When I tested the UUID-change patch on a read-only device it would fail silently. I tracked this down to the bug described below. (It is possible you already know about this and just choose to ignore it, because as far as I can tell it has no consequences for the current version of cryptsetup.)
The function write_blockwise() will return a negative value in case of an I/O error, or the number of bytes written otherwise. The return value is then tested in LUKS_write_phdr() to detect an error:
r = write_blockwise(devfd, &convHdr, sizeof(struct luks_phdr)) < sizeof(struct luks_phdr) ? -EIO : 0;
...but if write_blockwise() returns a (small) negative error number it will be cast to a (large) positive value in the comparison, because the return value of sizeof() is unsigned (size_t). So the test will fail, and r = 0, so the error will go undetected.
This is sneaky, because for older versions of gcc, sizeof() would return an ssize_t. At some point they changed it to be ANSI compliant.
There is a similar comparison in LUKS_read_phdr. The attached patch just casts the sizeof() value to ssize_t in these two cases.
Best regards,
Jacob Nielsen
Index: luks/keymanage.c
===================================================================
--- luks/keymanage.c (revision 58)
+++ luks/keymanage.c (working copy)
@@ -83,7 +83,7 @@
return -EINVAL;
}
- if(read_blockwise(devfd, hdr, sizeof(struct luks_phdr)) < sizeof(struct luks_phdr)) {
+ if(read_blockwise(devfd, hdr, sizeof(struct luks_phdr)) < (ssize_t)sizeof(struct luks_phdr)) {
r = -EIO;
} else if(memcmp(hdr->magic, luksMagic, LUKS_MAGIC_L)) { /* Check magic */
set_error(_("%s is not a LUKS partition\n"), device);
@@ -138,7 +138,7 @@
convHdr.keyblock[i].stripes = htonl(hdr->keyblock[i].stripes);
}
- r = write_blockwise(devfd, &convHdr, sizeof(struct luks_phdr)) < sizeof(struct luks_phdr) ? -EIO : 0;
+ r = write_blockwise(devfd, &convHdr, sizeof(struct luks_phdr)) < (ssize_t)sizeof(struct luks_phdr) ? -EIO : 0;
close(devfd);
return r;
--------------------------------------------------------------------- dm-crypt mailing list - http://www.saout.de/misc/dm-crypt/ To unsubscribe, e-mail: dm-crypt-unsubscribe@xxxxxxxx For additional commands, e-mail: dm-crypt-help@xxxxxxxx
