Fix memory allocation issue for BLAKE3 context

The kmem_alloc(sizeof (*ctx), KM_NOSLEEP) call on FreeBSD can't be
used in this code segment. Work around this by pre-allocating a percpu
context array for later use.
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tino Reichardt <milky-zfs@mcmilk.de>
Closes #13568 

include/sys/blake3.h

@@ -92,6 +92,11 @@ void Blake3_Final(const BLAKE3_CTX *ctx, uint8_t *out);
 void Blake3_FinalSeek(const BLAKE3_CTX *ctx, uint64_t seek, uint8_t *out,
     size_t out_len);
 
+/* these are pre-allocated contexts */
+extern void **blake3_per_cpu_ctx;
+extern void blake3_per_cpu_ctx_init(void);
+extern void blake3_per_cpu_ctx_fini(void);
+
 /* return number of supported implementations */
 extern int blake3_get_impl_count(void);
 

module/icp/algs/blake3/blake3_impl.c

@@ -201,6 +201,34 @@ blake3_impl_get_ops(void)
 	return (blake3_selected_impl);
 }
 
+#if defined(_KERNEL)
+void **blake3_per_cpu_ctx;
+
+void
+blake3_per_cpu_ctx_init(void)
+{
+	/*
+	 * Create "The Godfather" ptr to hold all blake3 ctx
+	 */
+	blake3_per_cpu_ctx = kmem_alloc(max_ncpus * sizeof (void *), KM_SLEEP);
+	for (int i = 0; i < max_ncpus; i++) {
+		blake3_per_cpu_ctx[i] = kmem_alloc(sizeof (BLAKE3_CTX),
+		    KM_SLEEP);
+	}
+}
+
+void
+blake3_per_cpu_ctx_fini(void)
+{
+	for (int i = 0; i < max_ncpus; i++) {
+		memset(blake3_per_cpu_ctx[i], 0, sizeof (BLAKE3_CTX));
+		kmem_free(blake3_per_cpu_ctx[i], sizeof (BLAKE3_CTX));
+	}
+	memset(blake3_per_cpu_ctx, 0, max_ncpus * sizeof (void *));
+	kmem_free(blake3_per_cpu_ctx, max_ncpus * sizeof (void *));
+}
+#endif
+
 #if defined(_KERNEL) && defined(__linux__)
 static int
 icp_blake3_impl_set(const char *name, zfs_kernel_param_t *kp)

module/zfs/blake3_zfs.c

@@ -47,18 +47,22 @@ void
 abd_checksum_blake3_native(abd_t *abd, uint64_t size, const void *ctx_template,
     zio_cksum_t *zcp)
 {
-	BLAKE3_CTX *ctx;
-
-	ctx = kmem_alloc(sizeof (*ctx), KM_NOSLEEP);
-	ASSERT(ctx != 0);
 	ASSERT(ctx_template != 0);
 
+#if defined(_KERNEL)
+	BLAKE3_CTX *ctx = blake3_per_cpu_ctx[CPU_SEQID_UNSTABLE];
+#else
+	BLAKE3_CTX *ctx = kmem_alloc(sizeof (*ctx), KM_SLEEP);
+#endif
+
 	memcpy(ctx, ctx_template, sizeof (*ctx));
 	(void) abd_iterate_func(abd, 0, size, blake3_incremental, ctx);
 	Blake3_Final(ctx, (uint8_t *)zcp);
 
+#if !defined(_KERNEL)
 	memset(ctx, 0, sizeof (*ctx));
 	kmem_free(ctx, sizeof (*ctx));
+#endif
 }
 
 /*

module/zfs/zfs_chksum.c

@@ -277,6 +277,9 @@ chksum_benchmark(void)
 void
 chksum_init(void)
 {
+#ifdef _KERNEL
+	blake3_per_cpu_ctx_init();
+#endif
 
 	/* Benchmark supported implementations */
 	chksum_benchmark();
@@ -313,4 +316,8 @@ chksum_fini(void)
 		chksum_stat_cnt = 0;
 		chksum_stat_data = 0;
 	}
+
+#ifdef _KERNEL
+	blake3_per_cpu_ctx_fini();
+#endif
 }