Query SMART attributes for all disks in spinup while restricting power operations to array members

cmdline/device.c

@@ -662,7 +662,7 @@ static void state_smart(struct snapraid_state* state, size_t n, tommy_list* low)
 		if (len > serial_pad)
 			serial_pad = len;
 
-		if (devinfo->parent && devinfo->parent->is_array)
+		if (devinfo->is_array)
 			have_array = 1;
 	}
 
@@ -756,10 +756,8 @@ static void state_smart(struct snapraid_state* state, size_t n, tommy_list* low)
 				/* this happens only if no data */
 				printf("    -");
 			} else {
-				int is_array = devinfo->parent && devinfo->parent->is_array;
-
 				/* compute failure rate only for array disks (or fallback to all disks if none are array members) */
-				if (is_array || !have_array)
+				if (devinfo->is_array || !have_array)
 					array_failure_rate += afr;
 
 				printf("%4.0f%%", poisson_prob_at_least_one_failure(afr) * 100);
@@ -1032,6 +1030,7 @@ int devtest(tommy_list* high, tommy_list* low, int operation)
 
 		/* fake device number */
 		entry->device = tommy_strhash_u32(0, devinfo->name);
+		entry->is_array = devinfo->is_array;
 
 		tommy_list_insert_tail(low, &entry->node, entry);
 
@@ -1162,36 +1161,25 @@ int state_device(struct snapraid_state* state, int operation, tommy_list* filter
 		}
 	}
 
-	/* extra disks are never spundown or spunup */
-	int include_extra = 1;
-	switch (operation) {
-	case DEVICE_UP :
-	case DEVICE_DOWN :
-	case DEVICE_DOWNIFUP :
-		include_extra = 0;
-		break;
-	}
-	if (include_extra) {
-		/* for all extra disks */
-		for (i = state->extralist; i != 0; i = i->next) {
-			struct snapraid_extra* extra = i->data;
-			devinfo_t* entry;
+	/* for all extra disks */
+	for (i = state->extralist; i != 0; i = i->next) {
+		struct snapraid_extra* extra = i->data;
+		devinfo_t* entry;
 
-			if (filterlist_disk != 0 && filter_path(filterlist_disk, 0, extra->name, 0) != 0)
-				continue;
+		if (filterlist_disk != 0 && filter_path(filterlist_disk, 0, extra->name, 0) != 0)
+			continue;
 
-			entry = calloc_nofail(1, sizeof(devinfo_t));
+		entry = calloc_nofail(1, sizeof(devinfo_t));
 
-			entry->device = extra->device;
-			device_name_set(entry, extra->name, 0);
-			entry->is_array = 0;
-			pathcpy(entry->mount, sizeof(entry->mount), extra->dir);
-			pathcpy(entry->smartctl, sizeof(entry->smartctl), extra->smartctl);
-			pathcpy(entry->smartctl_info, sizeof(entry->smartctl_info), extra->smartctl_info);
-			memcpy(entry->smartignore, extra->smartignore, sizeof(entry->smartignore));
+		entry->device = extra->device;
+		device_name_set(entry, extra->name, 0);
+		entry->is_array = 0;
+		pathcpy(entry->mount, sizeof(entry->mount), extra->dir);
+		pathcpy(entry->smartctl, sizeof(entry->smartctl), extra->smartctl);
+		pathcpy(entry->smartctl_info, sizeof(entry->smartctl_info), extra->smartctl_info);
+		memcpy(entry->smartignore, extra->smartignore, sizeof(entry->smartignore));
 
-			tommy_list_insert_tail(&high, &entry->node, entry);
-		}
+		tommy_list_insert_tail(&high, &entry->node, entry);
 	}
 
 	/* with a GUI always gives time reference */
@@ -1199,6 +1187,7 @@ int state_device(struct snapraid_state* state, int operation, tommy_list* filter
 		log_tag("unixtime:%" PRIi64 "\n", (int64_t)now);
 
 	int degraded = 0;
+	int query_failed = 0;
 
 	if (state->opt.fake_device) {
 		ret = devtest(&high, &low, operation);
@@ -1211,6 +1200,10 @@ int state_device(struct snapraid_state* state, int operation, tommy_list* filter
 		if (ret > 0) {
 			degraded = 1;
 			ret = 0;
+		} else if (ret < 0 && !tommy_list_empty(&low)) {
+			/* keep operating on devices resolved before or after a failed disk */
+			query_failed = 1;
+			ret = 0;
 		}
 	}
 
@@ -1259,6 +1252,8 @@ bail:
 
 	if (ret == 0 && degraded)
 		ret = 1;
+	if (ret == 0 && query_failed)
+		ret = -1;
 
 	return ret;
 }

cmdline/mingwapp.c

@@ -1049,6 +1049,7 @@ static int devtree(devinfo_t* parent, tommy_list* list)
 		pathprint(devinfo->file, sizeof(devinfo->file), "/dev/pd%" PRIu64, devinfo->device);
 		pathprint(devinfo->wfile, sizeof(devinfo->wfile), "\\\\.\\PhysicalDrive%" PRIu64, devinfo->device);
 		devinfo->parent = parent;
+		devinfo->is_array = parent->is_array;
 
 		/* insert in the list */
 		tommy_list_insert_tail(list, &devinfo->node, devinfo);
@@ -1959,27 +1960,34 @@ static int devup(uint64_t device, const char* name, const char* wfile)
 static void* thread_spinup(void* arg)
 {
 	devinfo_t* devinfo = arg;
-	uint64_t start;
+	int ret;
 
-	/* skip not rotational devices */
-	if (devpower(devinfo->device, devinfo->name, devinfo->wfile) == 0)
-		return 0;
+	/* only for members of the array and rotational devices */
+	if (devinfo->is_array && devpower(devinfo->device, devinfo->name, devinfo->wfile) != 0) {
+		uint64_t start = os_tick_ms();
 
-	start = os_tick_ms();
+		if (devup(devinfo->device, devinfo->name, devinfo->wfile) != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 
-	if (devup(devinfo->device, devinfo->name, devinfo->wfile) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
-	}
+		msg_status("Spunup device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
 
-	msg_status("Spunup device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
-
-	/* after the spin up, get SMART info */
-	if (devsmart(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, devinfo->smart, devinfo->info, devinfo->serial, devinfo->family, devinfo->model, devinfo->interf) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
+		ret = devsmart(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, devinfo->smart, devinfo->info, devinfo->serial, devinfo->family, devinfo->model, devinfo->interf);
+		if (ret != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
+	} else {
+		/* just probe others */
+		ret = devprobe(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, &devinfo->power, devinfo->smart, devinfo->info, devinfo->serial, devinfo->family, devinfo->model, devinfo->interf);
+		if (ret != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 	}
 
 	/*
@@ -1999,22 +2007,20 @@ static void* thread_spinup(void* arg)
 static void* thread_spindown(void* arg)
 {
 	devinfo_t* devinfo = arg;
-	uint64_t start;
 
-	/* skip not rotational devices */
-	if (devpower(devinfo->device, devinfo->name, devinfo->wfile) == 0)
-		return 0;
+	/* only for members of the array and rotational devices */
+	if (devinfo->is_array && devpower(devinfo->device, devinfo->name, devinfo->wfile) != 0) {
+		uint64_t start = os_tick_ms();
 
-	start = os_tick_ms();
+		if (devdown(devinfo->device, devinfo->name, devinfo->smartctl) != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 
-	if (devdown(devinfo->device, devinfo->name, devinfo->smartctl) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
+		msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
 	}
 
-	msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
-
 	return 0;
 }
 
@@ -2024,24 +2030,22 @@ static void* thread_spindown(void* arg)
 static void* thread_spindownifup(void* arg)
 {
 	devinfo_t* devinfo = arg;
-	uint64_t start;
-	int power;
 
-	/* skip not rotational devices */
-	if (devpower(devinfo->device, devinfo->name, devinfo->wfile) == 0)
-		return 0;
+	/* only for members of the array and rotational devices */
+	if (devinfo->is_array && devpower(devinfo->device, devinfo->name, devinfo->wfile) != 0) {
+		uint64_t start = os_tick_ms();
+		int power;
 
-	start = os_tick_ms();
+		if (devdownifup(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, &power) != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 
-	if (devdownifup(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, &power) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
+		if (power == POWER_ACTIVE)
+			msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
 	}
 
-	if (power == POWER_ACTIVE)
-		msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
-
 	return 0;
 }
 
@@ -2133,6 +2137,7 @@ void devsync(tommy_list* high)
 int devquery(tommy_list* high, tommy_list* low)
 {
 	tommy_node* i;
+	int failed = 0;
 
 	/* for each device */
 	for (i = tommy_list_head(high); i != 0; i = i->next) {
@@ -2142,7 +2147,8 @@ int devquery(tommy_list* high, tommy_list* low)
 		if (devresolve(devinfo->mount, devinfo->file, sizeof(devinfo->file), devinfo->wfile, sizeof(devinfo->wfile)) != 0) {
 			/* LCOV_EXCL_START */
 			log_error(EEXTERNAL, "Failed to resolve path '%s'.\n", devinfo->mount);
-			return -1;
+			failed = 1;
+			continue;
 			/* LCOV_EXCL_STOP */
 		}
 
@@ -2159,12 +2165,13 @@ int devquery(tommy_list* high, tommy_list* low)
 		if (devtree(devinfo, low) != 0) {
 			/* LCOV_EXCL_START */
 			log_error(EEXTERNAL, "Failed to expand device '%s'.\n", devinfo->file);
-			return -1;
+			failed = 1;
+			continue;
 			/* LCOV_EXCL_STOP */
 		}
 	}
 
-	return 0;
+	return failed ? -1 : 0;
 }
 
 int devrun(tommy_list* low, int operation)

cmdline/unixapp.c

@@ -3181,6 +3181,7 @@ static int devtree(devinfo_t* parent, dev_t device, tommy_list* list, int* degra
 	char path[PATH_MAX];
 	DIR* d;
 	int slaves = 0;
+	int failed = 0;
 
 	/* check if this device or parent partition is a degraded MD array */
 	pathprint(path, sizeof(path), "/sys/dev/block/%u:%u/md/degraded", major(device), minor(device));
@@ -3201,6 +3202,7 @@ static int devtree(devinfo_t* parent, dev_t device, tommy_list* list, int* degra
 		while ((dd = readdir(d)) != 0) {
 			if (dd->d_name[0] != '.') {
 				dev_t subdev;
+				++slaves;
 
 				/* for each slave, expand the full potential tree */
 				pathprint(path, sizeof(path), "/sys/dev/block/%u:%u/slaves/%s/dev", major(device), minor(device), dd->d_name);
@@ -3208,19 +3210,17 @@ static int devtree(devinfo_t* parent, dev_t device, tommy_list* list, int* degra
 				subdev = devread(path);
 				if (!subdev) {
 					/* LCOV_EXCL_START */
-					closedir(d);
-					return -1;
+					failed = 1;
+					continue;
 					/* LCOV_EXCL_STOP */
 				}
 
 				if (devtree(parent, subdev, list, degraded) != 0) {
 					/* LCOV_EXCL_START */
-					closedir(d);
-					return -1;
+					failed = 1;
+					continue;
 					/* LCOV_EXCL_STOP */
 				}
-
-				++slaves;
 			}
 		}
 
@@ -3263,12 +3263,14 @@ static int devtree(devinfo_t* parent, dev_t device, tommy_list* list, int* degra
 		memcpy(devinfo->smartignore, parent->smartignore, sizeof(devinfo->smartignore));
 		pathcpy(devinfo->file, sizeof(devinfo->file), path);
 		devinfo->parent = parent;
+		devinfo->is_array = parent->is_array;
 
 		/* insert in the list */
 		tommy_list_insert_tail(list, &devinfo->node, devinfo);
 	}
 
-	return 0;
+	/* retain resolved siblings even when another branch failed */
+	return failed ? -1 : 0;
 }
 #endif
 
@@ -3896,27 +3898,34 @@ static void* thread_spinup(void* arg)
 {
 #if HAVE_LINUX_DEVICE
 	devinfo_t* devinfo = arg;
-	uint64_t start;
+	int ret;
 
-	/* skip not rotational devices */
-	if (devpower(devinfo->device) == 0)
-		return 0;
+	/* only for members of the array and rotational devices */
+	if (devinfo->is_array && devpower(devinfo->device) != 0) {
+		uint64_t start = os_tick_ms();
 
-	start = os_tick_ms();
+		if (devup(devinfo->device, devinfo->name) != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 
-	if (devup(devinfo->device, devinfo->name) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
-	}
+		msg_status("Spunup device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
 
-	msg_status("Spunup device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
-
-	/* after the spin up, get SMART info */
-	if (devsmart(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, devinfo->smart, devinfo->info, devinfo->serial, devinfo->family, devinfo->model, devinfo->interf) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
+		ret = devsmart(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, devinfo->smart, devinfo->info, devinfo->serial, devinfo->family, devinfo->model, devinfo->interf);
+		if (ret != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
+	} else {
+		/* just probe others */
+		ret = devprobe(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, &devinfo->power, devinfo->smart, devinfo->info, devinfo->serial, devinfo->family, devinfo->model, devinfo->interf);
+		if (ret != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 	}
 
 	/*
@@ -3941,22 +3950,20 @@ static void* thread_spindown(void* arg)
 {
 #if HAVE_LINUX_DEVICE
 	devinfo_t* devinfo = arg;
-	uint64_t start;
 
-	/* skip not rotational devices */
-	if (devpower(devinfo->device) == 0)
-		return 0;
+	/* only for members of the array and rotational devices */
+	if (devinfo->is_array && devpower(devinfo->device) != 0) {
+		uint64_t start = os_tick_ms();
 
-	start = os_tick_ms();
+		if (devdown(devinfo->device, devinfo->name, devinfo->smartctl) != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 
-	if (devdown(devinfo->device, devinfo->name, devinfo->smartctl) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
+		msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
 	}
 
-	msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
-
 	return 0;
 #else
 	(void)arg;
@@ -3971,24 +3978,22 @@ static void* thread_spindownifup(void* arg)
 {
 #if HAVE_LINUX_DEVICE
 	devinfo_t* devinfo = arg;
-	uint64_t start;
-	int power;
 
-	/* skip not rotational devices */
-	if (devpower(devinfo->device) == 0)
-		return 0;
+	/* only for members of the array and rotational devices */
+	if (devinfo->is_array && devpower(devinfo->device) != 0) {
+		uint64_t start = os_tick_ms();
+		int power;
 
-	start = os_tick_ms();
+		if (devdownifup(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, &power) != 0) {
+			/* LCOV_EXCL_START */
+			return (void*)-1;
+			/* LCOV_EXCL_STOP */
+		}
 
-	if (devdownifup(devinfo->device, devinfo->name, devinfo->smartctl, devinfo->smartctl_info, &power) != 0) {
-		/* LCOV_EXCL_START */
-		return (void*)-1;
-		/* LCOV_EXCL_STOP */
+		if (power == POWER_ACTIVE)
+			msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
 	}
 
-	if (power == POWER_ACTIVE)
-		msg_status("Spundown device '%s' for disk '%s' in %" PRIu64 " ms.\n", devinfo->file, devinfo->name, os_tick_ms() - start);
-
 	return 0;
 #else
 	(void)arg;
@@ -4103,6 +4108,10 @@ void devsync(tommy_list* high)
 		devinfo_t* devinfo = i->data;
 		int f;
 
+		/* only for members of the array */
+		if (!devinfo->is_array)
+			continue;
+
 		f = open(devinfo->mount, O_RDONLY);
 		if (f >= 0) {
 			syncfs(f);
@@ -4117,6 +4126,7 @@ void devsync(tommy_list* high)
 int devquery(tommy_list* high, tommy_list* low)
 {
 	int degraded = 0;
+	int failed = 0;
 #if HAVE_LINUX_DEVICE
 	tommy_node* i;
 	struct stat st;
@@ -4142,7 +4152,8 @@ int devquery(tommy_list* high, tommy_list* low)
 		if (ret < 0) {
 			/* LCOV_EXCL_START */
 			log_error(EEXTERNAL, "Failed to dereference device '%u:%u' at '%s'.\n", major(device), minor(device), devinfo->mount);
-			return -1;
+			failed = 1;
+			continue;
 			/* LCOV_EXCL_STOP */
 		}
 
@@ -4167,7 +4178,8 @@ int devquery(tommy_list* high, tommy_list* low)
 			if (devresolve(dev->device, file, sizeof(file)) != 0) {
 				/* LCOV_EXCL_START */
 				log_error(EEXTERNAL, "Failed to resolve device '%u:%u'.\n", major(dev->device), minor(dev->device));
-				return -1;
+				failed = 1;
+				continue;
 				/* LCOV_EXCL_STOP */
 			}
 
@@ -4180,7 +4192,8 @@ int devquery(tommy_list* high, tommy_list* low)
 			if (devtree(devinfo, dev->device, low, &disk_degraded) != 0) {
 				/* LCOV_EXCL_START */
 				log_error(EEXTERNAL, "Failed to expand device '%u:%u'.\n", major(dev->device), minor(dev->device));
-				return -1;
+				failed = 1;
+				continue;
 				/* LCOV_EXCL_STOP */
 			}
 		}
@@ -4198,7 +4211,7 @@ int devquery(tommy_list* high, tommy_list* low)
 	(void)low;
 #endif
 
-	return degraded ? 1 : 0;
+	return failed ? -1 : degraded ? 1 : 0;
 }
 
 int devrun(tommy_list* low, int operation)