forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2023-12-11 d2ccde1c8e90d38cee87a1b0309ad2827f3fd30d 
 kernel/drivers/scsi/scsi_error.c
....@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
34  *
....@@ -115,6 +116,14 @@
115116 	return 1;
116117 }
117118 
119
+static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd)
120
+{
121
+	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
122
+		return true;
123
+
124
+	return ++cmd->retries <= cmd->allowed;
125
+}
126
+
118127 /**
119128  * scmd_eh_abort_handler - Handle command aborts
120129  * @work:	command to be aborted.
....@@ -150,7 +159,7 @@
150159 						    "eh timeout, not retrying "
151160 						    "aborted command\n"));
152161 			} else if (!scsi_noretry_cmd(scmd) &&
153
-			    (++scmd->retries <= scmd->allowed)) {
162
+				   scsi_cmd_retry_allowed(scmd)) {
154163 				SCSI_LOG_ERROR_RECOVERY(3,
155164 					scmd_printk(KERN_WARNING, scmd,
156165 						    "retry aborted command\n"));
....@@ -297,19 +306,19 @@
297306 
298307 	if (rtn == BLK_EH_DONE) {
299308 		/*
300
-		 * For blk-mq, we must set the request state to complete now
301
-		 * before sending the request to the scsi error handler. This
302
-		 * will prevent a use-after-free in the event the LLD manages
303
-		 * to complete the request before the error handler finishes
304
-		 * processing this timed out request.
309
+		 * Set the command to complete first in order to prevent a real
310
+		 * completion from releasing the command while error handling
311
+		 * is using it. If the command was already completed, then the
312
+		 * lower level driver beat the timeout handler, and it is safe
313
+		 * to return without escalating error recovery.
305314 		 *
306
-		 * If the request was already completed, then the LLD beat the
307
-		 * time out handler from transferring the request to the scsi
308
-		 * error handler. In that case we can return immediately as no
309
-		 * further action is required.
315
+		 * If timeout handling lost the race to a real completion, the
316
+		 * block layer may ignore that due to a fake timeout injection,
317
+		 * so return RESET_TIMER to allow error handling another shot
318
+		 * at this command.
310319 		 */
311
-		if (req->q->mq_ops && !blk_mq_mark_complete(req))
312
-			return rtn;
320
+		if (test_and_set_bit(SCMD_STATE_COMPLETE, &scmd->state))
321
+			return BLK_EH_RESET_TIMER;
313322 		if (scsi_abort_command(scmd) != SUCCESS) {
314323 			set_host_byte(scmd, DID_TIME_OUT);
315324 			scsi_eh_scmd_add(scmd);
....@@ -337,9 +346,6 @@
337346 	wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
338347 
339348 	online = scsi_device_online(sdev);
340
-
341
-	SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
342
-		"%s: rtn: %d\n", __func__, online));
343349 
344350 	return online;
345351 }
....@@ -601,7 +607,7 @@
601607 			set_host_byte(scmd, DID_ALLOC_FAILURE);
602608 			return SUCCESS;
603609 		}
604
-		/* FALLTHROUGH */
610
+		fallthrough;
605611 	case COPY_ABORTED:
606612 	case VOLUME_OVERFLOW:
607613 	case MISCOMPARE:
....@@ -623,7 +629,7 @@
623629 			return ADD_TO_MLQUEUE;
624630 		else
625631 			set_host_byte(scmd, DID_TARGET_FAILURE);
626
-		/* FALLTHROUGH */
632
+		fallthrough;
627633 
628634 	case ILLEGAL_REQUEST:
629635 		if (sshdr.asc == 0x20 || /* Invalid command operation code */
....@@ -736,7 +742,7 @@
736742 	switch (status_byte(scmd->result)) {
737743 	case GOOD:
738744 		scsi_handle_queue_ramp_up(scmd->device);
739
-		/* FALLTHROUGH */
745
+		fallthrough;
740746 	case COMMAND_TERMINATED:
741747 		return SUCCESS;
742748 	case CHECK_CONDITION:
....@@ -757,7 +763,7 @@
757763 		return FAILED;
758764 	case QUEUE_FULL:
759765 		scsi_handle_queue_full(scmd->device);
760
-		/* fall through */
766
+		fallthrough;
761767 	case BUSY:
762768 		return NEEDS_RETRY;
763769 	default:
....@@ -968,7 +974,6 @@
968974 	ses->cmnd = scmd->cmnd;
969975 	ses->data_direction = scmd->sc_data_direction;
970976 	ses->sdb = scmd->sdb;
971
-	ses->next_rq = scmd->request->next_rq;
972977 	ses->result = scmd->result;
973978 	ses->resid_len = scmd->req.resid_len;
974979 	ses->underflow = scmd->underflow;
....@@ -980,7 +985,6 @@
980985 	scmd->cmnd = ses->eh_cmnd;
981986 	memset(scmd->cmnd, 0, BLK_MAX_CDB);
982987 	memset(&scmd->sdb, 0, sizeof(scmd->sdb));
983
-	scmd->request->next_rq = NULL;
984988 	scmd->result = 0;
985989 	scmd->req.resid_len = 0;
986990 
....@@ -1034,7 +1038,6 @@
10341038 	scmd->cmnd = ses->cmnd;
10351039 	scmd->sc_data_direction = ses->data_direction;
10361040 	scmd->sdb = ses->sdb;
1037
-	scmd->request->next_rq = ses->next_rq;
10381041 	scmd->result = ses->result;
10391042 	scmd->req.resid_len = ses->resid_len;
10401043 	scmd->underflow = ses->underflow;
....@@ -1063,7 +1066,7 @@
10631066 	struct scsi_device *sdev = scmd->device;
10641067 	struct Scsi_Host *shost = sdev->host;
10651068 	DECLARE_COMPLETION_ONSTACK(done);
1066
-	unsigned long timeleft = timeout;
1069
+	unsigned long timeleft = timeout, delay;
10671070 	struct scsi_eh_save ses;
10681071 	const unsigned long stall_for = msecs_to_jiffies(100);
10691072 	int rtn;
....@@ -1074,7 +1077,29 @@
10741077 
10751078 	scsi_log_send(scmd);
10761079 	scmd->scsi_done = scsi_eh_done;
1077
-	rtn = shost->hostt->queuecommand(shost, scmd);
1080
+
1081
+	/*
1082
+	 * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can
1083
+	 * change the SCSI device state after we have examined it and before
1084
+	 * .queuecommand() is called.
1085
+	 */
1086
+	mutex_lock(&sdev->state_mutex);
1087
+	while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) {
1088
+		mutex_unlock(&sdev->state_mutex);
1089
+		SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev,
1090
+			"%s: state %d <> %d\n", __func__, sdev->sdev_state,
1091
+			SDEV_BLOCK));
1092
+		delay = min(timeleft, stall_for);
1093
+		timeleft -= delay;
1094
+		msleep(jiffies_to_msecs(delay));
1095
+		mutex_lock(&sdev->state_mutex);
1096
+	}
1097
+	if (sdev->sdev_state != SDEV_BLOCK)
1098
+		rtn = shost->hostt->queuecommand(shost, scmd);
1099
+	else
1100
+		rtn = SCSI_MLQUEUE_DEVICE_BUSY;
1101
+	mutex_unlock(&sdev->state_mutex);
1102
+
10781103 	if (rtn) {
10791104 		if (timeleft > stall_for) {
10801105 			scsi_eh_restore_cmnd(scmd, &ses);
....@@ -1247,11 +1272,18 @@
12471272 		 * upper level.
12481273 		 */
12491274 		if (rtn == SUCCESS)
1250
-			/* we don't want this command reissued, just
1251
-			 * finished with the sense data, so set
1252
-			 * retries to the max allowed to ensure it
1253
-			 * won't get reissued */
1254
-			scmd->retries = scmd->allowed;
1275
+			/*
1276
+			 * We don't want this command reissued, just finished
1277
+			 * with the sense data, so set retries to the max
1278
+			 * allowed to ensure it won't get reissued. If the user
1279
+			 * has requested infinite retries, we also want to
1280
+			 * finish this command, so force completion by setting
1281
+			 * retries and allowed to the same value.
1282
+			 */
1283
+			if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
1284
+				scmd->retries = scmd->allowed = 1;
1285
+			else
1286
+				scmd->retries = scmd->allowed;
12551287 		else if (rtn != NEEDS_RETRY)
12561288 			continue;
12571289 
....@@ -1285,7 +1317,7 @@
12851317 	case NEEDS_RETRY:
12861318 		if (retry_cnt--)
12871319 			goto retry_tur;
1288
-		/*FALLTHRU*/
1320
+		fallthrough;
12891321 	case SUCCESS:
12901322 		return 0;
12911323 	default:
....@@ -1395,6 +1427,7 @@
13951427 				sdev_printk(KERN_INFO, sdev,
13961428 					    "%s: skip START_UNIT, past eh deadline\n",
13971429 					    current->comm));
1430
+			scsi_device_put(sdev);
13981431 			break;
13991432 		}
14001433 		stu_scmd = NULL;
....@@ -1461,6 +1494,7 @@
14611494 				sdev_printk(KERN_INFO, sdev,
14621495 					    "%s: skip BDR, past eh deadline\n",
14631496 					     current->comm));
1497
+			scsi_device_put(sdev);
14641498 			break;
14651499 		}
14661500 		bdr_scmd = NULL;
....@@ -1720,7 +1754,7 @@
17201754 		if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
17211755 		    status_byte(scmd->result) == RESERVATION_CONFLICT)
17221756 			return 0;
1723
-		/* fall through */
1757
+		fallthrough;
17241758 	case DID_SOFT_ERROR:
17251759 		return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
17261760 	}
....@@ -1736,8 +1770,8 @@
17361770 	if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
17371771 	    blk_rq_is_passthrough(scmd->request))
17381772 		return 1;
1739
-	else
1740
-		return 0;
1773
+
1774
+	return 0;
17411775 }
17421776 
17431777 /**
....@@ -1791,7 +1825,7 @@
17911825 			set_host_byte(scmd, DID_TIME_OUT);
17921826 			return SUCCESS;
17931827 		}
1794
-		/* FALLTHROUGH */
1828
+		fallthrough;
17951829 	case DID_NO_CONNECT:
17961830 	case DID_BAD_TARGET:
17971831 		/*
....@@ -1835,7 +1869,7 @@
18351869 			 * lower down
18361870 			 */
18371871 			break;
1838
-		/* fallthrough */
1872
+		fallthrough;
18391873 	case DID_BUS_BUSY:
18401874 	case DID_PARITY:
18411875 		goto maybe_retry;
....@@ -1873,7 +1907,7 @@
18731907 		 * the case of trying to send too many commands to a
18741908 		 * tagged queueing device.
18751909 		 */
1876
-		/* FALLTHROUGH */
1910
+		fallthrough;
18771911 	case BUSY:
18781912 		/*
18791913 		 * device can't talk to us at the moment.  Should only
....@@ -1886,7 +1920,7 @@
18861920 		if (scmd->cmnd[0] == REPORT_LUNS)
18871921 			scmd->device->sdev_target->expecting_lun_change = 0;
18881922 		scsi_handle_queue_ramp_up(scmd->device);
1889
-		/* FALLTHROUGH */
1923
+		fallthrough;
18901924 	case COMMAND_TERMINATED:
18911925 		return SUCCESS;
18921926 	case TASK_ABORTED:
....@@ -1925,8 +1959,7 @@
19251959 	 * the request was not marked fast fail.  Note that above,
19261960 	 * even if the request is marked fast fail, we still requeue
19271961 	 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1928
-	if ((++scmd->retries) <= scmd->allowed
1929
-	    && !scsi_noretry_cmd(scmd)) {
1962
+	if (scsi_cmd_retry_allowed(scmd) && !scsi_noretry_cmd(scmd)) {
19301963 		return NEEDS_RETRY;
19311964 	} else {
19321965 		/*
....@@ -1938,7 +1971,7 @@
19381971 
19391972 static void eh_lock_door_done(struct request *req, blk_status_t status)
19401973 {
1941
-	__blk_put_request(req->q, req);
1974
+	blk_put_request(req);
19421975 }
19431976 
19441977 /**
....@@ -2072,8 +2105,7 @@
20722105 	list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
20732106 		list_del_init(&scmd->eh_entry);
20742107 		if (scsi_device_online(scmd->device) &&
2075
-		    !scsi_noretry_cmd(scmd) &&
2076
-		    (++scmd->retries <= scmd->allowed)) {
2108
+		    !scsi_noretry_cmd(scmd) && scsi_cmd_retry_allowed(scmd)) {
20772109 			SCSI_LOG_ERROR_RECOVERY(3,
20782110 				scmd_printk(KERN_INFO, scmd,
20792111 					     "%s: flush retry cmd\n",
....@@ -2357,22 +2389,22 @@
23572389 		rtn = scsi_try_bus_device_reset(scmd);
23582390 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
23592391 			break;
2360
-		/* FALLTHROUGH */
2392
+		fallthrough;
23612393 	case SG_SCSI_RESET_TARGET:
23622394 		rtn = scsi_try_target_reset(scmd);
23632395 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
23642396 			break;
2365
-		/* FALLTHROUGH */
2397
+		fallthrough;
23662398 	case SG_SCSI_RESET_BUS:
23672399 		rtn = scsi_try_bus_reset(scmd);
23682400 		if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
23692401 			break;
2370
-		/* FALLTHROUGH */
2402
+		fallthrough;
23712403 	case SG_SCSI_RESET_HOST:
23722404 		rtn = scsi_try_host_reset(scmd);
23732405 		if (rtn == SUCCESS)
23742406 			break;
2375
-		/* FALLTHROUGH */
2407
+		fallthrough;
23762408 	default:
23772409 		rtn = FAILED;
23782410 		break;
....@@ -2395,14 +2427,12 @@
23952427 	wake_up(&shost->host_wait);
23962428 	scsi_run_host_queues(shost);
23972429 
2398
-	scsi_put_command(scmd);
23992430 	kfree(rq);
24002431 
24012432 out_put_autopm_host:
24022433 	scsi_autopm_put_host(shost);
24032434 	return error;
24042435 }
2405
-EXPORT_SYMBOL(scsi_ioctl_reset);
24062436 
24072437 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
24082438 				  struct scsi_sense_hdr *sshdr)