s390/dasd: Add new ioctl BIODASDCHECKFMT

Implement new DASD IOCTL BIODASDCHECKFMT to check a range of tracks on a
DASD volume for correct formatting. The following characteristics are
checked:
- Block size
- ECKD key length
- ECKD record ID
- Number of records per track

Signed-off-by: Jan Höppner <hoeppner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/uapi/asm/dasd.h

@@ -187,6 +187,36 @@ typedef struct format_data_t {
 #define DASD_FMT_INT_INVAL  4 /* invalidate tracks */
 #define DASD_FMT_INT_COMPAT 8 /* use OS/390 compatible disk layout */
 
+/*
+ * struct format_check_t
+ * represents all data necessary to evaluate the format of
+ * different tracks of a dasd
+ */
+typedef struct format_check_t {
+	/* Input */
+	struct format_data_t expect;
+
+	/* Output */
+	unsigned int result;		/* Error indication (DASD_FMT_ERR_*) */
+	unsigned int unit;		/* Track that is in error */
+	unsigned int rec;		/* Record that is in error */
+	unsigned int num_records;	/* Records in the track in error */
+	unsigned int blksize;		/* Blocksize of first record in error */
+	unsigned int key_length;	/* Key length of first record in error */
+} format_check_t;
+
+/* Values returned in format_check_t when a format error is detected: */
+/* Too few records were found on a single track */
+#define DASD_FMT_ERR_TOO_FEW_RECORDS	1
+/* Too many records were found on a single track */
+#define DASD_FMT_ERR_TOO_MANY_RECORDS	2
+/* Blocksize/data-length of a record was wrong */
+#define DASD_FMT_ERR_BLKSIZE		3
+/* A record ID is defined by cylinder, head, and record number (CHR). */
+/* On mismatch, this error is set */
+#define DASD_FMT_ERR_RECORD_ID		4
+/* If key-length was != 0 */
+#define DASD_FMT_ERR_KEY_LENGTH		5
 
 /* 
  * struct attrib_data_t
@@ -288,6 +318,8 @@ struct dasd_snid_ioctl_data {
 
 /* Get Sense Path Group ID (SNID) data */
 #define BIODASDSNID    _IOWR(DASD_IOCTL_LETTER, 1, struct dasd_snid_ioctl_data)
+/* Check device format according to format_check_t */
+#define BIODASDCHECKFMT _IOWR(DASD_IOCTL_LETTER, 2, format_check_t)
 
 #define BIODASDSYMMIO  _IOWR(DASD_IOCTL_LETTER, 240, dasd_symmio_parms_t)
 

drivers/s390/block/dasd.c

@@ -1638,6 +1638,9 @@ void dasd_int_handler(struct ccw_device *cdev, unsigned long intparm,
 	struct dasd_ccw_req *cqr, *next;
 	struct dasd_device *device;
 	unsigned long long now;
+	int nrf_suppressed = 0;
+	int fp_suppressed = 0;
+	u8 *sense = NULL;
 	int expires;
 
 	if (IS_ERR(irb)) {
@@ -1673,7 +1676,23 @@ void dasd_int_handler(struct ccw_device *cdev, unsigned long intparm,
 			dasd_put_device(device);
 			return;
 		}
+
+		/*
+		 * In some cases 'File Protected' or 'No Record Found' errors
+		 * might be expected and debug log messages for the
+		 * corresponding interrupts shouldn't be written then.
+		 * Check if either of the according suppress bits is set.
+		 */
+		sense = dasd_get_sense(irb);
+		if (sense) {
+			fp_suppressed = (sense[1] & SNS1_FILE_PROTECTED) &&
+				test_bit(DASD_CQR_SUPPRESS_FP, &cqr->flags);
+			nrf_suppressed = (sense[1] & SNS1_NO_REC_FOUND) &&
+				test_bit(DASD_CQR_SUPPRESS_NRF, &cqr->flags);
+		}
+		if (!(fp_suppressed || nrf_suppressed))
 			device->discipline->dump_sense_dbf(device, irb, "int");
+
 		if (device->features & DASD_FEATURE_ERPLOG)
 			device->discipline->dump_sense(device, cqr, irb);
 		device->discipline->check_for_device_change(device, cqr, irb);
@@ -2312,6 +2331,7 @@ static int _dasd_sleep_on_queue(struct list_head *ccw_queue, int interruptible)
 {
 	struct dasd_device *device;
 	struct dasd_ccw_req *cqr, *n;
+	u8 *sense = NULL;
 	int rc;
 
 retry:
@@ -2357,6 +2377,20 @@ retry:
 
 	rc = 0;
 	list_for_each_entry_safe(cqr, n, ccw_queue, blocklist) {
+		/*
+		 * In some cases the 'File Protected' or 'Incorrect Length'
+		 * error might be expected and error recovery would be
+		 * unnecessary in these cases.	Check if the according suppress
+		 * bit is set.
+		 */
+		sense = dasd_get_sense(&cqr->irb);
+		if (sense && sense[1] & SNS1_FILE_PROTECTED &&
+		    test_bit(DASD_CQR_SUPPRESS_FP, &cqr->flags))
+			continue;
+		if (scsw_cstat(&cqr->irb.scsw) == 0x40 &&
+		    test_bit(DASD_CQR_SUPPRESS_IL, &cqr->flags))
+			continue;
+
 		/*
 		 * for alias devices simplify error recovery and
 		 * return to upper layer

drivers/s390/block/dasd_3990_erp.c

@@ -1367,6 +1367,12 @@ dasd_3990_erp_no_rec(struct dasd_ccw_req * default_erp, char *sense)
 
 	struct dasd_device *device = default_erp->startdev;
 
+	/*
+	 * In some cases the 'No Record Found' error might be expected and
+	 * log messages shouldn't be written then.
+	 * Check if the according suppress bit is set.
+	 */
+	if (!test_bit(DASD_CQR_SUPPRESS_NRF, &default_erp->flags))
 		dev_err(&device->cdev->dev,
 			"The specified record was not found\n");
 
@@ -1393,8 +1399,14 @@ dasd_3990_erp_file_prot(struct dasd_ccw_req * erp)
 
 	struct dasd_device *device = erp->startdev;
 
-	dev_err(&device->cdev->dev, "Accessing the DASD failed because of "
+	/*
-		"a hardware error\n");
+	 * In some cases the 'File Protected' error might be expected and
+	 * log messages shouldn't be written then.
+	 * Check if the according suppress bit is set.
+	 */
+	if (!test_bit(DASD_CQR_SUPPRESS_FP, &erp->flags))
+		dev_err(&device->cdev->dev,
+			"Accessing the DASD failed because of a hardware error\n");
 
 	return dasd_3990_erp_cleanup(erp, DASD_CQR_FAILED);
 

drivers/s390/block/dasd_eckd.c

@@ -121,6 +121,11 @@ struct check_attention_work_data {
 	__u8 lpum;
 };
 
+static int prepare_itcw(struct itcw *, unsigned int, unsigned int, int,
+			struct dasd_device *, struct dasd_device *,
+			unsigned int, int, unsigned int, unsigned int,
+			unsigned int, unsigned int);
+
 /* initial attempt at a probe function. this can be simplified once
  * the other detection code is gone */
 static int
@@ -257,10 +262,13 @@ define_extent(struct ccw1 *ccw, struct DE_eckd_data *data, unsigned int trk,
 	case DASD_ECKD_CCW_READ_CKD_MT:
 	case DASD_ECKD_CCW_READ_KD:
 	case DASD_ECKD_CCW_READ_KD_MT:
-	case DASD_ECKD_CCW_READ_COUNT:
 		data->mask.perm = 0x1;
 		data->attributes.operation = private->attrib.operation;
 		break;
+	case DASD_ECKD_CCW_READ_COUNT:
+		data->mask.perm = 0x1;
+		data->attributes.operation = DASD_BYPASS_CACHE;
+		break;
 	case DASD_ECKD_CCW_WRITE:
 	case DASD_ECKD_CCW_WRITE_MT:
 	case DASD_ECKD_CCW_WRITE_KD:
@@ -529,10 +537,13 @@ static int prefix_LRE(struct ccw1 *ccw, struct PFX_eckd_data *pfxdata,
 	case DASD_ECKD_CCW_READ_CKD_MT:
 	case DASD_ECKD_CCW_READ_KD:
 	case DASD_ECKD_CCW_READ_KD_MT:
-	case DASD_ECKD_CCW_READ_COUNT:
 		dedata->mask.perm = 0x1;
 		dedata->attributes.operation = basepriv->attrib.operation;
 		break;
+	case DASD_ECKD_CCW_READ_COUNT:
+		dedata->mask.perm = 0x1;
+		dedata->attributes.operation = DASD_BYPASS_CACHE;
+		break;
 	case DASD_ECKD_CCW_READ_TRACK:
 	case DASD_ECKD_CCW_READ_TRACK_DATA:
 		dedata->mask.perm = 0x1;
@@ -2096,6 +2107,180 @@ dasd_eckd_fill_geometry(struct dasd_block *block, struct hd_geometry *geo)
 	return 0;
 }
 
+/*
+ * Build the TCW request for the format check
+ */
+static struct dasd_ccw_req *
+dasd_eckd_build_check_tcw(struct dasd_device *base, struct format_data_t *fdata,
+			  int enable_pav, struct eckd_count *fmt_buffer,
+			  int rpt)
+{
+	struct dasd_eckd_private *start_priv;
+	struct dasd_device *startdev = NULL;
+	struct tidaw *last_tidaw = NULL;
+	struct dasd_ccw_req *cqr;
+	struct itcw *itcw;
+	int itcw_size;
+	int count;
+	int rc;
+	int i;
+
+	if (enable_pav)
+		startdev = dasd_alias_get_start_dev(base);
+
+	if (!startdev)
+		startdev = base;
+
+	start_priv = startdev->private;
+
+	count = rpt * (fdata->stop_unit - fdata->start_unit + 1);
+
+	/*
+	 * we're adding 'count' amount of tidaw to the itcw.
+	 * calculate the corresponding itcw_size
+	 */
+	itcw_size = itcw_calc_size(0, count, 0);
+
+	cqr = dasd_smalloc_request(DASD_ECKD_MAGIC, 0, itcw_size, startdev);
+	if (IS_ERR(cqr))
+		return cqr;
+
+	start_priv->count++;
+
+	itcw = itcw_init(cqr->data, itcw_size, ITCW_OP_READ, 0, count, 0);
+	if (IS_ERR(itcw)) {
+		rc = -EINVAL;
+		goto out_err;
+	}
+
+	cqr->cpaddr = itcw_get_tcw(itcw);
+	rc = prepare_itcw(itcw, fdata->start_unit, fdata->stop_unit,
+			  DASD_ECKD_CCW_READ_COUNT_MT, base, startdev, 0, count,
+			  sizeof(struct eckd_count),
+			  count * sizeof(struct eckd_count), 0, rpt);
+	if (rc)
+		goto out_err;
+
+	for (i = 0; i < count; i++) {
+		last_tidaw = itcw_add_tidaw(itcw, 0, fmt_buffer++,
+					    sizeof(struct eckd_count));
+		if (IS_ERR(last_tidaw)) {
+			rc = -EINVAL;
+			goto out_err;
+		}
+	}
+
+	last_tidaw->flags |= TIDAW_FLAGS_LAST;
+	itcw_finalize(itcw);
+
+	cqr->cpmode = 1;
+	cqr->startdev = startdev;
+	cqr->memdev = startdev;
+	cqr->basedev = base;
+	cqr->retries = startdev->default_retries;
+	cqr->expires = startdev->default_expires * HZ;
+	cqr->buildclk = get_tod_clock();
+	cqr->status = DASD_CQR_FILLED;
+	/* Set flags to suppress output for expected errors */
+	set_bit(DASD_CQR_SUPPRESS_FP, &cqr->flags);
+	set_bit(DASD_CQR_SUPPRESS_IL, &cqr->flags);
+
+	return cqr;
+
+out_err:
+	dasd_sfree_request(cqr, startdev);
+
+	return ERR_PTR(rc);
+}
+
+/*
+ * Build the CCW request for the format check
+ */
+static struct dasd_ccw_req *
+dasd_eckd_build_check(struct dasd_device *base, struct format_data_t *fdata,
+		      int enable_pav, struct eckd_count *fmt_buffer, int rpt)
+{
+	struct dasd_eckd_private *start_priv;
+	struct dasd_eckd_private *base_priv;
+	struct dasd_device *startdev = NULL;
+	struct dasd_ccw_req *cqr;
+	struct ccw1 *ccw;
+	void *data;
+	int cplength, datasize;
+	int use_prefix;
+	int count;
+	int i;
+
+	if (enable_pav)
+		startdev = dasd_alias_get_start_dev(base);
+
+	if (!startdev)
+		startdev = base;
+
+	start_priv = startdev->private;
+	base_priv = base->private;
+
+	count = rpt * (fdata->stop_unit - fdata->start_unit + 1);
+
+	use_prefix = base_priv->features.feature[8] & 0x01;
+
+	if (use_prefix) {
+		cplength = 1;
+		datasize = sizeof(struct PFX_eckd_data);
+	} else {
+		cplength = 2;
+		datasize = sizeof(struct DE_eckd_data) +
+			sizeof(struct LO_eckd_data);
+	}
+	cplength += count;
+
+	cqr = dasd_smalloc_request(DASD_ECKD_MAGIC, cplength, datasize,
+				  startdev);
+	if (IS_ERR(cqr))
+		return cqr;
+
+	start_priv->count++;
+	data = cqr->data;
+	ccw = cqr->cpaddr;
+
+	if (use_prefix) {
+		prefix_LRE(ccw++, data, fdata->start_unit, fdata->stop_unit,
+			   DASD_ECKD_CCW_READ_COUNT, base, startdev, 1, 0,
+			   count, 0, 0);
+	} else {
+		define_extent(ccw++, data, fdata->start_unit, fdata->stop_unit,
+			      DASD_ECKD_CCW_READ_COUNT, startdev);
+
+		data += sizeof(struct DE_eckd_data);
+		ccw[-1].flags |= CCW_FLAG_CC;
+
+		locate_record(ccw++, data, fdata->start_unit, 0, count,
+			      DASD_ECKD_CCW_READ_COUNT, base, 0);
+	}
+
+	for (i = 0; i < count; i++) {
+		ccw[-1].flags |= CCW_FLAG_CC;
+		ccw->cmd_code = DASD_ECKD_CCW_READ_COUNT;
+		ccw->flags = CCW_FLAG_SLI;
+		ccw->count = 8;
+		ccw->cda = (__u32)(addr_t) fmt_buffer;
+		ccw++;
+		fmt_buffer++;
+	}
+
+	cqr->startdev = startdev;
+	cqr->memdev = startdev;
+	cqr->basedev = base;
+	cqr->retries = DASD_RETRIES;
+	cqr->expires = startdev->default_expires * HZ;
+	cqr->buildclk = get_tod_clock();
+	cqr->status = DASD_CQR_FILLED;
+	/* Set flags to suppress output for expected errors */
+	set_bit(DASD_CQR_SUPPRESS_NRF, &cqr->flags);
+
+	return cqr;
+}
+
 static struct dasd_ccw_req *
 dasd_eckd_build_format(struct dasd_device *base,
 		       struct format_data_t *fdata,
@@ -2363,9 +2548,24 @@ dasd_eckd_build_format(struct dasd_device *base,
  */
 static struct dasd_ccw_req *
 dasd_eckd_format_build_ccw_req(struct dasd_device *base,
-			       struct format_data_t *fdata, int enable_pav)
+			       struct format_data_t *fdata, int enable_pav,
+			       int tpm, struct eckd_count *fmt_buffer, int rpt)
 {
-	return dasd_eckd_build_format(base, fdata, enable_pav);
+	struct dasd_ccw_req *ccw_req;
+
+	if (!fmt_buffer) {
+		ccw_req = dasd_eckd_build_format(base, fdata, enable_pav);
+	} else {
+		if (tpm)
+			ccw_req = dasd_eckd_build_check_tcw(base, fdata,
+							    enable_pav,
+							    fmt_buffer, rpt);
+		else
+			ccw_req = dasd_eckd_build_check(base, fdata, enable_pav,
+							fmt_buffer, rpt);
+	}
+
+	return ccw_req;
 }
 
 /*
@@ -2410,12 +2610,15 @@ static int dasd_eckd_format_sanity_checks(struct dasd_device *base,
  */
 static int dasd_eckd_format_process_data(struct dasd_device *base,
 					 struct format_data_t *fdata,
-					 int enable_pav)
+					 int enable_pav, int tpm,
+					 struct eckd_count *fmt_buffer, int rpt,
+					 struct irb *irb)
 {
 	struct dasd_eckd_private *private = base->private;
 	struct dasd_ccw_req *cqr, *n;
 	struct list_head format_queue;
 	struct dasd_device *device;
+	char *sense = NULL;
 	int old_start, old_stop, format_step;
 	int step, retry;
 	int rc;
@@ -2429,8 +2632,18 @@ static int dasd_eckd_format_process_data(struct dasd_device *base,
 	old_start = fdata->start_unit;
 	old_stop = fdata->stop_unit;
 
-	format_step = DASD_CQR_MAX_CCW / recs_per_track(&private->rdc_data, 0,
+	if (!tpm && fmt_buffer != NULL) {
-							fdata->blksize);
+		/* Command Mode / Format Check */
+		format_step = 1;
+	} else if (tpm && fmt_buffer != NULL) {
+		/* Transport Mode / Format Check */
+		format_step = DASD_CQR_MAX_CCW / rpt;
+	} else {
+		/* Normal Formatting */
+		format_step = DASD_CQR_MAX_CCW /
+			recs_per_track(&private->rdc_data, 0, fdata->blksize);
+	}
+
 	do {
 		retry = 0;
 		while (fdata->start_unit <= old_stop) {
@@ -2441,7 +2654,8 @@ static int dasd_eckd_format_process_data(struct dasd_device *base,
 			}
 
 			cqr = dasd_eckd_format_build_ccw_req(base, fdata,
-							     enable_pav);
+							     enable_pav, tpm,
+							     fmt_buffer, rpt);
 			if (IS_ERR(cqr)) {
 				rc = PTR_ERR(cqr);
 				if (rc == -ENOMEM) {
@@ -2459,6 +2673,10 @@ static int dasd_eckd_format_process_data(struct dasd_device *base,
 			}
 			list_add_tail(&cqr->blocklist, &format_queue);
 
+			if (fmt_buffer) {
+				step = fdata->stop_unit - fdata->start_unit + 1;
+				fmt_buffer += rpt * step;
+			}
 			fdata->start_unit = fdata->stop_unit + 1;
 			fdata->stop_unit = old_stop;
 		}
@@ -2469,15 +2687,41 @@ out_err:
 		list_for_each_entry_safe(cqr, n, &format_queue, blocklist) {
 			device = cqr->startdev;
 			private = device->private;
-			if (cqr->status == DASD_CQR_FAILED)
+
+			if (cqr->status == DASD_CQR_FAILED) {
+				/*
+				 * Only get sense data if called by format
+				 * check
+				 */
+				if (fmt_buffer && irb) {
+					sense = dasd_get_sense(&cqr->irb);
+					memcpy(irb, &cqr->irb, sizeof(*irb));
+				}
 				rc = -EIO;
+			}
 			list_del_init(&cqr->blocklist);
 			dasd_sfree_request(cqr, device);
 			private->count--;
 		}
 
-		if (rc)
+		if (rc && rc != -EIO)
 			goto out;
+		if (rc == -EIO) {
+			/*
+			 * In case fewer than the expected records are on the
+			 * track, we will most likely get a 'No Record Found'
+			 * error (in command mode) or a 'File Protected' error
+			 * (in transport mode). Those particular cases shouldn't
+			 * pass the -EIO to the IOCTL, therefore reset the rc
+			 * and continue.
+			 */
+			if (sense &&
+			    (sense[1] & SNS1_NO_REC_FOUND ||
+			     sense[1] & SNS1_FILE_PROTECTED))
+				retry = 1;
+			else
+				goto out;
+		}
 
 	} while (retry);
 
@@ -2491,7 +2735,225 @@ out:
 static int dasd_eckd_format_device(struct dasd_device *base,
 				   struct format_data_t *fdata, int enable_pav)
 {
-	return dasd_eckd_format_process_data(base, fdata, enable_pav);
+	return dasd_eckd_format_process_data(base, fdata, enable_pav, 0, NULL,
+					     0, NULL);
+}
+
+/*
+ * Helper function to count consecutive records of a single track.
+ */
+static int dasd_eckd_count_records(struct eckd_count *fmt_buffer, int start,
+				   int max)
+{
+	int head;
+	int i;
+
+	head = fmt_buffer[start].head;
+
+	/*
+	 * There are 3 conditions where we stop counting:
+	 * - if data reoccurs (same head and record may reoccur), which may
+	 *   happen due to the way DASD_ECKD_CCW_READ_COUNT works
+	 * - when the head changes, because we're iterating over several tracks
+	 *   then (DASD_ECKD_CCW_READ_COUNT_MT)
+	 * - when we've reached the end of sensible data in the buffer (the
+	 *   record will be 0 then)
+	 */
+	for (i = start; i < max; i++) {
+		if (i > start) {
+			if ((fmt_buffer[i].head == head &&
+			    fmt_buffer[i].record == 1) ||
+			    fmt_buffer[i].head != head ||
+			    fmt_buffer[i].record == 0)
+				break;
+		}
+	}
+
+	return i - start;
+}
+
+/*
+ * Evaluate a given range of tracks. Data like number of records, blocksize,
+ * record ids, and key length are compared with expected data.
+ *
+ * If a mismatch occurs, the corresponding error bit is set, as well as
+ * additional information, depending on the error.
+ */
+static void dasd_eckd_format_evaluate_tracks(struct eckd_count *fmt_buffer,
+					     struct format_check_t *cdata,
+					     int rpt_max, int rpt_exp,
+					     int trk_per_cyl, int tpm)
+{
+	struct ch_t geo;
+	int max_entries;
+	int count = 0;
+	int trkcount;
+	int blksize;
+	int pos = 0;
+	int i, j;
+	int kl;
+
+	trkcount = cdata->expect.stop_unit - cdata->expect.start_unit + 1;
+	max_entries = trkcount * rpt_max;
+
+	for (i = cdata->expect.start_unit; i <= cdata->expect.stop_unit; i++) {
+		/* Calculate the correct next starting position in the buffer */
+		if (tpm) {
+			while (fmt_buffer[pos].record == 0 &&
+			       fmt_buffer[pos].dl == 0) {
+				if (pos++ > max_entries)
+					break;
+			}
+		} else {
+			if (i != cdata->expect.start_unit)
+				pos += rpt_max - count;
+		}
+
+		/* Calculate the expected geo values for the current track */
+		set_ch_t(&geo, i / trk_per_cyl, i % trk_per_cyl);
+
+		/* Count and check number of records */
+		count = dasd_eckd_count_records(fmt_buffer, pos, pos + rpt_max);
+
+		if (count < rpt_exp) {
+			cdata->result = DASD_FMT_ERR_TOO_FEW_RECORDS;
+			break;
+		}
+		if (count > rpt_exp) {
+			cdata->result = DASD_FMT_ERR_TOO_MANY_RECORDS;
+			break;
+		}
+
+		for (j = 0; j < count; j++, pos++) {
+			blksize = cdata->expect.blksize;
+			kl = 0;
+
+			/*
+			 * Set special values when checking CDL formatted
+			 * devices.
+			 */
+			if ((cdata->expect.intensity & 0x08) &&
+			    geo.cyl == 0 && geo.head == 0) {
+				if (j < 3) {
+					blksize = sizes_trk0[j] - 4;
+					kl = 4;
+				}
+			}
+			if ((cdata->expect.intensity & 0x08) &&
+			    geo.cyl == 0 && geo.head == 1) {
+				blksize = LABEL_SIZE - 44;
+				kl = 44;
+			}
+
+			/* Check blocksize */
+			if (fmt_buffer[pos].dl != blksize) {
+				cdata->result = DASD_FMT_ERR_BLKSIZE;
+				goto out;
+			}
+			/* Check if key length is 0 */
+			if (fmt_buffer[pos].kl != kl) {
+				cdata->result = DASD_FMT_ERR_KEY_LENGTH;
+				goto out;
+			}
+			/* Check if record_id is correct */
+			if (fmt_buffer[pos].cyl != geo.cyl ||
+			    fmt_buffer[pos].head != geo.head ||
+			    fmt_buffer[pos].record != (j + 1)) {
+				cdata->result = DASD_FMT_ERR_RECORD_ID;
+				goto out;
+			}
+		}
+	}
+
+out:
+	/*
+	 * In case of no errors, we need to decrease by one
+	 * to get the correct positions.
+	 */
+	if (!cdata->result) {
+		i--;
+		pos--;
+	}
+
+	cdata->unit = i;
+	cdata->num_records = count;
+	cdata->rec = fmt_buffer[pos].record;
+	cdata->blksize = fmt_buffer[pos].dl;
+	cdata->key_length = fmt_buffer[pos].kl;
+}
+
+/*
+ * Check the format of a range of tracks of a DASD.
+ */
+static int dasd_eckd_check_device_format(struct dasd_device *base,
+					 struct format_check_t *cdata,
+					 int enable_pav)
+{
+	struct dasd_eckd_private *private = base->private;
+	struct eckd_count *fmt_buffer;
+	struct irb irb;
+	int rpt_max, rpt_exp;
+	int fmt_buffer_size;
+	int trk_per_cyl;
+	int trkcount;
+	int tpm = 0;
+	int rc;
+
+	trk_per_cyl = private->rdc_data.trk_per_cyl;
+
+	/* Get maximum and expected amount of records per track */
+	rpt_max = recs_per_track(&private->rdc_data, 0, 512) + 1;
+	rpt_exp = recs_per_track(&private->rdc_data, 0, cdata->expect.blksize);
+
+	trkcount = cdata->expect.stop_unit - cdata->expect.start_unit + 1;
+	fmt_buffer_size = trkcount * rpt_max * sizeof(struct eckd_count);
+
+	fmt_buffer = kzalloc(fmt_buffer_size, GFP_KERNEL | GFP_DMA);
+	if (!fmt_buffer)
+		return -ENOMEM;
+
+	/*
+	 * A certain FICON feature subset is needed to operate in transport
+	 * mode. Additionally, the support for transport mode is implicitly
+	 * checked by comparing the buffer size with fcx_max_data. As long as
+	 * the buffer size is smaller we can operate in transport mode and
+	 * process multiple tracks. If not, only one track at once is being
+	 * processed using command mode.
+	 */
+	if ((private->features.feature[40] & 0x04) &&
+	    fmt_buffer_size <= private->fcx_max_data)
+		tpm = 1;
+
+	rc = dasd_eckd_format_process_data(base, &cdata->expect, enable_pav,
+					   tpm, fmt_buffer, rpt_max, &irb);
+	if (rc && rc != -EIO)
+		goto out;
+	if (rc == -EIO) {
+		/*
+		 * If our first attempt with transport mode enabled comes back
+		 * with an incorrect length error, we're going to retry the
+		 * check with command mode.
+		 */
+		if (tpm && scsw_cstat(&irb.scsw) == 0x40) {
+			tpm = 0;
+			rc = dasd_eckd_format_process_data(base, &cdata->expect,
+							   enable_pav, tpm,
+							   fmt_buffer, rpt_max,
+							   &irb);
+			if (rc)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
+
+	dasd_eckd_format_evaluate_tracks(fmt_buffer, cdata, rpt_max, rpt_exp,
+					 trk_per_cyl, tpm);
+
+out:
+	kfree(fmt_buffer);
+
+	return rc;
 }
 
 static void dasd_eckd_handle_terminated_request(struct dasd_ccw_req *cqr)
@@ -3038,6 +3500,16 @@ static int prepare_itcw(struct itcw *itcw,
 		lredata->auxiliary.check_bytes = 0x2;
 		pfx_cmd = DASD_ECKD_CCW_PFX;
 		break;
+	case DASD_ECKD_CCW_READ_COUNT_MT:
+		dedata->mask.perm = 0x1;
+		dedata->attributes.operation = DASD_BYPASS_CACHE;
+		dedata->ga_extended |= 0x42;
+		dedata->blk_size = blksize;
+		lredata->operation.orientation = 0x2;
+		lredata->operation.operation = 0x16;
+		lredata->auxiliary.check_bytes = 0x01;
+		pfx_cmd = DASD_ECKD_CCW_PFX_READ;
+		break;
 	default:
 		DBF_DEV_EVENT(DBF_ERR, basedev,
 			      "prepare itcw, unknown opcode 0x%x", cmd);
@@ -3085,13 +3557,19 @@ static int prepare_itcw(struct itcw *itcw,
 		}
 	}
 
+	if (cmd == DASD_ECKD_CCW_READ_COUNT_MT) {
+		lredata->auxiliary.length_valid = 0;
+		lredata->auxiliary.length_scope = 0;
+		lredata->sector = 0xff;
+	} else {
 		lredata->auxiliary.length_valid = 1;
 		lredata->auxiliary.length_scope = 1;
+		lredata->sector = sector;
+	}
 	lredata->auxiliary.imbedded_ccw_valid = 1;
 	lredata->length = tlf;
 	lredata->imbedded_ccw = cmd;
 	lredata->count = count;
-	lredata->sector = sector;
 	set_ch_t(&lredata->seek_addr, begcyl, beghead);
 	lredata->search_arg.cyl = lredata->seek_addr.cyl;
 	lredata->search_arg.head = lredata->seek_addr.head;
@@ -4413,11 +4891,35 @@ static void dasd_eckd_dump_sense_tcw(struct dasd_device *device,
 static void dasd_eckd_dump_sense(struct dasd_device *device,
 				 struct dasd_ccw_req *req, struct irb *irb)
 {
-	if (scsw_is_tm(&irb->scsw))
+	u8 *sense = dasd_get_sense(irb);
+
+	if (scsw_is_tm(&irb->scsw)) {
+		/*
+		 * In some cases the 'File Protected' or 'Incorrect Length'
+		 * error might be expected and log messages shouldn't be written
+		 * then. Check if the according suppress bit is set.
+		 */
+		if (sense && (sense[1] & SNS1_FILE_PROTECTED) &&
+		    test_bit(DASD_CQR_SUPPRESS_FP, &req->flags))
+			return;
+		if (scsw_cstat(&irb->scsw) == 0x40 &&
+		    test_bit(DASD_CQR_SUPPRESS_IL, &req->flags))
+			return;
+
 		dasd_eckd_dump_sense_tcw(device, req, irb);
-	else
+	} else {
+		/*
+		 * In some cases the 'No Record Found' error might be expected
+		 * and log messages shouldn't be written then. Check if the
+		 * according suppress bit is set.
+		 */
+		if (sense && sense[1] & SNS1_NO_REC_FOUND &&
+		    test_bit(DASD_CQR_SUPPRESS_NRF, &req->flags))
+			return;
+
 		dasd_eckd_dump_sense_ccw(device, req, irb);
 	}
+}
 
 static int dasd_eckd_pm_freeze(struct dasd_device *device)
 {
@@ -5246,6 +5748,7 @@ static struct dasd_discipline dasd_eckd_discipline = {
 	.term_IO = dasd_term_IO,
 	.handle_terminated_request = dasd_eckd_handle_terminated_request,
 	.format_device = dasd_eckd_format_device,
+	.check_device_format = dasd_eckd_check_device_format,
 	.erp_action = dasd_eckd_erp_action,
 	.erp_postaction = dasd_eckd_erp_postaction,
 	.check_for_device_change = dasd_eckd_check_for_device_change,

drivers/s390/block/dasd_eckd.h

@@ -35,6 +35,7 @@
 #define DASD_ECKD_CCW_READ_MT		 0x86
 #define DASD_ECKD_CCW_WRITE_KD_MT	 0x8d
 #define DASD_ECKD_CCW_READ_KD_MT	 0x8e
+#define DASD_ECKD_CCW_READ_COUNT_MT	 0x92
 #define DASD_ECKD_CCW_RELEASE		 0x94
 #define DASD_ECKD_CCW_WRITE_FULL_TRACK	 0x95
 #define DASD_ECKD_CCW_READ_CKD_MT	 0x9e

drivers/s390/block/dasd_int.h

@@ -236,6 +236,13 @@ struct dasd_ccw_req {
 					 * stolen. Should not be combined with
 					 * DASD_CQR_FLAGS_USE_ERP
 					 */
+/*
+ * The following flags are used to suppress output of certain errors.
+ * These flags should only be used for format checks!
+ */
+#define DASD_CQR_SUPPRESS_NRF	4	/* Suppress 'No Record Found' error */
+#define DASD_CQR_SUPPRESS_FP	5	/* Suppress 'File Protected' error*/
+#define DASD_CQR_SUPPRESS_IL	6	/* Suppress 'Incorrect Length' error */
 
 /* Signature for error recovery functions. */
 typedef struct dasd_ccw_req *(*dasd_erp_fn_t) (struct dasd_ccw_req *);
@@ -318,7 +325,8 @@ struct dasd_discipline {
 	 * Device operation functions. build_cp creates a ccw chain for
 	 * a block device request, start_io starts the request and
 	 * term_IO cancels it (e.g. in case of a timeout). format_device
-	 * returns a ccw chain to be used to format the device.
+	 * formats the device and check_device_format compares the format of
+	 * a device with the expected format_data.
 	 * handle_terminated_request allows to examine a cqr and prepare
 	 * it for retry.
 	 */
@@ -329,7 +337,9 @@ struct dasd_discipline {
 	int (*term_IO) (struct dasd_ccw_req *);
 	void (*handle_terminated_request) (struct dasd_ccw_req *);
 	int (*format_device) (struct dasd_device *,
-			      struct format_data_t *, int enable_pav);
+			      struct format_data_t *, int);
+	int (*check_device_format)(struct dasd_device *,
+				   struct format_check_t *, int);
 	int (*free_cp) (struct dasd_ccw_req *, struct request *);
 
 	/*

drivers/s390/block/dasd_ioctl.c

@@ -238,6 +238,23 @@ dasd_format(struct dasd_block *block, struct format_data_t *fdata)
 	return rc;
 }
 
+static int dasd_check_format(struct dasd_block *block,
+			     struct format_check_t *cdata)
+{
+	struct dasd_device *base;
+	int rc;
+
+	base = block->base;
+	if (!base->discipline->check_device_format)
+		return -ENOTTY;
+
+	rc = base->discipline->check_device_format(base, cdata, 1);
+	if (rc == -EAGAIN)
+		rc = base->discipline->check_device_format(base, cdata, 0);
+
+	return rc;
+}
+
 /*
  * Format device.
  */
@@ -272,6 +289,47 @@ dasd_ioctl_format(struct block_device *bdev, void __user *argp)
 	}
 	rc = dasd_format(base->block, &fdata);
 	dasd_put_device(base);
+
+	return rc;
+}
+
+/*
+ * Check device format
+ */
+static int dasd_ioctl_check_format(struct block_device *bdev, void __user *argp)
+{
+	struct format_check_t cdata;
+	struct dasd_device *base;
+	int rc = 0;
+
+	if (!argp)
+		return -EINVAL;
+
+	base = dasd_device_from_gendisk(bdev->bd_disk);
+	if (!base)
+		return -ENODEV;
+	if (bdev != bdev->bd_contains) {
+		pr_warn("%s: The specified DASD is a partition and cannot be checked\n",
+			dev_name(&base->cdev->dev));
+		rc = -EINVAL;
+		goto out_err;
+	}
+
+	if (copy_from_user(&cdata, argp, sizeof(cdata))) {
+		rc = -EFAULT;
+		goto out_err;
+	}
+
+	rc = dasd_check_format(base->block, &cdata);
+	if (rc)
+		goto out_err;
+
+	if (copy_to_user(argp, &cdata, sizeof(cdata)))
+		rc = -EFAULT;
+
+out_err:
+	dasd_put_device(base);
+
 	return rc;
 }
 
@@ -519,6 +577,9 @@ int dasd_ioctl(struct block_device *bdev, fmode_t mode,
 	case BIODASDFMT:
 		rc = dasd_ioctl_format(bdev, argp);
 		break;
+	case BIODASDCHECKFMT:
+		rc = dasd_ioctl_check_format(bdev, argp);
+		break;
 	case BIODASDINFO:
 		rc = dasd_ioctl_information(block, cmd, argp);
 		break;