On Tue, Aug 05, 2025 at 05:46:00PM +0200, Laurent Vivier wrote:

These utilities enhance iov_tail manipulation, useful for efficient packet processing by enabling iovec array cloning and header stripping without data copies.

- iov_tail_drop(): Discards a specified number of bytes from the beginning of an iov_tail by advancing its internal offset and pruning consumed elements.

- iov_tail_clone(): Clone an iov_tail into an iovec array, adjusting the first iovec entry to remove the iov_tail offset.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson To the extent that the functions look useful and the implementations look correct; a couple of suggested improvements below.

--- iov.c | 52 ++++++++++++++++++++++++++++++++++++++++++++++++++++ iov.h | 3 +++ 2 files changed, 55 insertions(+)

diff --git a/iov.c b/iov.c index edf0444d1955..9d282d4af461 100644 --- a/iov.c +++ b/iov.c @@ -192,6 +192,21 @@ size_t iov_tail_size(struct iov_tail *tail) return iov_size(tail->iov, tail->cnt) - tail->off; }

+/** + * iov_tail_drop() - Discard a header from an IOV tail

Maybe _drop_hdr() to clarify that it's dropping specifically from the start. Or maybe not, I'm torn between brevity and specificity.

+ * @tail: IO vector tail + * @len: length to move the head of the tail + * + * Return: true if the item still contains any bytes, otherwise false + */ +/* cppcheck-suppress unusedFunction */ +bool iov_tail_drop(struct iov_tail *tail, size_t len) +{ + tail->off = tail->off + len; + + return iov_tail_prune(tail); +} + /** * iov_peek_header_() - Get pointer to a header from an IOV tail * @tail: IOV tail to get header from @@ -248,3 +263,40 @@ void *iov_remove_header_(struct iov_tail *tail, size_t len, size_t align) tail->off = tail->off + len; return p; } + +/** + * iov_tail_clone() - Assign iov references referencing a subset of the data + * in an iov_tail + * + * @dst_iov: Pointer to the destination array of struct iovec describing + * the scatter/gather I/O vector to shallow copy to. + * @dst_iov_cnt: Maximum number of elements in the destination iov array. + * @tail: Pointer to the source iov_tail + * + * Return: the number of elements successfully referenced from the destination + * iov array, a negative value if there is not enough room in the + * destination iov array + */ +/* cppcheck-suppress unusedFunction */ +ssize_t iov_tail_clone(struct iovec *dst_iov, size_t dst_iov_cnt, + struct iov_tail *tail) +{ + const struct iovec *iov = &tail->iov[0]; + size_t iov_cnt = tail->cnt; + size_t offset = tail->off; + unsigned int i, j; + + i = iov_skip_bytes(iov, iov_cnt, offset, &offset);

If you prune first, then you can assume that offset is within the first entry, and you'll also know that you need exactly tail->cnt entries in the destination. You could then: if (tail->off != 0) /* copy partial first entry */ /* memcpy the rest */

+ /* assign iov references referencing a subset of the source one */ + for (j = 0; i < iov_cnt && j < dst_iov_cnt; i++, j++) { + dst_iov[j].iov_base = (char *)iov[i].iov_base + offset; + dst_iov[j].iov_len = iov[i].iov_len - offset; + offset = 0; + } + + if (j == dst_iov_cnt && i != iov_cnt) + return -1; + + return j; +} diff --git a/iov.h b/iov.h index 3fc96ab9755a..bf9820ac52ab 100644 --- a/iov.h +++ b/iov.h @@ -72,8 +72,11 @@ struct iov_tail {

bool iov_tail_prune(struct iov_tail *tail); size_t iov_tail_size(struct iov_tail *tail); +bool iov_tail_drop(struct iov_tail *tail, size_t len); void *iov_peek_header_(struct iov_tail *tail, size_t len, size_t align); void *iov_remove_header_(struct iov_tail *tail, size_t len, size_t align); +ssize_t iov_tail_clone(struct iovec *dst_iov, size_t dst_iov_cnt, + struct iov_tail *tail);

/** * IOV_PEEK_HEADER() - Get typed pointer to a header from an IOV tail

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:01PM +0200, Laurent Vivier wrote:

Provide a temporary variable of the wanted type to store the header if the memory in the iovec array is not contiguous.

Signed-off-by: Laurent Vivier --- iov.c | 55 +++++++++++++++++++++++++++++++++++++++++++++---------- iov.h | 55 +++++++++++++++++++++++++++++++++++++++++-------------- tcp_buf.c | 2 +- 3 files changed, 87 insertions(+), 25 deletions(-)

diff --git a/iov.c b/iov.c index 9d282d4af461..d39bb099fa69 100644 --- a/iov.c +++ b/iov.c @@ -109,7 +109,7 @@ size_t iov_from_buf(const struct iovec *iov, size_t iov_cnt, * * Return: the number of bytes successfully copied. */ -/* cppcheck-suppress unusedFunction */ +/* cppcheck-suppress [staticFunction] */ size_t iov_to_buf(const struct iovec *iov, size_t iov_cnt, size_t offset, void *buf, size_t bytes) { @@ -127,6 +127,7 @@ size_t iov_to_buf(const struct iovec *iov, size_t iov_cnt, /* copying data */ for (copied = 0; copied < bytes && i < iov_cnt; i++) { size_t len = MIN(iov[i].iov_len - offset, bytes - copied); + /* NOLINTNEXTLINE(clang-analyzer-core.NonNullParamChecker) */

Which parameter was cppcheck complaining about?

memcpy((char *)buf + copied, (char *)iov[i].iov_base + offset, len); copied += len; @@ -208,7 +209,7 @@ bool iov_tail_drop(struct iov_tail *tail, size_t len) }

/** - * iov_peek_header_() - Get pointer to a header from an IOV tail + * iov_check_header() - Check if a header can be accessed * @tail: IOV tail to get header from * @len: Length of header to get, in bytes * @align: Required alignment of header, in bytes @@ -219,8 +220,7 @@ bool iov_tail_drop(struct iov_tail *tail, size_t len) * overruns the IO vector, is not contiguous or doesn't have the * requested alignment. */ -/* cppcheck-suppress [staticFunction,unmatchedSuppression] */ -void *iov_peek_header_(struct iov_tail *tail, size_t len, size_t align) +static void *iov_check_header(struct iov_tail *tail, size_t len, size_t align) { char *p;

@@ -240,27 +240,62 @@ void *iov_peek_header_(struct iov_tail *tail, size_t len, size_t align) return p; }

+/** + * iov_peek_header_() - Get pointer to a header from an IOV tail + * @tail: IOV tail to get header from + * @v: Temporary memory to use if the memory in @tail + * is discontinuous + * @len: Length of header to get, in bytes + * @align: Required alignment of header, in bytes + * + * @tail may be pruned, but will represent the same bytes as before. + * + * Return: pointer to the first @len logical bytes of the tail, or to + * a copy if that overruns the IO vector, is not contiguous or + * doesn't have the requested alignment. NULL if that overruns the + * IO vector. + */ +/* cppcheck-suppress [staticFunction,unmatchedSuppression] */ +void *iov_peek_header_(struct iov_tail *tail, void *v, size_t len, size_t align) +{ + char *p = iov_check_header(tail, len, align); + size_t l; + + if (p) + return p; + + l = iov_to_buf(tail->iov, tail->cnt, tail->off, v, len); + if (l != len) + return NULL; + + return v; +} + /** * iov_remove_header_() - Remove a header from an IOV tail * @tail: IOV tail to remove header from (modified) + * @v: Temporary memory to use if the memory in @tail + * is discontinuous * @len: Length of header to remove, in bytes * @align: Required alignment of header, in bytes * * On success, @tail is updated so that it longer includes the bytes of the * returned header. * - * Return: pointer to the first @len logical bytes of the tail, NULL if that - * overruns the IO vector, is not contiguous or doesn't have the - * requested alignment. + * Return: pointer to the first @len logical bytes of the tail, or to + * a copy if that overruns the IO vector, is not contiguous or + * doesn't have the requested alignment. NULL if that overruns the + * IO vector. */ -void *iov_remove_header_(struct iov_tail *tail, size_t len, size_t align) +void *iov_remove_header_(struct iov_tail *tail, void *v, size_t len, size_t align) { - char *p = iov_peek_header_(tail, len, align); + char *p = iov_peek_header_(tail, v, len, align);

if (!p) return NULL;

tail->off = tail->off + len;

Do you want to use your new iov_tail_drop() here?

+ return p; }

@@ -275,7 +310,7 @@ void *iov_remove_header_(struct iov_tail *tail, size_t len, size_t align) * * Return: the number of elements successfully referenced from the destination * iov array, a negative value if there is not enough room in the - * destination iov array + * destination iov array */ /* cppcheck-suppress unusedFunction */ ssize_t iov_tail_clone(struct iovec *dst_iov, size_t dst_iov_cnt, diff --git a/iov.h b/iov.h index bf9820ac52ab..ccdb690ef3f1 100644 --- a/iov.h +++ b/iov.h @@ -70,41 +70,68 @@ struct iov_tail { #define IOV_TAIL(iov_, cnt_, off_) \ (struct iov_tail){ .iov = (iov_), .cnt = (cnt_), .off = (off_) }

+/** + * IOV_TAIL_FROM_BUF() - Create a new IOV tail from a buffer + * @buf_: Buffer address to use in the iovec + * @len_: Buffer size + * @off_: Byte offset in the buffer where the tail begins + */ +#define IOV_TAIL_FROM_BUF(buf_, len_, off_) \ + IOV_TAIL((&(const struct iovec){ .iov_base = (buf_), \ + .iov_len = (len_) }), \ + 1, \ + (off_)) +

This seems unrelated to the rest of the patch, does it belong in a different one? Also, given that you're constructing the iovec entries yourself, why not set tail->off to zero and update iov_base and len accordingly.

bool iov_tail_prune(struct iov_tail *tail); size_t iov_tail_size(struct iov_tail *tail); bool iov_tail_drop(struct iov_tail *tail, size_t len); -void *iov_peek_header_(struct iov_tail *tail, size_t len, size_t align); -void *iov_remove_header_(struct iov_tail *tail, size_t len, size_t align); +void *iov_peek_header_(struct iov_tail *tail, void *v, size_t len, size_t align); +void *iov_remove_header_(struct iov_tail *tail, void *v, size_t len, size_t align); ssize_t iov_tail_clone(struct iovec *dst_iov, size_t dst_iov_cnt, struct iov_tail *tail);

/** * IOV_PEEK_HEADER() - Get typed pointer to a header from an IOV tail * @tail_: IOV tail to get header from - * @type_: Data type of the header + * @var_: Temporary buffer of the type of the header to use if + * the memory in the iovec array is not contiguous. * * @tail_ may be pruned, but will represent the same bytes as before. * - * Return: pointer of type (@type_ *) located at the start of @tail_, NULL if - * we can't get a contiguous and aligned pointer. + * Return: pointer of type (@type_ *) located at the start of @tail_ + * or to @var_ if iovec memory is not contiguous, NULL if + * that overruns the iovec.

The NULL case can't happen if given a variable of the right type.

*/ -#define IOV_PEEK_HEADER(tail_, type_) \ - ((type_ *)(iov_peek_header_((tail_), \ - sizeof(type_), __alignof__(type_)))) + +#define IOV_PEEK_HEADER(tail_, var_) \ + ((__typeof__(var_) *)(iov_peek_header_((tail_), &(var_), \ + sizeof(var_), \ + __alignof__(var_))))

/** * IOV_REMOVE_HEADER() - Remove and return typed header from an IOV tail * @tail_: IOV tail to remove header from (modified) - * @type_: Data type of the header to remove + * @var_: Temporary buffer of the type of the header to use if + * the memory in the iovec array is not contiguous. * * On success, @tail_ is updated so that it longer includes the bytes of the * returned header. * - * Return: pointer of type (@type_ *) located at the old start of @tail_, NULL - * if we can't get a contiguous and aligned pointer. + * Return: pointer of type (@type_ *) located at the start of @tail_ + * or to @var_ if iovec memory is not contiguous, NULL if + * that overruns the iovec.

Again, NULL case can't happen.

+ */ + +#define IOV_REMOVE_HEADER(tail_, var_) \ + ((__typeof__(var_) *)(iov_remove_header_((tail_), &(var_), \ + sizeof(var_), __alignof__(var_)))) + +/** IOV_DROP_HEADER() - Remove a typed header from an IOV tail + * @tail_: IOV tail to remove header from (modified) + * @type_: Data type of the header to remove + * + * Return: true if the tail still contains any bytes, otherwise false */ -#define IOV_REMOVE_HEADER(tail_, type_) \ - ((type_ *)(iov_remove_header_((tail_), \ - sizeof(type_), __alignof__(type_)))) +#define IOV_DROP_HEADER(tail_, type_) iov_tail_drop((tail_), sizeof(type_))

#endif /* IOVEC_H */ diff --git a/tcp_buf.c b/tcp_buf.c index d1fca676c9a7..bc898de86919 100644 --- a/tcp_buf.c +++ b/tcp_buf.c @@ -160,7 +160,7 @@ static void tcp_l2_buf_fill_headers(const struct tcp_tap_conn *conn, uint32_t seq, bool no_tcp_csum) { struct iov_tail tail = IOV_TAIL(&iov[TCP_IOV_PAYLOAD], 1, 0); - struct tcphdr *th = IOV_REMOVE_HEADER(&tail, struct tcphdr); + struct tcphdr th_storage, *th = IOV_REMOVE_HEADER(&tail, th_storage); struct tap_hdr *taph = iov[TCP_IOV_TAP].iov_base; const struct flowside *tapside = TAPFLOW(conn); const struct in_addr *a4 = inany_v4(&tapside->oaddr);

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:02PM +0200, Laurent Vivier wrote:

Use IOV_PEEK_HEADER() to get the ethernet header from the iovec.

Move the workaround about multiple iovec array from vu_handle_tx() to tap_add_packet(). Removing the offset out of the iovec array should reduce the iovec count to 1.

Signed-off-by: Laurent Vivier Reviewed-by: David Gibson --- iov.c | 1 - pcap.c | 1 + tap.c | 30 +++++++++++++++++++++--------- tap.h | 3 +-- vu_common.c | 26 +++++--------------------- 5 files changed, 28 insertions(+), 33 deletions(-)

diff --git a/iov.c b/iov.c index d39bb099fa69..97e4ea733540 100644 --- a/iov.c +++ b/iov.c @@ -200,7 +200,6 @@ size_t iov_tail_size(struct iov_tail *tail) * * Return: true if the item still contains any bytes, otherwise false */ -/* cppcheck-suppress unusedFunction */ bool iov_tail_drop(struct iov_tail *tail, size_t len) { tail->off = tail->off + len; diff --git a/pcap.c b/pcap.c index 46d11a2a6daa..03adc4c55f4b 100644 --- a/pcap.c +++ b/pcap.c @@ -74,6 +74,7 @@ static void pcap_frame(const struct iovec *iov, size_t iovcnt, * @pkt: Pointer to data buffer, including L2 headers * @l2len: L2 frame length */ +/* cppcheck-suppress unusedFunction */

Are we going to get any new users of this? Maybe we can just remove it now.

void pcap(const char *pkt, size_t l2len) { struct iovec iov = { (char *)pkt, l2len }; diff --git a/tap.c b/tap.c index 6db5d88b1760..c5520bf3bc76 100644 --- a/tap.c +++ b/tap.c @@ -1070,24 +1070,29 @@ void tap_handler(struct ctx *c, const struct timespec *now) /** * tap_add_packet() - Queue/capture packet, update notion of guest MAC address * @c: Execution context - * @l2len: Total L2 packet length - * @p: Packet buffer + * @data: Packet to add to the pool * @now: Current timestamp */ -void tap_add_packet(struct ctx *c, ssize_t l2len, char *p, +void tap_add_packet(struct ctx *c, struct iov_tail *data, const struct timespec *now) { + struct ethhdr eh_storage; const struct ethhdr *eh;

- pcap(p, l2len); + pcap_iov(data->iov, data->cnt, data->off);

- eh = (struct ethhdr *)p; + eh = IOV_PEEK_HEADER(data, eh_storage); + if (!eh) + return;

if (memcmp(c->guest_mac, eh->h_source, ETH_ALEN)) { memcpy(c->guest_mac, eh->h_source, ETH_ALEN); proto_update_l2_buf(c->guest_mac, NULL); }

+ iov_tail_prune(data); + ASSERT(data->cnt == 1); /* packet_add() doesn't support iovec */

So.. the IOV_PEEK_HEADER will have already pruned, but I think it would be a layering violation to assume that here. I'm wondering if we should change the invariants for the iov_tail structure and say they must *always* be pruned, outside of the innards of the iov_tail handling. That would mean pruning in all "constructors" of an iov_tail. It would also mean moving the prune to the end of iov_tail_drop() instead of the beginning. It would have the nice bonus that peek_header could take the iov_tail as a const *.

switch (ntohs(eh->h_proto)) { case ETH_P_ARP: case ETH_P_IP: @@ -1095,14 +1100,16 @@ void tap_add_packet(struct ctx *c, ssize_t l2len, char *p, tap4_handler(c, pool_tap4, now); pool_flush(pool_tap4); } - packet_add(pool_tap4, l2len, p); + packet_add(pool_tap4, data->iov[0].iov_len - data->off, + (char *)data->iov[0].iov_base + data->off); break; case ETH_P_IPV6: if (pool_full(pool_tap6)) { tap6_handler(c, pool_tap6, now); pool_flush(pool_tap6); } - packet_add(pool_tap6, l2len, p); + packet_add(pool_tap6, data->iov[0].iov_len - data->off, + (char *)data->iov[0].iov_base + data->off); break; default: break; @@ -1168,6 +1175,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec *now)

while (n >= (ssize_t)sizeof(uint32_t)) { uint32_t l2len = ntohl_unaligned(p); + struct iov_tail data;

if (l2len < sizeof(struct ethhdr) || l2len > L2_MAX_LEN_PASST) { err("Bad frame size from guest, resetting connection"); @@ -1182,7 +1190,8 @@ static void tap_passt_input(struct ctx *c, const struct timespec *now) p += sizeof(uint32_t); n -= sizeof(uint32_t);

- tap_add_packet(c, l2len, p, now); + data = IOV_TAIL_FROM_BUF(p, l2len, 0); + tap_add_packet(c, &data, now);

p += l2len; n -= l2len; @@ -1226,6 +1235,8 @@ static void tap_pasta_input(struct ctx *c, const struct timespec *now) for (n = 0; n <= (ssize_t)(sizeof(pkt_buf) - L2_MAX_LEN_PASTA); n += len) { + struct iov_tail data; + len = read(c->fd_tap, pkt_buf + n, L2_MAX_LEN_PASTA);

if (len == 0) { @@ -1247,7 +1258,8 @@ static void tap_pasta_input(struct ctx *c, const struct timespec *now) len > (ssize_t)L2_MAX_LEN_PASTA) continue;

- tap_add_packet(c, len, pkt_buf + n, now); + data = IOV_TAIL_FROM_BUF(pkt_buf + n, len, 0); + tap_add_packet(c, &data, now); }

tap_handler(c, now); diff --git a/tap.h b/tap.h index 936ae9371fd6..ce5510882d5d 100644 --- a/tap.h +++ b/tap.h @@ -119,7 +119,6 @@ void tap_sock_update_pool(void *base, size_t size); void tap_backend_init(struct ctx *c); void tap_flush_pools(void); void tap_handler(struct ctx *c, const struct timespec *now); -void tap_add_packet(struct ctx *c, ssize_t l2len, char *p, +void tap_add_packet(struct ctx *c, struct iov_tail *data, const struct timespec *now); - #endif /* TAP_H */ diff --git a/vu_common.c b/vu_common.c index 5e6fd4a8261f..b77b21420c57 100644 --- a/vu_common.c +++ b/vu_common.c @@ -163,7 +163,6 @@ static void vu_handle_tx(struct vu_dev *vdev, int index, struct vu_virtq_element elem[VIRTQUEUE_MAX_SIZE]; struct iovec out_sg[VIRTQUEUE_MAX_SIZE]; struct vu_virtq *vq = &vdev->vq[index]; - int hdrlen = sizeof(struct virtio_net_hdr_mrg_rxbuf); int out_sg_count; int count;

@@ -176,6 +175,7 @@ static void vu_handle_tx(struct vu_dev *vdev, int index, while (count < VIRTQUEUE_MAX_SIZE && out_sg_count + VU_MAX_TX_BUFFER_NB <= VIRTQUEUE_MAX_SIZE) { int ret; + struct iov_tail data;

elem[count].out_num = VU_MAX_TX_BUFFER_NB; elem[count].out_sg = &out_sg[out_sg_count]; @@ -191,26 +191,10 @@ static void vu_handle_tx(struct vu_dev *vdev, int index, warn("virtio-net transmit queue contains no out buffers"); break; } - if (elem[count].out_num == 1) { - tap_add_packet(vdev->context, - elem[count].out_sg[0].iov_len - hdrlen, - (char *)elem[count].out_sg[0].iov_base + - hdrlen, now); - } else { - /* vnet header can be in a separate iovec */ - if (elem[count].out_num != 2) { - debug("virtio-net transmit queue contains more than one buffer ([%d]: %u)", - count, elem[count].out_num); - } else if (elem[count].out_sg[0].iov_len != (size_t)hdrlen) { - debug("virtio-net transmit queue entry not aligned on hdrlen ([%d]: %d != %zu)", - count, hdrlen, elem[count].out_sg[0].iov_len); - } else { - tap_add_packet(vdev->context, - elem[count].out_sg[1].iov_len, - (char *)elem[count].out_sg[1].iov_base, - now); - } - } + + data = IOV_TAIL(elem[count].out_sg, elem[count].out_num, 0); + if (IOV_DROP_HEADER(&data, struct virtio_net_hdr_mrg_rxbuf)) + tap_add_packet(vdev->context, &data, now);

Nice simplification here.

count++; }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On 06/08/2025 04:17, David Gibson wrote:

On Tue, Aug 05, 2025 at 05:46:05PM +0200, Laurent Vivier wrote:

...

Use packet_data() and extract headers using IOV_REMOVE_HEADER() rather than packet_get().

Signed-off-by: Laurent Vivier Reviewed-by: David Gibson

Still R-b, but making an observation below that's perhaps more relevant to the previous patch.

...

--- arp.c | 12 +++++++++--- packet.c | 1 - 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/arp.c b/arp.c index 9f1fedeafec0..b3ac42082841 100644 --- a/arp.c +++ b/arp.c @@ -74,14 +74,20 @@ int arp(const struct ctx *c, const struct pool *p) struct arphdr ah; struct arpmsg am; } __attribute__((__packed__)) resp; + struct arphdr ah_storage; + struct ethhdr eh_storage; + struct arpmsg am_storage; const struct ethhdr *eh; const struct arphdr *ah; const struct arpmsg *am; + struct iov_tail data;

- eh = packet_get(p, 0, 0, sizeof(*eh), NULL); - ah = packet_get(p, 0, sizeof(*eh), sizeof(*ah), NULL); - am = packet_get(p, 0, sizeof(*eh) + sizeof(*ah), sizeof(*am), NULL); + if (!packet_data(p, 0, &data)) + return -1;

The only case where packet_data() will return false is if you give it a bad packet index. That should never happen, by construction. So I'm wondering if that should be an ASSSERT() in packet_data() rather than a return value.

Stefano, why do you think of this idea? Thanks, Laurent

On Thu, Aug 07, 2025 at 03:11:32PM +0200, Stefano Brivio wrote:

On Thu, 7 Aug 2025 14:58:34 +0200 Laurent Vivier wrote:

...

On 06/08/2025 04:17, David Gibson wrote:

...

On Tue, Aug 05, 2025 at 05:46:05PM +0200, Laurent Vivier wrote:

...

Use packet_data() and extract headers using IOV_REMOVE_HEADER() rather than packet_get().

Signed-off-by: Laurent Vivier Reviewed-by: David Gibson

Still R-b, but making an observation below that's perhaps more relevant to the previous patch.

...

--- arp.c | 12 +++++++++--- packet.c | 1 - 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/arp.c b/arp.c index 9f1fedeafec0..b3ac42082841 100644 --- a/arp.c +++ b/arp.c @@ -74,14 +74,20 @@ int arp(const struct ctx *c, const struct pool *p) struct arphdr ah; struct arpmsg am; } __attribute__((__packed__)) resp; + struct arphdr ah_storage; + struct ethhdr eh_storage; + struct arpmsg am_storage; const struct ethhdr *eh; const struct arphdr *ah; const struct arpmsg *am; + struct iov_tail data;

- eh = packet_get(p, 0, 0, sizeof(*eh), NULL); - ah = packet_get(p, 0, sizeof(*eh), sizeof(*ah), NULL); - am = packet_get(p, 0, sizeof(*eh) + sizeof(*ah), sizeof(*am), NULL); + if (!packet_data(p, 0, &data)) + return -1;

The only case where packet_data() will return false is if you give it a bad packet index. That should never happen, by construction. So I'm wondering if that should be an ASSSERT() in packet_data() rather than a return value.

Stefano, why do you think of this idea?

Well, yes, it *should* be by construction, but somewhere we might eventually calculate that index (indirectly) using data we receive, and I don't think we want to ASSERT() if somebody finds a way to make us calculate a bad index.

I really can't imagine a scenario where we'd want to do that. The order of things in the packet pool is entirely arbitrary, so anything from outside can't have any data that would be relevant to finding an index. So, in all cases we're either passing a (valid) index from one part of our code to another, or scanning the entire pool. Any failure in either case would represent a pretty unlikely bug on our side, which makes ASSERT() the appropriate choice IMO.

It's not a strong objection against ASSERT(), though. It makes the code marginally more terse and might help us find issues, too. I just have a slight preference for a return value in this case anyway (better to dodge a security issue and hide a functional issue than risking hitting both, I think).

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:14PM +0200, Laurent Vivier wrote:

dhcpv6_opt() and its callers are refactored for iov_tail option parsing, replacing direct offset management for improved robustness.

Its signature is now `bool dhcpv6_opt(iov_tail *data, type)`. `*data` (in/out) points to a found option on `true` return or is restored on `false`. The main dhcpv6() function uses IOV_REMOVE_HEADER for the msg_hdr, then passes the iov_tail (now at options start) to the new dhcpv6_opt().

Signed-off-by: Laurent Vivier

Hmm. I'm not sure this is a great use case for iov_tail - the code is repeatedly scanning the same options, so there's a whole bunch of rewinding. It works, but it seems awkward. DHCP is a slow path, anyway, so maybe we'd be better off just linearizing the entire packet and using plain old pointers to scan through the options.

--- dhcpv6.c | 179 ++++++++++++++++++++++++++++++++----------------------- iov.c | 1 - 2 files changed, 104 insertions(+), 76 deletions(-)

diff --git a/dhcpv6.c b/dhcpv6.c index ae06e646f92f..e93acaf9955e 100644 --- a/dhcpv6.c +++ b/dhcpv6.c @@ -280,112 +280,125 @@ static struct resp_not_on_link_t {

/** * dhcpv6_opt() - Get option from DHCPv6 message - * @p: Packet pool, single packet with UDP header - * @offset: Offset to look at, 0: end of header, set to option start + * @data: Buffer with options, set to matching option on return * @type: Option type to look up, network order * - * Return: pointer to option header, or NULL on malformed or missing option + * Return: true if found and @data points to the option header, + * or false on malformed or missing option and @data is + * unmodified. */ -static struct opt_hdr *dhcpv6_opt(const struct pool *p, size_t *offset, - uint16_t type) +static bool dhcpv6_opt(struct iov_tail *data, uint16_t type) { - struct opt_hdr *o; - size_t left; + struct iov_tail head = *data; + struct opt_hdr o_storage; + const struct opt_hdr *o;

- ASSERT(*offset >= UDP_MSG_HDR_SIZE); - - while ((o = packet_get_try(p, 0, *offset, sizeof(*o), &left))) { + while ((o = IOV_PEEK_HEADER(data, o_storage))) { unsigned int opt_len = ntohs(o->l) + sizeof(*o);

- if (ntohs(o->l) > left) - return NULL; + if (opt_len > iov_tail_size(data)) + break;

if (o->t == type) - return o; + return true;

- *offset += opt_len; + iov_tail_drop(data, opt_len); }

- return NULL; + *data = head; + return false; }

/** * dhcpv6_ia_notonlink() - Check if any IA contains non-appropriate addresses - * @p: Packet pool, single packet starting from UDP header + * @data: Data to look at, packet starting from UDP header (input/output) * @la: Address we want to lease to the client * - * Return: pointer to non-appropriate IA_NA or IA_TA, if any, NULL otherwise + * Return: true and @data points to non-appropriate IA_NA or IA_TA, if any, + * false otherwise and @data is unmodified */ -static struct opt_hdr *dhcpv6_ia_notonlink(const struct pool *p, - struct in6_addr *la) +static bool dhcpv6_ia_notonlink(struct iov_tail *data, + struct in6_addr *la) { int ia_types[2] = { OPT_IA_NA, OPT_IA_TA }, *ia_type; + struct opt_ia_addr opt_addr_storage; const struct opt_ia_addr *opt_addr; + struct iov_tail current, ia_base; + struct opt_ia_na ia_storage; char buf[INET6_ADDRSTRLEN]; + const struct opt_ia_na *ia; struct in6_addr req_addr; + struct opt_hdr h_storage; const struct opt_hdr *h; - struct opt_hdr *ia; - size_t offset;

foreach(ia_type, ia_types) { - offset = UDP_MSG_HDR_SIZE; - while ((ia = dhcpv6_opt(p, &offset, *ia_type))) { - if (ntohs(ia->l) < OPT_VSIZE(ia_na)) - return NULL; - - offset += sizeof(struct opt_ia_na); + current = *data; + while (dhcpv6_opt(¤t, *ia_type)) { + ia_base = current; + ia = IOV_REMOVE_HEADER(¤t, ia_storage); + if (!ia || ntohs(ia->hdr.l) < OPT_VSIZE(ia_na)) + goto notfound; + + while (dhcpv6_opt(¤t, OPT_IAAADR)) { + h = IOV_PEEK_HEADER(¤t, h_storage); + if (!h || ntohs(h->l) != OPT_VSIZE(ia_addr)) + goto notfound; + + opt_addr = IOV_REMOVE_HEADER(¤t, + opt_addr_storage); + if (!opt_addr) + goto notfound;

- while ((h = dhcpv6_opt(p, &offset, OPT_IAAADR))) { - if (ntohs(h->l) != OPT_VSIZE(ia_addr)) - return NULL; - - opt_addr = (const struct opt_ia_addr *)h; req_addr = opt_addr->addr; if (!IN6_ARE_ADDR_EQUAL(la, &req_addr)) - goto err; - - offset += sizeof(struct opt_ia_addr); + goto notonlink; } } }

- return NULL; +notfound: + return false;

-err: +notonlink: info("DHCPv6: requested address %s not on link", inet_ntop(AF_INET6, &req_addr, buf, sizeof(buf))); - return ia; + *data = ia_base; + return true; }

/** * dhcpv6_send_ia_notonlink() - Send NotOnLink status - * @c: Execution context - * @ia: Pointer to non-appropriate IA_NA or IA_TA - * @client_id: Client ID message option - * xid: Transaction ID for message exchange + * @c: Execution context + * @ia_base: Non-appropriate IA_NA or IA_TA base + * @client_id_base: Client ID message option base + * @len: Client ID length + * @xid: Transaction ID for message exchange */ -static void dhcpv6_send_ia_notonlink(struct ctx *c, struct opt_hdr *ia, - const struct opt_hdr *client_id, - uint32_t xid) +static void dhcpv6_send_ia_notonlink(struct ctx *c, + const struct iov_tail *ia_base, + const struct iov_tail *client_id_base, + int len, uint32_t xid) { const struct in6_addr *src = &c->ip6.our_tap_ll; + struct opt_hdr *ia = (struct opt_hdr *)resp_not_on_link.var; size_t n;

info("DHCPv6: received CONFIRM with inappropriate IA," " sending NotOnLink status in REPLY");

- ia->l = htons(OPT_VSIZE(ia_na) + sizeof(sc_not_on_link)); - n = sizeof(struct opt_ia_na); - memcpy(resp_not_on_link.var, ia, n); + iov_to_buf(&ia_base->iov[0], ia_base->cnt, ia_base->off, + resp_not_on_link.var, n); + ia->l = htons(OPT_VSIZE(ia_na) + sizeof(sc_not_on_link)); memcpy(resp_not_on_link.var + n, &sc_not_on_link, sizeof(sc_not_on_link));

n += sizeof(sc_not_on_link); - memcpy(resp_not_on_link.var + n, client_id, - sizeof(struct opt_hdr) + ntohs(client_id->l)); + iov_to_buf(&client_id_base->iov[0], client_id_base->cnt, + client_id_base->off, resp_not_on_link.var + n, + sizeof(struct opt_hdr) + len);

- n += sizeof(struct opt_hdr) + ntohs(client_id->l); + n += sizeof(struct opt_hdr) + len;

n = offsetof(struct resp_not_on_link_t, var) + n;

@@ -474,17 +487,19 @@ search:

/** * dhcpv6_client_fqdn_fill() - Fill in client FQDN option + * @data: Data to look at * @c: Execution context * @buf: Response message buffer where options will be appended * @offset: Offset in message buffer for new options * * Return: updated length of response message buffer. */ -static size_t dhcpv6_client_fqdn_fill(const struct pool *p, const struct ctx *c, +static size_t dhcpv6_client_fqdn_fill(const struct iov_tail *data, + const struct ctx *c, char *buf, int offset)

{ - struct opt_client_fqdn const *req_opt; + struct iov_tail current = *data; struct opt_client_fqdn *o; size_t opt_len;

@@ -502,14 +517,16 @@ static size_t dhcpv6_client_fqdn_fill(const struct pool *p, const struct ctx *c, }

o = (struct opt_client_fqdn *)(buf + offset); + o->flags = 0x00; encode_domain_name(o->domain_name, c->fqdn); - req_opt = (struct opt_client_fqdn *)dhcpv6_opt(p, - &(size_t){ UDP_MSG_HDR_SIZE }, - OPT_CLIENT_FQDN); - if (req_opt && req_opt->flags & 0x01 /* S flag */) - o->flags = 0x02 /* O flag */; - else - o->flags = 0x00; + if (dhcpv6_opt(¤t, OPT_CLIENT_FQDN)) { + struct opt_client_fqdn req_opt_storage; + struct opt_client_fqdn const *req_opt; + + req_opt = IOV_PEEK_HEADER(¤t, req_opt_storage); + if (req_opt && req_opt->flags & 0x01 /* S flag */) + o->flags = 0x02 /* O flag */; + }

opt_len++;

@@ -531,14 +548,18 @@ static size_t dhcpv6_client_fqdn_fill(const struct pool *p, const struct ctx *c, int dhcpv6(struct ctx *c, const struct pool *p, const struct in6_addr *saddr, const struct in6_addr *daddr) { - const struct opt_hdr *client_id, *server_id, *ia; + const struct opt_server_id *server_id = NULL; + struct iov_tail data, opt, client_id_base; + const struct opt_hdr *client_id = NULL; + struct opt_server_id server_id_storage; + const struct opt_ia_na *ia = NULL; + struct opt_hdr client_id_storage; + struct opt_ia_na ia_storage; const struct in6_addr *src; struct msg_hdr mh_storage; const struct msg_hdr *mh; struct udphdr uh_storage; const struct udphdr *uh; - struct opt_hdr *bad_ia; - struct iov_tail data; size_t mlen, n;

if (!packet_data(p, 0, &data)) @@ -565,20 +586,26 @@ int dhcpv6(struct ctx *c, const struct pool *p,

src = &c->ip6.our_tap_ll;

- mh = IOV_PEEK_HEADER(&data, mh_storage); + mh = IOV_REMOVE_HEADER(&data, mh_storage); if (!mh) return -1;

- client_id = dhcpv6_opt(p, &(size_t){ UDP_MSG_HDR_SIZE }, OPT_CLIENTID); + client_id_base = data; + if (dhcpv6_opt(&client_id_base, OPT_CLIENTID)) + client_id = IOV_PEEK_HEADER(&client_id_base, client_id_storage); if (!client_id || ntohs(client_id->l) > OPT_VSIZE(client_id)) return -1;

- server_id = dhcpv6_opt(p, &(size_t){ UDP_MSG_HDR_SIZE }, OPT_SERVERID); - if (server_id && ntohs(server_id->l) != OPT_VSIZE(server_id)) + opt = data; + if (dhcpv6_opt(&opt, OPT_SERVERID)) + server_id = IOV_PEEK_HEADER(&opt, server_id_storage); + if (server_id && ntohs(server_id->hdr.l) != OPT_VSIZE(server_id)) return -1;

- ia = dhcpv6_opt(p, &(size_t){ UDP_MSG_HDR_SIZE }, OPT_IA_NA); - if (ia && ntohs(ia->l) < MIN(OPT_VSIZE(ia_na), OPT_VSIZE(ia_ta))) + opt = data; + if (dhcpv6_opt(&opt, OPT_IA_NA)) + ia = IOV_PEEK_HEADER(&opt, ia_storage); + if (ia && ntohs(ia->hdr.l) < MIN(OPT_VSIZE(ia_na), OPT_VSIZE(ia_ta))) return -1;

resp.hdr.type = TYPE_REPLY; @@ -593,9 +620,10 @@ int dhcpv6(struct ctx *c, const struct pool *p, if (mh->type == TYPE_CONFIRM && server_id) return -1;

- if ((bad_ia = dhcpv6_ia_notonlink(p, &c->ip6.addr))) { + if (dhcpv6_ia_notonlink(&data, &c->ip6.addr)) {

- dhcpv6_send_ia_notonlink(c, bad_ia, client_id, mh->xid); + dhcpv6_send_ia_notonlink(c, &data, &client_id_base, + ntohs(client_id->l), mh->xid);

return 1; } @@ -607,7 +635,7 @@ int dhcpv6(struct ctx *c, const struct pool *p, memcmp(&resp.server_id, server_id, sizeof(resp.server_id))) return -1;

- if (ia || dhcpv6_opt(p, &(size_t){ UDP_MSG_HDR_SIZE }, OPT_IA_TA)) + if (ia || dhcpv6_opt(&data, OPT_IA_TA)) return -1;

info("DHCPv6: received INFORMATION_REQUEST, sending REPLY"); @@ -633,13 +661,14 @@ int dhcpv6(struct ctx *c, const struct pool *p, if (ia) resp.ia_na.iaid = ((struct opt_ia_na *)ia)->iaid;

- memcpy(&resp.client_id, client_id, - ntohs(client_id->l) + sizeof(struct opt_hdr)); + iov_to_buf(&client_id_base.iov[0], client_id_base.cnt, + client_id_base.off, &resp.client_id, + ntohs(client_id->l) + sizeof(struct opt_hdr));

n = offsetof(struct resp_t, client_id) + sizeof(struct opt_hdr) + ntohs(client_id->l); n = dhcpv6_dns_fill(c, (char *)&resp, n); - n = dhcpv6_client_fqdn_fill(p, c, (char *)&resp, n); + n = dhcpv6_client_fqdn_fill(&data, c, (char *)&resp, n);

resp.hdr.xid = mh->xid;

diff --git a/iov.c b/iov.c index f519eb3cfeaf..d17d4dd3da09 100644 --- a/iov.c +++ b/iov.c @@ -109,7 +109,6 @@ size_t iov_from_buf(const struct iovec *iov, size_t iov_cnt, * * Return: the number of bytes successfully copied. */ -/* cppcheck-suppress [staticFunction] */ size_t iov_to_buf(const struct iovec *iov, size_t iov_cnt, size_t offset, void *buf, size_t bytes) {

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:15PM +0200, Laurent Vivier wrote:

Use packet_data() and extract headers using IOV_REMOVE_HEADER() and IOV_PEEK_HEADER() rather than packet_get().

Unlike the previous patch, I think using iov_tail does work here, because there's a single scan through the options, rather than repeatedly scanning for options of specific types.

Signed-off-by: Laurent Vivier --- dhcp.c | 46 ++++++++++++++++++++++++++++------------------ 1 file changed, 28 insertions(+), 18 deletions(-)

diff --git a/dhcp.c b/dhcp.c index b0de04be6f27..cf73d4b07767 100644 --- a/dhcp.c +++ b/dhcp.c @@ -302,27 +302,33 @@ static void opt_set_dns_search(const struct ctx *c, size_t max_len) */ int dhcp(const struct ctx *c, const struct pool *p) { - size_t mlen, dlen, offset = 0, opt_len, opt_off = 0; char macstr[ETH_ADDRSTRLEN]; + size_t mlen, dlen, opt_len; struct in_addr mask, dst; + struct ethhdr eh_storage; + struct iphdr iph_storage; + struct udphdr uh_storage; const struct ethhdr *eh; const struct iphdr *iph; const struct udphdr *uh; + struct iov_tail data; struct msg const *m;

Pre-existing, but I'm a bit baffled as to what the (const *) is doing here.

struct msg reply; unsigned int i; + struct msg m_storage;

- eh = packet_get(p, 0, offset, sizeof(*eh), NULL); - offset += sizeof(*eh); + if (!packet_data(p, 0, &data)) + return -1;

- iph = packet_get(p, 0, offset, sizeof(*iph), NULL); + eh = IOV_REMOVE_HEADER(&data, eh_storage); + iph = IOV_PEEK_HEADER(&data, iph_storage); if (!eh || !iph) return -1;

- offset += iph->ihl * 4UL; - uh = packet_get(p, 0, offset, sizeof(*uh), &mlen); - offset += sizeof(*uh); + if (!iov_tail_drop(&data, iph->ihl * 4UL)) + return -1;

+ uh = IOV_REMOVE_HEADER(&data, uh_storage); if (!uh) return -1;

@@ -332,7 +338,10 @@ int dhcp(const struct ctx *c, const struct pool *p) if (c->no_dhcp) return 1;

- m = packet_get(p, 0, offset, offsetof(struct msg, o), &opt_len); + mlen = iov_tail_size(&data); + m = (struct msg const *)iov_remove_header_(&data, &m_storage, + offsetof(struct msg, o), + __alignof__(struct msg)); if (!m || mlen != ntohs(uh->len) - sizeof(*uh) || mlen < offsetof(struct msg, o) || @@ -355,27 +364,28 @@ int dhcp(const struct ctx *c, const struct pool *p) memset(&reply.file, 0, sizeof(reply.file)); reply.magic = m->magic;

- offset += offsetof(struct msg, o); - for (i = 0; i < ARRAY_SIZE(opts); i++) opts[i].clen = -1;

- while (opt_off + 2 < opt_len) { - const uint8_t *olen, *val; + opt_len = iov_tail_size(&data); + while (opt_len >= 2) { + uint8_t olen_storage, type_storage; + const uint8_t *olen; uint8_t *type;

- type = packet_get(p, 0, offset + opt_off, 1, NULL); - olen = packet_get(p, 0, offset + opt_off + 1, 1, NULL); + type = IOV_REMOVE_HEADER(&data, type_storage); + olen = IOV_REMOVE_HEADER(&data, olen_storage);

It seems a bit mad to access single bytes via 8-byte pointers, but it's probably not worth the hassle of handling it differently in this one case.

if (!type || !olen) return -1;

- val = packet_get(p, 0, offset + opt_off + 2, *olen, NULL); - if (!val) + opt_len = iov_tail_size(&data); + if (opt_len < *olen) return -1;

- memcpy(&opts[*type].c, val, *olen); + iov_to_buf(&data.iov[0], data.cnt, data.off, &opts[*type].c, *olen);

So, IIUC, if *olen is much too big, this is still safe..

opts[*type].clen = *olen;

.. but recording *olen unedited as the length of the option is probably wrong in that case.

- opt_off += *olen + 2; + iov_tail_drop(&data, *olen); + opt_len -= *olen;

Isn't the stanza above doing the equivalent of an iov_remove_header_()?

}

opts[80].slen = -1;

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Fri, Aug 08, 2025 at 11:33:24AM +0200, Laurent Vivier wrote:

On 06/08/2025 06:38, David Gibson wrote:

...

On Tue, Aug 05, 2025 at 05:46:15PM +0200, Laurent Vivier wrote:

...

Use packet_data() and extract headers using IOV_REMOVE_HEADER() and IOV_PEEK_HEADER() rather than packet_get().

Unlike the previous patch, I think using iov_tail does work here, because there's a single scan through the options, rather than repeatedly scanning for options of specific types.

...

Signed-off-by: Laurent Vivier --- dhcp.c | 46 ++++++++++++++++++++++++++++------------------ 1 file changed, 28 insertions(+), 18 deletions(-)

diff --git a/dhcp.c b/dhcp.c index b0de04be6f27..cf73d4b07767 100644 --- a/dhcp.c +++ b/dhcp.c @@ -302,27 +302,33 @@ static void opt_set_dns_search(const struct ctx *c, size_t max_len) */ int dhcp(const struct ctx *c, const struct pool *p) { - size_t mlen, dlen, offset = 0, opt_len, opt_off = 0; char macstr[ETH_ADDRSTRLEN]; + size_t mlen, dlen, opt_len; struct in_addr mask, dst; + struct ethhdr eh_storage; + struct iphdr iph_storage; + struct udphdr uh_storage; const struct ethhdr *eh; const struct iphdr *iph; const struct udphdr *uh; + struct iov_tail data; struct msg const *m;

Pre-existing, but I'm a bit baffled as to what the (const *) is doing here.

...

struct msg reply; unsigned int i; + struct msg m_storage; - eh = packet_get(p, 0, offset, sizeof(*eh), NULL); - offset += sizeof(*eh); + if (!packet_data(p, 0, &data)) + return -1; - iph = packet_get(p, 0, offset, sizeof(*iph), NULL); + eh = IOV_REMOVE_HEADER(&data, eh_storage); + iph = IOV_PEEK_HEADER(&data, iph_storage); if (!eh || !iph) return -1; - offset += iph->ihl * 4UL; - uh = packet_get(p, 0, offset, sizeof(*uh), &mlen); - offset += sizeof(*uh); + if (!iov_tail_drop(&data, iph->ihl * 4UL)) + return -1; + uh = IOV_REMOVE_HEADER(&data, uh_storage); if (!uh) return -1; @@ -332,7 +338,10 @@ int dhcp(const struct ctx *c, const struct pool *p) if (c->no_dhcp) return 1; - m = packet_get(p, 0, offset, offsetof(struct msg, o), &opt_len); + mlen = iov_tail_size(&data); + m = (struct msg const *)iov_remove_header_(&data, &m_storage, + offsetof(struct msg, o), + __alignof__(struct msg)); if (!m || mlen != ntohs(uh->len) - sizeof(*uh) || mlen < offsetof(struct msg, o) || @@ -355,27 +364,28 @@ int dhcp(const struct ctx *c, const struct pool *p) memset(&reply.file, 0, sizeof(reply.file)); reply.magic = m->magic; - offset += offsetof(struct msg, o); - for (i = 0; i < ARRAY_SIZE(opts); i++) opts[i].clen = -1; - while (opt_off + 2 < opt_len) { - const uint8_t *olen, *val; + opt_len = iov_tail_size(&data); + while (opt_len >= 2) { + uint8_t olen_storage, type_storage; + const uint8_t *olen; uint8_t *type; - type = packet_get(p, 0, offset + opt_off, 1, NULL); - olen = packet_get(p, 0, offset + opt_off + 1, 1, NULL); + type = IOV_REMOVE_HEADER(&data, type_storage); + olen = IOV_REMOVE_HEADER(&data, olen_storage);

It seems a bit mad to access single bytes via 8-byte pointers, but it's probably not worth the hassle of handling it differently in this one case.

...

if (!type || !olen) return -1; - val = packet_get(p, 0, offset + opt_off + 2, *olen, NULL); - if (!val) + opt_len = iov_tail_size(&data); + if (opt_len < *olen) return -1; - memcpy(&opts[*type].c, val, *olen); + iov_to_buf(&data.iov[0], data.cnt, data.off, &opts[*type].c, *olen);

So, IIUC, if *olen is much too big, this is still safe..

...

opts[*type].clen = *olen;

.. but recording *olen unedited as the length of the option is probably wrong in that case.

I don't understand how to edit *olen. There is no change regarding the original code.

Sorry, I was concerned about a malformed packet which gave a long olen when there isn't actually that much that. I missed that you tested for (opt_len < *olen) above, which handles that case.

...

...

- opt_off += *olen + 2; + iov_tail_drop(&data, *olen); + opt_len -= *olen;

Isn't the stanza above doing the equivalent of an iov_remove_header_()?

No, in fact iov_remove_header_() copy to the buffer only if the data are discontinuous, in this case we want to copy the data unconditionally to edit them later. We don't want to edit the data in the iovec buffer.

Ah, right. Unconditionally linearizing / copying a header seems like it might be a useful thing in more places. I wonder if we should make that its own helper and we can use it both here and in the slow path of iov_remove_header_(). -- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:16PM +0200, Laurent Vivier wrote:

Use packet_data() and extract headers using IOV_REMOVE_HEADER() and IOV_PEEK_HEADER() rather than packet_get().

Signed-off-by: Laurent Vivier --- ip.c | 32 +++++++++++++++----------------- ip.h | 3 +-- packet.c | 1 + tap.c | 4 +++- 4 files changed, 20 insertions(+), 20 deletions(-)

diff --git a/ip.c b/ip.c index 2cc7f6548aff..50bd69a70596 100644 --- a/ip.c +++ b/ip.c @@ -23,50 +23,48 @@

/** * ipv6_l4hdr() - Find pointer to L4 header in IPv6 packet and extract protocol - * @p: Packet pool, packet number @idx has IPv6 header at @offset - * @idx: Index of packet in pool - * @offset: Pre-calculated IPv6 header offset + * @data: IPv6 packet * @proto: Filled with L4 protocol number * @dlen: Data length (payload excluding header extensions), set on return * - * Return: pointer to L4 header, NULL if not found + * Return: true if the L4 header is found and @data, @proto, @dlen are set, + * false on error. Outputs are indeterminate on failure. */ -char *ipv6_l4hdr(const struct pool *p, int idx, size_t offset, uint8_t *proto, - size_t *dlen) +bool ipv6_l4hdr(struct iov_tail *data, uint8_t *proto, size_t *dlen) { + struct ipv6_opt_hdr o_storage; const struct ipv6_opt_hdr *o; + struct ipv6hdr ip6h_storage; const struct ipv6hdr *ip6h; - char *base; int hdrlen; uint8_t nh;

- base = packet_get(p, idx, 0, 0, NULL); - ip6h = packet_get(p, idx, offset, sizeof(*ip6h), dlen); + ip6h = IOV_REMOVE_HEADER(data, ip6h_storage); if (!ip6h) - return NULL; - - offset += sizeof(*ip6h); + return false; + *dlen = iov_tail_size(data);

nh = ip6h->nexthdr; if (!IPV6_NH_OPT(nh)) goto found;

- while ((o = packet_get_try(p, idx, offset, sizeof(*o), dlen))) { + while ((o = IOV_PEEK_HEADER(data, o_storage))) { + *dlen = iov_tail_size(data) - sizeof(*o);

I don't think this is quite right. This removes the option header from the total, but not the option body (if any). It will be corrected on the next loop, via iova_tail_size() - at the cost of rescanning the IOV. However, I think that means the body of the last option will be incorrectly included in dlen. You could update *dlen incrementally, but AFAICT you don't need it internally, so you could just compute from iov_tail_size() after the found label.

nh = o->nexthdr; hdrlen = (o->hdrlen + 1) * 8;

if (IPV6_NH_OPT(nh)) - offset += hdrlen; + iov_tail_drop(data, hdrlen); else goto found; }

- return NULL; + return false;

found: if (nh == 59)

Pre-existing: it'd be nice to have a name for that bare 59.

- return NULL; + return false;

*proto = nh; - return base + offset; + return true; } diff --git a/ip.h b/ip.h index 24509d9c11cd..5830b92302e2 100644 --- a/ip.h +++ b/ip.h @@ -115,8 +115,7 @@ static inline uint32_t ip6_get_flow_lbl(const struct ipv6hdr *ip6h) ip6h->flow_lbl[2]; }

-char *ipv6_l4hdr(const struct pool *p, int idx, size_t offset, uint8_t *proto, - size_t *dlen); +bool ipv6_l4hdr(struct iov_tail *data, uint8_t *proto, size_t *dlen);

/* IPv6 link-local all-nodes multicast address, ff02::1 */ static const struct in6_addr in6addr_ll_all_nodes = { diff --git a/packet.c b/packet.c index 34b1722b9a03..014b353cdf8b 100644 --- a/packet.c +++ b/packet.c @@ -133,6 +133,7 @@ void packet_add_do(struct pool *p, struct iov_tail *data, * * Return: pointer to start of data range, NULL on invalid range or descriptor */ +/* cppcheck-suppress [staticFunction] */ void *packet_get_try_do(const struct pool *p, size_t idx, size_t offset, size_t len, size_t *left, const char *func, int line) { diff --git a/tap.c b/tap.c index 8d2b118152f1..d7852fad6069 100644 --- a/tap.c +++ b/tap.c @@ -911,8 +911,10 @@ resume: if (plen != check) continue;

- if (!(l4h = ipv6_l4hdr(in, i, sizeof(*eh), &proto, &l4len))) + data = IOV_TAIL_FROM_BUF(ip6h, sizeof(*ip6h) + check, 0); + if (!ipv6_l4hdr(&data, &proto, &l4len)) continue; + l4h = (char *)data.iov[0].iov_base + data.off;

if (IN6_IS_ADDR_LOOPBACK(saddr) || IN6_IS_ADDR_LOOPBACK(daddr)) { char sstr[INET6_ADDRSTRLEN], dstr[INET6_ADDRSTRLEN];

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:17PM +0200, Laurent Vivier wrote:

Use packet_data() and extract headers using IOV_PEEK_HEADER() rather than packet_get().

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- tap.c | 33 ++++++++++++++++++++------------- 1 file changed, 20 insertions(+), 13 deletions(-)

diff --git a/tap.c b/tap.c index d7852fad6069..4fbcad3b385f 100644 --- a/tap.c +++ b/tap.c @@ -706,28 +706,34 @@ static int tap4_handler(struct ctx *c, const struct pool *in, i = 0; resume: for (seq_count = 0, seq = NULL; i < in->count; i++) { - size_t l2len, l3len, hlen, l4len; + size_t l3len, hlen, l4len; + struct ethhdr eh_storage; + struct iphdr iph_storage; + struct udphdr uh_storage; const struct ethhdr *eh; const struct udphdr *uh; struct iov_tail data; struct iphdr *iph; - const char *l4h;

- packet_get(in, i, 0, 0, &l2len); + if (!packet_data(in, i, &data)) + continue;

- eh = packet_get(in, i, 0, sizeof(*eh), &l3len); + eh = IOV_PEEK_HEADER(&data, eh_storage); if (!eh) continue; if (ntohs(eh->h_proto) == ETH_P_ARP) { PACKET_POOL_P(pkt, 1, in->buf, in->buf_size);

- data = IOV_TAIL_FROM_BUF((void *)eh, l2len, 0); packet_add(pkt, &data); arp(c, pkt); continue; }

- iph = packet_get(in, i, sizeof(*eh), sizeof(*iph), NULL); + if (!iov_tail_drop(&data, sizeof(*eh))) + continue; + l3len = iov_tail_size(&data); + + iph = IOV_PEEK_HEADER(&data, iph_storage); if (!iph) continue;

@@ -755,8 +761,9 @@ resume: if (iph->saddr && c->ip4.addr_seen.s_addr != iph->saddr) c->ip4.addr_seen.s_addr = iph->saddr;

- l4h = packet_get(in, i, sizeof(*eh) + hlen, l4len, NULL); - if (!l4h) + if (!iov_tail_drop(&data, hlen)) + continue; + if (iov_tail_size(&data) != l4len) continue;

if (iph->protocol == IPPROTO_ICMP) { @@ -767,7 +774,6 @@ resume:

tap_packet_debug(iph, NULL, NULL, 0, NULL, 1);

- data = IOV_TAIL_FROM_BUF((void *)l4h, l4len, 0); packet_add(pkt, &data); icmp_tap_handler(c, PIF_TAP, AF_INET, &iph->saddr, &iph->daddr, @@ -775,15 +781,17 @@ resume: continue; }

- uh = packet_get(in, i, sizeof(*eh) + hlen, sizeof(*uh), NULL); + uh = IOV_PEEK_HEADER(&data, uh_storage); if (!uh) continue;

if (iph->protocol == IPPROTO_UDP) { + struct iov_tail eh_data; + PACKET_POOL_P(pkt, 1, in->buf, in->buf_size);

- data = IOV_TAIL_FROM_BUF((void *)eh, l2len, 0); - packet_add(pkt, &data); + packet_data(in, i, &eh_data); + packet_add(pkt, &eh_data); if (dhcp(c, pkt)) continue; } @@ -834,7 +842,6 @@ resume: #undef L4_SET

append: - data = IOV_TAIL_FROM_BUF((void *)l4h, l4len, 0); packet_add((struct pool *)&seq->p, &data); }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:18PM +0200, Laurent Vivier wrote:

Use packet_data() and extract headers using IOV_REMOVE_HEADER() and IOV_PEEK_HEADER() rather than packet_get().

Remove packet_get() as it is not used anymore.

[snip]

@@ -896,21 +896,28 @@ resume: for (seq_count = 0, seq = NULL; i < in->count; i++) { size_t l4len, plen, check; struct in6_addr *saddr, *daddr; + struct ipv6hdr ip6h_storage; + struct ethhdr eh_storage; + struct udphdr uh_storage; const struct ethhdr *eh; const struct udphdr *uh; struct iov_tail data; struct ipv6hdr *ip6h; uint8_t proto; - char *l4h;

- eh = packet_get(in, i, 0, sizeof(*eh), NULL); + if (!packet_data(in, i, &data)) + return -1; + + eh = IOV_REMOVE_HEADER(&data, eh_storage); if (!eh) continue;

- ip6h = packet_get(in, i, sizeof(*eh), sizeof(*ip6h), &check); + ip6h = IOV_PEEK_HEADER(&data, ip6h_storage); if (!ip6h) continue;

You peek the IPv6 header here, but I haven't spotted where you remove / drop it before...

+ check = iov_tail_size(&data) - sizeof(*ip6h); + saddr = &ip6h->saddr; daddr = &ip6h->daddr;

@@ -918,10 +925,8 @@ resume: if (plen != check) continue;

- data = IOV_TAIL_FROM_BUF(ip6h, sizeof(*ip6h) + check, 0); if (!ipv6_l4hdr(&data, &proto, &l4len)) continue; - l4h = (char *)data.iov[0].iov_base + data.off;

if (IN6_IS_ADDR_LOOPBACK(saddr) || IN6_IS_ADDR_LOOPBACK(daddr)) { char sstr[INET6_ADDRSTRLEN], dstr[INET6_ADDRSTRLEN]; @@ -946,6 +951,8 @@ resume: }

if (proto == IPPROTO_ICMPV6) { + struct icmp6hdr l4h_storage; + const struct icmp6hdr *l4h; PACKET_POOL_P(pkt, 1, in->buf, in->buf_size);

if (c->no_icmp) @@ -954,9 +961,9 @@ resume: if (l4len < sizeof(struct icmp6hdr)) continue;

- data = IOV_TAIL_FROM_BUF(l4h, l4len, 0); packet_add(pkt, &data);

+ l4h = IOV_PEEK_HEADER(&data, l4h_storage);

... peeking the next header here.

if (ndp(c, (struct icmp6hdr *)l4h, saddr, pkt)) continue;

@@ -969,12 +976,13 @@ resume:

if (l4len < sizeof(*uh)) continue; - uh = (struct udphdr *)l4h; + uh = IOV_PEEK_HEADER(&data, uh_storage);

And here.

+ if (!uh) + continue;

if (proto == IPPROTO_UDP) { PACKET_POOL_P(pkt, 1, in->buf, in->buf_size);

- data = IOV_TAIL_FROM_BUF(l4h, l4len, 0); packet_add(pkt, &data);

if (dhcpv6(c, pkt, saddr, daddr)) @@ -1031,7 +1039,6 @@ resume: #undef L4_SET

append: - data = IOV_TAIL_FROM_BUF(l4h, l4len, 0); packet_add((struct pool *)&seq->p, &data); }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Fri, Aug 08, 2025 at 03:57:04PM +0200, Laurent Vivier wrote:

On 06/08/2025 08:21, David Gibson wrote:

...

On Tue, Aug 05, 2025 at 05:46:18PM +0200, Laurent Vivier wrote:

...

Use packet_data() and extract headers using IOV_REMOVE_HEADER() and IOV_PEEK_HEADER() rather than packet_get().

Remove packet_get() as it is not used anymore.

[snip]

...

@@ -896,21 +896,28 @@ resume: for (seq_count = 0, seq = NULL; i < in->count; i++) { size_t l4len, plen, check; struct in6_addr *saddr, *daddr; + struct ipv6hdr ip6h_storage; + struct ethhdr eh_storage; + struct udphdr uh_storage; const struct ethhdr *eh; const struct udphdr *uh; struct iov_tail data; struct ipv6hdr *ip6h; uint8_t proto; - char *l4h; - eh = packet_get(in, i, 0, sizeof(*eh), NULL); + if (!packet_data(in, i, &data)) + return -1; + + eh = IOV_REMOVE_HEADER(&data, eh_storage); if (!eh) continue; - ip6h = packet_get(in, i, sizeof(*eh), sizeof(*ip6h), &check); + ip6h = IOV_PEEK_HEADER(&data, ip6h_storage); if (!ip6h) continue;

You peek the IPv6 header here, but I haven't spotted where you remove / drop it before...

In fact, data offset is modified by ipv6_l4hdr() that scans the headers and the data provided to ipv6_l4hdr() must point to an ipv6hdr (it provides also the proto that is alloed to do the correct IOV_PEEK_HEADER()).

Ah, right. -- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:19PM +0200, Laurent Vivier wrote:

As we have removed packet_get(), we can rename packet_data() to packet_get() as the name is clearer.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- arp.c | 2 +- dhcp.c | 2 +- dhcpv6.c | 2 +- icmp.c | 2 +- ndp.c | 2 +- packet.c | 8 ++++---- packet.h | 9 ++++----- tap.c | 6 +++--- tcp.c | 4 ++-- udp.c | 4 ++-- 10 files changed, 20 insertions(+), 21 deletions(-)

diff --git a/arp.c b/arp.c index b3ac42082841..8b97df633e70 100644 --- a/arp.c +++ b/arp.c @@ -82,7 +82,7 @@ int arp(const struct ctx *c, const struct pool *p) const struct arpmsg *am; struct iov_tail data;

- if (!packet_data(p, 0, &data)) + if (!packet_get(p, 0, &data)) return -1;

eh = IOV_REMOVE_HEADER(&data, eh_storage); diff --git a/dhcp.c b/dhcp.c index cf73d4b07767..47317f334945 100644 --- a/dhcp.c +++ b/dhcp.c @@ -317,7 +317,7 @@ int dhcp(const struct ctx *c, const struct pool *p) unsigned int i; struct msg m_storage;

- if (!packet_data(p, 0, &data)) + if (!packet_get(p, 0, &data)) return -1;

eh = IOV_REMOVE_HEADER(&data, eh_storage); diff --git a/dhcpv6.c b/dhcpv6.c index e93acaf9955e..f54a75c642df 100644 --- a/dhcpv6.c +++ b/dhcpv6.c @@ -562,7 +562,7 @@ int dhcpv6(struct ctx *c, const struct pool *p, const struct udphdr *uh; size_t mlen, n;

- if (!packet_data(p, 0, &data)) + if (!packet_get(p, 0, &data)) return -1;

uh = IOV_REMOVE_HEADER(&data, uh_storage); diff --git a/icmp.c b/icmp.c index fdfc857b5ae8..71c496540310 100644 --- a/icmp.c +++ b/icmp.c @@ -251,7 +251,7 @@ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, (void)saddr; ASSERT(pif == PIF_TAP);

- if (!packet_data(p, 0, &data)) + if (!packet_get(p, 0, &data)) return -1;

if (af == AF_INET) { diff --git a/ndp.c b/ndp.c index 5de4e508dc52..ba87a0aaa6e9 100644 --- a/ndp.c +++ b/ndp.c @@ -354,7 +354,7 @@ int ndp(const struct ctx *c, const struct icmp6hdr *ih, const struct ndp_ns *ns; struct iov_tail data;

- if (!packet_data(p, 0, &data)) + if (!packet_get(p, 0, &data)) return -1;

ns = IOV_REMOVE_HEADER(&data, ns_storage); diff --git a/packet.c b/packet.c index 5da18bafa576..cbc43c2fc22d 100644 --- a/packet.c +++ b/packet.c @@ -122,7 +122,7 @@ void packet_add_do(struct pool *p, struct iov_tail *data, }

/** - * packet_data_do() - Get data range from packet descriptor from given pool + * packet_get_do() - Get data range from packet descriptor from given pool * @p: Packet pool * @idx: Index of packet descriptor in pool * @data: IOV tail to store the address of the data (output) @@ -132,9 +132,9 @@ void packet_add_do(struct pool *p, struct iov_tail *data, * Return: false if packet index is invalid, true otherwise. * If something wrong with @data, don't return at all (assert). */ -bool packet_data_do(const struct pool *p, size_t idx, - struct iov_tail *data, - const char *func, int line) +bool packet_get_do(const struct pool *p, size_t idx, + struct iov_tail *data, + const char *func, int line) { size_t i;

diff --git a/packet.h b/packet.h index dab8274fa5c5..7afe80ef3fcf 100644 --- a/packet.h +++ b/packet.h @@ -33,16 +33,15 @@ struct pool { int vu_packet_check_range(void *buf, const char *ptr, size_t len); void packet_add_do(struct pool *p, struct iov_tail *data, const char *func, int line); -bool packet_data_do(const struct pool *p, const size_t idx, - struct iov_tail *data, - const char *func, int line); +bool packet_get_do(const struct pool *p, const size_t idx, + struct iov_tail *data, const char *func, int line); bool pool_full(const struct pool *p); void pool_flush(struct pool *p);

#define packet_add(p, data) \ packet_add_do(p, data, __func__, __LINE__) -#define packet_data(p, idx, data) \ - packet_data_do(p, idx, data, __func__, __LINE__) +#define packet_get(p, idx, data) \ + packet_get_do(p, idx, data, __func__, __LINE__)

#define PACKET_POOL_DECL(_name, _size, _buf) \ struct _name ## _t { \ diff --git a/tap.c b/tap.c index 983f39ee8ee8..1d2e6fd802e9 100644 --- a/tap.c +++ b/tap.c @@ -715,7 +715,7 @@ resume: struct iov_tail data; struct iphdr *iph;

- if (!packet_data(in, i, &data)) + if (!packet_get(in, i, &data)) continue;

eh = IOV_PEEK_HEADER(&data, eh_storage); @@ -790,7 +790,7 @@ resume:

PACKET_POOL_P(pkt, 1, in->buf, in->buf_size);

- packet_data(in, i, &eh_data); + packet_get(in, i, &eh_data); packet_add(pkt, &eh_data); if (dhcp(c, pkt)) continue; @@ -905,7 +905,7 @@ resume: struct ipv6hdr *ip6h; uint8_t proto;

- if (!packet_data(in, i, &data)) + if (!packet_get(in, i, &data)) return -1;

eh = IOV_REMOVE_HEADER(&data, eh_storage); diff --git a/tcp.c b/tcp.c index e0efc4cacb9b..4ba066fd1cac 100644 --- a/tcp.c +++ b/tcp.c @@ -1657,7 +1657,7 @@ static int tcp_data_from_tap(const struct ctx *c, struct tcp_tap_conn *conn, size_t off, size; int count;

- if (!packet_data(p, i, &data)) + if (!packet_get(p, i, &data)) return -1;

th = IOV_PEEK_HEADER(&data, th_storage); @@ -1988,7 +1988,7 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,

(void)pif;

- if (!packet_data(p, idx, &data)) + if (!packet_get(p, idx, &data)) return 1;

l4len = iov_tail_size(&data); diff --git a/udp.c b/udp.c index 3c25f2e0ae97..86585b7e0942 100644 --- a/udp.c +++ b/udp.c @@ -990,7 +990,7 @@ int udp_tap_handler(const struct ctx *c, uint8_t pif,

ASSERT(!c->no_udp);

- if (!packet_data(p, idx, &data)) + if (!packet_get(p, idx, &data)) return 1;

uh = IOV_PEEK_HEADER(&data, uh_storage); @@ -1033,7 +1033,7 @@ int udp_tap_handler(const struct ctx *c, uint8_t pif, for (i = 0, j = 0; i < (int)p->count - idx && j < UIO_MAXIOV; i++) { const struct udphdr *uh_send;

- if (!packet_data(p, idx + i, &data)) + if (!packet_get(p, idx + i, &data)) return p->count - idx;

uh_send = IOV_REMOVE_HEADER(&data, uh_storage);

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:20PM +0200, Laurent Vivier wrote:

The arp() function signature is changed to accept `struct iov_tail *data` directly, replacing the previous `const struct pool *p` parameter. Consequently, arp() no longer fetches packet data internally using packet_data(), streamlining its logic.

This simplifies callers like tap4_handler(), which now pass the iov_tail for the L2 ARP frame directly, removing intermediate pool handling.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- arp.c | 14 +++++--------- arp.h | 2 +- tap.c | 5 +---- 3 files changed, 7 insertions(+), 14 deletions(-)

diff --git a/arp.c b/arp.c index 8b97df633e70..44677ad15b93 100644 --- a/arp.c +++ b/arp.c @@ -63,11 +63,11 @@ static bool ignore_arp(const struct ctx *c, /** * arp() - Check if this is a supported ARP message, reply as needed * @c: Execution context - * @p: Packet pool, single packet with Ethernet buffer + * @data: Single packet with Ethernet buffer * * Return: 1 if handled, -1 on failure */ -int arp(const struct ctx *c, const struct pool *p) +int arp(const struct ctx *c, struct iov_tail *data) { struct { struct ethhdr eh; @@ -80,14 +80,10 @@ int arp(const struct ctx *c, const struct pool *p) const struct ethhdr *eh; const struct arphdr *ah; const struct arpmsg *am; - struct iov_tail data;

- if (!packet_get(p, 0, &data)) - return -1; - - eh = IOV_REMOVE_HEADER(&data, eh_storage); - ah = IOV_REMOVE_HEADER(&data, ah_storage); - am = IOV_REMOVE_HEADER(&data, am_storage); + eh = IOV_REMOVE_HEADER(data, eh_storage); + ah = IOV_REMOVE_HEADER(data, ah_storage); + am = IOV_REMOVE_HEADER(data, am_storage); if (!eh || !ah || !am) return -1;

diff --git a/arp.h b/arp.h index ac5cd16e47f4..86bcbf878eda 100644 --- a/arp.h +++ b/arp.h @@ -20,6 +20,6 @@ struct arpmsg { unsigned char tip[4]; } __attribute__((__packed__));

-int arp(const struct ctx *c, const struct pool *p); +int arp(const struct ctx *c, struct iov_tail *data);

#endif /* ARP_H */ diff --git a/tap.c b/tap.c index 1d2e6fd802e9..ace735cfc136 100644 --- a/tap.c +++ b/tap.c @@ -722,10 +722,7 @@ resume: if (!eh) continue; if (ntohs(eh->h_proto) == ETH_P_ARP) { - PACKET_POOL_P(pkt, 1, in->buf, in->buf_size); - - packet_add(pkt, &data); - arp(c, pkt); + arp(c, &data); continue; }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:21PM +0200, Laurent Vivier wrote:

This patch refactors the dhcp() function to accept `struct iov_tail *data` directly as its packet input, replacing the previous `const struct pool *p` parameter. Consequently, dhcp() no longer fetches packet data internally using packet_data().

This change simplifies callers, such as tap4_handler(), which now pass the iov_tail representing the L2 frame directly to dhcp(). This removes the need for intermediate packet pool handling for DHCP processing.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- dhcp.c | 32 ++++++++++++++------------------ dhcp.h | 2 +- tap.c | 5 +---- 3 files changed, 16 insertions(+), 23 deletions(-)

diff --git a/dhcp.c b/dhcp.c index 47317f334945..cad1d037cee1 100644 --- a/dhcp.c +++ b/dhcp.c @@ -296,11 +296,11 @@ static void opt_set_dns_search(const struct ctx *c, size_t max_len) /** * dhcp() - Check if this is a DHCP message, reply as needed * @c: Execution context - * @p: Packet pool, single packet with Ethernet buffer + * @data: Single packet with Ethernet buffer * * Return: 0 if it's not a DHCP message, 1 if handled, -1 on failure */ -int dhcp(const struct ctx *c, const struct pool *p) +int dhcp(const struct ctx *c, struct iov_tail *data) { char macstr[ETH_ADDRSTRLEN]; size_t mlen, dlen, opt_len; @@ -311,24 +311,20 @@ int dhcp(const struct ctx *c, const struct pool *p) const struct ethhdr *eh; const struct iphdr *iph; const struct udphdr *uh; - struct iov_tail data; struct msg const *m; struct msg reply; unsigned int i; struct msg m_storage;

- if (!packet_get(p, 0, &data)) - return -1; - - eh = IOV_REMOVE_HEADER(&data, eh_storage); - iph = IOV_PEEK_HEADER(&data, iph_storage); + eh = IOV_REMOVE_HEADER(data, eh_storage); + iph = IOV_PEEK_HEADER(data, iph_storage); if (!eh || !iph) return -1;

- if (!iov_tail_drop(&data, iph->ihl * 4UL)) + if (!iov_tail_drop(data, iph->ihl * 4UL)) return -1;

- uh = IOV_REMOVE_HEADER(&data, uh_storage); + uh = IOV_REMOVE_HEADER(data, uh_storage); if (!uh) return -1;

@@ -338,8 +334,8 @@ int dhcp(const struct ctx *c, const struct pool *p) if (c->no_dhcp) return 1;

- mlen = iov_tail_size(&data); - m = (struct msg const *)iov_remove_header_(&data, &m_storage, + mlen = iov_tail_size(data); + m = (struct msg const *)iov_remove_header_(data, &m_storage, offsetof(struct msg, o), __alignof__(struct msg)); if (!m || @@ -367,24 +363,24 @@ int dhcp(const struct ctx *c, const struct pool *p) for (i = 0; i < ARRAY_SIZE(opts); i++) opts[i].clen = -1;

- opt_len = iov_tail_size(&data); + opt_len = iov_tail_size(data); while (opt_len >= 2) { uint8_t olen_storage, type_storage; const uint8_t *olen; uint8_t *type;

- type = IOV_REMOVE_HEADER(&data, type_storage); - olen = IOV_REMOVE_HEADER(&data, olen_storage); + type = IOV_REMOVE_HEADER(data, type_storage); + olen = IOV_REMOVE_HEADER(data, olen_storage); if (!type || !olen) return -1;

- opt_len = iov_tail_size(&data); + opt_len = iov_tail_size(data); if (opt_len < *olen) return -1;

- iov_to_buf(&data.iov[0], data.cnt, data.off, &opts[*type].c, *olen); + iov_to_buf(&data->iov[0], data->cnt, data->off, &opts[*type].c, *olen); opts[*type].clen = *olen; - iov_tail_drop(&data, *olen); + iov_tail_drop(data, *olen); opt_len -= *olen; }

diff --git a/dhcp.h b/dhcp.h index 87aeecd8dec8..cd50c99b8856 100644 --- a/dhcp.h +++ b/dhcp.h @@ -6,7 +6,7 @@ #ifndef DHCP_H #define DHCP_H

-int dhcp(const struct ctx *c, const struct pool *p); +int dhcp(const struct ctx *c, struct iov_tail *data); void dhcp_init(void);

#endif /* DHCP_H */ diff --git a/tap.c b/tap.c index ace735cfc136..7d7e89304723 100644 --- a/tap.c +++ b/tap.c @@ -785,11 +785,8 @@ resume: if (iph->protocol == IPPROTO_UDP) { struct iov_tail eh_data;

- PACKET_POOL_P(pkt, 1, in->buf, in->buf_size); - packet_get(in, i, &eh_data); - packet_add(pkt, &eh_data); - if (dhcp(c, pkt)) + if (dhcp(c, &eh_data)) continue; }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:22PM +0200, Laurent Vivier wrote:

This patch refactors the dhcpv6() function to accept `struct iov_tail *data` directly as its packet input, replacing the `const struct pool *p` parameter. Consequently, dhcpv6() no longer fetches packet data internally using packet_data().

This change simplifies callers, such as tap6_handler(), which now pass the iov_tail representing the L4 UDP segment (DHCPv6 message) directly. This removes the need for intermediate packet pool handling.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- dhcpv6.c | 29 +++++++++++++---------------- dhcpv6.h | 2 +- tap.c | 6 ++---- 3 files changed, 16 insertions(+), 21 deletions(-)

diff --git a/dhcpv6.c b/dhcpv6.c index f54a75c642df..52611d742b0b 100644 --- a/dhcpv6.c +++ b/dhcpv6.c @@ -539,19 +539,19 @@ static size_t dhcpv6_client_fqdn_fill(const struct iov_tail *data, /** * dhcpv6() - Check if this is a DHCPv6 message, reply as needed * @c: Execution context - * @p: Packet pool, single packet starting from UDP header + * @data: Single packet starting from UDP header * @saddr: Source IPv6 address of original message * @daddr: Destination IPv6 address of original message * * Return: 0 if it's not a DHCPv6 message, 1 if handled, -1 on failure */ -int dhcpv6(struct ctx *c, const struct pool *p, +int dhcpv6(struct ctx *c, struct iov_tail *data, const struct in6_addr *saddr, const struct in6_addr *daddr) { const struct opt_server_id *server_id = NULL; - struct iov_tail data, opt, client_id_base; const struct opt_hdr *client_id = NULL; struct opt_server_id server_id_storage; + struct iov_tail opt, client_id_base; const struct opt_ia_na *ia = NULL; struct opt_hdr client_id_storage; struct opt_ia_na ia_storage; @@ -562,10 +562,7 @@ int dhcpv6(struct ctx *c, const struct pool *p, const struct udphdr *uh; size_t mlen, n;

- if (!packet_get(p, 0, &data)) - return -1; - - uh = IOV_REMOVE_HEADER(&data, uh_storage); + uh = IOV_REMOVE_HEADER(data, uh_storage); if (!uh) return -1;

@@ -578,7 +575,7 @@ int dhcpv6(struct ctx *c, const struct pool *p, if (!IN6_IS_ADDR_MULTICAST(daddr)) return -1;

- mlen = iov_tail_size(&data); + mlen = iov_tail_size(data); if (mlen + sizeof(*uh) != ntohs(uh->len) || mlen < sizeof(*mh)) return -1;

@@ -586,23 +583,23 @@ int dhcpv6(struct ctx *c, const struct pool *p,

src = &c->ip6.our_tap_ll;

- mh = IOV_REMOVE_HEADER(&data, mh_storage); + mh = IOV_REMOVE_HEADER(data, mh_storage); if (!mh) return -1;

- client_id_base = data; + client_id_base = *data; if (dhcpv6_opt(&client_id_base, OPT_CLIENTID)) client_id = IOV_PEEK_HEADER(&client_id_base, client_id_storage); if (!client_id || ntohs(client_id->l) > OPT_VSIZE(client_id)) return -1;

- opt = data; + opt = *data; if (dhcpv6_opt(&opt, OPT_SERVERID)) server_id = IOV_PEEK_HEADER(&opt, server_id_storage); if (server_id && ntohs(server_id->hdr.l) != OPT_VSIZE(server_id)) return -1;

- opt = data; + opt = *data; if (dhcpv6_opt(&opt, OPT_IA_NA)) ia = IOV_PEEK_HEADER(&opt, ia_storage); if (ia && ntohs(ia->hdr.l) < MIN(OPT_VSIZE(ia_na), OPT_VSIZE(ia_ta))) @@ -620,9 +617,9 @@ int dhcpv6(struct ctx *c, const struct pool *p, if (mh->type == TYPE_CONFIRM && server_id) return -1;

- if (dhcpv6_ia_notonlink(&data, &c->ip6.addr)) { + if (dhcpv6_ia_notonlink(data, &c->ip6.addr)) {

- dhcpv6_send_ia_notonlink(c, &data, &client_id_base, + dhcpv6_send_ia_notonlink(c, data, &client_id_base, ntohs(client_id->l), mh->xid);

return 1; @@ -635,7 +632,7 @@ int dhcpv6(struct ctx *c, const struct pool *p, memcmp(&resp.server_id, server_id, sizeof(resp.server_id))) return -1;

- if (ia || dhcpv6_opt(&data, OPT_IA_TA)) + if (ia || dhcpv6_opt(data, OPT_IA_TA)) return -1;

info("DHCPv6: received INFORMATION_REQUEST, sending REPLY"); @@ -668,7 +665,7 @@ int dhcpv6(struct ctx *c, const struct pool *p, n = offsetof(struct resp_t, client_id) + sizeof(struct opt_hdr) + ntohs(client_id->l); n = dhcpv6_dns_fill(c, (char *)&resp, n); - n = dhcpv6_client_fqdn_fill(&data, c, (char *)&resp, n); + n = dhcpv6_client_fqdn_fill(data, c, (char *)&resp, n);

resp.hdr.xid = mh->xid;

diff --git a/dhcpv6.h b/dhcpv6.h index 580998862227..c706dfdbb2ac 100644 --- a/dhcpv6.h +++ b/dhcpv6.h @@ -6,7 +6,7 @@ #ifndef DHCPV6_H #define DHCPV6_H

-int dhcpv6(struct ctx *c, const struct pool *p, +int dhcpv6(struct ctx *c, struct iov_tail *data, struct in6_addr *saddr, struct in6_addr *daddr); void dhcpv6_init(const struct ctx *c);

diff --git a/tap.c b/tap.c index 7d7e89304723..3262b44c4287 100644 --- a/tap.c +++ b/tap.c @@ -975,11 +975,9 @@ resume: continue;

if (proto == IPPROTO_UDP) { - PACKET_POOL_P(pkt, 1, in->buf, in->buf_size); - - packet_add(pkt, &data); + struct iov_tail uh_data = data;

- if (dhcpv6(c, pkt, saddr, daddr)) + if (dhcpv6(c, &uh_data, saddr, daddr)) continue; }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:23PM +0200, Laurent Vivier wrote:

As the iov_tail has a non zero offset (because of the presence of packet headers in the iov array), we must copy it to a new iov array (using iov_tail_splice()) to pass it to sendmsg().

We cannot use anymore iov_tail_msghdr(), so remove it.

Right, as I mentioned on that patch, it was always kind of a foot gun. Was this the only user? Maybe just open code it earlier in the series rather than introducing then removing the helper.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- icmp.c | 30 +++++++++++++++++++----------- icmp.h | 2 +- iov.c | 23 ----------------------- iov.h | 2 -- tap.c | 7 ++----- 5 files changed, 22 insertions(+), 42 deletions(-)

diff --git a/icmp.c b/icmp.c index 71c496540310..be800e30c369 100644 --- a/icmp.c +++ b/icmp.c @@ -44,6 +44,7 @@

#define ICMP_ECHO_TIMEOUT 60 /* s, timeout for ICMP socket activity */ #define ICMP_NUM_IDS (1U << 16) +#define MAX_IOV_ICMP 16 /* Arbitrary, should be enough */

/** * ping_at_sidx() - Get ping specific flow at given sidx @@ -229,36 +230,33 @@ cancel: * @af: Address family, AF_INET or AF_INET6 * @saddr: Source address * @daddr: Destination address - * @p: Packet pool, single packet with ICMP/ICMPv6 header + * @data: Single packet with ICMP/ICMPv6 header * @now: Current timestamp * * Return: count of consumed packets (always 1, even if malformed) */ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, const void *saddr, const void *daddr, - const struct pool *p, const struct timespec *now) + struct iov_tail *data, const struct timespec *now) { + struct iovec iov[MAX_IOV_ICMP]; struct icmp_ping_flow *pingf; const struct flowside *tgt; union sockaddr_inany sa; - struct iov_tail data; struct msghdr msh; uint16_t id, seq; union flow *flow; uint8_t proto; - socklen_t sl; + int cnt;

(void)saddr; ASSERT(pif == PIF_TAP);

- if (!packet_get(p, 0, &data)) - return -1; - if (af == AF_INET) { struct icmphdr ih_storage; const struct icmphdr *ih;

- ih = IOV_PEEK_HEADER(&data, ih_storage); + ih = IOV_PEEK_HEADER(data, ih_storage); if (!ih) return 1;

@@ -272,7 +270,7 @@ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, struct icmp6hdr ih_storage; const struct icmp6hdr *ih;

- ih = IOV_PEEK_HEADER(&data, ih_storage); + ih = IOV_PEEK_HEADER(data, ih_storage); if (!ih) return 1;

@@ -286,6 +284,10 @@ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, ASSERT(0); }

+ cnt = iov_tail_clone(&iov[0], MAX_IOV_ICMP, data); + if (cnt < 0) + return 1; + flow = flow_at_sidx(flow_lookup_af(c, proto, PIF_TAP, af, saddr, daddr, id, id));

@@ -300,8 +302,14 @@ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, pingf->ts = now->tv_sec;

- pif_sockaddr(c, &sa, &sl, PIF_HOST, &tgt->eaddr, 0); - iov_tail_msghdr(&msh, &data, &sa, sl); + pif_sockaddr(c, &sa, &msh.msg_namelen, PIF_HOST, &tgt->eaddr, 0); + msh.msg_name = &sa; + msh.msg_iov = iov; + msh.msg_iovlen = cnt; + msh.msg_control = NULL; + msh.msg_controllen = 0; + msh.msg_flags = 0; + if (sendmsg(pingf->sock, &msh, MSG_NOSIGNAL) < 0) { flow_dbg_perror(pingf, "failed to relay request to socket"); } else { diff --git a/icmp.h b/icmp.h index 5ce22b5eca1f..d1cecb20e29d 100644 --- a/icmp.h +++ b/icmp.h @@ -14,7 +14,7 @@ struct icmp_ping_flow; void icmp_sock_handler(const struct ctx *c, union epoll_ref ref); int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af, const void *saddr, const void *daddr, - const struct pool *p, const struct timespec *now); + struct iov_tail *data, const struct timespec *now); void icmp_init(void);

/** diff --git a/iov.c b/iov.c index d17d4dd3da09..1d734acdfea6 100644 --- a/iov.c +++ b/iov.c @@ -157,29 +157,6 @@ size_t iov_size(const struct iovec *iov, size_t iov_cnt) return len; }

-/** - * iov_tail_msghdr - Initialize a msghdr from an IOV tail structure - * @msh: msghdr to initialize - * @tail: iov_tail to use to set msg_iov and msg_iovlen - * @msg_name: Pointer to set to msg_name - * @msg_namelen: Size of @msg_name - */ -void iov_tail_msghdr(struct msghdr *msh, struct iov_tail *tail, - void *msg_name, socklen_t msg_namelen) -{ - iov_tail_prune(tail); - - ASSERT(tail->off == 0); - - msh->msg_name = msg_name; - msh->msg_namelen = msg_namelen; - msh->msg_iov = (struct iovec *)tail->iov; - msh->msg_iovlen = tail->cnt; - msh->msg_control = NULL; - msh->msg_controllen = 0; - msh->msg_flags = 0; -} - /** * iov_tail_prune() - Remove any unneeded buffers from an IOV tail * @tail: IO vector tail (modified) diff --git a/iov.h b/iov.h index 75c3b07a87e3..ccdb690ef3f1 100644 --- a/iov.h +++ b/iov.h @@ -82,8 +82,6 @@ struct iov_tail { 1, \ (off_))

-void iov_tail_msghdr(struct msghdr *msh, struct iov_tail *tail, - void *msg_name, socklen_t msg_namelen); bool iov_tail_prune(struct iov_tail *tail); size_t iov_tail_size(struct iov_tail *tail); bool iov_tail_drop(struct iov_tail *tail, size_t len); diff --git a/tap.c b/tap.c index 3262b44c4287..48152a84674c 100644 --- a/tap.c +++ b/tap.c @@ -764,17 +764,14 @@ resume: continue;

if (iph->protocol == IPPROTO_ICMP) { - PACKET_POOL_P(pkt, 1, in->buf, in->buf_size); - if (c->no_icmp) continue;

tap_packet_debug(iph, NULL, NULL, 0, NULL, 1);

- packet_add(pkt, &data); icmp_tap_handler(c, PIF_TAP, AF_INET, &iph->saddr, &iph->daddr, - pkt, now); + &data, now); continue; }

@@ -964,7 +961,7 @@ resume: tap_packet_debug(NULL, ip6h, NULL, proto, NULL, 1);

icmp_tap_handler(c, PIF_TAP, AF_INET6, - saddr, daddr, pkt, now); + saddr, daddr, &data, now); continue; }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:24PM +0200, Laurent Vivier wrote:

The ndp() function signature is changed to accept `struct iov_tail *data` directly, replacing the previous `const struct pool *p` and `const struct icmp6hdr *ih` parameters.

This change simplifies callers, like tap6_handler(), which now provide the iov_tail representing the L4 ICMPv6 segment directly to ndp().

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- ndp.c | 19 +++++++++++-------- ndp.h | 4 ++-- tap.c | 10 +++------- 3 files changed, 16 insertions(+), 17 deletions(-)

diff --git a/ndp.c b/ndp.c index ba87a0aaa6e9..eb090cd2c5a7 100644 --- a/ndp.c +++ b/ndp.c @@ -336,13 +336,20 @@ static void ndp_ra(const struct ctx *c, const struct in6_addr *dst) * ndp() - Check for NDP solicitations, reply as needed * @c: Execution context * @saddr: Source IPv6 address - * @p: Packet pool + * @data: Single packet with ICMPv6 header * * Return: 0 if not handled here, 1 if handled, -1 on failure */ -int ndp(const struct ctx *c, const struct icmp6hdr *ih, - const struct in6_addr *saddr, const struct pool *p) +int ndp(const struct ctx *c, const struct in6_addr *saddr, + struct iov_tail *data) { + struct icmp6hdr ih_storage; + const struct icmp6hdr *ih; + + ih = IOV_PEEK_HEADER(data, ih_storage); + if (!ih) + return -1; + if (ih->icmp6_type < RS || ih->icmp6_type > NA) return 0;

@@ -352,12 +359,8 @@ int ndp(const struct ctx *c, const struct icmp6hdr *ih, if (ih->icmp6_type == NS) { struct ndp_ns ns_storage; const struct ndp_ns *ns; - struct iov_tail data; - - if (!packet_get(p, 0, &data)) - return -1;

- ns = IOV_REMOVE_HEADER(&data, ns_storage); + ns = IOV_REMOVE_HEADER(data, ns_storage); if (!ns) return -1;

diff --git a/ndp.h b/ndp.h index 41c2000356ec..b1dd5e82c085 100644 --- a/ndp.h +++ b/ndp.h @@ -8,8 +8,8 @@

struct icmp6hdr;

-int ndp(const struct ctx *c, const struct icmp6hdr *ih, - const struct in6_addr *saddr, const struct pool *p); +int ndp(const struct ctx *c, const struct in6_addr *saddr, + struct iov_tail *data); void ndp_timer(const struct ctx *c, const struct timespec *now);

#endif /* NDP_H */ diff --git a/tap.c b/tap.c index 48152a84674c..d327ec0c3d54 100644 --- a/tap.c +++ b/tap.c @@ -942,9 +942,7 @@ resume: }

if (proto == IPPROTO_ICMPV6) { - struct icmp6hdr l4h_storage; - const struct icmp6hdr *l4h; - PACKET_POOL_P(pkt, 1, in->buf, in->buf_size); + struct iov_tail ndp_data;

if (c->no_icmp) continue; @@ -952,10 +950,8 @@ resume: if (l4len < sizeof(struct icmp6hdr)) continue;

- packet_add(pkt, &data); - - l4h = IOV_PEEK_HEADER(&data, l4h_storage); - if (ndp(c, (struct icmp6hdr *)l4h, saddr, pkt)) + ndp_data = data; + if (ndp(c, saddr, &ndp_data)) continue;

tap_packet_debug(NULL, ip6h, NULL, proto, NULL, 1);

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:25PM +0200, Laurent Vivier wrote:

These macros are no longer used following the refactoring of packet handlers to directly use iov_tail. Callers no longer require PACKET_POOL_P for temporary pools, and PACKET_POOL can be replaced by PACKET_POOL_DECL and separate initialization if needed.

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- packet.h | 9 --------- 1 file changed, 9 deletions(-)

diff --git a/packet.h b/packet.h index 7afe80ef3fcf..286b6b9994db 100644 --- a/packet.h +++ b/packet.h @@ -59,19 +59,10 @@ struct _name ## _t { \ .size = _size, \ }

-#define PACKET_POOL(name, size, buf, buf_size) \ - PACKET_POOL_DECL(name, size, buf) name = \ - PACKET_POOL_INIT_NOCAST(size, buf, buf_size) - #define PACKET_INIT(name, size, buf, buf_size) \ (struct name ## _t) PACKET_POOL_INIT_NOCAST(size, buf, buf_size)

#define PACKET_POOL_NOINIT(name, size, buf) \ PACKET_POOL_DECL(name, size, buf) name ## _storage; \ static struct pool *name = (struct pool *)&name ## _storage - -#define PACKET_POOL_P(name, size, buf, buf_size) \ - PACKET_POOL(name ## _storage, size, buf, buf_size); \ - struct pool *name = (struct pool *)&name ## _storage - #endif /* PACKET_H */

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:26PM +0200, Laurent Vivier wrote:

_buf is not used in the macro. Remove it. Remove it also from PACKET_POOL_NOINIT() as it was needed for PACKET_POOL_DECL().

Signed-off-by: Laurent Vivier

Reviewed-by: David Gibson

--- packet.h | 6 +++--- tap.c | 6 +++--- 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/packet.h b/packet.h index 286b6b9994db..43b9022075d1 100644 --- a/packet.h +++ b/packet.h @@ -43,7 +43,7 @@ void pool_flush(struct pool *p); #define packet_get(p, idx, data) \ packet_get_do(p, idx, data, __func__, __LINE__)

-#define PACKET_POOL_DECL(_name, _size, _buf) \ +#define PACKET_POOL_DECL(_name, _size) \ struct _name ## _t { \ char *buf; \ size_t buf_size; \ @@ -62,7 +62,7 @@ struct _name ## _t { \ #define PACKET_INIT(name, size, buf, buf_size) \ (struct name ## _t) PACKET_POOL_INIT_NOCAST(size, buf, buf_size)

-#define PACKET_POOL_NOINIT(name, size, buf) \ - PACKET_POOL_DECL(name, size, buf) name ## _storage; \ +#define PACKET_POOL_NOINIT(name, size) \ + PACKET_POOL_DECL(name, size) name ## _storage; \ static struct pool *name = (struct pool *)&name ## _storage #endif /* PACKET_H */ diff --git a/tap.c b/tap.c index d327ec0c3d54..bbc786468455 100644 --- a/tap.c +++ b/tap.c @@ -95,8 +95,8 @@ CHECK_FRAME_LEN(L2_MAX_LEN_VU); ETH_HLEN + sizeof(struct ipv6hdr) + sizeof(struct udphdr))

/* IPv4 (plus ARP) and IPv6 message batches from tap/guest to IP handlers */ -static PACKET_POOL_NOINIT(pool_tap4, TAP_MSGS_IP4, pkt_buf); -static PACKET_POOL_NOINIT(pool_tap6, TAP_MSGS_IP6, pkt_buf); +static PACKET_POOL_NOINIT(pool_tap4, TAP_MSGS_IP4); +static PACKET_POOL_NOINIT(pool_tap6, TAP_MSGS_IP6);

#define TAP_SEQS 128 /* Different L4 tuples in one batch */ #define FRAGMENT_MSG_RATE 10 /* # seconds between fragment warnings */ @@ -555,7 +555,7 @@ void eth_update_mac(struct ethhdr *eh, memcpy(eh->h_source, eth_s, sizeof(eh->h_source)); }

-PACKET_POOL_DECL(pool_l4, UIO_MAXIOV, pkt_buf); +PACKET_POOL_DECL(pool_l4, UIO_MAXIOV);

/** * struct l4_seq4_t - Message sequence for one protocol handler call, IPv4

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:27PM +0200, Laurent Vivier wrote:

This patch refactors the handling of vhost-user memory regions by introducing a new `struct vdev_memory` to encapsulate the regions array and their count (`nregions`) within the main `vu_dev` structure.

This new `vdev_memory` structure is then passed to the packet pool by re-using the existing `p->buf` field. A `p->buf_size` of 0 indicates that `p->buf` holds a pointer to `struct vdev_memory` instead of a regular packet buffer. A new helper, `get_vdev_memory()`, is added to abstract this access pattern.

Signed-off-by: Laurent Vivier --- packet.c | 18 ++++++++++++++++-- packet.h | 6 ++++-- tap.c | 4 ++-- tap.h | 1 - vhost_user.c | 28 +++++++++++----------------- virtio.c | 4 ++-- virtio.h | 18 ++++++++++++++---- vu_common.c | 22 ++++++++++++---------- 8 files changed, 61 insertions(+), 40 deletions(-)

diff --git a/packet.c b/packet.c index cbc43c2fc22d..4b93688509a4 100644 --- a/packet.c +++ b/packet.c @@ -22,6 +22,20 @@ #include "util.h" #include "log.h"

+/** + * get_vdev_memory() - Return a pointer to the memory regions of the pool + * @p: Packet pool + * + * Return: Null if none, otherwise a pointer to vdev_memory structure + */ +static struct vdev_memory *get_vdev_memory(const struct pool *p) +{ + if (p->buf_size) + return NULL; + + return (struct vdev_memory *)p->buf; +} + /** * packet_check_range() - Check if a memory range is valid for a pool * @p: Packet pool @@ -41,10 +55,10 @@ static int packet_check_range(const struct pool *p, const char *ptr, size_t len, return -1; }

- if (p->buf_size == 0) { + if (get_vdev_memory(p)) { int ret;

- ret = vu_packet_check_range((void *)p->buf, ptr, len); + ret = vu_packet_check_range(get_vdev_memory(p), ptr, len);

Seems like it would be marginally more natural to assign get_vdev_memory() to a temporary in the if, then re-use it here.

if (ret == -1) debug("cannot find region, %s:%i", func, line); diff --git a/packet.h b/packet.h index 43b9022075d1..e51cbd19fdc4 100644 --- a/packet.h +++ b/packet.h @@ -8,6 +8,7 @@

#include #include "iov.h" +#include "virtio.h"

/* Maximum size of a single packet stored in pool, including headers */ #define PACKET_MAX_LEN ((size_t)UINT16_MAX) @@ -15,7 +16,7 @@ /** * struct pool - Generic pool of packets stored in a buffer * @buf: Buffer storing packet descriptors, - * a struct vu_dev_region array for passt vhost-user mode + * a struct vdev_region for passt vhost-user mode * @buf_size: Total size of buffer, * 0 for passt vhost-user mode * @size: Number of usable descriptors for the pool @@ -30,7 +31,8 @@ struct pool { struct iovec pkt[]; };

-int vu_packet_check_range(void *buf, const char *ptr, size_t len); +int vu_packet_check_range(struct vdev_memory *memory, + const char *ptr, size_t len); void packet_add_do(struct pool *p, struct iov_tail *data, const char *func, int line); bool packet_get_do(const struct pool *p, const size_t idx, diff --git a/tap.c b/tap.c index bbc786468455..9fd00915bb01 100644 --- a/tap.c +++ b/tap.c @@ -1458,7 +1458,7 @@ static void tap_sock_tun_init(struct ctx *c) * @base: Buffer base * @size Buffer size */ -void tap_sock_update_pool(void *base, size_t size) +static void tap_sock_update_pool(void *base, size_t size) { int i;

@@ -1479,8 +1479,8 @@ void tap_sock_update_pool(void *base, size_t size) void tap_backend_init(struct ctx *c) { if (c->mode == MODE_VU) { - tap_sock_update_pool(NULL, 0); vu_init(c); + tap_sock_update_pool(&c->vdev->memory, 0); } else { tap_sock_update_pool(pkt_buf, sizeof(pkt_buf)); } diff --git a/tap.h b/tap.h index ce5510882d5d..21db4d219ecb 100644 --- a/tap.h +++ b/tap.h @@ -115,7 +115,6 @@ void tap_handler_passt(struct ctx *c, uint32_t events, const struct timespec *now); int tap_sock_unix_open(char *sock_path); void tap_sock_reset(struct ctx *c); -void tap_sock_update_pool(void *base, size_t size); void tap_backend_init(struct ctx *c); void tap_flush_pools(void); void tap_handler(struct ctx *c, const struct timespec *now); diff --git a/vhost_user.c b/vhost_user.c index c1522d549f00..f97ec6064cac 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -137,8 +137,8 @@ static void *qva_to_va(struct vu_dev *dev, uint64_t qemu_addr) unsigned int i;

/* Find matching memory region. */ - for (i = 0; i < dev->nregions; i++) { - const struct vu_dev_region *r = &dev->regions[i]; + for (i = 0; i < dev->memory.nregions; i++) { + const struct vu_dev_region *r = &dev->memory.regions[i];

if ((qemu_addr >= r->qva) && (qemu_addr < (r->qva + r->size))) { /* NOLINTNEXTLINE(performance-no-int-to-ptr) */ @@ -428,8 +428,8 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, struct vhost_user_memory m = vmsg->payload.memory, *memory = &m; unsigned int i;

- for (i = 0; i < vdev->nregions; i++) { - const struct vu_dev_region *r = &vdev->regions[i]; + for (i = 0; i < vdev->memory.nregions; i++) { + const struct vu_dev_region *r = &vdev->memory.regions[i];

if (r->mmap_addr) { /* NOLINTNEXTLINE(performance-no-int-to-ptr) */ @@ -437,12 +437,12 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, r->size + r->mmap_offset); } } - vdev->nregions = memory->nregions; + vdev->memory.nregions = memory->nregions;

debug("vhost-user nregions: %u", memory->nregions); - for (i = 0; i < vdev->nregions; i++) { + for (i = 0; i < vdev->memory.nregions; i++) { struct vhost_user_memory_region *msg_region = &memory->regions[i]; - struct vu_dev_region *dev_region = &vdev->regions[i]; + struct vu_dev_region *dev_region = &vdev->memory.regions[i]; void *mmap_addr;

debug("vhost-user region %d", i); @@ -484,13 +484,7 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, } }

- /* As vu_packet_check_range() has no access to the number of - * memory regions, mark the end of the array with mmap_addr = 0 - */ - ASSERT(vdev->nregions < VHOST_USER_MAX_RAM_SLOTS - 1); - vdev->regions[vdev->nregions].mmap_addr = 0; - - tap_sock_update_pool(vdev->regions, 0); + ASSERT(vdev->memory.nregions < VHOST_USER_MAX_RAM_SLOTS);

It looks like the assertion is changing threshold by one, and I'm not sure why.

return false; } @@ -1106,8 +1100,8 @@ void vu_cleanup(struct vu_dev *vdev) vq->vring.avail = 0; }

- for (i = 0; i < vdev->nregions; i++) { - const struct vu_dev_region *r = &vdev->regions[i]; + for (i = 0; i < vdev->memory.nregions; i++) { + const struct vu_dev_region *r = &vdev->memory.regions[i];

if (r->mmap_addr) { /* NOLINTNEXTLINE(performance-no-int-to-ptr) */ @@ -1115,7 +1109,7 @@ void vu_cleanup(struct vu_dev *vdev) r->size + r->mmap_offset); } } - vdev->nregions = 0; + vdev->memory.nregions = 0;

vu_close_log(vdev);

diff --git a/virtio.c b/virtio.c index ed7842b4c78a..bd388c2dfc7f 100644 --- a/virtio.c +++ b/virtio.c @@ -102,8 +102,8 @@ static void *vu_gpa_to_va(const struct vu_dev *dev, uint64_t *plen, return NULL;

/* Find matching memory region. */ - for (i = 0; i < dev->nregions; i++) { - const struct vu_dev_region *r = &dev->regions[i]; + for (i = 0; i < dev->memory.nregions; i++) { + const struct vu_dev_region *r = &dev->memory.regions[i];

if ((guest_addr >= r->gpa) && (guest_addr < (r->gpa + r->size))) { diff --git a/virtio.h b/virtio.h index 32757458ea95..b55cc4042521 100644 --- a/virtio.h +++ b/virtio.h @@ -96,11 +96,22 @@ struct vu_dev_region { */ #define VHOST_USER_MAX_RAM_SLOTS 32

+/** + * struct vdev_memory - Describes the shared memory regions for a vhost-user + * device + * @nregions: Number of shared memory regions + * @regions: Guest shared memory regions + */ +struct vdev_memory { + uint32_t nregions; + struct vu_dev_region regions[VHOST_USER_MAX_RAM_SLOTS]; +}; + /** * struct vu_dev - vhost-user device information * @context: Execution context - * @nregions: Number of shared memory regions - * @regions: Guest shared memory regions + * @memory: Shared memory regions + * @vq: Virtqueues of the device * @features: Vhost-user features * @protocol_features: Vhost-user protocol features * @log_call_fd: Eventfd to report logging update @@ -109,8 +120,7 @@ struct vu_dev_region { */ struct vu_dev { struct ctx *context; - uint32_t nregions; - struct vu_dev_region regions[VHOST_USER_MAX_RAM_SLOTS]; + struct vdev_memory memory; struct vu_virtq vq[VHOST_USER_MAX_QUEUES]; uint64_t features; uint64_t protocol_features; diff --git a/vu_common.c b/vu_common.c index b77b21420c57..b716070ea3c3 100644 --- a/vu_common.c +++ b/vu_common.c @@ -25,26 +25,28 @@ /** * vu_packet_check_range() - Check if a given memory zone is contained in * a mapped guest memory region - * @buf: Array of the available memory regions + * @memory: Array of the available memory regions * @ptr: Start of desired data range - * @size: Length of desired data range + * @len: Length of desired data range * * Return: 0 if the zone is in a mapped memory region, -1 otherwise */ -int vu_packet_check_range(void *buf, const char *ptr, size_t len) +int vu_packet_check_range(struct vdev_memory *memory, + const char *ptr, size_t len) { - struct vu_dev_region *dev_region; + struct vu_dev_region *dev_region = memory->regions; + unsigned int i;

- for (dev_region = buf; dev_region->mmap_addr; dev_region++) { - uintptr_t base_addr = dev_region->mmap_addr + - dev_region->mmap_offset; + for (i = 0; i < memory->nregions; i++) { + uintptr_t base_addr = dev_region[i].mmap_addr + + dev_region[i].mmap_offset; /* NOLINTNEXTLINE(performance-no-int-to-ptr) */ const char *base = (const char *)base_addr;

- ASSERT(base_addr >= dev_region->mmap_addr); + ASSERT(base_addr >= dev_region[i].mmap_addr);

- if (len <= dev_region->size && base <= ptr && - (size_t)(ptr - base) <= dev_region->size - len) + if (len <= dev_region[i].size && base <= ptr && + (size_t)(ptr - base) <= dev_region[i].size - len) return 0; }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Thu, Aug 07, 2025 at 11:05:46AM +0200, Laurent Vivier wrote:

On 07/08/2025 08:10, David Gibson wrote:

...

...

@@ -437,12 +437,12 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, r->size + r->mmap_offset); } } - vdev->nregions = memory->nregions; + vdev->memory.nregions = memory->nregions; debug("vhost-user nregions: %u", memory->nregions); - for (i = 0; i < vdev->nregions; i++) { + for (i = 0; i < vdev->memory.nregions; i++) { struct vhost_user_memory_region *msg_region = &memory->regions[i]; - struct vu_dev_region *dev_region = &vdev->regions[i]; + struct vu_dev_region *dev_region = &vdev->memory.regions[i]; void *mmap_addr; debug("vhost-user region %d", i); @@ -484,13 +484,7 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev, } } - /* As vu_packet_check_range() has no access to the number of - * memory regions, mark the end of the array with mmap_addr = 0 - */ - ASSERT(vdev->nregions < VHOST_USER_MAX_RAM_SLOTS - 1); - vdev->regions[vdev->nregions].mmap_addr = 0; - - tap_sock_update_pool(vdev->regions, 0); + ASSERT(vdev->memory.nregions < VHOST_USER_MAX_RAM_SLOTS); It looks like the assertion is changing threshold by one, and I'm not sure why.

It's because previous version was using the last slot to mark the end of the array with a NULL memory region. Now, we have a counter, it's not needed anymore.

Ah, ok. Might be worth mentioning that in the commit message. -- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Tue, Aug 05, 2025 at 05:46:28PM +0200, Laurent Vivier wrote:

The packet pool was previously limited to handling packets contained within a single buffer.

This patch extends the packet pool to support iovec array, allowing a single logical packet to be composed of multiple iovec.

To accommodate this, the storage format within the pool is modified. For a multi-vector packet, a header entry is now stored first with iov_base = NULL and iov_len holding the number of subsequent vectors. The actual data vectors are then stored in the following pool slots.

The packet_add_do() and packet_get_do() functions are updated to manage this new format for storing and retrieving packets. The pool_full() check is also adjusted to ensure there is enough space for all vectors of a new packet before adding it.

Signed-off-by: Laurent Vivier --- packet.c | 50 +++++++++++++++++++++++++++++++++----------------- packet.h | 2 +- tap.c | 4 ++-- 3 files changed, 36 insertions(+), 20 deletions(-)

diff --git a/packet.c b/packet.c index 4b93688509a4..d697232d951a 100644 --- a/packet.c +++ b/packet.c @@ -90,12 +90,13 @@ static int packet_check_range(const struct pool *p, const char *ptr, size_t len, /** * pool_full() - Is a packet pool full? * @p: Pointer to packet pool + * @data: check data can fit in the pool * - * Return: true if the pool is full, false if more packets can be added + * Return: true if the pool is full, false if data can be added */ -bool pool_full(const struct pool *p) +bool pool_full(const struct pool *p, const struct iov_tail *data)

Given the slightly changed semantics, I wonder if 'pool_can_fit()' might be a better name now.

{ - return p->count >= p->size; + return p->count + data->cnt + (data->cnt > 1) >= p->size;

This test is only correct if data is already pruned. As I've said elsewhere, it might be worth changing to the assumption that iov_tails are pruned everywhere outside the iov_tail internal handling. Oh.. also I think the new check is off by one (in the relatively safe direction). It will say there's no room when there is just exactly enough room.

}

/** @@ -108,11 +109,9 @@ bool pool_full(const struct pool *p) void packet_add_do(struct pool *p, struct iov_tail *data, const char *func, int line) { - size_t idx = p->count; - const char *start; - size_t len; + size_t idx = p->count, i, offset;

- if (pool_full(p)) { + if (pool_full(p, data)) { debug("add packet index %zu to pool with size %zu, %s:%i", idx, p->size, func, line); return; @@ -121,18 +120,30 @@ void packet_add_do(struct pool *p, struct iov_tail *data, if (!iov_tail_prune(data)) return;

- ASSERT(data->cnt == 1); /* we don't support iovec */ + if (data->cnt > 1) { + p->pkt[idx].iov_base = NULL; + p->pkt[idx].iov_len = data->cnt; + idx++; + }

- len = data->iov[0].iov_len - data->off; - start = (char *)data->iov[0].iov_base + data->off; + offset = data->off; + for (i = 0; i < data->cnt; i++) { + const char *start; + size_t len;

- if (packet_check_range(p, start, len, func, line)) - return; + len = data->iov[i].iov_len - offset; + start = (char *)data->iov[i].iov_base + offset; + offset = 0;

- p->pkt[idx].iov_base = (void *)start; - p->pkt[idx].iov_len = len; + if (packet_check_range(p, start, len, func, line)) + return;

- p->count++; + p->pkt[idx].iov_base = (void *)start; + p->pkt[idx].iov_len = len; + idx++;

Hm. Isn't the above equivalent to iov_tail_clone()? Is calling packet_check_range() on each chunk the only reason for open-coding it here?

+ } + + p->count = idx; }

/** @@ -162,9 +173,14 @@ bool packet_get_do(const struct pool *p, size_t idx, return false; }

- data->cnt = 1; + if (p->pkt[idx].iov_base) { + data->cnt = 1; + data->iov = &p->pkt[idx]; + } else { + data->cnt = p->pkt[idx].iov_len; + data->iov = &p->pkt[idx + 1]; + } data->off = 0; - data->iov = &p->pkt[idx];

for (i = 0; i < data->cnt; i++) { ASSERT_WITH_MSG(!packet_check_range(p, data->iov[i].iov_base, diff --git a/packet.h b/packet.h index e51cbd19fdc4..67dc7deb17db 100644 --- a/packet.h +++ b/packet.h @@ -37,7 +37,7 @@ void packet_add_do(struct pool *p, struct iov_tail *data, const char *func, int line); bool packet_get_do(const struct pool *p, const size_t idx, struct iov_tail *data, const char *func, int line); -bool pool_full(const struct pool *p); +bool pool_full(const struct pool *p, const struct iov_tail *data); void pool_flush(struct pool *p);

#define packet_add(p, data) \ diff --git a/tap.c b/tap.c index 9fd00915bb01..95688b22fcb7 100644 --- a/tap.c +++ b/tap.c @@ -1103,14 +1103,14 @@ void tap_add_packet(struct ctx *c, struct iov_tail *data, switch (ntohs(eh->h_proto)) { case ETH_P_ARP: case ETH_P_IP: - if (pool_full(pool_tap4)) { + if (pool_full(pool_tap4, data)) { tap4_handler(c, pool_tap4, now); pool_flush(pool_tap4); } packet_add(pool_tap4, data); break; case ETH_P_IPV6: - if (pool_full(pool_tap6)) { + if (pool_full(pool_tap6, data)) { tap6_handler(c, pool_tap6, now); pool_flush(pool_tap6); }

-- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson

On Thu, Aug 14, 2025 at 11:02:57AM +0200, Laurent Vivier wrote:

On 07/08/2025 08:17, David Gibson wrote:

...

On Tue, Aug 05, 2025 at 05:46:28PM +0200, Laurent Vivier wrote:

...

The packet pool was previously limited to handling packets contained within a single buffer.

This patch extends the packet pool to support iovec array, allowing a single logical packet to be composed of multiple iovec.

To accommodate this, the storage format within the pool is modified. For a multi-vector packet, a header entry is now stored first with iov_base = NULL and iov_len holding the number of subsequent vectors. The actual data vectors are then stored in the following pool slots.

The packet_add_do() and packet_get_do() functions are updated to manage this new format for storing and retrieving packets. The pool_full() check is also adjusted to ensure there is enough space for all vectors of a new packet before adding it.

Signed-off-by: Laurent Vivier --- packet.c | 50 +++++++++++++++++++++++++++++++++----------------- packet.h | 2 +- tap.c | 4 ++-- 3 files changed, 36 insertions(+), 20 deletions(-)

diff --git a/packet.c b/packet.c index 4b93688509a4..d697232d951a 100644 --- a/packet.c +++ b/packet.c @@ -90,12 +90,13 @@ static int packet_check_range(const struct pool *p, const char *ptr, size_t len, /** * pool_full() - Is a packet pool full? * @p: Pointer to packet pool + * @data: check data can fit in the pool * - * Return: true if the pool is full, false if more packets can be added + * Return: true if the pool is full, false if data can be added */ -bool pool_full(const struct pool *p) +bool pool_full(const struct pool *p, const struct iov_tail *data)

Given the slightly changed semantics, I wonder if 'pool_can_fit()' might be a better name now.

okay

...

...

{ - return p->count >= p->size; + return p->count + data->cnt + (data->cnt > 1) >= p->size;

This test is only correct if data is already pruned. As I've said elsewhere, it might be worth changing to the assumption that iov_tails are pruned everywhere outside the iov_tail internal handling.

Oh.. also I think the new check is off by one (in the relatively safe direction). It will say there's no room when there is just exactly enough room.

Could you explain why you think it's off by 1?

Loosely, because the old version needed to check if the pool was _already_ full, but the new version needs to check if it _will_ be full with the addition. Suppose there's exactly one free slot in the pool, say p->count==99, p->size==100. We attempt to add a single buffer packet, data->cnt==1. 99 + 1 + 0 >= 100 But it should (just) fit.

...

...

} /** @@ -108,11 +109,9 @@ bool pool_full(const struct pool *p) void packet_add_do(struct pool *p, struct iov_tail *data, const char *func, int line) { - size_t idx = p->count; - const char *start; - size_t len; + size_t idx = p->count, i, offset; - if (pool_full(p)) { + if (pool_full(p, data)) { debug("add packet index %zu to pool with size %zu, %s:%i", idx, p->size, func, line); return; @@ -121,18 +120,30 @@ void packet_add_do(struct pool *p, struct iov_tail *data, if (!iov_tail_prune(data)) return; - ASSERT(data->cnt == 1); /* we don't support iovec */ + if (data->cnt > 1) { + p->pkt[idx].iov_base = NULL; + p->pkt[idx].iov_len = data->cnt; + idx++; + } - len = data->iov[0].iov_len - data->off; - start = (char *)data->iov[0].iov_base + data->off; + offset = data->off; + for (i = 0; i < data->cnt; i++) { + const char *start; + size_t len; - if (packet_check_range(p, start, len, func, line)) - return; + len = data->iov[i].iov_len - offset; + start = (char *)data->iov[i].iov_base + offset; + offset = 0; - p->pkt[idx].iov_base = (void *)start; - p->pkt[idx].iov_len = len; + if (packet_check_range(p, start, len, func, line)) + return; - p->count++; + p->pkt[idx].iov_base = (void *)start; + p->pkt[idx].iov_len = len; + idx++;

Hm. Isn't the above equivalent to iov_tail_clone()? Is calling packet_check_range() on each chunk the only reason for open-coding it here?

Yes, I think the code is clearer like this. And it avoids to scan the iovec array twice (with the offset management).

Ok. -- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson