Date: Thu, 6 Nov 2025 16:03:01 GMT From: Mark Johnston <markj@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org Subject: git: 4d6801a6b5bd - main - stand: Teach the zfs loader about dynamic gang headers Message-ID: <202511061603.5A6G311Z000989@gitrepo.freebsd.org>
index | next in thread | raw e-mail
The branch main has been updated by markj: URL: https://cgit.FreeBSD.org/src/commit/?id=4d6801a6b5bdd4d055a00484a743cb4ada659669 commit 4d6801a6b5bdd4d055a00484a743cb4ada659669 Author: Mark Johnston <markj@FreeBSD.org> AuthorDate: 2025-11-06 16:00:50 +0000 Commit: Mark Johnston <markj@FreeBSD.org> CommitDate: 2025-11-06 16:02:33 +0000 stand: Teach the zfs loader about dynamic gang headers There is a pool feature, dynamic_gang_header, that is enabled by default in new pools. When this feature is active, gang headers may be larger than 512 bytes. The loader needs to be taught to cope with that. Try using the vdev ashift to pick the gang block header size. If the checksum fails, fall back to the old gang block header size. This is based on a patch by Paul Dagnelie, with testing, bug-fixing and some simplifications from me. PR: 289690 Co-authored by: Paul Dagnelie <paul.dagnelie@klarasystems.com> Reviewed by: imp MFC after: 1 week Differential Revision: https://reviews.freebsd.org/D53578 --- stand/libsa/zfs/zfsimpl.c | 67 ++++++++++++++++++++++++++++++++++++--------- sys/cddl/boot/zfs/zfsimpl.h | 15 ++-------- 2 files changed, 56 insertions(+), 26 deletions(-) diff --git a/stand/libsa/zfs/zfsimpl.c b/stand/libsa/zfs/zfsimpl.c index f15d9b016068..e5920004bd9d 100644 --- a/stand/libsa/zfs/zfsimpl.c +++ b/stand/libsa/zfs/zfsimpl.c @@ -128,6 +128,7 @@ static const char *features_for_read[] = { "org.open-zfs:large_blocks", "org.openzfs:blake3", "org.zfsonlinux:large_dnode", + "com.klarasystems:dynamic_gang_header", NULL }; @@ -141,6 +142,8 @@ static uint64_t dnode_cache_bn; static char *dnode_cache_buf; static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf); +static int zio_read_impl(const spa_t *spa, const blkptr_t *bp, void *buf, + bool print); static int zfs_get_root(const spa_t *spa, uint64_t *objid); static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result); static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, @@ -530,7 +533,7 @@ vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset, } static vdev_t * -vdev_lookup_top(spa_t *spa, uint64_t vdev) +vdev_lookup_top(const spa_t *spa, uint64_t vdev) { vdev_t *rvd; vdev_list_t *vlist; @@ -2270,45 +2273,77 @@ ilog2(int n) return (-1); } +static inline uint64_t +gbh_nblkptrs(uint64_t size) +{ + ASSERT(IS_P2ALIGNED(size, sizeof(blkptr_t))); + return ((size - sizeof(zio_eck_t)) / sizeof(blkptr_t)); +} + static int zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf) { blkptr_t gbh_bp; - zio_gbh_phys_t zio_gb; + void *gbuf; char *pbuf; - int i; + uint64_t gangblocksize; + int err, i; + + gangblocksize = UINT64_MAX; + for (int dva = 0; dva < BP_GET_NDVAS(bp); dva++) { + vdev_t *vd = vdev_lookup_top(spa, + DVA_GET_VDEV(&bp->blk_dva[dva])); + gangblocksize = MIN(gangblocksize, 1ULL << vd->v_ashift); + } /* Artificial BP for gang block header. */ gbh_bp = *bp; - BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE); - BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE); + BP_SET_PSIZE(&gbh_bp, gangblocksize); + BP_SET_LSIZE(&gbh_bp, gangblocksize); BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER); BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF); for (i = 0; i < SPA_DVAS_PER_BP; i++) DVA_SET_GANG(&gbh_bp.blk_dva[i], 0); + gbuf = malloc(gangblocksize); + if (gbuf == NULL) + return (ENOMEM); /* Read gang header block using the artificial BP. */ - if (zio_read(spa, &gbh_bp, &zio_gb)) + err = zio_read_impl(spa, &gbh_bp, gbuf, false); + if ((err == EIO || err == ECKSUM) && + gangblocksize > SPA_OLD_GANGBLOCKSIZE) { + /* This might be a legacy gang block header, try again. */ + gangblocksize = SPA_OLD_GANGBLOCKSIZE; + BP_SET_PSIZE(&gbh_bp, gangblocksize); + BP_SET_LSIZE(&gbh_bp, gangblocksize); + err = zio_read(spa, &gbh_bp, gbuf); + } + if (err != 0) { + free(gbuf); return (EIO); + } pbuf = buf; - for (i = 0; i < SPA_GBH_NBLKPTRS; i++) { - blkptr_t *gbp = &zio_gb.zg_blkptr[i]; + for (i = 0; i < gbh_nblkptrs(gangblocksize); i++) { + blkptr_t *gbp = &((blkptr_t *)gbuf)[i]; if (BP_IS_HOLE(gbp)) continue; - if (zio_read(spa, gbp, pbuf)) + if (zio_read(spa, gbp, pbuf)) { + free(gbuf); return (EIO); + } pbuf += BP_GET_PSIZE(gbp); } + free(gbuf); if (zio_checksum_verify(spa, bp, buf)) return (EIO); return (0); } static int -zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) +zio_read_impl(const spa_t *spa, const blkptr_t *bp, void *buf, bool print) { int cpfunc = BP_GET_COMPRESS(bp); uint64_t align, size; @@ -2340,7 +2375,7 @@ zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) size, buf, BP_GET_LSIZE(bp)); free(pbuf); } - if (error != 0) + if (error != 0 && print) printf("ZFS: i/o error - unable to decompress " "block pointer data, error %d\n", error); return (error); @@ -2394,7 +2429,7 @@ zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp)); else if (size != BP_GET_PSIZE(bp)) bcopy(pbuf, buf, BP_GET_PSIZE(bp)); - } else { + } else if (print) { printf("zio_read error: %d\n", error); } if (buf != pbuf) @@ -2402,12 +2437,18 @@ zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) if (error == 0) break; } - if (error != 0) + if (error != 0 && print) printf("ZFS: i/o error - all block copies unavailable\n"); return (error); } +static int +zio_read(const spa_t *spa, const blkptr_t *bp, void *buf) +{ + return (zio_read_impl(spa, bp, buf, true)); +} + static int dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen) diff --git a/sys/cddl/boot/zfs/zfsimpl.h b/sys/cddl/boot/zfs/zfsimpl.h index c9de1fe4c391..d3ae3c32635d 100644 --- a/sys/cddl/boot/zfs/zfsimpl.h +++ b/sys/cddl/boot/zfs/zfsimpl.h @@ -94,6 +94,7 @@ typedef enum { B_FALSE, B_TRUE } boolean_t; #define P2END(x, align) (-(~(x) & -(align))) #define P2PHASEUP(x, align, phase) ((phase) - (((phase) - (x)) & -(align))) #define P2BOUNDARY(off, len, align) (((off) ^ ((off) + (len) - 1)) > (align) - 1) +#define IS_P2ALIGNED(v, a) ((((uintptr_t)(v)) & ((uintptr_t)(a) - 1)) == 0) /* * General-purpose 32-bit and 64-bit bitfield encodings. @@ -498,19 +499,7 @@ typedef struct zio_eck { * Gang block headers are self-checksumming and contain an array * of block pointers. */ -#define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE -#define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \ - sizeof (zio_eck_t)) / sizeof (blkptr_t)) -#define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \ - sizeof (zio_eck_t) - \ - (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\ - sizeof (uint64_t)) - -typedef struct zio_gbh { - blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS]; - uint64_t zg_filler[SPA_GBH_FILLER]; - zio_eck_t zg_tail; -} zio_gbh_phys_t; +#define SPA_OLD_GANGBLOCKSIZE SPA_MINBLOCKSIZE #define VDEV_RAIDZ_MAXPARITY 3help
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202511061603.5A6G311Z000989>