diff options
Diffstat (limited to 'sys/contrib/openzfs/module')
48 files changed, 3230 insertions, 1118 deletions
diff --git a/sys/contrib/openzfs/module/Kbuild.in b/sys/contrib/openzfs/module/Kbuild.in index 3d6f288fa5da..58a80dc4402c 100644 --- a/sys/contrib/openzfs/module/Kbuild.in +++ b/sys/contrib/openzfs/module/Kbuild.in @@ -4,7 +4,7 @@ ZFS_MODULE_CFLAGS += -std=gnu99 -Wno-declaration-after-statement ZFS_MODULE_CFLAGS += -Wmissing-prototypes -ZFS_MODULE_CFLAGS += @KERNEL_DEBUG_CFLAGS@ @NO_FORMAT_ZERO_LENGTH@ +ZFS_MODULE_CFLAGS += @KERNEL_DEBUG_CFLAGS@ @KERNEL_NO_FORMAT_ZERO_LENGTH@ ifneq ($(KBUILD_EXTMOD),) zfs_include = @abs_top_srcdir@/include @@ -135,6 +135,7 @@ ICP_OBJS_X86_64 := \ asm-x86_64/sha2/sha256-x86_64.o \ asm-x86_64/sha2/sha512-x86_64.o \ asm-x86_64/modes/aesni-gcm-x86_64.o \ + asm-x86_64/modes/aesni-gcm-avx2-vaes.o \ asm-x86_64/modes/gcm_pclmulqdq.o \ asm-x86_64/modes/ghash-x86_64.o diff --git a/sys/contrib/openzfs/module/icp/algs/modes/gcm.c b/sys/contrib/openzfs/module/icp/algs/modes/gcm.c index c2a982b5a376..3cfa5b8165ce 100644 --- a/sys/contrib/openzfs/module/icp/algs/modes/gcm.c +++ b/sys/contrib/openzfs/module/icp/algs/modes/gcm.c @@ -46,6 +46,9 @@ #define IMPL_CYCLE (UINT32_MAX-1) #ifdef CAN_USE_GCM_ASM #define IMPL_AVX (UINT32_MAX-2) +#if CAN_USE_GCM_ASM >= 2 +#define IMPL_AVX2 (UINT32_MAX-3) +#endif #endif #define GCM_IMPL_READ(i) (*(volatile uint32_t *) &(i)) static uint32_t icp_gcm_impl = IMPL_FASTEST; @@ -56,17 +59,16 @@ static uint32_t user_sel_impl = IMPL_FASTEST; boolean_t gcm_avx_can_use_movbe = B_FALSE; /* * Whether to use the optimized openssl gcm and ghash implementations. - * Set to true if module parameter icp_gcm_impl == "avx". */ -static boolean_t gcm_use_avx = B_FALSE; -#define GCM_IMPL_USE_AVX (*(volatile boolean_t *)&gcm_use_avx) +static gcm_impl gcm_impl_used = GCM_IMPL_GENERIC; +#define GCM_IMPL_USED (*(volatile gcm_impl *)&gcm_impl_used) extern boolean_t ASMABI atomic_toggle_boolean_nv(volatile boolean_t *); static inline boolean_t gcm_avx_will_work(void); -static inline void gcm_set_avx(boolean_t); -static inline boolean_t gcm_toggle_avx(void); -static inline size_t gcm_simd_get_htab_size(boolean_t); +static inline boolean_t gcm_avx2_will_work(void); +static inline void gcm_use_impl(gcm_impl impl); +static inline gcm_impl gcm_toggle_impl(void); static int gcm_mode_encrypt_contiguous_blocks_avx(gcm_ctx_t *, char *, size_t, crypto_data_t *, size_t); @@ -89,7 +91,7 @@ gcm_mode_encrypt_contiguous_blocks(gcm_ctx_t *ctx, char *data, size_t length, void (*xor_block)(uint8_t *, uint8_t *)) { #ifdef CAN_USE_GCM_ASM - if (ctx->gcm_use_avx == B_TRUE) + if (ctx->impl != GCM_IMPL_GENERIC) return (gcm_mode_encrypt_contiguous_blocks_avx( ctx, data, length, out, block_size)); #endif @@ -208,7 +210,7 @@ gcm_encrypt_final(gcm_ctx_t *ctx, crypto_data_t *out, size_t block_size, { (void) copy_block; #ifdef CAN_USE_GCM_ASM - if (ctx->gcm_use_avx == B_TRUE) + if (ctx->impl != GCM_IMPL_GENERIC) return (gcm_encrypt_final_avx(ctx, out, block_size)); #endif @@ -374,7 +376,7 @@ gcm_decrypt_final(gcm_ctx_t *ctx, crypto_data_t *out, size_t block_size, void (*xor_block)(uint8_t *, uint8_t *)) { #ifdef CAN_USE_GCM_ASM - if (ctx->gcm_use_avx == B_TRUE) + if (ctx->impl != GCM_IMPL_GENERIC) return (gcm_decrypt_final_avx(ctx, out, block_size)); #endif @@ -631,23 +633,23 @@ gcm_init_ctx(gcm_ctx_t *gcm_ctx, char *param, ((aes_key_t *)gcm_ctx->gcm_keysched)->ops->needs_byteswap; if (GCM_IMPL_READ(icp_gcm_impl) != IMPL_CYCLE) { - gcm_ctx->gcm_use_avx = GCM_IMPL_USE_AVX; + gcm_ctx->impl = GCM_IMPL_USED; } else { /* - * Handle the "cycle" implementation by creating avx and - * non-avx contexts alternately. + * Handle the "cycle" implementation by creating different + * contexts, one per implementation. */ - gcm_ctx->gcm_use_avx = gcm_toggle_avx(); + gcm_ctx->impl = gcm_toggle_impl(); - /* The avx impl. doesn't handle byte swapped key schedules. */ - if (gcm_ctx->gcm_use_avx == B_TRUE && needs_bswap == B_TRUE) { - gcm_ctx->gcm_use_avx = B_FALSE; + /* The AVX impl. doesn't handle byte swapped key schedules. */ + if (needs_bswap == B_TRUE) { + gcm_ctx->impl = GCM_IMPL_GENERIC; } /* - * If this is a GCM context, use the MOVBE and the BSWAP + * If this is an AVX context, use the MOVBE and the BSWAP * variants alternately. */ - if (gcm_ctx->gcm_use_avx == B_TRUE && + if (gcm_ctx->impl == GCM_IMPL_AVX && zfs_movbe_available() == B_TRUE) { (void) atomic_toggle_boolean_nv( (volatile boolean_t *)&gcm_avx_can_use_movbe); @@ -658,12 +660,13 @@ gcm_init_ctx(gcm_ctx_t *gcm_ctx, char *param, * still they could be created by the aes generic implementation. * Make sure not to use them since we'll corrupt data if we do. */ - if (gcm_ctx->gcm_use_avx == B_TRUE && needs_bswap == B_TRUE) { - gcm_ctx->gcm_use_avx = B_FALSE; + if (gcm_ctx->impl != GCM_IMPL_GENERIC && needs_bswap == B_TRUE) { + gcm_ctx->impl = GCM_IMPL_GENERIC; cmn_err_once(CE_WARN, "ICP: Can't use the aes generic or cycle implementations " - "in combination with the gcm avx implementation!"); + "in combination with the gcm avx or avx2-vaes " + "implementation!"); cmn_err_once(CE_WARN, "ICP: Falling back to a compatible implementation, " "aes-gcm performance will likely be degraded."); @@ -672,36 +675,20 @@ gcm_init_ctx(gcm_ctx_t *gcm_ctx, char *param, "restore performance."); } - /* Allocate Htab memory as needed. */ - if (gcm_ctx->gcm_use_avx == B_TRUE) { - size_t htab_len = gcm_simd_get_htab_size(gcm_ctx->gcm_use_avx); - - if (htab_len == 0) { - return (CRYPTO_MECHANISM_PARAM_INVALID); - } - gcm_ctx->gcm_htab_len = htab_len; - gcm_ctx->gcm_Htable = - kmem_alloc(htab_len, KM_SLEEP); - - if (gcm_ctx->gcm_Htable == NULL) { - return (CRYPTO_HOST_MEMORY); - } + /* + * AVX implementations use Htable with sizes depending on + * implementation. + */ + if (gcm_ctx->impl != GCM_IMPL_GENERIC) { + rv = gcm_init_avx(gcm_ctx, iv, iv_len, aad, aad_len, + block_size); } - /* Avx and non avx context initialization differs from here on. */ - if (gcm_ctx->gcm_use_avx == B_FALSE) { + else #endif /* ifdef CAN_USE_GCM_ASM */ - if (gcm_init(gcm_ctx, iv, iv_len, aad, aad_len, block_size, - encrypt_block, copy_block, xor_block) != CRYPTO_SUCCESS) { - rv = CRYPTO_MECHANISM_PARAM_INVALID; - } -#ifdef CAN_USE_GCM_ASM - } else { - if (gcm_init_avx(gcm_ctx, iv, iv_len, aad, aad_len, - block_size) != CRYPTO_SUCCESS) { - rv = CRYPTO_MECHANISM_PARAM_INVALID; - } + if (gcm_init(gcm_ctx, iv, iv_len, aad, aad_len, block_size, + encrypt_block, copy_block, xor_block) != CRYPTO_SUCCESS) { + rv = CRYPTO_MECHANISM_PARAM_INVALID; } -#endif /* ifdef CAN_USE_GCM_ASM */ return (rv); } @@ -767,6 +754,9 @@ gcm_impl_get_ops(void) break; #ifdef CAN_USE_GCM_ASM case IMPL_AVX: +#if CAN_USE_GCM_ASM >= 2 + case IMPL_AVX2: +#endif /* * Make sure that we return a valid implementation while * switching to the avx implementation since there still @@ -828,6 +818,13 @@ gcm_impl_init(void) * Use the avx implementation if it's available and the implementation * hasn't changed from its default value of fastest on module load. */ +#if CAN_USE_GCM_ASM >= 2 + if (gcm_avx2_will_work()) { + if (GCM_IMPL_READ(user_sel_impl) == IMPL_FASTEST) { + gcm_use_impl(GCM_IMPL_AVX2); + } + } else +#endif if (gcm_avx_will_work()) { #ifdef HAVE_MOVBE if (zfs_movbe_available() == B_TRUE) { @@ -835,7 +832,7 @@ gcm_impl_init(void) } #endif if (GCM_IMPL_READ(user_sel_impl) == IMPL_FASTEST) { - gcm_set_avx(B_TRUE); + gcm_use_impl(GCM_IMPL_AVX); } } #endif @@ -852,6 +849,7 @@ static const struct { { "fastest", IMPL_FASTEST }, #ifdef CAN_USE_GCM_ASM { "avx", IMPL_AVX }, + { "avx2-vaes", IMPL_AVX2 }, #endif }; @@ -887,7 +885,13 @@ gcm_impl_set(const char *val) /* Check mandatory options */ for (i = 0; i < ARRAY_SIZE(gcm_impl_opts); i++) { #ifdef CAN_USE_GCM_ASM +#if CAN_USE_GCM_ASM >= 2 /* Ignore avx implementation if it won't work. */ + if (gcm_impl_opts[i].sel == IMPL_AVX2 && + !gcm_avx2_will_work()) { + continue; + } +#endif if (gcm_impl_opts[i].sel == IMPL_AVX && !gcm_avx_will_work()) { continue; } @@ -915,11 +919,17 @@ gcm_impl_set(const char *val) * Use the avx implementation if available and the requested one is * avx or fastest. */ +#if CAN_USE_GCM_ASM >= 2 + if (gcm_avx2_will_work() == B_TRUE && + (impl == IMPL_AVX2 || impl == IMPL_FASTEST)) { + gcm_use_impl(GCM_IMPL_AVX2); + } else +#endif if (gcm_avx_will_work() == B_TRUE && (impl == IMPL_AVX || impl == IMPL_FASTEST)) { - gcm_set_avx(B_TRUE); + gcm_use_impl(GCM_IMPL_AVX); } else { - gcm_set_avx(B_FALSE); + gcm_use_impl(GCM_IMPL_GENERIC); } #endif @@ -952,6 +962,12 @@ icp_gcm_impl_get(char *buffer, zfs_kernel_param_t *kp) for (i = 0; i < ARRAY_SIZE(gcm_impl_opts); i++) { #ifdef CAN_USE_GCM_ASM /* Ignore avx implementation if it won't work. */ +#if CAN_USE_GCM_ASM >= 2 + if (gcm_impl_opts[i].sel == IMPL_AVX2 && + !gcm_avx2_will_work()) { + continue; + } +#endif if (gcm_impl_opts[i].sel == IMPL_AVX && !gcm_avx_will_work()) { continue; } @@ -993,9 +1009,6 @@ MODULE_PARM_DESC(icp_gcm_impl, "Select gcm implementation."); /* Clear the FPU registers since they hold sensitive internal state. */ #define clear_fpu_regs() clear_fpu_regs_avx() -#define GHASH_AVX(ctx, in, len) \ - gcm_ghash_avx((ctx)->gcm_ghash, (const uint64_t *)(ctx)->gcm_Htable, \ - in, len) #define gcm_incr_counter_block(ctx) gcm_incr_counter_block_by(ctx, 1) @@ -1010,20 +1023,77 @@ MODULE_PARM_DESC(icp_gcm_impl, "Select gcm implementation."); static uint32_t gcm_avx_chunk_size = ((32 * 1024) / GCM_AVX_MIN_DECRYPT_BYTES) * GCM_AVX_MIN_DECRYPT_BYTES; +/* + * GCM definitions: uint128_t is copied from include/crypto/modes.h + * Avoiding u128 because it is already defined in kernel sources. + */ +typedef struct { + uint64_t hi, lo; +} uint128_t; + extern void ASMABI clear_fpu_regs_avx(void); extern void ASMABI gcm_xor_avx(const uint8_t *src, uint8_t *dst); extern void ASMABI aes_encrypt_intel(const uint32_t rk[], int nr, const uint32_t pt[4], uint32_t ct[4]); extern void ASMABI gcm_init_htab_avx(uint64_t *Htable, const uint64_t H[2]); +#if CAN_USE_GCM_ASM >= 2 +extern void ASMABI gcm_init_vpclmulqdq_avx2(uint128_t Htable[16], + const uint64_t H[2]); +#endif extern void ASMABI gcm_ghash_avx(uint64_t ghash[2], const uint64_t *Htable, const uint8_t *in, size_t len); +#if CAN_USE_GCM_ASM >= 2 +extern void ASMABI gcm_ghash_vpclmulqdq_avx2(uint64_t ghash[2], + const uint64_t *Htable, const uint8_t *in, size_t len); +#endif +static inline void GHASH_AVX(gcm_ctx_t *ctx, const uint8_t *in, size_t len) +{ + switch (ctx->impl) { +#if CAN_USE_GCM_ASM >= 2 + case GCM_IMPL_AVX2: + gcm_ghash_vpclmulqdq_avx2(ctx->gcm_ghash, + (const uint64_t *)ctx->gcm_Htable, in, len); + break; +#endif + + case GCM_IMPL_AVX: + gcm_ghash_avx(ctx->gcm_ghash, + (const uint64_t *)ctx->gcm_Htable, in, len); + break; + + default: + VERIFY(B_FALSE); + } +} +typedef size_t ASMABI aesni_gcm_encrypt_impl(const uint8_t *, uint8_t *, + size_t, const void *, uint64_t *, const uint64_t *Htable, uint64_t *); extern size_t ASMABI aesni_gcm_encrypt(const uint8_t *, uint8_t *, size_t, const void *, uint64_t *, uint64_t *); +#if CAN_USE_GCM_ASM >= 2 +extern void ASMABI aes_gcm_enc_update_vaes_avx2(const uint8_t *in, + uint8_t *out, size_t len, const void *key, const uint8_t ivec[16], + const uint128_t Htable[16], uint8_t Xi[16]); +#endif +typedef size_t ASMABI aesni_gcm_decrypt_impl(const uint8_t *, uint8_t *, + size_t, const void *, uint64_t *, const uint64_t *Htable, uint64_t *); extern size_t ASMABI aesni_gcm_decrypt(const uint8_t *, uint8_t *, size_t, const void *, uint64_t *, uint64_t *); +#if CAN_USE_GCM_ASM >= 2 +extern void ASMABI aes_gcm_dec_update_vaes_avx2(const uint8_t *in, + uint8_t *out, size_t len, const void *key, const uint8_t ivec[16], + const uint128_t Htable[16], uint8_t Xi[16]); +#endif + +static inline boolean_t +gcm_avx2_will_work(void) +{ + return (kfpu_allowed() && + zfs_avx2_available() && zfs_vaes_available() && + zfs_vpclmulqdq_available()); +} static inline boolean_t gcm_avx_will_work(void) @@ -1035,33 +1105,67 @@ gcm_avx_will_work(void) } static inline void -gcm_set_avx(boolean_t val) +gcm_use_impl(gcm_impl impl) { - if (gcm_avx_will_work() == B_TRUE) { - atomic_swap_32(&gcm_use_avx, val); + switch (impl) { +#if CAN_USE_GCM_ASM >= 2 + case GCM_IMPL_AVX2: + if (gcm_avx2_will_work() == B_TRUE) { + atomic_swap_32(&gcm_impl_used, impl); + return; + } + + zfs_fallthrough; +#endif + + case GCM_IMPL_AVX: + if (gcm_avx_will_work() == B_TRUE) { + atomic_swap_32(&gcm_impl_used, impl); + return; + } + + zfs_fallthrough; + + default: + atomic_swap_32(&gcm_impl_used, GCM_IMPL_GENERIC); } } static inline boolean_t -gcm_toggle_avx(void) +gcm_impl_will_work(gcm_impl impl) { - if (gcm_avx_will_work() == B_TRUE) { - return (atomic_toggle_boolean_nv(&GCM_IMPL_USE_AVX)); - } else { - return (B_FALSE); + switch (impl) { +#if CAN_USE_GCM_ASM >= 2 + case GCM_IMPL_AVX2: + return (gcm_avx2_will_work()); +#endif + + case GCM_IMPL_AVX: + return (gcm_avx_will_work()); + + default: + return (B_TRUE); } } -static inline size_t -gcm_simd_get_htab_size(boolean_t simd_mode) +static inline gcm_impl +gcm_toggle_impl(void) { - switch (simd_mode) { - case B_TRUE: - return (2 * 6 * 2 * sizeof (uint64_t)); + gcm_impl current_impl, new_impl; + do { /* handle races */ + current_impl = atomic_load_32(&gcm_impl_used); + new_impl = current_impl; + while (B_TRUE) { /* handle incompatble implementations */ + new_impl = (new_impl + 1) % GCM_IMPL_MAX; + if (gcm_impl_will_work(new_impl)) { + break; + } + } - default: - return (0); - } + } while (atomic_cas_32(&gcm_impl_used, current_impl, new_impl) != + current_impl); + + return (new_impl); } @@ -1077,6 +1181,50 @@ gcm_incr_counter_block_by(gcm_ctx_t *ctx, int n) ctx->gcm_cb[1] = (ctx->gcm_cb[1] & ~counter_mask) | counter; } +static size_t aesni_gcm_encrypt_avx(const uint8_t *in, uint8_t *out, + size_t len, const void *key, uint64_t *iv, const uint64_t *Htable, + uint64_t *Xip) +{ + (void) Htable; + return (aesni_gcm_encrypt(in, out, len, key, iv, Xip)); +} + +#if CAN_USE_GCM_ASM >= 2 +// kSizeTWithoutLower4Bits is a mask that can be used to zero the lower four +// bits of a |size_t|. +// This is from boringssl/crypto/fipsmodule/aes/gcm.cc.inc +static const size_t kSizeTWithoutLower4Bits = (size_t)-16; + +/* The following CRYPTO methods are from boringssl/crypto/internal.h */ +static inline uint32_t CRYPTO_bswap4(uint32_t x) { + return (__builtin_bswap32(x)); +} + +static inline uint32_t CRYPTO_load_u32_be(const void *in) { + uint32_t v; + memcpy(&v, in, sizeof (v)); + return (CRYPTO_bswap4(v)); +} + +static inline void CRYPTO_store_u32_be(void *out, uint32_t v) { + v = CRYPTO_bswap4(v); + memcpy(out, &v, sizeof (v)); +} + +static size_t aesni_gcm_encrypt_avx2(const uint8_t *in, uint8_t *out, + size_t len, const void *key, uint64_t *iv, const uint64_t *Htable, + uint64_t *Xip) +{ + uint8_t *ivec = (uint8_t *)iv; + len &= kSizeTWithoutLower4Bits; + aes_gcm_enc_update_vaes_avx2(in, out, len, key, ivec, + (const uint128_t *)Htable, (uint8_t *)Xip); + CRYPTO_store_u32_be(&ivec[12], + CRYPTO_load_u32_be(&ivec[12]) + len / 16); + return (len); +} +#endif /* if CAN_USE_GCM_ASM >= 2 */ + /* * Encrypt multiple blocks of data in GCM mode. * This is done in gcm_avx_chunk_size chunks, utilizing AVX assembler routines @@ -1091,8 +1239,15 @@ gcm_mode_encrypt_contiguous_blocks_avx(gcm_ctx_t *ctx, char *data, size_t done = 0; uint8_t *datap = (uint8_t *)data; size_t chunk_size = (size_t)GCM_CHUNK_SIZE_READ; + aesni_gcm_encrypt_impl *encrypt_blocks = +#if CAN_USE_GCM_ASM >= 2 + ctx->impl == GCM_IMPL_AVX2 ? + aesni_gcm_encrypt_avx2 : +#endif + aesni_gcm_encrypt_avx; const aes_key_t *key = ((aes_key_t *)ctx->gcm_keysched); uint64_t *ghash = ctx->gcm_ghash; + uint64_t *htable = ctx->gcm_Htable; uint64_t *cb = ctx->gcm_cb; uint8_t *ct_buf = NULL; uint8_t *tmp = (uint8_t *)ctx->gcm_tmp; @@ -1156,8 +1311,8 @@ gcm_mode_encrypt_contiguous_blocks_avx(gcm_ctx_t *ctx, char *data, /* Do the bulk encryption in chunk_size blocks. */ for (; bleft >= chunk_size; bleft -= chunk_size) { kfpu_begin(); - done = aesni_gcm_encrypt( - datap, ct_buf, chunk_size, key, cb, ghash); + done = encrypt_blocks( + datap, ct_buf, chunk_size, key, cb, htable, ghash); clear_fpu_regs(); kfpu_end(); @@ -1180,7 +1335,8 @@ gcm_mode_encrypt_contiguous_blocks_avx(gcm_ctx_t *ctx, char *data, /* Bulk encrypt the remaining data. */ kfpu_begin(); if (bleft >= GCM_AVX_MIN_ENCRYPT_BYTES) { - done = aesni_gcm_encrypt(datap, ct_buf, bleft, key, cb, ghash); + done = encrypt_blocks(datap, ct_buf, bleft, key, cb, htable, + ghash); if (done == 0) { rv = CRYPTO_FAILED; goto out; @@ -1293,6 +1449,29 @@ gcm_encrypt_final_avx(gcm_ctx_t *ctx, crypto_data_t *out, size_t block_size) return (CRYPTO_SUCCESS); } +static size_t aesni_gcm_decrypt_avx(const uint8_t *in, uint8_t *out, + size_t len, const void *key, uint64_t *iv, const uint64_t *Htable, + uint64_t *Xip) +{ + (void) Htable; + return (aesni_gcm_decrypt(in, out, len, key, iv, Xip)); +} + +#if CAN_USE_GCM_ASM >= 2 +static size_t aesni_gcm_decrypt_avx2(const uint8_t *in, uint8_t *out, + size_t len, const void *key, uint64_t *iv, const uint64_t *Htable, + uint64_t *Xip) +{ + uint8_t *ivec = (uint8_t *)iv; + len &= kSizeTWithoutLower4Bits; + aes_gcm_dec_update_vaes_avx2(in, out, len, key, ivec, + (const uint128_t *)Htable, (uint8_t *)Xip); + CRYPTO_store_u32_be(&ivec[12], + CRYPTO_load_u32_be(&ivec[12]) + len / 16); + return (len); +} +#endif /* if CAN_USE_GCM_ASM >= 2 */ + /* * Finalize decryption: We just have accumulated crypto text, so now we * decrypt it here inplace. @@ -1306,10 +1485,17 @@ gcm_decrypt_final_avx(gcm_ctx_t *ctx, crypto_data_t *out, size_t block_size) B_FALSE); size_t chunk_size = (size_t)GCM_CHUNK_SIZE_READ; + aesni_gcm_decrypt_impl *decrypt_blocks = +#if CAN_USE_GCM_ASM >= 2 + ctx->impl == GCM_IMPL_AVX2 ? + aesni_gcm_decrypt_avx2 : +#endif + aesni_gcm_decrypt_avx; size_t pt_len = ctx->gcm_processed_data_len - ctx->gcm_tag_len; uint8_t *datap = ctx->gcm_pt_buf; const aes_key_t *key = ((aes_key_t *)ctx->gcm_keysched); uint32_t *cb = (uint32_t *)ctx->gcm_cb; + uint64_t *htable = ctx->gcm_Htable; uint64_t *ghash = ctx->gcm_ghash; uint32_t *tmp = (uint32_t *)ctx->gcm_tmp; int rv = CRYPTO_SUCCESS; @@ -1322,8 +1508,8 @@ gcm_decrypt_final_avx(gcm_ctx_t *ctx, crypto_data_t *out, size_t block_size) */ for (bleft = pt_len; bleft >= chunk_size; bleft -= chunk_size) { kfpu_begin(); - done = aesni_gcm_decrypt(datap, datap, chunk_size, - (const void *)key, ctx->gcm_cb, ghash); + done = decrypt_blocks(datap, datap, chunk_size, + (const void *)key, ctx->gcm_cb, htable, ghash); clear_fpu_regs(); kfpu_end(); if (done != chunk_size) { @@ -1334,8 +1520,8 @@ gcm_decrypt_final_avx(gcm_ctx_t *ctx, crypto_data_t *out, size_t block_size) /* Decrypt remainder, which is less than chunk size, in one go. */ kfpu_begin(); if (bleft >= GCM_AVX_MIN_DECRYPT_BYTES) { - done = aesni_gcm_decrypt(datap, datap, bleft, - (const void *)key, ctx->gcm_cb, ghash); + done = decrypt_blocks(datap, datap, bleft, + (const void *)key, ctx->gcm_cb, htable, ghash); if (done == 0) { clear_fpu_regs(); kfpu_end(); @@ -1424,13 +1610,42 @@ gcm_init_avx(gcm_ctx_t *ctx, const uint8_t *iv, size_t iv_len, ASSERT3S(((aes_key_t *)ctx->gcm_keysched)->ops->needs_byteswap, ==, B_FALSE); + size_t htab_len = 0; +#if CAN_USE_GCM_ASM >= 2 + if (ctx->impl == GCM_IMPL_AVX2) { + /* + * BoringSSL's API specifies uint128_t[16] for htab; but only + * uint128_t[12] are used. + * See https://github.com/google/boringssl/blob/ + * 813840dd094f9e9c1b00a7368aa25e656554221f1/crypto/fipsmodule/ + * modes/asm/aes-gcm-avx2-x86_64.pl#L198-L200 + */ + htab_len = (2 * 8 * sizeof (uint128_t)); + } else +#endif /* CAN_USE_GCM_ASM >= 2 */ + { + htab_len = (2 * 6 * sizeof (uint128_t)); + } + + ctx->gcm_Htable = kmem_alloc(htab_len, KM_SLEEP); + if (ctx->gcm_Htable == NULL) { + return (CRYPTO_HOST_MEMORY); + } + /* Init H (encrypt zero block) and create the initial counter block. */ memset(H, 0, sizeof (ctx->gcm_H)); kfpu_begin(); aes_encrypt_intel(keysched, aes_rounds, (const uint32_t *)H, (uint32_t *)H); - gcm_init_htab_avx(ctx->gcm_Htable, H); +#if CAN_USE_GCM_ASM >= 2 + if (ctx->impl == GCM_IMPL_AVX2) { + gcm_init_vpclmulqdq_avx2((uint128_t *)ctx->gcm_Htable, H); + } else +#endif /* if CAN_USE_GCM_ASM >= 2 */ + { + gcm_init_htab_avx(ctx->gcm_Htable, H); + } if (iv_len == 12) { memcpy(cb, iv, 12); diff --git a/sys/contrib/openzfs/module/icp/algs/modes/modes.c b/sys/contrib/openzfs/module/icp/algs/modes/modes.c index 343591cd9691..ef3c1806e4b6 100644 --- a/sys/contrib/openzfs/module/icp/algs/modes/modes.c +++ b/sys/contrib/openzfs/module/icp/algs/modes/modes.c @@ -171,7 +171,7 @@ gcm_clear_ctx(gcm_ctx_t *ctx) explicit_memset(ctx->gcm_remainder, 0, sizeof (ctx->gcm_remainder)); explicit_memset(ctx->gcm_H, 0, sizeof (ctx->gcm_H)); #if defined(CAN_USE_GCM_ASM) - if (ctx->gcm_use_avx == B_TRUE) { + if (ctx->impl != GCM_IMPL_GENERIC) { ASSERT3P(ctx->gcm_Htable, !=, NULL); explicit_memset(ctx->gcm_Htable, 0, ctx->gcm_htab_len); kmem_free(ctx->gcm_Htable, ctx->gcm_htab_len); diff --git a/sys/contrib/openzfs/module/icp/algs/sha2/sha256_impl.c b/sys/contrib/openzfs/module/icp/algs/sha2/sha256_impl.c index 6d3bcca9f995..dcb0a391dda4 100644 --- a/sys/contrib/openzfs/module/icp/algs/sha2/sha256_impl.c +++ b/sys/contrib/openzfs/module/icp/algs/sha2/sha256_impl.c @@ -38,11 +38,14 @@ kfpu_begin(); E(s, d, b); kfpu_end(); \ } +#if defined(__x86_64) || defined(__aarch64__) || defined(__arm__) || \ + defined(__PPC64__) /* some implementation is always okay */ static inline boolean_t sha2_is_supported(void) { return (B_TRUE); } +#endif #if defined(__x86_64) diff --git a/sys/contrib/openzfs/module/icp/algs/sha2/sha512_impl.c b/sys/contrib/openzfs/module/icp/algs/sha2/sha512_impl.c index 2efd9fcf4c99..a85a71a83df4 100644 --- a/sys/contrib/openzfs/module/icp/algs/sha2/sha512_impl.c +++ b/sys/contrib/openzfs/module/icp/algs/sha2/sha512_impl.c @@ -38,11 +38,14 @@ kfpu_begin(); E(s, d, b); kfpu_end(); \ } +#if defined(__x86_64) || defined(__aarch64__) || defined(__arm__) || \ + defined(__aarch64__) || defined(__arm__) || defined(__PPC64__) /* some implementation is always okay */ static inline boolean_t sha2_is_supported(void) { return (B_TRUE); } +#endif #if defined(__x86_64) diff --git a/sys/contrib/openzfs/module/icp/asm-x86_64/modes/THIRDPARTYLICENSE.boringssl b/sys/contrib/openzfs/module/icp/asm-x86_64/modes/THIRDPARTYLICENSE.boringssl new file mode 100644 index 000000000000..04c03a37e0cb --- /dev/null +++ b/sys/contrib/openzfs/module/icp/asm-x86_64/modes/THIRDPARTYLICENSE.boringssl @@ -0,0 +1,253 @@ +BoringSSL is a fork of OpenSSL. As such, large parts of it fall under OpenSSL +licensing. Files that are completely new have a Google copyright and an ISC +license. This license is reproduced at the bottom of this file. + +Contributors to BoringSSL are required to follow the CLA rules for Chromium: +https://cla.developers.google.com/clas + +Files in third_party/ have their own licenses, as described therein. The MIT +license, for third_party/fiat, which, unlike other third_party directories, is +compiled into non-test libraries, is included below. + +The OpenSSL toolkit stays under a dual license, i.e. both the conditions of the +OpenSSL License and the original SSLeay license apply to the toolkit. See below +for the actual license texts. Actually both licenses are BSD-style Open Source +licenses. In case of any license issues related to OpenSSL please contact +openssl-core@openssl.org. + +The following are Google-internal bug numbers where explicit permission from +some authors is recorded for use of their work. (This is purely for our own +record keeping.) + 27287199 + 27287880 + 27287883 + 263291445 + + + OpenSSL License + --------------- + +/* ==================================================================== + * Copyright (c) 1998-2011 The OpenSSL Project. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * + * 3. All advertising materials mentioning features or use of this + * software must display the following acknowledgment: + * "This product includes software developed by the OpenSSL Project + * for use in the OpenSSL Toolkit. (http://www.openssl.org/)" + * + * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to + * endorse or promote products derived from this software without + * prior written permission. For written permission, please contact + * openssl-core@openssl.org. + * + * 5. Products derived from this software may not be called "OpenSSL" + * nor may "OpenSSL" appear in their names without prior written + * permission of the OpenSSL Project. + * + * 6. Redistributions of any form whatsoever must retain the following + * acknowledgment: + * "This product includes software developed by the OpenSSL Project + * for use in the OpenSSL Toolkit (http://www.openssl.org/)" + * + * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY + * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR + * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT + * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + * ==================================================================== + * + * This product includes cryptographic software written by Eric Young + * (eay@cryptsoft.com). This product includes software written by Tim + * Hudson (tjh@cryptsoft.com). + * + */ + + Original SSLeay License + ----------------------- + +/* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com) + * All rights reserved. + * + * This package is an SSL implementation written + * by Eric Young (eay@cryptsoft.com). + * The implementation was written so as to conform with Netscapes SSL. + * + * This library is free for commercial and non-commercial use as long as + * the following conditions are aheared to. The following conditions + * apply to all code found in this distribution, be it the RC4, RSA, + * lhash, DES, etc., code; not just the SSL code. The SSL documentation + * included with this distribution is covered by the same copyright terms + * except that the holder is Tim Hudson (tjh@cryptsoft.com). + * + * Copyright remains Eric Young's, and as such any Copyright notices in + * the code are not to be removed. + * If this package is used in a product, Eric Young should be given attribution + * as the author of the parts of the library used. + * This can be in the form of a textual message at program startup or + * in documentation (online or textual) provided with the package. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * "This product includes cryptographic software written by + * Eric Young (eay@cryptsoft.com)" + * The word 'cryptographic' can be left out if the rouines from the library + * being used are not cryptographic related :-). + * 4. If you include any Windows specific code (or a derivative thereof) from + * the apps directory (application code) you must include an acknowledgement: + * "This product includes software written by Tim Hudson (tjh@cryptsoft.com)" + * + * THIS SOFTWARE IS PROVIDED BY ERIC YOUNG ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * The licence and distribution terms for any publically available version or + * derivative of this code cannot be changed. i.e. this code cannot simply be + * copied and put under another distribution licence + * [including the GNU Public Licence.] + */ + + +ISC license used for completely new code in BoringSSL: + +/* Copyright 2015 The BoringSSL Authors + * + * Permission to use, copy, modify, and/or distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY + * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION + * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN + * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ + + +The code in third_party/fiat carries the MIT license: + +Copyright (c) 2015-2016 the fiat-crypto authors (see +https://github.com/mit-plv/fiat-crypto/blob/master/AUTHORS). + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. + + +Licenses for support code +------------------------- + +Parts of the TLS test suite are under the Go license. This code is not included +in BoringSSL (i.e. libcrypto and libssl) when compiled, however, so +distributing code linked against BoringSSL does not trigger this license: + +Copyright (c) 2009 The Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + +BoringSSL uses the Chromium test infrastructure to run a continuous build, +trybots etc. The scripts which manage this, and the script for generating build +metadata, are under the Chromium license. Distributing code linked against +BoringSSL does not trigger this license. + +Copyright 2015 The Chromium Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/sys/contrib/openzfs/module/icp/asm-x86_64/modes/THIRDPARTYLICENSE.boringssl.descrip b/sys/contrib/openzfs/module/icp/asm-x86_64/modes/THIRDPARTYLICENSE.boringssl.descrip new file mode 100644 index 000000000000..f63a67a4d2ae --- /dev/null +++ b/sys/contrib/openzfs/module/icp/asm-x86_64/modes/THIRDPARTYLICENSE.boringssl.descrip @@ -0,0 +1 @@ +PORTIONS OF AES GCM and GHASH FUNCTIONALITY diff --git a/sys/contrib/openzfs/module/icp/asm-x86_64/modes/aesni-gcm-avx2-vaes.S b/sys/contrib/openzfs/module/icp/asm-x86_64/modes/aesni-gcm-avx2-vaes.S new file mode 100644 index 000000000000..3d1b045127e2 --- /dev/null +++ b/sys/contrib/openzfs/module/icp/asm-x86_64/modes/aesni-gcm-avx2-vaes.S @@ -0,0 +1,1323 @@ +// SPDX-License-Identifier: Apache-2.0 +// This file is generated from a similarly-named Perl script in the BoringSSL +// source tree. Do not edit by hand. + +#if defined(__x86_64__) && defined(HAVE_AVX) && \ + defined(HAVE_VAES) && defined(HAVE_VPCLMULQDQ) + +#define _ASM +#include <sys/asm_linkage.h> + +/* Windows userland links with OpenSSL */ +#if !defined (_WIN32) || defined (_KERNEL) + +.section .rodata +.balign 16 + + +.Lbswap_mask: +.quad 0x08090a0b0c0d0e0f, 0x0001020304050607 + + + + + + + + +.Lgfpoly: +.quad 1, 0xc200000000000000 + + +.Lgfpoly_and_internal_carrybit: +.quad 1, 0xc200000000000001 + +.balign 32 + +.Lctr_pattern: +.quad 0, 0 +.quad 1, 0 +.Linc_2blocks: +.quad 2, 0 +.quad 2, 0 + +ENTRY_ALIGN(gcm_init_vpclmulqdq_avx2, 32) +.cfi_startproc + +ENDBR + + + + + + vmovdqu (%rsi),%xmm3 + // KCF/ICP stores H in network byte order with the hi qword first + // so we need to swap all bytes, not the 2 qwords. + vmovdqu .Lbswap_mask(%rip),%xmm4 + vpshufb %xmm4,%xmm3,%xmm3 + + + + + + vpshufd $0xd3,%xmm3,%xmm0 + vpsrad $31,%xmm0,%xmm0 + vpaddq %xmm3,%xmm3,%xmm3 + vpand .Lgfpoly_and_internal_carrybit(%rip),%xmm0,%xmm0 + vpxor %xmm0,%xmm3,%xmm3 + + vbroadcasti128 .Lgfpoly(%rip),%ymm6 + + + vpclmulqdq $0x00,%xmm3,%xmm3,%xmm0 + vpclmulqdq $0x11,%xmm3,%xmm3,%xmm5 + vpclmulqdq $0x01,%xmm0,%xmm6,%xmm1 + vpshufd $0x4e,%xmm0,%xmm0 + vpxor %xmm0,%xmm1,%xmm1 + vpclmulqdq $0x01,%xmm1,%xmm6,%xmm0 + vpshufd $0x4e,%xmm1,%xmm1 + vpxor %xmm1,%xmm5,%xmm5 + vpxor %xmm0,%xmm5,%xmm5 + + + + vinserti128 $1,%xmm3,%ymm5,%ymm3 + vinserti128 $1,%xmm5,%ymm5,%ymm5 + + + vpclmulqdq $0x00,%ymm5,%ymm3,%ymm0 + vpclmulqdq $0x01,%ymm5,%ymm3,%ymm1 + vpclmulqdq $0x10,%ymm5,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpclmulqdq $0x01,%ymm0,%ymm6,%ymm2 + vpshufd $0x4e,%ymm0,%ymm0 + vpxor %ymm0,%ymm1,%ymm1 + vpxor %ymm2,%ymm1,%ymm1 + vpclmulqdq $0x11,%ymm5,%ymm3,%ymm4 + vpclmulqdq $0x01,%ymm1,%ymm6,%ymm0 + vpshufd $0x4e,%ymm1,%ymm1 + vpxor %ymm1,%ymm4,%ymm4 + vpxor %ymm0,%ymm4,%ymm4 + + + + vmovdqu %ymm3,96(%rdi) + vmovdqu %ymm4,64(%rdi) + + + + vpunpcklqdq %ymm3,%ymm4,%ymm0 + vpunpckhqdq %ymm3,%ymm4,%ymm1 + vpxor %ymm1,%ymm0,%ymm0 + vmovdqu %ymm0,128+32(%rdi) + + + vpclmulqdq $0x00,%ymm5,%ymm4,%ymm0 + vpclmulqdq $0x01,%ymm5,%ymm4,%ymm1 + vpclmulqdq $0x10,%ymm5,%ymm4,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpclmulqdq $0x01,%ymm0,%ymm6,%ymm2 + vpshufd $0x4e,%ymm0,%ymm0 + vpxor %ymm0,%ymm1,%ymm1 + vpxor %ymm2,%ymm1,%ymm1 + vpclmulqdq $0x11,%ymm5,%ymm4,%ymm3 + vpclmulqdq $0x01,%ymm1,%ymm6,%ymm0 + vpshufd $0x4e,%ymm1,%ymm1 + vpxor %ymm1,%ymm3,%ymm3 + vpxor %ymm0,%ymm3,%ymm3 + + vpclmulqdq $0x00,%ymm5,%ymm3,%ymm0 + vpclmulqdq $0x01,%ymm5,%ymm3,%ymm1 + vpclmulqdq $0x10,%ymm5,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpclmulqdq $0x01,%ymm0,%ymm6,%ymm2 + vpshufd $0x4e,%ymm0,%ymm0 + vpxor %ymm0,%ymm1,%ymm1 + vpxor %ymm2,%ymm1,%ymm1 + vpclmulqdq $0x11,%ymm5,%ymm3,%ymm4 + vpclmulqdq $0x01,%ymm1,%ymm6,%ymm0 + vpshufd $0x4e,%ymm1,%ymm1 + vpxor %ymm1,%ymm4,%ymm4 + vpxor %ymm0,%ymm4,%ymm4 + + vmovdqu %ymm3,32(%rdi) + vmovdqu %ymm4,0(%rdi) + + + + vpunpcklqdq %ymm3,%ymm4,%ymm0 + vpunpckhqdq %ymm3,%ymm4,%ymm1 + vpxor %ymm1,%ymm0,%ymm0 + vmovdqu %ymm0,128(%rdi) + + vzeroupper + RET + +.cfi_endproc +SET_SIZE(gcm_init_vpclmulqdq_avx2) +ENTRY_ALIGN(gcm_gmult_vpclmulqdq_avx2, 32) +.cfi_startproc + +ENDBR + + + + vmovdqu (%rdi),%xmm0 + vmovdqu .Lbswap_mask(%rip),%xmm1 + vmovdqu 128-16(%rsi),%xmm2 + vmovdqu .Lgfpoly(%rip),%xmm3 + vpshufb %xmm1,%xmm0,%xmm0 + + vpclmulqdq $0x00,%xmm2,%xmm0,%xmm4 + vpclmulqdq $0x01,%xmm2,%xmm0,%xmm5 + vpclmulqdq $0x10,%xmm2,%xmm0,%xmm6 + vpxor %xmm6,%xmm5,%xmm5 + vpclmulqdq $0x01,%xmm4,%xmm3,%xmm6 + vpshufd $0x4e,%xmm4,%xmm4 + vpxor %xmm4,%xmm5,%xmm5 + vpxor %xmm6,%xmm5,%xmm5 + vpclmulqdq $0x11,%xmm2,%xmm0,%xmm0 + vpclmulqdq $0x01,%xmm5,%xmm3,%xmm4 + vpshufd $0x4e,%xmm5,%xmm5 + vpxor %xmm5,%xmm0,%xmm0 + vpxor %xmm4,%xmm0,%xmm0 + + + vpshufb %xmm1,%xmm0,%xmm0 + vmovdqu %xmm0,(%rdi) + + + RET + +.cfi_endproc +SET_SIZE(gcm_gmult_vpclmulqdq_avx2) +ENTRY_ALIGN(gcm_ghash_vpclmulqdq_avx2, 32) +.cfi_startproc + +ENDBR + + + + + + + vmovdqu .Lbswap_mask(%rip),%xmm6 + vmovdqu .Lgfpoly(%rip),%xmm7 + + + vmovdqu (%rdi),%xmm5 + vpshufb %xmm6,%xmm5,%xmm5 + + + cmpq $32,%rcx + jb .Lghash_lastblock + + + + vinserti128 $1,%xmm6,%ymm6,%ymm6 + vinserti128 $1,%xmm7,%ymm7,%ymm7 + + cmpq $127,%rcx + jbe .Lghash_loop_1x + + + vmovdqu 128(%rsi),%ymm8 + vmovdqu 128+32(%rsi),%ymm9 +.Lghash_loop_4x: + + vmovdqu 0(%rdx),%ymm1 + vpshufb %ymm6,%ymm1,%ymm1 + vmovdqu 0(%rsi),%ymm2 + vpxor %ymm5,%ymm1,%ymm1 + vpclmulqdq $0x00,%ymm2,%ymm1,%ymm3 + vpclmulqdq $0x11,%ymm2,%ymm1,%ymm5 + vpunpckhqdq %ymm1,%ymm1,%ymm0 + vpxor %ymm1,%ymm0,%ymm0 + vpclmulqdq $0x00,%ymm8,%ymm0,%ymm4 + + vmovdqu 32(%rdx),%ymm1 + vpshufb %ymm6,%ymm1,%ymm1 + vmovdqu 32(%rsi),%ymm2 + vpclmulqdq $0x00,%ymm2,%ymm1,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vpclmulqdq $0x11,%ymm2,%ymm1,%ymm0 + vpxor %ymm0,%ymm5,%ymm5 + vpunpckhqdq %ymm1,%ymm1,%ymm0 + vpxor %ymm1,%ymm0,%ymm0 + vpclmulqdq $0x10,%ymm8,%ymm0,%ymm0 + vpxor %ymm0,%ymm4,%ymm4 + + vmovdqu 64(%rdx),%ymm1 + vpshufb %ymm6,%ymm1,%ymm1 + vmovdqu 64(%rsi),%ymm2 + vpclmulqdq $0x00,%ymm2,%ymm1,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vpclmulqdq $0x11,%ymm2,%ymm1,%ymm0 + vpxor %ymm0,%ymm5,%ymm5 + vpunpckhqdq %ymm1,%ymm1,%ymm0 + vpxor %ymm1,%ymm0,%ymm0 + vpclmulqdq $0x00,%ymm9,%ymm0,%ymm0 + vpxor %ymm0,%ymm4,%ymm4 + + + vmovdqu 96(%rdx),%ymm1 + vpshufb %ymm6,%ymm1,%ymm1 + vmovdqu 96(%rsi),%ymm2 + vpclmulqdq $0x00,%ymm2,%ymm1,%ymm0 + vpxor %ymm0,%ymm3,%ymm3 + vpclmulqdq $0x11,%ymm2,%ymm1,%ymm0 + vpxor %ymm0,%ymm5,%ymm5 + vpunpckhqdq %ymm1,%ymm1,%ymm0 + vpxor %ymm1,%ymm0,%ymm0 + vpclmulqdq $0x10,%ymm9,%ymm0,%ymm0 + vpxor %ymm0,%ymm4,%ymm4 + + vpxor %ymm3,%ymm4,%ymm4 + vpxor %ymm5,%ymm4,%ymm4 + + + vbroadcasti128 .Lgfpoly(%rip),%ymm2 + vpclmulqdq $0x01,%ymm3,%ymm2,%ymm0 + vpshufd $0x4e,%ymm3,%ymm3 + vpxor %ymm3,%ymm4,%ymm4 + vpxor %ymm0,%ymm4,%ymm4 + + vpclmulqdq $0x01,%ymm4,%ymm2,%ymm0 + vpshufd $0x4e,%ymm4,%ymm4 + vpxor %ymm4,%ymm5,%ymm5 + vpxor %ymm0,%ymm5,%ymm5 + vextracti128 $1,%ymm5,%xmm0 + vpxor %xmm0,%xmm5,%xmm5 + + subq $-128,%rdx + addq $-128,%rcx + cmpq $127,%rcx + ja .Lghash_loop_4x + + + cmpq $32,%rcx + jb .Lghash_loop_1x_done +.Lghash_loop_1x: + vmovdqu (%rdx),%ymm0 + vpshufb %ymm6,%ymm0,%ymm0 + vpxor %ymm0,%ymm5,%ymm5 + vmovdqu 128-32(%rsi),%ymm0 + vpclmulqdq $0x00,%ymm0,%ymm5,%ymm1 + vpclmulqdq $0x01,%ymm0,%ymm5,%ymm2 + vpclmulqdq $0x10,%ymm0,%ymm5,%ymm3 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x01,%ymm1,%ymm7,%ymm3 + vpshufd $0x4e,%ymm1,%ymm1 + vpxor %ymm1,%ymm2,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x11,%ymm0,%ymm5,%ymm5 + vpclmulqdq $0x01,%ymm2,%ymm7,%ymm1 + vpshufd $0x4e,%ymm2,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpxor %ymm1,%ymm5,%ymm5 + + vextracti128 $1,%ymm5,%xmm0 + vpxor %xmm0,%xmm5,%xmm5 + addq $32,%rdx + subq $32,%rcx + cmpq $32,%rcx + jae .Lghash_loop_1x +.Lghash_loop_1x_done: + + +.Lghash_lastblock: + testq %rcx,%rcx + jz .Lghash_done + vmovdqu (%rdx),%xmm0 + vpshufb %xmm6,%xmm0,%xmm0 + vpxor %xmm0,%xmm5,%xmm5 + vmovdqu 128-16(%rsi),%xmm0 + vpclmulqdq $0x00,%xmm0,%xmm5,%xmm1 + vpclmulqdq $0x01,%xmm0,%xmm5,%xmm2 + vpclmulqdq $0x10,%xmm0,%xmm5,%xmm3 + vpxor %xmm3,%xmm2,%xmm2 + vpclmulqdq $0x01,%xmm1,%xmm7,%xmm3 + vpshufd $0x4e,%xmm1,%xmm1 + vpxor %xmm1,%xmm2,%xmm2 + vpxor %xmm3,%xmm2,%xmm2 + vpclmulqdq $0x11,%xmm0,%xmm5,%xmm5 + vpclmulqdq $0x01,%xmm2,%xmm7,%xmm1 + vpshufd $0x4e,%xmm2,%xmm2 + vpxor %xmm2,%xmm5,%xmm5 + vpxor %xmm1,%xmm5,%xmm5 + + +.Lghash_done: + + vpshufb %xmm6,%xmm5,%xmm5 + vmovdqu %xmm5,(%rdi) + + vzeroupper + RET + +.cfi_endproc +SET_SIZE(gcm_ghash_vpclmulqdq_avx2) +ENTRY_ALIGN(aes_gcm_enc_update_vaes_avx2, 32) +.cfi_startproc + +ENDBR + pushq %r12 +.cfi_adjust_cfa_offset 8 +.cfi_offset %r12,-16 + + movq 16(%rsp),%r12 +#ifdef BORINGSSL_DISPATCH_TEST +.extern BORINGSSL_function_hit +.hidden BORINGSSL_function_hit + movb $1,BORINGSSL_function_hit+6(%rip) +#endif + vbroadcasti128 .Lbswap_mask(%rip),%ymm0 + + + + vmovdqu (%r12),%xmm1 + vpshufb %xmm0,%xmm1,%xmm1 + vbroadcasti128 (%r8),%ymm11 + vpshufb %ymm0,%ymm11,%ymm11 + + + + movl 504(%rcx),%r10d // ICP has a larger offset for rounds. + leal -24(,%r10,4),%r10d // ICP uses 10,12,14 not 9,11,13 for rounds. + + + + + leaq 96(%rcx,%r10,4),%r11 + vbroadcasti128 (%rcx),%ymm9 + vbroadcasti128 (%r11),%ymm10 + + + vpaddd .Lctr_pattern(%rip),%ymm11,%ymm11 + + + + cmpq $127,%rdx + jbe .Lcrypt_loop_4x_done__func1 + + vmovdqu 128(%r9),%ymm7 + vmovdqu 128+32(%r9),%ymm8 + + + + vmovdqu .Linc_2blocks(%rip),%ymm2 + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm14 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm15 + vpaddd %ymm2,%ymm11,%ymm11 + + + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + vpxor %ymm9,%ymm14,%ymm14 + vpxor %ymm9,%ymm15,%ymm15 + + leaq 16(%rcx),%rax +.Lvaesenc_loop_first_4_vecs__func1: + vbroadcasti128 (%rax),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + addq $16,%rax + cmpq %rax,%r11 + jne .Lvaesenc_loop_first_4_vecs__func1 + vpxor 0(%rdi),%ymm10,%ymm2 + vpxor 32(%rdi),%ymm10,%ymm3 + vpxor 64(%rdi),%ymm10,%ymm5 + vpxor 96(%rdi),%ymm10,%ymm6 + vaesenclast %ymm2,%ymm12,%ymm12 + vaesenclast %ymm3,%ymm13,%ymm13 + vaesenclast %ymm5,%ymm14,%ymm14 + vaesenclast %ymm6,%ymm15,%ymm15 + vmovdqu %ymm12,0(%rsi) + vmovdqu %ymm13,32(%rsi) + vmovdqu %ymm14,64(%rsi) + vmovdqu %ymm15,96(%rsi) + + subq $-128,%rdi + addq $-128,%rdx + cmpq $127,%rdx + jbe .Lghash_last_ciphertext_4x__func1 +.balign 16 +.Lcrypt_loop_4x__func1: + + + + + vmovdqu .Linc_2blocks(%rip),%ymm2 + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm14 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm15 + vpaddd %ymm2,%ymm11,%ymm11 + + + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + vpxor %ymm9,%ymm14,%ymm14 + vpxor %ymm9,%ymm15,%ymm15 + + cmpl $24,%r10d + jl .Laes128__func1 + je .Laes192__func1 + + vbroadcasti128 -208(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vbroadcasti128 -192(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + +.Laes192__func1: + vbroadcasti128 -176(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vbroadcasti128 -160(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + +.Laes128__func1: + prefetcht0 512(%rdi) + prefetcht0 512+64(%rdi) + + vmovdqu 0(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 0(%r9),%ymm4 + vpxor %ymm1,%ymm3,%ymm3 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x00,%ymm7,%ymm2,%ymm6 + + vbroadcasti128 -144(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vbroadcasti128 -128(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vmovdqu 32(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 32(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x10,%ymm7,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + vbroadcasti128 -112(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vmovdqu 64(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 64(%r9),%ymm4 + + vbroadcasti128 -96(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + + vbroadcasti128 -80(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x00,%ymm8,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + + vmovdqu 96(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + + vbroadcasti128 -64(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vmovdqu 96(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x10,%ymm8,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + vbroadcasti128 -48(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm1,%ymm6,%ymm6 + + + vbroadcasti128 .Lgfpoly(%rip),%ymm4 + vpclmulqdq $0x01,%ymm5,%ymm4,%ymm2 + vpshufd $0x4e,%ymm5,%ymm5 + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm2,%ymm6,%ymm6 + + vbroadcasti128 -32(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vpclmulqdq $0x01,%ymm6,%ymm4,%ymm2 + vpshufd $0x4e,%ymm6,%ymm6 + vpxor %ymm6,%ymm1,%ymm1 + vpxor %ymm2,%ymm1,%ymm1 + + vbroadcasti128 -16(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vextracti128 $1,%ymm1,%xmm2 + vpxor %xmm2,%xmm1,%xmm1 + + + subq $-128,%rsi + vpxor 0(%rdi),%ymm10,%ymm2 + vpxor 32(%rdi),%ymm10,%ymm3 + vpxor 64(%rdi),%ymm10,%ymm5 + vpxor 96(%rdi),%ymm10,%ymm6 + vaesenclast %ymm2,%ymm12,%ymm12 + vaesenclast %ymm3,%ymm13,%ymm13 + vaesenclast %ymm5,%ymm14,%ymm14 + vaesenclast %ymm6,%ymm15,%ymm15 + vmovdqu %ymm12,0(%rsi) + vmovdqu %ymm13,32(%rsi) + vmovdqu %ymm14,64(%rsi) + vmovdqu %ymm15,96(%rsi) + + subq $-128,%rdi + + addq $-128,%rdx + cmpq $127,%rdx + ja .Lcrypt_loop_4x__func1 +.Lghash_last_ciphertext_4x__func1: + + vmovdqu 0(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 0(%r9),%ymm4 + vpxor %ymm1,%ymm3,%ymm3 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x00,%ymm7,%ymm2,%ymm6 + + vmovdqu 32(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 32(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x10,%ymm7,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + vmovdqu 64(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 64(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x00,%ymm8,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + + vmovdqu 96(%rsi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 96(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x10,%ymm8,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm1,%ymm6,%ymm6 + + + vbroadcasti128 .Lgfpoly(%rip),%ymm4 + vpclmulqdq $0x01,%ymm5,%ymm4,%ymm2 + vpshufd $0x4e,%ymm5,%ymm5 + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm2,%ymm6,%ymm6 + + vpclmulqdq $0x01,%ymm6,%ymm4,%ymm2 + vpshufd $0x4e,%ymm6,%ymm6 + vpxor %ymm6,%ymm1,%ymm1 + vpxor %ymm2,%ymm1,%ymm1 + vextracti128 $1,%ymm1,%xmm2 + vpxor %xmm2,%xmm1,%xmm1 + + subq $-128,%rsi +.Lcrypt_loop_4x_done__func1: + + testq %rdx,%rdx + jz .Ldone__func1 + + + + + + leaq 128(%r9),%r8 + subq %rdx,%r8 + + + vpxor %xmm5,%xmm5,%xmm5 + vpxor %xmm6,%xmm6,%xmm6 + vpxor %xmm7,%xmm7,%xmm7 + + cmpq $64,%rdx + jb .Llessthan64bytes__func1 + + + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd .Linc_2blocks(%rip),%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpaddd .Linc_2blocks(%rip),%ymm11,%ymm11 + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + leaq 16(%rcx),%rax +.Lvaesenc_loop_tail_1__func1: + vbroadcasti128 (%rax),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + addq $16,%rax + cmpq %rax,%r11 + jne .Lvaesenc_loop_tail_1__func1 + vaesenclast %ymm10,%ymm12,%ymm12 + vaesenclast %ymm10,%ymm13,%ymm13 + + + vmovdqu 0(%rdi),%ymm2 + vmovdqu 32(%rdi),%ymm3 + vpxor %ymm2,%ymm12,%ymm12 + vpxor %ymm3,%ymm13,%ymm13 + vmovdqu %ymm12,0(%rsi) + vmovdqu %ymm13,32(%rsi) + + + vpshufb %ymm0,%ymm12,%ymm12 + vpshufb %ymm0,%ymm13,%ymm13 + vpxor %ymm1,%ymm12,%ymm12 + vmovdqu (%r8),%ymm2 + vmovdqu 32(%r8),%ymm3 + vpclmulqdq $0x00,%ymm2,%ymm12,%ymm5 + vpclmulqdq $0x01,%ymm2,%ymm12,%ymm6 + vpclmulqdq $0x10,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%ymm2,%ymm12,%ymm7 + vpclmulqdq $0x00,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm5,%ymm5 + vpclmulqdq $0x01,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x10,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm7,%ymm7 + + addq $64,%r8 + addq $64,%rdi + addq $64,%rsi + subq $64,%rdx + jz .Lreduce__func1 + + vpxor %xmm1,%xmm1,%xmm1 + + +.Llessthan64bytes__func1: + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd .Linc_2blocks(%rip),%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + leaq 16(%rcx),%rax +.Lvaesenc_loop_tail_2__func1: + vbroadcasti128 (%rax),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + addq $16,%rax + cmpq %rax,%r11 + jne .Lvaesenc_loop_tail_2__func1 + vaesenclast %ymm10,%ymm12,%ymm12 + vaesenclast %ymm10,%ymm13,%ymm13 + + + + + cmpq $32,%rdx + jb .Lxor_one_block__func1 + je .Lxor_two_blocks__func1 + +.Lxor_three_blocks__func1: + vmovdqu 0(%rdi),%ymm2 + vmovdqu 32(%rdi),%xmm3 + vpxor %ymm2,%ymm12,%ymm12 + vpxor %xmm3,%xmm13,%xmm13 + vmovdqu %ymm12,0(%rsi) + vmovdqu %xmm13,32(%rsi) + + vpshufb %ymm0,%ymm12,%ymm12 + vpshufb %xmm0,%xmm13,%xmm13 + vpxor %ymm1,%ymm12,%ymm12 + vmovdqu (%r8),%ymm2 + vmovdqu 32(%r8),%xmm3 + vpclmulqdq $0x00,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm5,%ymm5 + vpclmulqdq $0x01,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x10,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm7,%ymm7 + jmp .Lghash_mul_one_vec_unreduced__func1 + +.Lxor_two_blocks__func1: + vmovdqu (%rdi),%ymm2 + vpxor %ymm2,%ymm12,%ymm12 + vmovdqu %ymm12,(%rsi) + vpshufb %ymm0,%ymm12,%ymm12 + vpxor %ymm1,%ymm12,%ymm12 + vmovdqu (%r8),%ymm2 + jmp .Lghash_mul_one_vec_unreduced__func1 + +.Lxor_one_block__func1: + vmovdqu (%rdi),%xmm2 + vpxor %xmm2,%xmm12,%xmm12 + vmovdqu %xmm12,(%rsi) + vpshufb %xmm0,%xmm12,%xmm12 + vpxor %xmm1,%xmm12,%xmm12 + vmovdqu (%r8),%xmm2 + +.Lghash_mul_one_vec_unreduced__func1: + vpclmulqdq $0x00,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm5,%ymm5 + vpclmulqdq $0x01,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x10,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm7,%ymm7 + +.Lreduce__func1: + + vbroadcasti128 .Lgfpoly(%rip),%ymm2 + vpclmulqdq $0x01,%ymm5,%ymm2,%ymm3 + vpshufd $0x4e,%ymm5,%ymm5 + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm3,%ymm6,%ymm6 + vpclmulqdq $0x01,%ymm6,%ymm2,%ymm3 + vpshufd $0x4e,%ymm6,%ymm6 + vpxor %ymm6,%ymm7,%ymm7 + vpxor %ymm3,%ymm7,%ymm7 + vextracti128 $1,%ymm7,%xmm1 + vpxor %xmm7,%xmm1,%xmm1 + +.Ldone__func1: + + vpshufb %xmm0,%xmm1,%xmm1 + vmovdqu %xmm1,(%r12) + + vzeroupper + popq %r12 +.cfi_adjust_cfa_offset -8 +.cfi_restore %r12 + RET + +.cfi_endproc +SET_SIZE(aes_gcm_enc_update_vaes_avx2) +ENTRY_ALIGN(aes_gcm_dec_update_vaes_avx2, 32) +.cfi_startproc + +ENDBR + pushq %r12 +.cfi_adjust_cfa_offset 8 +.cfi_offset %r12,-16 + + movq 16(%rsp),%r12 + vbroadcasti128 .Lbswap_mask(%rip),%ymm0 + + + + vmovdqu (%r12),%xmm1 + vpshufb %xmm0,%xmm1,%xmm1 + vbroadcasti128 (%r8),%ymm11 + vpshufb %ymm0,%ymm11,%ymm11 + + + + movl 504(%rcx),%r10d // ICP has a larger offset for rounds. + leal -24(,%r10,4),%r10d // ICP uses 10,12,14 not 9,11,13 for rounds. + + + + + leaq 96(%rcx,%r10,4),%r11 + vbroadcasti128 (%rcx),%ymm9 + vbroadcasti128 (%r11),%ymm10 + + + vpaddd .Lctr_pattern(%rip),%ymm11,%ymm11 + + + + cmpq $127,%rdx + jbe .Lcrypt_loop_4x_done__func2 + + vmovdqu 128(%r9),%ymm7 + vmovdqu 128+32(%r9),%ymm8 +.balign 16 +.Lcrypt_loop_4x__func2: + + + + + vmovdqu .Linc_2blocks(%rip),%ymm2 + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm14 + vpaddd %ymm2,%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm15 + vpaddd %ymm2,%ymm11,%ymm11 + + + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + vpxor %ymm9,%ymm14,%ymm14 + vpxor %ymm9,%ymm15,%ymm15 + + cmpl $24,%r10d + jl .Laes128__func2 + je .Laes192__func2 + + vbroadcasti128 -208(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vbroadcasti128 -192(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + +.Laes192__func2: + vbroadcasti128 -176(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vbroadcasti128 -160(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + +.Laes128__func2: + prefetcht0 512(%rdi) + prefetcht0 512+64(%rdi) + + vmovdqu 0(%rdi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 0(%r9),%ymm4 + vpxor %ymm1,%ymm3,%ymm3 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x00,%ymm7,%ymm2,%ymm6 + + vbroadcasti128 -144(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vbroadcasti128 -128(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vmovdqu 32(%rdi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 32(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x10,%ymm7,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + vbroadcasti128 -112(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vmovdqu 64(%rdi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + vmovdqu 64(%r9),%ymm4 + + vbroadcasti128 -96(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + + vbroadcasti128 -80(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x00,%ymm8,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + + vmovdqu 96(%rdi),%ymm3 + vpshufb %ymm0,%ymm3,%ymm3 + + vbroadcasti128 -64(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vmovdqu 96(%r9),%ymm4 + vpclmulqdq $0x00,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm5,%ymm5 + vpclmulqdq $0x11,%ymm4,%ymm3,%ymm2 + vpxor %ymm2,%ymm1,%ymm1 + vpunpckhqdq %ymm3,%ymm3,%ymm2 + vpxor %ymm3,%ymm2,%ymm2 + vpclmulqdq $0x10,%ymm8,%ymm2,%ymm2 + vpxor %ymm2,%ymm6,%ymm6 + + vbroadcasti128 -48(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm1,%ymm6,%ymm6 + + + vbroadcasti128 .Lgfpoly(%rip),%ymm4 + vpclmulqdq $0x01,%ymm5,%ymm4,%ymm2 + vpshufd $0x4e,%ymm5,%ymm5 + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm2,%ymm6,%ymm6 + + vbroadcasti128 -32(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + + vpclmulqdq $0x01,%ymm6,%ymm4,%ymm2 + vpshufd $0x4e,%ymm6,%ymm6 + vpxor %ymm6,%ymm1,%ymm1 + vpxor %ymm2,%ymm1,%ymm1 + + vbroadcasti128 -16(%r11),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + vaesenc %ymm2,%ymm14,%ymm14 + vaesenc %ymm2,%ymm15,%ymm15 + + vextracti128 $1,%ymm1,%xmm2 + vpxor %xmm2,%xmm1,%xmm1 + + + + vpxor 0(%rdi),%ymm10,%ymm2 + vpxor 32(%rdi),%ymm10,%ymm3 + vpxor 64(%rdi),%ymm10,%ymm5 + vpxor 96(%rdi),%ymm10,%ymm6 + vaesenclast %ymm2,%ymm12,%ymm12 + vaesenclast %ymm3,%ymm13,%ymm13 + vaesenclast %ymm5,%ymm14,%ymm14 + vaesenclast %ymm6,%ymm15,%ymm15 + vmovdqu %ymm12,0(%rsi) + vmovdqu %ymm13,32(%rsi) + vmovdqu %ymm14,64(%rsi) + vmovdqu %ymm15,96(%rsi) + + subq $-128,%rdi + subq $-128,%rsi + addq $-128,%rdx + cmpq $127,%rdx + ja .Lcrypt_loop_4x__func2 +.Lcrypt_loop_4x_done__func2: + + testq %rdx,%rdx + jz .Ldone__func2 + + + + + + leaq 128(%r9),%r8 + subq %rdx,%r8 + + + vpxor %xmm5,%xmm5,%xmm5 + vpxor %xmm6,%xmm6,%xmm6 + vpxor %xmm7,%xmm7,%xmm7 + + cmpq $64,%rdx + jb .Llessthan64bytes__func2 + + + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd .Linc_2blocks(%rip),%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpaddd .Linc_2blocks(%rip),%ymm11,%ymm11 + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + leaq 16(%rcx),%rax +.Lvaesenc_loop_tail_1__func2: + vbroadcasti128 (%rax),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + addq $16,%rax + cmpq %rax,%r11 + jne .Lvaesenc_loop_tail_1__func2 + vaesenclast %ymm10,%ymm12,%ymm12 + vaesenclast %ymm10,%ymm13,%ymm13 + + + vmovdqu 0(%rdi),%ymm2 + vmovdqu 32(%rdi),%ymm3 + vpxor %ymm2,%ymm12,%ymm12 + vpxor %ymm3,%ymm13,%ymm13 + vmovdqu %ymm12,0(%rsi) + vmovdqu %ymm13,32(%rsi) + + + vpshufb %ymm0,%ymm2,%ymm12 + vpshufb %ymm0,%ymm3,%ymm13 + vpxor %ymm1,%ymm12,%ymm12 + vmovdqu (%r8),%ymm2 + vmovdqu 32(%r8),%ymm3 + vpclmulqdq $0x00,%ymm2,%ymm12,%ymm5 + vpclmulqdq $0x01,%ymm2,%ymm12,%ymm6 + vpclmulqdq $0x10,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%ymm2,%ymm12,%ymm7 + vpclmulqdq $0x00,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm5,%ymm5 + vpclmulqdq $0x01,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x10,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%ymm3,%ymm13,%ymm4 + vpxor %ymm4,%ymm7,%ymm7 + + addq $64,%r8 + addq $64,%rdi + addq $64,%rsi + subq $64,%rdx + jz .Lreduce__func2 + + vpxor %xmm1,%xmm1,%xmm1 + + +.Llessthan64bytes__func2: + vpshufb %ymm0,%ymm11,%ymm12 + vpaddd .Linc_2blocks(%rip),%ymm11,%ymm11 + vpshufb %ymm0,%ymm11,%ymm13 + vpxor %ymm9,%ymm12,%ymm12 + vpxor %ymm9,%ymm13,%ymm13 + leaq 16(%rcx),%rax +.Lvaesenc_loop_tail_2__func2: + vbroadcasti128 (%rax),%ymm2 + vaesenc %ymm2,%ymm12,%ymm12 + vaesenc %ymm2,%ymm13,%ymm13 + addq $16,%rax + cmpq %rax,%r11 + jne .Lvaesenc_loop_tail_2__func2 + vaesenclast %ymm10,%ymm12,%ymm12 + vaesenclast %ymm10,%ymm13,%ymm13 + + + + + cmpq $32,%rdx + jb .Lxor_one_block__func2 + je .Lxor_two_blocks__func2 + +.Lxor_three_blocks__func2: + vmovdqu 0(%rdi),%ymm2 + vmovdqu 32(%rdi),%xmm3 + vpxor %ymm2,%ymm12,%ymm12 + vpxor %xmm3,%xmm13,%xmm13 + vmovdqu %ymm12,0(%rsi) + vmovdqu %xmm13,32(%rsi) + + vpshufb %ymm0,%ymm2,%ymm12 + vpshufb %xmm0,%xmm3,%xmm13 + vpxor %ymm1,%ymm12,%ymm12 + vmovdqu (%r8),%ymm2 + vmovdqu 32(%r8),%xmm3 + vpclmulqdq $0x00,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm5,%ymm5 + vpclmulqdq $0x01,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x10,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%xmm3,%xmm13,%xmm4 + vpxor %ymm4,%ymm7,%ymm7 + jmp .Lghash_mul_one_vec_unreduced__func2 + +.Lxor_two_blocks__func2: + vmovdqu (%rdi),%ymm2 + vpxor %ymm2,%ymm12,%ymm12 + vmovdqu %ymm12,(%rsi) + vpshufb %ymm0,%ymm2,%ymm12 + vpxor %ymm1,%ymm12,%ymm12 + vmovdqu (%r8),%ymm2 + jmp .Lghash_mul_one_vec_unreduced__func2 + +.Lxor_one_block__func2: + vmovdqu (%rdi),%xmm2 + vpxor %xmm2,%xmm12,%xmm12 + vmovdqu %xmm12,(%rsi) + vpshufb %xmm0,%xmm2,%xmm12 + vpxor %xmm1,%xmm12,%xmm12 + vmovdqu (%r8),%xmm2 + +.Lghash_mul_one_vec_unreduced__func2: + vpclmulqdq $0x00,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm5,%ymm5 + vpclmulqdq $0x01,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x10,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm6,%ymm6 + vpclmulqdq $0x11,%ymm2,%ymm12,%ymm4 + vpxor %ymm4,%ymm7,%ymm7 + +.Lreduce__func2: + + vbroadcasti128 .Lgfpoly(%rip),%ymm2 + vpclmulqdq $0x01,%ymm5,%ymm2,%ymm3 + vpshufd $0x4e,%ymm5,%ymm5 + vpxor %ymm5,%ymm6,%ymm6 + vpxor %ymm3,%ymm6,%ymm6 + vpclmulqdq $0x01,%ymm6,%ymm2,%ymm3 + vpshufd $0x4e,%ymm6,%ymm6 + vpxor %ymm6,%ymm7,%ymm7 + vpxor %ymm3,%ymm7,%ymm7 + vextracti128 $1,%ymm7,%xmm1 + vpxor %xmm7,%xmm1,%xmm1 + +.Ldone__func2: + + vpshufb %xmm0,%xmm1,%xmm1 + vmovdqu %xmm1,(%r12) + + vzeroupper + popq %r12 +.cfi_adjust_cfa_offset -8 +.cfi_restore %r12 + RET + +.cfi_endproc +SET_SIZE(aes_gcm_dec_update_vaes_avx2) + +#endif /* !_WIN32 || _KERNEL */ + +/* Mark the stack non-executable. */ +#if defined(__linux__) && defined(__ELF__) +.section .note.GNU-stack,"",%progbits +#endif + +#endif /* defined(__x86_64__) && defined(HAVE_AVX) && defined(HAVE_AES) ... */ diff --git a/sys/contrib/openzfs/module/icp/include/modes/modes.h b/sys/contrib/openzfs/module/icp/include/modes/modes.h index ca734cf4f045..de11d9eafafb 100644 --- a/sys/contrib/openzfs/module/icp/include/modes/modes.h +++ b/sys/contrib/openzfs/module/icp/include/modes/modes.h @@ -42,7 +42,7 @@ extern "C" { */ #if defined(__x86_64__) && defined(HAVE_AVX) && \ defined(HAVE_AES) && defined(HAVE_PCLMULQDQ) -#define CAN_USE_GCM_ASM +#define CAN_USE_GCM_ASM (HAVE_VAES && HAVE_VPCLMULQDQ ? 2 : 1) extern boolean_t gcm_avx_can_use_movbe; #endif @@ -129,6 +129,15 @@ typedef struct ccm_ctx { #define ccm_copy_to ccm_common.cc_copy_to #define ccm_flags ccm_common.cc_flags +#ifdef CAN_USE_GCM_ASM +typedef enum gcm_impl { + GCM_IMPL_GENERIC = 0, + GCM_IMPL_AVX, + GCM_IMPL_AVX2, + GCM_IMPL_MAX, +} gcm_impl; +#endif + /* * gcm_tag_len: Length of authentication tag. * @@ -174,7 +183,7 @@ typedef struct gcm_ctx { uint64_t gcm_len_a_len_c[2]; uint8_t *gcm_pt_buf; #ifdef CAN_USE_GCM_ASM - boolean_t gcm_use_avx; + enum gcm_impl impl; #endif } gcm_ctx_t; diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c b/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c index c114db14a916..b218c0da8125 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/kmod_core.c @@ -112,7 +112,6 @@ static int zfs__fini(void); static void zfs_shutdown(void *, int); static eventhandler_tag zfs_shutdown_event_tag; -static eventhandler_tag zfs_mountroot_event_tag; #define ZFS_MIN_KSTACK_PAGES 4 @@ -311,9 +310,6 @@ zfs_modevent(module_t mod, int type, void *unused __unused) zfs_shutdown_event_tag = EVENTHANDLER_REGISTER( shutdown_post_sync, zfs_shutdown, NULL, SHUTDOWN_PRI_FIRST); - zfs_mountroot_event_tag = EVENTHANDLER_REGISTER( - mountroot, spa_boot_init, NULL, - SI_ORDER_ANY); } return (err); case MOD_UNLOAD: @@ -322,9 +318,6 @@ zfs_modevent(module_t mod, int type, void *unused __unused) if (zfs_shutdown_event_tag != NULL) EVENTHANDLER_DEREGISTER(shutdown_post_sync, zfs_shutdown_event_tag); - if (zfs_mountroot_event_tag != NULL) - EVENTHANDLER_DEREGISTER(mountroot, - zfs_mountroot_event_tag); } return (err); case MOD_SHUTDOWN: diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c index ace2360c032d..393bfaa65ff5 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/sysctl_os.c @@ -188,11 +188,6 @@ param_set_arc_max(SYSCTL_HANDLER_ARGS) return (0); } -SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_max, - CTLTYPE_ULONG | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_max, "LU", - "Maximum ARC size in bytes (LEGACY)"); - int param_set_arc_min(SYSCTL_HANDLER_ARGS) { @@ -217,11 +212,6 @@ param_set_arc_min(SYSCTL_HANDLER_ARGS) return (0); } -SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_min, - CTLTYPE_ULONG | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_min, "LU", - "Minimum ARC size in bytes (LEGACY)"); - extern uint_t zfs_arc_free_target; int @@ -245,16 +235,6 @@ param_set_arc_free_target(SYSCTL_HANDLER_ARGS) return (0); } -/* - * NOTE: This sysctl is CTLFLAG_RW not CTLFLAG_RWTUN due to its dependency on - * pagedaemon initialization. - */ -SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_free_target, - CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_free_target, "IU", - "Desired number of free pages below which ARC triggers reclaim" - " (LEGACY)"); - int param_set_arc_no_grow_shift(SYSCTL_HANDLER_ARGS) { @@ -273,187 +253,6 @@ param_set_arc_no_grow_shift(SYSCTL_HANDLER_ARGS) return (0); } -SYSCTL_PROC(_vfs_zfs, OID_AUTO, arc_no_grow_shift, - CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - NULL, 0, param_set_arc_no_grow_shift, "I", - "log2(fraction of ARC which must be free to allow growing) (LEGACY)"); - -extern uint64_t l2arc_write_max; - -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_write_max, - CTLFLAG_RWTUN, &l2arc_write_max, 0, - "Max write bytes per interval (LEGACY)"); - -extern uint64_t l2arc_write_boost; - -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_write_boost, - CTLFLAG_RWTUN, &l2arc_write_boost, 0, - "Extra write bytes during device warmup (LEGACY)"); - -extern uint64_t l2arc_headroom; - -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_headroom, - CTLFLAG_RWTUN, &l2arc_headroom, 0, - "Number of max device writes to precache (LEGACY)"); - -extern uint64_t l2arc_headroom_boost; - -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_headroom_boost, - CTLFLAG_RWTUN, &l2arc_headroom_boost, 0, - "Compressed l2arc_headroom multiplier (LEGACY)"); - -extern uint64_t l2arc_feed_secs; - -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_feed_secs, - CTLFLAG_RWTUN, &l2arc_feed_secs, 0, - "Seconds between L2ARC writing (LEGACY)"); - -extern uint64_t l2arc_feed_min_ms; - -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, l2arc_feed_min_ms, - CTLFLAG_RWTUN, &l2arc_feed_min_ms, 0, - "Min feed interval in milliseconds (LEGACY)"); - -extern int l2arc_noprefetch; - -SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_noprefetch, - CTLFLAG_RWTUN, &l2arc_noprefetch, 0, - "Skip caching prefetched buffers (LEGACY)"); - -extern int l2arc_feed_again; - -SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_feed_again, - CTLFLAG_RWTUN, &l2arc_feed_again, 0, - "Turbo L2ARC warmup (LEGACY)"); - -extern int l2arc_norw; - -SYSCTL_INT(_vfs_zfs, OID_AUTO, l2arc_norw, - CTLFLAG_RWTUN, &l2arc_norw, 0, - "No reads during writes (LEGACY)"); - -static int -param_get_arc_state_size(SYSCTL_HANDLER_ARGS) -{ - arc_state_t *state = (arc_state_t *)arg1; - int64_t val; - - val = zfs_refcount_count(&state->arcs_size[ARC_BUFC_DATA]) + - zfs_refcount_count(&state->arcs_size[ARC_BUFC_METADATA]); - return (sysctl_handle_64(oidp, &val, 0, req)); -} - -extern arc_state_t ARC_anon; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, anon_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_anon, 0, param_get_arc_state_size, "Q", - "size of anonymous state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_metadata_esize, CTLFLAG_RD, - &ARC_anon.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, - "size of evictable metadata in anonymous state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, anon_data_esize, CTLFLAG_RD, - &ARC_anon.arcs_esize[ARC_BUFC_DATA].rc_count, 0, - "size of evictable data in anonymous state"); - -extern arc_state_t ARC_mru; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, mru_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_mru, 0, param_get_arc_state_size, "Q", - "size of mru state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_metadata_esize, CTLFLAG_RD, - &ARC_mru.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, - "size of evictable metadata in mru state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_data_esize, CTLFLAG_RD, - &ARC_mru.arcs_esize[ARC_BUFC_DATA].rc_count, 0, - "size of evictable data in mru state"); - -extern arc_state_t ARC_mru_ghost; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, mru_ghost_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_mru_ghost, 0, param_get_arc_state_size, "Q", - "size of mru ghost state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_metadata_esize, CTLFLAG_RD, - &ARC_mru_ghost.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, - "size of evictable metadata in mru ghost state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mru_ghost_data_esize, CTLFLAG_RD, - &ARC_mru_ghost.arcs_esize[ARC_BUFC_DATA].rc_count, 0, - "size of evictable data in mru ghost state"); - -extern arc_state_t ARC_mfu; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, mfu_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_mfu, 0, param_get_arc_state_size, "Q", - "size of mfu state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_metadata_esize, CTLFLAG_RD, - &ARC_mfu.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, - "size of evictable metadata in mfu state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_data_esize, CTLFLAG_RD, - &ARC_mfu.arcs_esize[ARC_BUFC_DATA].rc_count, 0, - "size of evictable data in mfu state"); - -extern arc_state_t ARC_mfu_ghost; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, mfu_ghost_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_mfu_ghost, 0, param_get_arc_state_size, "Q", - "size of mfu ghost state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_metadata_esize, CTLFLAG_RD, - &ARC_mfu_ghost.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, - "size of evictable metadata in mfu ghost state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, mfu_ghost_data_esize, CTLFLAG_RD, - &ARC_mfu_ghost.arcs_esize[ARC_BUFC_DATA].rc_count, 0, - "size of evictable data in mfu ghost state"); - -extern arc_state_t ARC_uncached; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, uncached_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_uncached, 0, param_get_arc_state_size, "Q", - "size of uncached state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, uncached_metadata_esize, CTLFLAG_RD, - &ARC_uncached.arcs_esize[ARC_BUFC_METADATA].rc_count, 0, - "size of evictable metadata in uncached state"); -SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, uncached_data_esize, CTLFLAG_RD, - &ARC_uncached.arcs_esize[ARC_BUFC_DATA].rc_count, 0, - "size of evictable data in uncached state"); - -extern arc_state_t ARC_l2c_only; - -SYSCTL_PROC(_vfs_zfs, OID_AUTO, l2c_only_size, - CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, - &ARC_l2c_only, 0, param_get_arc_state_size, "Q", - "size of l2c_only state"); - -/* dbuf.c */ - -/* dmu.c */ - -/* dmu_zfetch.c */ - -SYSCTL_NODE(_vfs_zfs, OID_AUTO, zfetch, CTLFLAG_RW, 0, "ZFS ZFETCH (LEGACY)"); - -extern uint32_t zfetch_max_distance; - -SYSCTL_UINT(_vfs_zfs_zfetch, OID_AUTO, max_distance, - CTLFLAG_RWTUN, &zfetch_max_distance, 0, - "Max bytes to prefetch per stream (LEGACY)"); - -extern uint32_t zfetch_max_idistance; - -SYSCTL_UINT(_vfs_zfs_zfetch, OID_AUTO, max_idistance, - CTLFLAG_RWTUN, &zfetch_max_idistance, 0, - "Max bytes to prefetch indirects for per stream (LEGACY)"); - -/* dsl_pool.c */ - -/* dnode.c */ - -/* dsl_scan.c */ - /* metaslab.c */ int @@ -514,19 +313,6 @@ SYSCTL_UINT(_vfs_zfs, OID_AUTO, condense_pct, "Condense on-disk spacemap when it is more than this many percents" " of in-memory counterpart"); -extern uint_t zfs_remove_max_segment; - -SYSCTL_UINT(_vfs_zfs, OID_AUTO, remove_max_segment, - CTLFLAG_RWTUN, &zfs_remove_max_segment, 0, - "Largest contiguous segment ZFS will attempt to allocate when removing" - " a device"); - -extern int zfs_removal_suspend_progress; - -SYSCTL_INT(_vfs_zfs, OID_AUTO, removal_suspend_progress, - CTLFLAG_RWTUN, &zfs_removal_suspend_progress, 0, - "Ensures certain actions can happen while in the middle of a removal"); - /* * Minimum size which forces the dynamic allocator to change * it's allocation strategy. Once the space map cannot satisfy @@ -749,12 +535,6 @@ param_set_min_auto_ashift(SYSCTL_HANDLER_ARGS) return (0); } -SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift, - CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - &zfs_vdev_min_auto_ashift, sizeof (zfs_vdev_min_auto_ashift), - param_set_min_auto_ashift, "IU", - "Min ashift used when creating new top-level vdev. (LEGACY)"); - int param_set_max_auto_ashift(SYSCTL_HANDLER_ARGS) { @@ -774,13 +554,6 @@ param_set_max_auto_ashift(SYSCTL_HANDLER_ARGS) return (0); } -SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift, - CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, - &zfs_vdev_max_auto_ashift, sizeof (zfs_vdev_max_auto_ashift), - param_set_max_auto_ashift, "IU", - "Max ashift used when optimizing for logical -> physical sector size on" - " new top-level vdevs. (LEGACY)"); - /* * Since the DTL space map of a vdev is not expected to have a lot of * entries, we default its block size to 4K. @@ -802,23 +575,6 @@ SYSCTL_INT(_vfs_zfs, OID_AUTO, standard_sm_blksz, CTLFLAG_RDTUN, &zfs_vdev_standard_sm_blksz, 0, "Block size for standard space map. Power of 2 greater than 4096."); -extern int vdev_validate_skip; - -SYSCTL_INT(_vfs_zfs, OID_AUTO, validate_skip, - CTLFLAG_RDTUN, &vdev_validate_skip, 0, - "Enable to bypass vdev_validate()."); - -/* vdev_mirror.c */ - -/* vdev_queue.c */ - -extern uint_t zfs_vdev_max_active; - -SYSCTL_UINT(_vfs_zfs, OID_AUTO, top_maxinflight, - CTLFLAG_RWTUN, &zfs_vdev_max_active, 0, - "The maximum number of I/Os of all types active for each device." - " (LEGACY)"); - /* zio.c */ SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, exclude_metadata, diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c index b15a3e6e38c0..cb5787269db2 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c @@ -1175,7 +1175,7 @@ zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx) int count = 0; zfs_acl_phys_t acl_phys; - if (zp->z_zfsvfs->z_replay == B_FALSE) { + if (ZTOV(zp) != NULL && zp->z_zfsvfs->z_replay == B_FALSE) { ASSERT_VOP_IN_SEQC(ZTOV(zp)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c index 61d0bb26d1e5..a222c5de4a2a 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_ctldir.c @@ -494,7 +494,7 @@ zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) vap->va_uid = 0; vap->va_gid = 0; - vap->va_rdev = 0; + vap->va_rdev = NODEV; /* * We are a purely virtual object, so we have no * blocksize or allocated blocks. @@ -674,6 +674,7 @@ zfsctl_root_readdir(struct vop_readdir_args *ap) zfs_uio_t uio; int *eofp = ap->a_eofflag; off_t dots_offset; + ssize_t orig_resid; int error; zfs_uio_init(&uio, ap->a_uio); @@ -688,16 +689,16 @@ zfsctl_root_readdir(struct vop_readdir_args *ap) * count to return is 0. */ if (zfs_uio_offset(&uio) == 3 * sizeof (entry)) { + if (eofp != NULL) + *eofp = 1; return (0); } + orig_resid = zfs_uio_resid(&uio); error = sfs_readdir_common(zfsvfs->z_root, ZFSCTL_INO_ROOT, ap, &uio, &dots_offset); - if (error != 0) { - if (error == ENAMETOOLONG) /* ran out of destination space */ - error = 0; - return (error); - } + if (error != 0) + goto err; if (zfs_uio_offset(&uio) != dots_offset) return (SET_ERROR(EINVAL)); @@ -710,8 +711,11 @@ zfsctl_root_readdir(struct vop_readdir_args *ap) entry.d_reclen = sizeof (entry); error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio)); if (error != 0) { - if (error == ENAMETOOLONG) - error = 0; +err: + if (error == ENAMETOOLONG) { + error = orig_resid == zfs_uio_resid(&uio) ? + EINVAL : 0; + } return (SET_ERROR(error)); } if (eofp != NULL) @@ -1056,17 +1060,21 @@ zfsctl_snapdir_readdir(struct vop_readdir_args *ap) zfs_uio_t uio; int *eofp = ap->a_eofflag; off_t dots_offset; + ssize_t orig_resid; int error; zfs_uio_init(&uio, ap->a_uio); + orig_resid = zfs_uio_resid(&uio); ASSERT3S(vp->v_type, ==, VDIR); error = sfs_readdir_common(ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, ap, &uio, &dots_offset); if (error != 0) { - if (error == ENAMETOOLONG) /* ran out of destination space */ - error = 0; + if (error == ENAMETOOLONG) { /* ran out of destination space */ + error = orig_resid == zfs_uio_resid(&uio) ? + EINVAL : 0; + } return (error); } @@ -1084,9 +1092,13 @@ zfsctl_snapdir_readdir(struct vop_readdir_args *ap) dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); if (error != 0) { if (error == ENOENT) { - if (eofp != NULL) - *eofp = 1; - error = 0; + if (orig_resid == zfs_uio_resid(&uio)) { + error = EINVAL; + } else { + error = 0; + if (eofp != NULL) + *eofp = 1; + } } zfs_exit(zfsvfs, FTAG); return (error); @@ -1099,8 +1111,10 @@ zfsctl_snapdir_readdir(struct vop_readdir_args *ap) entry.d_reclen = sizeof (entry); error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio)); if (error != 0) { - if (error == ENAMETOOLONG) - error = 0; + if (error == ENAMETOOLONG) { + error = orig_resid == zfs_uio_resid(&uio) ? + EINVAL : 0; + } zfs_exit(zfsvfs, FTAG); return (SET_ERROR(error)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c index 21e5f7938f9f..ca13569a1235 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_file_os.c @@ -164,8 +164,9 @@ zfs_file_write(zfs_file_t *fp, const void *buf, size_t count, ssize_t *resid) int zfs_file_pwrite(zfs_file_t *fp, const void *buf, size_t count, loff_t off, - ssize_t *resid) + uint8_t ashift, ssize_t *resid) { + (void) ashift; return (zfs_file_write_impl(fp, buf, count, &off, resid)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c index 1813c411b013..8dce97baba66 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c @@ -61,6 +61,7 @@ #include <sys/fs/zfs.h> #include <sys/dmu.h> #include <sys/dmu_objset.h> +#include <sys/dsl_dataset.h> #include <sys/spa.h> #include <sys/txg.h> #include <sys/dbuf.h> @@ -388,7 +389,9 @@ zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, error = vn_lock(vp, LK_EXCLUSIVE); if (error) return (error); + vn_seqc_write_begin(vp); error = zfs_ioctl_setxattr(vp, fsx, cred); + vn_seqc_write_end(vp); VOP_UNLOCK(vp); return (error); } @@ -1695,6 +1698,7 @@ zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, objset_t *os; caddr_t outbuf; size_t bufsize; + ssize_t orig_resid; zap_cursor_t zc; zap_attribute_t *zap; uint_t bytes_wanted; @@ -1735,7 +1739,7 @@ zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, /* * Quit if directory has been removed (posix) */ - if ((*eofp = zp->z_unlinked) != 0) { + if ((*eofp = (zp->z_unlinked != 0)) != 0) { zfs_exit(zfsvfs, FTAG); return (0); } @@ -1743,6 +1747,7 @@ zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, error = 0; os = zfsvfs->z_os; offset = zfs_uio_offset(uio); + orig_resid = zfs_uio_resid(uio); prefetch = zp->z_zn_prefetch; zap = zap_attribute_long_alloc(); @@ -1922,7 +1927,7 @@ update: kmem_free(outbuf, bufsize); if (error == ENOENT) - error = 0; + error = orig_resid == zfs_uio_resid(uio) ? EINVAL : 0; ZFS_ACCESSTIME_STAMP(zfsvfs, zp); @@ -2013,7 +2018,7 @@ zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) if (vp->v_type == VBLK || vp->v_type == VCHR) vap->va_rdev = zfs_cmpldev(rdev); else - vap->va_rdev = 0; + vap->va_rdev = NODEV; vap->va_gen = zp->z_gen; vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ vap->va_filerev = zp->z_seq; @@ -2203,6 +2208,7 @@ zfs_setattr_dir(znode_t *dzp) if (err) break; + vn_seqc_write_begin(ZTOV(zp)); mutex_enter(&dzp->z_lock); if (zp->z_uid != dzp->z_uid) { @@ -2252,6 +2258,7 @@ sa_add_projid_err: dmu_tx_abort(tx); } tx = NULL; + vn_seqc_write_end(ZTOV(zp)); if (err != 0 && err != ENOENT) break; @@ -5727,6 +5734,9 @@ zfs_freebsd_pathconf(struct vop_pathconf_args *ap) { ulong_t val; int error; +#ifdef _PC_CLONE_BLKSIZE + zfsvfs_t *zfsvfs; +#endif error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); @@ -5773,6 +5783,21 @@ zfs_freebsd_pathconf(struct vop_pathconf_args *ap) *ap->a_retval = 1; return (0); #endif +#ifdef _PC_CLONE_BLKSIZE + case _PC_CLONE_BLKSIZE: + zfsvfs = (zfsvfs_t *)ap->a_vp->v_mount->mnt_data; + if (zfs_bclone_enabled && + spa_feature_is_enabled(dmu_objset_spa(zfsvfs->z_os), + SPA_FEATURE_BLOCK_CLONING)) + *ap->a_retval = dsl_dataset_feature_is_active( + zfsvfs->z_os->os_dsl_dataset, + SPA_FEATURE_LARGE_BLOCKS) ? + SPA_MAXBLOCKSIZE : + SPA_OLD_MAXBLOCKSIZE; + else + *ap->a_retval = 0; + return (0); +#endif default: return (vop_stdpathconf(ap)); } diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c index 7cd0a153577c..649022ab5bcb 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_znode_os.c @@ -817,6 +817,10 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, (*zpp)->z_dnodesize = dnodesize; (*zpp)->z_projid = projid; + vnode_t *vp = ZTOV(*zpp); + if (!(flag & IS_ROOT_NODE)) + vn_seqc_write_begin(vp); + if (vap->va_mask & AT_XVATTR) zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); @@ -825,7 +829,7 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); } if (!(flag & IS_ROOT_NODE)) { - vnode_t *vp = ZTOV(*zpp); + vn_seqc_write_end(vp); vp->v_vflag |= VV_FORCEINSMQ; int err = insmntque(vp, zfsvfs->z_vfs); vp->v_vflag &= ~VV_FORCEINSMQ; diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c b/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c index 265dfd55fc4d..0dd2ecd7fd8d 100644 --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zvol_os.c @@ -31,7 +31,7 @@ * Copyright (c) 2012, 2017 by Delphix. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. * Copyright (c) 2014 Integros [integros.com] - * Copyright (c) 2024, Klara, Inc. + * Copyright (c) 2024, 2025, Klara, Inc. */ /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ @@ -196,7 +196,6 @@ DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); static int zvol_geom_open(struct g_provider *pp, int flag, int count); static int zvol_geom_close(struct g_provider *pp, int flag, int count); -static void zvol_geom_destroy(zvol_state_t *zv); static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); static void zvol_geom_bio_start(struct bio *bp); static int zvol_geom_bio_getattr(struct bio *bp); @@ -226,25 +225,14 @@ zvol_geom_open(struct g_provider *pp, int flag, int count) } retry: - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - /* - * Obtain a copy of private under zvol_state_lock to make sure either - * the result of zvol free code setting private to NULL is observed, - * or the zv is protected from being freed because of the positive - * zv_open_count. - */ - zv = pp->private; - if (zv == NULL) { - rw_exit(&zvol_state_lock); - err = SET_ERROR(ENXIO); - goto out_locked; - } + zv = atomic_load_ptr(&pp->private); + if (zv == NULL) + return (SET_ERROR(ENXIO)); mutex_enter(&zv->zv_state_lock); if (zv->zv_zso->zso_dying || zv->zv_flags & ZVOL_REMOVING) { - rw_exit(&zvol_state_lock); err = SET_ERROR(ENXIO); - goto out_zv_locked; + goto out_locked; } ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM); @@ -257,8 +245,24 @@ retry: drop_suspend = B_TRUE; if (!rw_tryenter(&zv->zv_suspend_lock, ZVOL_RW_READER)) { mutex_exit(&zv->zv_state_lock); + + /* + * Removal may happen while the locks are down, so + * we can't trust zv any longer; we have to start over. + */ + zv = atomic_load_ptr(&pp->private); + if (zv == NULL) + return (SET_ERROR(ENXIO)); + rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + if (zv->zv_zso->zso_dying || + zv->zv_flags & ZVOL_REMOVING) { + err = SET_ERROR(ENXIO); + goto out_locked; + } + /* Check to see if zv_suspend_lock is needed. */ if (zv->zv_open_count != 0) { rw_exit(&zv->zv_suspend_lock); @@ -266,7 +270,6 @@ retry: } } } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -294,7 +297,7 @@ retry: if (drop_namespace) mutex_exit(&spa_namespace_lock); if (err) - goto out_zv_locked; + goto out_locked; pp->mediasize = zv->zv_volsize; pp->stripeoffset = 0; pp->stripesize = zv->zv_volblocksize; @@ -329,9 +332,8 @@ out_opened: zvol_last_close(zv); wakeup(zv); } -out_zv_locked: - mutex_exit(&zv->zv_state_lock); out_locked: + mutex_exit(&zv->zv_state_lock); if (drop_suspend) rw_exit(&zv->zv_suspend_lock); return (err); @@ -345,12 +347,9 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) boolean_t drop_suspend = B_TRUE; int new_open_count; - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - zv = pp->private; - if (zv == NULL) { - rw_exit(&zvol_state_lock); + zv = atomic_load_ptr(&pp->private); + if (zv == NULL) return (SET_ERROR(ENXIO)); - } mutex_enter(&zv->zv_state_lock); if (zv->zv_flags & ZVOL_EXCL) { @@ -377,6 +376,15 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + /* + * Unlike in zvol_geom_open(), we don't check if + * removal started here, because we might be one of the + * openers that needs to be thrown out! If we're the + * last, we need to call zvol_last_close() below to + * finish cleanup. So, no special treatment for us. + */ + /* Check to see if zv_suspend_lock is needed. */ new_open_count = zv->zv_open_count - count; if (new_open_count != 0) { @@ -387,7 +395,6 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) } else { drop_suspend = B_FALSE; } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -408,20 +415,6 @@ zvol_geom_close(struct g_provider *pp, int flag, int count) return (0); } -static void -zvol_geom_destroy(zvol_state_t *zv) -{ - struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom; - struct g_provider *pp = zsg->zsg_provider; - - ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_GEOM); - - g_topology_assert(); - - zsg->zsg_provider = NULL; - g_wither_geom(pp->geom, ENXIO); -} - void zvol_wait_close(zvol_state_t *zv) { @@ -454,7 +447,7 @@ zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", pp->name, acr, acw, ace)); - if (pp->private == NULL) { + if (atomic_load_ptr(&pp->private) == NULL) { if (acr <= 0 && acw <= 0 && ace <= 0) return (0); return (pp->error); @@ -921,25 +914,14 @@ zvol_cdev_open(struct cdev *dev, int flags, int fmt, struct thread *td) boolean_t drop_suspend = B_FALSE; retry: - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - /* - * Obtain a copy of si_drv2 under zvol_state_lock to make sure either - * the result of zvol free code setting si_drv2 to NULL is observed, - * or the zv is protected from being freed because of the positive - * zv_open_count. - */ - zv = dev->si_drv2; - if (zv == NULL) { - rw_exit(&zvol_state_lock); - err = SET_ERROR(ENXIO); - goto out_locked; - } + zv = atomic_load_ptr(&dev->si_drv2); + if (zv == NULL) + return (SET_ERROR(ENXIO)); mutex_enter(&zv->zv_state_lock); - if (zv->zv_zso->zso_dying) { - rw_exit(&zvol_state_lock); + if (zv->zv_zso->zso_dying || zv->zv_flags & ZVOL_REMOVING) { err = SET_ERROR(ENXIO); - goto out_zv_locked; + goto out_locked; } ASSERT3S(zv->zv_volmode, ==, ZFS_VOLMODE_DEV); @@ -954,6 +936,13 @@ retry: mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { + /* Removal started while locks were down. */ + err = SET_ERROR(ENXIO); + goto out_locked; + } + /* Check to see if zv_suspend_lock is needed. */ if (zv->zv_open_count != 0) { rw_exit(&zv->zv_suspend_lock); @@ -961,7 +950,6 @@ retry: } } } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -989,7 +977,7 @@ retry: if (drop_namespace) mutex_exit(&spa_namespace_lock); if (err) - goto out_zv_locked; + goto out_locked; } ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -1016,9 +1004,8 @@ out_opened: zvol_last_close(zv); wakeup(zv); } -out_zv_locked: - mutex_exit(&zv->zv_state_lock); out_locked: + mutex_exit(&zv->zv_state_lock); if (drop_suspend) rw_exit(&zv->zv_suspend_lock); return (err); @@ -1030,12 +1017,9 @@ zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td) zvol_state_t *zv; boolean_t drop_suspend = B_TRUE; - rw_enter(&zvol_state_lock, ZVOL_RW_READER); - zv = dev->si_drv2; - if (zv == NULL) { - rw_exit(&zvol_state_lock); + zv = atomic_load_ptr(&dev->si_drv2); + if (zv == NULL) return (SET_ERROR(ENXIO)); - } mutex_enter(&zv->zv_state_lock); if (zv->zv_flags & ZVOL_EXCL) { @@ -1060,6 +1044,15 @@ zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td) mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); mutex_enter(&zv->zv_state_lock); + + /* + * Unlike in zvol_cdev_open(), we don't check if + * removal started here, because we might be one of the + * openers that needs to be thrown out! If we're the + * last, we need to call zvol_last_close() below to + * finish cleanup. So, no special treatment for us. + */ + /* Check to see if zv_suspend_lock is needed. */ if (zv->zv_open_count != 1) { rw_exit(&zv->zv_suspend_lock); @@ -1069,7 +1062,6 @@ zvol_cdev_close(struct cdev *dev, int flags, int fmt, struct thread *td) } else { drop_suspend = B_FALSE; } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -1101,7 +1093,8 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, int error; boolean_t sync; - zv = dev->si_drv2; + zv = atomic_load_ptr(&dev->si_drv2); + ASSERT3P(zv, !=, NULL); error = 0; KASSERT(zv->zv_open_count > 0, @@ -1162,6 +1155,7 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, *(off_t *)data = 0; break; case DIOCGATTR: { + rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); spa_t *spa = dmu_objset_spa(zv->zv_objset); struct diocgattr_arg *arg = (struct diocgattr_arg *)data; uint64_t refd, avail, usedobjs, availobjs; @@ -1186,6 +1180,7 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, arg->value.off = refd / DEV_BSIZE; } else error = SET_ERROR(ENOIOCTL); + rw_exit(&zv->zv_suspend_lock); break; } case FIOSEEKHOLE: @@ -1196,10 +1191,12 @@ zvol_cdev_ioctl(struct cdev *dev, ulong_t cmd, caddr_t data, hole = (cmd == FIOSEEKHOLE); noff = *off; + rw_enter(&zv->zv_suspend_lock, ZVOL_RW_READER); lr = zfs_rangelock_enter(&zv->zv_rangelock, 0, UINT64_MAX, RL_READER); error = dmu_offset_next(zv->zv_objset, ZVOL_OBJ, hole, &noff); zfs_rangelock_exit(lr); + rw_exit(&zv->zv_suspend_lock); *off = noff; break; } @@ -1400,42 +1397,65 @@ zvol_alloc(const char *name, uint64_t volsize, uint64_t volblocksize, * Remove minor node for the specified volume. */ void -zvol_os_free(zvol_state_t *zv) +zvol_os_remove_minor(zvol_state_t *zv) { - ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); - ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + ASSERT(MUTEX_HELD(&zv->zv_state_lock)); ASSERT0(zv->zv_open_count); + ASSERT0(atomic_read(&zv->zv_suspend_ref)); + ASSERT(zv->zv_flags & ZVOL_REMOVING); - ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); - - rw_destroy(&zv->zv_suspend_lock); - zfs_rangelock_fini(&zv->zv_rangelock); + struct zvol_state_os *zso = zv->zv_zso; + zv->zv_zso = NULL; if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { - struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom; - struct g_provider *pp __maybe_unused = zsg->zsg_provider; - - ASSERT0P(pp->private); + struct zvol_state_geom *zsg = &zso->zso_geom; + struct g_provider *pp = zsg->zsg_provider; + atomic_store_ptr(&pp->private, NULL); + mutex_exit(&zv->zv_state_lock); g_topology_lock(); - zvol_geom_destroy(zv); + g_wither_geom(pp->geom, ENXIO); g_topology_unlock(); } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { - struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev; + struct zvol_state_dev *zsd = &zso->zso_dev; struct cdev *dev = zsd->zsd_cdev; + if (dev != NULL) + atomic_store_ptr(&dev->si_drv2, NULL); + mutex_exit(&zv->zv_state_lock); + if (dev != NULL) { - ASSERT0P(dev->si_drv2); destroy_dev(dev); knlist_clear(&zsd->zsd_selinfo.si_note, 0); knlist_destroy(&zsd->zsd_selinfo.si_note); } } + kmem_free(zso, sizeof (struct zvol_state_os)); + + mutex_enter(&zv->zv_state_lock); +} + +void +zvol_os_free(zvol_state_t *zv) +{ + ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); + ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + ASSERT0(zv->zv_open_count); + ASSERT0P(zv->zv_zso); + + ASSERT0P(zv->zv_objset); + ASSERT0P(zv->zv_zilog); + ASSERT0P(zv->zv_dn); + + ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); + + rw_destroy(&zv->zv_suspend_lock); + zfs_rangelock_fini(&zv->zv_rangelock); + mutex_destroy(&zv->zv_state_lock); cv_destroy(&zv->zv_removing_cv); dataset_kstats_destroy(&zv->zv_kstat); - kmem_free(zv->zv_zso, sizeof (struct zvol_state_os)); kmem_free(zv, sizeof (zvol_state_t)); zvol_minors--; } @@ -1538,28 +1558,6 @@ out_doi: return (error); } -void -zvol_os_clear_private(zvol_state_t *zv) -{ - ASSERT(RW_LOCK_HELD(&zvol_state_lock)); - if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { - struct zvol_state_geom *zsg = &zv->zv_zso->zso_geom; - struct g_provider *pp = zsg->zsg_provider; - - if (pp->private == NULL) /* already cleared */ - return; - - pp->private = NULL; - ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); - } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { - struct zvol_state_dev *zsd = &zv->zv_zso->zso_dev; - struct cdev *dev = zsd->zsd_cdev; - - if (dev != NULL) - dev->si_drv2 = NULL; - } -} - int zvol_os_update_volsize(zvol_state_t *zv, uint64_t volsize) { diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zfs_file_os.c b/sys/contrib/openzfs/module/os/linux/zfs/zfs_file_os.c index c729947369c2..3fdcdbac6f68 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zfs_file_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zfs_file_os.c @@ -115,8 +115,9 @@ zfs_file_write(zfs_file_t *fp, const void *buf, size_t count, ssize_t *resid) */ int zfs_file_pwrite(zfs_file_t *fp, const void *buf, size_t count, loff_t off, - ssize_t *resid) + uint8_t ashift, ssize_t *resid) { + (void) ashift; ssize_t rc; rc = kernel_write(fp, buf, count, &off); diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zpl_ctldir.c b/sys/contrib/openzfs/module/os/linux/zfs/zpl_ctldir.c index 48dae79a2373..81ac26cb0c93 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zpl_ctldir.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zpl_ctldir.c @@ -202,7 +202,7 @@ zpl_snapdir_revalidate(struct dentry *dentry, unsigned int flags) return (!!dentry->d_inode); } -static dentry_operations_t zpl_dops_snapdirs = { +static const struct dentry_operations zpl_dops_snapdirs = { /* * Auto mounting of snapshots is only supported for 2.6.37 and * newer kernels. Prior to this kernel the ops->follow_link() @@ -215,6 +215,51 @@ static dentry_operations_t zpl_dops_snapdirs = { .d_revalidate = zpl_snapdir_revalidate, }; +/* + * For the .zfs control directory to work properly we must be able to override + * the default operations table and register custom .d_automount and + * .d_revalidate callbacks. + */ +static void +set_snapdir_dentry_ops(struct dentry *dentry, unsigned int extraflags) { + static const unsigned int op_flags = + DCACHE_OP_HASH | DCACHE_OP_COMPARE | + DCACHE_OP_REVALIDATE | DCACHE_OP_DELETE | + DCACHE_OP_PRUNE | DCACHE_OP_WEAK_REVALIDATE | DCACHE_OP_REAL; + +#ifdef HAVE_D_SET_D_OP + /* + * d_set_d_op() will set the DCACHE_OP_ flags according to what it + * finds in the passed dentry_operations, so we don't have to. + * + * We clear the flags and the old op table before calling d_set_d_op() + * because issues a warning when the dentry operations table is already + * set. + */ + dentry->d_op = NULL; + dentry->d_flags &= ~op_flags; + d_set_d_op(dentry, &zpl_dops_snapdirs); + dentry->d_flags |= extraflags; +#else + /* + * Since 6.17 there's no exported way to modify dentry ops, so we have + * to reach in and do it ourselves. This should be safe for our very + * narrow use case, which is to create or splice in an entry to give + * access to a snapshot. + * + * We need to set the op flags directly. We hardcode + * DCACHE_OP_REVALIDATE because that's the only operation we have; if + * we ever extend zpl_dops_snapdirs we will need to update the op flags + * to match. + */ + spin_lock(&dentry->d_lock); + dentry->d_op = &zpl_dops_snapdirs; + dentry->d_flags &= ~op_flags; + dentry->d_flags |= DCACHE_OP_REVALIDATE | extraflags; + spin_unlock(&dentry->d_lock); +#endif +} + static struct dentry * zpl_snapdir_lookup(struct inode *dip, struct dentry *dentry, unsigned int flags) @@ -236,10 +281,7 @@ zpl_snapdir_lookup(struct inode *dip, struct dentry *dentry, return (ERR_PTR(error)); ASSERT(error == 0 || ip == NULL); - d_clear_d_op(dentry); - d_set_d_op(dentry, &zpl_dops_snapdirs); - dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; - + set_snapdir_dentry_ops(dentry, DCACHE_NEED_AUTOMOUNT); return (d_splice_alias(ip, dentry)); } @@ -373,8 +415,7 @@ zpl_snapdir_mkdir(struct inode *dip, struct dentry *dentry, umode_t mode) error = -zfsctl_snapdir_mkdir(dip, dname(dentry), vap, &ip, cr, 0); if (error == 0) { - d_clear_d_op(dentry); - d_set_d_op(dentry, &zpl_dops_snapdirs); + set_snapdir_dentry_ops(dentry, 0); d_instantiate(dentry, ip); } diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c b/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c index a73acdad34ae..bac166fcd89e 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c @@ -22,7 +22,7 @@ /* * Copyright (c) 2012, 2020 by Delphix. All rights reserved. * Copyright (c) 2024, Rob Norris <robn@despairlabs.com> - * Copyright (c) 2024, Klara, Inc. + * Copyright (c) 2024, 2025, Klara, Inc. */ #include <sys/dataset_kstats.h> @@ -679,28 +679,19 @@ zvol_open(struct block_device *bdev, fmode_t flag) retry: #endif - rw_enter(&zvol_state_lock, RW_READER); - /* - * Obtain a copy of private_data under the zvol_state_lock to make - * sure that either the result of zvol free code path setting - * disk->private_data to NULL is observed, or zvol_os_free() - * is not called on this zv because of the positive zv_open_count. - */ + #ifdef HAVE_BLK_MODE_T - zv = disk->private_data; + zv = atomic_load_ptr(&disk->private_data); #else - zv = bdev->bd_disk->private_data; + zv = atomic_load_ptr(&bdev->bd_disk->private_data); #endif if (zv == NULL) { - rw_exit(&zvol_state_lock); return (-SET_ERROR(ENXIO)); } mutex_enter(&zv->zv_state_lock); - if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { mutex_exit(&zv->zv_state_lock); - rw_exit(&zvol_state_lock); return (-SET_ERROR(ENXIO)); } @@ -712,8 +703,28 @@ retry: if (zv->zv_open_count == 0) { if (!rw_tryenter(&zv->zv_suspend_lock, RW_READER)) { mutex_exit(&zv->zv_state_lock); + + /* + * Removal may happen while the locks are down, so + * we can't trust zv any longer; we have to start over. + */ +#ifdef HAVE_BLK_MODE_T + zv = atomic_load_ptr(&disk->private_data); +#else + zv = atomic_load_ptr(&bdev->bd_disk->private_data); +#endif + if (zv == NULL) + return (-SET_ERROR(ENXIO)); + rw_enter(&zv->zv_suspend_lock, RW_READER); mutex_enter(&zv->zv_state_lock); + + if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { + mutex_exit(&zv->zv_state_lock); + rw_exit(&zv->zv_suspend_lock); + return (-SET_ERROR(ENXIO)); + } + /* check to see if zv_suspend_lock is needed */ if (zv->zv_open_count != 0) { rw_exit(&zv->zv_suspend_lock); @@ -724,7 +735,6 @@ retry: drop_suspend = B_TRUE; } } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -821,11 +831,11 @@ zvol_release(struct gendisk *disk, fmode_t unused) #if !defined(HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_1ARG) (void) unused; #endif - zvol_state_t *zv; boolean_t drop_suspend = B_TRUE; - rw_enter(&zvol_state_lock, RW_READER); - zv = disk->private_data; + zvol_state_t *zv = atomic_load_ptr(&disk->private_data); + if (zv == NULL) + return; mutex_enter(&zv->zv_state_lock); ASSERT3U(zv->zv_open_count, >, 0); @@ -839,6 +849,15 @@ zvol_release(struct gendisk *disk, fmode_t unused) mutex_exit(&zv->zv_state_lock); rw_enter(&zv->zv_suspend_lock, RW_READER); mutex_enter(&zv->zv_state_lock); + + /* + * Unlike in zvol_open(), we don't check if removal + * started here, because we might be one of the openers + * that needs to be thrown out! If we're the last, we + * need to call zvol_last_close() below to finish + * cleanup. So, no special treatment for us. + */ + /* check to see if zv_suspend_lock is needed */ if (zv->zv_open_count != 1) { rw_exit(&zv->zv_suspend_lock); @@ -848,7 +867,6 @@ zvol_release(struct gendisk *disk, fmode_t unused) } else { drop_suspend = B_FALSE; } - rw_exit(&zvol_state_lock); ASSERT(MUTEX_HELD(&zv->zv_state_lock)); @@ -868,9 +886,10 @@ static int zvol_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg) { - zvol_state_t *zv = bdev->bd_disk->private_data; int error = 0; + zvol_state_t *zv = atomic_load_ptr(&bdev->bd_disk->private_data); + ASSERT3P(zv, !=, NULL); ASSERT3U(zv->zv_open_count, >, 0); switch (cmd) { @@ -923,9 +942,8 @@ zvol_check_events(struct gendisk *disk, unsigned int clearing) { unsigned int mask = 0; - rw_enter(&zvol_state_lock, RW_READER); + zvol_state_t *zv = atomic_load_ptr(&disk->private_data); - zvol_state_t *zv = disk->private_data; if (zv != NULL) { mutex_enter(&zv->zv_state_lock); mask = zv->zv_changed ? DISK_EVENT_MEDIA_CHANGE : 0; @@ -933,17 +951,14 @@ zvol_check_events(struct gendisk *disk, unsigned int clearing) mutex_exit(&zv->zv_state_lock); } - rw_exit(&zvol_state_lock); - return (mask); } static int zvol_revalidate_disk(struct gendisk *disk) { - rw_enter(&zvol_state_lock, RW_READER); + zvol_state_t *zv = atomic_load_ptr(&disk->private_data); - zvol_state_t *zv = disk->private_data; if (zv != NULL) { mutex_enter(&zv->zv_state_lock); set_capacity(zv->zv_zso->zvo_disk, @@ -951,8 +966,6 @@ zvol_revalidate_disk(struct gendisk *disk) mutex_exit(&zv->zv_state_lock); } - rw_exit(&zvol_state_lock); - return (0); } @@ -971,16 +984,6 @@ zvol_os_update_volsize(zvol_state_t *zv, uint64_t volsize) return (0); } -void -zvol_os_clear_private(zvol_state_t *zv) -{ - /* - * Cleared while holding zvol_state_lock as a writer - * which will prevent zvol_open() from opening it. - */ - zv->zv_zso->zvo_disk->private_data = NULL; -} - /* * Provide a simple virtual geometry for legacy compatibility. For devices * smaller than 1 MiB a small head and sector count is used to allow very @@ -990,9 +993,10 @@ zvol_os_clear_private(zvol_state_t *zv) static int zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) { - zvol_state_t *zv = bdev->bd_disk->private_data; sector_t sectors; + zvol_state_t *zv = atomic_load_ptr(&bdev->bd_disk->private_data); + ASSERT3P(zv, !=, NULL); ASSERT3U(zv->zv_open_count, >, 0); sectors = get_capacity(zv->zv_zso->zvo_disk); @@ -1417,53 +1421,70 @@ out_kmem: return (ret); } -/* - * Cleanup then free a zvol_state_t which was created by zvol_alloc(). - * At this time, the structure is not opened by anyone, is taken off - * the zvol_state_list, and has its private data set to NULL. - * The zvol_state_lock is dropped. - * - * This function may take many milliseconds to complete (e.g. we've seen - * it take over 256ms), due to the calls to "blk_cleanup_queue" and - * "del_gendisk". Thus, consumers need to be careful to account for this - * latency when calling this function. - */ void -zvol_os_free(zvol_state_t *zv) +zvol_os_remove_minor(zvol_state_t *zv) { - - ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); - ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + ASSERT(MUTEX_HELD(&zv->zv_state_lock)); ASSERT0(zv->zv_open_count); - ASSERT0P(zv->zv_zso->zvo_disk->private_data); + ASSERT0(atomic_read(&zv->zv_suspend_ref)); + ASSERT(zv->zv_flags & ZVOL_REMOVING); - rw_destroy(&zv->zv_suspend_lock); - zfs_rangelock_fini(&zv->zv_rangelock); + struct zvol_state_os *zso = zv->zv_zso; + zv->zv_zso = NULL; + + /* Clearing private_data will make new callers return immediately. */ + atomic_store_ptr(&zso->zvo_disk->private_data, NULL); + + /* + * Drop the state lock before calling del_gendisk(). There may be + * callers waiting to acquire it, but del_gendisk() will block until + * they exit, which would deadlock. + */ + mutex_exit(&zv->zv_state_lock); - del_gendisk(zv->zv_zso->zvo_disk); + del_gendisk(zso->zvo_disk); #if defined(HAVE_SUBMIT_BIO_IN_BLOCK_DEVICE_OPERATIONS) && \ (defined(HAVE_BLK_ALLOC_DISK) || defined(HAVE_BLK_ALLOC_DISK_2ARG)) #if defined(HAVE_BLK_CLEANUP_DISK) - blk_cleanup_disk(zv->zv_zso->zvo_disk); + blk_cleanup_disk(zso->zvo_disk); #else - put_disk(zv->zv_zso->zvo_disk); + put_disk(zso->zvo_disk); #endif #else - blk_cleanup_queue(zv->zv_zso->zvo_queue); - put_disk(zv->zv_zso->zvo_disk); + blk_cleanup_queue(zso->zvo_queue); + put_disk(zso->zvo_disk); #endif - if (zv->zv_zso->use_blk_mq) - blk_mq_free_tag_set(&zv->zv_zso->tag_set); + if (zso->use_blk_mq) + blk_mq_free_tag_set(&zso->tag_set); + + ida_simple_remove(&zvol_ida, MINOR(zso->zvo_dev) >> ZVOL_MINOR_BITS); - ida_simple_remove(&zvol_ida, - MINOR(zv->zv_zso->zvo_dev) >> ZVOL_MINOR_BITS); + kmem_free(zso, sizeof (struct zvol_state_os)); + + mutex_enter(&zv->zv_state_lock); +} + +void +zvol_os_free(zvol_state_t *zv) +{ + + ASSERT(!RW_LOCK_HELD(&zv->zv_suspend_lock)); + ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); + ASSERT0(zv->zv_open_count); + ASSERT0P(zv->zv_zso); + + ASSERT0P(zv->zv_objset); + ASSERT0P(zv->zv_zilog); + ASSERT0P(zv->zv_dn); + + rw_destroy(&zv->zv_suspend_lock); + zfs_rangelock_fini(&zv->zv_rangelock); cv_destroy(&zv->zv_removing_cv); mutex_destroy(&zv->zv_state_lock); dataset_kstats_destroy(&zv->zv_kstat); - kmem_free(zv->zv_zso, sizeof (struct zvol_state_os)); kmem_free(zv, sizeof (zvol_state_t)); } diff --git a/sys/contrib/openzfs/module/zcommon/simd_stat.c b/sys/contrib/openzfs/module/zcommon/simd_stat.c index 11e2080ff9f2..007ae9e4fbbc 100644 --- a/sys/contrib/openzfs/module/zcommon/simd_stat.c +++ b/sys/contrib/openzfs/module/zcommon/simd_stat.c @@ -118,6 +118,10 @@ simd_stat_kstat_data(char *buf, size_t size, void *data) "pclmulqdq", zfs_pclmulqdq_available()); off += SIMD_STAT_PRINT(simd_stat_kstat_payload, "movbe", zfs_movbe_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "vaes", zfs_vaes_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "vpclmulqdq", zfs_vpclmulqdq_available()); off += SIMD_STAT_PRINT(simd_stat_kstat_payload, "osxsave", boot_cpu_has(X86_FEATURE_OSXSAVE)); diff --git a/sys/contrib/openzfs/module/zcommon/zfs_deleg.c b/sys/contrib/openzfs/module/zcommon/zfs_deleg.c index 49bb534ca26c..87596558c9a1 100644 --- a/sys/contrib/openzfs/module/zcommon/zfs_deleg.c +++ b/sys/contrib/openzfs/module/zcommon/zfs_deleg.c @@ -59,6 +59,7 @@ const zfs_deleg_perm_tab_t zfs_deleg_perm_tab[] = { {ZFS_DELEG_PERM_SNAPSHOT}, {ZFS_DELEG_PERM_SHARE}, {ZFS_DELEG_PERM_SEND}, + {ZFS_DELEG_PERM_SEND_RAW}, {ZFS_DELEG_PERM_USERPROP}, {ZFS_DELEG_PERM_USERQUOTA}, {ZFS_DELEG_PERM_GROUPQUOTA}, diff --git a/sys/contrib/openzfs/module/zcommon/zpool_prop.c b/sys/contrib/openzfs/module/zcommon/zpool_prop.c index 04ae9f986d8f..07819ba2be8b 100644 --- a/sys/contrib/openzfs/module/zcommon/zpool_prop.c +++ b/sys/contrib/openzfs/module/zcommon/zpool_prop.c @@ -467,9 +467,15 @@ vdev_prop_init(void) zprop_register_index(VDEV_PROP_RAIDZ_EXPANDING, "raidz_expanding", 0, PROP_READONLY, ZFS_TYPE_VDEV, "on | off", "RAIDZ_EXPANDING", boolean_table, sfeatures); + zprop_register_index(VDEV_PROP_SIT_OUT, "sit_out", 0, + PROP_DEFAULT, ZFS_TYPE_VDEV, "on | off", "SIT_OUT", boolean_table, + sfeatures); zprop_register_index(VDEV_PROP_TRIM_SUPPORT, "trim_support", 0, PROP_READONLY, ZFS_TYPE_VDEV, "on | off", "TRIMSUP", boolean_table, sfeatures); + zprop_register_index(VDEV_PROP_AUTOSIT, "autosit", 0, + PROP_DEFAULT, ZFS_TYPE_VDEV, "on | off", "AUTOSIT", boolean_table, + sfeatures); /* default index properties */ zprop_register_index(VDEV_PROP_FAILFAST, "failfast", B_TRUE, diff --git a/sys/contrib/openzfs/module/zfs/arc.c b/sys/contrib/openzfs/module/zfs/arc.c index df41e3b49204..bd6dc8edd8ca 100644 --- a/sys/contrib/openzfs/module/zfs/arc.c +++ b/sys/contrib/openzfs/module/zfs/arc.c @@ -486,13 +486,13 @@ static taskq_t *arc_flush_taskq; static uint_t zfs_arc_evict_threads = 0; /* The 7 states: */ -arc_state_t ARC_anon; -arc_state_t ARC_mru; -arc_state_t ARC_mru_ghost; -arc_state_t ARC_mfu; -arc_state_t ARC_mfu_ghost; -arc_state_t ARC_l2c_only; -arc_state_t ARC_uncached; +static arc_state_t ARC_anon; +/* */ arc_state_t ARC_mru; +static arc_state_t ARC_mru_ghost; +/* */ arc_state_t ARC_mfu; +static arc_state_t ARC_mfu_ghost; +static arc_state_t ARC_l2c_only; +static arc_state_t ARC_uncached; arc_stats_t arc_stats = { { "hits", KSTAT_DATA_UINT64 }, @@ -832,15 +832,15 @@ typedef struct arc_async_flush { #define L2ARC_FEED_TYPES 4 /* L2ARC Performance Tunables */ -uint64_t l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ -uint64_t l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ -uint64_t l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ -uint64_t l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; -uint64_t l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ -uint64_t l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ -int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ -int l2arc_feed_again = B_TRUE; /* turbo warmup */ -int l2arc_norw = B_FALSE; /* no reads during writes */ +static uint64_t l2arc_write_max = L2ARC_WRITE_SIZE; /* def max write size */ +static uint64_t l2arc_write_boost = L2ARC_WRITE_SIZE; /* extra warmup write */ +static uint64_t l2arc_headroom = L2ARC_HEADROOM; /* # of dev writes */ +static uint64_t l2arc_headroom_boost = L2ARC_HEADROOM_BOOST; +static uint64_t l2arc_feed_secs = L2ARC_FEED_SECS; /* interval seconds */ +static uint64_t l2arc_feed_min_ms = L2ARC_FEED_MIN_MS; /* min interval msecs */ +static int l2arc_noprefetch = B_TRUE; /* don't cache prefetch bufs */ +static int l2arc_feed_again = B_TRUE; /* turbo warmup */ +static int l2arc_norw = B_FALSE; /* no reads during writes */ static uint_t l2arc_meta_percent = 33; /* limit on headers size */ /* diff --git a/sys/contrib/openzfs/module/zfs/dbuf.c b/sys/contrib/openzfs/module/zfs/dbuf.c index 3d0f88b36336..fccc4c5b5b94 100644 --- a/sys/contrib/openzfs/module/zfs/dbuf.c +++ b/sys/contrib/openzfs/module/zfs/dbuf.c @@ -2270,14 +2270,6 @@ dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) if (dn->dn_objset->os_dsl_dataset != NULL) rrw_exit(&dn->dn_objset->os_dsl_dataset->ds_bp_rwlock, FTAG); #endif - /* - * We make this assert for private objects as well, but after we - * check if we're already dirty. They are allowed to re-dirty - * in syncing context. - */ - ASSERT(dn->dn_object == DMU_META_DNODE_OBJECT || - dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == - (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); mutex_enter(&db->db_mtx); /* @@ -2289,12 +2281,6 @@ dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) db->db_state == DB_CACHED || db->db_state == DB_FILL || db->db_state == DB_NOFILL); - mutex_enter(&dn->dn_mtx); - dnode_set_dirtyctx(dn, tx, db); - if (tx->tx_txg > dn->dn_dirty_txg) - dn->dn_dirty_txg = tx->tx_txg; - mutex_exit(&dn->dn_mtx); - if (db->db_blkid == DMU_SPILL_BLKID) dn->dn_have_spill = B_TRUE; @@ -2313,13 +2299,6 @@ dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx) return (dr_next); } - /* - * Only valid if not already dirty. - */ - ASSERT(dn->dn_object == 0 || - dn->dn_dirtyctx == DN_UNDIRTIED || dn->dn_dirtyctx == - (dmu_tx_is_syncing(tx) ? DN_DIRTY_SYNC : DN_DIRTY_OPEN)); - ASSERT3U(dn->dn_nlevels, >, db->db_level); /* @@ -2557,12 +2536,13 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx) /* * Due to our use of dn_nlevels below, this can only be called - * in open context, unless we are operating on the MOS. - * From syncing context, dn_nlevels may be different from the - * dn_nlevels used when dbuf was dirtied. + * in open context, unless we are operating on the MOS or it's + * a special object. From syncing context, dn_nlevels may be + * different from the dn_nlevels used when dbuf was dirtied. */ ASSERT(db->db_objset == dmu_objset_pool(db->db_objset)->dp_meta_objset || + DMU_OBJECT_IS_SPECIAL(db->db.db_object) || txg != spa_syncing_txg(dmu_objset_spa(db->db_objset))); ASSERT(db->db_blkid != DMU_BONUS_BLKID); ASSERT0(db->db_level); diff --git a/sys/contrib/openzfs/module/zfs/ddt.c b/sys/contrib/openzfs/module/zfs/ddt.c index d6658375f810..0dc9adc7fd4f 100644 --- a/sys/contrib/openzfs/module/zfs/ddt.c +++ b/sys/contrib/openzfs/module/zfs/ddt.c @@ -1701,9 +1701,11 @@ ddt_load(spa_t *spa) } } - error = ddt_log_load(ddt); - if (error != 0 && error != ENOENT) - return (error); + if (ddt->ddt_flags & DDT_FLAG_LOG) { + error = ddt_log_load(ddt); + if (error != 0 && error != ENOENT) + return (error); + } DDT_KSTAT_SET(ddt, dds_log_active_entries, avl_numnodes(&ddt->ddt_log_active->ddl_tree)); diff --git a/sys/contrib/openzfs/module/zfs/ddt_log.c b/sys/contrib/openzfs/module/zfs/ddt_log.c index 3d30e244c1f7..c7a2426f3a77 100644 --- a/sys/contrib/openzfs/module/zfs/ddt_log.c +++ b/sys/contrib/openzfs/module/zfs/ddt_log.c @@ -176,11 +176,13 @@ ddt_log_update_stats(ddt_t *ddt) * that's reasonable to expect anyway. */ dmu_object_info_t doi; - uint64_t nblocks; - dmu_object_info(ddt->ddt_os, ddt->ddt_log_active->ddl_object, &doi); - nblocks = doi.doi_physical_blocks_512; - dmu_object_info(ddt->ddt_os, ddt->ddt_log_flushing->ddl_object, &doi); - nblocks += doi.doi_physical_blocks_512; + uint64_t nblocks = 0; + if (dmu_object_info(ddt->ddt_os, ddt->ddt_log_active->ddl_object, + &doi) == 0) + nblocks += doi.doi_physical_blocks_512; + if (dmu_object_info(ddt->ddt_os, ddt->ddt_log_flushing->ddl_object, + &doi) == 0) + nblocks += doi.doi_physical_blocks_512; ddt_object_t *ddo = &ddt->ddt_log_stats; ddo->ddo_count = @@ -243,6 +245,13 @@ ddt_log_alloc_entry(ddt_t *ddt) } static void +ddt_log_free_entry(ddt_t *ddt, ddt_log_entry_t *ddle) +{ + kmem_cache_free(ddt->ddt_flags & DDT_FLAG_FLAT ? + ddt_log_entry_flat_cache : ddt_log_entry_trad_cache, ddle); +} + +static void ddt_log_update_entry(ddt_t *ddt, ddt_log_t *ddl, ddt_lightweight_entry_t *ddlwe) { /* Create the log tree entry from a live or stored entry */ @@ -347,8 +356,7 @@ ddt_log_take_first(ddt_t *ddt, ddt_log_t *ddl, ddt_lightweight_entry_t *ddlwe) ddt_histogram_sub_entry(ddt, &ddt->ddt_log_histogram, ddlwe); avl_remove(&ddl->ddl_tree, ddle); - kmem_cache_free(ddt->ddt_flags & DDT_FLAG_FLAT ? - ddt_log_entry_flat_cache : ddt_log_entry_trad_cache, ddle); + ddt_log_free_entry(ddt, ddle); return (B_TRUE); } @@ -365,8 +373,7 @@ ddt_log_remove_key(ddt_t *ddt, ddt_log_t *ddl, const ddt_key_t *ddk) ddt_histogram_sub_entry(ddt, &ddt->ddt_log_histogram, &ddlwe); avl_remove(&ddl->ddl_tree, ddle); - kmem_cache_free(ddt->ddt_flags & DDT_FLAG_FLAT ? - ddt_log_entry_flat_cache : ddt_log_entry_trad_cache, ddle); + ddt_log_free_entry(ddt, ddle); return (B_TRUE); } @@ -527,8 +534,7 @@ ddt_log_empty(ddt_t *ddt, ddt_log_t *ddl) IMPLY(ddt->ddt_version == UINT64_MAX, avl_is_empty(&ddl->ddl_tree)); while ((ddle = avl_destroy_nodes(&ddl->ddl_tree, &cookie)) != NULL) { - kmem_cache_free(ddt->ddt_flags & DDT_FLAG_FLAT ? - ddt_log_entry_flat_cache : ddt_log_entry_trad_cache, ddle); + ddt_log_free_entry(ddt, ddle); } ASSERT(avl_is_empty(&ddl->ddl_tree)); } @@ -727,7 +733,7 @@ ddt_log_load(ddt_t *ddt) ddle = fe; fe = AVL_NEXT(fl, fe); avl_remove(fl, ddle); - + ddt_log_free_entry(ddt, ddle); ddle = ae; ae = AVL_NEXT(al, ae); } diff --git a/sys/contrib/openzfs/module/zfs/dmu.c b/sys/contrib/openzfs/module/zfs/dmu.c index f7f808d5b8f7..a7a5c89bdafb 100644 --- a/sys/contrib/openzfs/module/zfs/dmu.c +++ b/sys/contrib/openzfs/module/zfs/dmu.c @@ -759,6 +759,8 @@ dmu_prefetch_by_dnode(dnode_t *dn, int64_t level, uint64_t offset, */ uint8_t ibps = ibs - SPA_BLKPTRSHIFT; limit = P2ROUNDUP(dmu_prefetch_max, 1 << ibs) >> ibs; + if (limit == 0) + end2 = start2; do { level2++; start2 = P2ROUNDUP(start2, 1 << ibps) >> ibps; @@ -1689,8 +1691,8 @@ dmu_object_cached_size(objset_t *os, uint64_t object, dmu_object_info_from_dnode(dn, &doi); - for (uint64_t off = 0; off < doi.doi_max_offset; - off += dmu_prefetch_max) { + for (uint64_t off = 0; off < doi.doi_max_offset && + dmu_prefetch_max > 0; off += dmu_prefetch_max) { /* dbuf_read doesn't prefetch L1 blocks. */ dmu_prefetch_by_dnode(dn, 1, off, dmu_prefetch_max, ZIO_PRIORITY_SYNC_READ); diff --git a/sys/contrib/openzfs/module/zfs/dmu_objset.c b/sys/contrib/openzfs/module/zfs/dmu_objset.c index a77f338bdfd3..8e6b569c2100 100644 --- a/sys/contrib/openzfs/module/zfs/dmu_objset.c +++ b/sys/contrib/openzfs/module/zfs/dmu_objset.c @@ -2037,6 +2037,8 @@ userquota_updates_task(void *arg) dn->dn_id_flags |= DN_ID_CHKED_BONUS; } dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); + ASSERT3U(dn->dn_dirtycnt, >, 0); + dn->dn_dirtycnt--; mutex_exit(&dn->dn_mtx); multilist_sublist_remove(list, dn); @@ -2070,6 +2072,10 @@ dnode_rele_task(void *arg) dnode_t *dn; while ((dn = multilist_sublist_head(list)) != NULL) { + mutex_enter(&dn->dn_mtx); + ASSERT3U(dn->dn_dirtycnt, >, 0); + dn->dn_dirtycnt--; + mutex_exit(&dn->dn_mtx); multilist_sublist_remove(list, dn); dnode_rele(dn, &os->os_synced_dnodes); } diff --git a/sys/contrib/openzfs/module/zfs/dmu_zfetch.c b/sys/contrib/openzfs/module/zfs/dmu_zfetch.c index 51165d0bf723..3d3a9c713568 100644 --- a/sys/contrib/openzfs/module/zfs/dmu_zfetch.c +++ b/sys/contrib/openzfs/module/zfs/dmu_zfetch.c @@ -57,19 +57,19 @@ static unsigned int zfetch_max_sec_reap = 2; /* min bytes to prefetch per stream (default 2MB) */ static unsigned int zfetch_min_distance = 2 * 1024 * 1024; /* max bytes to prefetch per stream (default 8MB) */ -unsigned int zfetch_max_distance = 8 * 1024 * 1024; +static unsigned int zfetch_max_distance = 8 * 1024 * 1024; #else /* min bytes to prefetch per stream (default 4MB) */ static unsigned int zfetch_min_distance = 4 * 1024 * 1024; /* max bytes to prefetch per stream (default 64MB) */ -unsigned int zfetch_max_distance = 64 * 1024 * 1024; +static unsigned int zfetch_max_distance = 64 * 1024 * 1024; #endif /* max bytes to prefetch indirects for per stream (default 128MB) */ -unsigned int zfetch_max_idistance = 128 * 1024 * 1024; +static unsigned int zfetch_max_idistance = 128 * 1024 * 1024; /* max request reorder distance within a stream (default 16MB) */ -unsigned int zfetch_max_reorder = 16 * 1024 * 1024; +static unsigned int zfetch_max_reorder = 16 * 1024 * 1024; /* Max log2 fraction of holes in a stream */ -unsigned int zfetch_hole_shift = 2; +static unsigned int zfetch_hole_shift = 2; typedef struct zfetch_stats { kstat_named_t zfetchstat_hits; diff --git a/sys/contrib/openzfs/module/zfs/dnode.c b/sys/contrib/openzfs/module/zfs/dnode.c index 963ff41232a3..6c150d31c669 100644 --- a/sys/contrib/openzfs/module/zfs/dnode.c +++ b/sys/contrib/openzfs/module/zfs/dnode.c @@ -173,9 +173,7 @@ dnode_cons(void *arg, void *unused, int kmflag) dn->dn_allocated_txg = 0; dn->dn_free_txg = 0; dn->dn_assigned_txg = 0; - dn->dn_dirty_txg = 0; - dn->dn_dirtyctx = 0; - dn->dn_dirtyctx_firstset = NULL; + dn->dn_dirtycnt = 0; dn->dn_bonus = NULL; dn->dn_have_spill = B_FALSE; dn->dn_zio = NULL; @@ -229,9 +227,7 @@ dnode_dest(void *arg, void *unused) ASSERT0(dn->dn_allocated_txg); ASSERT0(dn->dn_free_txg); ASSERT0(dn->dn_assigned_txg); - ASSERT0(dn->dn_dirty_txg); - ASSERT0(dn->dn_dirtyctx); - ASSERT0P(dn->dn_dirtyctx_firstset); + ASSERT0(dn->dn_dirtycnt); ASSERT0P(dn->dn_bonus); ASSERT(!dn->dn_have_spill); ASSERT0P(dn->dn_zio); @@ -692,10 +688,8 @@ dnode_destroy(dnode_t *dn) dn->dn_allocated_txg = 0; dn->dn_free_txg = 0; dn->dn_assigned_txg = 0; - dn->dn_dirty_txg = 0; + dn->dn_dirtycnt = 0; - dn->dn_dirtyctx = 0; - dn->dn_dirtyctx_firstset = NULL; if (dn->dn_bonus != NULL) { mutex_enter(&dn->dn_bonus->db_mtx); dbuf_destroy(dn->dn_bonus); @@ -800,11 +794,9 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, dn->dn_bonuslen = bonuslen; dn->dn_checksum = ZIO_CHECKSUM_INHERIT; dn->dn_compress = ZIO_COMPRESS_INHERIT; - dn->dn_dirtyctx = 0; dn->dn_free_txg = 0; - dn->dn_dirtyctx_firstset = NULL; - dn->dn_dirty_txg = 0; + dn->dn_dirtycnt = 0; dn->dn_allocated_txg = tx->tx_txg; dn->dn_id_flags = 0; @@ -955,9 +947,7 @@ dnode_move_impl(dnode_t *odn, dnode_t *ndn) ndn->dn_allocated_txg = odn->dn_allocated_txg; ndn->dn_free_txg = odn->dn_free_txg; ndn->dn_assigned_txg = odn->dn_assigned_txg; - ndn->dn_dirty_txg = odn->dn_dirty_txg; - ndn->dn_dirtyctx = odn->dn_dirtyctx; - ndn->dn_dirtyctx_firstset = odn->dn_dirtyctx_firstset; + ndn->dn_dirtycnt = odn->dn_dirtycnt; ASSERT0(zfs_refcount_count(&odn->dn_tx_holds)); zfs_refcount_transfer(&ndn->dn_holds, &odn->dn_holds); ASSERT(avl_is_empty(&ndn->dn_dbufs)); @@ -1020,9 +1010,7 @@ dnode_move_impl(dnode_t *odn, dnode_t *ndn) odn->dn_allocated_txg = 0; odn->dn_free_txg = 0; odn->dn_assigned_txg = 0; - odn->dn_dirty_txg = 0; - odn->dn_dirtyctx = 0; - odn->dn_dirtyctx_firstset = NULL; + odn->dn_dirtycnt = 0; odn->dn_have_spill = B_FALSE; odn->dn_zio = NULL; odn->dn_oldused = 0; @@ -1273,8 +1261,8 @@ dnode_check_slots_free(dnode_children_t *children, int idx, int slots) } else if (DN_SLOT_IS_PTR(dn)) { mutex_enter(&dn->dn_mtx); boolean_t can_free = (dn->dn_type == DMU_OT_NONE && - zfs_refcount_is_zero(&dn->dn_holds) && - !DNODE_IS_DIRTY(dn)); + dn->dn_dirtycnt == 0 && + zfs_refcount_is_zero(&dn->dn_holds)); mutex_exit(&dn->dn_mtx); if (!can_free) @@ -1757,17 +1745,23 @@ dnode_hold(objset_t *os, uint64_t object, const void *tag, dnode_t **dnp) * reference on the dnode. Returns FALSE if unable to add a * new reference. */ +static boolean_t +dnode_add_ref_locked(dnode_t *dn, const void *tag) +{ + ASSERT(MUTEX_HELD(&dn->dn_mtx)); + if (zfs_refcount_is_zero(&dn->dn_holds)) + return (FALSE); + VERIFY(1 < zfs_refcount_add(&dn->dn_holds, tag)); + return (TRUE); +} + boolean_t dnode_add_ref(dnode_t *dn, const void *tag) { mutex_enter(&dn->dn_mtx); - if (zfs_refcount_is_zero(&dn->dn_holds)) { - mutex_exit(&dn->dn_mtx); - return (FALSE); - } - VERIFY(1 < zfs_refcount_add(&dn->dn_holds, tag)); + boolean_t r = dnode_add_ref_locked(dn, tag); mutex_exit(&dn->dn_mtx); - return (TRUE); + return (r); } void @@ -1830,31 +1824,20 @@ dnode_try_claim(objset_t *os, uint64_t object, int slots) } /* - * Checks if the dnode itself is dirty, or is carrying any uncommitted records. - * It is important to check both conditions, as some operations (eg appending - * to a file) can dirty both as a single logical unit, but they are not synced - * out atomically, so checking one and not the other can result in an object - * appearing to be clean mid-way through a commit. + * Test if the dnode is dirty, or carrying uncommitted records. * - * Do not change this lightly! If you get it wrong, dmu_offset_next() can - * detect a hole where there is really data, leading to silent corruption. + * dn_dirtycnt is the number of txgs this dnode is dirty on. It's incremented + * in dnode_setdirty() the first time the dnode is dirtied on a txg, and + * decremented in either dnode_rele_task() or userquota_updates_task() when the + * txg is synced out. */ boolean_t dnode_is_dirty(dnode_t *dn) { mutex_enter(&dn->dn_mtx); - - for (int i = 0; i < TXG_SIZE; i++) { - if (multilist_link_active(&dn->dn_dirty_link[i]) || - !list_is_empty(&dn->dn_dirty_records[i])) { - mutex_exit(&dn->dn_mtx); - return (B_TRUE); - } - } - + boolean_t dirty = (dn->dn_dirtycnt != 0); mutex_exit(&dn->dn_mtx); - - return (B_FALSE); + return (dirty); } void @@ -1916,7 +1899,11 @@ dnode_setdirty(dnode_t *dn, dmu_tx_t *tx) * dnode will hang around after we finish processing its * children. */ - VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg)); + mutex_enter(&dn->dn_mtx); + VERIFY(dnode_add_ref_locked(dn, (void *)(uintptr_t)tx->tx_txg)); + dn->dn_dirtycnt++; + ASSERT3U(dn->dn_dirtycnt, <=, 3); + mutex_exit(&dn->dn_mtx); (void) dbuf_dirty(dn->dn_dbuf, tx); @@ -2221,32 +2208,6 @@ dnode_dirty_l1range(dnode_t *dn, uint64_t start_blkid, uint64_t end_blkid, mutex_exit(&dn->dn_dbufs_mtx); } -void -dnode_set_dirtyctx(dnode_t *dn, dmu_tx_t *tx, const void *tag) -{ - /* - * Don't set dirtyctx to SYNC if we're just modifying this as we - * initialize the objset. - */ - if (dn->dn_dirtyctx == DN_UNDIRTIED) { - dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset; - - if (ds != NULL) { - rrw_enter(&ds->ds_bp_rwlock, RW_READER, tag); - } - if (!BP_IS_HOLE(dn->dn_objset->os_rootbp)) { - if (dmu_tx_is_syncing(tx)) - dn->dn_dirtyctx = DN_DIRTY_SYNC; - else - dn->dn_dirtyctx = DN_DIRTY_OPEN; - dn->dn_dirtyctx_firstset = tag; - } - if (ds != NULL) { - rrw_exit(&ds->ds_bp_rwlock, tag); - } - } -} - static void dnode_partial_zero(dnode_t *dn, uint64_t off, uint64_t blkoff, uint64_t len, dmu_tx_t *tx) diff --git a/sys/contrib/openzfs/module/zfs/dsl_deadlist.c b/sys/contrib/openzfs/module/zfs/dsl_deadlist.c index 475db3c89508..41ac72bf1c16 100644 --- a/sys/contrib/openzfs/module/zfs/dsl_deadlist.c +++ b/sys/contrib/openzfs/module/zfs/dsl_deadlist.c @@ -1049,7 +1049,8 @@ dsl_livelist_iterate(void *arg, const blkptr_t *bp, boolean_t bp_freed, ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(&found->le_bp)); ASSERT3U(BP_GET_CHECKSUM(bp), ==, BP_GET_CHECKSUM(&found->le_bp)); - ASSERT3U(BP_GET_BIRTH(bp), ==, BP_GET_BIRTH(&found->le_bp)); + ASSERT3U(BP_GET_PHYSICAL_BIRTH(bp), ==, + BP_GET_PHYSICAL_BIRTH(&found->le_bp)); } if (bp_freed) { if (found == NULL) { diff --git a/sys/contrib/openzfs/module/zfs/multilist.c b/sys/contrib/openzfs/module/zfs/multilist.c index 7b85d19e19ee..46fb79269310 100644 --- a/sys/contrib/openzfs/module/zfs/multilist.c +++ b/sys/contrib/openzfs/module/zfs/multilist.c @@ -81,7 +81,7 @@ multilist_create_impl(multilist_t *ml, size_t size, size_t offset, ml->ml_num_sublists = num; ml->ml_index_func = index_func; - ml->ml_sublists = kmem_zalloc(sizeof (multilist_sublist_t) * + ml->ml_sublists = vmem_zalloc(sizeof (multilist_sublist_t) * ml->ml_num_sublists, KM_SLEEP); ASSERT3P(ml->ml_sublists, !=, NULL); @@ -134,7 +134,7 @@ multilist_destroy(multilist_t *ml) } ASSERT3P(ml->ml_sublists, !=, NULL); - kmem_free(ml->ml_sublists, + vmem_free(ml->ml_sublists, sizeof (multilist_sublist_t) * ml->ml_num_sublists); ml->ml_num_sublists = 0; diff --git a/sys/contrib/openzfs/module/zfs/spa_config.c b/sys/contrib/openzfs/module/zfs/spa_config.c index 7d4d06659146..cf28955b0c50 100644 --- a/sys/contrib/openzfs/module/zfs/spa_config.c +++ b/sys/contrib/openzfs/module/zfs/spa_config.c @@ -48,18 +48,17 @@ /* * Pool configuration repository. * - * Pool configuration is stored as a packed nvlist on the filesystem. By - * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot - * (when the ZFS module is loaded). Pools can also have the 'cachefile' - * property set that allows them to be stored in an alternate location until - * the control of external software. + * Pool configuration is stored as a packed nvlist on the filesystem. When + * pools are imported they are added to the /etc/zfs/zpool.cache file and + * removed from it when exported. For each cache file, we have a single nvlist + * which holds all the configuration information. Pools can also have the + * 'cachefile' property set which allows this config to be stored in an + * alternate location under the control of external software. * - * For each cache file, we have a single nvlist which holds all the - * configuration information. When the module loads, we read this information - * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is - * maintained independently in spa.c. Whenever the namespace is modified, or - * the configuration of a pool is changed, we call spa_write_cachefile(), which - * walks through all the active pools and writes the configuration to disk. + * The kernel independantly maintains an AVL tree of imported pools. See the + * "SPA locking" comment in spa.c. Whenever a pool configuration is modified + * we call spa_write_cachefile() which walks through all the active pools and + * writes the updated configuration to to /etc/zfs/zpool.cache file. */ static uint64_t spa_config_generation = 1; @@ -69,94 +68,6 @@ static uint64_t spa_config_generation = 1; * userland pools when doing testing. */ char *spa_config_path = (char *)ZPOOL_CACHE; -#ifdef _KERNEL -static int zfs_autoimport_disable = B_TRUE; -#endif - -/* - * Called when the module is first loaded, this routine loads the configuration - * file into the SPA namespace. It does not actually open or load the pools; it - * only populates the namespace. - */ -void -spa_config_load(void) -{ - void *buf = NULL; - nvlist_t *nvlist, *child; - nvpair_t *nvpair; - char *pathname; - zfs_file_t *fp; - zfs_file_attr_t zfa; - uint64_t fsize; - int err; - -#ifdef _KERNEL - if (zfs_autoimport_disable) - return; -#endif - - /* - * Open the configuration file. - */ - pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP); - - (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path); - - err = zfs_file_open(pathname, O_RDONLY, 0, &fp); - -#ifdef __FreeBSD__ - if (err) - err = zfs_file_open(ZPOOL_CACHE_BOOT, O_RDONLY, 0, &fp); -#endif - kmem_free(pathname, MAXPATHLEN); - - if (err) - return; - - if (zfs_file_getattr(fp, &zfa)) - goto out; - - fsize = zfa.zfa_size; - buf = kmem_alloc(fsize, KM_SLEEP); - - /* - * Read the nvlist from the file. - */ - if (zfs_file_read(fp, buf, fsize, NULL) < 0) - goto out; - - /* - * Unpack the nvlist. - */ - if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0) - goto out; - - /* - * Iterate over all elements in the nvlist, creating a new spa_t for - * each one with the specified configuration. - */ - mutex_enter(&spa_namespace_lock); - nvpair = NULL; - while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) { - if (nvpair_type(nvpair) != DATA_TYPE_NVLIST) - continue; - - child = fnvpair_value_nvlist(nvpair); - - if (spa_lookup(nvpair_name(nvpair)) != NULL) - continue; - (void) spa_add(nvpair_name(nvpair), child, NULL); - } - mutex_exit(&spa_namespace_lock); - - nvlist_free(nvlist); - -out: - if (buf != NULL) - kmem_free(buf, fsize); - - zfs_file_close(fp); -} static int spa_config_remove(spa_config_dirent_t *dp) @@ -623,7 +534,6 @@ spa_config_update(spa_t *spa, int what) spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS); } -EXPORT_SYMBOL(spa_config_load); EXPORT_SYMBOL(spa_all_configs); EXPORT_SYMBOL(spa_config_set); EXPORT_SYMBOL(spa_config_generate); @@ -634,8 +544,3 @@ EXPORT_SYMBOL(spa_config_update); ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD, "SPA config file (/etc/zfs/zpool.cache)"); #endif - -#ifdef _KERNEL -ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW, - "Disable pool import at module load"); -#endif diff --git a/sys/contrib/openzfs/module/zfs/spa_misc.c b/sys/contrib/openzfs/module/zfs/spa_misc.c index cce772eae598..6f7c060f97f8 100644 --- a/sys/contrib/openzfs/module/zfs/spa_misc.c +++ b/sys/contrib/openzfs/module/zfs/spa_misc.c @@ -251,11 +251,11 @@ spa_mode_t spa_mode_global = SPA_MODE_UNINIT; #ifdef ZFS_DEBUG /* - * Everything except dprintf, set_error, spa, and indirect_remap is on - * by default in debug builds. + * Everything except dprintf, set_error, indirect_remap, and raidz_reconstruct + * is on by default in debug builds. */ int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SET_ERROR | - ZFS_DEBUG_INDIRECT_REMAP); + ZFS_DEBUG_INDIRECT_REMAP | ZFS_DEBUG_RAIDZ_RECONSTRUCT); #else int zfs_flags = 0; #endif @@ -2548,13 +2548,6 @@ spa_name_compare(const void *a1, const void *a2) } void -spa_boot_init(void *unused) -{ - (void) unused; - spa_config_load(); -} - -void spa_init(spa_mode_t mode) { mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL); @@ -2607,7 +2600,6 @@ spa_init(spa_mode_t mode) chksum_init(); zpool_prop_init(); zpool_feature_init(); - spa_config_load(); vdev_prop_init(); l2arc_start(); scan_init(); diff --git a/sys/contrib/openzfs/module/zfs/vdev.c b/sys/contrib/openzfs/module/zfs/vdev.c index 9cf35e379000..fc6d445f9785 100644 --- a/sys/contrib/openzfs/module/zfs/vdev.c +++ b/sys/contrib/openzfs/module/zfs/vdev.c @@ -29,7 +29,7 @@ * Copyright 2017 Joyent, Inc. * Copyright (c) 2017, Intel Corporation. * Copyright (c) 2019, Datto Inc. All rights reserved. - * Copyright (c) 2021, Klara Inc. + * Copyright (c) 2021, 2025, Klara, Inc. * Copyright (c) 2021, 2023 Hewlett Packard Enterprise Development LP. */ @@ -100,7 +100,7 @@ static uint_t zfs_vdev_default_ms_shift = 29; /* upper limit for metaslab size (16G) */ static uint_t zfs_vdev_max_ms_shift = 34; -int vdev_validate_skip = B_FALSE; +static int vdev_validate_skip = B_FALSE; /* * Since the DTL space map of a vdev is not expected to have a lot of @@ -1086,6 +1086,10 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, } } + if (top_level && (ops == &vdev_raidz_ops || ops == &vdev_draid_ops)) + vd->vdev_autosit = + vdev_prop_default_numeric(VDEV_PROP_AUTOSIT); + /* * Add ourselves to the parent's list of children. */ @@ -1187,6 +1191,9 @@ vdev_free(vdev_t *vd) spa_spare_remove(vd); if (vd->vdev_isl2cache) spa_l2cache_remove(vd); + if (vd->vdev_prev_histo) + kmem_free(vd->vdev_prev_histo, + sizeof (uint64_t) * VDEV_L_HISTO_BUCKETS); txg_list_destroy(&vd->vdev_ms_list); txg_list_destroy(&vd->vdev_dtl_list); @@ -3857,6 +3864,26 @@ vdev_load(vdev_t *vd) } } + if (vd == vd->vdev_top && vd->vdev_top_zap != 0) { + spa_t *spa = vd->vdev_spa; + uint64_t autosit; + + error = zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap, + vdev_prop_to_name(VDEV_PROP_AUTOSIT), sizeof (autosit), + 1, &autosit); + if (error == 0) { + vd->vdev_autosit = autosit == 1; + } else if (error == ENOENT) { + vd->vdev_autosit = vdev_prop_default_numeric( + VDEV_PROP_AUTOSIT); + } else { + vdev_dbgmsg(vd, + "vdev_load: zap_lookup(top_zap=%llu) " + "failed [error=%d]", + (u_longlong_t)vd->vdev_top_zap, error); + } + } + /* * Load any rebuild state from the top-level vdev zap. */ @@ -4616,6 +4643,8 @@ vdev_clear(spa_t *spa, vdev_t *vd) vd->vdev_stat.vs_checksum_errors = 0; vd->vdev_stat.vs_dio_verify_errors = 0; vd->vdev_stat.vs_slow_ios = 0; + atomic_store_64(&vd->vdev_outlier_count, 0); + vd->vdev_read_sit_out_expire = 0; for (int c = 0; c < vd->vdev_children; c++) vdev_clear(spa, vd->vdev_child[c]); @@ -6107,6 +6136,56 @@ vdev_prop_set(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl) } vd->vdev_failfast = intval & 1; break; + case VDEV_PROP_SIT_OUT: + /* Only expose this for a draid or raidz leaf */ + if (!vd->vdev_ops->vdev_op_leaf || + vd->vdev_top == NULL || + (vd->vdev_top->vdev_ops != &vdev_raidz_ops && + vd->vdev_top->vdev_ops != &vdev_draid_ops)) { + error = ENOTSUP; + break; + } + if (nvpair_value_uint64(elem, &intval) != 0) { + error = EINVAL; + break; + } + if (intval == 1) { + vdev_t *ancestor = vd; + while (ancestor->vdev_parent != vd->vdev_top) + ancestor = ancestor->vdev_parent; + vdev_t *pvd = vd->vdev_top; + uint_t sitouts = 0; + for (int i = 0; i < pvd->vdev_children; i++) { + if (pvd->vdev_child[i] == ancestor) + continue; + if (vdev_sit_out_reads( + pvd->vdev_child[i], 0)) { + sitouts++; + } + } + if (sitouts >= vdev_get_nparity(pvd)) { + error = ZFS_ERR_TOO_MANY_SITOUTS; + break; + } + if (error == 0) + vdev_raidz_sit_child(vd, + INT64_MAX - gethrestime_sec()); + } else { + vdev_raidz_unsit_child(vd); + } + break; + case VDEV_PROP_AUTOSIT: + if (vd->vdev_ops != &vdev_raidz_ops && + vd->vdev_ops != &vdev_draid_ops) { + error = ENOTSUP; + break; + } + if (nvpair_value_uint64(elem, &intval) != 0) { + error = EINVAL; + break; + } + vd->vdev_autosit = intval == 1; + break; case VDEV_PROP_CHECKSUM_N: if (nvpair_value_uint64(elem, &intval) != 0) { error = EINVAL; @@ -6456,6 +6535,19 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl) ZPROP_SRC_NONE); } continue; + case VDEV_PROP_SIT_OUT: + /* Only expose this for a draid or raidz leaf */ + if (vd->vdev_ops->vdev_op_leaf && + vd->vdev_top != NULL && + (vd->vdev_top->vdev_ops == + &vdev_raidz_ops || + vd->vdev_top->vdev_ops == + &vdev_draid_ops)) { + vdev_prop_add_list(outnvl, propname, + NULL, vdev_sit_out_reads(vd, 0), + ZPROP_SRC_NONE); + } + continue; case VDEV_PROP_TRIM_SUPPORT: /* only valid for leaf vdevs */ if (vd->vdev_ops->vdev_op_leaf) { @@ -6506,6 +6598,29 @@ vdev_prop_get(vdev_t *vd, nvlist_t *innvl, nvlist_t *outnvl) vdev_prop_add_list(outnvl, propname, strval, intval, src); break; + case VDEV_PROP_AUTOSIT: + /* Only raidz vdevs cannot have this property */ + if (vd->vdev_ops != &vdev_raidz_ops && + vd->vdev_ops != &vdev_draid_ops) { + src = ZPROP_SRC_NONE; + intval = ZPROP_BOOLEAN_NA; + } else { + err = vdev_prop_get_int(vd, prop, + &intval); + if (err && err != ENOENT) + break; + + if (intval == + vdev_prop_default_numeric(prop)) + src = ZPROP_SRC_DEFAULT; + else + src = ZPROP_SRC_LOCAL; + } + + vdev_prop_add_list(outnvl, propname, NULL, + intval, src); + break; + case VDEV_PROP_CHECKSUM_N: case VDEV_PROP_CHECKSUM_T: case VDEV_PROP_IO_N: diff --git a/sys/contrib/openzfs/module/zfs/vdev_draid.c b/sys/contrib/openzfs/module/zfs/vdev_draid.c index a05289102af2..8588cfee3f7d 100644 --- a/sys/contrib/openzfs/module/zfs/vdev_draid.c +++ b/sys/contrib/openzfs/module/zfs/vdev_draid.c @@ -22,6 +22,7 @@ /* * Copyright (c) 2018 Intel Corporation. * Copyright (c) 2020 by Lawrence Livermore National Security, LLC. + * Copyright (c) 2025, Klara, Inc. */ #include <sys/zfs_context.h> @@ -1996,6 +1997,33 @@ vdev_draid_io_start_read(zio_t *zio, raidz_row_t *rr) rc->rc_allow_repair = 1; } } + + if (vdev_sit_out_reads(cvd, zio->io_flags)) { + rr->rr_outlier_cnt++; + ASSERT0(rc->rc_latency_outlier); + rc->rc_latency_outlier = 1; + } + } + + /* + * When the row contains a latency outlier and sufficient parity + * exists to reconstruct the column data, then skip reading the + * known slow child vdev as a performance optimization. + */ + if (rr->rr_outlier_cnt > 0 && + (rr->rr_firstdatacol - rr->rr_missingparity) >= + (rr->rr_missingdata + 1)) { + + for (int c = rr->rr_cols - 1; c >= rr->rr_firstdatacol; c--) { + raidz_col_t *rc = &rr->rr_col[c]; + + if (rc->rc_error == 0 && rc->rc_latency_outlier) { + rr->rr_missingdata++; + rc->rc_error = SET_ERROR(EAGAIN); + rc->rc_skipped = 1; + break; + } + } } /* diff --git a/sys/contrib/openzfs/module/zfs/vdev_file.c b/sys/contrib/openzfs/module/zfs/vdev_file.c index f457669bc809..20b4db65ec06 100644 --- a/sys/contrib/openzfs/module/zfs/vdev_file.c +++ b/sys/contrib/openzfs/module/zfs/vdev_file.c @@ -228,7 +228,8 @@ vdev_file_io_strategy(void *arg) abd_return_buf_copy(zio->io_abd, buf, size); } else { buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size); - err = zfs_file_pwrite(vf->vf_file, buf, size, off, &resid); + err = zfs_file_pwrite(vf->vf_file, buf, size, off, + vd->vdev_ashift, &resid); abd_return_buf(zio->io_abd, buf, size); } zio->io_error = err; diff --git a/sys/contrib/openzfs/module/zfs/vdev_queue.c b/sys/contrib/openzfs/module/zfs/vdev_queue.c index c12713b107bf..e69e5598939e 100644 --- a/sys/contrib/openzfs/module/zfs/vdev_queue.c +++ b/sys/contrib/openzfs/module/zfs/vdev_queue.c @@ -122,7 +122,7 @@ * The maximum number of i/os active to each device. Ideally, this will be >= * the sum of each queue's max_active. */ -uint_t zfs_vdev_max_active = 1000; +static uint_t zfs_vdev_max_active = 1000; /* * Per-queue limits on the number of i/os active to each device. If the diff --git a/sys/contrib/openzfs/module/zfs/vdev_raidz.c b/sys/contrib/openzfs/module/zfs/vdev_raidz.c index b597d6daefde..80727b0d8f91 100644 --- a/sys/contrib/openzfs/module/zfs/vdev_raidz.c +++ b/sys/contrib/openzfs/module/zfs/vdev_raidz.c @@ -24,6 +24,7 @@ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012, 2020 by Delphix. All rights reserved. * Copyright (c) 2016 Gvozden Nešković. All rights reserved. + * Copyright (c) 2025, Klara, Inc. */ #include <sys/zfs_context.h> @@ -356,6 +357,32 @@ unsigned long raidz_expand_max_reflow_bytes = 0; uint_t raidz_expand_pause_point = 0; /* + * This represents the duration for a slow drive read sit out. + */ +static unsigned long vdev_read_sit_out_secs = 600; + +/* + * How often each RAID-Z and dRAID vdev will check for slow disk outliers. + * Increasing this interval will reduce the sensitivity of detection (since all + * I/Os since the last check are included in the statistics), but will slow the + * response to a disk developing a problem. + * + * Defaults to once per second; setting extremely small values may cause + * negative performance effects. + */ +static hrtime_t vdev_raidz_outlier_check_interval_ms = 1000; + +/* + * When performing slow outlier checks for RAID-Z and dRAID vdevs, this value is + * used to determine how far out an outlier must be before it counts as an event + * worth consdering. + * + * Smaller values will result in more aggressive sitting out of disks that may + * have problems, but may significantly increase the rate of spurious sit-outs. + */ +static uint32_t vdev_raidz_outlier_insensitivity = 50; + +/* * Maximum amount of copy io's outstanding at once. */ #ifdef _ILP32 @@ -2311,6 +2338,41 @@ vdev_raidz_min_asize(vdev_t *vd) vd->vdev_children); } +/* + * return B_TRUE if a read should be skipped due to being too slow. + * + * In vdev_child_slow_outlier() it looks for outliers based on disk + * latency from the most recent child reads. Here we're checking if, + * over time, a disk has has been an outlier too many times and is + * now in a sit out period. + */ +boolean_t +vdev_sit_out_reads(vdev_t *vd, zio_flag_t io_flags) +{ + if (vdev_read_sit_out_secs == 0) + return (B_FALSE); + + /* Avoid skipping a data column read when scrubbing */ + if (io_flags & ZIO_FLAG_SCRUB) + return (B_FALSE); + + if (!vd->vdev_ops->vdev_op_leaf) { + boolean_t sitting = B_FALSE; + for (int c = 0; c < vd->vdev_children; c++) { + sitting |= vdev_sit_out_reads(vd->vdev_child[c], + io_flags); + } + return (sitting); + } + + if (vd->vdev_read_sit_out_expire >= gethrestime_sec()) + return (B_TRUE); + + vd->vdev_read_sit_out_expire = 0; + + return (B_FALSE); +} + void vdev_raidz_child_done(zio_t *zio) { @@ -2475,6 +2537,45 @@ vdev_raidz_io_start_read_row(zio_t *zio, raidz_row_t *rr, boolean_t forceparity) rc->rc_skipped = 1; continue; } + + if (vdev_sit_out_reads(cvd, zio->io_flags)) { + rr->rr_outlier_cnt++; + ASSERT0(rc->rc_latency_outlier); + rc->rc_latency_outlier = 1; + } + } + + /* + * When the row contains a latency outlier and sufficient parity + * exists to reconstruct the column data, then skip reading the + * known slow child vdev as a performance optimization. + */ + if (rr->rr_outlier_cnt > 0 && + (rr->rr_firstdatacol - rr->rr_missingparity) >= + (rr->rr_missingdata + 1)) { + + for (int c = rr->rr_cols - 1; c >= 0; c--) { + raidz_col_t *rc = &rr->rr_col[c]; + + if (rc->rc_error == 0 && rc->rc_latency_outlier) { + if (c >= rr->rr_firstdatacol) + rr->rr_missingdata++; + else + rr->rr_missingparity++; + rc->rc_error = SET_ERROR(EAGAIN); + rc->rc_skipped = 1; + break; + } + } + } + + for (int c = rr->rr_cols - 1; c >= 0; c--) { + raidz_col_t *rc = &rr->rr_col[c]; + vdev_t *cvd = vd->vdev_child[rc->rc_devidx]; + + if (rc->rc_error || rc->rc_size == 0) + continue; + if (forceparity || c >= rr->rr_firstdatacol || rr->rr_missingdata > 0 || (zio->io_flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER))) { @@ -2498,6 +2599,7 @@ vdev_raidz_io_start_read_phys_cols(zio_t *zio, raidz_map_t *rm) ASSERT3U(prc->rc_devidx, ==, i); vdev_t *cvd = vd->vdev_child[i]; + if (!vdev_readable(cvd)) { prc->rc_error = SET_ERROR(ENXIO); prc->rc_tried = 1; /* don't even try */ @@ -2774,6 +2876,239 @@ vdev_raidz_worst_error(raidz_row_t *rr) return (error); } +/* + * Find the median value from a set of n values + */ +static uint64_t +latency_median_value(const uint64_t *data, size_t n) +{ + uint64_t m; + + if (n % 2 == 0) + m = (data[(n >> 1) - 1] + data[n >> 1]) >> 1; + else + m = data[((n + 1) >> 1) - 1]; + + return (m); +} + +/* + * Calculate the outlier fence from a set of n latency values + * + * fence = Q3 + vdev_raidz_outlier_insensitivity x (Q3 - Q1) + */ +static uint64_t +latency_quartiles_fence(const uint64_t *data, size_t n, uint64_t *iqr) +{ + uint64_t q1 = latency_median_value(&data[0], n >> 1); + uint64_t q3 = latency_median_value(&data[(n + 1) >> 1], n >> 1); + + /* + * To avoid detecting false positive outliers when N is small and + * and the latencies values are very close, make sure the IQR + * is at least 25% larger than Q1. + */ + *iqr = MAX(q3 - q1, q1 / 4); + + return (q3 + (*iqr * vdev_raidz_outlier_insensitivity)); +} +#define LAT_CHILDREN_MIN 5 +#define LAT_OUTLIER_LIMIT 20 + +static int +latency_compare(const void *arg1, const void *arg2) +{ + const uint64_t *l1 = (uint64_t *)arg1; + const uint64_t *l2 = (uint64_t *)arg2; + + return (TREE_CMP(*l1, *l2)); +} + +void +vdev_raidz_sit_child(vdev_t *svd, uint64_t secs) +{ + for (int c = 0; c < svd->vdev_children; c++) + vdev_raidz_sit_child(svd->vdev_child[c], secs); + + if (!svd->vdev_ops->vdev_op_leaf) + return; + + /* Begin a sit out period for this slow drive */ + svd->vdev_read_sit_out_expire = gethrestime_sec() + + secs; + + /* Count each slow io period */ + mutex_enter(&svd->vdev_stat_lock); + svd->vdev_stat.vs_slow_ios++; + mutex_exit(&svd->vdev_stat_lock); +} + +void +vdev_raidz_unsit_child(vdev_t *vd) +{ + for (int c = 0; c < vd->vdev_children; c++) + vdev_raidz_unsit_child(vd->vdev_child[c]); + + if (!vd->vdev_ops->vdev_op_leaf) + return; + + vd->vdev_read_sit_out_expire = 0; +} + +/* + * Check for any latency outlier from latest set of child reads. + * + * Uses a Tukey's fence, with K = 50, for detecting extreme outliers. This + * rule defines extreme outliers as data points outside the fence of the + * third quartile plus fifty times the Interquartile Range (IQR). This range + * is the distance between the first and third quartile. + * + * Fifty is an extremely large value for Tukey's fence, but the outliers we're + * attempting to detect here are orders of magnitude times larger than the + * median. This large value should capture any truly fault disk quickly, + * without causing spurious sit-outs. + * + * To further avoid spurious sit-outs, vdevs must be detected multiple times + * as an outlier before they are sat, and outlier counts will gradually decay. + * Every nchildren times we have detected an outlier, we subtract 2 from the + * outlier count of all children. If detected outliers are close to uniformly + * distributed, this will result in the outlier count remaining close to 0 + * (in expectation; over long enough time-scales, spurious sit-outs are still + * possible). + */ +static void +vdev_child_slow_outlier(zio_t *zio) +{ + vdev_t *vd = zio->io_vd; + if (!vd->vdev_autosit || vdev_read_sit_out_secs == 0 || + vd->vdev_children < LAT_CHILDREN_MIN) + return; + + hrtime_t now = getlrtime(); + uint64_t last = atomic_load_64(&vd->vdev_last_latency_check); + + if ((now - last) < MSEC2NSEC(vdev_raidz_outlier_check_interval_ms)) + return; + + /* Allow a single winner when there are racing callers. */ + if (atomic_cas_64(&vd->vdev_last_latency_check, last, now) != last) + return; + + int children = vd->vdev_children; + uint64_t *lat_data = kmem_alloc(sizeof (uint64_t) * children, KM_SLEEP); + + for (int c = 0; c < children; c++) { + vdev_t *cvd = vd->vdev_child[c]; + if (cvd->vdev_prev_histo == NULL) { + mutex_enter(&cvd->vdev_stat_lock); + size_t size = + sizeof (cvd->vdev_stat_ex.vsx_disk_histo[0]); + cvd->vdev_prev_histo = kmem_zalloc(size, KM_SLEEP); + memcpy(cvd->vdev_prev_histo, + cvd->vdev_stat_ex.vsx_disk_histo[ZIO_TYPE_READ], + size); + mutex_exit(&cvd->vdev_stat_lock); + } + } + uint64_t max = 0; + vdev_t *svd = NULL; + uint_t sitouts = 0; + boolean_t skip = B_FALSE, svd_sitting = B_FALSE; + for (int c = 0; c < children; c++) { + vdev_t *cvd = vd->vdev_child[c]; + boolean_t sitting = vdev_sit_out_reads(cvd, 0) || + cvd->vdev_state != VDEV_STATE_HEALTHY; + + /* We can't sit out more disks than we have parity */ + if (sitting && ++sitouts >= vdev_get_nparity(vd)) + skip = B_TRUE; + + mutex_enter(&cvd->vdev_stat_lock); + + uint64_t *prev_histo = cvd->vdev_prev_histo; + uint64_t *histo = + cvd->vdev_stat_ex.vsx_disk_histo[ZIO_TYPE_READ]; + if (skip) { + size_t size = + sizeof (cvd->vdev_stat_ex.vsx_disk_histo[0]); + memcpy(prev_histo, histo, size); + mutex_exit(&cvd->vdev_stat_lock); + continue; + } + uint64_t count = 0; + lat_data[c] = 0; + for (int i = 0; i < VDEV_L_HISTO_BUCKETS; i++) { + uint64_t this_count = histo[i] - prev_histo[i]; + lat_data[c] += (1ULL << i) * this_count; + count += this_count; + } + size_t size = sizeof (cvd->vdev_stat_ex.vsx_disk_histo[0]); + memcpy(prev_histo, histo, size); + mutex_exit(&cvd->vdev_stat_lock); + lat_data[c] /= MAX(1, count); + + /* Wait until all disks have been read from */ + if (lat_data[c] == 0 && !sitting) { + skip = B_TRUE; + continue; + } + + /* Keep track of the vdev with largest value */ + if (lat_data[c] > max) { + max = lat_data[c]; + svd = cvd; + svd_sitting = sitting; + } + } + + if (skip) { + kmem_free(lat_data, sizeof (uint64_t) * children); + return; + } + + qsort((void *)lat_data, children, sizeof (uint64_t), latency_compare); + + uint64_t iqr; + uint64_t fence = latency_quartiles_fence(lat_data, children, &iqr); + + ASSERT3U(lat_data[children - 1], ==, max); + if (max > fence && !svd_sitting) { + ASSERT3U(iqr, >, 0); + uint64_t incr = MAX(1, MIN((max - fence) / iqr, + LAT_OUTLIER_LIMIT / 4)); + vd->vdev_outlier_count += incr; + if (vd->vdev_outlier_count >= children) { + for (int c = 0; c < children; c++) { + vdev_t *cvd = vd->vdev_child[c]; + cvd->vdev_outlier_count -= 2; + cvd->vdev_outlier_count = MAX(0, + cvd->vdev_outlier_count); + } + vd->vdev_outlier_count = 0; + } + /* + * Keep track of how many times this child has had + * an outlier read. A disk that persitently has a + * higher than peers outlier count will be considered + * a slow disk. + */ + svd->vdev_outlier_count += incr; + if (svd->vdev_outlier_count > LAT_OUTLIER_LIMIT) { + ASSERT0(svd->vdev_read_sit_out_expire); + vdev_raidz_sit_child(svd, vdev_read_sit_out_secs); + (void) zfs_ereport_post(FM_EREPORT_ZFS_SITOUT, + zio->io_spa, svd, NULL, NULL, 0); + vdev_dbgmsg(svd, "begin read sit out for %d secs", + (int)vdev_read_sit_out_secs); + + for (int c = 0; c < vd->vdev_children; c++) + vd->vdev_child[c]->vdev_outlier_count = 0; + } + } + + kmem_free(lat_data, sizeof (uint64_t) * children); +} + static void vdev_raidz_io_done_verified(zio_t *zio, raidz_row_t *rr) { @@ -3515,6 +3850,9 @@ vdev_raidz_io_done(zio_t *zio) raidz_row_t *rr = rm->rm_row[i]; vdev_raidz_io_done_verified(zio, rr); } + /* Periodically check for a read outlier */ + if (zio->io_type == ZIO_TYPE_READ) + vdev_child_slow_outlier(zio); zio_checksum_verified(zio); } else { /* @@ -5155,3 +5493,10 @@ ZFS_MODULE_PARAM(zfs_vdev, raidz_, io_aggregate_rows, ULONG, ZMOD_RW, ZFS_MODULE_PARAM(zfs, zfs_, scrub_after_expand, INT, ZMOD_RW, "For expanded RAIDZ, automatically start a pool scrub when expansion " "completes"); +ZFS_MODULE_PARAM(zfs_vdev, vdev_, read_sit_out_secs, ULONG, ZMOD_RW, + "Raidz/draid slow disk sit out time period in seconds"); +ZFS_MODULE_PARAM(zfs_vdev, vdev_, raidz_outlier_check_interval_ms, ULONG, + ZMOD_RW, "Interval to check for slow raidz/draid children"); +ZFS_MODULE_PARAM(zfs_vdev, vdev_, raidz_outlier_insensitivity, UINT, + ZMOD_RW, "How insensitive the slow raidz/draid child check should be"); +/* END CSTYLED */ diff --git a/sys/contrib/openzfs/module/zfs/vdev_removal.c b/sys/contrib/openzfs/module/zfs/vdev_removal.c index 2f7a739da241..2ce0121324ad 100644 --- a/sys/contrib/openzfs/module/zfs/vdev_removal.c +++ b/sys/contrib/openzfs/module/zfs/vdev_removal.c @@ -105,7 +105,7 @@ static const uint_t zfs_remove_max_copy_bytes = 64 * 1024 * 1024; * * See also the accessor function spa_remove_max_segment(). */ -uint_t zfs_remove_max_segment = SPA_MAXBLOCKSIZE; +static uint_t zfs_remove_max_segment = SPA_MAXBLOCKSIZE; /* * Ignore hard IO errors during device removal. When set if a device @@ -137,7 +137,7 @@ uint_t vdev_removal_max_span = 32 * 1024; * This is used by the test suite so that it can ensure that certain * actions happen while in the middle of a removal. */ -int zfs_removal_suspend_progress = 0; +static int zfs_removal_suspend_progress = 0; #define VDEV_REMOVAL_ZAP_OBJS "lzap" diff --git a/sys/contrib/openzfs/module/zfs/zfeature.c b/sys/contrib/openzfs/module/zfs/zfeature.c index 0816ea134bf3..4cf9e0dbb405 100644 --- a/sys/contrib/openzfs/module/zfs/zfeature.c +++ b/sys/contrib/openzfs/module/zfs/zfeature.c @@ -308,6 +308,7 @@ feature_sync(spa_t *spa, zfeature_info_t *feature, uint64_t refcount, ASSERT(VALID_FEATURE_OR_NONE(feature->fi_feature)); uint64_t zapobj = (feature->fi_flags & ZFEATURE_FLAG_READONLY_COMPAT) ? spa->spa_feat_for_write_obj : spa->spa_feat_for_read_obj; + ASSERT(MUTEX_HELD(&spa->spa_feat_stats_lock)); VERIFY0(zap_update(spa->spa_meta_objset, zapobj, feature->fi_guid, sizeof (uint64_t), 1, &refcount, tx)); @@ -360,7 +361,9 @@ feature_enable_sync(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx) feature->fi_guid, 1, strlen(feature->fi_desc) + 1, feature->fi_desc, tx)); + mutex_enter(&spa->spa_feat_stats_lock); feature_sync(spa, feature, initial_refcount, tx); + mutex_exit(&spa->spa_feat_stats_lock); if (spa_feature_is_enabled(spa, SPA_FEATURE_ENABLED_TXG)) { uint64_t enabling_txg = dmu_tx_get_txg(tx); @@ -416,6 +419,7 @@ feature_do_action(spa_t *spa, spa_feature_t fid, feature_action_t action, ASSERT(dmu_tx_is_syncing(tx)); ASSERT3U(spa_version(spa), >=, SPA_VERSION_FEATURES); + mutex_enter(&spa->spa_feat_stats_lock); VERIFY3U(feature_get_refcount(spa, feature, &refcount), !=, ENOTSUP); switch (action) { @@ -433,6 +437,7 @@ feature_do_action(spa_t *spa, spa_feature_t fid, feature_action_t action, } feature_sync(spa, feature, refcount, tx); + mutex_exit(&spa->spa_feat_stats_lock); } void diff --git a/sys/contrib/openzfs/module/zfs/zfs_crrd.c b/sys/contrib/openzfs/module/zfs/zfs_crrd.c index f9267ed41d71..30d4c7c36897 100644 --- a/sys/contrib/openzfs/module/zfs/zfs_crrd.c +++ b/sys/contrib/openzfs/module/zfs/zfs_crrd.c @@ -162,9 +162,9 @@ dbrrd_add(dbrrd_t *db, hrtime_t time, uint64_t txg) daydiff = time - rrd_tail(&db->dbr_days); monthdiff = time - rrd_tail(&db->dbr_months); - if (monthdiff >= 0 && monthdiff >= SEC2NSEC(30 * 24 * 60 * 60)) + if (monthdiff >= 0 && monthdiff >= 30 * 24 * 60 * 60) rrd_add(&db->dbr_months, time, txg); - else if (daydiff >= 0 && daydiff >= SEC2NSEC(24 * 60 * 60)) + else if (daydiff >= 0 && daydiff >= 24 * 60 * 60) rrd_add(&db->dbr_days, time, txg); else if (minutedif >= 0) rrd_add(&db->dbr_minutes, time, txg); @@ -208,7 +208,8 @@ dbrrd_closest(hrtime_t tv, const rrd_data_t *r1, const rrd_data_t *r2) if (r2 == NULL) return (r1); - return (ABS(tv - r1->rrdd_time) < ABS(tv - r2->rrdd_time) ? r1 : r2); + return (ABS(tv - (hrtime_t)r1->rrdd_time) < + ABS(tv - (hrtime_t)r2->rrdd_time) ? r1 : r2); } uint64_t diff --git a/sys/contrib/openzfs/module/zfs/zfs_ioctl.c b/sys/contrib/openzfs/module/zfs/zfs_ioctl.c index 121b966b9864..5ca7c2320c4e 100644 --- a/sys/contrib/openzfs/module/zfs/zfs_ioctl.c +++ b/sys/contrib/openzfs/module/zfs/zfs_ioctl.c @@ -683,6 +683,7 @@ zfs_secpolicy_send(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) dsl_dataset_t *ds; const char *cp; int error; + boolean_t rawok = (zc->zc_flags & 0x8); /* * Generate the current snapshot name from the given objsetid, then @@ -705,6 +706,10 @@ zfs_secpolicy_send(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) error = zfs_secpolicy_write_perms_ds(zc->zc_name, ds, ZFS_DELEG_PERM_SEND, cr); + if (error != 0 && rawok == B_TRUE) { + error = zfs_secpolicy_write_perms_ds(zc->zc_name, ds, + ZFS_DELEG_PERM_SEND_RAW, cr); + } dsl_dataset_rele(ds, FTAG); dsl_pool_rele(dp, FTAG); @@ -714,9 +719,17 @@ zfs_secpolicy_send(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) static int zfs_secpolicy_send_new(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr) { + boolean_t rawok = nvlist_exists(innvl, "rawok"); + int error; + (void) innvl; - return (zfs_secpolicy_write_perms(zc->zc_name, - ZFS_DELEG_PERM_SEND, cr)); + error = zfs_secpolicy_write_perms(zc->zc_name, + ZFS_DELEG_PERM_SEND, cr); + if (error != 0 && rawok == B_TRUE) { + error = zfs_secpolicy_write_perms(zc->zc_name, + ZFS_DELEG_PERM_SEND_RAW, cr); + } + return (error); } static int @@ -4726,7 +4739,7 @@ zfs_ioc_rollback(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl) error = error ? error : resume_err; } zfs_vfs_rele(zfsvfs); - } else if ((zv = zvol_suspend(fsname)) != NULL) { + } else if (zvol_suspend(fsname, &zv) == 0) { error = dsl_dataset_rollback(fsname, target, zvol_tag(zv), outnvl); zvol_resume(zv); @@ -5448,7 +5461,7 @@ zfs_ioc_recv_impl(char *tofs, char *tosnap, const char *origin, } error = error ? error : end_err; zfs_vfs_rele(zfsvfs); - } else if ((zv = zvol_suspend(tofs)) != NULL) { + } else if (zvol_suspend(tofs, &zv) == 0) { error = dmu_recv_end(&drc, zvol_tag(zv)); zvol_resume(zv); } else { @@ -7619,7 +7632,7 @@ zfs_ioctl_init(void) zfs_ioctl_register("scrub", ZFS_IOC_POOL_SCRUB, zfs_ioc_pool_scrub, zfs_secpolicy_config, POOL_NAME, - POOL_CHECK_NONE, B_TRUE, B_TRUE, + POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_TRUE, B_TRUE, zfs_keys_pool_scrub, ARRAY_SIZE(zfs_keys_pool_scrub)); zfs_ioctl_register("get_props", ZFS_IOC_POOL_GET_PROPS, diff --git a/sys/contrib/openzfs/module/zfs/zil.c b/sys/contrib/openzfs/module/zfs/zil.c index 31b59c55f17b..0307df55aa21 100644 --- a/sys/contrib/openzfs/module/zfs/zil.c +++ b/sys/contrib/openzfs/module/zfs/zil.c @@ -819,34 +819,37 @@ zil_lwb_vdev_compare(const void *x1, const void *x2) * we choose them here and later make the block allocation match. */ static lwb_t * -zil_alloc_lwb(zilog_t *zilog, int sz, blkptr_t *bp, boolean_t slog, - uint64_t txg, lwb_state_t state) +zil_alloc_lwb(zilog_t *zilog, blkptr_t *bp, int min_sz, int sz, + boolean_t slog, uint64_t txg) { lwb_t *lwb; lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP); + lwb->lwb_flags = 0; lwb->lwb_zilog = zilog; if (bp) { lwb->lwb_blk = *bp; - lwb->lwb_slim = (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2); + if (BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_ZILOG2) + lwb->lwb_flags |= LWB_FLAG_SLIM; sz = BP_GET_LSIZE(bp); + lwb->lwb_min_sz = sz; } else { BP_ZERO(&lwb->lwb_blk); - lwb->lwb_slim = (spa_version(zilog->zl_spa) >= - SPA_VERSION_SLIM_ZIL); + if (spa_version(zilog->zl_spa) >= SPA_VERSION_SLIM_ZIL) + lwb->lwb_flags |= LWB_FLAG_SLIM; + lwb->lwb_min_sz = min_sz; } - lwb->lwb_slog = slog; + if (slog) + lwb->lwb_flags |= LWB_FLAG_SLOG; lwb->lwb_error = 0; - if (lwb->lwb_slim) { - lwb->lwb_nmax = sz; - lwb->lwb_nused = lwb->lwb_nfilled = sizeof (zil_chain_t); - } else { - lwb->lwb_nmax = sz - sizeof (zil_chain_t); - lwb->lwb_nused = lwb->lwb_nfilled = 0; - } + /* + * Buffer allocation and capacity setup will be done in + * zil_lwb_write_open() when the LWB is opened for ITX assignment. + */ + lwb->lwb_nmax = lwb->lwb_nused = lwb->lwb_nfilled = 0; lwb->lwb_sz = sz; - lwb->lwb_state = state; - lwb->lwb_buf = zio_buf_alloc(sz); + lwb->lwb_buf = NULL; + lwb->lwb_state = LWB_STATE_NEW; lwb->lwb_child_zio = NULL; lwb->lwb_write_zio = NULL; lwb->lwb_root_zio = NULL; @@ -857,8 +860,6 @@ zil_alloc_lwb(zilog_t *zilog, int sz, blkptr_t *bp, boolean_t slog, mutex_enter(&zilog->zl_lock); list_insert_tail(&zilog->zl_lwb_list, lwb); - if (state != LWB_STATE_NEW) - zilog->zl_last_lwb_opened = lwb; mutex_exit(&zilog->zl_lock); return (lwb); @@ -878,7 +879,7 @@ zil_free_lwb(zilog_t *zilog, lwb_t *lwb) VERIFY(list_is_empty(&lwb->lwb_itxs)); VERIFY(list_is_empty(&lwb->lwb_waiters)); ASSERT(avl_is_empty(&lwb->lwb_vdev_tree)); - ASSERT(!MUTEX_HELD(&lwb->lwb_vdev_lock)); + ASSERT(!MUTEX_HELD(&lwb->lwb_lock)); /* * Clear the zilog's field to indicate this lwb is no longer @@ -1019,7 +1020,7 @@ zil_create(zilog_t *zilog) } error = zio_alloc_zil(zilog->zl_spa, zilog->zl_os, txg, &blk, - ZIL_MIN_BLKSZ, &slog); + ZIL_MIN_BLKSZ, ZIL_MIN_BLKSZ, &slog, B_TRUE); if (error == 0) zil_init_log_chain(zilog, &blk); } @@ -1028,7 +1029,7 @@ zil_create(zilog_t *zilog) * Allocate a log write block (lwb) for the first log block. */ if (error == 0) - lwb = zil_alloc_lwb(zilog, 0, &blk, slog, txg, LWB_STATE_NEW); + lwb = zil_alloc_lwb(zilog, &blk, 0, 0, slog, txg); /* * If we just allocated the first log block, commit our transaction @@ -1324,10 +1325,12 @@ zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds, void *tx) * zil_commit() is racing with spa_sync(). */ static void -zil_commit_waiter_skip(zil_commit_waiter_t *zcw) +zil_commit_waiter_done(zil_commit_waiter_t *zcw, int err) { mutex_enter(&zcw->zcw_lock); ASSERT3B(zcw->zcw_done, ==, B_FALSE); + zcw->zcw_lwb = NULL; + zcw->zcw_error = err; zcw->zcw_done = B_TRUE; cv_broadcast(&zcw->zcw_cv); mutex_exit(&zcw->zcw_lock); @@ -1389,7 +1392,7 @@ zil_lwb_add_block(lwb_t *lwb, const blkptr_t *bp) if (zil_nocacheflush) return; - mutex_enter(&lwb->lwb_vdev_lock); + mutex_enter(&lwb->lwb_lock); for (i = 0; i < ndvas; i++) { zvsearch.zv_vdev = DVA_GET_VDEV(&bp->blk_dva[i]); if (avl_find(t, &zvsearch, &where) == NULL) { @@ -1398,7 +1401,7 @@ zil_lwb_add_block(lwb_t *lwb, const blkptr_t *bp) avl_insert(t, zv, where); } } - mutex_exit(&lwb->lwb_vdev_lock); + mutex_exit(&lwb->lwb_lock); } static void @@ -1415,12 +1418,12 @@ zil_lwb_flush_defer(lwb_t *lwb, lwb_t *nlwb) /* * While 'lwb' is at a point in its lifetime where lwb_vdev_tree does - * not need the protection of lwb_vdev_lock (it will only be modified + * not need the protection of lwb_lock (it will only be modified * while holding zilog->zl_lock) as its writes and those of its * children have all completed. The younger 'nlwb' may be waiting on * future writes to additional vdevs. */ - mutex_enter(&nlwb->lwb_vdev_lock); + mutex_enter(&nlwb->lwb_lock); /* * Tear down the 'lwb' vdev tree, ensuring that entries which do not * exist in 'nlwb' are moved to it, freeing any would-be duplicates. @@ -1434,7 +1437,7 @@ zil_lwb_flush_defer(lwb_t *lwb, lwb_t *nlwb) kmem_free(zv, sizeof (*zv)); } } - mutex_exit(&nlwb->lwb_vdev_lock); + mutex_exit(&nlwb->lwb_lock); } void @@ -1491,10 +1494,6 @@ zil_lwb_flush_vdevs_done(zio_t *zio) zil_itx_destroy(itx, 0); while ((zcw = list_remove_head(&lwb->lwb_waiters)) != NULL) { - mutex_enter(&zcw->zcw_lock); - - ASSERT3P(zcw->zcw_lwb, ==, lwb); - zcw->zcw_lwb = NULL; /* * We expect any ZIO errors from child ZIOs to have been * propagated "up" to this specific LWB's root ZIO, in @@ -1509,14 +1508,7 @@ zil_lwb_flush_vdevs_done(zio_t *zio) * errors not being handled correctly here. See the * comment above the call to "zio_flush" for details. */ - - zcw->zcw_zio_error = zio->io_error; - - ASSERT3B(zcw->zcw_done, ==, B_FALSE); - zcw->zcw_done = B_TRUE; - cv_broadcast(&zcw->zcw_cv); - - mutex_exit(&zcw->zcw_lock); + zil_commit_waiter_done(zcw, zio->io_error); } uint64_t txg = lwb->lwb_issued_txg; @@ -1588,7 +1580,7 @@ zil_lwb_write_done(zio_t *zio) avl_tree_t *t = &lwb->lwb_vdev_tree; void *cookie = NULL; zil_vdev_node_t *zv; - lwb_t *nlwb; + lwb_t *nlwb = NULL; ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), !=, 0); @@ -1608,9 +1600,11 @@ zil_lwb_write_done(zio_t *zio) * its write ZIO a parent this ZIO. In such case we can not defer * our flushes or below may be a race between the done callbacks. */ - nlwb = list_next(&zilog->zl_lwb_list, lwb); - if (nlwb && nlwb->lwb_state != LWB_STATE_ISSUED) - nlwb = NULL; + if (!(lwb->lwb_flags & LWB_FLAG_CRASHED)) { + nlwb = list_next(&zilog->zl_lwb_list, lwb); + if (nlwb && nlwb->lwb_state != LWB_STATE_ISSUED) + nlwb = NULL; + } mutex_exit(&zilog->zl_lock); if (avl_numnodes(t) == 0) @@ -1624,12 +1618,17 @@ zil_lwb_write_done(zio_t *zio) * written out. * * Additionally, we don't perform any further error handling at - * this point (e.g. setting "zcw_zio_error" appropriately), as - * we expect that to occur in "zil_lwb_flush_vdevs_done" (thus, - * we expect any error seen here, to have been propagated to - * that function). + * this point (e.g. setting "zcw_error" appropriately), as we + * expect that to occur in "zil_lwb_flush_vdevs_done" (thus, we + * expect any error seen here, to have been propagated to that + * function). + * + * Note that we treat a "crashed" LWB as though it was in error, + * even if it did appear to succeed, because we've already + * signaled error and cleaned up waiters and committers in + * zil_crash(); we just want to clean up and get out of here. */ - if (zio->io_error != 0) { + if (zio->io_error != 0 || (lwb->lwb_flags & LWB_FLAG_CRASHED)) { while ((zv = avl_destroy_nodes(t, &cookie)) != NULL) kmem_free(zv, sizeof (*zv)); return; @@ -1742,10 +1741,26 @@ zil_lwb_write_open(zilog_t *zilog, lwb_t *lwb) return; } + mutex_enter(&lwb->lwb_lock); mutex_enter(&zilog->zl_lock); lwb->lwb_state = LWB_STATE_OPENED; zilog->zl_last_lwb_opened = lwb; mutex_exit(&zilog->zl_lock); + mutex_exit(&lwb->lwb_lock); + + /* + * Allocate buffer and set up LWB capacities. + */ + ASSERT0P(lwb->lwb_buf); + ASSERT3U(lwb->lwb_sz, >, 0); + lwb->lwb_buf = zio_buf_alloc(lwb->lwb_sz); + if (lwb->lwb_flags & LWB_FLAG_SLIM) { + lwb->lwb_nmax = lwb->lwb_sz; + lwb->lwb_nused = lwb->lwb_nfilled = sizeof (zil_chain_t); + } else { + lwb->lwb_nmax = lwb->lwb_sz - sizeof (zil_chain_t); + lwb->lwb_nused = lwb->lwb_nfilled = 0; + } } /* @@ -1762,6 +1777,8 @@ static uint_t zil_lwb_plan(zilog_t *zilog, uint64_t size, uint_t *minsize) { uint_t md = zilog->zl_max_block_size - sizeof (zil_chain_t); + uint_t waste = zil_max_waste_space(zilog); + waste = MAX(waste, zilog->zl_cur_max); if (size <= md) { /* @@ -1772,9 +1789,10 @@ zil_lwb_plan(zilog_t *zilog, uint64_t size, uint_t *minsize) } else if (size > 8 * md) { /* * Big bursts use maximum blocks. The first block size - * is hard to predict, but it does not really matter. + * is hard to predict, but we need at least enough space + * to make reasonable progress. */ - *minsize = 0; + *minsize = waste; return (md); } @@ -1787,57 +1805,52 @@ zil_lwb_plan(zilog_t *zilog, uint64_t size, uint_t *minsize) uint_t s = size; uint_t n = DIV_ROUND_UP(s, md - sizeof (lr_write_t)); uint_t chunk = DIV_ROUND_UP(s, n); - uint_t waste = zil_max_waste_space(zilog); - waste = MAX(waste, zilog->zl_cur_max); if (chunk <= md - waste) { *minsize = MAX(s - (md - waste) * (n - 1), waste); return (chunk); } else { - *minsize = 0; + *minsize = waste; return (md); } } /* * Try to predict next block size based on previous history. Make prediction - * sufficient for 7 of 8 previous bursts. Don't try to save if the saving is - * less then 50%, extra writes may cost more, but we don't want single spike - * to badly affect our predictions. + * sufficient for 7 of 8 previous bursts, but don't try to save if the saving + * is less then 50%. Extra writes may cost more, but we don't want single + * spike to badly affect our predictions. */ -static uint_t -zil_lwb_predict(zilog_t *zilog) +static void +zil_lwb_predict(zilog_t *zilog, uint64_t *min_predict, uint64_t *max_predict) { - uint_t m, o; + uint_t m1 = 0, m2 = 0, o; - /* If we are in the middle of a burst, take it into account also. */ - if (zilog->zl_cur_size > 0) { - o = zil_lwb_plan(zilog, zilog->zl_cur_size, &m); - } else { + /* If we are in the middle of a burst, take it as another data point. */ + if (zilog->zl_cur_size > 0) + o = zil_lwb_plan(zilog, zilog->zl_cur_size, &m1); + else o = UINT_MAX; - m = 0; - } - /* Find minimum optimal size. We don't need to go below that. */ - for (int i = 0; i < ZIL_BURSTS; i++) - o = MIN(o, zilog->zl_prev_opt[i]); - - /* Find two biggest minimal first block sizes above the optimal. */ - uint_t m1 = MAX(m, o), m2 = o; + /* Find two largest minimal first block sizes. */ for (int i = 0; i < ZIL_BURSTS; i++) { - m = zilog->zl_prev_min[i]; - if (m >= m1) { + uint_t cur = zilog->zl_prev_min[i]; + if (cur >= m1) { m2 = m1; - m1 = m; - } else if (m > m2) { - m2 = m; + m1 = cur; + } else if (cur > m2) { + m2 = cur; } } - /* - * If second minimum size gives 50% saving -- use it. It may cost us - * one additional write later, but the space saving is just too big. - */ - return ((m1 < m2 * 2) ? m1 : m2); + /* Minimum should guarantee progress in most cases. */ + *min_predict = (m1 < m2 * 2) ? m1 : m2; + + /* Maximum doesn't need to go below the minimum optimal size. */ + for (int i = 0; i < ZIL_BURSTS; i++) + o = MIN(o, zilog->zl_prev_opt[i]); + m1 = MAX(m1, o); + m2 = MAX(m2, o); + *max_predict = (m1 < m2 * 2) ? m1 : m2; } /* @@ -1845,12 +1858,13 @@ zil_lwb_predict(zilog_t *zilog) * Has to be called under zl_issuer_lock to chain more lwbs. */ static lwb_t * -zil_lwb_write_close(zilog_t *zilog, lwb_t *lwb, lwb_state_t state) +zil_lwb_write_close(zilog_t *zilog, lwb_t *lwb) { - uint64_t blksz, plan, plan2; + uint64_t minbs, maxbs; ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT3S(lwb->lwb_state, ==, LWB_STATE_OPENED); + membar_producer(); lwb->lwb_state = LWB_STATE_CLOSED; /* @@ -1875,27 +1889,34 @@ zil_lwb_write_close(zilog_t *zilog, lwb_t *lwb, lwb_state_t state) * Try to predict what can it be and plan for the worst case. */ uint_t m; - plan = zil_lwb_plan(zilog, zilog->zl_cur_left, &m); + maxbs = zil_lwb_plan(zilog, zilog->zl_cur_left, &m); + minbs = m; if (zilog->zl_parallel) { - plan2 = zil_lwb_plan(zilog, zilog->zl_cur_left + - zil_lwb_predict(zilog), &m); - if (plan < plan2) - plan = plan2; + uint64_t minp, maxp; + zil_lwb_predict(zilog, &minp, &maxp); + maxp = zil_lwb_plan(zilog, zilog->zl_cur_left + maxp, + &m); + if (maxbs < maxp) + maxbs = maxp; } } else { /* * The previous burst is done and we can only predict what * will come next. */ - plan = zil_lwb_predict(zilog); + zil_lwb_predict(zilog, &minbs, &maxbs); } - blksz = plan + sizeof (zil_chain_t); - blksz = P2ROUNDUP_TYPED(blksz, ZIL_MIN_BLKSZ, uint64_t); - blksz = MIN(blksz, zilog->zl_max_block_size); - DTRACE_PROBE3(zil__block__size, zilog_t *, zilog, uint64_t, blksz, - uint64_t, plan); - return (zil_alloc_lwb(zilog, blksz, NULL, 0, 0, state)); + minbs += sizeof (zil_chain_t); + maxbs += sizeof (zil_chain_t); + minbs = P2ROUNDUP_TYPED(minbs, ZIL_MIN_BLKSZ, uint64_t); + maxbs = P2ROUNDUP_TYPED(maxbs, ZIL_MIN_BLKSZ, uint64_t); + maxbs = MIN(maxbs, zilog->zl_max_block_size); + minbs = MIN(minbs, maxbs); + DTRACE_PROBE3(zil__block__size, zilog_t *, zilog, uint64_t, minbs, + uint64_t, maxbs); + + return (zil_alloc_lwb(zilog, NULL, minbs, maxbs, 0, 0)); } /* @@ -1944,14 +1965,16 @@ zil_lwb_write_issue(zilog_t *zilog, lwb_t *lwb) mutex_exit(&zilog->zl_lock); next_lwb: - if (lwb->lwb_slim) + if (lwb->lwb_flags & LWB_FLAG_SLIM) zilc = (zil_chain_t *)lwb->lwb_buf; else zilc = (zil_chain_t *)(lwb->lwb_buf + lwb->lwb_nmax); - int wsz = lwb->lwb_sz; + uint64_t alloc_size = BP_GET_LSIZE(&lwb->lwb_blk); + int wsz = alloc_size; if (lwb->lwb_error == 0) { abd_t *lwb_abd = abd_get_from_buf(lwb->lwb_buf, lwb->lwb_sz); - if (!lwb->lwb_slog || zilog->zl_cur_size <= zil_slog_bulk) + if (!(lwb->lwb_flags & LWB_FLAG_SLOG) || + zilog->zl_cur_size <= zil_slog_bulk) prio = ZIO_PRIORITY_SYNC_WRITE; else prio = ZIO_PRIORITY_ASYNC_WRITE; @@ -1959,16 +1982,17 @@ next_lwb: ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]); lwb->lwb_write_zio = zio_rewrite(lwb->lwb_root_zio, spa, 0, - &lwb->lwb_blk, lwb_abd, lwb->lwb_sz, zil_lwb_write_done, + &lwb->lwb_blk, lwb_abd, alloc_size, zil_lwb_write_done, lwb, prio, ZIO_FLAG_CANFAIL, &zb); zil_lwb_add_block(lwb, &lwb->lwb_blk); - if (lwb->lwb_slim) { + if (lwb->lwb_flags & LWB_FLAG_SLIM) { /* For Slim ZIL only write what is used. */ wsz = P2ROUNDUP_TYPED(lwb->lwb_nused, ZIL_MIN_BLKSZ, int); - ASSERT3S(wsz, <=, lwb->lwb_sz); - zio_shrink(lwb->lwb_write_zio, wsz); + ASSERT3S(wsz, <=, alloc_size); + if (wsz < alloc_size) + zio_shrink(lwb->lwb_write_zio, wsz); wsz = lwb->lwb_write_zio->io_size; } memset(lwb->lwb_buf + lwb->lwb_nused, 0, wsz - lwb->lwb_nused); @@ -2004,13 +2028,53 @@ next_lwb: BP_ZERO(bp); error = lwb->lwb_error; if (error == 0) { - error = zio_alloc_zil(spa, zilog->zl_os, txg, bp, nlwb->lwb_sz, - &slog); + /* + * Allocation flexibility depends on LWB state: + * if NEW: allow range allocation and larger sizes; + * if OPENED: use fixed predetermined allocation size; + * if CLOSED + Slim: allocate precisely for actual usage. + */ + boolean_t flexible = (nlwb->lwb_state == LWB_STATE_NEW); + if (flexible) { + /* We need to prevent opening till we update lwb_sz. */ + mutex_enter(&nlwb->lwb_lock); + flexible = (nlwb->lwb_state == LWB_STATE_NEW); + if (!flexible) + mutex_exit(&nlwb->lwb_lock); /* We lost. */ + } + boolean_t closed_slim = (nlwb->lwb_state == LWB_STATE_CLOSED && + (lwb->lwb_flags & LWB_FLAG_SLIM)); + + uint64_t min_size, max_size; + if (closed_slim) { + /* This transition is racy, but only one way. */ + membar_consumer(); + min_size = max_size = P2ROUNDUP_TYPED(nlwb->lwb_nused, + ZIL_MIN_BLKSZ, uint64_t); + } else if (flexible) { + min_size = nlwb->lwb_min_sz; + max_size = nlwb->lwb_sz; + } else { + min_size = max_size = nlwb->lwb_sz; + } + + error = zio_alloc_zil(spa, zilog->zl_os, txg, bp, + min_size, max_size, &slog, flexible); + if (error == 0) { + if (closed_slim) + ASSERT3U(BP_GET_LSIZE(bp), ==, max_size); + else if (flexible) + nlwb->lwb_sz = BP_GET_LSIZE(bp); + else + ASSERT3U(BP_GET_LSIZE(bp), ==, nlwb->lwb_sz); + } + if (flexible) + mutex_exit(&nlwb->lwb_lock); } if (error == 0) { ASSERT3U(BP_GET_BIRTH(bp), ==, txg); - BP_SET_CHECKSUM(bp, nlwb->lwb_slim ? ZIO_CHECKSUM_ZILOG2 : - ZIO_CHECKSUM_ZILOG); + BP_SET_CHECKSUM(bp, (nlwb->lwb_flags & LWB_FLAG_SLIM) ? + ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); bp->blk_cksum = lwb->lwb_blk.blk_cksum; bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++; } @@ -2039,14 +2103,15 @@ next_lwb: if (nlwb) { nlwb->lwb_blk = *bp; nlwb->lwb_error = error; - nlwb->lwb_slog = slog; + if (slog) + nlwb->lwb_flags |= LWB_FLAG_SLOG; nlwb->lwb_alloc_txg = txg; if (nlwb->lwb_state != LWB_STATE_READY) nlwb = NULL; } mutex_exit(&zilog->zl_lock); - if (lwb->lwb_slog) { + if (lwb->lwb_flags & LWB_FLAG_SLOG) { ZIL_STAT_BUMP(zilog, zil_itx_metaslab_slog_count); ZIL_STAT_INCR(zilog, zil_itx_metaslab_slog_bytes, lwb->lwb_nused); @@ -2220,7 +2285,6 @@ zil_lwb_assign(zilog_t *zilog, lwb_t *lwb, itx_t *itx, list_t *ilwbs) ASSERT(MUTEX_HELD(&zilog->zl_issuer_lock)); ASSERT3P(lwb, !=, NULL); - ASSERT3P(lwb->lwb_buf, !=, NULL); zil_lwb_write_open(zilog, lwb); @@ -2262,9 +2326,10 @@ cont: (dlen % max_log_data == 0 || lwb_sp < reclen + dlen % max_log_data))) { list_insert_tail(ilwbs, lwb); - lwb = zil_lwb_write_close(zilog, lwb, LWB_STATE_OPENED); + lwb = zil_lwb_write_close(zilog, lwb); if (lwb == NULL) return (NULL); + zil_lwb_write_open(zilog, lwb); lwb_sp = lwb->lwb_nmax - lwb->lwb_nused; } @@ -2554,7 +2619,7 @@ zil_itxg_clean(void *arg) * called) we will hit this case. */ if (itx->itx_lr.lrc_txtype == TX_COMMIT) - zil_commit_waiter_skip(itx->itx_private); + zil_commit_waiter_done(itx->itx_private, 0); zil_itx_destroy(itx, 0); } @@ -2742,6 +2807,7 @@ zil_crash_clean(zilog_t *zilog, uint64_t synced_txg) } /* This LWB is from the past, so we can clean it up now. */ + ASSERT(lwb->lwb_flags & LWB_FLAG_CRASHED); list_remove(&zilog->zl_lwb_crash_list, lwb); if (lwb->lwb_buf != NULL) zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); @@ -2981,7 +3047,7 @@ zil_prune_commit_list(zilog_t *zilog) * never any itx's for it to wait on), so it's * safe to skip this waiter and mark it done. */ - zil_commit_waiter_skip(itx->itx_private); + zil_commit_waiter_done(itx->itx_private, 0); } else { zil_commit_waiter_link_lwb(itx->itx_private, last_lwb); } @@ -3212,15 +3278,21 @@ zil_process_commit_list(zilog_t *zilog, zil_commit_waiter_t *zcw, list_t *ilwbs) * "next" lwb on-disk. When this happens, we must stall * the ZIL write pipeline; see the comment within * zil_commit_writer_stall() for more details. + * + * ESHUTDOWN has to be handled carefully here. If we get it, + * then the pool suspended and zil_crash() was called, so we + * need to stop trying and just get an error back to the + * callers. */ int err = 0; while ((lwb = list_remove_head(ilwbs)) != NULL) { - err = zil_lwb_write_issue(zilog, lwb); - if (err != 0) - break; + if (err == 0) + err = zil_lwb_write_issue(zilog, lwb); } - if (err == 0) + if (err != ESHUTDOWN) err = zil_commit_writer_stall(zilog); + if (err == ESHUTDOWN) + err = SET_ERROR(EIO); /* * Additionally, we have to signal and mark the "nolwb" @@ -3230,7 +3302,7 @@ zil_process_commit_list(zilog_t *zilog, zil_commit_waiter_t *zcw, list_t *ilwbs) */ zil_commit_waiter_t *zcw; while ((zcw = list_remove_head(&nolwb_waiters)) != NULL) - zil_commit_waiter_skip(zcw); + zil_commit_waiter_done(zcw, err); /* * And finally, we have to destroy the itx's that @@ -3238,7 +3310,7 @@ zil_process_commit_list(zilog_t *zilog, zil_commit_waiter_t *zcw, list_t *ilwbs) * the itx's callback if one exists for the itx. */ while ((itx = list_remove_head(&nolwb_itxs)) != NULL) - zil_itx_destroy(itx, 0); + zil_itx_destroy(itx, err); } else { ASSERT(list_is_empty(&nolwb_waiters)); ASSERT3P(lwb, !=, NULL); @@ -3292,17 +3364,17 @@ zil_process_commit_list(zilog_t *zilog, zil_commit_waiter_t *zcw, list_t *ilwbs) (!zilog->zl_parallel || zilog->zl_suspend > 0)) { zil_burst_done(zilog); list_insert_tail(ilwbs, lwb); - lwb = zil_lwb_write_close(zilog, lwb, LWB_STATE_NEW); + lwb = zil_lwb_write_close(zilog, lwb); if (lwb == NULL) { int err = 0; while ((lwb = list_remove_head(ilwbs)) != NULL) { - err = zil_lwb_write_issue(zilog, lwb); - if (err != 0) - break; + if (err == 0) + err = zil_lwb_write_issue( + zilog, lwb); } - if (err == 0) - zil_commit_writer_stall(zilog); + if (err != ESHUTDOWN) + (void) zil_commit_writer_stall(zilog); } } } @@ -3470,7 +3542,7 @@ zil_commit_waiter_timeout(zilog_t *zilog, zil_commit_waiter_t *zcw) * hasn't been issued. */ zil_burst_done(zilog); - lwb_t *nlwb = zil_lwb_write_close(zilog, lwb, LWB_STATE_NEW); + lwb_t *nlwb = zil_lwb_write_close(zilog, lwb); ASSERT3S(lwb->lwb_state, ==, LWB_STATE_CLOSED); @@ -3546,7 +3618,7 @@ zil_commit_waiter(zilog_t *zilog, zil_commit_waiter_t *zcw) * commit itxs. When this occurs, the commit waiters linked * off of these commit itxs will not be committed to an * lwb. Additionally, these commit waiters will not be - * marked done until zil_commit_waiter_skip() is called via + * marked done until zil_commit_waiter_done() is called via * zil_itxg_clean(). * * Thus, it's possible for this commit waiter (i.e. the @@ -3624,7 +3696,7 @@ zil_alloc_commit_waiter(void) list_link_init(&zcw->zcw_node); zcw->zcw_lwb = NULL; zcw->zcw_done = B_FALSE; - zcw->zcw_zio_error = 0; + zcw->zcw_error = 0; return (zcw); } @@ -3728,6 +3800,9 @@ zil_crash(zilog_t *zilog) */ for (lwb_t *lwb = list_head(&zilog->zl_lwb_crash_list); lwb != NULL; lwb = list_next(&zilog->zl_lwb_crash_list, lwb)) { + ASSERT(!(lwb->lwb_flags & LWB_FLAG_CRASHED)); + lwb->lwb_flags |= LWB_FLAG_CRASHED; + itx_t *itx; while ((itx = list_remove_head(&lwb->lwb_itxs)) != NULL) zil_itx_destroy(itx, EIO); @@ -3736,7 +3811,7 @@ zil_crash(zilog_t *zilog) while ((zcw = list_remove_head(&lwb->lwb_waiters)) != NULL) { mutex_enter(&zcw->zcw_lock); zcw->zcw_lwb = NULL; - zcw->zcw_zio_error = EIO; + zcw->zcw_error = EIO; zcw->zcw_done = B_TRUE; cv_broadcast(&zcw->zcw_cv); mutex_exit(&zcw->zcw_lock); @@ -4014,7 +4089,7 @@ zil_commit_impl(zilog_t *zilog, uint64_t foid) zil_commit_waiter(zilog, zcw); int err = 0; - if (zcw->zcw_zio_error != 0) { + if (zcw->zcw_error != 0) { /* * If there was an error writing out the ZIL blocks that * this thread is waiting on, then we fallback to @@ -4149,7 +4224,7 @@ zil_lwb_cons(void *vbuf, void *unused, int kmflag) offsetof(zil_commit_waiter_t, zcw_node)); avl_create(&lwb->lwb_vdev_tree, zil_lwb_vdev_compare, sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node)); - mutex_init(&lwb->lwb_vdev_lock, NULL, MUTEX_DEFAULT, NULL); + mutex_init(&lwb->lwb_lock, NULL, MUTEX_DEFAULT, NULL); return (0); } @@ -4158,7 +4233,7 @@ zil_lwb_dest(void *vbuf, void *unused) { (void) unused; lwb_t *lwb = vbuf; - mutex_destroy(&lwb->lwb_vdev_lock); + mutex_destroy(&lwb->lwb_lock); avl_destroy(&lwb->lwb_vdev_tree); list_destroy(&lwb->lwb_waiters); list_destroy(&lwb->lwb_itxs); @@ -4381,7 +4456,7 @@ zil_close(zilog_t *zilog) if (lwb != NULL) { ASSERT(list_is_empty(&zilog->zl_lwb_list)); ASSERT3S(lwb->lwb_state, ==, LWB_STATE_NEW); - zio_buf_free(lwb->lwb_buf, lwb->lwb_sz); + ASSERT0P(lwb->lwb_buf); zil_free_lwb(zilog, lwb); } mutex_exit(&zilog->zl_lock); @@ -4472,16 +4547,16 @@ zil_suspend(const char *osname, void **cookiep) cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock); mutex_exit(&zilog->zl_lock); - if (cookiep == NULL) + if (zilog->zl_restart_txg > 0) { + /* ZIL crashed while we were waiting. */ + zil_resume(os); + error = SET_ERROR(EBUSY); + } else if (cookiep == NULL) zil_resume(os); else *cookiep = os; - if (zilog->zl_restart_txg > 0) - /* ZIL crashed while we were waiting. */ - return (SET_ERROR(EBUSY)); - - return (0); + return (error); } /* diff --git a/sys/contrib/openzfs/module/zfs/zio.c b/sys/contrib/openzfs/module/zfs/zio.c index 3f0ddb63249d..4cf8912d4269 100644 --- a/sys/contrib/openzfs/module/zfs/zio.c +++ b/sys/contrib/openzfs/module/zfs/zio.c @@ -4434,12 +4434,15 @@ zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) */ int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp, - uint64_t size, boolean_t *slog) + uint64_t min_size, uint64_t max_size, boolean_t *slog, + boolean_t allow_larger) { int error; zio_alloc_list_t io_alloc_list; + uint64_t alloc_size = 0; ASSERT(txg > spa_syncing_txg(spa)); + ASSERT3U(min_size, <=, max_size); metaslab_trace_init(&io_alloc_list); @@ -4448,7 +4451,7 @@ zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp, * Fill in the obvious ones before calling into metaslab_alloc(). */ BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); - BP_SET_PSIZE(new_bp, size); + BP_SET_PSIZE(new_bp, max_size); BP_SET_LEVEL(new_bp, 0); /* @@ -4463,43 +4466,51 @@ zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp, ZIOSTAT_BUMP(ziostat_total_allocations); /* Try log class (dedicated slog devices) first */ - error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1, - txg, NULL, flags, &io_alloc_list, allocator, NULL); + error = metaslab_alloc_range(spa, spa_log_class(spa), min_size, + max_size, new_bp, 1, txg, NULL, flags, &io_alloc_list, allocator, + NULL, &alloc_size); *slog = (error == 0); /* Try special_embedded_log class (reserved on special vdevs) */ if (error != 0) { - error = metaslab_alloc(spa, spa_special_embedded_log_class(spa), - size, new_bp, 1, txg, NULL, flags, &io_alloc_list, - allocator, NULL); + error = metaslab_alloc_range(spa, + spa_special_embedded_log_class(spa), min_size, max_size, + new_bp, 1, txg, NULL, flags, &io_alloc_list, allocator, + NULL, &alloc_size); } /* Try special class (general special vdev allocation) */ if (error != 0) { - error = metaslab_alloc(spa, spa_special_class(spa), size, - new_bp, 1, txg, NULL, flags, &io_alloc_list, allocator, - NULL); + error = metaslab_alloc_range(spa, spa_special_class(spa), + min_size, max_size, new_bp, 1, txg, NULL, flags, + &io_alloc_list, allocator, NULL, &alloc_size); } /* Try embedded_log class (reserved on normal vdevs) */ if (error != 0) { - error = metaslab_alloc(spa, spa_embedded_log_class(spa), size, - new_bp, 1, txg, NULL, flags, &io_alloc_list, allocator, - NULL); + error = metaslab_alloc_range(spa, spa_embedded_log_class(spa), + min_size, max_size, new_bp, 1, txg, NULL, flags, + &io_alloc_list, allocator, NULL, &alloc_size); } /* Finally fall back to normal class */ if (error != 0) { ZIOSTAT_BUMP(ziostat_alloc_class_fallbacks); - error = metaslab_alloc(spa, spa_normal_class(spa), size, - new_bp, 1, txg, NULL, flags, &io_alloc_list, allocator, - NULL); + error = metaslab_alloc_range(spa, spa_normal_class(spa), + min_size, max_size, new_bp, 1, txg, NULL, flags, + &io_alloc_list, allocator, NULL, &alloc_size); } metaslab_trace_fini(&io_alloc_list); if (error == 0) { - BP_SET_LSIZE(new_bp, size); - BP_SET_PSIZE(new_bp, size); + if (!allow_larger) + alloc_size = MIN(alloc_size, max_size); + else if (max_size <= SPA_OLD_MAXBLOCKSIZE) + alloc_size = MIN(alloc_size, SPA_OLD_MAXBLOCKSIZE); + alloc_size = P2ALIGN_TYPED(alloc_size, ZIL_MIN_BLKSZ, uint64_t); + + BP_SET_LSIZE(new_bp, alloc_size); + BP_SET_PSIZE(new_bp, alloc_size); BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); BP_SET_CHECKSUM(new_bp, spa_version(spa) >= SPA_VERSION_SLIM_ZIL @@ -4527,8 +4538,8 @@ zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp, } } else { zfs_dbgmsg("%s: zil block allocation failure: " - "size %llu, error %d", spa_name(spa), (u_longlong_t)size, - error); + "min_size %llu, max_size %llu, error %d", spa_name(spa), + (u_longlong_t)min_size, (u_longlong_t)max_size, error); } return (error); diff --git a/sys/contrib/openzfs/module/zfs/zvol.c b/sys/contrib/openzfs/module/zfs/zvol.c index 29f51e230a37..faced0db7e9e 100644 --- a/sys/contrib/openzfs/module/zfs/zvol.c +++ b/sys/contrib/openzfs/module/zfs/zvol.c @@ -38,25 +38,36 @@ * Copyright 2014 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2016 Actifio, Inc. All rights reserved. * Copyright (c) 2012, 2019 by Delphix. All rights reserved. - * Copyright (c) 2024, Klara, Inc. + * Copyright (c) 2024, 2025, Klara, Inc. */ /* * Note on locking of zvol state structures. * - * These structures are used to maintain internal state used to emulate block - * devices on top of zvols. In particular, management of device minor number - * operations - create, remove, rename, and set_snapdev - involves access to - * these structures. The zvol_state_lock is primarily used to protect the - * zvol_state_list. The zv->zv_state_lock is used to protect the contents - * of the zvol_state_t structures, as well as to make sure that when the - * time comes to remove the structure from the list, it is not in use, and - * therefore, it can be taken off zvol_state_list and freed. + * zvol_state_t represents the connection between a single dataset + * (DMU_OST_ZVOL) and the device "minor" (some OS-specific representation of a + * "disk" or "device" or "volume", eg, a /dev/zdXX node, a GEOM object, etc). * - * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol, - * e.g. for the duration of receive and rollback operations. This lock can be - * held for significant periods of time. Given that it is undesirable to hold - * mutexes for long periods of time, the following lock ordering applies: + * The global zvol_state_lock is used to protect access to zvol_state_list and + * zvol_htable, which are the primary way to obtain a zvol_state_t from a name. + * It should not be used for anything not name-relateds, and you should avoid + * sleeping or waiting while its held. See zvol_find_by_name(), zvol_insert(), + * zvol_remove(). + * + * The zv_state_lock is used to protect the contents of the associated + * zvol_state_t. Most of the zvol_state_t is dedicated to control and + * configuration; almost none of it is needed for data operations (that is, + * read, write, flush) so this lock is rarely taken during general IO. It + * should be released quickly; you should avoid sleeping or waiting while its + * held. + * + * zv_suspend_lock is used to suspend IO/data operations to a zvol. The read + * half should held for the duration of an IO operation. The write half should + * be taken when something to wait for IO to complete and the block further IO, + * eg for the duration of receive and rollback operations. This lock can be + * held for long periods of time. + * + * Thus, the following lock ordering appies. * - take zvol_state_lock if necessary, to protect zvol_state_list * - take zv_suspend_lock if necessary, by the code path in question * - take zv_state_lock to protect zvol_state_t @@ -67,9 +78,8 @@ * these operations are serialized per pool. Consequently, we can be certain * that for a given zvol, there is only one operation at a time in progress. * That is why one can be sure that first, zvol_state_t for a given zvol is - * allocated and placed on zvol_state_list, and then other minor operations - * for this zvol are going to proceed in the order of issue. - * + * allocated and placed on zvol_state_list, and then other minor operations for + * this zvol are going to proceed in the order of issue. */ #include <sys/dataset_kstats.h> @@ -1135,20 +1145,34 @@ zvol_tag(zvol_state_t *zv) /* * Suspend the zvol for recv and rollback. */ -zvol_state_t * -zvol_suspend(const char *name) +int +zvol_suspend(const char *name, zvol_state_t **zvp) { zvol_state_t *zv; zv = zvol_find_by_name(name, RW_WRITER); if (zv == NULL) - return (NULL); + return (SET_ERROR(ENOENT)); /* block all I/O, release in zvol_resume. */ ASSERT(MUTEX_HELD(&zv->zv_state_lock)); ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock)); + /* + * If it's being removed, unlock and return error. It doesn't make any + * sense to try to suspend a zvol being removed, but being here also + * means that zvol_remove_minors_impl() is about to call zvol_remove() + * and then destroy the zvol_state_t, so returning a pointer to it for + * the caller to mess with would be a disaster anyway. + */ + if (zv->zv_flags & ZVOL_REMOVING) { + mutex_exit(&zv->zv_state_lock); + rw_exit(&zv->zv_suspend_lock); + /* NB: Returning EIO here to match zfsvfs_teardown() */ + return (SET_ERROR(EIO)); + } + atomic_inc(&zv->zv_suspend_ref); if (zv->zv_open_count > 0) @@ -1161,7 +1185,8 @@ zvol_suspend(const char *name) mutex_exit(&zv->zv_state_lock); /* zv_suspend_lock is released in zvol_resume() */ - return (zv); + *zvp = zv; + return (0); } int @@ -1570,184 +1595,156 @@ zvol_create_minors_impl(zvol_task_t *task) } /* - * Remove minors for specified dataset including children and snapshots. - */ - -/* - * Remove the minor for a given zvol. This will do it all: - * - flag the zvol for removal, so new requests are rejected - * - wait until outstanding requests are completed - * - remove it from lists - * - free it - * It's also usable as a taskq task, and smells nice too. + * Remove minors for specified dataset and, optionally, its children and + * snapshots. */ static void -zvol_remove_minor_task(void *arg) -{ - zvol_state_t *zv = (zvol_state_t *)arg; - - ASSERT(!RW_LOCK_HELD(&zvol_state_lock)); - ASSERT(!MUTEX_HELD(&zv->zv_state_lock)); - - mutex_enter(&zv->zv_state_lock); - while (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) { - zv->zv_flags |= ZVOL_REMOVING; - cv_wait(&zv->zv_removing_cv, &zv->zv_state_lock); - } - mutex_exit(&zv->zv_state_lock); - - rw_enter(&zvol_state_lock, RW_WRITER); - mutex_enter(&zv->zv_state_lock); - - zvol_remove(zv); - zvol_os_clear_private(zv); - - mutex_exit(&zv->zv_state_lock); - rw_exit(&zvol_state_lock); - - zvol_os_free(zv); -} - -static void -zvol_free_task(void *arg) -{ - zvol_os_free(arg); -} - -static void zvol_remove_minors_impl(zvol_task_t *task) { zvol_state_t *zv, *zv_next; const char *name = task ? task->zt_name1 : NULL; int namelen = ((name) ? strlen(name) : 0); - taskqid_t t; - list_t delay_list, free_list; + boolean_t children = task ? !!task->zt_value : B_TRUE; if (zvol_inhibit_dev) return; - list_create(&delay_list, sizeof (zvol_state_t), - offsetof(zvol_state_t, zv_next)); - list_create(&free_list, sizeof (zvol_state_t), - offsetof(zvol_state_t, zv_next)); + /* + * We collect up zvols that we want to remove on a separate list, so + * that we don't have to hold zvol_state_lock for the whole time. + * + * We can't remove them from the global lists until we're completely + * done with them, because that would make them appear to ZFS-side ops + * that they don't exist, and the name might be reused, which can't be + * good. + */ + list_t remove_list; + list_create(&remove_list, sizeof (zvol_state_t), + offsetof(zvol_state_t, zv_remove_node)); - rw_enter(&zvol_state_lock, RW_WRITER); + rw_enter(&zvol_state_lock, RW_READER); for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { zv_next = list_next(&zvol_state_list, zv); mutex_enter(&zv->zv_state_lock); + if (zv->zv_flags & ZVOL_REMOVING) { + /* Another thread is handling shutdown, skip it. */ + mutex_exit(&zv->zv_state_lock); + continue; + } + + /* + * This zvol should be removed if: + * - no name was offered (ie removing all at shutdown); or + * - name matches exactly; or + * - we were asked to remove children, and + * - the start of the name matches, and + * - there is a '/' immediately after the matched name; or + * - there is a '@' immediately after the matched name + */ if (name == NULL || strcmp(zv->zv_name, name) == 0 || - (strncmp(zv->zv_name, name, namelen) == 0 && + (children && strncmp(zv->zv_name, name, namelen) == 0 && (zv->zv_name[namelen] == '/' || zv->zv_name[namelen] == '@'))) { - /* - * By holding zv_state_lock here, we guarantee that no - * one is currently using this zv - */ /* - * If in use, try to throw everyone off and try again - * later. + * Matched, so mark it removal. We want to take the + * write half of the suspend lock to make sure that + * the zvol is not suspended, and give any data ops + * chance to finish. */ - if (zv->zv_open_count > 0 || - atomic_read(&zv->zv_suspend_ref)) { - zv->zv_flags |= ZVOL_REMOVING; - t = taskq_dispatch( - zv->zv_objset->os_spa->spa_zvol_taskq, - zvol_remove_minor_task, zv, TQ_SLEEP); - if (t == TASKQID_INVALID) { - /* - * Couldn't create the task, so we'll - * do it in place once the loop is - * finished. - */ - list_insert_head(&delay_list, zv); - } + mutex_exit(&zv->zv_state_lock); + rw_enter(&zv->zv_suspend_lock, RW_WRITER); + mutex_enter(&zv->zv_state_lock); + + if (zv->zv_flags & ZVOL_REMOVING) { + /* Another thread has taken it, let them. */ mutex_exit(&zv->zv_state_lock); + rw_exit(&zv->zv_suspend_lock); continue; } - zvol_remove(zv); - /* - * Cleared while holding zvol_state_lock as a writer - * which will prevent zvol_open() from opening it. + * Mark it and unlock. New entries will see the flag + * and return ENXIO. */ - zvol_os_clear_private(zv); - - /* Drop zv_state_lock before zvol_free() */ + zv->zv_flags |= ZVOL_REMOVING; mutex_exit(&zv->zv_state_lock); + rw_exit(&zv->zv_suspend_lock); - /* Try parallel zv_free, if failed do it in place */ - t = taskq_dispatch(system_taskq, zvol_free_task, zv, - TQ_SLEEP); - if (t == TASKQID_INVALID) - list_insert_head(&free_list, zv); - } else { + /* Put it on the list for the next stage. */ + list_insert_head(&remove_list, zv); + } else mutex_exit(&zv->zv_state_lock); - } } - rw_exit(&zvol_state_lock); - - /* Wait for zvols that we couldn't create a remove task for */ - while ((zv = list_remove_head(&delay_list)) != NULL) - zvol_remove_minor_task(zv); - - /* Free any that we couldn't free in parallel earlier */ - while ((zv = list_remove_head(&free_list)) != NULL) - zvol_os_free(zv); -} - -/* Remove minor for this specific volume only */ -static int -zvol_remove_minor_impl(const char *name) -{ - zvol_state_t *zv = NULL, *zv_next; - if (zvol_inhibit_dev) - return (0); + rw_exit(&zvol_state_lock); - rw_enter(&zvol_state_lock, RW_WRITER); + /* Didn't match any, nothing to do! */ + if (list_is_empty(&remove_list)) { + if (task) + task->zt_error = SET_ERROR(ENOENT); + return; + } - for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { - zv_next = list_next(&zvol_state_list, zv); + /* Actually shut them all down. */ + for (zv = list_head(&remove_list); zv != NULL; zv = zv_next) { + zv_next = list_next(&remove_list, zv); mutex_enter(&zv->zv_state_lock); - if (strcmp(zv->zv_name, name) == 0) - /* Found, leave the the loop with zv_lock held */ - break; - mutex_exit(&zv->zv_state_lock); - } - - if (zv == NULL) { - rw_exit(&zvol_state_lock); - return (SET_ERROR(ENOENT)); - } - ASSERT(MUTEX_HELD(&zv->zv_state_lock)); + /* + * Still open or suspended, just wait. This can happen if, for + * example, we managed to acquire zv_state_lock in the moments + * where zvol_open() or zvol_release() are trading locks to + * call zvol_first_open() or zvol_last_close(). + */ + while (zv->zv_open_count > 0 || + atomic_read(&zv->zv_suspend_ref)) + cv_wait(&zv->zv_removing_cv, &zv->zv_state_lock); - if (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) { /* - * In use, so try to throw everyone off, then wait - * until finished. + * No users, shut down the OS side. This may not remove the + * minor from view immediately, depending on the kernel + * specifics, but it will ensure that it is unusable and that + * this zvol_state_t can never again be reached from an OS-side + * operation. */ - zv->zv_flags |= ZVOL_REMOVING; + zvol_os_remove_minor(zv); mutex_exit(&zv->zv_state_lock); + + /* Remove it from the name lookup lists */ + rw_enter(&zvol_state_lock, RW_WRITER); + zvol_remove(zv); rw_exit(&zvol_state_lock); - zvol_remove_minor_task(zv); - return (0); } - zvol_remove(zv); - zvol_os_clear_private(zv); + /* + * Our own references on remove_list is the last one, free them and + * we're done. + */ + while ((zv = list_remove_head(&remove_list)) != NULL) + zvol_os_free(zv); - mutex_exit(&zv->zv_state_lock); - rw_exit(&zvol_state_lock); + list_destroy(&remove_list); +} - zvol_os_free(zv); +/* Remove minor for this specific volume only */ +static int +zvol_remove_minor_impl(const char *name) +{ + if (zvol_inhibit_dev) + return (0); - return (0); + zvol_task_t task; + memset(&task, 0, sizeof (zvol_task_t)); + strlcpy(task.zt_name1, name, sizeof (task.zt_name1)); + task.zt_value = B_FALSE; + + zvol_remove_minors_impl(&task); + + return (task.zt_error); } /* @@ -2067,6 +2064,7 @@ zvol_remove_minors(spa_t *spa, const char *name, boolean_t async) task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP); task->zt_op = ZVOL_ASYNC_REMOVE_MINORS; strlcpy(task->zt_name1, name, sizeof (task->zt_name1)); + task->zt_value = B_TRUE; id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); if ((async == B_FALSE) && (id != TASKQID_INVALID)) taskq_wait_id(spa->spa_zvol_taskq, id); @@ -2188,14 +2186,6 @@ zvol_fini_impl(void) zvol_remove_minors_impl(NULL); - /* - * The call to "zvol_remove_minors_impl" may dispatch entries to - * the system_taskq, but it doesn't wait for those entries to - * complete before it returns. Thus, we must wait for all of the - * removals to finish, before we can continue. - */ - taskq_wait_outstanding(system_taskq, 0); - kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head)); list_destroy(&zvol_state_list); rw_destroy(&zvol_state_lock); diff --git a/sys/contrib/openzfs/module/zstd/zfs_zstd.c b/sys/contrib/openzfs/module/zstd/zfs_zstd.c index 391216d6e263..3db196953f74 100644 --- a/sys/contrib/openzfs/module/zstd/zfs_zstd.c +++ b/sys/contrib/openzfs/module/zstd/zfs_zstd.c @@ -876,9 +876,9 @@ static void __init zstd_mempool_init(void) { zstd_mempool_cctx = - kmem_zalloc(ZSTD_POOL_MAX * sizeof (struct zstd_pool), KM_SLEEP); + vmem_zalloc(ZSTD_POOL_MAX * sizeof (struct zstd_pool), KM_SLEEP); zstd_mempool_dctx = - kmem_zalloc(ZSTD_POOL_MAX * sizeof (struct zstd_pool), KM_SLEEP); + vmem_zalloc(ZSTD_POOL_MAX * sizeof (struct zstd_pool), KM_SLEEP); for (int i = 0; i < ZSTD_POOL_MAX; i++) { mutex_init(&zstd_mempool_cctx[i].barrier, NULL, @@ -924,8 +924,8 @@ zstd_mempool_deinit(void) release_pool(&zstd_mempool_dctx[i]); } - kmem_free(zstd_mempool_dctx, ZSTD_POOL_MAX * sizeof (struct zstd_pool)); - kmem_free(zstd_mempool_cctx, ZSTD_POOL_MAX * sizeof (struct zstd_pool)); + vmem_free(zstd_mempool_dctx, ZSTD_POOL_MAX * sizeof (struct zstd_pool)); + vmem_free(zstd_mempool_cctx, ZSTD_POOL_MAX * sizeof (struct zstd_pool)); zstd_mempool_dctx = NULL; zstd_mempool_cctx = NULL; } |