Re: PHP dies on URL fopen

Stanislav Malyshev wrote: > Recently, PHP started to segfault for me on URL fopen, e.g.: > fopen("http://www.host.com","r") brings > immediate segfault. > > Investigation showed that the problem happens in php_sock_fgets_internal > (file fsock.c), in the first SEARCHCR() statement and is because when sock > parameter is passed to php_sock_fgets_internal, both readbuf and readpos > are 0, and SEARCHCR does p = READPTR(sock), which makes p be 0 too, and > then uses *p != '\n' as a condition. > Could anybody who know the code look at this and check what has gone > wrong? > Here's what I came up with: What it was (latest CVS I have, fsock.c revision 1.74): #define SEARCHCR() do { \ for (p = READPTR(sock), pe = p + MIN(TOREAD(sock), maxlen); \ *p != '\n'; ) if (++p >= pe) { p = NULL; break; } \ } while (0) What I did instead: #define SEARCHCR() \ if (MIN(TOREAD(sock), maxlen) == 0) \ p = NULL; \ else { \ for (p = READPTR(sock), pe = p + MIN(TOREAD(sock), maxlen); *p != '\n'; p++) { \ if (p >= pe) { \ p = NULL; \ break; \ } \ } \ } That seems to work nicely. Looks like crap, but it works. It also fires off some warnings about comparing READPTR(sock) to 0 since you're comparing a pointer address to an integer, but that shouldn't be a big deal. A patch follows if anybody's interested. btw, this is my first ever patch submitted for PHP, so if it totally sucks, be kind. J Patch: --- ext/standard/fsock.c Tue Jul 31 17:27:42 2001 +++ ext/standard/newfsock.c Tue Jul 31 17:34:31 2001 @@ -89,10 +89,18 @@ efree(key); \ } -#define SEARCHCR() do { \ - for (p = READPTR(sock), pe = p + MIN(TOREAD(sock), maxlen); \ - *p != '\n'; ) if (++p >= pe) { p = NULL; break; } \ -} while (0) +#define SEARCHCR() \ + if (MIN(TOREAD(sock), maxlen) == 0) \ + p = NULL; \ + else { \ + for (p = READPTR(sock), pe = p + MIN(TOREAD(sock),maxlen); *p != '\n'; p++) { \ + if (p >= pe) { \ + p = NULL; \ + break; \ + } \ + } \ + } +