[PATCH v3 0/6] Refactor epoll handling in preparation for multithreading
This series refactors how epoll file descriptors are managed throughout the codebase in preparation for introducing multithreading support. Currently, passt uses a single global epollfd accessed through the context structure. With multithreading, each thread will need its own epollfd managing its subset of flows. The key changes are: 1. Centralize epoll management by extracting helper functions into a new epoll_ctl.c/h module and moving union epoll_ref from passt.h to its more logical location in epoll_ctl.h. 2. Simplify epoll_del() to take the epollfd directly rather than extracting it from the context structure, reducing coupling between epoll operations and the global context. 3. Move epoll registration out of sock_l4_sa() into protocol-specific code, giving callers explicit control over which epoll instance manages each socket. 4. Replace the boolean in_epoll flag in TCP connections with a threadnb (thread number) field in flow_common. This serves dual purposes: tracking registration status (FLOW_THREADNB_INVALID = not registered) and identifying the owning thread. A thread-to-epollfd mapping allows retrieving the actual epoll file descriptor. The threadnb field is 8 bits, limiting values to 0-254 (255 = FLOW_THREADNB_INVALID). 5. Apply this pattern consistently across all protocol handlers (TCP, ICMP, UDP), storing the managing thread number in each flow's common structure. These changes make thread ownership explicit in the flow tracking system, enabling flows to be managed by different threads (and their epoll instances), a prerequisite for per-thread epollfd design in upcoming multithreading work. Changes since v1: - New patch: "epoll_ctl: Extract epoll operations" - centralizes epoll helpers into a dedicated module and relocates union epoll_ref from passt.h to epoll_ctl.h - Changed epollfd type in flow_common from int to 8-bit bitfield to avoid exceeding cacheline size threshold - Added flow_epollfd_valid() helper to check epoll registration status - Added flow_set_epollfd() helper to set the epoll instance for a flow Changes since v2: - Renamed field in flow_common from epollfd to threadnb (thread number) to better reflect that flows are now associated with threads rather than directly with epoll file descriptors - Introduced thread-to-epollfd mapping via threadnb_to_epollfd[] array in flow.c, providing an indirection layer between threads and their epoll instances - Renamed/refactored helper functions: - flow_set_epollfd() -> flow_epollfd_set() - now takes thread number and epollfd parameters - Added flow_epollfd_get() - retrieves the epoll fd for a flow's thread - flow_epollfd_valid() - now checks threadnb instead of epollfd - Updated constants: replaced EPOLLFD_* with FLOW_THREADNB_* equivalents (e.g., EPOLLFD_INVALID -> FLOW_THREADNB_INVALID) - Applied thread-based pattern consistently across TCP, ICMP, and UDP Laurent Vivier (6): util: Simplify epoll_del() interface to take epollfd directly epoll_ctl: Extract epoll operations util: Move epoll registration out of sock_l4_sa() tcp, flow: Replace per-connection in_epoll flag with threadnb in flow_common icmp: Use thread-based epoll management for ICMP flows udp: Use thread-based epoll management for UDP flows Makefile | 22 +++++++++++----------- epoll_ctl.c | 45 ++++++++++++++++++++++++++++++++++++++++++++ epoll_ctl.h | 51 ++++++++++++++++++++++++++++++++++++++++++++++++++ flow.c | 53 +++++++++++++++++++++++++++++++++++++++++++++------- flow.h | 13 ++++++++++++- icmp.c | 23 +++++++++++++++-------- passt.c | 2 +- passt.h | 34 --------------------------------- pasta.c | 7 +++---- pif.c | 32 ++++++++++++++++++++++++------- repair.c | 14 +++++--------- tap.c | 15 +++++---------- tcp.c | 41 +++++++++++++++++++++------------------- tcp_conn.h | 8 +------- tcp_splice.c | 26 +++++++++++++------------- udp.c | 2 +- udp_flow.c | 23 +++++++++++++++++++---- util.c | 28 ++------------------------- util.h | 6 ++++-- vhost_user.c | 14 +++++--------- vu_common.c | 2 +- 21 files changed, 287 insertions(+), 174 deletions(-) create mode 100644 epoll_ctl.c create mode 100644 epoll_ctl.h -- 2.50.1
Centralize epoll_add() and epoll_del() helper functions into new
epoll_ctl.c/h files.
This also moves the union epoll_ref definition from passt.h to
epoll_ctl.h where it's more logically placed.
The new epoll_add() helper simplifies adding file descriptors to epoll
by taking an epoll_ref and events, handling error reporting
consistently across all call sites.
Signed-off-by: Laurent Vivier
Move epoll_add() calls from sock_l4_sa() to the protocol-specific code
(icmp.c, pif.c, udp_flow.c) to give callers more control over epoll
registration. This allows sock_l4_sa() to focus solely on socket
creation and binding, while epoll management happens at a higher level.
Remove the data parameter from sock_l4_sa() and flowside_sock_l4() as
it's no longer needed - callers now construct the full epoll_ref and
register the socket themselves after creation.
Signed-off-by: Laurent Vivier
Change epoll_del() to accept the epoll file descriptor directly instead
of the full context structure. This simplifies the interface and aligns
with the threading refactoring by reducing dependency on the context
structure for basic epoll operations as we will manage an epollfd per
thread.
Signed-off-by: Laurent Vivier
The in_epoll boolean flag in tcp_tap_conn and tcp_splice_conn only tracked
whether a connection was registered with epoll, not which epoll instance.
This limited flexibility for future multi-epoll support.
Replace the boolean with a threadnb field in flow_common that identifies
which thread (and thus which epoll instance) the flow is registered with.
Use FLOW_THREADNB_INVALID to indicate when a flow is not registered with
any epoll instance. A threadnb_to_epollfd[] mapping table translates
thread numbers to their corresponding epoll file descriptors.
Add helper functions:
- flow_epollfd_valid() to check if a flow is registered with epoll
- flow_epollfd_get() to retrieve the epoll fd for a flow's thread
- flow_epollfd_set() to register a flow with a thread's epoll instance
This change also simplifies tcp_timer_ctl() and conn_flag_do() by removing
the need to pass the context 'c', since the epoll fd is now directly
accessible from the flow structure via flow_epollfd_get().
Signed-off-by: Laurent Vivier
Store the thread number in the flow_common structure for UDP
flows using flow_epollfd_set() and retrieve the corresponding epoll
file descriptor with flow_epollfd_get() instead of passing c->epollfd
directly. This makes UDP consistent with the recent TCP and ICMP
changes.
Signed-off-by: Laurent Vivier
Store the thread number in the flow_common structure for ICMP ping
flows using flow_epollfd_set() and retrieve the corresponding epoll
file descriptor with flow_epollfd_get() instead of passing c->epollfd
directly. This makes ICMP consistent with the recent TCP changes and
follows the pattern established in previous commit.
Signed-off-by: Laurent Vivier
On Thu, Oct 09, 2025 at 03:04:05PM +0200, Laurent Vivier wrote:
Centralize epoll_add() and epoll_del() helper functions into new epoll_ctl.c/h files.
This also moves the union epoll_ref definition from passt.h to epoll_ctl.h where it's more logically placed.
The new epoll_add() helper simplifies adding file descriptors to epoll by taking an epoll_ref and events, handling error reporting consistently across all call sites.
Signed-off-by: Laurent Vivier
Nice cleanup. One nit noted below.
--- Makefile | 22 +++++++++++----------- epoll_ctl.c | 45 +++++++++++++++++++++++++++++++++++++++++++++ epoll_ctl.h | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++ icmp.c | 4 +--- passt.c | 2 +- passt.h | 34 ---------------------------------- pasta.c | 7 +++---- repair.c | 14 +++++--------- tap.c | 13 ++++--------- tcp.c | 2 +- tcp_splice.c | 2 +- udp.c | 2 +- udp_flow.c | 1 + util.c | 22 +++------------------- util.h | 4 +++- vhost_user.c | 8 ++------ vu_common.c | 2 +- 17 files changed, 134 insertions(+), 101 deletions(-) create mode 100644 epoll_ctl.c create mode 100644 epoll_ctl.h
diff --git a/Makefile b/Makefile index 3328f8324140..91e037b8fd3c 100644 --- a/Makefile +++ b/Makefile @@ -37,23 +37,23 @@ FLAGS += -DPAGE_SIZE=$(shell getconf PAGE_SIZE) FLAGS += -DVERSION=\"$(VERSION)\" FLAGS += -DDUAL_STACK_SOCKETS=$(DUAL_STACK_SOCKETS)
-PASST_SRCS = arch.c arp.c checksum.c conf.c dhcp.c dhcpv6.c flow.c fwd.c \ - icmp.c igmp.c inany.c iov.c ip.c isolation.c lineread.c log.c mld.c \ - ndp.c netlink.c migrate.c packet.c passt.c pasta.c pcap.c pif.c \ - repair.c tap.c tcp.c tcp_buf.c tcp_splice.c tcp_vu.c udp.c udp_flow.c \ - udp_vu.c util.c vhost_user.c virtio.c vu_common.c +PASST_SRCS = arch.c arp.c checksum.c conf.c dhcp.c dhcpv6.c epoll_ctl.c \ + flow.c fwd.c icmp.c igmp.c inany.c iov.c ip.c isolation.c lineread.c \ + log.c mld.c ndp.c netlink.c migrate.c packet.c passt.c pasta.c pcap.c \ + pif.c repair.c tap.c tcp.c tcp_buf.c tcp_splice.c tcp_vu.c udp.c \ + udp_flow.c udp_vu.c util.c vhost_user.c virtio.c vu_common.c QRAP_SRCS = qrap.c PASST_REPAIR_SRCS = passt-repair.c SRCS = $(PASST_SRCS) $(QRAP_SRCS) $(PASST_REPAIR_SRCS)
MANPAGES = passt.1 pasta.1 qrap.1 passt-repair.1
-PASST_HEADERS = arch.h arp.h checksum.h conf.h dhcp.h dhcpv6.h flow.h fwd.h \ - flow_table.h icmp.h icmp_flow.h inany.h iov.h ip.h isolation.h \ - lineread.h log.h migrate.h ndp.h netlink.h packet.h passt.h pasta.h \ - pcap.h pif.h repair.h siphash.h tap.h tcp.h tcp_buf.h tcp_conn.h \ - tcp_internal.h tcp_splice.h tcp_vu.h udp.h udp_flow.h udp_internal.h \ - udp_vu.h util.h vhost_user.h virtio.h vu_common.h +PASST_HEADERS = arch.h arp.h checksum.h conf.h dhcp.h dhcpv6.h epoll_ctl.h \ + flow.h fwd.h flow_table.h icmp.h icmp_flow.h inany.h iov.h ip.h \ + isolation.h lineread.h log.h migrate.h ndp.h netlink.h packet.h \ + passt.h pasta.h pcap.h pif.h repair.h siphash.h tap.h tcp.h tcp_buf.h \ + tcp_conn.h tcp_internal.h tcp_splice.h tcp_vu.h udp.h udp_flow.h \ + udp_internal.h udp_vu.h util.h vhost_user.h virtio.h vu_common.h HEADERS = $(PASST_HEADERS) seccomp.h
C := \#include
\nint main(){int a=getrandom(0, 0, 0);} diff --git a/epoll_ctl.c b/epoll_ctl.c new file mode 100644 index 000000000000..0a06350f87a5 --- /dev/null +++ b/epoll_ctl.c @@ -0,0 +1,45 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* epoll_ctl.c - epoll manipulation helpers + * + * Copyright Red Hat + * Author: Laurent Vivier + */ + +#include + +#include "epoll_ctl.h" + +/** + * epoll_add() - Add a file descriptor to an epollfd + * @epollfd: epoll file descriptor to add to + * @events: epoll events + * @ref: epoll reference for the file descriptor (includes fd and metadata) + * + * Return: 0 on success, negative errno on failure + */ +int epoll_add(int epollfd, uint32_t events, union epoll_ref *ref) +{ + struct epoll_event ev; + int ret; + + ev.events = events; + ev.data.u64 = ref->u64; + + ret = epoll_ctl(epollfd, EPOLL_CTL_ADD, ref->fd, &ev); + if (ret == -1) { + ret = -errno; + warn("Failed to add fd to epoll: %s", strerror_(-ret));
I think this should be err() not warn(). If we're unable to add an fd to epoll things are almost certainly going to go badly wrong.
+ } + + return ret; +} + +/** + * epoll_del() - Remove a file descriptor from an epollfd + * @epollfd: epoll file descriptor to remove from + * @fd: File descriptor to remove + */ +void epoll_del(int epollfd, int fd) +{ + epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, NULL); +} diff --git a/epoll_ctl.h b/epoll_ctl.h new file mode 100644 index 000000000000..cf92b0f63f26 --- /dev/null +++ b/epoll_ctl.h @@ -0,0 +1,51 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later + * Copyright Red Hat + * Author: Laurent Vivier
+ */ + +#ifndef EPOLL_CTL_H +#define EPOLL_CTL_H + +#include + +#include "util.h" +#include "passt.h" +#include "epoll_type.h" +#include "flow.h" +#include "tcp.h" +#include "udp.h" + +/** + * union epoll_ref - Breakdown of reference for epoll fd bookkeeping + * @type: Type of fd (tells us what to do with events) + * @fd: File descriptor number (implies < 2^24 total descriptors) + * @flow: Index of the flow this fd is linked to + * @tcp_listen: TCP-specific reference part for listening sockets + * @udp: UDP-specific reference part + * @data: Data handled by protocol handlers + * @nsdir_fd: netns dirfd for fallback timer checking if namespace is gone + * @queue: vhost-user queue index for this fd + * @u64: Opaque reference for epoll_ctl() and epoll_wait() + */ +union epoll_ref { + struct { + enum epoll_type type:8; + int32_t fd:FD_REF_BITS; + union { + uint32_t flow; + flow_sidx_t flowside; + union tcp_listen_epoll_ref tcp_listen; + union udp_listen_epoll_ref udp; + uint32_t data; + int nsdir_fd; + int queue; + }; + }; + uint64_t u64; +}; +static_assert(sizeof(union epoll_ref) <= sizeof(union epoll_data), + "epoll_ref must have same size as epoll_data"); + +int epoll_add(int epollfd, uint32_t events, union epoll_ref *ref); +void epoll_del(int epollfd, int fd); +#endif /* EPOLL_CTL_H */ diff --git a/icmp.c b/icmp.c index bd3108a21675..c26561da80bf 100644 --- a/icmp.c +++ b/icmp.c @@ -15,7 +15,6 @@ #include #include #include -#include #include #include #include @@ -23,10 +22,8 @@ #include #include #include -#include #include #include -#include #include #include
@@ -41,6 +38,7 @@ #include "inany.h" #include "icmp.h" #include "flow_table.h" +#include "epoll_ctl.h" #define ICMP_ECHO_TIMEOUT 60 /* s, timeout for ICMP socket activity */ #define ICMP_NUM_IDS (1U << 16) diff --git a/passt.c b/passt.c index 31fbb75b1b12..e595f13a56f7 100644 --- a/passt.c +++ b/passt.c @@ -19,7 +19,6 @@ * created in a separate network namespace). */
-#include
#include #include #include @@ -53,6 +52,7 @@ #include "vu_common.h" #include "migrate.h" #include "repair.h" +#include "epoll_ctl.h" #define EPOLL_EVENTS 8
diff --git a/passt.h b/passt.h index 0075eb4b3b16..befe56bb167b 100644 --- a/passt.h +++ b/passt.h @@ -35,40 +35,6 @@ union epoll_ref; #define MAC_OUR_LAA \ ((uint8_t [ETH_ALEN]){0x9a, 0x55, 0x9a, 0x55, 0x9a, 0x55})
-/** - * union epoll_ref - Breakdown of reference for epoll fd bookkeeping - * @type: Type of fd (tells us what to do with events) - * @fd: File descriptor number (implies < 2^24 total descriptors) - * @flow: Index of the flow this fd is linked to - * @tcp_listen: TCP-specific reference part for listening sockets - * @udp: UDP-specific reference part - * @icmp: ICMP-specific reference part - * @data: Data handled by protocol handlers - * @nsdir_fd: netns dirfd for fallback timer checking if namespace is gone - * @queue: vhost-user queue index for this fd - * @u64: Opaque reference for epoll_ctl() and epoll_wait() - */ -union epoll_ref { - struct { - enum epoll_type type:8; -#define FD_REF_BITS 24 -#define FD_REF_MAX ((int)MAX_FROM_BITS(FD_REF_BITS)) - int32_t fd:FD_REF_BITS; - union { - uint32_t flow; - flow_sidx_t flowside; - union tcp_listen_epoll_ref tcp_listen; - union udp_listen_epoll_ref udp; - uint32_t data; - int nsdir_fd; - int queue; - }; - }; - uint64_t u64; -}; -static_assert(sizeof(union epoll_ref) <= sizeof(union epoll_data), - "epoll_ref must have same size as epoll_data"); - /* Large enough for ~128 maximum size frames */ #define PKT_BUF_BYTES (8UL << 20)
diff --git a/pasta.c b/pasta.c index 687406b6e736..e905f6d33b95 100644 --- a/pasta.c +++ b/pasta.c @@ -27,7 +27,6 @@ #include
#include #include -#include #include #include #include @@ -49,6 +48,7 @@ #include "isolation.h" #include "netlink.h" #include "log.h" +#include "epoll_ctl.h" #define HOSTNAME_PREFIX "pasta-"
@@ -444,7 +444,6 @@ static int pasta_netns_quit_timer(void) */ void pasta_netns_quit_init(const struct ctx *c) { - struct epoll_event ev = { .events = EPOLLIN }; int flags = O_NONBLOCK | O_CLOEXEC; struct statfs s = { 0 }; bool try_inotify = true; @@ -487,8 +486,8 @@ void pasta_netns_quit_init(const struct ctx *c) die("netns monitor file number %i too big, exiting", fd);
ref.fd = fd; - ev.data.u64 = ref.u64; - epoll_ctl(c->epollfd, EPOLL_CTL_ADD, fd, &ev); + + epoll_add(c->epollfd, EPOLLIN, &ref); }
/** diff --git a/repair.c b/repair.c index f6b1bf36479c..c8f4737fa62a 100644 --- a/repair.c +++ b/repair.c @@ -22,6 +22,7 @@ #include "inany.h" #include "flow.h" #include "flow_table.h" +#include "epoll_ctl.h"
#include "repair.h"
@@ -47,7 +48,6 @@ static int repair_nfds; void repair_sock_init(const struct ctx *c) { union epoll_ref ref = { .type = EPOLL_TYPE_REPAIR_LISTEN }; - struct epoll_event ev = { 0 };
if (c->fd_repair_listen == -1) return; @@ -58,9 +58,7 @@ void repair_sock_init(const struct ctx *c) }
ref.fd = c->fd_repair_listen; - ev.events = EPOLLIN | EPOLLHUP | EPOLLET; - ev.data.u64 = ref.u64; - if (epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_repair_listen, &ev)) + if (epoll_add(c->epollfd, EPOLLIN | EPOLLHUP | EPOLLET, &ref)) err_perror("repair helper socket epoll_ctl(), won't migrate"); }
@@ -74,7 +72,6 @@ void repair_sock_init(const struct ctx *c) int repair_listen_handler(struct ctx *c, uint32_t events) { union epoll_ref ref = { .type = EPOLL_TYPE_REPAIR }; - struct epoll_event ev = { 0 }; struct ucred ucred; socklen_t len; int rc; @@ -112,10 +109,9 @@ int repair_listen_handler(struct ctx *c, uint32_t events) info("Accepted TCP_REPAIR helper, PID %i", ucred.pid);
ref.fd = c->fd_repair; - ev.events = EPOLLHUP | EPOLLET; - ev.data.u64 = ref.u64; - if (epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_repair, &ev)) { - rc = errno; + + rc = epoll_add(c->epollfd, EPOLLHUP | EPOLLET, &ref); + if (rc < 0) { debug_perror("epoll_ctl() on TCP_REPAIR helper socket"); close(c->fd_repair); c->fd_repair = -1; diff --git a/tap.c b/tap.c index 134c37a72979..873874796f79 100644 --- a/tap.c +++ b/tap.c @@ -26,7 +26,6 @@ #include
#include #include -#include #include #include #include @@ -61,6 +60,7 @@ #include "log.h" #include "vhost_user.h" #include "vu_common.h" +#include "epoll_ctl.h" /* Maximum allowed frame lengths (including L2 header) */
@@ -1327,14 +1327,12 @@ static void tap_backend_show_hints(struct ctx *c) static void tap_sock_unix_init(const struct ctx *c) { union epoll_ref ref = { .type = EPOLL_TYPE_TAP_LISTEN }; - struct epoll_event ev = { 0 };
listen(c->fd_tap_listen, 0);
ref.fd = c->fd_tap_listen; - ev.events = EPOLLIN | EPOLLET; - ev.data.u64 = ref.u64; - epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_tap_listen, &ev); + + epoll_add(c->epollfd, EPOLLIN | EPOLLET, &ref); }
/** @@ -1343,7 +1341,6 @@ static void tap_sock_unix_init(const struct ctx *c) */ static void tap_start_connection(const struct ctx *c) { - struct epoll_event ev = { 0 }; union epoll_ref ref = { 0 };
ref.fd = c->fd_tap; @@ -1359,9 +1356,7 @@ static void tap_start_connection(const struct ctx *c) break; }
- ev.events = EPOLLIN | EPOLLRDHUP; - ev.data.u64 = ref.u64; - epoll_ctl(c->epollfd, EPOLL_CTL_ADD, c->fd_tap, &ev); + epoll_add(c->epollfd, EPOLLIN | EPOLLRDHUP, &ref);
if (c->ifi4) arp_send_init_req(c); diff --git a/tcp.c b/tcp.c index 9100c67e57e0..c2d842bbdf4f 100644 --- a/tcp.c +++ b/tcp.c @@ -279,7 +279,6 @@ #include
#include #include -#include #include #include #include @@ -309,6 +308,7 @@ #include "tcp_internal.h" #include "tcp_buf.h" #include "tcp_vu.h" +#include "epoll_ctl.h" /* * The size of TCP header (including options) is given by doff (Data Offset) diff --git a/tcp_splice.c b/tcp_splice.c index 666ee62b738f..6f21184bdc55 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -44,7 +44,6 @@ #include
#include #include -#include #include #include @@ -56,6 +55,7 @@ #include "siphash.h" #include "inany.h" #include "flow.h" +#include "epoll_ctl.h"
#include "flow_table.h"
diff --git a/udp.c b/udp.c index 86585b7e0942..3812d5c2336f 100644 --- a/udp.c +++ b/udp.c @@ -94,7 +94,6 @@ #include
#include #include -#include #include #include #include @@ -115,6 +114,7 @@ #include "flow_table.h" #include "udp_internal.h" #include "udp_vu.h" +#include "epoll_ctl.h" #define UDP_MAX_FRAMES 32 /* max # of frames to receive at once */
diff --git a/udp_flow.c b/udp_flow.c index 84973f807167..d9c75f1bb1d8 100644 --- a/udp_flow.c +++ b/udp_flow.c @@ -15,6 +15,7 @@ #include "passt.h" #include "flow_table.h" #include "udp_internal.h" +#include "epoll_ctl.h"
#define UDP_CONN_TIMEOUT 180 /* s, timeout for ephemeral or local bind */
diff --git a/util.c b/util.c index 88a91b1100f5..b2490123590a 100644 --- a/util.c +++ b/util.c @@ -18,7 +18,6 @@ #include
#include #include -#include #include #include #include @@ -35,6 +34,7 @@ #include "packet.h" #include "log.h" #include "pcap.h" +#include "epoll_ctl.h" #ifdef HAS_GETRANDOM #include #endif @@ -58,7 +58,6 @@ int sock_l4_sa(const struct ctx *c, enum epoll_type type, sa_family_t af = ((const struct sockaddr *)sa)->sa_family; union epoll_ref ref = { .type = type, .data = data }; bool freebind = false; - struct epoll_event ev; int fd, y = 1, ret; uint8_t proto; int socktype; @@ -172,13 +171,9 @@ int sock_l4_sa(const struct ctx *c, enum epoll_type type, return ret; } - ev.events = EPOLLIN; - ev.data.u64 = ref.u64; - if (epoll_ctl(c->epollfd, EPOLL_CTL_ADD, fd, &ev) == -1) { - ret = -errno; - warn("L4 epoll_ctl: %s", strerror_(-ret)); + ret = epoll_add(c->epollfd, EPOLLIN, &ref); + if (ret < 0) return ret; - }
return fd; } @@ -994,17 +989,6 @@ void raw_random(void *buf, size_t buflen) die("Unexpected EOF on random data source"); }
-/** - * epoll_del() - Remove a file descriptor from our passt epoll - * @epollfd: epoll file descriptor to add to - * @fd: File descriptor to remove - */ -void epoll_del(int epollfd, int fd) -{ - epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, NULL); - -} - /** * encode_domain_name() - Encode domain name according to RFC 1035, section 3.1 * @buf: Buffer to fill in with encoded domain name diff --git a/util.h b/util.h index c61cbef357aa..8e4b4c5c6032 100644 --- a/util.h +++ b/util.h @@ -193,6 +193,9 @@ int do_clone(int (*fn)(void *), char *stack_area, size_t stack_size, int flags, #define SNDBUF_BIG (4ULL * 1024 * 1024) #define SNDBUF_SMALL (128ULL * 1024)
+#define FD_REF_BITS 24 +#define FD_REF_MAX ((int)MAX_FROM_BITS(FD_REF_BITS)) + #include
#include #include @@ -300,7 +303,6 @@ static inline bool mod_between(unsigned x, unsigned i, unsigned j, unsigned m) #define FPRINTF(f, ...) (void)fprintf(f, __VA_ARGS__) void raw_random(void *buf, size_t buflen); -void epoll_del(int epollfd, int fd);
/* * Starting from glibc 2.40.9000 and commit 25a5eb4010df ("string: strerror, diff --git a/vhost_user.c b/vhost_user.c index f8324c59cc6c..aea1e2cbcea5 100644 --- a/vhost_user.c +++ b/vhost_user.c @@ -32,8 +32,6 @@ #include
#include #include -#include -#include #include #include #include @@ -45,6 +43,7 @@ #include "vhost_user.h" #include "pcap.h" #include "migrate.h" +#include "epoll_ctl.h" /* vhost-user version we are compatible with */ #define VHOST_USER_VERSION 1 @@ -753,11 +752,8 @@ static void vu_set_watch(const struct vu_dev *vdev, int idx) .fd = vdev->vq[idx].kick_fd, .queue = idx }; - struct epoll_event ev = { 0 };
- ev.data.u64 = ref.u64; - ev.events = EPOLLIN; - epoll_ctl(vdev->context->epollfd, EPOLL_CTL_ADD, ref.fd, &ev); + epoll_add(vdev->context->epollfd, EPOLLIN, &ref); }
/** diff --git a/vu_common.c b/vu_common.c index b716070ea3c3..b13b7c308fd8 100644 --- a/vu_common.c +++ b/vu_common.c @@ -6,7 +6,6 @@ */
#include
-#include #include #include #include @@ -19,6 +18,7 @@ #include "pcap.h" #include "vu_common.h" #include "migrate.h" +#include "epoll_ctl.h" #define VU_MAX_TX_BUFFER_NB 2
-- 2.50.1
-- 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 Thu, Oct 09, 2025 at 03:04:06PM +0200, Laurent Vivier wrote:
Move epoll_add() calls from sock_l4_sa() to the protocol-specific code (icmp.c, pif.c, udp_flow.c) to give callers more control over epoll registration. This allows sock_l4_sa() to focus solely on socket creation and binding, while epoll management happens at a higher level.
Remove the data parameter from sock_l4_sa() and flowside_sock_l4() as it's no longer needed - callers now construct the full epoll_ref and register the socket themselves after creation.
Signed-off-by: Laurent Vivier
Reviewed-by: David Gibson
--- flow.c | 10 ++++------ flow.h | 2 +- icmp.c | 15 +++++++++++---- pif.c | 32 +++++++++++++++++++++++++------- udp_flow.c | 16 ++++++++++++++-- util.c | 10 +--------- util.h | 2 +- 7 files changed, 57 insertions(+), 30 deletions(-)
diff --git a/flow.c b/flow.c index feefda3ce74e..b14e9d8b63ff 100644 --- a/flow.c +++ b/flow.c @@ -163,7 +163,6 @@ static void flowside_from_af(struct flowside *side, sa_family_t af, * @type: Socket epoll type * @sa: Socket address * @sl: Length of @sa - * @data: epoll reference data */ struct flowside_sock_args { const struct ctx *c; @@ -173,7 +172,6 @@ struct flowside_sock_args { const struct sockaddr *sa; socklen_t sl; const char *path; - uint32_t data; };
/** flowside_sock_splice() - Create and bind socket for PIF_SPLICE based on flowside @@ -188,7 +186,7 @@ static int flowside_sock_splice(void *arg) ns_enter(a->c);
a->fd = sock_l4_sa(a->c, a->type, a->sa, a->sl, NULL, - a->sa->sa_family == AF_INET6, a->data); + a->sa->sa_family == AF_INET6); a->err = errno;
return 0; @@ -205,7 +203,7 @@ static int flowside_sock_splice(void *arg) * (if specified). */ int flowside_sock_l4(const struct ctx *c, enum epoll_type type, uint8_t pif, - const struct flowside *tgt, uint32_t data) + const struct flowside *tgt) { const char *ifname = NULL; union sockaddr_inany sa; @@ -225,12 +223,12 @@ int flowside_sock_l4(const struct ctx *c, enum epoll_type type, uint8_t pif, ifname = c->ip6.ifname_out;
return sock_l4_sa(c, type, &sa, sl, ifname, - sa.sa_family == AF_INET6, data); + sa.sa_family == AF_INET6);
case PIF_SPLICE: { struct flowside_sock_args args = { .c = c, .type = type, - .sa = &sa.sa, .sl = sl, .data = data, + .sa = &sa.sa, .sl = sl, }; NS_CALL(flowside_sock_splice, &args); errno = args.err; diff --git a/flow.h b/flow.h index cac618ad0ca1..ef138b83add8 100644 --- a/flow.h +++ b/flow.h @@ -167,7 +167,7 @@ static inline bool flowside_eq(const struct flowside *left, }
int flowside_sock_l4(const struct ctx *c, enum epoll_type type, uint8_t pif, - const struct flowside *tgt, uint32_t data); + const struct flowside *tgt); int flowside_connect(const struct ctx *c, int s, uint8_t pif, const struct flowside *tgt);
diff --git a/icmp.c b/icmp.c index c26561da80bf..56dfac6c958e 100644 --- a/icmp.c +++ b/icmp.c @@ -170,10 +170,10 @@ static struct icmp_ping_flow *icmp_ping_new(const struct ctx *c, { uint8_t proto = af == AF_INET ? IPPROTO_ICMP : IPPROTO_ICMPV6; uint8_t flowtype = af == AF_INET ? FLOW_PING4 : FLOW_PING6; - union epoll_ref ref = { .type = EPOLL_TYPE_PING }; union flow *flow = flow_alloc(); struct icmp_ping_flow *pingf; const struct flowside *tgt; + union epoll_ref ref;
if (!flow) return NULL; @@ -194,9 +194,7 @@ static struct icmp_ping_flow *icmp_ping_new(const struct ctx *c,
pingf->seq = -1;
- ref.flowside = FLOW_SIDX(flow, TGTSIDE); - pingf->sock = flowside_sock_l4(c, EPOLL_TYPE_PING, PIF_HOST, - tgt, ref.data); + pingf->sock = flowside_sock_l4(c, EPOLL_TYPE_PING, PIF_HOST, tgt);
if (pingf->sock < 0) { warn("Cannot open \"ping\" socket. You might need to:"); @@ -208,6 +206,15 @@ static struct icmp_ping_flow *icmp_ping_new(const struct ctx *c, if (pingf->sock > FD_REF_MAX) goto cancel;
+ ref.type = EPOLL_TYPE_PING; + ref.flowside = FLOW_SIDX(flow, TGTSIDE); + ref.fd = pingf->sock; + + if (epoll_add(c->epollfd, EPOLLIN, &ref) < 0) { + close(pingf->sock); + goto cancel; + } + flow_dbg(pingf, "new socket %i for echo ID %"PRIu16, pingf->sock, id);
flow_hash_insert(c, FLOW_SIDX(pingf, INISIDE)); diff --git a/pif.c b/pif.c index 592fafaab58a..17b41705183c 100644 --- a/pif.c +++ b/pif.c @@ -5,6 +5,7 @@ * Passt/pasta interface types and IDs */
+#include
#include #include #include @@ -14,7 +15,7 @@ #include "siphash.h" #include "ip.h" #include "inany.h" -#include "passt.h" +#include "epoll_ctl.h" const char *pif_type_str[] = { [PIF_NONE] = "<none>", @@ -83,7 +84,9 @@ int pif_sock_l4(const struct ctx *c, enum epoll_type type, uint8_t pif, .sa6.sin6_addr = in6addr_any, .sa6.sin6_port = htons(port), }; + union epoll_ref ref; socklen_t sl; + int ret;
ASSERT(pif_is_socket(pif));
@@ -93,11 +96,26 @@ int pif_sock_l4(const struct ctx *c, enum epoll_type type, uint8_t pif, ASSERT(addr && inany_is_loopback(addr)); }
- if (!addr) - return sock_l4_sa(c, type, &sa, sizeof(sa.sa6), - ifname, false, data); + if (!addr) { + ref.fd = sock_l4_sa(c, type, &sa, sizeof(sa.sa6), + ifname, false); + } else { + pif_sockaddr(c, &sa, &sl, pif, addr, port); + ref.fd = sock_l4_sa(c, type, &sa, sl, + ifname, sa.sa_family == AF_INET6); + } + + if (ref.fd < 0) + return ref.fd; + + ref.type = type; + ref.data = data; + + ret = epoll_add(c->epollfd, EPOLLIN, &ref); + if (ret < 0) { + close(ref.fd); + return ret; + }
- pif_sockaddr(c, &sa, &sl, pif, addr, port); - return sock_l4_sa(c, type, &sa, sl, - ifname, sa.sa_family == AF_INET6, data); + return ref.fd; } diff --git a/udp_flow.c b/udp_flow.c index d9c75f1bb1d8..caaf3dababb1 100644 --- a/udp_flow.c +++ b/udp_flow.c @@ -78,16 +78,28 @@ static int udp_flow_sock(const struct ctx *c, flow_sidx_t sidx; uint32_t data; } fref = { .sidx = FLOW_SIDX(uflow, sidei) }; + union epoll_ref ref; + int rc; int s;
- s = flowside_sock_l4(c, EPOLL_TYPE_UDP, pif, side, fref.data); + s = flowside_sock_l4(c, EPOLL_TYPE_UDP, pif, side); if (s < 0) { flow_dbg_perror(uflow, "Couldn't open flow specific socket"); return s; }
+ ref.type = EPOLL_TYPE_UDP; + ref.data = fref.data; + ref.fd = s; + + rc = epoll_add(c->epollfd, EPOLLIN, &ref); + if (rc < 0) { + close(s); + return rc; + } + if (flowside_connect(c, s, pif, side) < 0) { - int rc = -errno; + rc = -errno;
epoll_del(c->epollfd, s); close(s); diff --git a/util.c b/util.c index b2490123590a..707ab3388d13 100644 --- a/util.c +++ b/util.c @@ -47,16 +47,14 @@ * @sl: Length of @sa * @ifname: Interface for binding, NULL for any * @v6only: Set IPV6_V6ONLY socket option - * @data: epoll reference portion for protocol handlers * * Return: newly created socket, negative error code on failure */ int sock_l4_sa(const struct ctx *c, enum epoll_type type, const void *sa, socklen_t sl, - const char *ifname, bool v6only, uint32_t data) + const char *ifname, bool v6only) { sa_family_t af = ((const struct sockaddr *)sa)->sa_family; - union epoll_ref ref = { .type = type, .data = data }; bool freebind = false; int fd, y = 1, ret; uint8_t proto; @@ -99,8 +97,6 @@ int sock_l4_sa(const struct ctx *c, enum epoll_type type, return -EBADF; }
- ref.fd = fd; - if (v6only) if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &y, sizeof(y))) debug("Failed to set IPV6_V6ONLY on socket %i", fd); @@ -171,10 +167,6 @@ int sock_l4_sa(const struct ctx *c, enum epoll_type type, return ret; }
- ret = epoll_add(c->epollfd, EPOLLIN, &ref); - if (ret < 0) - return ret; - return fd; }
diff --git a/util.h b/util.h index 8e4b4c5c6032..9108f8f19e14 100644 --- a/util.h +++ b/util.h @@ -207,7 +207,7 @@ struct ctx;
int sock_l4_sa(const struct ctx *c, enum epoll_type type, const void *sa, socklen_t sl, - const char *ifname, bool v6only, uint32_t data); + const char *ifname, bool v6only); int sock_unix(char *sock_path); void sock_probe_mem(struct ctx *c); long timespec_diff_ms(const struct timespec *a, const struct timespec *b); -- 2.50.1
-- 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
participants (2)
-
David Gibson
-
Laurent Vivier