Date: Fri, 09 Jan 2026 14:16:19 +0000 From: Brooks Davis <brooks@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org Subject: git: 1b425afa8d5f - main - cache: avoid hardcoded CACHE_PATH_CUTOFF Message-ID: <69610db3.d405.630e1315@gitrepo.freebsd.org>
index | next in thread | raw e-mail
The branch main has been updated by brooks: URL: https://cgit.FreeBSD.org/src/commit/?id=1b425afa8d5fb611b17f951e5099c177fd97ffea commit 1b425afa8d5fb611b17f951e5099c177fd97ffea Author: Brooks Davis <brooks@FreeBSD.org> AuthorDate: 2026-01-09 14:15:43 +0000 Commit: Brooks Davis <brooks@FreeBSD.org> CommitDate: 2026-01-09 14:15:43 +0000 cache: avoid hardcoded CACHE_PATH_CUTOFF Compute the cutoff at compile time instead which will avoid the need for a CHERI case. Correct the comment about CACHE_PATH_CUTOFF. It dates to 5.2.0 not 4.4 BSD. Historic values are: 32 - introduced (c2935410f6d5f) 35 - NUL terminated and bumped (5d5c174869104) 39 - sized to alignment (bb48255cf5eff) 45 - bumped to improve efficency on 64-bit (3862838921eb8) No functional change. Reviewed by: markj Suggested by: jhb Effort: CHERI upstreaming Sponsored by: DARPA, AFRL Differential Revision: https://reviews.freebsd.org/D54554 --- sys/kern/vfs_cache.c | 32 ++++++++++++++++++-------------- 1 file changed, 18 insertions(+), 14 deletions(-) diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c index 404234861710..3f8591bd0ba7 100644 --- a/sys/kern/vfs_cache.c +++ b/sys/kern/vfs_cache.c @@ -406,31 +406,35 @@ TAILQ_HEAD(cache_freebatch, namecache); #define CACHE_ZONE_ALIGN_MASK UMA_ALIGNOF(struct namecache_ts) /* - * TODO: the initial value of CACHE_PATH_CUTOFF was inherited from the - * 4.4 BSD codebase. Later on struct namecache was tweaked to become - * smaller and the value was bumped to retain the total size, but it - * was never re-evaluated for suitability. A simple test counting - * lengths during package building shows that the value of 45 covers - * about 86% of all added entries, reaching 99% at 65. + * TODO: CACHE_PATH_CUTOFF was initially introduced with an arbitrary + * value of 32 in FreeBSD 5.2.0. It was bumped to 35 and the path was + * NUL terminated with the introduction of DTrace probes. Later, it was + * expanded to match the alignment allowing an increase to 39, but it + * was not re-evaluated for suitability. It was again bumped to 45 on + * 64-bit systems and 41 on 32-bit systems (the current values, now + * computed at compile time rather than hardcoded). A simple test + * counting lengths during package building in 2020 showed that the + * value of 45 covers about 86% of all added entries, reaching 99% + * at 65. * * Regardless of the above, use of dedicated zones instead of malloc may be * inducing additional waste. This may be hard to address as said zones are * tied to VFS SMR. Even if retaining them, the current split should be * re-evaluated. */ -#ifdef __LP64__ -#define CACHE_PATH_CUTOFF 45 -#else -#define CACHE_PATH_CUTOFF 41 -#endif +#define CACHE_PATH_CUTOFF_MIN 40 +#define CACHE_STRUCT_LEN(pathlen) \ + (offsetof(struct namecache, nc_name) + (pathlen) + 1) +#define CACHE_PATH_CUTOFF \ + (roundup2(CACHE_STRUCT_LEN(CACHE_PATH_CUTOFF_MIN), \ + _Alignof(struct namecache_ts)) - CACHE_STRUCT_LEN(0)) #define CACHE_ZONE_SMALL_SIZE \ - (offsetof(struct namecache, nc_name) + CACHE_PATH_CUTOFF + 1) + CACHE_STRUCT_LEN(CACHE_PATH_CUTOFF) #define CACHE_ZONE_SMALL_TS_SIZE \ (offsetof(struct namecache_ts, nc_nc) + CACHE_ZONE_SMALL_SIZE) #define CACHE_ZONE_LARGE_SIZE \ - roundup2(offsetof(struct namecache, nc_name) + NAME_MAX + 1, \ - _Alignof(struct namecache_ts)) + roundup2(CACHE_STRUCT_LEN(NAME_MAX), _Alignof(struct namecache_ts)) #define CACHE_ZONE_LARGE_TS_SIZE \ (offsetof(struct namecache_ts, nc_nc) + CACHE_ZONE_LARGE_SIZE)home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?69610db3.d405.630e1315>