DOC: htx: Update comments in HTX files
This patch may be backported to 2.0 to have accurate comments.
(cherry picked from commit 3b21972061754a9e099b984aa4eb10dd9045aa64)
Signed-off-by: Christopher Faulet <cfaulet@haproxy.com>
diff --git a/include/common/htx.h b/include/common/htx.h
index ca2309b..27f0e5e 100644
--- a/include/common/htx.h
+++ b/include/common/htx.h
@@ -31,54 +31,104 @@
#include <common/standard.h>
/*
- * The internal representation of an HTTP message is a contiguous array
- * containing both the blocks (htx_blk) and their contents. Blocks are stored
- * starting from the end of the array while their contents are stored at the
- * beginning.
+ * The internal representation of an HTTP message, called HTX, is a structure
+ * with useful information on the message followed by a contiguous array
+ * containing parts of the message, called blocks. A block is composed of
+ * metadata (htx_blk) and the associated payload. Blocks' metadata are stored
+ * starting from the end of the array while their payload are stored at the
+ * beginning. Blocks' metadata are often simply called blocks. it is a misuse of
+ * language that's simplify explainations.
*
- * As data are sent to the peer, blocks and contents are released at the
- * edges. This free space is reused when no more space left. So blocks and
- * contents may wrap, not necessarily the same time.
*
- * An HTTP block is as well a header as a body part or a trailer part. For all
- * these types of block, a content is attached to the block. It can also be a
- * mark, like the end-of-headers or end-of-message. For these blocks, there is
- * no content but it count for a byte. It is important to not skip it when data
- * are forwarded. An HTTP block is composed of 2 fields:
+ * +-----+---------------+------------------------------+--------------+
+ * | HTX | PAYLOADS ==> | | <== HTX_BLKs |
+ * +-----+---------------+------------------------------+--------------+
+ * ^
+ * blocks[] (the beginning of the bocks array)
+ *
+ *
+ * The blocks part remains linear and sorted. You may think about it as an array
+ * with negative indexes. But, instead of using negative indexes, we use
+ * positive positions to identify a block. This position is then converted to a
+ * address relatively to the beginning of the blocks array.
+ *
+ *
+ * .....--+------------------------------+-----+-----+
+ * | ... | BLK | BLK |
+ * .....--+------------------------------+-----+-----+
+ * ^ ^
+ * Addr of the block Addr of the block
+ * at the position 1 at the position 0
+ *
+ *
+ * The payloads part is a raw space that may wrap. You never access to a block's
+ * payload directly. Instead you get a block to retrieve the address of its
+ * payload. When no more space left between blocks and payloads parts, the free
+ * space at the beginning, if any, is used.
+ *
+ *
+ * +----------- WRAPPING ------------------------+
+ * | |
+ * V |
+ * +-----+-------------+---------------+---------------++--------------+
+ * | HTX | PAYLOAD ==> | | PAYLOADS ==X || X== HTX_BLKs |
+ * +-----+-------------+---------------+---------------++--------------+
+ *
+ *
+ * The blocks part, on its side, never wrap. If we have no space to allocate a
+ * new block and if there is a hole at the beginning of the blocks part (so at
+ * the end of the blocks array), we move back all blocks.x
+ *
+ *
+ * ...+--------------+----------+ blocks ...+----------+--------------+
+ * | X== HTX_BLKS | | defrag | | <== HTX_BLKS |
+ * ...+--------------+----------+ =====> ...+----------+--------------+
+ *
+ *
+ * At the end, if payload wrapping or blocks defragmenation is not enough, some
+ * free space may be get back with a full defragmenation. This way, the holes in
+ * the middle are not reusable but count in the available free space. The only
+ * way to reuse this lost space is to fully defragmenate the HTX message.
+ *
+ * - * -
+ *
+ * An HTX block is as well a header as a body part or a trailer. For all these
+ * types of block, a payload is attached to the block. It can also be a mark,
+ * like the end-of-headers or end-of-message. For these blocks, there is no
+ * payload but it count for a byte. It is important to not skip it when data are
+ * forwarded. Metadata of an HTX block are composed of 2 fields :
*
* - .info : It a 32 bits field containing the block's type on 4 bits
- * followed by content' length. See below for details.
+ * followed by the payload length. See below for details.
*
- * - .addr : The content's address, if any, relatively to the beginning the
- * array used to store the HTTP message itself.
+ * - .addr : The payload's address, if any, relatively to the beginning the
+ * array used to store the HTX message itself.
*
- * htx_blk.info representation:
+ * htx_blk.info representation :
*
* 0b 0000 0000 0000 0000 0000 0000 0000 0000
* ---- ------------------------ ---------
- * type value (1 MB max) name length (header)
+ * type value (1 MB max) name length (header/trailer)
* ----------------------------------
* data length (256 MB max)
- * (body, method, path, version, status, reason, trailers)
+ * (body, method, path, version, status, reason)
*
- * types:
+ * types :
* - 0000 = request start-line
* - 0001 = response start-line
* - 0010 = header
* - 0011 = pseudo-header ou "special" header
* - 0100 = end-of-headers
* - 0101 = data
- * - 0110 = end-of-data
- * - 0111 = trailer
+ * - 0110 = trailer
+ * - 0111 = end-of-trailers
* - 1000 = end-of-message
* ...
- * - 1101 = out-of-band
- * - 1110 = error
* - 1111 = unused
*
*/
-/*HTX start-line flags */
+/* HTX start-line flags */
#define HTX_SL_F_NONE 0x00000000
#define HTX_SL_F_IS_RESP 0x00000001 /* It is the response start-line (unset means the request one) */
#define HTX_SL_F_XFER_LEN 0x00000002 /* The message xfer size can be dertermined */
@@ -93,11 +143,11 @@
/* HTX flags */
#define HTX_FL_NONE 0x00000000
-#define HTX_FL_PARSING_ERROR 0x00000001
-#define HTX_FL_UPGRADE 0x00000002
+#define HTX_FL_PARSING_ERROR 0x00000001 /* Set when a parsing error occurred */
+#define HTX_FL_UPGRADE 0x00000002 /* Set when an upgrade is in progress */
-/* HTTP block's type (max 15). */
+/* HTX block's type (max 15). */
enum htx_blk_type {
HTX_BLK_REQ_SL = 0, /* Request start-line */
HTX_BLK_RES_SL = 1, /* Response start-line */
@@ -111,17 +161,19 @@
HTX_BLK_UNUSED = 15, /* unused/removed block */
};
-/* One HTTP block descriptor */
+/* One HTX block descriptor */
struct htx_blk {
- uint32_t addr; /* relative storage address of a data block */
- uint32_t info; /* information about data stored */
+ uint32_t addr; /* relative storage address of the block's payload */
+ uint32_t info; /* information about the block (type, length) */
};
+/* Composite return value used by some HTX functions */
struct htx_ret {
- int32_t ret;
- struct htx_blk *blk;
+ int32_t ret; /* A numerical value */
+ struct htx_blk *blk; /* An HTX block */
};
+/* HTX start-line */
struct htx_sl {
unsigned int flags; /* HTX_SL_F_* */
union {
@@ -135,8 +187,9 @@
/* XXX 2 bytes unused */
- int32_t hdrs_bytes; /* Bytes held by all headers from this start-line
- * to the corresponding EOH. -1 if unknown */
+ int32_t hdrs_bytes; /* Bytes held by all headers, as seen by the mux
+ * during parsing, from this start-line to the
+ * corresponding EOH. -1 if unknown */
unsigned int len[3]; /* length of differnt parts of the start-line */
char l[0];
@@ -150,8 +203,8 @@
* i.e. [ used * sizeof(struct htx_blk *) ] */
uint32_t used; /* number of blocks in use */
- uint32_t tail; /* last inserted block */
- uint32_t head; /* older inserted block */
+ uint32_t tail; /* newest inserted block. -1 if the HTX message is empty */
+ uint32_t head; /* oldest inserted block. -1 if the HTX message is empty */
uint32_t tail_addr; /* start address of the free space in front of the the blocks table */
uint32_t head_addr; /* start address of the free space at the beginning */
@@ -205,7 +258,7 @@
int htx_data_to_h1(const struct ist data, struct buffer *chk, int chunked);
/* Functions and macros to get parts of the start-line or legnth of these
- * parts
+ * parts. Request and response start-lines are both composed of 3 parts.
*/
#define HTX_SL_LEN(sl) ((sl)->len[0] + (sl)->len[1] + (sl)->len[2])
@@ -276,19 +329,21 @@
return htx_sl_p3(sl);
}
-/* Returns the array index of a block given its position <pos> */
+/* Converts a position to the corresponding relative address */
static inline uint32_t htx_pos_to_idx(const struct htx *htx, uint32_t pos)
{
return ((htx->size / sizeof(htx->blocks[0])) - pos - 1);
}
-/* Returns the position of the block <blk> */
+/* Returns the position of the block <blk>. It is the caller responsibility to
+ * be sure <blk> is part of <htx>. */
static inline uint32_t htx_get_blk_pos(const struct htx *htx, const struct htx_blk *blk)
{
return (htx->blocks + (htx->size / sizeof(htx->blocks[0])) - blk - 1);
}
-/* Returns the block at the position <pos> */
+/* Returns the block at the position <pos>. It is the caller responsibility to
+ * be sure the block at the position <pos> exists. */
static inline struct htx_blk *htx_get_blk(const struct htx *htx, uint32_t pos)
{
return ((struct htx_blk *)(htx->blocks) + htx_pos_to_idx(htx, pos));
@@ -532,8 +587,8 @@
return ((void *)htx->blocks + blk->addr);
}
-/* Returns the name of the block <blk>, only if it is a header. Otherwise it
- * returns an empty name.
+/* Returns the name of the block <blk>, only if it is a header or a
+ * trailer. Otherwise it returns an empty string.
*/
static inline struct ist htx_get_blk_name(const struct htx *htx, const struct htx_blk *blk)
{
@@ -555,7 +610,7 @@
/* Returns the value of the block <blk>, depending on its type. If there is no
- * value, an empty one is retruned.
+ * value (for end-of blocks), an empty one is retruned.
*/
static inline struct ist htx_get_blk_value(const struct htx *htx, const struct htx_blk *blk)
{
@@ -604,7 +659,7 @@
return (htx->used * sizeof(htx->blocks[0]));
}
-/* Returns the space used (data + metadata) in <htx> */
+/* Returns the space used (payload + metadata) in <htx> */
static inline uint32_t htx_used_space(const struct htx *htx)
{
return (htx->data + htx_meta_space(htx));
@@ -650,6 +705,7 @@
return 0;
}
+/* Resets an HTX message */
static inline void htx_reset(struct htx *htx)
{
htx->data = htx->used = htx->tail = htx->head = 0;
@@ -659,7 +715,7 @@
htx->first = -1;
}
-/* returns the available room for raw data in buffer <buf> once HTX overhead is
+/* Returns the available room for raw data in buffer <buf> once HTX overhead is
* taken into account (one HTX header and two blocks). The purpose is to figure
* the optimal fill length to avoid copies.
*/
@@ -678,9 +734,11 @@
/* Returns an HTX message using the buffer <buf>. Unlike htx_from_buf(), this
- * function does not update to the buffer.
- * Note that it always returns a valid pointer, either to an initialized buffer
- * or to the empty buffer.
+ * function does not update the buffer. So if the HTX message is updated, the
+ * caller must call htx_to_buf() to be sure to also update the underlying buffer
+ * accordingly. Note that it always returns a valid pointer, either to an
+ * initialized buffer or to the empty buffer. This function must always be
+ * called with a buffer containing an HTX message (or an empty buffer).
*/
static inline struct htx *htxbuf(const struct buffer *buf)
{
@@ -697,9 +755,11 @@
}
/* Returns an HTX message using the buffer <buf>. <buf> is updated to appear as
- * full. It is the caller responsibility to call htx_to_buf() when it finish to
- * manipulate the HTX message to update <buf> accordingly. The returned pointer
- * is always valid.
+ * full. It should be used when you want to add something into the HTX message,
+ * so the call to htx_to_buf() may be skipped. But, it is the caller
+ * responsibility to call htx_to_buf() to reset <buf> if it is relevant. The
+ * returned pointer is always valid. This function must always be called with a
+ * buffer containing an HTX message (or an empty buffer).
*
* The caller can call htxbuf() function to avoid any update of the buffer.
*/
@@ -711,7 +771,7 @@
return htx;
}
-/* Upate <buf> accordingly to the HTX message <htx> */
+/* Update <buf> accordingly to the HTX message <htx> */
static inline void htx_to_buf(struct htx *htx, struct buffer *buf)
{
if (!htx->used && !(htx->flags & (HTX_FL_PARSING_ERROR|HTX_FL_UPGRADE))) {
@@ -755,6 +815,7 @@
};
}
+/* For debugging purpose */
static inline void htx_dump(struct htx *htx)
{
int32_t pos;
diff --git a/src/htx.c b/src/htx.c
index 502184a..c29a66d 100644
--- a/src/htx.c
+++ b/src/htx.c
@@ -15,10 +15,10 @@
struct htx htx_empty = { .size = 0, .data = 0, .used = 0 };
-/* Defragments an HTTP message, removing unused blocks and unwrapping blocks and
- * their contents. A temporary message is used to do so. This function never
- * fails. if <blk> is not NULL, we replace it by the new block address, after
- * the defragmentation. The new <blk> is returned.
+/* Defragments an HTX message. It removes unused blocks and unwraps the payloads
+ * part. A temporary buffer is used to do so. This function never fails. if
+ * <blk> is not NULL, we replace it by the new block address, after the
+ * defragmentation. The new <blk> is returned.
*/
/* TODO: merge data blocks into one */
struct htx_blk *htx_defrag(struct htx *htx, struct htx_blk *blk)
@@ -54,7 +54,7 @@
if (htx->first == old)
first = new;
- /* if <blk> is defined, set its new position */
+ /* if <blk> is defined, save its new position */
if (blk != NULL && blk == oldblk)
blkpos = new;
@@ -75,6 +75,10 @@
return ((blkpos == -1) ? NULL : htx_get_blk(htx, blkpos));
}
+/* Degragments HTX blocks of an HTX message. Payloads part is keep untouched
+ * here. This function will move back all blocks starting at the position 0,
+ * removing unused blocks. It must never be called with an empty message.
+ */
static void htx_defrag_blks(struct htx *htx)
{
int32_t pos, new;
@@ -103,11 +107,11 @@
htx->tail = new - 1;
}
-/* Reserves a new block in the HTTP message <htx> with a content of <blksz>
+/* Reserves a new block in the HTX message <htx> with a content of <blksz>
* bytes. If there is not enough space, NULL is returned. Otherwise the reserved
- * block is returned and the HTTP message is updated. Space for this new block
- * is reserved in the HTTP message. But it is the caller responsibility to set
- * right info in the block to reflect the stored data.
+ * block is returned and the HTX message is updated. Space for this new block is
+ * reserved in the HTX message. But it is the caller responsibility to set right
+ * info in the block to reflect the stored data.
*/
static struct htx_blk *htx_reserve_nxblk(struct htx *htx, uint32_t blksz)
{
@@ -175,7 +179,7 @@
}
else {
defrag:
- /* need to defragment the table before inserting upfront */
+ /* need to defragment the message before inserting upfront */
htx_defrag(htx, NULL);
tail = htx->tail + 1;
used = htx->used + 1;
@@ -203,12 +207,12 @@
* expanded by <delta> bytes and we need find where this expansion will be
* performed. It can be a compression if <delta> is negative. This function only
* updates all addresses. The caller have the responsibility to performe the
- * expansion and update the block and the HTX message accordingly. No error msut
- * occurre. It returns following values:
+ * expansion and update the block and the HTX message accordingly. No error must
+ * occurr. It returns following values:
*
* 0: The expansion cannot be performed, there is not enough space.
*
- * 1: the expansion must be performed in place, there is enought space after
+ * 1: the expansion must be performed in place, there is enougth space after
* the block's payload to handle it. This is especially true if it is a
* compression and not an expension.
*
@@ -299,8 +303,8 @@
return ret;
}
-/* Adds a new block of type <type> in the HTTP message <htx>. Its content size
- * is passed but it is the caller responsibility to do the copy.
+/* Adds a new block of type <type> in the HTX message <htx>. Its content size is
+ * passed but it is the caller responsibility to do the copy.
*/
struct htx_blk *htx_add_blk(struct htx *htx, enum htx_blk_type type, uint32_t blksz)
{
@@ -315,9 +319,9 @@
return blk;
}
-/* Removes the block <blk> from the HTTP message <htx>. The function returns the
- * block following <blk> or NULL if <blk> is the last block or the last
- * inserted one.
+/* Removes the block <blk> from the HTX message <htx>. The function returns the
+ * block following <blk> or NULL if <blk> is the last block or the last inserted
+ * one.
*/
struct htx_blk *htx_remove_blk(struct htx *htx, struct htx_blk *blk)
{
@@ -340,7 +344,7 @@
blk->info = ((uint32_t)HTX_BLK_UNUSED << 28);
}
- /* There is at least 2 blocks, so tail is always >= 0 */
+ /* There is at least 2 blocks, so tail is always > 0 */
if (pos == htx->head) {
/* move the head forward */
htx->used--;
@@ -391,8 +395,8 @@
return blk;
}
-/* Truncate all blocks after the one containing the offset <offset>. This last
- * one is truncated too.
+/* Removes all blocks after the one containing the offset <offset>. This last
+ * one may be truncated if it is a DATA block.
*/
void htx_truncate(struct htx *htx, uint32_t offset)
{
@@ -414,11 +418,11 @@
blk = htx_remove_blk(htx, blk);
}
-/* Drain <count> bytes from the HTX message <htx>. DATA blocks will be cut if
- * necessary. Others blocks will be removed at once if <count> is large
- * enough. The function returns an htx_ret with the first block remaing in the
- * messsage and the amount of data drained. If everything is removed,
- * htx_ret.blk is set to NULL.
+/* Drains <count> bytes from the HTX message <htx>. If the last block is a DATA
+ * block, it will be cut if necessary. Others blocks will be removed at once if
+ * <count> is large enough. The function returns an htx_ret with the first block
+ * remaing in the messsage and the amount of data drained. If everything is
+ * removed, htx_ret.blk is set to NULL.
*/
struct htx_ret htx_drain(struct htx *htx, uint32_t count)
{
@@ -565,19 +569,19 @@
htx_set_blk_value_len(blk, v.len + delta);
htx->data += delta;
- if (ret == 1) {
+ if (ret == 1) { /* Replace in place */
if (delta <= 0) {
- /* compression: copy new data first */
+ /* compression: copy new data first then move the end */
memcpy(old.ptr, new.ptr, new.len);
memmove(old.ptr + new.len, old.ptr + old.len, (v.ptr + v.len) - (old.ptr + old.len));
}
else {
- /* expansion: move the end first */
+ /* expansion: move the end first then copy new data */
memmove(old.ptr + new.len, old.ptr + old.len, (v.ptr + v.len) - (old.ptr + old.len));
memcpy(old.ptr, new.ptr, new.len);
}
}
- else if (ret == 2) {
+ else if (ret == 2) { /* New address but no defrag */
void *ptr = htx_get_blk_ptr(htx, blk);
/* Copy the name, if any */
@@ -595,7 +599,7 @@
/* Copy value after old part, if any */
memcpy(ptr, old.ptr + old.len, (v.ptr + v.len) - (old.ptr + old.len));
}
- else {
+ else { /* Do a degrag first */
struct buffer *tmp = get_trash_chunk();
/* Copy the header name, if any */
@@ -671,7 +675,7 @@
if (!max)
break;
if (sz > max) {
- /* Headers must be fully copied */
+ /* Only DATA blocks can be partially xferred */
if (type != HTX_BLK_DATA)
break;
sz = max;
@@ -686,7 +690,7 @@
count -= sizeof(dstblk) + sz;
if (blk->info != info) {
- /* Partial move: don't remove <blk> from <src> but
+ /* Partial xfer: don't remove <blk> from <src> but
* resize its content */
htx_cut_data_blk(src, blk, sz);
break;
@@ -695,7 +699,6 @@
blk = htx_remove_blk(src, blk);
if (type == mark)
break;
-
}
end:
@@ -728,6 +731,9 @@
blk->info = (type << 28) + (value.len << 8) + name.len;
htx->data += delta;
+ /* Replace in place or at a new address is the same. We replace all the
+ * header (name+value). Only take care to defrag the message if
+ * necessary. */
if (ret == 3)
blk = htx_defrag(htx, blk);
@@ -772,6 +778,8 @@
htx_set_blk_value_len(blk, sz+delta);
htx->data += delta;
+ /* Replace in place or at a new address is the same. We replace all the
+ * start-line. Only take care to defrag the message if necessary. */
if (ret == 3)
blk = htx_defrag(htx, blk);
@@ -848,7 +856,7 @@
}
/* Adds an HTX block of type TLR in <htx>. It returns the new block on
- * success. Otherwise, it returns NULL. The header name is always lower cased.
+ * success. Otherwise, it returns NULL. The trailer name is always lower cased.
*/
struct htx_blk *htx_add_trailer(struct htx *htx, const struct ist name,
const struct ist value)
@@ -882,6 +890,9 @@
return blk;
}
+/* Add all headers from the list <hdrs> into the HTX message <htx>, followed by
+ * the EOH. On sucess, it returns the last block inserted (the EOH), otherwise
+ * NULL is returned. */
struct htx_blk *htx_add_all_headers(struct htx *htx, const struct http_hdr *hdrs)
{
int i;
@@ -893,6 +904,9 @@
return htx_add_endof(htx, HTX_BLK_EOH);
}
+/* Add all trailers from the list <hdrs> into the HTX message <htx>, followed by
+ * the EOT. On sucess, it returns the last block inserted (the EOT), otherwise
+ * NULL is returned. */
struct htx_blk *htx_add_all_trailers(struct htx *htx, const struct http_hdr *hdrs)
{
int i;
@@ -904,7 +918,7 @@
return htx_add_endof(htx, HTX_BLK_EOT);
}
-/* Adds an HTX block of type EOH or EOM in <htx>. It returns the new block
+/* Adds an HTX block of type EOH, EOT, or EOM in <htx>. It returns the new block
* on success. Otherwise, it returns NULL.
*/
struct htx_blk *htx_add_endof(struct htx *htx, enum htx_blk_type type)
@@ -1047,9 +1061,8 @@
*ref = pblk;
}
-/* Appends the H1 representation of the request line block <blk> to the
- * chunk <chk>. It returns 1 if data are successfully appended, otherwise it
- * returns 0.
+/* Appends the H1 representation of the request line <sl> to the chunk <chk>. It
+ * returns 1 if data are successfully appended, otherwise it returns 0.
*/
int htx_reqline_to_h1(const struct htx_sl *sl, struct buffer *chk)
{
@@ -1070,9 +1083,8 @@
return 1;
}
-/* Appends the H1 representation of the status line block <blk> to the chunk
- * <chk>. It returns 1 if data are successfully appended, otherwise it
- * returns 0.
+/* Appends the H1 representation of the status line <sl> to the chunk <chk>. It
+ * returns 1 if data are successfully appended, otherwise it returns 0.
*/
int htx_stline_to_h1(const struct htx_sl *sl, struct buffer *chk)
{
@@ -1092,9 +1104,9 @@
return 1;
}
-/* Appends the H1 representation of the header block <blk> to the chunk
- * <chk>. It returns 1 if data are successfully appended, otherwise it returns
- * 0.
+/* Appends the H1 representation of the header <n> witht the value <v> to the
+ * chunk <chk>. It returns 1 if data are successfully appended, otherwise it
+ * returns 0.
*/
int htx_hdr_to_h1(const struct ist n, const struct ist v, struct buffer *chk)
{
@@ -1109,9 +1121,9 @@
return 1;
}
-/* Appends the H1 representation of the data block <blk> to the chunk
- * <chk>. If <chunked> is non-zero, it emits HTTP/1 chunk-encoded data. It
- * returns 1 if data are successfully appended, otherwise it returns 0.
+/* Appends the H1 representation of the data <data> to the chunk <chk>. If
+ * <chunked> is non-zero, it emits HTTP/1 chunk-encoded data. It returns 1 if
+ * data are successfully appended, otherwise it returns 0.
*/
int htx_data_to_h1(const struct ist data, struct buffer *chk, int chunked)
{