Scope the cookie-wall retry to this URL's own cookies (#15)

The retry trigger compared the whole shared cookie jar before and after the
request. Because the jar is one process-wide store mutated by every in-flight
fetch, a concurrent slot's Set-Cookie landing mid-loop could flip the compare
and force a needless (though bounded) re-fetch of an unrelated self-redirect.

Compare instead the Cookie header THIS url would send, built for its own
host/path. A Set-Cookie from another host no longer trips the retry; a change
in this URL's own outgoing cookies (its self-redirect setting one) still does,
which is exactly the wall we want to re-fetch. No jar-wide snapshot, and the
same natural termination: once the cookie stabilizes the header stops changing.

Reuses append_cookie_header via http_cookie_header (renamed from the
_selftest-suffixed wrapper, now that engine code calls it too).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Signed-off-by: Xavier Roche <roche@httrack.com>

src/htsback.c

@@ -3908,10 +3908,11 @@ void back_wait(struct_back * sback, httrackp * opt, cache_back * cache,
                           if (fp) {
                             LLint alloc_mem = resume + 1;
 
-                            // Reject a hostile Content-Length that would
-                            // overflow the buffer size: drop the partial and
-                            // refetch.
-                            if (back[i].r.totalsize > INT64_MAX - alloc_mem) {
+                            // Bound the in-memory buffer to a 32-bit size (real
+                            // in-RAM resources are far smaller); a hostile
+                            // Content-Length that would overflow the add or the
+                            // (size_t) cast is dropped and refetched instead.
+                            if (back[i].r.totalsize > INT32_MAX - alloc_mem) {
                               url_savename_refname_remove(opt, back[i].url_adr,
                                                           back[i].url_fil);
                               UNLINK(back[i].url_sav);

src/htslib.c

@@ -1059,10 +1059,10 @@ static int append_cookie_header(buff_struct *bstr, t_cookie *cookie,
   return cook;
 }
 
-/* Self-test entry for append_cookie_header(): build the request Cookie line
-   into dst (always NUL-terminated). Returns the number of cookies emitted. */
-int http_cookie_header_selftest(t_cookie *cookie, const char *domain,
-                                const char *path, char *dst, size_t dst_size) {
+/* Build the request Cookie line for domain/path into dst (always
+   NUL-terminated). Returns the number of cookies emitted. */
+int http_cookie_header(t_cookie *cookie, const char *domain, const char *path,
+                       char *dst, size_t dst_size) {
   buff_struct bstr = {dst, dst_size, 0};
 
   assertf(dst != NULL && dst_size > 0);
@@ -2004,6 +2004,15 @@ LLint http_xfread1(htsblk * r, int bufl) {
 
   if (bufl > 0) {
     if (!r->is_write) {         // stocker en mémoire
+      // In-memory content must fit a 32-bit index (allocs below add 1, reads
+      // use int offsets): reject a hostile Content-Length or endless stream.
+      const LLint inmem_want =
+          (r->totalsize >= 0) ? r->totalsize : (r->size + bufl);
+      if (inmem_want >= INT32_MAX) {
+        r->statuscode = STATUSCODE_INVALID;
+        strcpybuff(r->msg, "In-memory content too large");
+        return READ_ERROR;
+      }
       if (r->totalsize >= 0) {  // totalsize déterminé ET ALLOUE
         if (r->adr == NULL) {
           r->adr = (char *) malloct((size_t) r->totalsize + 1);

src/htslib.h

@@ -184,10 +184,10 @@ int http_sendhead(httrackp * opt, t_cookie * cookie, int mode, const char *xsend
                   const char *referer_adr, const char *referer_fil,
                   htsblk * retour);
 /* Build the request "Cookie:" header line for stored cookies matching
-   domain/path into dst (NUL-terminated). Exposed for the -#Q self-test;
-   wraps the same logic http_sendhead() uses. Returns cookies emitted. */
-int http_cookie_header_selftest(t_cookie *cookie, const char *domain,
-                                const char *path, char *dst, size_t dst_size);
+   domain/path into dst (NUL-terminated), wrapping the logic http_sendhead()
+   uses. Returns cookies emitted. */
+int http_cookie_header(t_cookie *cookie, const char *domain, const char *path,
+                       char *dst, size_t dst_size);
 
 T_SOC newhttp(httrackp * opt, const char *iadr, htsblk * retour, int port,
               int waitconnect);

src/htsparse.c

@@ -4311,6 +4311,8 @@ int hts_wait_delayed(htsmoduleStruct * str, lien_adrfilsave *afs,
       *forbidden_url == 0 && IS_DELAYED_EXT(afs->save) && !opt->state.stop) {
     int loops;
     int continue_loop;
+    char BIGSTK
+        cookie_before[16384]; /* #15: this URL's Cookie header, pre-request */
 
     hts_log_print(opt, LOG_DEBUG, "Waiting for type to be known: %s%s", afs->af.adr,
                   afs->af.fil);
@@ -4319,6 +4321,12 @@ int hts_wait_delayed(htsmoduleStruct * str, lien_adrfilsave *afs,
     for(loops = 0, continue_loop = 1;
         IS_DELAYED_EXT(afs->save) && continue_loop && loops < 7; loops++) {
       continue_loop = 0;
+      /* #15: snapshot the Cookie header THIS url would send, so only its own
+         self-redirect Set-Cookie trips the retry, not a concurrent slot's. */
+      cookie_before[0] = '\0';
+      if (opt->accept_cookie && opt->cookie != NULL)
+        http_cookie_header(opt->cookie, jump_identification_const(afs->af.adr),
+                           afs->af.fil, cookie_before, sizeof(cookie_before));
 
       /* Wait for an available slot */
       if (!hts_wait_available_socket(sback, opt, cache, ptr))
@@ -4516,6 +4524,17 @@ int hts_wait_delayed(htsmoduleStruct * str, lien_adrfilsave *afs,
           /* Moved! */
           else if (HTTP_IS_REDIRECT(back[b].r.statuscode)) {
             char BIGSTK mov_url[HTS_URLMAXSIZE * 2];
+            char BIGSTK cookie_after[16384];
+            hts_boolean cookies_changed;
+
+            /* #15: cookie-wall signal: this URL's own Cookie header changed
+               across its self-redirect (a Set-Cookie it just set). */
+            cookie_after[0] = '\0';
+            if (opt->accept_cookie && opt->cookie != NULL)
+              http_cookie_header(
+                  opt->cookie, jump_identification_const(afs->af.adr),
+                  afs->af.fil, cookie_after, sizeof(cookie_after));
+            cookies_changed = strcmp(cookie_before, cookie_after) != 0;
 
             mov_url[0] = '\0';
             strcpybuff(mov_url, back[b].r.location);    // copier URL
@@ -4585,11 +4604,24 @@ int hts_wait_delayed(htsmoduleStruct * str, lien_adrfilsave *afs,
                   url_savename(afs, former, heap(ptr)->adr, heap(ptr)->fil, 
                                opt, sback, cache, hash, ptr, numero_passe,
                                &delayed_back);
+                } else if (cookies_changed) {
+                  // #15: cookie-wall self-redirect; evict the cached
+                  // fast-header so the re-issue refetches with the cookie.
+                  if (cache->cached_tests != NULL)
+                    coucal_remove(cache->cached_tests,
+                                  concat(OPT_GET_BUFF(opt),
+                                         OPT_GET_BUFF_SIZE(opt), afs->af.adr,
+                                         afs->af.fil));
+                  afs->save[0] = '\0';
+                  url_savename(afs, former, heap(ptr)->adr, heap(ptr)->fil, opt,
+                               sback, cache, hash, ptr, numero_passe,
+                               &delayed_back);
+                  continue_loop = 1;
                 } else {
                   hts_log_print(opt, LOG_WARNING,
                                 "Unable to test %s%s (loop to same filename)",
                                 afs->af.adr, afs->af.fil);
-                }               // loop to same location
+                } // loop to same location
               }                 // ident_url_relatif()
             }                   // location
           }                     // redirect

src/htsselftest.c

@@ -1310,6 +1310,59 @@ static int st_headerlong(httrackp *opt, int argc, char **argv) {
   return 0;
 }
 
+/* http_xfread1 must refuse an in-memory buffer whose size would exceed a 32-bit
+   index (hostile Content-Length or endless stream) rather than allocate it.
+   The guard returns before any socket read, so no real connection is needed. */
+static int st_xfread_limit(httrackp *opt, int argc, char **argv) {
+  htsblk r;
+
+  (void) opt;
+  (void) argc;
+  (void) argv;
+
+  // Content-Length just over 2 GiB.
+  memset(&r, 0, sizeof(r));
+  r.soc = INVALID_SOCKET;
+  r.totalsize = (LLint) INT32_MAX + 1;
+  printf("bylen: refused=%d adr=%s msg=%s\n",
+         http_xfread1(&r, 8192) == READ_ERROR, r.adr != NULL ? "alloc" : "null",
+         r.msg);
+  if (r.adr != NULL)
+    freet(r.adr);
+
+  // Unknown length, buffer already at the limit: the next read would exceed it.
+  memset(&r, 0, sizeof(r));
+  r.soc = INVALID_SOCKET;
+  r.totalsize = -1;
+  r.size = (LLint) INT32_MAX;
+  printf("bygrow: refused=%d adr=%s msg=%s\n",
+         http_xfread1(&r, 8192) == READ_ERROR, r.adr != NULL ? "alloc" : "null",
+         r.msg);
+  if (r.adr != NULL)
+    freet(r.adr);
+
+  // Exactly at the 2 GiB index (size + bufl == INT32_MAX): must also be
+  // refused, since the reallocs below add 1 (a `> INT32_MAX` check would let
+  // this through and overflow the int realloc size).
+  memset(&r, 0, sizeof(r));
+  r.soc = INVALID_SOCKET;
+  r.totalsize = -1;
+  r.size = (LLint) INT32_MAX - 8192;
+  http_xfread1(&r, 8192);
+  printf("boundary: msg=%s\n", r.msg);
+
+  // A legitimate small size must NOT be refused by the guard (the read then
+  // fails on the invalid socket, but the size-too-large msg must not be set).
+  memset(&r, 0, sizeof(r));
+  r.soc = INVALID_SOCKET;
+  r.totalsize = 1000;
+  http_xfread1(&r, 8192);
+  printf("accept: msg=%s\n", r.msg);
+  if (r.adr != NULL)
+    freet(r.adr);
+  return 0;
+}
+
 /* Parse a Content-Range header and print the sanitized triple. A hostile value
    (negative or INT64 extreme) must clamp to 0 without signed-overflow UB. */
 static int st_crange(httrackp *opt, int argc, char **argv) {
@@ -1702,7 +1755,7 @@ static int st_cookies(httrackp *opt, int argc, char **argv) {
     return 1;
   }
 
-  http_cookie_header_selftest(&cookie, dom, "/", hdr, sizeof(hdr));
+  http_cookie_header(&cookie, dom, "/", hdr, sizeof(hdr));
   if (strcmp(hdr, expected) != 0)
     err = 1;
   if (strstr(hdr, "$Version") != NULL || strstr(hdr, "$Path") != NULL)
@@ -2521,6 +2574,8 @@ static const struct selftest_entry {
      st_headerlong},
     {"crange", "<raw-content-range-line> ...",
      "Content-Range parse integer safety", st_crange},
+    {"xfread-limit", "", "in-memory receive buffer size bound",
+     st_xfread_limit},
     {"savename", "<fil> <content-type> [key=value ...]",
      "local save-name for a URL", st_savename},
     {"sniff", "<content-type> <hex:..|text>", "MIME magic consistency",

tests/01_engine-xfread.test

@@ -0,0 +1,29 @@
+#!/bin/bash
+#
+
+set -euo pipefail
+
+# http_xfread1 must refuse an in-memory receive buffer that would exceed a
+# 32-bit index (hostile Content-Length, or an endless stream) rather than
+# allocate it, while still accepting a normal small size. -#test=xfread-limit
+# drives both refusal paths and the accept path.
+
+out="$(httrack -O /dev/null -#test=xfread-limit)"
+for case in bylen bygrow; do
+    echo "$out" | grep -q "${case}: refused=1 adr=null msg=In-memory content too large" || {
+        echo "FAIL ${case}: $out"
+        exit 1
+    }
+done
+
+# Exactly INT32_MAX must be refused too (the reallocs add 1).
+echo "$out" | grep -q 'boundary: msg=In-memory content too large' || {
+    echo "FAIL boundary: $out"
+    exit 1
+}
+
+# The guard must NOT fire for a legitimate small size.
+if echo "$out" | grep -q 'accept: msg=In-memory content too large'; then
+    echo "FAIL accept (guard fired on a legit size): $out"
+    exit 1
+fi

tests/49_local-cookiewall.test

@@ -0,0 +1,34 @@
+#!/bin/bash
+#
+# Cookie-wall self-redirect (#15): httrack replays a Set-Cookie to mirror the
+# real page, gives up when nothing changes, and stays bounded when it never does.
+
+set -euo pipefail
+
+: "${top_srcdir:=..}"
+
+# Unknown-ext (wall.php) and known-ext (wall.html): the wall sets the cookie once,
+# then serves real content; httrack must replay it and mirror that content.
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
+    --found 'cookiewall/wall.html' \
+    --file-matches 'cookiewall/wall.html' 'REAL-CONTENT-BEHIND-COOKIE-WALL' \
+    httrack 'BASEURL/cookiewall/index.html'
+
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
+    --found 'cookiewall2/wall.html' \
+    --file-matches 'cookiewall2/wall.html' 'REAL-CONTENT-BEHIND-COOKIE-WALL' \
+    httrack 'BASEURL/cookiewall2/index.html'
+
+# No Set-Cookie: the jar never changes, so httrack must give up at once (retry
+# stays gated) and not mirror the wall. A dropped gate would skip this log line.
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
+    --not-found 'cookiewall3/wall.html' \
+    --log-found 'loop to same filename' \
+    httrack 'BASEURL/cookiewall3/index.html'
+
+# Ever-changing cookie never satisfies the wall: httrack must stop at the retry
+# cap (bounded, no early give-up) and not mirror it.
+bash "$top_srcdir/tests/local-crawl.sh" --errors 0 \
+    --not-found 'cookiewall4/wall.html' \
+    --log-not-found 'loop to same filename' \
+    httrack 'BASEURL/cookiewall4/index.html'

tests/Makefile.am

@@ -66,6 +66,7 @@ TESTS = \
 	01_engine-urlhack.test \
 	01_engine-unescape-bounds.test \
 	01_engine-useragent.test \
+	01_engine-xfread.test \
 	01_zlib-acceptencoding.test \
 	01_zlib-cache.test \
 	01_zlib-cache-corrupt.test \
@@ -118,6 +119,7 @@ TESTS = \
 	45_local-maxsize-recv.test \
 	46_local-update-304-resume.test \
 	47_local-crange-overflow.test \
-	48_local-crange-memresume.test
+	48_local-crange-memresume.test \
+	49_local-cookiewall.test
 
 CLEANFILES = check-network_sh.cache

tests/local-server.py

@@ -1143,6 +1143,56 @@ class Handler(SimpleHTTPRequestHandler):
     def route_delayed_empty(self):
         self.send_raw(b"", "text/html")  # 200 + Content-Length: 0
 
+    # --- /cookiewall/ (#15): a self-redirect that only sets a cookie is a
+    # consent wall; httrack must replay the cookie and fetch the real page.
+    WALL_MARK = b"REAL-CONTENT-BEHIND-COOKIE-WALL"
+
+    def route_cookiewall_index(self):
+        self.send_html('\t<a href="wall.php">wall</a>')
+
+    def route_cookiewall_wall(self):
+        self._cookiewall_reply("wall.php")
+
+    # Known-extension twin: .html so the type is not delayed-resolved.
+    def route_cookiewall2_index(self):
+        self.send_html('\t<a href="wall.html">wall</a>')
+
+    def route_cookiewall2_wall(self):
+        self._cookiewall_reply("wall.html")
+
+    def _cookiewall_reply(self, location):
+        if self.request_cookies().get("gate") == "1":
+            self.send_raw(
+                b"<html><body>" + self.WALL_MARK + b"</body></html>\n", "text/html"
+            )
+        else:
+            self._wall_redirect(location, "gate=1; Path=/")
+
+    # No-cookie self-redirect: the jar never changes, so httrack must give up at
+    # once rather than re-fetch (proves the cookie-wall retry stays gated on #15).
+    def route_cookiewall3_index(self):
+        self.send_html('\t<a href="wall.php">wall</a>')
+
+    def route_cookiewall3_wall(self):
+        self._wall_redirect("wall.php", None)
+
+    # Ever-changing cookie: every hit sets a fresh value, so the jar keeps
+    # changing; httrack must stop at the loops<7 cap, not spin forever.
+    def route_cookiewall4_index(self):
+        self.send_html('\t<a href="wall.php">wall</a>')
+
+    def route_cookiewall4_wall(self):
+        nonce = int(self.request_cookies().get("gate", "0")) + 1
+        self._wall_redirect("wall.php", f"gate={nonce}; Path=/")
+
+    def _wall_redirect(self, location, set_cookie):
+        self.send_response(302, "Found")
+        self.send_header("Location", location)
+        if set_cookie is not None:
+            self.send_header("Set-Cookie", set_cookie)
+        self.send_header("Content-Length", "0")
+        self.end_headers()
+
     # -E time-limit (#481): pages that trickle far longer than any -E budget,
     # so only an engine-side abort can end the crawl.
     TRICKLE_SECONDS = 60
@@ -1359,6 +1409,14 @@ class Handler(SimpleHTTPRequestHandler):
         "/delayed/chain7.php": route_delayed_chain,
         "/delayed/chain8.php": route_delayed_chain,
         "/delayed/chain9.php": route_delayed_chain,
+        "/cookiewall/index.html": route_cookiewall_index,
+        "/cookiewall/wall.php": route_cookiewall_wall,
+        "/cookiewall2/index.html": route_cookiewall2_index,
+        "/cookiewall2/wall.html": route_cookiewall2_wall,
+        "/cookiewall3/index.html": route_cookiewall3_index,
+        "/cookiewall3/wall.php": route_cookiewall3_wall,
+        "/cookiewall4/index.html": route_cookiewall4_index,
+        "/cookiewall4/wall.php": route_cookiewall4_wall,
         "/redir/index.html": route_redir_index,
         "/redir/go.php": route_redir_go,
         "/redir/target.html": route_redir_target,