[abd] remove dependency on linear abd

- vdev_raidz
- zio, zio_checksum
- zfs_fm
- change abd_alloc_for_io() to use abd_alloc()

Signed-off-by: Gvozden Neskovic <[email protected]>

cmd/zdb/zdb.c

@@ -3362,6 +3362,7 @@ zdb_read_block(char *thing, spa_t *spa)
 	void *lbuf, *buf;
 	char *s, *p, *dup, *vdev, *flagstr;
 	int i, error;
+	boolean_t borrowed = B_FALSE;
 
 	dup = strdup(thing);
 	s = strtok(dup, ":");
@@ -3432,7 +3433,7 @@ zdb_read_block(char *thing, spa_t *spa)
 	psize = size;
 	lsize = size;
 
-	pabd = abd_alloc_linear(SPA_MAXBLOCKSIZE, B_FALSE);
+	pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
 	lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
 
 	BP_ZERO(bp);
@@ -3533,8 +3534,9 @@ zdb_read_block(char *thing, spa_t *spa)
 		buf = lbuf;
 		size = lsize;
 	} else {
-		buf = abd_to_buf(pabd);
 		size = psize;
+		buf = abd_borrow_buf_copy(pabd, size);
+		borrowed = B_TRUE;
 	}
 
 	if (flags & ZDB_FLAG_PRINT_BLKPTR)
@@ -3550,6 +3552,9 @@ zdb_read_block(char *thing, spa_t *spa)
 	else
 		zdb_dump_block(thing, buf, size, flags);
 
+	if (borrowed)
+		abd_return_buf_copy(pabd, buf, size);
+
 out:
 	abd_free(pabd);
 	umem_free(lbuf, SPA_MAXBLOCKSIZE);

include/sys/zio.h

@@ -291,7 +291,7 @@ typedef struct zio_prop {
 typedef struct zio_cksum_report zio_cksum_report_t;
 
 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
-    const void *good_data);
+    const abd_t *good_data);
 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 
 struct zio_bad_cksum;				/* defined in zio_checksum.h */
@@ -573,14 +573,14 @@ extern hrtime_t zio_handle_io_delay(zio_t *zio);
 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
-    const void *good_data, const void *bad_data, boolean_t drop_if_identical);
+    const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical);
 
 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 
 /* If we have the good data in hand, this function can be used */
 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
     struct zio *zio, uint64_t offset, uint64_t length,
-    const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
+    const abd_t *good_data, const abd_t *bad_data, struct zio_bad_cksum *info);
 
 /* Called from spa_sync(), but primarily an injection handler */
 extern void spa_handle_ignored_writes(spa_t *spa);

include/sys/zio_checksum.h

@@ -100,7 +100,7 @@ extern int zio_checksum_equal(spa_t *, blkptr_t *, enum zio_checksum,
     void *, uint64_t, uint64_t, zio_bad_cksum_t *);
 extern void zio_checksum_compute(zio_t *, enum zio_checksum,
     struct abd *, uint64_t);
-extern int zio_checksum_error_impl(spa_t *, blkptr_t *, enum zio_checksum,
+extern int zio_checksum_error_impl(spa_t *, const blkptr_t *, enum zio_checksum,
     struct abd *, uint64_t, uint64_t, zio_bad_cksum_t *);
 extern int zio_checksum_error(zio_t *zio, zio_bad_cksum_t *out);
 extern enum zio_checksum spa_dedup_checksum(spa_t *spa);

module/zfs/abd.c

@@ -457,15 +457,11 @@ abd_alloc_sametype(abd_t *sabd, size_t size)
  * Currently this is linear ABDs, however if ldi_strategy() can ever issue I/Os
  * using a scatter/gather list we should switch to that and replace this call
  * with vanilla abd_alloc().
- *
- * LINUX ABD TODO - once vdev_disk.c has ABD page support change to vanilla
- * - TODO vdev_disk.c now has ABD page support, but some disk label checksumming
- *   code still assumes linear ABD.
  */
 abd_t *
 abd_alloc_for_io(size_t size, boolean_t is_metadata)
 {
-	return (abd_alloc_linear(size, is_metadata));
+	return (abd_alloc(size, is_metadata));
 }
 
 abd_t *

module/zfs/vdev_raidz.c

@@ -117,8 +117,7 @@ vdev_raidz_map_free(raidz_map_t *rm)
 		abd_free(rm->rm_col[c].rc_abd);
 
 		if (rm->rm_col[c].rc_gdata != NULL)
-			zio_buf_free(rm->rm_col[c].rc_gdata,
-			    rm->rm_col[c].rc_size);
+			abd_free(rm->rm_col[c].rc_gdata);
 	}
 
 	size = 0;
@@ -158,14 +157,14 @@ vdev_raidz_cksum_free(void *arg, size_t ignored)
 }
 
 static void
-vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const void *good_data)
+vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const abd_t *good_data)
 {
 	raidz_map_t *rm = zcr->zcr_cbdata;
-	size_t c = zcr->zcr_cbinfo;
-	size_t x;
+	const size_t c = zcr->zcr_cbinfo;
+	size_t x, offset;
 
-	const char *good = NULL;
-	char *bad;
+	const abd_t *good = NULL;
+	const abd_t *bad = rm->rm_col[c].rc_abd;
 
 	if (good_data == NULL) {
 		zfs_ereport_finish_checksum(zcr, NULL, NULL, B_FALSE);
@@ -180,8 +179,6 @@ vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const void *good_data)
 		 */
 		if (rm->rm_col[0].rc_gdata == NULL) {
 			abd_t *bad_parity[VDEV_RAIDZ_MAXPARITY];
-			char *buf;
-			int offset;
 
 			/*
 			 * Set up the rm_col[]s to generate the parity for
@@ -190,20 +187,21 @@ vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const void *good_data)
 			 */
 			for (x = 0; x < rm->rm_firstdatacol; x++) {
 				bad_parity[x] = rm->rm_col[x].rc_abd;
-				rm->rm_col[x].rc_gdata =
-				    zio_buf_alloc(rm->rm_col[x].rc_size);
 				rm->rm_col[x].rc_abd =
-				    abd_get_from_buf(rm->rm_col[x].rc_gdata,
-				    rm->rm_col[x].rc_size);
+				    rm->rm_col[x].rc_gdata =
+				    abd_alloc_for_io(rm->rm_col[x].rc_size,
+				    B_FALSE);
 			}
 
 			/* fill in the data columns from good_data */
-			buf = (char *)good_data;
+			offset = 0;
 			for (; x < rm->rm_cols; x++) {
 				abd_put(rm->rm_col[x].rc_abd);
-				rm->rm_col[x].rc_abd = abd_get_from_buf(buf,
-				    rm->rm_col[x].rc_size);
-				buf += rm->rm_col[x].rc_size;
+
+				rm->rm_col[x].rc_abd =
+				    abd_get_offset_size((abd_t *)good_data,
+				    offset, rm->rm_col[x].rc_size);
+				offset += rm->rm_col[x].rc_size;
 			}
 
 			/*
@@ -212,10 +210,8 @@ vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const void *good_data)
 			vdev_raidz_generate_parity(rm);
 
 			/* restore everything back to its original state */
-			for (x = 0; x < rm->rm_firstdatacol; x++) {
-				abd_put(rm->rm_col[x].rc_abd);
+			for (x = 0; x < rm->rm_firstdatacol; x++)
 				rm->rm_col[x].rc_abd = bad_parity[x];
-			}
 
 			offset = 0;
 			for (x = rm->rm_firstdatacol; x < rm->rm_cols; x++) {
@@ -228,19 +224,19 @@ vdev_raidz_cksum_finish(zio_cksum_report_t *zcr, const void *good_data)
 		}
 
 		ASSERT3P(rm->rm_col[c].rc_gdata, !=, NULL);
-		good = rm->rm_col[c].rc_gdata;
+		good = abd_get_offset(rm->rm_col[c].rc_gdata, 0);
 	} else {
 		/* adjust good_data to point at the start of our column */
-		good = good_data;
-
+		offset = 0;
 		for (x = rm->rm_firstdatacol; x < c; x++)
-			good += rm->rm_col[x].rc_size;
+			offset += rm->rm_col[x].rc_size;
+
+		good = abd_get_offset((abd_t *)good_data, offset);
 	}
 
-	bad = abd_borrow_buf_copy(rm->rm_col[c].rc_abd, rm->rm_col[c].rc_size);
 	/* we drop the ereport if it ends up that the data was good */
 	zfs_ereport_finish_checksum(zcr, good, bad, B_TRUE);
-	abd_return_buf(rm->rm_col[c].rc_abd, bad, rm->rm_col[c].rc_size);
+	abd_put((abd_t *)good);
 }
 
 /*
@@ -283,15 +279,15 @@ vdev_raidz_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *arg)
 	for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++)
 		size += rm->rm_col[c].rc_size;
 
-	rm->rm_abd_copy =
-	    abd_alloc_sametype(rm->rm_col[rm->rm_firstdatacol].rc_abd, size);
+	rm->rm_abd_copy = abd_alloc_for_io(size, B_FALSE);
 
 	for (offset = 0, c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
 		raidz_col_t *col = &rm->rm_col[c];
 		abd_t *tmp = abd_get_offset_size(rm->rm_abd_copy, offset,
 		    col->rc_size);
 
 		abd_copy(tmp, col->rc_abd, col->rc_size);
+
 		abd_put(col->rc_abd);
 		col->rc_abd = tmp;
 
@@ -712,9 +708,8 @@ vdev_raidz_io_start(zio_t *zio)
  * Report a checksum error for a child of a RAID-Z device.
  */
 static void
-raidz_checksum_error(zio_t *zio, raidz_col_t *rc, void *bad_data)
+raidz_checksum_error(zio_t *zio, raidz_col_t *rc, abd_t *bad_data)
 {
-	void *buf;
 	vdev_t *vd = zio->io_vd->vdev_child[rc->rc_devidx];
 
 	if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
@@ -728,11 +723,9 @@ raidz_checksum_error(zio_t *zio, raidz_col_t *rc, void *bad_data)
 		zbc.zbc_has_cksum = 0;
 		zbc.zbc_injected = rm->rm_ecksuminjected;
 
-		buf = abd_borrow_buf_copy(rc->rc_abd, rc->rc_size);
 		zfs_ereport_post_checksum(zio->io_spa, vd, zio,
-		    rc->rc_offset, rc->rc_size, buf, bad_data,
+		    rc->rc_offset, rc->rc_size, rc->rc_abd, bad_data,
 		    &zbc);
-		abd_return_buf(rc->rc_abd, buf, rc->rc_size);
 	}
 }
 
@@ -765,16 +758,17 @@ raidz_checksum_verify(zio_t *zio)
 static int
 raidz_parity_verify(zio_t *zio, raidz_map_t *rm)
 {
-	void *orig[VDEV_RAIDZ_MAXPARITY];
+	abd_t *orig[VDEV_RAIDZ_MAXPARITY];
 	int c, ret = 0;
 	raidz_col_t *rc;
 
 	for (c = 0; c < rm->rm_firstdatacol; c++) {
 		rc = &rm->rm_col[c];
 		if (!rc->rc_tried || rc->rc_error != 0)
 			continue;
-		orig[c] = zio_buf_alloc(rc->rc_size);
-		abd_copy_to_buf(orig[c], rc->rc_abd, rc->rc_size);
+
+		orig[c] = abd_alloc_sametype(rc->rc_abd, rc->rc_size);
+		abd_copy(orig[c], rc->rc_abd, rc->rc_size);
 	}
 
 	vdev_raidz_generate_parity(rm);
@@ -783,12 +777,12 @@ raidz_parity_verify(zio_t *zio, raidz_map_t *rm)
 		rc = &rm->rm_col[c];
 		if (!rc->rc_tried || rc->rc_error != 0)
 			continue;
-		if (bcmp(orig[c], abd_to_buf(rc->rc_abd), rc->rc_size) != 0) {
+		if (abd_cmp(orig[c], rc->rc_abd) != 0) {
 			raidz_checksum_error(zio, rc, orig[c]);
 			rc->rc_error = SET_ERROR(ECKSUM);
 			ret++;
 		}
-		zio_buf_free(orig[c], rc->rc_size);
+		abd_free(orig[c]);
 	}
 
 	return (ret);
@@ -818,7 +812,7 @@ vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
 {
 	raidz_map_t *rm = zio->io_vsd;
 	raidz_col_t *rc;
-	void *orig[VDEV_RAIDZ_MAXPARITY];
+	abd_t *orig[VDEV_RAIDZ_MAXPARITY] = { NULL };
 	int tstore[VDEV_RAIDZ_MAXPARITY + 2];
 	int *tgts = &tstore[1];
 	int curr, next, i, c, n;
@@ -867,7 +861,7 @@ vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
 			ASSERT(orig[i] != NULL);
 		}
 
-		orig[n - 1] = zio_buf_alloc(rm->rm_col[0].rc_size);
+		orig[n - 1] = abd_alloc_for_io(rm->rm_col[0].rc_size, B_FALSE);
 
 		curr = 0;
 		next = tgts[curr];
@@ -886,8 +880,7 @@ vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
 				ASSERT3S(c, >=, 0);
 				ASSERT3S(c, <, rm->rm_cols);
 				rc = &rm->rm_col[c];
-				abd_copy_to_buf(orig[i], rc->rc_abd,
-				    rc->rc_size);
+				abd_copy(orig[i], rc->rc_abd, rc->rc_size);
 			}
 
 			/*
@@ -917,8 +910,7 @@ vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
 			for (i = 0; i < n; i++) {
 				c = tgts[i];
 				rc = &rm->rm_col[c];
-				abd_copy_from_buf(rc->rc_abd, orig[i],
-				    rc->rc_size);
+				abd_copy(rc->rc_abd, orig[i], rc->rc_size);
 			}
 
 			do {
@@ -955,9 +947,8 @@ vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
 	}
 	n--;
 done:
-	for (i = 0; i < n; i++) {
-		zio_buf_free(orig[i], rm->rm_col[0].rc_size);
-	}
+	for (i = 0; i < n; i++)
+		abd_free(orig[i]);
 
 	return (ret);
 }

module/zfs/zfs_fm.c

@@ -561,11 +561,11 @@ zei_range_total_size(zfs_ecksum_info_t *eip)
 
 static zfs_ecksum_info_t *
 annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
-    const uint8_t *goodbuf, const uint8_t *badbuf, size_t size,
+    const abd_t *goodabd, const abd_t *badabd, size_t size,
     boolean_t drop_if_identical)
 {
-	const uint64_t *good = (const uint64_t *)goodbuf;
-	const uint64_t *bad = (const uint64_t *)badbuf;
+	const uint64_t *good;
+	const uint64_t *bad;
 
 	uint64_t allset = 0;
 	uint64_t allcleared = 0;
@@ -609,13 +609,16 @@ annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
 		}
 	}
 
-	if (badbuf == NULL || goodbuf == NULL)
+	if (badabd == NULL || goodabd == NULL)
 		return (eip);
 
 	ASSERT3U(size, ==, nui64s * sizeof (uint64_t));
 	ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
 	ASSERT3U(size, <=, UINT32_MAX);
 
+	good = (const uint64_t *) abd_borrow_buf_copy((abd_t *)goodabd, size);
+	bad = (const uint64_t *) abd_borrow_buf_copy((abd_t *)badabd, size);
+
 	/* build up the range list by comparing the two buffers. */
 	for (idx = 0; idx < nui64s; idx++) {
 		if (good[idx] == bad[idx]) {
@@ -645,6 +648,8 @@ annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
 	 */
 	if (inline_size == 0 && drop_if_identical) {
 		kmem_free(eip, sizeof (*eip));
+		abd_return_buf((abd_t *)goodabd, (void *)good, size);
+		abd_return_buf((abd_t *)badabd, (void *)bad, size);
 		return (NULL);
 	}
 
@@ -685,6 +690,10 @@ annotate_ecksum(nvlist_t *ereport, zio_bad_cksum_t *info,
 		eip->zei_ranges[range].zr_start	*= sizeof (uint64_t);
 		eip->zei_ranges[range].zr_end	*= sizeof (uint64_t);
 	}
+
+	abd_return_buf((abd_t *)goodabd, (void *)good, size);
+	abd_return_buf((abd_t *)badabd, (void *)bad, size);
+
 	eip->zei_allowed_mingap	*= sizeof (uint64_t);
 	inline_size		*= sizeof (uint64_t);
 
@@ -781,8 +790,8 @@ zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
 }
 
 void
-zfs_ereport_finish_checksum(zio_cksum_report_t *report,
-    const void *good_data, const void *bad_data, boolean_t drop_if_identical)
+zfs_ereport_finish_checksum(zio_cksum_report_t *report, const abd_t *good_data,
+    const abd_t *bad_data, boolean_t drop_if_identical)
 {
 #ifdef _KERNEL
 	zfs_ecksum_info_t *info;
@@ -824,7 +833,7 @@ zfs_ereport_free_checksum(zio_cksum_report_t *rpt)
 void
 zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
     struct zio *zio, uint64_t offset, uint64_t length,
-    const void *good_data, const void *bad_data, zio_bad_cksum_t *zbc)
+    const abd_t *good_data, const abd_t *bad_data, zio_bad_cksum_t *zbc)
 {
 #ifdef _KERNEL
 	nvlist_t *ereport = NULL;