search for: datum_len

...shqking at gmail.com QA Contact: rsync-qa at samba.org A suspicious integer overflow is found in xattrs.c:692. The code snippet is as follows. 684 for (num = 1; num <= count; num++) { 685 char *ptr, *name; 686 rsync_xa *rxa; 687 size_t name_len = read_varint(f); 688 size_t datum_len = read_varint(f); 689 size_t dget_len = datum_len > MAX_FULL_DATUM ? 1 + MAX_DIGEST_LEN : datum_len; 690 size_t extra_len = MIGHT_NEED_RPRE ? RPRE_LEN : 0; 691 if ((dget_len + extra_len < dget_len) 692 || (dget_len + extra_len + name_len < dget_len)) 693 overflow_exit(&...

...Jay Fenlason, vaguely based on the ACLs patch */ @@ -590,10 +590,10 @@ + strcpy(ptr, name); + if (datum_size) + memcpy(ptr + len, datumbuf, datum_size); -+ x->rxas[curr_rsync_xal.count].name_len = len; -+ x->rxas[curr_rsync_xal.count].name = ptr; -+ x->rxas[curr_rsync_xal.count].datum_len = datum_size; -+ x->rxas[curr_rsync_xal.count].datum = ptr + len; ++ x->rxas[x->count].name_len = len; ++ x->rxas[x->count].name = ptr; ++ x->rxas[x->count].datum_len = datum_size; ++ x->rxas[x->count].datum = ptr + len; + x->count++; + } + if (x->count &gt...

...ributes are copied w/o 2 -X options. */ @@ -358,6 +361,9 @@ int copy_xattrs(const char *source, const char *dest) if (user_only ? !HAS_PREFIX(name, USER_PREFIX) : HAS_PREFIX(name, SYSTEM_PREFIX)) continue; + + if (!strcmp(name, "security.evm")) + continue; #endif datum_len = 0; @@ -828,7 +834,9 @@ void receive_xattr(int f, struct file_struct *file) } #ifdef HAVE_LINUX_XATTRS /* Non-root can only save the user namespace. */ - if (am_root <= 0 && !HAS_PREFIX(name, USER_PREFIX)) { + if (am_root <= 0 && + (!HAS_PREFIX(name, USER_PREFIX...

...ws: 1. EAs on MacOS need not begin with 'user.' like they do on linux. Therefore, I can't backup from a mac->linux machine. However, linux- >mac->linux works fine since the Mac preserves the 'user.' namespace. 2. 2.6.9pre2 has some sort of bug where it zeros out the datum_len of the last rsync_xa in a file. I suspect a buffer overrun somewhere, but couldn't find it, so my patch doesn't work on 2.6.9pre2. My question: is this useful? should I prefix all MacOS EAs with 'user.'?

...+128,7 @@ } if (list_len >= 0) return list_len; - if (errno == ENOTSUP) + if (errno == ENOTSUP || errno == EPERM) return 0; if (errno == ERANGE) { list_len = sys_llistxattr(fname, NULL, 0); @@ -766,6 +766,8 @@ } if (sys_lsetxattr(fname, name, rxas[i].datum, rxas[i].datum_len) < 0) { + if (errno == EPERM) + return 0; rsyserr(FERROR_XFER, errno, "rsync_xal_set: lsetxattr(\"%s\",\"%s\") failed", fname, name); I plan to submit that, I just haven't had a spare moment. > > > However, trying a local rsync...

This is a PoC patch for NFSv4/ZFS ACLs. The objective of the patch is that rsync --acls support NFSv4/ZFS ACLs without requiring a new command line option NFSv4 ACLs can't be represented using POSIX draft ACLs, if an NFSv4 ACL is present a separate POSIX draft ACL will not be present and there are new APIs to access NFSv4 ACLs. So we need to distinguish between NFSv4 ACLs and POSIX ACLs in