umount ( int mtablock )
{
2012-05-15 21:24:43 +05:30
if ( mtablock > = 0 ) {
int res ;
res = lockf ( mtablock , F_ULOCK , 0 ) ;
if ( res < 0 ) {
fprintf ( stderr , " %s: error releasing lock: %s \n " ,
progname , strerror ( errno ) ) ;
}
close ( mtablock ) ;
}
2010-02-20 18:53:38 +00:00
}
2009-08-11 18:26:11 -07:00
static int add_mount ( const char * source , const char * mnt , const char * type ,
const char * opts )
{
return fuse_mnt_add_mount ( progname , source , mnt , type , opts ) ;
}
2010-02-20 18:53:38 +00:00
static int may_unmount ( const char * mnt , int quiet )
2009-08-11 18:26:11 -07:00
{
2010-02-20 18:53:38 +00:00
struct mntent * entp ;
FILE * fp ;
const char * user = NULL ;
char uidstr [ 32 ] ;
unsigned uidlen = 0 ;
int found ;
const char * mtab = _PATH_MOUNTED ;
2009-08-11 18:26:11 -07:00
2010-02-20 18:53:38 +00:00
user = get_user_name ( ) ;
if ( user = = NULL )
return - 1 ;
2009-08-11 18:26:11 -07:00
2010-02-20 18:53:38 +00:00
fp = setmntent ( mtab , " r " ) ;
if ( fp = = NULL ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " , progname , mtab ,
strerror ( errno ) ) ;
return - 1 ;
}
uidlen = sprintf ( uidstr , " %u " , getuid ( ) ) ;
found = 0 ;
while ( ( entp = getmntent ( fp ) ) ! = NULL ) {
if ( ! found & & strcmp ( entp - > mnt_dir , mnt ) = = 0 & &
( strcmp ( entp - > mnt_type , " fuse " ) = = 0 | |
strcmp ( entp - > mnt_type , " fuseblk " ) = = 0 | |
strncmp ( entp - > mnt_type , " fuse. " , 5 ) = = 0 | |
strncmp ( entp - > mnt_type , " fuseblk. " , 8 ) = = 0 ) ) {
char * p = strstr ( entp - > mnt_opts , " user= " ) ;
if ( p & &
( p = = entp - > mnt_opts | | * ( p - 1 ) = = ' , ' ) & &
strcmp ( p + 5 , user ) = = 0 ) {
found = 1 ;
break ;
}
/* /etc/mtab is a link pointing to
/ proc / mounts : */
else if ( ( p =
strstr ( entp - > mnt_opts , " user_id= " ) ) & &
( p = = entp - > mnt_opts | |
* ( p - 1 ) = = ' , ' ) & &
strncmp ( p + 8 , uidstr , uidlen ) = = 0 & &
( * ( p + 8 + uidlen ) = = ' , ' | |
* ( p + 8 + uidlen ) = = ' \0 ' ) ) {
found = 1 ;
break ;
2009-08-11 18:26:11 -07:00
}
}
2010-02-20 18:53:38 +00:00
}
endmntent ( fp ) ;
2009-08-11 18:26:11 -07:00
2010-02-20 18:53:38 +00:00
if ( ! found ) {
if ( ! quiet )
fprintf ( stderr ,
" %s: entry for %s not found in %s \n " ,
progname , mnt , mtab ) ;
return - 1 ;
}
return 0 ;
}
/*
* Check whether the file specified in " fusermount -u " is really a
* mountpoint and not a symlink . This is necessary otherwise the user
* could move the mountpoint away and replace it with a symlink
* pointing to an arbitrary mount , thereby tricking fusermount into
* unmounting that ( umount ( 2 ) will follow symlinks ) .
*
* This is the child process running in a separate mount namespace , so
* we don ' t mess with the global namespace and if the process is
* killed for any reason , mounts are automatically cleaned up .
*
* First make sure nothing is propagated back into the parent
2010-04-28 01:37:39 +00:00
* namespace by marking all mounts " private " .
2010-02-20 18:53:38 +00:00
*
* Then bind mount parent onto a stable base where the user can ' t move
2010-04-28 01:37:39 +00:00
* it around .
2010-02-20 18:53:38 +00:00
*
* Finally check / proc / mounts for an entry matching the requested
* mountpoint . If it ' s found then we are OK , and the user can ' t move
2010-04-28 01:37:39 +00:00
* it around within the parent directory as rename ( ) will return
* EBUSY . Be careful to ignore any mounts that existed before the
* bind .
2010-02-20 18:53:38 +00:00
*/
static int check_is_mount_child ( void * p )
{
const char * * a = p ;
const char * last = a [ 0 ] ;
const char * mnt = a [ 1 ] ;
int res ;
const char * procmounts = " /proc/mounts " ;
int found ;
FILE * fp ;
struct mntent * entp ;
2010-04-28 01:37:39 +00:00
int count ;
2010-02-20 18:53:38 +00:00
2010-04-28 01:37:39 +00:00
res = mount ( " " , " / " , " " , MS_PRIVATE | MS_REC , NULL ) ;
2010-02-20 18:53:38 +00:00
if ( res = = - 1 ) {
2010-04-28 01:37:39 +00:00
fprintf ( stderr , " %s: failed to mark mounts private: %s \n " ,
2010-02-20 18:53:38 +00:00
progname , strerror ( errno ) ) ;
return 1 ;
}
2010-04-28 01:37:39 +00:00
fp = setmntent ( procmounts , " r " ) ;
if ( fp = = NULL ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " , progname ,
procmounts , strerror ( errno ) ) ;
2010-02-20 18:53:38 +00:00
return 1 ;
}
2010-04-28 01:37:39 +00:00
count = 0 ;
2012-05-15 21:24:43 +05:30
while ( getmntent ( fp ) ! = NULL )
2010-04-28 01:37:39 +00:00
count + + ;
endmntent ( fp ) ;
2010-02-20 18:53:38 +00:00
fp = setmntent ( procmounts , " r " ) ;
if ( fp = = NULL ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " , progname ,
procmounts , strerror ( errno ) ) ;
return 1 ;
}
2010-04-28 01:37:39 +00:00
res = mount ( " . " , " / " , " " , MS_BIND | MS_REC , NULL ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: failed to bind parent to /: %s \n " ,
progname , strerror ( errno ) ) ;
return 1 ;
}
2010-02-20 18:53:38 +00:00
found = 0 ;
while ( ( entp = getmntent ( fp ) ) ! = NULL ) {
2010-04-28 01:37:39 +00:00
if ( count > 0 ) {
count - - ;
continue ;
}
if ( entp - > mnt_dir [ 0 ] = = ' / ' & &
strcmp ( entp - > mnt_dir + 1 , last ) = = 0 ) {
2010-02-20 18:53:38 +00:00
found = 1 ;
break ;
2009-08-11 18:26:11 -07:00
}
}
2010-02-20 18:53:38 +00:00
endmntent ( fp ) ;
if ( ! found ) {
fprintf ( stderr , " %s: %s not mounted \n " , progname , mnt ) ;
return 1 ;
}
return 0 ;
}
static pid_t clone_newns ( void * a )
{
char buf [ 131072 ] ;
char * stack = buf + ( sizeof ( buf ) / 2 - ( ( size_t ) buf & 15 ) ) ;
# ifdef __ia64__
extern int __clone2 ( int ( * fn ) ( void * ) ,
void * child_stack_base , size_t stack_size ,
int flags , void * arg , pid_t * ptid ,
void * tls , pid_t * ctid ) ;
return __clone2 ( check_is_mount_child , stack , sizeof ( buf ) / 2 ,
CLONE_NEWNS , a , NULL , NULL , NULL ) ;
# else
return clone ( check_is_mount_child , stack , CLONE_NEWNS , a ) ;
# endif
}
static int check_is_mount ( const char * last , const char * mnt )
{
pid_t pid , p ;
int status ;
const char * a [ 2 ] = { last , mnt } ;
pid = clone_newns ( ( void * ) a ) ;
if ( pid = = ( pid_t ) - 1 ) {
fprintf ( stderr , " %s: failed to clone namespace: %s \n " ,
progname , strerror ( errno ) ) ;
return - 1 ;
}
p = waitpid ( pid , & status , __WCLONE ) ;
if ( p = = ( pid_t ) - 1 ) {
fprintf ( stderr , " %s: waitpid failed: %s \n " ,
progname , strerror ( errno ) ) ;
return - 1 ;
}
if ( ! WIFEXITED ( status ) ) {
fprintf ( stderr , " %s: child terminated abnormally (status %i) \n " ,
progname , status ) ;
return - 1 ;
}
if ( WEXITSTATUS ( status ) ! = 0 )
return - 1 ;
return 0 ;
}
2012-05-15 21:24:43 +05:30
static int chdir_to_parent ( char * copy , const char * * lastp )
2010-02-20 18:53:38 +00:00
{
char * tmp ;
const char * parent ;
char buf [ 65536 ] ;
int res ;
tmp = strrchr ( copy , ' / ' ) ;
if ( tmp = = NULL | | tmp [ 1 ] = = ' \0 ' ) {
fprintf ( stderr , " %s: internal error: invalid abs path: <%s> \n " ,
progname , copy ) ;
return - 1 ;
}
if ( tmp ! = copy ) {
* tmp = ' \0 ' ;
parent = copy ;
* lastp = tmp + 1 ;
} else if ( tmp [ 1 ] ! = ' \0 ' ) {
* lastp = tmp + 1 ;
parent = " / " ;
} else {
* lastp = " . " ;
parent = " / " ;
}
res = chdir ( parent ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: failed to chdir to %s: %s \n " ,
progname , parent , strerror ( errno ) ) ;
return - 1 ;
}
if ( getcwd ( buf , sizeof ( buf ) ) = = NULL ) {
fprintf ( stderr , " %s: failed to obtain current directory: %s \n " ,
progname , strerror ( errno ) ) ;
return - 1 ;
}
if ( strcmp ( buf , parent ) ! = 0 ) {
fprintf ( stderr , " %s: mountpoint moved (%s -> %s) \n " , progname ,
parent , buf ) ;
return - 1 ;
}
return 0 ;
}
static int unmount_fuse_locked ( const char * mnt , int quiet , int lazy )
{
char * copy ;
const char * last ;
int res ;
if ( getuid ( ) ! = 0 ) {
res = may_unmount ( mnt , quiet ) ;
if ( res = = - 1 )
return - 1 ;
}
copy = strdup ( mnt ) ;
if ( copy = = NULL ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
return - 1 ;
}
2012-05-15 21:24:43 +05:30
res = chdir_to_parent ( copy , & last ) ;
2010-02-20 18:53:38 +00:00
if ( res = = - 1 )
goto out ;
res = check_is_mount ( last , mnt ) ;
if ( res = = - 1 )
goto out ;
res = fuse_mnt_umount ( progname , mnt , last , lazy ) ;
out :
free ( copy ) ;
return res ;
}
static int unmount_fuse ( const char * mnt , int quiet , int lazy )
{
int res ;
int mtablock = lock_umount ( ) ;
res = unmount_fuse_locked ( mnt , quiet , lazy ) ;
unlock_umount ( mtablock ) ;
2009-08-11 18:26:11 -07:00
2010-02-20 18:53:38 +00:00
return res ;
2009-08-11 18:26:11 -07:00
}
static int count_fuse_fs ( void )
{
struct mntent * entp ;
int count = 0 ;
const char * mtab = _PATH_MOUNTED ;
FILE * fp = setmntent ( mtab , " r " ) ;
if ( fp = = NULL ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " , progname , mtab ,
strerror ( errno ) ) ;
return - 1 ;
}
while ( ( entp = getmntent ( fp ) ) ! = NULL ) {
if ( strcmp ( entp - > mnt_type , " fuse " ) = = 0 | |
strncmp ( entp - > mnt_type , " fuse. " , 5 ) = = 0 )
count + + ;
}
endmntent ( fp ) ;
return count ;
}
# else /* IGNORE_MTAB */
static int count_fuse_fs ( )
{
return 0 ;
}
static int add_mount ( const char * source , const char * mnt , const char * type ,
const char * opts )
{
( void ) source ;
( void ) mnt ;
( void ) type ;
( void ) opts ;
return 0 ;
}
static int unmount_fuse ( const char * mnt , int quiet , int lazy )
{
2010-02-20 18:53:38 +00:00
return fuse_mnt_umount ( progname , mnt , mnt , lazy ) ;
2009-08-11 18:26:11 -07:00
}
# endif /* IGNORE_MTAB */
static void strip_line ( char * line )
{
char * s = strchr ( line , ' # ' ) ;
if ( s ! = NULL )
s [ 0 ] = ' \0 ' ;
for ( s = line + strlen ( line ) - 1 ;
s > = line & & isspace ( ( unsigned char ) * s ) ; s - - ) ;
s [ 1 ] = ' \0 ' ;
for ( s = line ; isspace ( ( unsigned char ) * s ) ; s + + ) ;
if ( s ! = line )
memmove ( line , s , strlen ( s ) + 1 ) ;
}
static void parse_line ( char * line , int linenum )
{
int tmp ;
if ( strcmp ( line , " user_allow_other " ) = = 0 )
user_allow_other = 1 ;
else if ( sscanf ( line , " mount_max = %i " , & tmp ) = = 1 )
mount_max = tmp ;
else if ( line [ 0 ] )
fprintf ( stderr ,
" %s: unknown parameter in %s at line %i: '%s' \n " ,
progname , FUSE_CONF , linenum , line ) ;
}
static void read_conf ( void )
{
FILE * fp = fopen ( FUSE_CONF , " r " ) ;
if ( fp ! = NULL ) {
int linenum = 1 ;
char line [ 256 ] ;
int isnewline = 1 ;
while ( fgets ( line , sizeof ( line ) , fp ) ! = NULL ) {
if ( isnewline ) {
if ( line [ strlen ( line ) - 1 ] = = ' \n ' ) {
strip_line ( line ) ;
parse_line ( line , linenum ) ;
} else {
isnewline = 0 ;
}
} else if ( line [ strlen ( line ) - 1 ] = = ' \n ' ) {
fprintf ( stderr , " %s: reading %s: line %i too long \n " , progname , FUSE_CONF , linenum ) ;
isnewline = 1 ;
}
if ( isnewline )
linenum + + ;
}
if ( ! isnewline ) {
fprintf ( stderr , " %s: reading %s: missing newline at end of file \n " , progname , FUSE_CONF ) ;
}
fclose ( fp ) ;
} else if ( errno ! = ENOENT ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " ,
progname , FUSE_CONF , strerror ( errno ) ) ;
}
}
static int begins_with ( const char * s , const char * beg )
{
if ( strncmp ( s , beg , strlen ( beg ) ) = = 0 )
return 1 ;
else
return 0 ;
}
struct mount_flags {
const char * opt ;
unsigned long flag ;
int on ;
int safe ;
} ;
static struct mount_flags mount_flags [ ] = {
{ " rw " , MS_RDONLY , 0 , 1 } ,
{ " ro " , MS_RDONLY , 1 , 1 } ,
{ " suid " , MS_NOSUID , 0 , 0 } ,
{ " nosuid " , MS_NOSUID , 1 , 1 } ,
{ " dev " , MS_NODEV , 0 , 0 } ,
{ " nodev " , MS_NODEV , 1 , 1 } ,
{ " exec " , MS_NOEXEC , 0 , 1 } ,
{ " noexec " , MS_NOEXEC , 1 , 1 } ,
{ " async " , MS_SYNCHRONOUS , 0 , 1 } ,
{ " sync " , MS_SYNCHRONOUS , 1 , 1 } ,
{ " atime " , MS_NOATIME , 0 , 1 } ,
{ " noatime " , MS_NOATIME , 1 , 1 } ,
{ " dirsync " , MS_DIRSYNC , 1 , 1 } ,
{ NULL , 0 , 0 , 0 }
} ;
static int find_mount_flag ( const char * s , unsigned len , int * on , int * flag )
{
int i ;
for ( i = 0 ; mount_flags [ i ] . opt ! = NULL ; i + + ) {
const char * opt = mount_flags [ i ] . opt ;
if ( strlen ( opt ) = = len & & strncmp ( opt , s , len ) = = 0 ) {
* on = mount_flags [ i ] . on ;
* flag = mount_flags [ i ] . flag ;
if ( ! mount_flags [ i ] . safe & & getuid ( ) ! = 0 ) {
* flag = 0 ;
fprintf ( stderr ,
" %s: unsafe option %s ignored \n " ,
progname , opt ) ;
}
return 1 ;
}
}
return 0 ;
}
static int add_option ( char * * optsp , const char * opt , unsigned expand )
{
char * newopts ;
if ( * optsp = = NULL )
newopts = strdup ( opt ) ;
else {
unsigned oldsize = strlen ( * optsp ) ;
unsigned newsize = oldsize + 1 + strlen ( opt ) + expand + 1 ;
newopts = ( char * ) realloc ( * optsp , newsize ) ;
if ( newopts )
sprintf ( newopts + oldsize , " ,%s " , opt ) ;
}
if ( newopts = = NULL ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
return - 1 ;
}
* optsp = newopts ;
return 0 ;
}
static int get_mnt_opts ( int flags , char * opts , char * * mnt_optsp )
{
int i ;
2012-06-20 14:03:14 +02:00
size_t l ;
2009-08-11 18:26:11 -07:00
if ( ! ( flags & MS_RDONLY ) & & add_option ( mnt_optsp , " rw " , 0 ) = = - 1 )
return - 1 ;
for ( i = 0 ; mount_flags [ i ] . opt ! = NULL ; i + + ) {
if ( mount_flags [ i ] . on & & ( flags & mount_flags [ i ] . flag ) & &
add_option ( mnt_optsp , mount_flags [ i ] . opt , 0 ) = = - 1 )
return - 1 ;
}
if ( add_option ( mnt_optsp , opts , 0 ) = = - 1 )
return - 1 ;
/* remove comma from end of opts*/
l = strlen ( * mnt_optsp ) ;
2012-06-20 14:03:14 +02:00
if ( l & & ( * mnt_optsp ) [ l - 1 ] = = ' , ' )
2009-08-11 18:26:11 -07:00
( * mnt_optsp ) [ l - 1 ] = ' \0 ' ;
if ( getuid ( ) ! = 0 ) {
const char * user = get_user_name ( ) ;
if ( user = = NULL )
return - 1 ;
if ( add_option ( mnt_optsp , " user= " , strlen ( user ) ) = = - 1 )
return - 1 ;
strcat ( * mnt_optsp , user ) ;
}
return 0 ;
}
static int opt_eq ( const char * s , unsigned len , const char * opt )
{
if ( strlen ( opt ) = = len & & strncmp ( s , opt , len ) = = 0 )
return 1 ;
else
return 0 ;
}
static int get_string_opt ( const char * s , unsigned len , const char * opt ,
char * * val )
{
2010-09-30 12:44:26 +00:00
int i ;
2009-08-11 18:26:11 -07:00
unsigned opt_len = strlen ( opt ) ;
2010-09-30 12:44:26 +00:00
char * d ;
2009-08-11 18:26:11 -07:00
if ( * val )
free ( * val ) ;
* val = ( char * ) malloc ( len - opt_len + 1 ) ;
if ( ! * val ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
return 0 ;
}
2010-09-30 12:44:26 +00:00
d = * val ;
s + = opt_len ;
len - = opt_len ;
for ( i = 0 ; i < len ; i + + ) {
if ( s [ i ] = = ' \\ ' & & i + 1 < len )
i + + ;
* d + + = s [ i ] ;
}
* d = ' \0 ' ;
2009-08-11 18:26:11 -07:00
return 1 ;
}
static int do_mount ( const char * mnt , char * * typep , mode_t rootmode ,
int fd , const char * opts , const char * dev , char * * sourcep ,
char * * mnt_optsp , off_t rootsize )
{
int res ;
int flags = MS_NOSUID | MS_NODEV ;
char * optbuf ;
char * mnt_opts = NULL ;
const char * s ;
char * d ;
char * fsname = NULL ;
char * subtype = NULL ;
char * source = NULL ;
char * type = NULL ;
int check_empty = 1 ;
int blkdev = 0 ;
optbuf = ( char * ) malloc ( strlen ( opts ) + 128 ) ;
if ( ! optbuf ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
return - 1 ;
}
for ( s = opts , d = optbuf ; * s ; ) {
unsigned len ;
const char * fsname_str = " fsname= " ;
const char * subtype_str = " subtype= " ;
2010-09-30 12:44:26 +00:00
for ( len = 0 ; s [ len ] ; len + + ) {
if ( s [ len ] = = ' \\ ' & & s [ len + 1 ] )
len + + ;
else if ( s [ len ] = = ' , ' )
break ;
}
2009-08-11 18:26:11 -07:00
if ( begins_with ( s , fsname_str ) ) {
if ( ! get_string_opt ( s , len , fsname_str , & fsname ) )
goto err ;
} else if ( begins_with ( s , subtype_str ) ) {
if ( ! get_string_opt ( s , len , subtype_str , & subtype ) )
goto err ;
} else if ( opt_eq ( s , len , " blkdev " ) ) {
if ( getuid ( ) ! = 0 ) {
fprintf ( stderr ,
" %s: option blkdev is privileged \n " ,
progname ) ;
goto err ;
}
blkdev = 1 ;
} else if ( opt_eq ( s , len , " nonempty " ) ) {
check_empty = 0 ;
} else if ( ! begins_with ( s , " fd= " ) & &
! begins_with ( s , " rootmode= " ) & &
! begins_with ( s , " user_id= " ) & &
! begins_with ( s , " group_id= " ) ) {
int on ;
int flag ;
int skip_option = 0 ;
if ( opt_eq ( s , len , " large_read " ) ) {
struct utsname utsname ;
unsigned kmaj , kmin ;
res = uname ( & utsname ) ;
if ( res = = 0 & &
sscanf ( utsname . release , " %u.%u " ,
& kmaj , & kmin ) = = 2 & &
( kmaj > 2 | | ( kmaj = = 2 & & kmin > 4 ) ) ) {
fprintf ( stderr , " %s: note: 'large_read' mount option is deprecated for %i.%i kernels \n " , progname , kmaj , kmin ) ;
skip_option = 1 ;
}
}
if ( getuid ( ) ! = 0 & & ! user_allow_other & &
( opt_eq ( s , len , " allow_other " ) | |
opt_eq ( s , len , " allow_root " ) ) ) {
fprintf ( stderr , " %s: option %.*s only allowed if 'user_allow_other' is set in /etc/fuse.conf \n " , progname , len , s ) ;
goto err ;
}
if ( ! skip_option ) {
if ( find_mount_flag ( s , len , & on , & flag ) ) {
if ( on )
flags | = flag ;
else
flags & = ~ flag ;
} else {
memcpy ( d , s , len ) ;
d + = len ;
* d + + = ' , ' ;
}
}
}
s + = len ;
if ( * s )
s + + ;
}
* d = ' \0 ' ;
res = get_mnt_opts ( flags , optbuf , & mnt_opts ) ;
if ( res = = - 1 )
goto err ;
sprintf ( d , " fd=%i,rootmode=%o,user_id=%i,group_id=%i " ,
fd , rootmode , getuid ( ) , getgid ( ) ) ;
if ( check_empty & &
fuse_mnt_check_empty ( progname , mnt , rootmode , rootsize ) = = - 1 )
goto err ;
source = malloc ( ( fsname ? strlen ( fsname ) : 0 ) +
( subtype ? strlen ( subtype ) : 0 ) + strlen ( dev ) + 32 ) ;
type = malloc ( ( subtype ? strlen ( subtype ) : 0 ) + 32 ) ;
if ( ! type | | ! source ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
goto err ;
}
if ( subtype )
sprintf ( type , " %s.%s " , blkdev ? " fuseblk " : " fuse " , subtype ) ;
else
strcpy ( type , blkdev ? " fuseblk " : " fuse " ) ;
if ( fsname )
strcpy ( source , fsname ) ;
else
strcpy ( source , subtype ? subtype : dev ) ;
res = mount ( source , mnt , type , flags , optbuf ) ;
if ( res = = - 1 & & errno = = ENODEV & & subtype ) {
/* Probably missing subtype support */
strcpy ( type , blkdev ? " fuseblk " : " fuse " ) ;
if ( fsname ) {
if ( ! blkdev )
sprintf ( source , " %s#%s " , subtype , fsname ) ;
} else {
strcpy ( source , type ) ;
}
res = mount ( source , mnt , type , flags , optbuf ) ;
}
if ( res = = - 1 & & errno = = EINVAL ) {
/* It could be an old version not supporting group_id */
sprintf ( d , " fd=%i,rootmode=%o,user_id=%i " ,
fd , rootmode , getuid ( ) ) ;
res = mount ( source , mnt , type , flags , optbuf ) ;
}
if ( res = = - 1 ) {
int errno_save = errno ;
if ( blkdev & & errno = = ENODEV & & ! fuse_mnt_check_fuseblk ( ) )
fprintf ( stderr , " %s: 'fuseblk' support missing \n " ,
progname ) ;
else
fprintf ( stderr , " %s: mount failed: %s \n " , progname ,
strerror ( errno_save ) ) ;
goto err ;
}
2012-05-15 21:24:43 +05:30
* sourcep = source ;
* typep = type ;
* mnt_optsp = mnt_opts ;
2009-08-11 18:26:11 -07:00
free ( fsname ) ;
free ( optbuf ) ;
2012-05-15 21:24:43 +05:30
return 0 ;
2009-08-11 18:26:11 -07:00
err :
free ( fsname ) ;
free ( subtype ) ;
free ( source ) ;
free ( type ) ;
free ( mnt_opts ) ;
free ( optbuf ) ;
return - 1 ;
}
static int check_version ( const char * dev )
{
int res ;
int majorver ;
int minorver ;
const char * version_file ;
FILE * vf ;
if ( strcmp ( dev , FUSE_DEV_OLD ) ! = 0 )
return 0 ;
version_file = FUSE_VERSION_FILE_OLD ;
vf = fopen ( version_file , " r " ) ;
if ( vf = = NULL ) {
fprintf ( stderr , " %s: kernel interface too old \n " , progname ) ;
return - 1 ;
}
res = fscanf ( vf , " %i.%i " , & majorver , & minorver ) ;
fclose ( vf ) ;
if ( res ! = 2 ) {
fprintf ( stderr , " %s: error reading %s \n " , progname ,
version_file ) ;
return - 1 ;
}
if ( majorver < 3 ) {
fprintf ( stderr , " %s: kernel interface too old \n " , progname ) ;
return - 1 ;
}
return 0 ;
}
2012-05-15 21:24:43 +05:30
static int check_perm ( const char * * mntp , struct stat * stbuf , int * mountpoint_fd )
2009-08-11 18:26:11 -07:00
{
int res ;
const char * mnt = * mntp ;
const char * origmnt = mnt ;
res = lstat ( mnt , stbuf ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: failed to access mountpoint %s: %s \n " ,
progname , mnt , strerror ( errno ) ) ;
return - 1 ;
}
/* No permission checking is done for root */
if ( getuid ( ) = = 0 )
return 0 ;
if ( S_ISDIR ( stbuf - > st_mode ) ) {
res = chdir ( mnt ) ;
if ( res = = - 1 ) {
fprintf ( stderr ,
" %s: failed to chdir to mountpoint: %s \n " ,
progname , strerror ( errno ) ) ;
return - 1 ;
}
mnt = * mntp = " . " ;
res = lstat ( mnt , stbuf ) ;
if ( res = = - 1 ) {
fprintf ( stderr ,
" %s: failed to access mountpoint %s: %s \n " ,
progname , origmnt , strerror ( errno ) ) ;
return - 1 ;
}
if ( ( stbuf - > st_mode & S_ISVTX ) & & stbuf - > st_uid ! = getuid ( ) ) {
fprintf ( stderr , " %s: mountpoint %s not owned by user \n " ,
progname , origmnt ) ;
return - 1 ;
}
res = access ( mnt , W_OK ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: user has no write access to mountpoint %s \n " ,
progname , origmnt ) ;
return - 1 ;
}
} else if ( S_ISREG ( stbuf - > st_mode ) ) {
static char procfile [ 256 ] ;
* mountpoint_fd = open ( mnt , O_WRONLY ) ;
if ( * mountpoint_fd = = - 1 ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " ,
progname , mnt , strerror ( errno ) ) ;
return - 1 ;
}
res = fstat ( * mountpoint_fd , stbuf ) ;
if ( res = = - 1 ) {
fprintf ( stderr ,
" %s: failed to access mountpoint %s: %s \n " ,
progname , mnt , strerror ( errno ) ) ;
return - 1 ;
}
if ( ! S_ISREG ( stbuf - > st_mode ) ) {
fprintf ( stderr ,
" %s: mountpoint %s is no longer a regular file \n " ,
progname , mnt ) ;
return - 1 ;
}
sprintf ( procfile , " /proc/self/fd/%i " , * mountpoint_fd ) ;
* mntp = procfile ;
} else {
fprintf ( stderr ,
" %s: mountpoint %s is not a directory or a regular file \n " ,
progname , mnt ) ;
return - 1 ;
}
return 0 ;
}
static int try_open ( const char * dev , char * * devp , int silent )
{
int fd = open ( dev , O_RDWR ) ;
if ( fd ! = - 1 ) {
* devp = strdup ( dev ) ;
if ( * devp = = NULL ) {
fprintf ( stderr , " %s: failed to allocate memory \n " ,
progname ) ;
close ( fd ) ;
fd = - 1 ;
}
} else if ( errno = = ENODEV | |
errno = = ENOENT ) /* check for ENOENT too, for the udev case */
return - 2 ;
else if ( ! silent ) {
fprintf ( stderr , " %s: failed to open %s: %s \n " , progname , dev ,
strerror ( errno ) ) ;
}
return fd ;
}
static int try_open_fuse_device ( char * * devp )
{
int fd ;
int err ;
drop_privs ( ) ;
fd = try_open ( FUSE_DEV_NEW , devp , 0 ) ;
restore_privs ( ) ;
if ( fd > = 0 )
return fd ;
err = fd ;
fd = try_open ( FUSE_DEV_OLD , devp , 1 ) ;
if ( fd > = 0 )
return fd ;
return err ;
}
static int open_fuse_device ( char * * devp )
{
int fd = try_open_fuse_device ( devp ) ;
if ( fd > = - 1 )
return fd ;
fprintf ( stderr ,
" %s: fuse device not found, try 'modprobe fuse' first \n " ,
progname ) ;
return - 1 ;
}
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
static int check_fuse_device ( char * devfd , char * * devp )
{
int res ;
char * devlink ;
res = asprintf ( & devlink , " /proc/self/fd/%s " , devfd ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
return - 1 ;
}
* devp = ( char * ) calloc ( 1 , PATH_MAX + 1 ) ;
if ( ! * devp ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , progname ) ;
free ( devlink ) ;
return - 1 ;
}
res = readlink ( devlink , * devp , PATH_MAX ) ;
free ( devlink ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: specified fuse fd is invalid \n " ,
progname ) ;
return - 1 ;
}
return atoi ( devfd ) ;
}
2009-08-11 18:26:11 -07:00
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
static int mount_fuse ( const char * mnt , const char * opts , char * devfd )
2009-08-11 18:26:11 -07:00
{
int res ;
int fd ;
char * dev ;
struct stat stbuf ;
char * type = NULL ;
char * source = NULL ;
char * mnt_opts = NULL ;
const char * real_mnt = mnt ;
int mountpoint_fd = - 1 ;
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
fd = devfd ? check_fuse_device ( devfd , & dev ) : open_fuse_device ( & dev ) ;
2009-08-11 18:26:11 -07:00
if ( fd = = - 1 )
return - 1 ;
drop_privs ( ) ;
read_conf ( ) ;
if ( getuid ( ) ! = 0 & & mount_max ! = - 1 ) {
int mount_count = count_fuse_fs ( ) ;
if ( mount_count > = mount_max ) {
fprintf ( stderr , " %s: too many FUSE filesystems mounted; mount_max=N can be set in /etc/fuse.conf \n " , progname ) ;
2012-05-15 21:24:43 +05:30
goto fail_close_fd ;
2009-08-11 18:26:11 -07:00
}
}
res = check_version ( dev ) ;
if ( res ! = - 1 ) {
2012-05-15 21:24:43 +05:30
res = check_perm ( & real_mnt , & stbuf , & mountpoint_fd ) ;
2009-08-11 18:26:11 -07:00
restore_privs ( ) ;
if ( res ! = - 1 )
res = do_mount ( real_mnt , & type , stbuf . st_mode & S_IFMT ,
fd , opts , dev , & source , & mnt_opts ,
stbuf . st_size ) ;
} else
restore_privs ( ) ;
if ( mountpoint_fd ! = - 1 )
close ( mountpoint_fd ) ;
2012-05-15 21:24:43 +05:30
if ( res = = - 1 )
goto fail_close_fd ;
res = chdir ( " / " ) ;
2009-08-11 18:26:11 -07:00
if ( res = = - 1 ) {
2012-05-15 21:24:43 +05:30
fprintf ( stderr , " %s: failed to chdir to '/' \n " , progname ) ;
goto fail_close_fd ;
2009-08-11 18:26:11 -07:00
}
if ( geteuid ( ) = = 0 ) {
res = add_mount ( source , mnt , type , mnt_opts ) ;
if ( res = = - 1 ) {
2012-05-15 21:24:43 +05:30
/* Can't clean up mount in a non-racy way */
goto fail_close_fd ;
2009-08-11 18:26:11 -07:00
}
}
2012-05-15 21:24:43 +05:30
out_free :
2009-08-11 18:26:11 -07:00
free ( source ) ;
free ( type ) ;
free ( mnt_opts ) ;
free ( dev ) ;
return fd ;
2012-05-15 21:24:43 +05:30
fail_close_fd :
close ( fd ) ;
fd = - 1 ;
goto out_free ;
2009-08-11 18:26:11 -07:00
}
static int send_fd ( int sock_fd , int fd )
{
int retval ;
struct msghdr msg ;
struct cmsghdr * p_cmsg ;
struct iovec vec ;
size_t cmsgbuf [ CMSG_SPACE ( sizeof ( fd ) ) / sizeof ( size_t ) ] ;
int * p_fds ;
char sendchar = 0 ;
msg . msg_control = cmsgbuf ;
msg . msg_controllen = sizeof ( cmsgbuf ) ;
p_cmsg = CMSG_FIRSTHDR ( & msg ) ;
p_cmsg - > cmsg_level = SOL_SOCKET ;
p_cmsg - > cmsg_type = SCM_RIGHTS ;
p_cmsg - > cmsg_len = CMSG_LEN ( sizeof ( fd ) ) ;
p_fds = ( int * ) CMSG_DATA ( p_cmsg ) ;
* p_fds = fd ;
msg . msg_controllen = p_cmsg - > cmsg_len ;
msg . msg_name = NULL ;
msg . msg_namelen = 0 ;
msg . msg_iov = & vec ;
msg . msg_iovlen = 1 ;
msg . msg_flags = 0 ;
/* "To pass file descriptors or credentials you need to send/read at
* least one byte " (man 7 unix) */
vec . iov_base = & sendchar ;
vec . iov_len = sizeof ( sendchar ) ;
while ( ( retval = sendmsg ( sock_fd , & msg , 0 ) ) = = - 1 & & errno = = EINTR ) ;
if ( retval ! = 1 ) {
perror ( " sending file descriptor " ) ;
return - 1 ;
}
return 0 ;
}
static void usage ( void )
{
fprintf ( stderr ,
" %s: [options] mountpoint \n "
" Options: \n "
" -h print help \n "
" -V print version \n "
" -o opt[,opt...] mount options \n "
" -u unmount \n "
" -q quiet \n "
" -z lazy unmount \n " ,
progname ) ;
exit ( 1 ) ;
}
static void show_version ( void )
{
printf ( " fusermount version: %s \n " , PACKAGE_VERSION ) ;
exit ( 0 ) ;
}
int main ( int argc , char * argv [ ] )
{
int ch ;
int fd ;
int res ;
char * origmnt ;
char * mnt ;
static int unmount = 0 ;
static int lazy = 0 ;
static int quiet = 0 ;
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
char * devfd ;
2009-08-11 18:26:11 -07:00
char * commfd ;
int cfd ;
const char * opts = " " ;
static const struct option long_opts [ ] = {
{ " unmount " , no_argument , NULL , ' u ' } ,
{ " lazy " , no_argument , NULL , ' z ' } ,
{ " quiet " , no_argument , NULL , ' q ' } ,
{ " help " , no_argument , NULL , ' h ' } ,
{ " version " , no_argument , NULL , ' V ' } ,
{ 0 , 0 , 0 , 0 } } ;
progname = strdup ( argv [ 0 ] ) ;
if ( progname = = NULL ) {
fprintf ( stderr , " %s: failed to allocate memory \n " , argv [ 0 ] ) ;
exit ( 1 ) ;
}
while ( ( ch = getopt_long ( argc , argv , " hVo:uzq " , long_opts ,
NULL ) ) ! = - 1 ) {
switch ( ch ) {
case ' h ' :
usage ( ) ;
break ;
case ' V ' :
show_version ( ) ;
break ;
case ' o ' :
opts = optarg ;
break ;
case ' u ' :
unmount = 1 ;
break ;
case ' z ' :
lazy = 1 ;
break ;
case ' q ' :
quiet = 1 ;
break ;
default :
exit ( 1 ) ;
}
}
if ( lazy & & ! unmount ) {
fprintf ( stderr , " %s: -z can only be used with -u \n " , progname ) ;
exit ( 1 ) ;
}
if ( optind > = argc ) {
fprintf ( stderr , " %s: missing mountpoint argument \n " , progname ) ;
exit ( 1 ) ;
} else if ( argc > optind + 1 ) {
fprintf ( stderr , " %s: extra arguments after the mountpoint \n " ,
progname ) ;
exit ( 1 ) ;
}
origmnt = argv [ optind ] ;
drop_privs ( ) ;
mnt = fuse_mnt_resolve_path ( progname , origmnt ) ;
2012-05-15 21:24:43 +05:30
if ( mnt ! = NULL ) {
res = chdir ( " / " ) ;
if ( res = = - 1 ) {
fprintf ( stderr , " %s: failed to chdir to '/' \n " , progname ) ;
exit ( 1 ) ;
}
}
2009-08-11 18:26:11 -07:00
restore_privs ( ) ;
if ( mnt = = NULL )
exit ( 1 ) ;
umask ( 033 ) ;
if ( unmount ) {
if ( geteuid ( ) = = 0 )
res = unmount_fuse ( mnt , quiet , lazy ) ;
else {
res = umount2 ( mnt , lazy ? 2 : 0 ) ;
if ( res = = - 1 & & ! quiet )
fprintf ( stderr ,
" %s: failed to unmount %s: %s \n " ,
progname , mnt , strerror ( errno ) ) ;
}
if ( res = = - 1 )
exit ( 1 ) ;
return 0 ;
}
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
devfd = getenv ( FUSE_DEVFD_ENV ) ;
if ( devfd = = NULL ) {
commfd = getenv ( FUSE_COMMFD_ENV ) ;
if ( commfd = = NULL ) {
fprintf ( stderr , " %s: old style mounting not supported \n " ,
progname ) ;
exit ( 1 ) ;
}
2009-08-11 18:26:11 -07:00
}
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
fd = mount_fuse ( mnt , opts , devfd ) ;
2009-08-11 18:26:11 -07:00
if ( fd = = - 1 )
exit ( 1 ) ;
fuse, glusterfsd: mount logic fixes
Commit 7d0397c2 introduced two issues:
i) broke the libfuse derived mount logic (details below)
ii) in case of a daemonized glusterfs client is ran as daemon, parent
process can return earlier than the mount is in place, which breaks
agents that programmatically do a gluster mount via a direct call to
glusterfs (ie. not via mount(8)).
This patch fixes these issues by a refactor that merges the approaches
sported by commits
7d0397c2 fuse: allow requests during mount (needed for SELinux labels)
c5d781e0 upon daemonizing, wait on mtab update to terminate in parent
Original daemonized libfuse event flow is as follows:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
mount(8) -f # manipulate mtab
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
where fusermount(1) does:
fd = open("/dev/fuse")
mount("-oopts,fd=%d" % fd ...)
sp = atoi(getenv("_FUSE_COMMFD"))
send_fd(sp, fd)
daemonize(
# in child
fuse_loop(fd)
)
# in parent
exit()
As of 013850c9 (instead of adopting FUSE's 47e61004¹), we went for async
mtab manipulation, and as of c5d781e0, still wanted keep that in sync
with termination of daemon parent, so we changed it to:
try:
fd = open("/dev/fuse")
mount("-oopts,fd=%s" % fd ...)
pid = fork(
# in child
mount(8) -f
)
except:
sp = socketpair()
env _FUSE_COMMFD=sp fusermount -oopts
fd = receive_fd(sp)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
(Note the new approch came only to direct [privileged] mount, so fusermount
based mounting was already partially broken.)
As of 7d0397c2, with the purpose of facilitating async mount, the event flow
was practically reduced to:
fd = open("/dev/fuse")
fork(
mount("-oopts,fd=%s" % fd ...)
fork(
mount(8) -n
)
)
daemonize(
fuse_loop(fd)
)
exit()
Thus fusermount based mounting become defunct; however, the dead
code was still kept around. So, we should either drop it or fix
it. Also, the mtab manipulator is forked into yet another child
with no purpose, while syncing with it in daemon parent is broken.
mount(2) is neither synced with parent.
Now we are coming to the following scheme:
fd = open("/dev/fuse")
pid = fork(
try:
mount("-oopts,fd=%s" % fd ...)
mount(8) -n
except:
env _FUSE_DEVFD=fd fusermount -oopts
)
where fusermount(1) does:
fd = getenv("_FUSE_DEVFD")
mount("-oopts,fd=%s" % fd ...)
daemonize(
fuse_loop(fd)
)
waitpid(pid)
exit()
Nb.:
- We can't help losing compatibility with upstream fusermount,
as it sends back the fd only when mount(2) is completed,
thus defeating the async mount approach. The
'getenv("_FUSE_DEVFD")' mechanism is specfic to glusterfs'
fusermount (at the moment -- sure we can talk about it with
upstream)
- fusermount opens /dev/fuse at same privilege level as of
original process², so we can bravely go on with doing the open
unconditionally in original process
- Original mounting code actually tries to mount through
fusermount _twice_: if first attempt fails, then, assuming
subtype support is missing in kernel, it tries again subtype
stripped. However, this is redundant, as fusermount internally
also performs the subtype check³. Therefore we simplified the
logic to have just a single fusermount call.
- we revert the changes to mount.glusterfs as of 7d0397c2, as
now there is no issue with glusterfs to work around in that scope
¹ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blobdiff;f=ChangeLog;h=47e61004;hb=4c3d9b19;hpb=e61b775a
² http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l1023
³ http://fuse.git.sourceforge.net/git/gitweb.cgi?p=fuse/fuse;a=blob;f=util/fusermount.c;h=b2e87d95#l839
Change-Id: I0c4ab70e0c5ad7b27337228749b266bcd0ba941d
BUG: 811217
Signed-off-by: Csaba Henk <csaba@redhat.com>
Reviewed-on: http://review.gluster.com/3341
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Anand Avati <avati@redhat.com>
2012-05-14 17:07:28 +05:30
if ( devfd = = NULL ) {
cfd = atoi ( commfd ) ;
res = send_fd ( cfd , fd ) ;
if ( res = = - 1 )
exit ( 1 ) ;
}
2009-08-11 18:26:11 -07:00
return 0 ;
}