search for: get_tmpnam

https://bugzilla.samba.org/show_bug.cgi?id=7816 Summary: get_tmpname() can create invalid UTF-8 filenames Product: rsync Version: 3.0.7 Platform: Sparc OS/Version: Solaris Status: NEW Severity: minor Priority: P3 Component: core AssignedTo: wayned at samba.org ReportedB...

...attempt is made to create a file with a leading ".." As a workaround I figured the simplest thing to do would be to ensure that rsync only ever adds a "." when creating tmp files for filenames that do not already have a leading dot. I've attached a modified version of the get_tmpname method which simply wraps the addition of a "." in an if statement. I'd like to ask the list if you think this is a likely to have any issues (eg filename comparisons relying on a dot always being added). On my system this change seems to work around the problem pretty well. Th...

...5.orig/receiver.c Thu Feb 14 03:41:58 2002 +++ rsync-2.5.5/receiver.c Sat Jun 8 03:41:22 2002 @@ -163,10 +163,16 @@ } } +#include <sys/vfs.h> +static inline int fname_syslimit(const char* path) { + struct statfs sb; + return !statfs(path,&sb) ? sb.f_namelen : 255; +} static int get_tmpname(char *fnametmp, char *fname) { char *f; + int fn_max; /* open tmp file */ if (tmpdir) { @@ -179,7 +185,12 @@ rprintf(FERROR,"filename too long\n"); return 0; } - snprintf(fnametmp,MAXPATHLEN, "%s/.%s.XXXXXX",tmpdir,f); + fn_max = fname_syslimit(tmpdir); +...

...truct map_struct *buf; @@ -314,6 +316,7 @@ extern int preserve_perms; extern int delete_after; struct stats initial_stats; + int write_flags; if (verbose > 2) { rprintf(FINFO,"recv_files(%d) starting\n",flist->count); @@ -404,22 +407,29 @@ buf = NULL; } - if (!get_tmpname(fnametmp,fname)) { - if (buf) unmap_file(buf); - if (fd1 != -1) close(fd1); - continue; - } + if(direct_write) { + fnametmp = fname; + write_flags = O_WRONLY|O_CREAT; + } else { + fnametmp = fnametmpbuf; + if (!get_tmpname(fnametmp,fname)) { + if (buf) unmap_file(buf); + if...

...h a leading ".." >> >> As a workaround I figured the simplest thing to do would be to ensure that rsync only ever adds a "." when creating tmp files for filenames that do not already have a leading dot. >> >> I've attached a modified version of the get_tmpname method which simply wraps the addition of a "." in an if statement. >> >> I'd like to ask the list if you think this is a likely to have any issues (eg filename comparisons relying on a dot always being added). >> >> On my system this change seems to work...

...offset += len; } flush_write_file(fd); + if (inplace) + ftruncate(fd, offset); + if (do_progress) end_progress(total_size); @@ -410,39 +424,52 @@ } else mapbuf = NULL; - if (!get_tmpname(fnametmp,fname)) { - if (mapbuf) unmap_file(mapbuf); - if (fd1 != -1) close(fd1); - continue; - } - - strlcpy(template, fnametmp, sizeof template); + /* We now check to see if we are writing file "inplace&q...

...for the filename being maintained (to restart the download? I don't know, it seems like this would fail for files in tmpdir...). Anyway... --- rsync-2.5.6-orig/receiver.c Mon Jan 20 23:32:17 2003 +++ rsync-2.5.6/receiver.c Mon Aug 25 10:13:37 2003 @@ -163,33 +163,51 @@ } } - static int get_tmpname(char *fnametmp, char *fname) { - char *f; + char *f, holder; + size_t len; + + if(strlen(fname) > MAXPATHLEN) { + rprintf(FERROR,"%s: filename too long\n", fname); + return 0; + } + + holder = 0; /* open tmp file */ if (tmpdir) { + int tlen; f = strrchr(fname,'/')...

...by the same name, so it works; however, if rsync is interrupted, data is lost. Using an rsync-style temporary file would be good, but copy_file would have to be changed. It occurs to me that --copy-dest local copies should probably also use a temporary file so that replaces are atomic. However, get_tmpname is currently private to receiver.c. That brings up another good point: maybe the receiver should perform both --copy-dest and --no-tweak-hlinked local copies since they might hold up the pipeline if done in the generator. But then the generator has to somehow indicate the source of the local cop...

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 I'm backing up a file named 'wh.txt' with the destination an aufs volume - rsync creates a partial filename like '.wh.txt.M8ucDx', which results in an Operation not permitted (1) error - aufs reserves '.wh.*' names for whiteout records I think. Other than ignoring the problem and using --inplace, it doesnt look like I can

...goto return_with_success; goto cleanup; + } else { + char fnametmp[MAXPATHLEN]; + + /* Replace the symlink atomically. */ + if (!get_tmpname(fnametmp, fname)) + goto cleanup; + if (!mktemp(fnametmp)) + goto cleanup; + if (do_symlink(sl, fnametmp) != 0) { + rsyserr(...

I've got a patch that changes f_name_to() to return an unsigned int (like snprintf() and strlcpy() do) and adds checking to ensure that we didn't overflow the name before we try to use it: http://www.blorf.net/name-overflow.patch If anyone would care to check out the following patch before I commit it, please do. ..wayne..