MINOR: hpack: implement the header tables management
This code deals with header insertion, retrieval and eviction, as well
as with dynamic header table defragmentation. It is functional for use
as a decoder and was heavily tested in this context. There's still some
room for optimization (eg: the defragmentation code currently does it
in place using a memcpy).
Also for now the dynamic header table is allocated using malloc() while
a pool needs to be created instead.
This code was mostly imported from https://github.com/wtarreau/http2-exp
with "hpack_" prepended in front of most names to avoid risks of conflicts.
Some small cleanups and renamings were applied during the import. This
version must be considered more recent.
Some HPACK error codes were placed here (HPACK_ERR_*), not exactly because
they're needed by the decoder but they'll be needed by all callers. Maybe
a different location should be found.
diff --git a/src/hpack-tbl.c b/src/hpack-tbl.c
new file mode 100644
index 0000000..4ca6796
--- /dev/null
+++ b/src/hpack-tbl.c
@@ -0,0 +1,380 @@
+/*
+ * HPACK header table management (RFC7541)
+ *
+ * Copyright (C) 2014-2017 Willy Tarreau <willy@haproxy.org>
+ * Copyright (C) 2017 HAProxy Technologies
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <common/hpack-huff.h>
+#include <common/hpack-tbl.h>
+#include <common/ist.h>
+
+#include <types/global.h>
+
+/* static header table as in RFC7541 Appendix A. [0] unused. */
+const struct http_hdr hpack_sht[HPACK_SHT_SIZE] = {
+ [ 1] = { .n = IST(":authority"), .v = IST("") },
+ [ 2] = { .n = IST(":method"), .v = IST("GET") },
+ [ 3] = { .n = IST(":method"), .v = IST("POST") },
+ [ 4] = { .n = IST(":path"), .v = IST("/") },
+ [ 5] = { .n = IST(":path"), .v = IST("/index.html") },
+ [ 6] = { .n = IST(":scheme"), .v = IST("http") },
+ [ 7] = { .n = IST(":scheme"), .v = IST("https") },
+ [ 8] = { .n = IST(":status"), .v = IST("200") },
+ [ 9] = { .n = IST(":status"), .v = IST("204") },
+ [10] = { .n = IST(":status"), .v = IST("206") },
+ [11] = { .n = IST(":status"), .v = IST("304") },
+ [12] = { .n = IST(":status"), .v = IST("400") },
+ [13] = { .n = IST(":status"), .v = IST("404") },
+ [14] = { .n = IST(":status"), .v = IST("500") },
+ [15] = { .n = IST("accept-charset"), .v = IST("") },
+ [16] = { .n = IST("accept-encoding"), .v = IST("gzip, deflate") },
+ [17] = { .n = IST("accept-language"), .v = IST("") },
+ [18] = { .n = IST("accept-ranges"), .v = IST("") },
+ [19] = { .n = IST("accept"), .v = IST("") },
+ [20] = { .n = IST("access-control-allow-origin"), .v = IST("") },
+ [21] = { .n = IST("age"), .v = IST("") },
+ [22] = { .n = IST("allow"), .v = IST("") },
+ [23] = { .n = IST("authorization"), .v = IST("") },
+ [24] = { .n = IST("cache-control"), .v = IST("") },
+ [25] = { .n = IST("content-disposition"), .v = IST("") },
+ [26] = { .n = IST("content-encoding"), .v = IST("") },
+ [27] = { .n = IST("content-language"), .v = IST("") },
+ [28] = { .n = IST("content-length"), .v = IST("") },
+ [29] = { .n = IST("content-location"), .v = IST("") },
+ [30] = { .n = IST("content-range"), .v = IST("") },
+ [31] = { .n = IST("content-type") , .v = IST("") },
+ [32] = { .n = IST("cookie"), .v = IST("") },
+ [33] = { .n = IST("date"), .v = IST("") },
+ [34] = { .n = IST("etag"), .v = IST("") },
+ [35] = { .n = IST("expect"), .v = IST("") },
+ [36] = { .n = IST("expires"), .v = IST("") },
+ [37] = { .n = IST("from"), .v = IST("") },
+ [38] = { .n = IST("host"), .v = IST("") },
+ [39] = { .n = IST("if-match"), .v = IST("") },
+ [40] = { .n = IST("if-modified-since"), .v = IST("") },
+ [41] = { .n = IST("if-none-match"), .v = IST("") },
+ [42] = { .n = IST("if-range"), .v = IST("") },
+ [43] = { .n = IST("if-unmodified-since"), .v = IST("") },
+ [44] = { .n = IST("last-modified"), .v = IST("") },
+ [45] = { .n = IST("link"), .v = IST("") },
+ [46] = { .n = IST("location"), .v = IST("") },
+ [47] = { .n = IST("max-forwards"), .v = IST("") },
+ [48] = { .n = IST("proxy-authenticate"), .v = IST("") },
+ [49] = { .n = IST("proxy-authorization"), .v = IST("") },
+ [50] = { .n = IST("range"), .v = IST("") },
+ [51] = { .n = IST("referer"), .v = IST("") },
+ [52] = { .n = IST("refresh"), .v = IST("") },
+ [53] = { .n = IST("retry-after"), .v = IST("") },
+ [54] = { .n = IST("server"), .v = IST("") },
+ [55] = { .n = IST("set-cookie"), .v = IST("") },
+ [56] = { .n = IST("strict-transport-security"), .v = IST("") },
+ [57] = { .n = IST("transfer-encoding"), .v = IST("") },
+ [58] = { .n = IST("user-agent"), .v = IST("") },
+ [59] = { .n = IST("vary"), .v = IST("") },
+ [60] = { .n = IST("via"), .v = IST("") },
+ [61] = { .n = IST("www-authenticate"), .v = IST("") },
+};
+
+/* returns the slot number of the oldest entry (tail). Must not be used on an
+ * empty table.
+ */
+static inline unsigned int hpack_dht_get_tail(const struct hpack_dht *dht)
+{
+ return ((dht->head + 1U < dht->used) ? dht->wrap : 0) + dht->head + 1U - dht->used;
+}
+
+#ifdef DEBUG_HPACK
+/* dump the whole dynamic header table */
+static void hpack_dht_dump(const struct hpack_dht *dht)
+{
+ int i;
+ unsigned int slot;
+ char name[4096], value[4096];
+
+ for (i = HPACK_SHT_SIZE; i < HPACK_SHT_SIZE + dht->used; i++) {
+ slot = (hpack_get_dte(dht, i - HPACK_SHT_SIZE + 1) - dht->dte);
+ fprintf(stderr, "idx=%d slot=%u name=<%s> value=<%s> addr=%u-%u\n",
+ i, slot,
+ istpad(name, hpack_idx_to_name(dht, i)).ptr,
+ istpad(value, hpack_idx_to_value(dht, i)).ptr,
+ dht->dte[slot].addr, dht->dte[slot].addr+dht->dte[slot].nlen+dht->dte[slot].vlen-1);
+ }
+}
+
+/* check for the whole dynamic header table consistency, abort on failures */
+static void hpack_dht_check_consistency(const struct hpack_dht *dht)
+{
+ unsigned slot = hpack_dht_get_tail(dht);
+ unsigned used2 = dht->used;
+ unsigned total = 0;
+
+ if (!dht->used)
+ return;
+
+ if (dht->front >= dht->wrap)
+ abort();
+
+ if (dht->used > dht->wrap)
+ abort();
+
+ if (dht->head >= dht->wrap)
+ abort();
+
+ while (used2--) {
+ total += dht->dte[slot].nlen + dht->dte[slot].vlen;
+ slot++;
+ if (slot >= dht->wrap)
+ slot = 0;
+ }
+
+ if (total != dht->total) {
+ fprintf(stderr, "%d: total=%u dht=%u\n", __LINE__, total, dht->total);
+ abort();
+ }
+}
+#endif // DEBUG_HPACK
+
+/* rebuild a new dynamic header table from <dht> with an unwrapped index and
+ * contents at the end. The new table is returned, the caller must not use the
+ * previous one anymore. NULL may be returned if no table could be allocated.
+ */
+static struct hpack_dht *hpack_dht_defrag(struct hpack_dht *dht)
+{
+ struct hpack_dht *alt_dht;
+ uint16_t old, new;
+ uint32_t addr;
+
+ /* Note: for small tables we could use alloca() instead but
+ * portability especially for large tables can be problematic.
+ */
+ alt_dht = hpack_dht_alloc(dht->size);
+ if (!alt_dht)
+ return NULL;
+
+ alt_dht->total = dht->total;
+ alt_dht->used = dht->used;
+ alt_dht->wrap = dht->used;
+
+ new = 0;
+ addr = alt_dht->size;
+
+ if (dht->used) {
+ /* start from the tail */
+ old = hpack_dht_get_tail(dht);
+ do {
+ alt_dht->dte[new].nlen = dht->dte[old].nlen;
+ alt_dht->dte[new].vlen = dht->dte[old].vlen;
+ addr -= dht->dte[old].nlen + dht->dte[old].vlen;
+ alt_dht->dte[new].addr = addr;
+
+ memcpy((void *)alt_dht + alt_dht->dte[new].addr,
+ (void *)dht + dht->dte[old].addr,
+ dht->dte[old].nlen + dht->dte[old].vlen);
+
+ old++;
+ if (old >= dht->wrap)
+ old = 0;
+ new++;
+ } while (new < dht->used);
+ }
+
+ alt_dht->front = alt_dht->head = new - 1;
+
+ memcpy(dht, alt_dht, dht->size);
+ hpack_dht_free(alt_dht);
+
+ return dht;
+}
+
+/* Purges table dht until a header field of <needed> bytes fits according to
+ * the protocol (adding 32 bytes overhead). Returns non-zero on success, zero
+ * on failure (ie: table empty but still not sufficient). It must only be
+ * called when the table is not large enough to suit the new entry and there
+ * are some entries left. In case of doubt, use dht_make_room() instead.
+ */
+int __hpack_dht_make_room(struct hpack_dht *dht, unsigned int needed)
+{
+ unsigned int used = dht->used;
+ unsigned int wrap = dht->wrap;
+ unsigned int tail;
+
+ do {
+ tail = ((dht->head + 1U < used) ? wrap : 0) + dht->head + 1U - used;
+ dht->total -= dht->dte[tail].nlen + dht->dte[tail].vlen;
+ if (tail == dht->front)
+ dht->front = dht->head;
+ used--;
+ } while (used && used * 32 + dht->total + needed + 32 > dht->size);
+
+ dht->used = used;
+
+ /* realign if empty */
+ if (!used)
+ dht->front = dht->head = 0;
+
+ /* pack the table if it doesn't wrap anymore */
+ if (dht->head + 1U >= used)
+ dht->wrap = dht->head + 1;
+
+ /* no need to check for 'used' here as if it doesn't fit, used==0 */
+ return needed + 32 <= dht->size;
+}
+
+/* tries to insert a new header <name>:<value> in front of the current head. A
+ * negative value is returned on error.
+ */
+int hpack_dht_insert(struct hpack_dht *dht, struct ist name, struct ist value)
+{
+ unsigned int used;
+ unsigned int head;
+ unsigned int prev;
+ unsigned int wrap;
+ unsigned int tail;
+ uint32_t headroom, tailroom;
+
+ if (!hpack_dht_make_room(dht, name.len + value.len))
+ return -1;
+
+ used = dht->used;
+ prev = head = dht->head;
+ wrap = dht->wrap;
+ tail = hpack_dht_get_tail(dht);
+
+ /* Now there is enough room in the table, that's guaranteed by the
+ * protocol, but not necessarily where we need it.
+ */
+
+ if (!used) {
+ /* easy, the table was empty */
+ dht->front = dht->head = 0;
+ dht->wrap = dht->used = 1;
+ dht->total = 0;
+ head = 0;
+ dht->dte[head].addr = dht->size - (name.len + value.len);
+ goto copy;
+ }
+
+ /* compute the new head, used and wrap position */
+ used++;
+ head++;
+
+ if (head >= wrap) {
+ /* head is leading the entries, we either need to push the
+ * table further or to loop back to released entries. We could
+ * force to loop back when at least half of the allocatable
+ * entries are free but in practice it never happens.
+ */
+ if ((sizeof(*dht) + (wrap + 1) * sizeof(dht->dte[0]) <= dht->dte[dht->front].addr))
+ wrap++;
+ else if (head >= used) /* there's a hole at the beginning */
+ head = 0;
+ else {
+ /* no more room, head hits tail and the index cannot be
+ * extended, we have to realign the whole table.
+ */
+ if (!hpack_dht_defrag(dht))
+ return -1;
+
+ wrap = dht->wrap + 1;
+ head = dht->head + 1;
+ prev = head - 1;
+ tail = 0;
+ }
+ }
+ else if (used >= wrap) {
+ /* we've hit the tail, we need to reorganize the index so that
+ * the head is at the end (but not necessarily move the data).
+ */
+ if (!hpack_dht_defrag(dht))
+ return -1;
+
+ wrap = dht->wrap + 1;
+ head = dht->head + 1;
+ prev = head - 1;
+ tail = 0;
+ }
+
+ /* Now we have updated head, used and wrap, we know that there is some
+ * available room at least from the protocol's perspective. This space
+ * is split in two areas :
+ *
+ * 1: if the previous head was the front cell, the space between the
+ * end of the index table and the front cell's address.
+ * 2: if the previous head was the front cell, the space between the
+ * end of the tail and the end of the table ; or if the previous
+ * head was not the front cell, the space between the end of the
+ * tail and the head's address.
+ */
+ if (prev == dht->front) {
+ /* the area was contiguous */
+ headroom = dht->dte[dht->front].addr - (sizeof(*dht) + wrap * sizeof(dht->dte[0]));
+ tailroom = dht->size - dht->dte[tail].addr - dht->dte[tail].nlen - dht->dte[tail].vlen;
+ }
+ else {
+ /* it's already wrapped so we can't store anything in the headroom */
+ headroom = 0;
+ tailroom = dht->dte[prev].addr - dht->dte[tail].addr - dht->dte[tail].nlen - dht->dte[tail].vlen;
+ }
+
+ /* We can decide to stop filling the headroom as soon as there's enough
+ * room left in the tail to suit the protocol, but tests show that in
+ * practice it almost never happens in other situations so the extra
+ * test is useless and we simply fill the headroom as long as it's
+ * available.
+ */
+ if (headroom >= name.len + value.len) {
+ /* install upfront and update ->front */
+ dht->dte[head].addr = dht->dte[dht->front].addr - (name.len + value.len);
+ dht->front = head;
+ }
+ else if (tailroom >= name.len + value.len) {
+ dht->dte[head].addr = dht->dte[tail].addr + dht->dte[tail].nlen + dht->dte[tail].vlen + tailroom - (name.len + value.len);
+ }
+ else {
+ /* need to defragment the table before inserting upfront */
+ dht = hpack_dht_defrag(dht);
+ wrap = dht->wrap + 1;
+ head = dht->head + 1;
+ dht->dte[head].addr = dht->dte[dht->front].addr - (name.len + value.len);
+ dht->front = head;
+ }
+
+ dht->wrap = wrap;
+ dht->head = head;
+ dht->used = used;
+
+ copy:
+ dht->total += name.len + value.len;
+ dht->dte[head].nlen = name.len;
+ dht->dte[head].vlen = value.len;
+
+ memcpy((void *)dht + dht->dte[head].addr, name.ptr, name.len);
+ memcpy((void *)dht + dht->dte[head].addr + name.len, value.ptr, value.len);
+ return 0;
+}