Martchus
/
mdadm
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Manage: fix checks for removal from a container. 

We must only remove from a container if the device isn't a
member of any member array.
To check we look at the 'holders' directory in sysfs.

We currently skip that check if ->devname is "detached", however
that can never be true since the change that introduced
add_detached().

Also sysfs_unique_holder returns status in 'errno' which isn't
entirely safe as e.g. closedir() is probably allowed to clear it.

So make sysfs_unique_holder return an unambigious value, and us
it to decide what to report.

Signed-off-by: NeilBrown <neilb@suse.de>
This commit is contained in:
NeilBrown 2012-09-24 12:26:03 +10:00
parent 65d0b4ce2d
commit aab15415ed
2 changed files with 39 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
Manage.c
View File

@ -857,6 +857,7 @@ int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv,
* hot spare while we are checking, we
* get an O_EXCL open on the container
*/
int ret;
int dnum = fd2devnum(fd);
lfd = open_dev_excl(dnum);
if (lfd < 0) {
@ -864,19 +865,26 @@ int Manage_remove(struct supertype *tst, int fd, struct mddev_dev *dv,
" to container - odd\n");
return -1;
}
/* In the detached case it is not possible to
* check if we are the unique holder, so just
* rely on the 'detached' checks
/* We may not be able to check on holders in
* sysfs, either because we don't have the dev num
* (rdev == 0) or because the device has been detached
* and the 'holders' directory no longer exists
* (ret == -1). In that case, assume it is OK to
* remove.
*/
if (strcmp(dv->devname, "detached") == 0 ||
sysfd >= 0 ||
sysfs_unique_holder(dnum, rdev))
/* pass */;
else {
pr_err("%s is %s, cannot remove.\n",
dv->devname,
errno == EEXIST ? "still in use":
"not a member");
if (rdev == 0)
ret = -1;
else
ret = sysfs_unique_holder(dnum, rdev);
if (ret == 0) {
pr_err("%s is not a member, cannot remove.\n",
dv->devname);
close(lfd);
return -1;
}
if (ret >= 2) {
pr_err("%s is still in use, cannot remove.\n",
dv->devname);
close(lfd);
return -1;
}
@ -1030,7 +1038,7 @@ int Manage_subdevs(char *devname, int fd,
struct mddev_dev **dp;
if (dv->disposition != 'A') {
pr_err("'missing' only meaningful "
"with --re-add\n");
"with --re-add\n");
goto abort;
}
add_devlist = conf_get_devs();
@ -1047,8 +1055,8 @@ int Manage_subdevs(char *devname, int fd,
}
if (strchr(dv->devname, '/') == NULL &&
strchr(dv->devname, ':') == NULL &&
strlen(dv->devname) < 50) {
strchr(dv->devname, ':') == NULL &&
strlen(dv->devname) < 50) {
/* Assume this is a kernel-internal name like 'sda1' */
int found = 0;
char dname[55];

34
sysfs.c
View File

@ -778,18 +778,22 @@ int sysfs_unique_holder(int devnum, long rdev)
* and is the only holder.
* we should be locked against races by
* an O_EXCL on devnum
* Return values:
* 0 - not unique, not even a holder
* 1 - unique, this is the only holder.
* 2/3 - not unique, there is another holder
* -1 - error, cannot find the holders
*/
DIR *dir;
struct dirent *de;
char dirname[100];
char l;
int found = 0;
int ret = 0;
sprintf(dirname, "/sys/dev/block/%d:%d/holders",
major(rdev), minor(rdev));
dir = opendir(dirname);
errno = ENOENT;
if (!dir)
return 0;
return -1;
l = strlen(dirname);
while ((de = readdir(dir)) != NULL) {
char buf[10];
@ -807,8 +811,8 @@ int sysfs_unique_holder(int devnum, long rdev)
strcat(dirname+l, "/dev");
fd = open(dirname, O_RDONLY);
if (fd < 0) {
errno = ENOENT;
break;
/* Probably a race, just ignore this */
continue;
}
n = read(fd, buf, sizeof(buf)-1);
close(fd);
@ -816,24 +820,18 @@ int sysfs_unique_holder(int devnum, long rdev)
continue;
buf[n] = 0;
if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
c != '\n') {
errno = ENOENT;
break;
}
c != '\n')
continue;
if (mj != MD_MAJOR)
mn = -1-(mn>>6);
if (devnum != mn) {
errno = EEXIST;
break;
}
found = 1;
if (devnum == mn)
ret |= 1;
else
ret |= 2;
}
closedir(dir);
if (de)
return 0;
else
return found;
return ret;
}
int sysfs_freeze_array(struct mdinfo *sra)