Date: Thu, 25 Jul 2024 13:41:46 GMT From: Kristof Provost <kp@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-branches@FreeBSD.org Subject: git: 428b8c820265 - stable/13 - pf: vnet-ify pf_hashsize, pf_hashmask, pf_srchashsize and V_pf_srchashmask Message-ID: <202407251341.46PDfkW6044292@gitrepo.freebsd.org>
next in thread | raw e-mail | index | archive | help
The branch stable/13 has been updated by kp: URL: https://cgit.FreeBSD.org/src/commit/?id=428b8c8202654bddbfe871eb412e272c23f86802 commit 428b8c8202654bddbfe871eb412e272c23f86802 Author: Kristof Provost <kp@FreeBSD.org> AuthorDate: 2024-07-17 13:52:13 +0000 Commit: Kristof Provost <kp@FreeBSD.org> CommitDate: 2024-07-25 08:19:55 +0000 pf: vnet-ify pf_hashsize, pf_hashmask, pf_srchashsize and V_pf_srchashmask These variables are tunables, so in principle they never change at runtime. That would mean they don't need to be tracked per-vnet. However, they both can be decreased (back to their default values) if the memory allocations for their respective tables fail, and these allocations are per-vnet. That is, it's possible for a few vnets to be started and have the tuned size for the hash and srchash tables only to have later vnets fail the initial allocation and fall back to smaller allocations. That would confuse the previously created vnets (because their actual table size and size/mask variables would no longer match). Avoid this by turning these into per-vnet variables. MFC after: 1 week Sponsored by: Rubicon Communications, LLC ("Netgate") (cherry picked from commit 271f146955641857c93705b5b1916b8004e5623c) --- sys/net/pfvar.h | 8 +++-- sys/netpfil/pf/if_pfsync.c | 12 +++++--- sys/netpfil/pf/pf.c | 74 ++++++++++++++++++++++++---------------------- sys/netpfil/pf/pf_ioctl.c | 22 +++++++------- 4 files changed, 62 insertions(+), 54 deletions(-) diff --git a/sys/net/pfvar.h b/sys/net/pfvar.h index 11bde6224eb3..5d72dabfe305 100644 --- a/sys/net/pfvar.h +++ b/sys/net/pfvar.h @@ -1851,8 +1851,10 @@ struct pf_idhash { }; extern u_long pf_ioctl_maxcount; -extern u_long pf_hashmask; -extern u_long pf_srchashmask; +VNET_DECLARE(u_long, pf_hashmask); +#define V_pf_hashmask VNET(pf_hashmask) +VNET_DECLARE(u_long, pf_srchashmask); +#define V_pf_srchashmask VNET(pf_srchashmask) #define PF_HASHSIZ (131072) #define PF_SRCHASHSIZ (PF_HASHSIZ/4) VNET_DECLARE(struct pf_keyhash *, pf_keyhash); @@ -1862,7 +1864,7 @@ VNET_DECLARE(struct pf_idhash *, pf_idhash); VNET_DECLARE(struct pf_srchash *, pf_srchash); #define V_pf_srchash VNET(pf_srchash) -#define PF_IDHASH(s) (be64toh((s)->id) % (pf_hashmask + 1)) +#define PF_IDHASH(s) (be64toh((s)->id) % (V_pf_hashmask + 1)) VNET_DECLARE(void *, pf_swi_cookie); #define V_pf_swi_cookie VNET(pf_swi_cookie) diff --git a/sys/netpfil/pf/if_pfsync.c b/sys/netpfil/pf/if_pfsync.c index b8632e1fd61b..6ca6db00e685 100644 --- a/sys/netpfil/pf/if_pfsync.c +++ b/sys/netpfil/pf/if_pfsync.c @@ -764,7 +764,7 @@ pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) pfi_kkif_find(clr[i].ifname) == NULL) continue; - for (int i = 0; i <= pf_hashmask; i++) { + for (int i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_kstate *s; relock: @@ -1847,7 +1847,11 @@ pfsync_defer_tmo(void *arg) struct pfsync_softc *sc = pd->pd_sc; struct mbuf *m = pd->pd_m; struct pf_kstate *st = pd->pd_st; - struct pfsync_bucket *b = pfsync_get_bucket(sc, st); + struct pfsync_bucket *b; + + CURVNET_SET(sc->sc_ifp->if_vnet); + + b = pfsync_get_bucket(sc, st); PFSYNC_BUCKET_LOCK_ASSERT(b); @@ -1860,11 +1864,11 @@ pfsync_defer_tmo(void *arg) if (sc->sc_sync_if == NULL) { pf_release_state(st); m_freem(m); + CURVNET_RESTORE(); return; } NET_EPOCH_ENTER(et); - CURVNET_SET(sc->sc_sync_if->if_vnet); pfsync_tx(sc, m); @@ -2211,7 +2215,7 @@ pfsync_bulk_update(void *arg) else i = sc->sc_bulk_hashid; - for (; i <= pf_hashmask; i++) { + for (; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; if (s != NULL) diff --git a/sys/netpfil/pf/pf.c b/sys/netpfil/pf/pf.c index e99886a2d120..0c3df612e3f8 100644 --- a/sys/netpfil/pf/pf.c +++ b/sys/netpfil/pf/pf.c @@ -431,16 +431,18 @@ VNET_DEFINE(struct pf_srchash *, pf_srchash); SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "pf(4)"); -u_long pf_hashmask; -u_long pf_srchashmask; -static u_long pf_hashsize; -static u_long pf_srchashsize; +VNET_DEFINE(u_long, pf_hashmask); +VNET_DEFINE(u_long, pf_srchashmask); +VNET_DEFINE_STATIC(u_long, pf_hashsize); +#define V_pf_hashsize VNET(pf_hashsize) +VNET_DEFINE_STATIC(u_long, pf_srchashsize); +#define V_pf_srchashsize VNET(pf_srchashsize) u_long pf_ioctl_maxcount = 65535; -SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, - &pf_hashsize, 0, "Size of pf(4) states hashtable"); -SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, - &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); +SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, + &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable"); +SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, + &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable"); SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); @@ -607,7 +609,7 @@ pf_hashkey(struct pf_state_key *sk) sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), V_pf_hashseed); - return (h & pf_hashmask); + return (h & V_pf_hashmask); } static __inline uint32_t @@ -628,7 +630,7 @@ pf_hashsrc(struct pf_addr *addr, sa_family_t af) panic("%s: unknown address family %u", __func__, af); } - return (h & pf_srchashmask); + return (h & V_pf_srchashmask); } #ifdef ALTQ @@ -829,7 +831,7 @@ pf_overload_task(void *v, int pending) return; } - for (int i = 0; i <= pf_hashmask; i++) { + for (int i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_state_key *sk; struct pf_kstate *s; @@ -1009,10 +1011,10 @@ pf_initialize(void) struct pf_srchash *sh; u_int i; - if (pf_hashsize == 0 || !powerof2(pf_hashsize)) - pf_hashsize = PF_HASHSIZ; - if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) - pf_srchashsize = PF_SRCHASHSIZ; + if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize)) + V_pf_hashsize = PF_HASHSIZ; + if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize)) + V_pf_srchashsize = PF_SRCHASHSIZ; V_pf_hashseed = arc4random(); @@ -1027,26 +1029,26 @@ pf_initialize(void) sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); - V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), + V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), M_PFHASH, M_NOWAIT | M_ZERO); - V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), + V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), M_PFHASH, M_NOWAIT | M_ZERO); if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { printf("pf: Unable to allocate memory for " - "state_hashsize %lu.\n", pf_hashsize); + "state_hashsize %lu.\n", V_pf_hashsize); free(V_pf_keyhash, M_PFHASH); free(V_pf_idhash, M_PFHASH); - pf_hashsize = PF_HASHSIZ; - V_pf_keyhash = mallocarray(pf_hashsize, + V_pf_hashsize = PF_HASHSIZ; + V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); - V_pf_idhash = mallocarray(pf_hashsize, + V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); } - pf_hashmask = pf_hashsize - 1; - for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; + V_pf_hashmask = V_pf_hashsize - 1; + for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; i++, kh++, ih++) { mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); @@ -1060,19 +1062,19 @@ pf_initialize(void) uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); - V_pf_srchash = mallocarray(pf_srchashsize, + V_pf_srchash = mallocarray(V_pf_srchashsize, sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); if (V_pf_srchash == NULL) { printf("pf: Unable to allocate memory for " - "source_hashsize %lu.\n", pf_srchashsize); + "source_hashsize %lu.\n", V_pf_srchashsize); - pf_srchashsize = PF_SRCHASHSIZ; - V_pf_srchash = mallocarray(pf_srchashsize, + V_pf_srchashsize = PF_SRCHASHSIZ; + V_pf_srchash = mallocarray(V_pf_srchashsize, sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); } - pf_srchashmask = pf_srchashsize - 1; - for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) + V_pf_srchashmask = V_pf_srchashsize - 1; + for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); /* ALTQ */ @@ -1111,7 +1113,7 @@ pf_cleanup(void) struct pf_send_entry *pfse, *next; u_int i; - for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; + for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask; i++, kh++, ih++) { KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", __func__)); @@ -1123,7 +1125,7 @@ pf_cleanup(void) free(V_pf_keyhash, M_PFHASH); free(V_pf_idhash, M_PFHASH); - for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { + for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { KASSERT(LIST_EMPTY(&sh->nodes), ("%s: source node hash not empty", __func__)); mtx_destroy(&sh->lock); @@ -1519,7 +1521,7 @@ pf_find_state_byid(uint64_t id, uint32_t creatorid) pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); - ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; + ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) @@ -1901,7 +1903,7 @@ pf_purge_thread(void *unused __unused) * table every run. */ V_pf_purge_idx = - pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / + pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask / (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); /* @@ -1945,7 +1947,7 @@ pf_unload_vnet_purge(void) /* * Now purge everything. */ - pf_purge_expired_states(0, pf_hashmask); + pf_purge_expired_states(0, V_pf_hashmask); pf_purge_fragments(UINT_MAX); pf_purge_expired_src_nodes(); @@ -2005,7 +2007,7 @@ pf_purge_expired_src_nodes(void) int i; LIST_INIT(&freelist); - for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { + for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) if (cur->states == 0 && cur->expire <= time_uptime) { @@ -2176,7 +2178,7 @@ relock: SDT_PROBE2(pf, purge, state, rowcount, i, count); /* Return when we hit end of hash. */ - if (++i > pf_hashmask) { + if (++i > V_pf_hashmask) { V_pf_status.states = uma_zone_get_cur(V_pf_state_z); return (0); } diff --git a/sys/netpfil/pf/pf_ioctl.c b/sys/netpfil/pf/pf_ioctl.c index a7f9f2604068..2ddb84642b7d 100644 --- a/sys/netpfil/pf/pf_ioctl.c +++ b/sys/netpfil/pf/pf_ioctl.c @@ -2961,7 +2961,7 @@ DIOCCHANGERULE_error: sizeof(struct pfsync_state), M_TEMP, M_WAITOK | M_ZERO); nr = 0; - for (i = 0; i <= pf_hashmask; i++) { + for (i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; DIOCGETSTATES_retry: @@ -3039,7 +3039,7 @@ DIOCGETSTATES_full: sizeof(struct pf_state_export), M_TEMP, M_WAITOK | M_ZERO); nr = 0; - for (i = 0; i <= pf_hashmask; i++) { + for (i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; DIOCGETSTATESV2_retry: @@ -4637,7 +4637,7 @@ DIOCCHANGEADDR_error: struct pf_src_node *p, *pstore; uint32_t i, nr = 0; - for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; + for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) @@ -4656,7 +4656,7 @@ DIOCCHANGEADDR_error: nr = 0; p = pstore = malloc(psn->psn_len, M_TEMP, M_WAITOK | M_ZERO); - for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; + for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) { @@ -5068,7 +5068,7 @@ pf_clear_all_states(void) struct pf_kstate *s; u_int i; - for (i = 0; i <= pf_hashmask; i++) { + for (i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock: PF_HASHROW_LOCK(ih); @@ -5104,7 +5104,7 @@ pf_clear_srcnodes(struct pf_ksrc_node *n) struct pf_kstate *s; int i; - for (i = 0; i <= pf_hashmask; i++) { + for (i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; PF_HASHROW_LOCK(ih); @@ -5120,7 +5120,7 @@ pf_clear_srcnodes(struct pf_ksrc_node *n) if (n == NULL) { struct pf_srchash *sh; - for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; + for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) { @@ -5142,7 +5142,7 @@ pf_kill_srcnodes(struct pfioc_src_node_kill *psnk) struct pf_ksrc_node_list kill; LIST_INIT(&kill); - for (int i = 0; i <= pf_srchashmask; i++) { + for (int i = 0; i <= V_pf_srchashmask; i++) { struct pf_srchash *sh = &V_pf_srchash[i]; struct pf_ksrc_node *sn, *tmp; @@ -5163,7 +5163,7 @@ pf_kill_srcnodes(struct pfioc_src_node_kill *psnk) PF_HASHROW_UNLOCK(sh); } - for (int i = 0; i <= pf_hashmask; i++) { + for (int i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_kstate *s; @@ -5224,7 +5224,7 @@ pf_clear_states(const struct pf_kstate_kill *kill) int idx; unsigned int killed = 0, dir; - for (unsigned int i = 0; i <= pf_hashmask; i++) { + for (unsigned int i = 0; i <= V_pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock_DIOCCLRSTATES: @@ -5298,7 +5298,7 @@ pf_killstates(struct pf_kstate_kill *kill, unsigned int *killed) return; } - for (unsigned int i = 0; i <= pf_hashmask; i++) + for (unsigned int i = 0; i <= V_pf_hashmask; i++) *killed += pf_killstates_row(kill, &V_pf_idhash[i]); return;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202407251341.46PDfkW6044292>