// SPDX-License-Identifier: GPL-3.0-or-later
// Copyright (C) 2011 Andrea Mazzoleni

#include "os/portable.h"

#include "elem.h"
#include "import.h"
#include "search.h"
#include "state.h"
#include "support.h"
#include "parity.h"
#include "stream.h"
#include "handle.h"
#include "io.h"
#include "thermal.h"
#include "raid/raid.h"
#include "raid/cpu.h"

/**
 * Configure the multithread support.
 *
 * Multi thread for write could be either faster or slower, depending
 * on the specific conditions. With multithreads it's likely faster
 * writing to disk, but you'll need to access multiple times the same data,
 * being potentially slower.
 *
 * Multi thread for verify is instead always generally faster,
 * so we enable it if possible.
 */
#if HAVE_THREAD
/* #define HAVE_MT_WRITE 1 */
#define HAVE_MT_VERIFY 1
#endif

/*
 * Start content-file writeback in bounded windows to avoid dirty-page
 * throttling; the subsequent fsync() remains responsible for durability.
 *
 * Best test-rewrite serialization/write times with four 4 GB content files:
 *   ADVISE_SEQUENTIAL:       128s
 *   ADVISE_FLUSH:             86s
 *   ADVISE_DISCARD_WINDOW:    73s
 *   ADVISE_FLUSH_WINDOW:      61s
 */
#define CONTENT_WRITE_ADVISE ADVISE_FLUSH_WINDOW

const char* lev_name(unsigned l)
{
	switch (l) {
	case 0 : return "Parity";
	case 1 : return "2-Parity";
	case 2 : return "3-Parity";
	case 3 : return "4-Parity";
	case 4 : return "5-Parity";
	case 5 : return "6-Parity";
	}

	return "invalid";
}

const char* lev_config_name(unsigned l)
{
	switch (l) {
	case 0 : return "parity";
	case 1 : return "2-parity";
	case 2 : return "3-parity";
	case 3 : return "4-parity";
	case 4 : return "5-parity";
	case 5 : return "6-parity";
	}

	return "invalid";
}

static int lev_config_scan(const char* s, unsigned* level, unsigned* mode)
{
	if (strcmp(s, "parity") == 0 || strcmp(s, "1-parity") == 0) {
		*level = 0;
		return 0;
	}

	if (strcmp(s, "q-parity") == 0 || strcmp(s, "2-parity") == 0) {
		*level = 1;
		return 0;
	}

	if (strcmp(s, "r-parity") == 0 || strcmp(s, "3-parity") == 0) {
		*level = 2;
		return 0;
	}

	if (strcmp(s, "4-parity") == 0) {
		*level = 3;
		return 0;
	}

	if (strcmp(s, "5-parity") == 0) {
		*level = 4;
		return 0;
	}

	if (strcmp(s, "6-parity") == 0) {
		*level = 5;
		return 0;
	}

	if (strcmp(s, "z-parity") == 0) {
		*level = 2;
		if (mode)
			*mode = RAID_MODE_VANDERMONDE_RAID;
		return 0;
	}

	return -1;
}

const char* disk_name_invalid_reason(const char* name)
{
	unsigned dummy_level;
#ifdef _WIN32
	char lower[DISK_NAME_MAX + 1];
#endif

	if (strlen(name) > DISK_NAME_MAX)
		return "Name is too long (maximum 127 bytes)";
	if (strchr(name, '/') != 0)
		return "Name contains '/'";
	if (strchr(name, '\\') != 0)
		return "Name contains '\\'";
	if (strchr(name, ':') != 0)
		return "Name contains ':'";

#ifdef _WIN32
	pathcpy(lower, sizeof(lower), name);
	strlwr(lower);
	if (lev_config_scan(lower, &dummy_level, 0) == 0)
		return "Name is a reserved parity name";
#else
	if (lev_config_scan(name, &dummy_level, 0) == 0)
		return "Name is a reserved parity name";
#endif

	return 0;
}

const char* lev_raid_name(unsigned mode, unsigned n)
{
	switch (n) {
	case 1 : return "par1";
	case 2 : return "par2";
	case 3 : if (mode == RAID_MODE_CAUCHY_RAID || mode == RAID_MODE_CAUCHY_AES)
			return "par3";
		else
			return "parz";
	case 4 : return "par4";
	case 5 : return "par5";
	case 6 : return "par6";
	}

	return 0;
}

void state_init(struct snapraid_state* state)
{
	unsigned l, s, i;

	memset(&state->opt, 0, sizeof(state->opt));
	state->mapped_device = 0;
	state->snapshot = 0;
	state->filter_hidden = 0;
	state->autosave = 0;
	state->need_write = 0;
	state->written = 0;
	state->checked_read = 0;
	state->block_size = 256 * KIBI; /* default 256 KiB */
	state->raid_mode = RAID_MODE_CAUCHY_RAID;
	state->file_mode = ADVISE_DEFAULT;
	for (l = 0; l < LEV_MAX; ++l) {
		state->parity[l].split_mac = 0;
		for (s = 0; s < SPLIT_MAX; ++s) {
			state->parity[l].split_map[s].path[0] = 0;
			state->parity[l].split_map[s].uuid[0] = 0;
			state->parity[l].split_map[s].content_uuid[0] = 0;
			state->parity[l].split_map[s].fstype[0] = 0;
			state->parity[l].split_map[s].fslabel[0] = 0;
			state->parity[l].split_map[s].size = PARITY_SIZE_INVALID;
			state->parity[l].split_map[s].device = 0;
		}
		state->parity[l].smartctl[0] = 0;
		state->parity[l].smartctl_info[0] = 0;
		for (i = 0; i < SMART_IGNORE_MAX; ++i) {
			state->parity[l].smartignore[i].attr_index = 0;
			state->parity[l].smartignore[i].attr_name[0] = 0;
		}
		state->parity[l].total_blocks = 0;
		state->parity[l].free_blocks = 0;
		state->parity[l].skip_access = 0;
		state->parity[l].tick = 0;
		state->parity[l].cached_blocks = 0;
		state->parity[l].is_excluded_by_filter = 0;
	}
	state->tick_io = 0;
	state->tick_misc = 0;
	state->tick_sched = 0;
	state->tick_raid = 0;
	state->tick_hash = 0;
	state->tick_last = os_tick();
	state->share[0] = 0;
	state->pool[0] = 0;
	state->pool_device = 0;
	state->lockfile[0] = 0;
	state->level = 1; /* default is the lowest protection */
	state->no_conf = 0;
	for (i = 0; i < SMART_IGNORE_MAX; ++i) {
		state->smartignore[i].attr_index = 0;
		state->smartignore[i].attr_name[0] = 0;
	}
	state->rehash_blocks = 0;
	state->bad_blocks = 0;
	state->unsynced_blocks = 0;
	state->unscrubbed_blocks = 0;
	state->thermal_stop_gathering = 0;
	state->thermal_system_temperature = 0;
	state->thermal_ambient_temperature = 0;
	state->thermal_highest_temperature = 0;
	state->thermal_first = 0;
	state->thermal_latest = 0;
	state->thermal_temperature_limit = 0;
	state->thermal_cooldown_time = 0;

	tommy_list_init(&state->disklist);
	tommy_list_init(&state->extralist);
	tommy_list_init(&state->maplist);
	tommy_list_init(&state->contentlist);
	tommy_list_init(&state->filterlist);
	tommy_list_init(&state->importlist);
	tommy_list_init(&state->thermallist);
	tommy_hashdyn_init(&state->importset);
	tommy_hashdyn_init(&state->previmportset);
	tommy_hashdyn_init(&state->searchset);
	tommy_arrayblkof_init(&state->infoarr, sizeof(snapraid_info));
	tommy_list_init(&state->bucketlist);
}

void state_done(struct snapraid_state* state)
{
	tommy_list_foreach(&state->disklist, disk_free);
	tommy_list_foreach(&state->extralist, extra_free);
	tommy_list_foreach(&state->maplist, map_free);
	tommy_list_foreach(&state->contentlist, content_free);
	tommy_list_foreach(&state->filterlist, filter_free);
	tommy_list_foreach(&state->importlist, import_file_free);
	tommy_list_foreach(&state->thermallist, thermal_free);
	tommy_hashdyn_foreach(&state->searchset, search_file_free);
	tommy_hashdyn_done(&state->importset);
	tommy_hashdyn_done(&state->previmportset);
	tommy_hashdyn_done(&state->searchset);
	tommy_arrayblkof_done(&state->infoarr);
	tommy_list_foreach(&state->bucketlist, bucket_free);
}

/**
 * Check the configuration.
 */
static void state_config_check(struct snapraid_state* state, const char* path, tommy_list* filterlist_disk)
{
	tommy_node* i;
	unsigned l, s;
	size_t disk_count;

	/* check total memory budget for block buffers (max 256 MiB) */
	disk_count = tommy_list_count(&state->disklist) + state->level;
	if (disk_count > (256 * 1024) / (state->block_size / KIBI)) {
		/* LCOV_EXCL_START */
		log_fatal(EUSER, "The combination of %zu disks and %u KiB 'blocksize' in '%s' exceeds the 256 MiB limit.\n",
			disk_count, state->block_size / KIBI, path);
		log_fatal(EUSER, "Please reduce the 'blocksize' or the number of disks.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* check for parity level */
	if (state->raid_mode == RAID_MODE_VANDERMONDE_RAID) {
		if (state->level > 3) {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Using z-parity limits you to a maximum of 3 parities.\n");
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STIO */
		}
	}

	for (l = 0; l < state->level; ++l) {
		if (state->parity[l].split_mac == 0) {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Missing '%s' specification in '%s'\n", lev_config_name(l), path);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
	}

	if (tommy_list_empty(&state->contentlist)) {
		/* LCOV_EXCL_START */
		log_fatal(EUSER, "Missing 'content' specification in '%s'\n", path);
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* check for equal paths */
	for (i = state->contentlist; i != 0; i = i->next) {
		struct snapraid_content* content = i->data;

		for (l = 0; l < state->level; ++l) {
			for (s = 0; s < state->parity[l].split_mac; ++s) {
				if (pathcmp(state->parity[l].split_map[s].path, content->content) == 0) {
					/* LCOV_EXCL_START */
					log_fatal(EUSER, "Same path used for '%s' and 'content' as '%s'\n", lev_config_name(l), content->content);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}
			}
		}
	}

	/* check device of data disks */
	if (!state->opt.skip_device && !state->opt.skip_disk_access) {
		for (i = state->disklist; i != 0; i = i->next) {
			tommy_node* j;
			struct snapraid_disk* disk = i->data;

			/* skip data disks that are not accessible */
			if (disk->skip_access)
				continue;

#ifdef _WIN32
			if (disk->mount_device == 0) {
				/* LCOV_EXCL_START */
				log_fatal(ESOFT, "Disk '%s' has a zero serial number.\n", disk->dir);
				log_fatal(ESOFT, "This is not necessarily wrong, but for using SnapRAID\n");
				log_fatal(ESOFT, "it's better to change the serial number of the disk.\n");
				log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'.\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
#endif

			for (j = i->next; j != 0; j = j->next) {
				struct snapraid_disk* other = j->data;
				if (disk->mount_device == other->mount_device) {
					if (state->opt.force_device) {
						/*
						 * Note that we just ignore the issue
						 * and we DON'T mark the disk to be skipped
						 * because we want to use these disks
						 */
						if (!state->opt.no_warnings)
							log_error(EUSER, "DANGER! Ignoring that disks '%s' and '%s' are on the same device\n", disk->name, other->name);
					} else {
						/* LCOV_EXCL_START */
						log_fatal(ESOFT, "Disks '%s' and '%s' are on the same device.\n", disk->mount_point, other->mount_point);
#ifdef _WIN32
						log_fatal(ESOFT, "Both have the serial number '%" PRIx64 "'.\n", disk->mount_device);
						log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'\n");
						log_fatal(ESOFT, "to change one of the disk serial.\n");
#endif
						/* in "fix" we allow to continue anyway */
						if (strcmp(state->command, "fix") == 0) {
							log_fatal(ESOFT, "You can '%s' anyway, using 'snapraid --force-device %s'.\n", state->command, state->command);
						}
						exit(EXIT_FAILURE);
						/* LCOV_EXCL_STOP */
					}
				}
			}

			if (!state->opt.skip_parity_access) {
				for (l = 0; l < state->level; ++l) {
					for (s = 0; s < state->parity[l].split_mac; ++s) {
						if (disk->mount_device == state->parity[l].split_map[s].device) {
							if (state->opt.force_device) {
								/*
								 * Note that we just ignore the issue
								 * and we DON'T mark the disk to be skipped
								 * because we want to use these disks
								 */
								if (!state->opt.no_warnings)
									log_error(EUSER, "DANGER! Ignoring that disks '%s' and %s '%s' are on the same device\n", disk->mount_point, lev_name(l), state->parity[l].split_map[s].path);
							} else {
								/* LCOV_EXCL_START */
								log_fatal(ESOFT, "Disk '%s' and %s '%s' are on the same device.\n", disk->mount_point, lev_name(l), state->parity[l].split_map[s].path);
#ifdef _WIN32
								log_fatal(ESOFT, "Both have the serial number '%" PRIx64 "'.\n", disk->mount_device);
								log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'\n");
								log_fatal(ESOFT, "to change one of the disk serial.\n");
#endif
								exit(EXIT_FAILURE);
								/* LCOV_EXCL_STOP */
							}
						}
					}
				}
			}

			if (state->pool[0] != 0 && disk->mount_device == state->pool_device) {
				/* LCOV_EXCL_START */
				log_fatal(ESOFT, "Disk '%s' and pool '%s' are on the same device.\n", disk->mount_point, state->pool);
#ifdef _WIN32
				log_fatal(ESOFT, "Both have the serial number '%" PRIx64 "'.\n", disk->mount_device);
				log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'\n");
				log_fatal(ESOFT, "to change one of the disk serial.\n");
#endif
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}
	}

	/* check device of parity disks */
	if (!state->opt.skip_device && !state->opt.skip_parity_access) {
		for (l = 0; l < state->level; ++l) {
			for (s = 0; s < state->parity[l].split_mac; ++s) {
				unsigned j, t;

				/* skip parity disks that are not accessible */
				if (state->parity[l].skip_access)
					continue;

#ifdef _WIN32
				if (state->parity[l].split_map[s].device == 0) {
					/* LCOV_EXCL_START */
					log_fatal(ESOFT, "Disk '%s' has a zero serial number.\n", state->parity[l].split_map[s].path);
					log_fatal(ESOFT, "This is not necessarily wrong, but for using SnapRAID\n");
					log_fatal(ESOFT, "it's better to change the serial number of the disk.\n");
					log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'.\n");
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}
#endif

				for (j = l + 1; j < state->level; ++j) {
					for (t = 0; t < state->parity[j].split_mac; ++t) {
						if (state->parity[l].split_map[s].device == state->parity[j].split_map[t].device) {
							if (state->opt.force_device) {
								/*
								 * Note that we just ignore the issue
								 * and we DON'T mark the disk to be skipped
								 * because we want to use these disks
								 */
								if (!state->opt.no_warnings)
									log_error(EUSER, "DANGER! Skipping parities '%s' and '%s' on the same device\n", lev_config_name(l), lev_config_name(j));
							} else {
								/* LCOV_EXCL_START */
								log_fatal(ESOFT, "Parity '%s' and '%s' are on the same device.\n", state->parity[l].split_map[s].path, state->parity[j].split_map[t].path);
#ifdef _WIN32
								log_fatal(ESOFT, "Both have the serial number '%" PRIx64 "'.\n", state->parity[l].split_map[s].device);
								log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'\n");
								log_fatal(ESOFT, "to change one of the disk serial.\n");
#endif
								/* in "fix" we allow to continue anyway */
								if (strcmp(state->command, "fix") == 0) {
									log_fatal(ESOFT, "You can '%s' anyway, using 'snapraid --force-device %s'.\n", state->command, state->command);
								}
								exit(EXIT_FAILURE);
								/* LCOV_EXCL_STOP */
							}
						}
					}
				}

				if (state->pool[0] != 0 && state->pool_device == state->parity[l].split_map[s].device) {
					/* LCOV_EXCL_START */
					log_fatal(ESOFT, "Pool '%s' and parity '%s' are on the same device.\n", state->pool, state->parity[l].split_map[s].path);
#ifdef _WIN32
					log_fatal(ESOFT, "Both have the serial number '%" PRIx64 "'.\n", state->pool_device);
					log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'\n");
					log_fatal(ESOFT, "to change one of the disk serial.\n");
#endif
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}
			}
		}
	}

	/* check device of pool disk */
#ifdef _WIN32
	if (!state->opt.skip_device) {
		if (state->pool[0] != 0 && state->pool_device == 0) {
			/* LCOV_EXCL_START */
			log_fatal(ESOFT, "Disk '%s' has a zero serial number.\n", state->pool);
			log_fatal(ESOFT, "This is not necessarily wrong, but for using SnapRAID\n");
			log_fatal(ESOFT, "it's better to change the serial number of the disk.\n");
			log_fatal(ESOFT, "Try using the 'VolumeID' tool by 'Mark Russinovich'.\n");
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
	}
#endif

	/* count the content files */
	if (!state->opt.skip_device && !state->opt.skip_content_access) {
		unsigned content_count;

		content_count = 0;
		for (i = state->contentlist; i != 0; i = i->next) {
			tommy_node* j;
			struct snapraid_content* content = i->data;

			/* check if there are others in the same disk */
			for (j = i->next; j != 0; j = j->next) {
				struct snapraid_content* other = j->data;
				if (content->device == other->device) {
					log_error(ESOFT, "WARNING! Content files on the same disk: '%s' and '%s'.\n", content->content, other->content);
					break;
				}
			}
			if (j != 0) {
				/* skip it */
				continue;
			}

			++content_count;
		}

		if (content_count < state->level + 1) {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "You must have at least %d 'content' files in different disks.\n", state->level + 1);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
	}

	/* check for speed */
#ifdef CONFIG_X86
	if (!raid_cpu_has_ssse3())
#endif
	if ((state->raid_mode == RAID_MODE_CAUCHY_RAID || state->raid_mode == RAID_MODE_CAUCHY_AES) && !state->opt.no_warnings) {
		if (state->level == 3) {
			log_info(EUSER, "WARNING! Your CPU doesn't have a fast implementation for triple parity.\n");
			log_info(EUSER, "WARNING! It's recommended to switch to 'z-parity' instead than '3-parity'.\n");
		} else if (state->level > 3) {
			log_info(EUSER, "WARNING! Your CPU doesn't have a fast implementation beyond triple parity.\n");
			log_info(EUSER, "WARNING! It's recommended to reduce the parity levels to triple parity.\n");
		}
	}

	/* ensure that specified filter disks are valid ones */
	for (i = tommy_list_head(filterlist_disk); i != 0; i = i->next) {
		tommy_node* j;
		struct snapraid_filter* filter = i->data;
		for (j = state->disklist; j != 0; j = j->next) {
			struct snapraid_disk* disk = j->data;
			if (wnmatch(filter->pattern, disk->name) == 0)
				break;
		}
		if (j == 0) {
			for (j = state->extralist; j != 0; j = j->next) {
				struct snapraid_extra* extra = j->data;
				if (wnmatch(filter->pattern, extra->name) == 0)
					break;
			}
		}
		if (j == 0) {
			/* check matching with parity disks */
			for (l = 0; l < state->level; ++l)
				if (wnmatch(filter->pattern, lev_config_name(l)) == 0)
					break;
			if (l == state->level) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Option -d, --filter-disk %s doesn't match any data or parity or extra disk.\n", filter->pattern);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}
	}
}

/**
 * Parse the smartctl command.
 *
 * It extracts the optional [info: xxx] tag, validates, and returns
 * the cleaned smartctl command and smartctl_info options.
 */
int parse_smartctl(const char* custom, char* smartctl, size_t smartctl_size, char* smartctl_info, size_t smartctl_info_size)
{
	const char* s;
	const char* info_start = 0;
	size_t info_len = 0;
	size_t smartctl_len = 0;
	int arg = 0;

	/* initialize output buffers */
	smartctl[0] = 0;
	smartctl_info[0] = 0;

	/* look for "[info:" */
	s = custom;
	while (*s) {
		if (strncmp(s, "[info:", 6) == 0) {
			if (info_start) {
				/* multiple info tags not allowed */
				return -1;
			}

			info_start = s;
			s += 6;

			/* skip spaces inside bracket */
			while (isspace((unsigned char)*s))
				++s;

			/* find matching ']' */
			const char* val_start = s;
			while (*s && *s != ']')
				++s;
			if (*s != ']') {
				/* missing closing bracket */
				return -1;
			}

			/* trim trailing spaces inside bracket */
			const char* val_end = s;
			while (val_end > val_start && isspace((unsigned char)val_end[-1]))
				--val_end;
			info_len = val_end - val_start;
			if (info_len >= smartctl_info_size) {
				/* info string too long */
				return -1;
			}

			memcpy(smartctl_info, val_start, info_len);
			smartctl_info[info_len] = 0;
			++s; /* skip ']' */
		} else {
			if (smartctl_len + 1 >= smartctl_size) {
				return -1;
			}
			smartctl[smartctl_len++] = *s++;
		}
	}
	smartctl[smartctl_len] = 0;

	/* trim extra spaces from the smartctl options */
	strtrim(smartctl);

	/* validate the cleaned smartctl command */
	s = smartctl;
	while (*s) {
		if (s[0] == '%' && s[1] == 's') {
			if (arg) {
				/* LCOV_EXCL_START */
				return -1;
				/* LCOV_EXCL_STOP */
			}
			arg = 1;
			++s;
		} else if (s[0] == '%') {
			/* LCOV_EXCL_START */
			return -1;
			/* LCOV_EXCL_STOP */
		}
		++s;
	}

	/* validate the smartctl_info command */
	s = smartctl_info;
	while (*s) {
		if (s[0] == '%') {
			/* LCOV_EXCL_START */
			return -1;
			/* LCOV_EXCL_STOP */
		}
		++s;
	}

	return 0;
}

void state_config(struct snapraid_state* state, const char* path, const char* command, struct snapraid_option* opt, tommy_list* filterlist_disk)
{
	STREAM* f;
	unsigned line;
	tommy_node* i;
	unsigned l, s;
	/* copy the options */
	state->opt = *opt;

	/* if unset, sort by directory order */
	if (!state->opt.force_order)
		state->opt.force_order = SORT_DIR;

	/* adjust file mode */
	if (state->opt.file_mode != ADVISE_DEFAULT) {
		state->file_mode = state->opt.file_mode;
	} else {
		/* default mode, if nothing is specified */
		state->file_mode = ADVISE_DISCARD;
	}

	/* store current command */
	state->command = command;

	log_tag("conf:file:%s\n", esc_tag(path));

	f = sopen_read(path, 0);
	if (!f) {
		/* LCOV_EXCL_START */
		if (errno == ENOENT) {
			log_fatal(errno, "No configuration file found at '%s'\n", path);
		} else if (errno == EACCES) {
			log_fatal(errno, "You do not have rights to access the configuration file '%s'\n", path);
		} else {
			log_fatal(errno, "Error opening the configuration file '%s'. %s.\n", path, strerror(errno));
		}
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	line = 1;
	while (1) {
		char tag[PATH_MAX];
		char buffer[PATH_MAX];
		ssize_t ret;
		int c;
		unsigned level;

		/* skip initial spaces */
		sgetspace(f);

		/* read the command */
		ret = sgettok(f, tag, sizeof(tag));
		if (ret < 0) {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Error reading the configuration file '%s' at line %u\n", path, line);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		/* skip spaces after the command */
		sgetspace(f);

		if (strcmp(tag, "blocksize") == 0
		        /* block_size is the old format of the option */
			|| strcmp(tag, "block_size") == 0) {

			ret = sgetu32(f, &state->block_size);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'blocksize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (state->block_size < 1) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Too small 'blocksize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (state->block_size > 16 * KIBI) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Too big 'blocksize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			/* check if it's a power of 2 */
			if ((state->block_size & (state->block_size - 1)) != 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Not power of 2 'blocksize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			state->block_size *= KIBI;
		} else if (strcmp(tag, "hashsize") == 0
			|| strcmp(tag, "hash_size") == 0 /* v11.0 used incorrectly this one, kept now for backward compatibility */
		) {
			uint32_t hash_size;

			ret = sgetu32(f, &hash_size);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'hashsize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (hash_size == 2) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Unsupported 'hashsize 2' specification in '%s' at line %u! The minimum supported hash size is 4.\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (hash_size < 4) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Too small 'hashsize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (hash_size > HASH_MAX) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Too big 'hashsize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			/* check if it's a power of 2 */
			if ((hash_size & (hash_size - 1)) != 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Not power of 2 'hashsize' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			BLOCK_HASH_SIZE = hash_size;
		} else if (lev_config_scan(tag, &level, &state->raid_mode) == 0) {
			char device[PATH_MAX];
			char* split_map[SPLIT_MAX + 1];
			unsigned split_mac;
			uint64_t dev;
			int skip_access;

			if (state->parity[level].split_mac != 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Multiple '%s' specification in '%s' at line %u\n", tag, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid '%s' specification in '%s' at line %u\n", tag, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty '%s' specification in '%s' at line %u\n", tag, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			split_mac = strsplit(split_map, SPLIT_MAX + 1, buffer, ",", " \t\r\n", 0);

			if (split_mac > SPLIT_MAX) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Too many files in '%s' specification in '%s' at line %u\n", tag, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			skip_access = 0;
			state->parity[level].split_mac = split_mac;
			for (s = 0; s < split_mac; ++s) {
				char uuid[UUID_MAX];
				pathimport(state->parity[level].split_map[s].path, sizeof(state->parity[level].split_map[s].path), split_map[s]);

				if (!state->opt.skip_parity_access) {
					struct stat st;

					/* get the device of the directory containing the parity file */
					pathimport(device, sizeof(device), split_map[s]);

					/* use only the dir, as the parity file may not exist yet */
					pathcut(device);

					if (stat(device, &st) == 0) {
						dev = st.st_dev;

						/* read the uuid, if unsupported use an empty one */
						if (devuuid(dev, device, uuid, sizeof(uuid)) != 0) {
							*uuid = 0;
						}
					} else {
						/* if the disk can be skipped */
						if (state->opt.force_device) {
							/* use a fake device, and mark the disk to be skipped */
							dev = 0;
							*uuid = 0;
							skip_access = 1;
							log_error(errno, "DANGER! Skipping inaccessible parity disk '%s'...\n", tag);
						} else {
							/* LCOV_EXCL_START */
							log_fatal(errno, "Error accessing 'parity' dir '%s' specification in '%s' at line %u\n", device, path, line);

							/* in "fix" we allow to continue anyway */
							if (strcmp(state->command, "fix") == 0) {
								log_fatal(errno, "You can '%s' anyway, using 'snapraid --force-device %s'.\n", state->command, state->command);
							}
							exit(EXIT_FAILURE);
							/* LCOV_EXCL_STOP */
						}
					}
				} else {
					/* use a fake device */
					dev = 0;
					*uuid = 0;
				}

				state->parity[level].split_map[s].device = dev;
				pathcpy(state->parity[level].split_map[s].uuid, sizeof(state->parity[level].split_map[s].uuid), uuid);
			}

			/* store the global parity skip_access */
			state->parity[level].skip_access = skip_access;

			/* adjust the level */
			if (state->level < level + 1)
				state->level = level + 1;
		} else if (strcmp(tag, "share") == 0) {
			if (*state->share) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Multiple 'share' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'share' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'share' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			pathimport(state->share, sizeof(state->share), buffer);
		} else if (strcmp(tag, "pool") == 0) {
			struct stat st;

			if (*state->pool) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Multiple 'pool' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'pool' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'pool' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			pathimport(state->pool, sizeof(state->pool), buffer);

			/* get the device of the directory containing the pool tree */
			if (stat(buffer, &st) != 0) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error accessing 'pool' dir '%s' specification in '%s' at line %u\n", buffer, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			state->pool_device = st.st_dev;
		} else if (strcmp(tag, "content") == 0) {
			struct snapraid_content* content;
			char device[PATH_MAX];
			char* slash;
			struct stat st;
			uint64_t dev;

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'content' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (pathcmp(buffer, "/dev/null") == 0 || pathcmp(buffer, "NUL") == 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "You cannot use the null device as 'content' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'content' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* check if the content file is already specified */
			for (i = state->contentlist; i != 0; i = i->next) {
				content = i->data;
				if (pathcmp(content->content, buffer) == 0)
					break;
			}
			if (i) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Duplicate 'content' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* get the device of the directory containing the content file */
			pathimport(device, sizeof(device), buffer);
			slash = strrchr(device, '/');
			if (slash)
				slash[1] = 0;
			else
				pathcpy(device, sizeof(device), ".");
			if (stat(device, &st) == 0) {
				dev = st.st_dev;
			} else {
				if (state->opt.skip_content_access) {
					/* use a fake device */
					dev = 0;
					log_error(errno, "WARNING! Skipping inaccessible content file '%s'...\n", buffer);
				} else {
					/* LCOV_EXCL_START */
					log_fatal(errno, "Error accessing 'content' dir '%s' specification in '%s' at line %u\n", device, path, line);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}
			}

			/* set the lock file at the first accessible content file */
			if (state->lockfile[0] == 0 && dev != 0) {
				pathcpy(state->lockfile, sizeof(state->lockfile), buffer);
				pathcat(state->lockfile, sizeof(state->lockfile), ".lock");
			}

			content = content_alloc(buffer, dev);

			tommy_list_insert_tail(&state->contentlist, &content->node, content);
		} else if (strcmp(tag, "data") == 0 || strcmp(tag, "disk") == 0) {
			/* "disk" is the deprecated name up to SnapRAID 9.x */
			char dir[PATH_MAX];
			char device[PATH_MAX];
			char uuid[UUID_MAX];
			struct snapraid_disk* disk;
			uint64_t dev;
			int skip_access;

			ret = sgettok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'data' name specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'data' name specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			const char* reason = disk_name_invalid_reason(buffer);
			if (reason) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'data' name '%s' in '%s' at line %u. %s.\n", buffer, path, line, reason);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			sgetspace(f);

			ret = sgetlasttok(f, dir, sizeof(dir));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'data' dir specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*dir) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'data' dir specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* get the device of the dir */
			pathimport(device, sizeof(device), dir);

			/* check if the disk name already exists */
			for (i = state->disklist; i != 0; i = i->next) {
				disk = i->data;
				if (pathcmp(disk->name, buffer) == 0)
					break;
			}
			if (i) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Duplicate 'data' name '%s' at line %u\n", buffer, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			for (i = state->extralist; i != 0; i = i->next) {
				struct snapraid_extra* found_extra = i->data;
				if (pathcmp(found_extra->name, buffer) == 0)
					break;
			}
			if (i) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Duplicate 'data' name '%s' already used in 'extra' at line %u\n", buffer, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* if the disk has to be present */
			skip_access = 0;
			if (!state->opt.skip_disk_access) {
				struct stat st;

				if (stat(device, &st) == 0) {
					dev = st.st_dev;

					/* read the uuid, if unsupported use an empty one */
					if (devuuid(dev, device, uuid, sizeof(uuid)) != 0) {
						*uuid = 0;
					}

					/* fake a different UUID when testing */
					if (state->opt.fake_uuid) {
						snprintf(uuid, sizeof(uuid), "fake-uuid-%d", state->opt.fake_uuid);
						--state->opt.fake_uuid;
					}
				} else {
					/* if the disk can be skipped */
					if (state->opt.force_device) {
						/* use a fake device, and mark the disk to be skipped */
						dev = 0;
						*uuid = 0;
						skip_access = 1;
						log_error(errno, "DANGER! Skipping inaccessible data disk '%s'...\n", buffer);
					} else {
						/* LCOV_EXCL_START */
						log_fatal(errno, "Error accessing 'disk' '%s' specification in '%s' at line %u\n", dir, device, line);

						/* in "fix" we allow to continue anyway */
						if (strcmp(state->command, "fix") == 0) {
							log_fatal(errno, "You can '%s' anyway, using 'snapraid --force-device %s'.\n", state->command, state->command);
						}
						exit(EXIT_FAILURE);
						/* LCOV_EXCL_STOP */
					}
				}
			} else {
				/* use a fake device */
				dev = 0;
				*uuid = 0;
			}

			disk = disk_alloc(buffer, dir, dev, uuid, skip_access);

			tommy_list_insert_tail(&state->disklist, &disk->node, disk);
		} else if (strcmp(tag, "extra") == 0) {
			char dir[PATH_MAX];
			char device[PATH_MAX];
			char uuid[UUID_MAX];
			struct snapraid_extra* extra;
			uint64_t dev;

			ret = sgettok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'extra' name specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'extra' name specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			const char* reason = disk_name_invalid_reason(buffer);
			if (reason) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'extra' name '%s' in '%s' at line %u. %s.\n", buffer, path, line, reason);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			sgetspace(f);

			ret = sgetlasttok(f, dir, sizeof(dir));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'extra' dir specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*dir) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'extra' dir specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* get the device of the dir */
			pathimport(device, sizeof(device), dir);

			/* check if the disk name already exists */
			for (i = state->extralist; i != 0; i = i->next) {
				extra = i->data;
				if (pathcmp(extra->name, buffer) == 0)
					break;
			}
			if (i) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Duplicate 'extra' name '%s' at line %u\n", buffer, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			for (i = state->disklist; i != 0; i = i->next) {
				struct snapraid_disk* found_disk = i->data;
				if (pathcmp(found_disk->name, buffer) == 0)
					break;
			}
			if (i) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Duplicate 'extra' name '%s' already used in 'data' at line %u\n", buffer, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* if the disk has to be present */
			if (!state->opt.skip_disk_access) {
				struct stat st;

				if (stat(device, &st) == 0) {
					dev = st.st_dev;

					/* read the uuid, if unsupported use an empty one */
					if (devuuid(dev, device, uuid, sizeof(uuid)) != 0) {
						*uuid = 0;
					}

					/* fake a different UUID when testing */
					if (state->opt.fake_uuid) {
						snprintf(uuid, sizeof(uuid), "fake-uuid-%d", state->opt.fake_uuid);
						--state->opt.fake_uuid;
					}
				} else {
					/* use a fake device, and mark the disk to be skipped */
					dev = 0;
					*uuid = 0;
				}
			} else {
				/* use a fake device */
				dev = 0;
				*uuid = 0;
			}

			extra = extra_alloc(buffer, dir, dev, uuid);

			tommy_list_insert_tail(&state->extralist, &extra->node, extra);
		} else if (strcmp(tag, "smartctl") == 0) {
			char custom[PATH_MAX];
			char parsed_smartctl[SMARTCTL_MAX];
			char parsed_smartctl_info[SMARTCTL_MAX];

			ret = sgettok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'smartctl' name in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'smartctl' name in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			sgetspace(f);

			ret = sgetlasttok(f, custom, sizeof(custom));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'smartctl' option in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*custom) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'smartctl' option in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (parse_smartctl(custom, parsed_smartctl, sizeof(parsed_smartctl), parsed_smartctl_info, sizeof(parsed_smartctl_info)) != 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'smartctl' option in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* search for parity */
			if (lev_config_scan(buffer, &level, 0) == 0) {
				if (state->parity[level].smartctl[0] != 0) {
					/* LCOV_EXCL_START */
					log_fatal(EUSER, "Duplicate parity 'smartctl' '%s' at line %u\n", buffer, line);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}

				pathcpy(state->parity[level].smartctl, sizeof(state->parity[level].smartctl), parsed_smartctl);
				pathcpy(state->parity[level].smartctl_info, sizeof(state->parity[level].smartctl_info), parsed_smartctl_info);
			} else {
				/* search the disk */
				struct snapraid_disk* found_disk = 0;
				for (i = state->disklist; i != 0; i = i->next) {
					struct snapraid_disk* disk = i->data;
					if (pathcmp(disk->name, buffer) == 0) {
						found_disk = disk;
						break;
					}
				}
				struct snapraid_extra* found_extra = 0;
				for (i = state->extralist; i != 0; i = i->next) {
					struct snapraid_extra* extra = i->data;
					if (pathcmp(extra->name, buffer) == 0) {
						found_extra = extra;
						break;
					}
				}
				if (found_disk) {
					if (found_disk->smartctl[0] != 0) {
						/* LCOV_EXCL_START */
						log_fatal(EUSER, "Duplicate 'smartctl' name '%s' at line %u\n", buffer, line);
						exit(EXIT_FAILURE);
						/* LCOV_EXCL_STOP */
					}

					pathcpy(found_disk->smartctl, sizeof(found_disk->smartctl), parsed_smartctl);
					pathcpy(found_disk->smartctl_info, sizeof(found_disk->smartctl_info), parsed_smartctl_info);
				} else if (found_extra) {
					if (found_extra->smartctl[0] != 0) {
						/* LCOV_EXCL_START */
						log_fatal(EUSER, "Duplicate 'smartctl' name '%s' at line %u\n", buffer, line);
						exit(EXIT_FAILURE);
						/* LCOV_EXCL_STOP */
					}

					pathcpy(found_extra->smartctl, sizeof(found_extra->smartctl), parsed_smartctl);
					pathcpy(found_extra->smartctl_info, sizeof(found_extra->smartctl_info), parsed_smartctl_info);
				} else {
					/* LCOV_EXCL_START */
					log_fatal(EUSER, "Unknown 'smartctl' name '%s' at line %u\n", buffer, line);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}
			}
		} else if (strcmp(tag, "smartignore") == 0) {
			struct smartignore_struct* smart;

			ret = sgettok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'smartignore' name in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'smartignore' name in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			sgetspace(f);

			if (strcmp(buffer, "*") == 0) {
				smart = state->smartignore;
			} else if (lev_config_scan(buffer, &level, 0) == 0) { /* search for parity */
				smart = state->parity[level].smartignore;
			} else {
				/* search the disk */
				struct snapraid_disk* found_disk = 0;
				for (i = state->disklist; i != 0; i = i->next) {
					struct snapraid_disk* disk = i->data;
					if (pathcmp(disk->name, buffer) == 0) {
						found_disk = disk;
						break;
					}
				}
				struct snapraid_extra* found_extra = 0;
				for (i = state->extralist; i != 0; i = i->next) {
					struct snapraid_extra* extra = i->data;
					if (pathcmp(extra->name, buffer) == 0) {
						found_extra = extra;
						break;
					}
				}
				if (found_disk) {
					smart = found_disk->smartignore;
				} else if (found_extra) {
					smart = found_extra->smartignore;
				} else {
					/* LCOV_EXCL_START */
					log_fatal(EUSER, "Unknown 'smartignore' name '%s' at line %u\n", buffer, line);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}
			}

			int si = 0;

			while (1) {
				char* e;

				ret = sgettok(f, buffer, sizeof(buffer));
				if (ret < 0) {
					/* LCOV_EXCL_START */
					log_fatal(EUSER, "Invalid 'smartignore' specification in '%s' at line %u\n", path, line);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}

				if (!*buffer)
					break;

				if (si == SMART_IGNORE_MAX) {
					/* LCOV_EXCL_START */
					log_fatal(EUSER, "Too many 'smartignore' specification in '%s' at line %u\n", path, line);
					exit(EXIT_FAILURE);
					/* LCOV_EXCL_STOP */
				}

				int val = strton(buffer, &e, 0);

				if (e && !*e) {
					/* numerical form */
					if (val < 1 || val >= 256) {
						/* LCOV_EXCL_START */
						log_fatal(EUSER,
							"Invalid 'smartignore' specification in '%s' at line %u. It should be between 1 and 255.\n",
							path, line);
						exit(EXIT_FAILURE);
						/* LCOV_EXCL_STOP */
					}
					smart[si].attr_index = val;
					smart[si].attr_name[0] = 0;
				} else {
					/* name form */
					if (strlen(buffer) >= SMART_MAX) {
						/* LCOV_EXCL_START */
						log_fatal(EUSER,
							"Invalid 'smartignore' specification in '%s' at line %u. Attribute name too long.\n",
							path, line);
						exit(EXIT_FAILURE);
						/* LCOV_EXCL_STOP */
					}
					smart[si].attr_index = 0;
					pathcpy(smart[si].attr_name, sizeof(smart[si].attr_name), buffer);
				}

				++si;

				sgetspace(f);
			}

			if (si == 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'smartignore' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		} else if (strcmp(tag, "temp_limit") == 0) {
			unsigned temp;

			ret = sgetu32(f, &temp);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'temp_limit' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (temp < 40 || temp > 70) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'temp_limit' temperature specification in '%s' at line %u. It must be between 40 and 70\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			state->thermal_temperature_limit = temp;
		} else if (strcmp(tag, "temp_sleep") == 0) {
			unsigned time;

			ret = sgetu32(f, &time);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'temp_sleep' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (time < 3 || time > 120) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'temp_sleep' temperature specification in '%s' at line %u. It must be between 3 and 120\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			state->thermal_cooldown_time = time * 60;
		} else if (strcmp(tag, "nohidden") == 0) {
			state->filter_hidden = 1;
		} else if (strcmp(tag, "snapshot") == 0) {
			state->snapshot = 1;
		} else if (strcmp(tag, "exclude") == 0) {
			struct snapraid_filter* filter;

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'exclude' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'exclude' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			filter = filter_alloc_file(-1, "", buffer);
			if (!filter) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'exclude' specification '%s' in '%s' at line %u\n", buffer, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			tommy_list_insert_tail(&state->filterlist, &filter->node, filter);
		} else if (strcmp(tag, "include") == 0) {
			struct snapraid_filter* filter;

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'include' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'include' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			filter = filter_alloc_file(1, "", buffer);
			if (!filter) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'include' specification '%s' in '%s' at line %u\n", buffer, path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			tommy_list_insert_tail(&state->filterlist, &filter->node, filter);
		} else if (strcmp(tag, "autosave") == 0) {
			char* e;

			ret = sgetlasttok(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'autosave' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (!*buffer) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Empty 'autosave' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			state->autosave = strtonl(buffer, &e, 0);

			if (!e || *e) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid 'autosave' specification in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* convert to GB */
			state->autosave *= GIGA;
		} else if (tag[0] == 0) {
			/* allow empty lines */
		} else if (tag[0] == '#') {
			ret = sgetline(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid comment in '%s' at line %u\n", path, line);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		} else {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Invalid command '%s' in '%s' at line %u\n", tag, path, line);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		/* skip final spaces */
		sgetspace(f);

		/* next line */
		c = sgeteol(f);
		if (c == EOF) {
			break;
		}
		if (c != '\n') {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Extra data in '%s' at line %u\n", path, line);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
		++line;
	}

	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error reading the configuration file '%s' at line %u\n", path, line);
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	sclose(f);

	state_config_check(state, path, filterlist_disk);

	/* select the default hash */
	if (state->opt.force_murmur3) {
		state->besthash = HASH_MURMUR3;
	} else if (state->opt.force_spooky2) {
		state->besthash = HASH_SPOOKY2;
	} else {
		state->besthash = membesthash();
	}

	/* by default use the best hash */
	state->hash = state->besthash;

	/* by default use a random hash seed */
	if (os_randomize(state->hashseed, HASH_MAX) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Failed to retrieve random values.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* no previous hash by default */
	state->prevhash = HASH_UNDEFINED;

	/* intentionally not set the prevhashseed, if used valgrind will warn about it */

	log_tag("blocksize:%u\n", state->block_size);
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		log_tag("data:%s:%s:%s\n", disk->name, esc_tag(disk->mount_point), disk->uuid);
	}
	for (i = state->extralist; i != 0; i = i->next) {
		struct snapraid_extra* extra = i->data;
		log_tag("extra:%s:%s:%s\n", esc_tag(extra->name), esc_tag(extra->dir), extra->uuid);
	}

	log_tag("mode:%s\n", lev_raid_name(state->raid_mode, state->level));
	for (l = 0; l < state->level; ++l) {
		log_tag("%s:%s:%s\n", lev_config_name(l), esc_tag(state->parity[l].split_map[0].path), state->parity[l].split_map[0].uuid);
		for (s = 1; s < state->parity[l].split_mac; ++s) {
			log_tag("%s/%u:%s:%s\n", lev_config_name(l), s, esc_tag(state->parity[l].split_map[s].path), state->parity[l].split_map[s].uuid);
		}
	}
	if (state->pool[0] != 0)
		log_tag("pool:%s\n", esc_tag(state->pool));
	if (state->share[0] != 0)
		log_tag("share:%s\n", esc_tag(state->share));
	if (state->autosave != 0)
		log_tag("autosave:%" PRIu64 "\n", state->autosave);
	for (i = tommy_list_head(&state->filterlist); i != 0; i = i->next) {
		char out[PATH_MAX];
		struct snapraid_filter* filter = i->data;
		log_tag("filter:%s\n", esc_tag(filter_type(filter, out, sizeof(out))));
	}
	if (state->filter_hidden)
		log_tag("filter:nohidden:\n");
	log_flush();
}

/**
 * Find a disk by name.
 */
static struct snapraid_disk* find_disk_by_name(struct snapraid_state* state, const char* name)
{
	tommy_node* i;

	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		if (pathcmp(disk->name, name) == 0)
			return disk;
	}

	if (state->no_conf) {
		/* without a configuration file, add disks automatically */
		struct snapraid_disk* disk;

		disk = disk_alloc(name, "DUMMY/", -1, "", 0);

		tommy_list_insert_tail(&state->disklist, &disk->node, disk);

		return disk;
	}

	return 0;
}

/**
 * Find a disk by UUID.
 */
static struct snapraid_disk* find_disk_by_uuid(struct snapraid_state* state, const char* uuid)
{
	tommy_node* i;
	struct snapraid_disk* found = 0;

	/*
	 * Special test case to find the first matching UUID
	 * when testing UUID are all equal or not supported
	 * and we should handle this case specifically
	 */
	if (state->opt.match_first_uuid)
		return state->disklist->data;

	/* LCOV_EXCL_START */
	/* never find an empty uuid */
	if (!*uuid)
		return 0;

	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		if (strcmp(disk->uuid, uuid) == 0) {
			/* never match duplicate UUID */
			if (found)
				return 0;
			found = disk;
		}
	}

	return found;
	/* LCOV_EXCL_STOP */
}

/**
 * Update the disk mapping if required.
 */
static void state_map(struct snapraid_state* state)
{
	unsigned hole;
	tommy_node* i;
	unsigned uuid_mismatch;
	size_t diskcount;
	unsigned l, s;

	/*
	 * Maps each unmapped disk present in the configuration file in the first available hole
	 * this happens when you add disks for the first time in the configuration file
	 */
	hole = 0; /* first position to try */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		struct snapraid_map* map;
		tommy_node* j;

		/* check if the disk is already mapped */
		for (j = state->maplist; j != 0; j = j->next) {
			map = j->data;
			if (pathcmp(disk->name, map->name) == 0) {
				/* mapping found */
				break;
			}
		}
		if (j != 0) {
			/* mapping is present, then copy the free blocks into to disk */
			disk->total_blocks = map->total_blocks;
			disk->free_blocks = map->free_blocks;
			continue;
		}

		/* mapping not found, search for an hole */
		while (1) {
			for (j = state->maplist; j != 0; j = j->next) {
				map = j->data;
				if (map->position == hole) {
					/* position already used */
					break;
				}
			}
			if (j == 0) {
				/* hole found */
				break;
			}

			/* try with the next one */
			++hole;
		}

		/* insert the new mapping */
		map = map_alloc(disk->name, hole, 0, 0, "");

		tommy_list_insert_tail(&state->maplist, &map->node, map);
	}

	/* without configuration don't check for number of data disks or uuid changes */
	if (state->no_conf)
		return;

	/* counter for the number of UUID mismatches */
	uuid_mismatch = 0;

	/* check if mapping match the disk uuid */
	if (!state->opt.skip_disk_access) {
		for (i = state->maplist; i != 0; i = i->next) {
			struct snapraid_map* map = i->data;
			struct snapraid_disk* disk;

			disk = find_disk_by_name(state, map->name);
			if (disk == 0) {
				/* LCOV_EXCL_START */
				log_fatal(EINTERNAL, "Internal inconsistency: Mapping not found for '%s'\n", map->name);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (disk->uuid[0] == 0) {
				/* if uuid is not available, skip this one */
				continue;
			}

			/* if the uuid is changed */
			if (strcmp(disk->uuid, map->content_uuid) != 0) {
				/* mark the disk as with an UUID change */
				disk->has_different_uuid = 1;

				/* if the previous uuid is available */
				if (map->content_uuid[0] != 0) {
					/* count the number of uuid change */
					++uuid_mismatch;
					log_error(ESOFT, "UUID change for disk '%s' from '%s' to '%s'\n", disk->name, map->content_uuid, disk->uuid);
				} else {
					/*
					 * No message here, because having a disk without
					 * UUID is the normal state of an empty disk
					 */
					disk->had_empty_uuid = 1;
				}

				/* write the new state with the new uuid */
				state->need_write = 1;
			}
		}
	}

	/* check the parity uuid */
	if (!state->opt.skip_parity_access) {
		for (l = 0; l < state->level; ++l) {
			for (s = 0; s < state->parity[l].split_mac; ++s) {
				if (state->parity[l].split_map[s].uuid[0] == 0) {
					/* if uuid is not available, skip this one */
					continue;
				}

				/* if the uuid is changed */
				if (strcmp(state->parity[l].split_map[s].uuid, state->parity[l].split_map[s].content_uuid) != 0) {
					/* if the previous uuid is available */
					if (state->parity[l].split_map[s].content_uuid[0] != 0) {
						/* count the number of uuid change */
						++uuid_mismatch;
						log_error(ESOFT, "UUID change for parity '%s/%u' from '%s' to '%s'\n", lev_config_name(l), s, state->parity[l].split_map[s].content_uuid, state->parity[l].split_map[s].uuid);
					}

					/* write the new state with the new uuid */
					state->need_write = 1;
				}
			}
		}
	}

	if (!state->opt.force_uuid && uuid_mismatch > state->level) {
		/* LCOV_EXCL_START */
		log_fatal(ESOFT, "Too many disks have changed UUIDs since the last 'sync'.\n");
		log_fatal(ESOFT, "If this happened because you actually replaced them,\n");
		log_fatal(ESOFT, "you can still '%s', using 'snapraid --force-uuid %s'.\n", state->command, state->command);
		log_fatal(ESOFT, "Alternatively, you may have misconfigured the disk mount points,\n");
		log_fatal(ESOFT, "and you need to restore the mount points to the state of the latest sync.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* count the number of data disks, including holes left after removing some */
	diskcount = 0;
	for (i = state->maplist; i != 0; i = i->next) {
		struct snapraid_map* map = i->data;

		if (map->position + 1 > diskcount)
			diskcount = map->position + 1;
	}

	/* ensure to don't go over the limit of the RAID engine */
	if (diskcount > RAID_DATA_MAX) {
		/* LCOV_EXCL_START */
		log_fatal(ESOFT, "Too many data disks. Maximum allowed is %u.\n", RAID_DATA_MAX);
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* now count the real number of data disks, excluding holes left after removing some */
	diskcount = tommy_list_count(&state->maplist);

	/* recommend number of parities */
	if (!state->opt.no_warnings) {
		/* LCOV_EXCL_START */
		if (diskcount >= 36 && state->level < 6) {
			log_info(EUSER, "WARNING! For %zu disks, it's recommended to use six parity levels.\n", diskcount);
		} else if (diskcount >= 29 && state->level < 5) {
			log_info(EUSER, "WARNING! For %zu disks, it's recommended to use five parity levels.\n", diskcount);
		} else if (diskcount >= 22 && state->level < 4) {
			log_info(EUSER, "WARNING! For %zu disks, it's recommended to use four parity levels.\n", diskcount);
		} else if (diskcount >= 15 && state->level < 3) {
			log_info(EUSER, "WARNING! For %zu disks, it's recommended to use three parity levels.\n", diskcount);
		} else if (diskcount >= 5 && state->level < 2) {
			log_info(EUSER, "WARNING! For %zu disks, it's recommended to use two parity levels.\n", diskcount);
		}
		/* LCOV_EXCL_STOP */
	}
}

void state_refresh(struct snapraid_state* state)
{
	tommy_node* i;
	unsigned l, s;
	uint64_t bs = (uint64_t)state->block_size;

	/* for all disks */
	for (i = state->maplist; i != 0; i = i->next) {
		struct snapraid_map* map = i->data;
		struct snapraid_disk* disk;
		uint64_t total_space;
		uint64_t free_space;
		int ret;

		disk = find_disk_by_name(state, map->name);
		if (disk == 0) {
			/* LCOV_EXCL_START */
			log_fatal(EINTERNAL, "Internal inconsistency: Mapping not found for '%s'\n", map->name);
			os_abort();
			/* LCOV_EXCL_STOP */
		}

		if (disk->skip_access)
			continue;

		ret = fsinfo(disk->mount_point, 0, 0, &total_space, &free_space, disk->fstype, sizeof(disk->fstype), disk->fslabel, sizeof(disk->fslabel));
		if (ret != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error accessing disk '%s' to retrieve file-system info. %s.\n", disk->mount_point, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		/* set the new free blocks */
		map->total_blocks = total_space / bs;
		map->free_blocks = free_space / bs;

		/* also update the disk info */
		disk->total_blocks = map->total_blocks;
		disk->free_blocks = map->free_blocks;

		log_tag("fsinfo_data_split:%s:%" PRIu64 ":%" PRIu64 ":%s:%s\n", disk->name, disk->total_blocks * bs, disk->free_blocks * bs, disk->fstype, esc_tag(disk->fslabel));
		log_tag("fsinfo_data:%s:%" PRIu64 ":%" PRIu64 "\n", disk->name, disk->total_blocks * bs, disk->free_blocks * bs);
	}

	/* for all parities */
	for (l = 0; l < state->level; ++l) {
		/* set the new free blocks */
		state->parity[l].total_blocks = 0;
		state->parity[l].free_blocks = 0;

		if (state->parity[l].skip_access)
			continue;

		for (s = 0; s < state->parity[l].split_mac; ++s) {
			struct snapraid_split* split = &state->parity[l].split_map[s];
			uint64_t total_space;
			uint64_t free_space;
			int ret;

			ret = fsinfo(split->path, 0, 0, &total_space, &free_space, split->fstype, sizeof(split->fstype), split->fslabel, sizeof(split->fslabel));
			if (ret != 0) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error accessing file '%s' to retrieve file-system info. %s.\n", split->path, strerror(errno));
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

#ifdef WIN32
			/*
			 * Take into account the space we have to leave free
			 * to avoid the waring about low space
			 */
			if (free_space >= WINDOWS_SPACEHOLDER_SIZE)
				free_space -= WINDOWS_SPACEHOLDER_SIZE;
			else
				free_space = 0;
#endif

			/* add the new free blocks */
			uint64_t split_total_blocks = total_space / bs;
			uint64_t split_free_blocks = free_space / bs;

			state->parity[l].total_blocks += split_total_blocks;
			state->parity[l].free_blocks += split_free_blocks;

			if (s == 0)
				log_tag("fsinfo_parity_split:%s:%" PRIu64 ":%" PRIu64 ":%s:%s\n", lev_config_name(l), split_total_blocks * bs, split_free_blocks * bs, split->fstype, esc_tag(split->fslabel));
			else
				log_tag("fsinfo_parity_split:%s/%u:%" PRIu64 ":%" PRIu64 ":%s:%s\n", lev_config_name(l), s, split_total_blocks * bs, split_free_blocks * bs, split->fstype, esc_tag(split->fslabel));
		}

		log_tag("fsinfo_parity:%s:%" PRIu64 ":%" PRIu64 "\n", lev_config_name(l), state->parity[l].total_blocks * bs, state->parity[l].free_blocks * bs);
	}

	/* for all extra disks */
	for (i = state->extralist; i != 0; i = i->next) {
		struct snapraid_extra* extra = i->data;
		uint64_t total_space;
		uint64_t free_space;
		int ret;

		ret = fsinfo(extra->dir, 0, 0, &total_space, &free_space, extra->fstype, sizeof(extra->fstype), extra->fslabel, sizeof(extra->fslabel));
		if (ret != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error accessing disk '%s' to retrieve file-system info. %s.\n", extra->dir, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		log_tag("fsinfo_extra:%s:%" PRIu64 ":%" PRIu64 "\n", esc_tag(extra->name), total_space, free_space);
	}

	/*
	 * Note what we don't set need_write = 1, because we don't want
	 * to update the content file only for the free space info.
	 */
}

/**
 * Check the content.
 */
static void state_content_check(struct snapraid_state* state, const char* path)
{
	tommy_node* i;

	/* check that any map has different name and position */
	for (i = state->maplist; i != 0; i = i->next) {
		struct snapraid_map* map = i->data;
		tommy_node* j;
		for (j = i->next; j != 0; j = j->next) {
			struct snapraid_map* other = j->data;
			if (pathcmp(map->name, other->name) == 0) {
				/* LCOV_EXCL_START */
				log_fatal(ECONTENT, "Conflicting 'map' disk specification in '%s'\n", path);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
			if (map->position == other->position) {
				/* LCOV_EXCL_START */
				log_fatal(ECONTENT, "Conflicting 'map' index specification in '%s'\n", path);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}
	}
}

/**
 * Analyze if the position is REQUIRED, if parity is unsynced, and if it has
 * DELETED or REBUILD blocks.
 *
 * A DELETED block makes parity invalid, but it counts as unsynced only when at
 * least one disk still has a file at the same position. This is the same logic
 * used by "status" to detect an incomplete sync.
 */
static int fs_position_analyze(struct snapraid_state* state, block_off_t pos, int* is_unsynced, int* has_deleted, int* has_rebuild)
{
	tommy_node* i;
	int one_file = 0;
	int one_invalid = 0;
	int one_deleted = 0;
	int one_rebuild = 0;

	/* check for each disk */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		struct snapraid_block* block = fs_par2block_find(disk, pos);
		unsigned block_state = block_state_get(block);

		if (block_has_file(block))
			one_file = 1;
		if (block_has_invalid_parity(block))
			one_invalid = 1;
		if (block_state == BLOCK_STATE_DELETED)
			one_deleted = 1;
		if (block_state == BLOCK_STATE_REBUILD)
			one_rebuild = 1;
	}

	*is_unsynced = one_file && one_invalid;
	*has_deleted = one_deleted;
	*has_rebuild = one_rebuild;

	return one_file;
}

/**
 * Check if the info block is REQUIREQ.
 *
 * This is used to ensure that we keep the last check used for scrubbing.
 * and that we add it when importing old context files.
 *
 * Note that you can have position without info blocks, for example
 * if all the blocks are not synced.
 *
 * Note also that not requiring an info block, doesn't mean that if present it
 * can be discarded.
 */
static int fs_info_is_required(struct snapraid_state* state, block_off_t pos)
{
	tommy_node* i;

	/* check for each disk */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		struct snapraid_block* block = fs_par2block_find(disk, pos);

		/* if we have at least one synced file, the info is required */
		if (block_state_get(block) == BLOCK_STATE_BLK)
			return 1;
	}

	return 0;
}

static void fs_position_clear_deleted(struct snapraid_state* state, block_off_t pos)
{
	tommy_node* i;

	/* check for each disk if block is really used */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		struct snapraid_block* block = fs_par2block_find(disk, pos);

		/* if the block is deleted */
		if (block_state_get(block) == BLOCK_STATE_DELETED) {
			/* set it to empty */
			fs_deallocate(disk, pos);
		}
	}
}

/**
 * Check if a block position in a disk is deleted.
 */
static inline int fs_is_block_deleted(struct snapraid_disk* disk, block_off_t pos)
{
	struct snapraid_block* block = fs_par2block_find(disk, pos);

	return block_state_get(block) == BLOCK_STATE_DELETED;
}

/**
 * Check if a block parity needs sync
 *
 * This is the same logic used in "status" to detect an incomplete "sync",
 * that ignores invalid block, if they are not used by a file in any disk.
 *
 * This means that DELETED blocks won't necessarily imply an invalid parity.
 */
static int fs_is_block_unsynced(struct snapraid_state* state, block_off_t pos)
{
	int one_invalid = 0;
	int one_valid = 0;
	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		struct snapraid_block* block = fs_par2block_find(disk, pos);

		if (block_has_file(block))
			one_valid = 1;
		if (block_has_invalid_parity(block))
			one_invalid = 1;

		if (one_invalid && one_valid)
			return 1;
	}

	return 0;
}

/**
 * Flush the file checking the final CRC and abort on error.
 * We exploit the fact that the CRC is always stored in the last 4 bytes.
 *
 * Notice that on corruption, decoding error, or CRC mismatch we intentionally
 * exit with a fatal error rather than automatically falling back to another
 * content file. Automatic fallback could silently mask disk or memory corruption
 * or inadvertently load a stale state. We require the user to explicitly rename
 * or delete the corrupted file so SnapRAID will fall back to the surviving copy
 * on the next run.
 */
static void decoding_error(const char* path, STREAM* f)
{
	unsigned char buf[4];
	uint32_t crc_stored;
	uint32_t crc_computed;

	if (seof(f)) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Unexpected end of content file '%s' at offset %" PRIi64 "\n", path, stell(f));
		log_fatal(ECONTENT, "The content file '%s' is truncated!\n", path);
		log_fatal(ECONTENT, "To recover, rename or delete it and rerun the command.\n");
		log_fatal(ECONTENT, "SnapRAID will automatically fall back to the next healthy copy.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error reading the content file '%s' at offset %" PRIi64 "\n", path, stell(f));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	log_fatal(ECONTENT, "Error decoding '%s' at offset %" PRIi64 "\n", path, stell(f));

	if (sdeplete(f, buf) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Failed to flush content file '%s' at offset %" PRIi64 "\n", path, stell(f));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* get the stored crc from the last four bytes */
	crc_stored = buf[0] | (uint32_t)buf[1] << 8 | (uint32_t)buf[2] << 16 | (uint32_t)buf[3] << 24;

	/* get the computed crc */
	crc_computed = scrc(f);

	/* adjust the stored crc to include itself */
	crc_stored = crc32c(crc_stored, buf, 4);

	if (crc_computed != crc_stored) {
		log_fatal(ECONTENT, "CRC mismatch in '%s'\n", path);
		log_fatal(ECONTENT, "The content file '%s' is damaged or corrupted (CRC mismatch)!\n", path);
		log_fatal(ECONTENT, "To recover, rename or delete it and rerun the command.\n");
		log_fatal(ECONTENT, "SnapRAID will automatically fall back to the next healthy copy.\n");
		exit(EXIT_FAILURE);
	} else {
		log_fatal(ECONTENT, "The CRC of the file is correct!\n");
	}
}

/**
 * Read and initialize a complete run of file blocks.
 *
 * CRC is computed when the stream buffer is filled. Advancing the stream
 * pointer directly over cached hashes therefore preserves the same integrity
 * check performed by sread().
 */
static int state_read_block_run(STREAM* f, struct snapraid_file* file, block_off_t file_pos, block_off_t count, unsigned state, int discard_hash)
{
	unsigned char* block_ptr = (unsigned char*)file_block(file, file_pos);
	size_t block_stride = block_sizeof();

	while (count) {
		size_t available = (size_t)(f->end - f->pos);
		block_off_t cached_count = available / BLOCK_HASH_SIZE;
		unsigned char* input;

		if (cached_count > count)
			cached_count = count;

		if (cached_count == 0) {
			struct snapraid_block* block = (struct snapraid_block*)block_ptr;
			int ret;

			block_state_set(block, state);
			ret = sread(f, block->hash, BLOCK_HASH_SIZE);
			if (ret < 0)
				return -1;
			if (discard_hash)
				hash_invalid_set(block->hash);

			block_ptr += block_stride;
			--count;
			continue;
		}

		input = f->pos;
		for (block_off_t i = 0; i < cached_count; ++i) {
			struct snapraid_block* block = (struct snapraid_block*)block_ptr;

			block_state_set(block, state);
			if (discard_hash)
				hash_invalid_set(block->hash);
			else
				hash_copy(block->hash, input);

			input += BLOCK_HASH_SIZE;
			block_ptr += block_stride;
		}

		f->pos = input;
		count -= cached_count;
	}

	return 0;
}

/**
 * Enable or disable lock-free access to the extent trees of all disks.
 * Callers must use this helper in matched, local pairs around operations that
 * are known to be the only extent-tree users.
 */
static void fs_single_thread(struct snapraid_state* state, int single_thread)
{
#if HAVE_THREAD
	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		disk->single_thread = single_thread;
	}
#else
	(void)state;
	(void)single_thread;
#endif
}

/**
 * Context for reading the content file.
 */
struct state_read_context {
	block_off_t blockmax;
	uint32_t count_map;
	uint64_t count_file;
	uint64_t count_hardlink;
	uint64_t count_symlink;
	uint64_t count_dir;
	uint64_t count_bad;
	uint64_t count_rehash;
	uint64_t count_unsynced;
	uint64_t count_unscrubbed;
};

static void state_read_content(struct snapraid_state* state, const char* path, STREAM* f, struct state_read_context* context)
{
	block_off_t blockmax;
	uint64_t count_file;
	uint64_t count_hardlink;
	uint64_t count_symlink;
	uint64_t count_dir;
	uint64_t count_bad;
	uint64_t count_rehash;
	uint64_t count_unsynced;
	uint64_t count_unscrubbed;
	int crc_checked;
	int has_block_size;
	int has_blockmax;
	int has_hash_size;
	int has_hash;
	int has_prevhash;
	int has_info;
	int has_invalid_parity;
	char buffer[PATH_MAX];
	int ret;
	tommy_array disk_mapping;
	uint32_t mapping_max;
	tommy_hashdyn bucket_hash;

	blockmax = 0;
	count_file = 0;
	count_hardlink = 0;
	count_symlink = 0;
	count_dir = 0;
	count_bad = 0;
	count_rehash = 0;
	count_unsynced = 0;
	count_unscrubbed = 0;
	crc_checked = 0;
	has_block_size = 0;
	has_blockmax = 0;
	has_hash_size = 0;
	has_hash = 0;
	has_prevhash = 0;
	has_info = 0;
	has_invalid_parity = 0;
	mapping_max = 0;
	tommy_array_init(&disk_mapping);
	tommy_hashdyn_init(&bucket_hash);

	ret = sread(f, buffer, 12);
	if (ret < 0) {
		/* LCOV_EXCL_START */
		decoding_error(path, f);
		log_fatal(ECONTENT, "Invalid header: read failed!\n");
		os_abort();
		/* LCOV_EXCL_STOP */
	}

	/*
	 * File format versions:
	 *  - SNAPCNT1/2 are legacy formats without 'Q' parity split sizes and are unsupported.
	 *  - SNAPCNT3/SnapRAID 11.0 Adds entry 'y' for hash size.
	 *  - SNAPCNT3/SnapRAID 11.0 Adds entry 'Q' for multi parity file.
	 *  - SNAPCNT4/SnapRAID 15.0 Adds entry 'd' for deallocated files;
	 *  - SNAPCNT4/SnapRAID 15.0 Adds block type 'r' / BLOCK_STATE_REBUILD for persistent in-place parity rebuilds.
	 */
	if (memcmp(buffer, "SNAPCNT1\n\3\0\0", 12) == 0
		|| memcmp(buffer, "SNAPCNT2\n\3\0\0", 12) == 0
	) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Legacy content format is no longer supported.\n");
		log_fatal(ECONTENT, "Run 'snapraid sync' with SnapRAID 14.x to upgrade the content file.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (memcmp(buffer, "SNAPCNT3\n\3\0\0", 12) != 0
		&& memcmp(buffer, "SNAPCNT4\n\3\0\0", 12) != 0
	) {
		/* LCOV_EXCL_START */
		if (memcmp(buffer, "SNAPCNT", 7) != 0) {
			decoding_error(path, f);
			log_fatal(ECONTENT, "Invalid header: magic mismatch!\n");
			os_abort();
		} else {
			log_fatal(ECONTENT, "The content file '%s' was generated with a newer version of SnapRAID!\n", path);
			exit(EXIT_FAILURE);
		}
		/* LCOV_EXCL_STOP */
	}

	while (1) {
		int c;

		/* read the command */
		c = sgetc(f);
		if (c == EOF) {
			break;
		}

		if (c == 'f') {
			/* file */
			char sub[PATH_MAX];
			uint64_t v_size;
			uint64_t v_mtime_sec;
			uint32_t v_mtime_nsec;
			uint64_t v_inode;
			block_off_t v_idx;
			struct snapraid_file* file;
			struct snapraid_disk* disk;
			uint32_t mapping;

			ret = sgetb32(f, &mapping);
			if (ret < 0 || mapping >= mapping_max) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' mapping index out of range\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			disk = tommy_array_get(&disk_mapping, mapping);

			ret = sgetb64(f, &v_size);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' has an invalid size\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* ensure the unsigned serialized size fits in signed data_off_t before conversion. */
			if (v_size > INT64_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'f' size %" PRIu64 " exceeds maximum!\n", v_size);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (state->block_size == 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' encountered with zero blocksize\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (!has_blockmax) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' missing 'blockmax' in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* check for impossible file size to avoid to crash for a too big allocation */
			if (v_size / state->block_size > blockmax) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' size is too big!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb64(f, &v_mtime_sec);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' has an invalid mtime_sec\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb32(f, &v_mtime_nsec);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' has an invalid mtime_nsec\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/*
			 * STAT_NSEC_INVALID is encoded as 0.
			 * Values > 1000000000 can occur in content files written by SnapRAID <= 14.10
			 * on Windows for files with pre-1970 timestamps due to negative nanoseconds.
			 * Normalize them to STAT_NSEC_INVALID.
			 */
			if (v_mtime_nsec == 0 || v_mtime_nsec > 1000000000)
				v_mtime_nsec = STAT_NSEC_INVALID;
			else
				--v_mtime_nsec;

			ret = sgetb64(f, &v_inode);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' has an invalid inode\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetbs(f, sub, sizeof(sub));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' has an invalid subpath\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			if (!path_is_sub(sub)) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' has invalid subpath '%s'!\n", sub);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/*
			 * Allocate without initializing blocks. The runs below cover the whole
			 * file and set every state and hash before exposing the loaded state.
			 */
			file = file_alloc(state->block_size, sub, v_size, v_mtime_sec, v_mtime_nsec, v_inode);

			/* insert the file in the file containers */
			if (file->inode != INODE_INVALID)
				tommy_hashdyn_insert(&disk->inodeset, &file->nodeset, file, file_inode_hash(file->inode));
			tommy_hashdyn_insert(&disk->pathset, &file->pathset, file, file_path_hash(file->sub));
			tommy_hashdyn_insert(&disk->stampset, &file->stampset, file, file_stamp_hash(file->size, file->mtime_sec, file->mtime_nsec));
			tommy_list_insert_tail(&disk->filelist, &file->nodelist, file);

			/* read all the blocks */
			v_idx = 0;
			while (v_idx < file->blockmax) {
				block_off_t v_pos;
				block_off_t v_count;
				block_off_t v_file_pos;
				unsigned v_state;
				int discard_hash;

				/* get the "subcommand */
				c = sgetc(f);

				ret = sgetb64(f, &v_pos);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' block run has an invalid pos\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetb64(f, &v_count);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' block run has an invalid count\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				if (v_count == 0 || v_count > file->blockmax || v_idx > file->blockmax - v_count) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' block number out of range\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				if (v_count > blockmax || v_pos > blockmax - v_count) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' block position out of range %" PRIu64 "/%" PRIu64 "!\n", blockmax, v_pos + v_count);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				discard_hash = 0;
				switch (c) {
				case 'b' :
					v_state = BLOCK_STATE_BLK;
					break;
				case 'r' :
					v_state = BLOCK_STATE_REBUILD;
					break;
				case 'g' :
					v_state = BLOCK_STATE_CHG;
					break;
				case 'p' :
					if (state->opt.force_nocopy) {
						/* discard stored copy hashes and convert REP blocks to CHG */
						v_state = BLOCK_STATE_CHG;
						discard_hash = 1;
					} else {
						v_state = BLOCK_STATE_REP;
					}
					break;
				default :
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'f' has invalid block type '%c'!\n", (char)c);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/* all file block states except BLK require a parity update */
				if (v_state != BLOCK_STATE_BLK)
					has_invalid_parity = 1;

				v_file_pos = v_idx;
				ret = state_read_block_run(f, file, v_file_pos, v_count, v_state, discard_hash);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'f' error reading block run\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}
				v_idx += v_count;

				/* set the parity association for the whole run */
				fs_allocate(disk, v_pos, file, v_file_pos, v_count);
			}

			/* stat */
			++count_file;
		} else if (c == 'i') {
			/* "inf" command */
			snapraid_info info;
			block_off_t v_pos;
			uint64_t v_oldest;

			if (has_info || !has_blockmax) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' duplicate or misplaced in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			has_info = 1;

			ret = sgetb64(f, &v_oldest);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' has an invalid oldest timestamp\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* ensure the unsigned serialized timestamp fits in signed time_t before conversion. */
			if (v_oldest > INT64_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'i' has an out-of-range oldest timestamp %" PRIu64 "!\n", v_oldest);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			v_pos = 0;
			while (v_pos < blockmax) {
				int bad;
				int rehash;
				int justsynced;
				uint64_t t64;
				uint32_t flag;
				block_off_t v_count;

				ret = sgetb64(f, &v_count);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' run has an invalid count\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				if (v_count > blockmax || v_pos > blockmax - v_count) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' run size out of range %" PRIu64 "/%" PRIu64 "!\n", blockmax, v_pos + v_count);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetb32(f, &flag);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' has invalid flags\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/* if there is an info */
				if ((flag & 1) != 0) {
					/* read the time */
					ret = sgetb64(f, &t64);
					if (ret < 0) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' has an invalid timestamp\n");
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					/* ensure the timestamp addition fits in signed time_t without overflow. */
					if (t64 > INT64_MAX - v_oldest) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(ECONTENT, "Internal inconsistency: Field 'i' has an overflowing timestamp %" PRIu64 "!\n", t64);
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					/* analyze the flags */
					bad = (flag & 2) != 0;
					rehash = (flag & 4) != 0;
					justsynced = (flag & 8) != 0;

					if (bad)
						count_bad += v_count;
					if (rehash)
						count_rehash += v_count;
					if (justsynced)
						count_unscrubbed += v_count;

					if (rehash && state->prevhash == HASH_UNDEFINED) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' requires previous checksum but none defined\n");
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					info = info_make(t64 + v_oldest, bad, rehash, justsynced);
					if (!info) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' failed to create info object\n");
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					bucket_insert(&bucket_hash, t64 + v_oldest, v_count, justsynced);

					/* insert the complete run in the info array */
					info_set_run(&state->infoarr, v_pos, v_count, info);
					v_pos += v_count;
				} else {
					while (v_count) {
						/* ensure that an info is present only for used positions */
						if (fs_info_is_required(state, v_pos)) {
							/* LCOV_EXCL_START */
							decoding_error(path, f);
							log_fatal(EINTERNAL, "Internal inconsistency: Field 'i' missing required info for position %" PRIu64 "\n", (uint64_t)v_pos);
							os_abort();
							/* LCOV_EXCL_STOP */
						}

						/* go to next block */
						++v_pos;
						--v_count;
					}
				}
			}

			/*
			 * A synchronized content has no invalid block state, so its unsynced
			 * count is certainly zero. Scan the extent maps only when needed.
			 */
			if (has_invalid_parity) {
				for (v_pos = 0; v_pos < blockmax; ++v_pos)
					if (fs_is_block_unsynced(state, v_pos))
						++count_unsynced;
			}
		} else if (c == 'h') {
			/* hole */
			block_off_t v_pos;
			struct snapraid_disk* disk;
			uint32_t mapping;

			ret = sgetb32(f, &mapping);
			if (ret < 0 || mapping >= mapping_max) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'h' mapping index out of range\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			disk = tommy_array_get(&disk_mapping, mapping);

			if (!has_blockmax) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'h' missing 'blockmax' in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			v_pos = 0;
			while (v_pos < blockmax) {
				block_off_t v_count;
				struct snapraid_file* deleted;

				ret = sgetb64(f, &v_count);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'h' run has an invalid count\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				if (v_count == 0 || v_count > blockmax || v_pos > blockmax - v_count) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'h' size out of range %" PRIu64 "/%" PRIu64 "!\n", blockmax, v_pos + v_count);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/* get the sub-command */
				c = sgetc(f);

				switch (c) {
				case 'o' :
					/* if it's a run of deleted blocks */
					has_invalid_parity = 1;

					/* ensure the converted byte size fits in signed data_off_t without overflow */
					if (v_count > INT64_MAX / (uint64_t)state->block_size) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(ECONTENT, "Internal inconsistency: Field 'h' deleted block count %" PRIu64 " overflows size!\n", v_count);
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					/*
					 * Allocate without initializing blocks. This deleted run sets every
					 * state and hash before the file is inserted in the extent map.
					 */
					deleted = file_alloc(state->block_size, "<deleted>", v_count * state->block_size, 0, 0, 0);

					/* mark the file as deleted */
					file_flag_set(deleted, FILE_IS_DELETED);

					/* insert it in the list of deleted files */
					tommy_list_insert_tail(&disk->deletedlist, &deleted->nodelist, deleted);

					/* read all blocks in the deleted run */
					ret = state_read_block_run(f, deleted, 0, v_count, BLOCK_STATE_DELETED, 0);
					if (ret < 0) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(EINTERNAL, "Internal inconsistency: Field 'h' error reading deleted block run\n");
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					/* insert the whole deleted run in the extent map */
					fs_allocate(disk, v_pos, deleted, 0, v_count);
					v_pos += v_count;
					break;
				case 'O' :
					/* go to the next run */
					v_pos += v_count;
					break;
				default :
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'h' has invalid hole type '%c'!\n", (char)c);
					os_abort();
					/* LCOV_EXCL_STOP */
				}
			}
		} else if (c == 'd') {
			/* dealloc */
			block_off_t v_count;
			block_off_t v_pos;
			struct snapraid_disk* disk;
			uint32_t mapping;

			ret = sgetb32(f, &mapping);
			if (ret < 0 || mapping >= mapping_max) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' mapping index out of range\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			disk = tommy_array_get(&disk_mapping, mapping);

			ret = sgetb64(f, &v_count);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' has an invalid count\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			log_tag("content_info:dealloc:%s:%" PRIu64 "\n", disk->name, v_count);

			for (v_pos = 0; v_pos < v_count; ++v_pos) {
				char sub[PATH_MAX];
				uint64_t v_size;
				uint64_t v_mtime_sec;
				uint32_t v_mtime_nsec;

				ret = sgetbs(f, sub, sizeof(sub));
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' entry has an invalid subpath\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}
				if (!path_is_sub(sub)) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' entry has invalid subpath '%s'!\n", sub);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetb64(f, &v_size);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' entry has an invalid size\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/* ensure the unsigned serialized size fits in signed data_off_t before conversion. */
				if (v_size > INT64_MAX) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'd' entry has an out-of-range size %" PRIu64 "!\n", v_size);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetb64(f, &v_mtime_sec);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' entry has an invalid mtime_sec\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetb32(f, &v_mtime_nsec);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' entry has an invalid mtime_nsec\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/*
				 * STAT_NSEC_INVALID is encoded as 0.
				 * Values > 1000000000 can occur in content files written by SnapRAID <= 14.10
				 * on Windows for files with pre-1970 timestamps due to negative nanoseconds.
				 * Normalize them to STAT_NSEC_INVALID.
				 */
				if (v_mtime_nsec == 0 || v_mtime_nsec > 1000000000)
					v_mtime_nsec = STAT_NSEC_INVALID;
				else
					--v_mtime_nsec;

				/* allocate the file */
				struct snapraid_dealloc* dealloc = dealloc_alloc(state->block_size, sub, v_size, v_mtime_sec, v_mtime_nsec);

				log_tag("content_info:dealloc_entry:%s:%s:%" PRIu64 ":%" PRIu64 ":%u\n", disk->name, esc_tag(dealloc->sub), dealloc->size, dealloc->mtime_sec, dealloc->mtime_nsec);

				/* read all hashes with their generation */
				for (block_off_t k = 0; k < dealloc->blockmax; ++k) {
					struct snapraid_dealloc_block* block = &dealloc->block[k];

					c = sgetc(f);
					if (c == 'c') {
						block->prev = 0;
					} else if (c == 'p') {
						if (state->prevhash == HASH_UNDEFINED) {
							/* LCOV_EXCL_START */
							decoding_error(path, f);
							log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' previous hash without previous checksum\n");
							os_abort();
							/* LCOV_EXCL_STOP */
						}
						block->prev = 1;
					} else {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(ECONTENT, "Internal inconsistency: Field 'd' has invalid hash type '%c'!\n", (char)c);
						os_abort();
						/* LCOV_EXCL_STOP */
					}

					ret = sread(f, block->hash, BLOCK_HASH_SIZE);
					if (ret < 0) {
						/* LCOV_EXCL_START */
						decoding_error(path, f);
						log_fatal(EINTERNAL, "Internal inconsistency: Field 'd' error reading block hash\n");
						os_abort();
						/* LCOV_EXCL_STOP */
					}
				}

				/* insert the dealloc in the dealloc containers */
				tommy_list_insert_tail(&disk->dealloclist, &dealloc->nodelist, dealloc);
			}
		} else if (c == 's') {
			/* symlink */
			char sub[PATH_MAX];
			char linkto[PATH_MAX];
			struct snapraid_link* slink;
			struct snapraid_disk* disk;
			uint32_t mapping;

			ret = sgetb32(f, &mapping);
			if (ret < 0 || mapping >= mapping_max) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 's' mapping index out of range\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			disk = tommy_array_get(&disk_mapping, mapping);

			ret = sgetbs(f, sub, sizeof(sub));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 's' has an invalid subpath\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (!path_is_sub(sub)) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 's' has invalid subpath '%s'!\n", sub);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetbs(f, linkto, sizeof(linkto));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 's' has an invalid target path\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* allocate the link as symbolic link */
			slink = link_alloc(sub, linkto, FILE_IS_SYMLINK);

			/* insert the link in the link containers */
			tommy_hashdyn_insert(&disk->linkset, &slink->nodeset, slink, link_name_hash(slink->sub));
			tommy_list_insert_tail(&disk->linklist, &slink->nodelist, slink);

			/* stat */
			++count_symlink;
		} else if (c == 'a') {
			/* hardlink */
			char sub[PATH_MAX];
			char linkto[PATH_MAX];
			struct snapraid_link* slink;
			struct snapraid_disk* disk;
			uint32_t mapping;

			ret = sgetb32(f, &mapping);
			if (ret < 0 || mapping >= mapping_max) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'a' mapping index out of range\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			disk = tommy_array_get(&disk_mapping, mapping);

			ret = sgetbs(f, sub, sizeof(sub));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'a' has an invalid subpath\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (!path_is_sub(sub)) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'a' has invalid subpath '%s'!\n", sub);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetbs(f, linkto, sizeof(linkto));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'a' has an invalid target path\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (!path_is_sub(linkto)) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'a' has invalid target '%s' for '%s'!\n", linkto, sub);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* allocate the link as hard link */
			slink = link_alloc(sub, linkto, FILE_IS_HARDLINK);

			/* insert the link in the link containers */
			tommy_hashdyn_insert(&disk->linkset, &slink->nodeset, slink, link_name_hash(slink->sub));
			tommy_list_insert_tail(&disk->linklist, &slink->nodelist, slink);

			/* stat */
			++count_hardlink;
		} else if (c == 'r') {
			/* dir */
			char sub[PATH_MAX];
			struct snapraid_dir* dir;
			struct snapraid_disk* disk;
			uint32_t mapping;

			ret = sgetb32(f, &mapping);
			if (ret < 0 || mapping >= mapping_max) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'r' mapping index out of range\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			disk = tommy_array_get(&disk_mapping, mapping);

			ret = sgetbs(f, sub, sizeof(sub));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'r' has an invalid subpath\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (!path_is_sub(sub)) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'r' has invalid subpath '%s'!\n", sub);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* allocate the dir */
			dir = dir_alloc(sub);

			/* insert the dir in the dir containers */
			tommy_hashdyn_insert(&disk->dirset, &dir->nodeset, dir, dir_name_hash(dir->sub));
			tommy_list_insert_tail(&disk->dirlist, &dir->nodelist, dir);

			/* stat */
			++count_dir;
		} else if (c == 'c') {
			if (has_hash || mapping_max != 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'c' duplicate or misplaced the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			has_hash = 1;

			/* get the subcommand */
			c = sgetc(f);

			switch (c) {
			case 'u' :
				state->hash = HASH_MURMUR3;
				break;
			case 'k' :
				state->hash = HASH_SPOOKY2;
				break;
			default :
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'c' has invalid checksum type '%c'!\n", (char)c);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* read the seed */
			ret = sread(f, state->hashseed, HASH_MAX);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'c' error reading hash seed\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
		} else if (c == 'C') {
			if (has_prevhash || mapping_max != 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'C' duplicate or misplaced in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			has_prevhash = 1;

			/* get the sub-command */
			c = sgetc(f);

			switch (c) {
			case 'u' :
				state->prevhash = HASH_MURMUR3;
				break;
			case 'k' :
				state->prevhash = HASH_SPOOKY2;
				break;
			default :
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'C' has invalid checksum type '%c'!\n", (char)c);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* read the seed */
			ret = sread(f, state->prevhashseed, HASH_MAX);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'C' error reading hash seed\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
		} else if (c == 'z') {
			uint32_t block_size;

			if (has_block_size || mapping_max != 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'z' duplicate or misplaced in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			has_block_size = 1;

			ret = sgetb32(f, &block_size);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'z' has an invalid block size\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (block_size == 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'z' has zero 'blocksize' specification in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* without configuration, auto assign the block size */
			if (state->no_conf) {
				state->block_size = block_size;
			}

			if (block_size != state->block_size) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EUSER, "Mismatching 'blocksize' specification in the content file!\n");
				log_fatal(EUSER, "Please restore the 'blocksize' value in the configuration file to '%u'\n", block_size / KIBI);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		} else if (c == 'y') {
			uint64_t hash_size;

			if (has_hash_size || mapping_max != 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'y' duplicate or misplaced in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			has_hash_size = 1;

			ret = sgetb64(f, &hash_size);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'y' has an invalid hash size\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (hash_size == 2) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Unsupported 'hashsize 2' specification in the content file! The minimum supported hash size is 4.\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (hash_size < 4 || hash_size > HASH_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Invalid 'hashsize' specification %" PRIu64 " in the content file!\n", hash_size);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* without configuration, auto assign the hash size */
			if (state->no_conf) {
				BLOCK_HASH_SIZE = hash_size;
			}

			if (hash_size != BLOCK_HASH_SIZE) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EUSER, "Mismatching 'hashsize' specification in the content file!\n");
				log_fatal(EUSER, "Please restore the 'hashsize' value in the configuration file to '%" PRIu64 "'\n", hash_size);
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		} else if (c == 'x') {
			if (has_blockmax || mapping_max != 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'x' duplicate or misplaced in the content file!\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			has_blockmax = 1;

			ret = sgetb64(f, &blockmax);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'x' has an invalid blockmax\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}
			if (blockmax > BLOCK_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
#if SIZE_MAX == UINT32_MAX
				log_fatal(ESOFT, "Array with %" PRIu64 " blocks in '%s' is too large for a 32-bit build. Use a 64-bit build or increase the block size.\n", blockmax, path);
#else
				log_fatal(ESOFT, "Array with %" PRIu64 " blocks in '%s' is too large. Increase the block size.\n", blockmax, path);
#endif
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		} else if (c == 'M') {
			struct snapraid_map* map;
			char uuid[UUID_MAX];
			uint32_t v_idx;
			block_off_t v_total_blocks;
			block_off_t v_free_blocks;
			struct snapraid_disk* disk;

			if (mapping_max >= RAID_DATA_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Too many data disk mappings in the content file!\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			ret = sgetbs(f, buffer, sizeof(buffer));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'M' has an invalid disk name\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb32(f, &v_idx);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'M' has an invalid disk index\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (v_idx >= RAID_DATA_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'M' invalid disk position '%u' in the content file!\n", v_idx);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb64(f, &v_total_blocks);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'M' has an invalid total_blocks count\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb64(f, &v_free_blocks);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'M' has an invalid free_blocks count\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* read the uuid */
			ret = sgetbs(f, uuid, sizeof(uuid));
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'M' has an invalid uuid\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* find the disk */
			disk = find_disk_by_name(state, buffer);
			if (!disk) {
				/* search by UUID if renamed */
				disk = find_disk_by_uuid(state, uuid);
				if (disk) {
					log_info(EUSER, "WARNING! Renaming disk '%s' to '%s'\n", buffer, disk->name);

					/* write the new state with the new name */
					state->need_write = 1;
				}
			}
			if (!disk) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EUSER, "Disk '%s' with uuid '%s' not present in the configuration file!\n", buffer, uuid);
				log_fatal(EUSER, "If you have removed it from the configuration file, please restore it\n");
				/* if it's a command without UUID, it cannot autorename using UUID */
				if (state->opt.skip_disk_access)
					log_fatal(EUSER, "If you have renamed it, run 'sync' to update the new name\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			map = map_alloc(disk->name, v_idx, v_total_blocks, v_free_blocks, uuid);

			tommy_list_insert_tail(&state->maplist, &map->node, map);

			/* insert in the mapping vector */
			tommy_array_grow(&disk_mapping, mapping_max + 1);
			tommy_array_set(&disk_mapping, mapping_max, disk);
			++mapping_max;
		} else if (c == 'Q') {
			uint32_t v_level;
			block_off_t v_total_blocks;
			block_off_t v_free_blocks;
			uint32_t v_split_mac;
			uint64_t v_size_total = 0;
			unsigned s;

			ret = sgetb32(f, &v_level);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' has an invalid parity level\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb64(f, &v_total_blocks);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' has an invalid total_blocks count\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb64(f, &v_free_blocks);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' has an invalid free_blocks count\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			ret = sgetb32(f, &v_split_mac);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' has an invalid split count\n");
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (v_split_mac > SPLIT_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'Q' has too many splits '%u' in the content file!\n", v_split_mac);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			if (v_level >= LEV_MAX) {
				/* LCOV_EXCL_START */
				decoding_error(path, f);
				log_fatal(ECONTENT, "Internal inconsistency: Field 'Q' has invalid parity level '%u' in the content file!\n", v_level);
				os_abort();
				/* LCOV_EXCL_STOP */
			}

			/* auto configure if configuration is missing */
			if (state->no_conf) {
				if (v_level >= state->level)
					state->level = v_level + 1;
				if (state->parity[v_level].split_mac < v_split_mac)
					state->parity[v_level].split_mac = v_split_mac;
			}

			/* if we use this parity entry */
			if (v_level < state->level) {
				/* set the parity info */
				state->parity[v_level].total_blocks = v_total_blocks;
				state->parity[v_level].free_blocks = v_free_blocks;
			}

			for (s = 0; s < v_split_mac; ++s) {
				char v_path[PATH_MAX];
				char v_uuid[UUID_MAX];
				uint64_t v_size;

				ret = sgetbs(f, v_path, sizeof(v_path));
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' split has an invalid path\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetbs(f, v_uuid, sizeof(v_uuid));
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' split has an invalid uuid\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				ret = sgetb64(f, &v_size);
				if (ret < 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(EINTERNAL, "Internal inconsistency: Field 'Q' split has an invalid size\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/* ensure the unsigned serialized size fits in signed data_off_t before conversion. */
				if (v_size > INT64_MAX) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'Q' split has an out-of-range size %" PRIu64 "!\n", v_size);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				if (v_size > INT64_MAX - v_size_total) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'Q' total parity split size is too large!\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				v_size_total += v_size;

				if (!has_block_size) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'Q' missing 'blocksize' parity split information in the content file!\n");
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				if (v_size % state->block_size != 0) {
					/* LCOV_EXCL_START */
					decoding_error(path, f);
					log_fatal(ECONTENT, "Internal inconsistency: Field 'Q' parity split size %" PRIu64 " not aligned to block size %u!\n", v_size, state->block_size);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				/* if we use this parity entry */
				if (v_level < state->level) {
					/* if this split was removed from the configuration */
					if (s >= state->parity[v_level].split_mac) {
						/* if the file is used, we really need it */
						if (v_size != 0) {
							/* LCOV_EXCL_START */
							decoding_error(path, f);
							log_fatal(EUSER, "Parity '%s' misses used file '%u'!\n", lev_config_name(v_level), s);
							log_fatal(EUSER, "If you have removed it from the configuration file, please restore it\n");
							exit(EXIT_FAILURE);
							/* LCOV_EXCL_STOP */
						}

						/* otherwise we can drop it */
						log_info(EUSER, "WARNING! Dropping from '%s' unused split '%u'\n", lev_config_name(v_level), s);
					} else {
						/* we copy the path and uuid only if without configuration file */
						if (state->no_conf) {
							pathcpy(state->parity[v_level].split_map[s].path, sizeof(state->parity[v_level].split_map[s].path), v_path);
							pathcpy(state->parity[v_level].split_map[s].uuid, sizeof(state->parity[v_level].split_map[s].uuid), v_uuid);
						}

						/* set the split info */
						pathcpy(state->parity[v_level].split_map[s].content_uuid, sizeof(state->parity[v_level].split_map[s].content_uuid), v_uuid);
						state->parity[v_level].split_map[s].size = v_size;
					}
				}
			}
		} else if (c == 'N') {
			uint32_t crc_stored;
			uint32_t crc_computed;

			/* get the crc before reading it from the file */
			crc_computed = scrc(f);

			ret = sgetble32(f, &crc_stored);
			if (ret < 0) {
				/* LCOV_EXCL_START */
				/* here don't call decoding_error() because it's too late to get the crc */
				log_fatal(errno, "Error reading the CRC in '%s' at offset %" PRIi64 "\n", path, stell(f));
				log_fatal(errno, "The content file '%s' is damaged or corrupted!\n", path);
				log_fatal(errno, "To recover, rename or delete it and rerun the command.\n");
				log_fatal(errno, "SnapRAID will automatically fall back to the next healthy copy.\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			if (crc_stored != crc_computed) {
				/* LCOV_EXCL_START */
				/* here don't call decoding_error() because it's too late to get the crc */
				log_fatal(ECONTENT, "CRC mismatch in '%s'\n", path);
				log_fatal(ECONTENT, "The content file '%s' is damaged or corrupted (CRC mismatch)!\n", path);
				log_fatal(ECONTENT, "To recover, rename or delete it and rerun the command.\n");
				log_fatal(ECONTENT, "SnapRAID will automatically fall back to the next healthy copy.\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			crc_checked = 1;

			/* trailing bytes would lie outside the content boundary verified by the terminal CRC */
			c = sgetc(f);
			if (c != EOF) {
				/* LCOV_EXCL_START */
				log_fatal(ECONTENT, "Unexpected data after CRC in content file '%s' at offset %" PRIi64 "\n", path, stell(f));
				log_fatal(ECONTENT, "The content file '%s' is damaged or corrupted!\n", path);
				log_fatal(ECONTENT, "To recover, rename or delete it and rerun the command.\n");
				log_fatal(ECONTENT, "SnapRAID will automatically fall back to the next healthy copy.\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			break;
		} else {
			/* LCOV_EXCL_START */
			decoding_error(path, f);
			log_fatal(ECONTENT, "Internal inconsistency: Invalid field command '%c'!\n", (char)c);
			os_abort();
			/* LCOV_EXCL_STOP */
		}
	}

	tommy_array_done(&disk_mapping);

	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error reading the content file '%s' at offset %" PRIi64 "\n", path, stell(f));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (!crc_checked) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Reached the end of '%s' without finding the expected CRC\n", path);
		log_fatal(ECONTENT, "The content file '%s' is truncated or damaged!\n", path);
		log_fatal(ECONTENT, "To recover, rename or delete it and rerun the command.\n");
		log_fatal(ECONTENT, "SnapRAID will automatically fall back to the next healthy copy.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (!has_blockmax) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Missing 'blockmax' in the content file!\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (!has_block_size) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Missing 'blocksize' in the content file!\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (!has_hash_size) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Missing 'hashsize' in the content file!\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (!has_hash) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "Missing 'hash' in the content file!\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* check the file-system on all disks */
	state_fscheck(state, "after read");

	/* check that the stored parity size matches the loaded state */
	if (blockmax != parity_allocated_size(state)) {
		/* LCOV_EXCL_START */
		log_fatal(EINTERNAL, "Wrong parity size %" PRIu64 "/%" PRIu64 " in '%s' at offset %" PRIi64 "\n", blockmax, parity_allocated_size(state), path, stell(f));
		if (state->opt.skip_content_check) {
			log_fatal(ECONTENT, "Overriding.\n");
			blockmax = parity_allocated_size(state);
		} else {
			exit(EXIT_FAILURE);
		}
		/* LCOV_EXCL_STOP */
	}

	context->blockmax = blockmax;
	context->count_map = mapping_max;
	context->count_file = count_file;
	context->count_hardlink = count_hardlink;
	context->count_symlink = count_symlink;
	context->count_dir = count_dir;
	context->count_bad = count_bad;
	context->count_rehash = count_rehash;
	context->count_unsynced = count_unsynced;
	context->count_unscrubbed = count_unscrubbed;

	/* store the blocks counters */
	state->rehash_blocks = count_rehash;
	state->bad_blocks = count_bad;
	state->unsynced_blocks = count_unsynced;
	state->unscrubbed_blocks = count_unscrubbed;

	/* store the bucket info list */
	bucket_to_list(&bucket_hash, &state->bucketlist, &state->bucketcount);
	tommy_hashdyn_done(&bucket_hash);
}

struct state_write_context {
	struct snapraid_state* state;
#if HAVE_MT_WRITE
	thread_id_t thread;
#endif
	/* input */
	int content_version;
	block_off_t blockmax;
	time_t info_oldest;
	time_t info_now;
	int has_prevhash_data;
	uint64_t count_unsynced;
	STREAM* f;
	int first;
#if HAVE_MT_WRITE
	const char* content;
#endif

	/* output (required to postpone the output to the terminal after the latest write) */
	uint32_t crc;
	uint64_t count_file;
	uint64_t count_hardlink;
	uint64_t count_symlink;
	uint64_t count_dir;
	uint64_t count_bad;
	uint64_t count_rehash;
	uint64_t count_unscrubbed;
};

/**
 * Write all hashes in a contiguous run of file blocks.
 *
 * The source hashes are separated by the block state byte, so they cannot be
 * copied with a single memcpy(). Keep the stream pointer local while filling
 * each buffered group to avoid the per-block stream call and field updates.
 */
static int state_write_block_run(STREAM* f, struct snapraid_file* file, block_off_t file_pos, block_off_t count)
{
	unsigned char* block_ptr = (unsigned char*)file_block(file, file_pos);
	size_t block_stride = block_sizeof();

	while (count) {
		size_t available = (size_t)(f->end - f->pos);
		block_off_t cached_count = available / BLOCK_HASH_SIZE;
		unsigned char* output;

		if (cached_count > count)
			cached_count = count;

		if (cached_count == 0) {
			struct snapraid_block* block = (struct snapraid_block*)block_ptr;

			if (swrite(block->hash, BLOCK_HASH_SIZE, f) != 0)
				return -1;

			block_ptr += block_stride;
			--count;
			continue;
		}

		output = f->pos;
		for (block_off_t i = 0; i < cached_count; ++i) {
			struct snapraid_block* block = (struct snapraid_block*)block_ptr;

			hash_copy(output, block->hash);

			output += BLOCK_HASH_SIZE;
			block_ptr += block_stride;
		}

		f->pos = output;
		count -= cached_count;
	}

	return 0;
}

static void* state_write_thread(void* arg)
{
	struct state_write_context* context = arg;
	struct snapraid_state* state = context->state;
	block_off_t blockmax = context->blockmax;
	time_t info_oldest = context->info_oldest;
	time_t info_now = context->info_now;
	int has_prevhash_data = context->has_prevhash_data;
	STREAM* f = context->f;
	uint32_t crc;
	uint64_t t64;
	uint64_t count_file;
	uint64_t count_hardlink;
	uint64_t count_symlink;
	uint64_t count_dir;
	uint64_t count_bad;
	uint64_t count_rehash;
	uint64_t count_unscrubbed;
	tommy_node* i;
	block_off_t begin;
	unsigned l, s;
	tommy_hashdyn bucket_hash;
#if HAVE_MT_WRITE
	uint64_t start;

	start = os_tick_ms();
#endif

	count_file = 0;
	count_hardlink = 0;
	count_symlink = 0;
	count_dir = 0;
	count_bad = 0;
	count_rehash = 0;
	count_unscrubbed = 0;
	tommy_hashdyn_init(&bucket_hash);

	/* write header */
	if (context->content_version >= 4)
		swrite("SNAPCNT4\n\3\0\0", 12, f);
	else
		swrite("SNAPCNT3\n\3\0\0", 12, f);

	/* write block size and block max */
	sputc('z', f);
	sputb32(state->block_size, f);
	sputc('x', f);
	sputb64(blockmax, f);

	/* hash size */
	sputc('y', f);
	sputb64(BLOCK_HASH_SIZE, f);

	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
		goto bail;
		/* LCOV_EXCL_STOP */
	}

	sputc('c', f);
	if (state->hash == HASH_MURMUR3) {
		sputc('u', f);
	} else if (state->hash == HASH_SPOOKY2) {
		sputc('k', f);
	} else {
		/* LCOV_EXCL_START */
		log_fatal(EINTERNAL, "Unexpected hash when writing the content file '%s'.\n", serrorfile(f));
		goto bail;
		/* LCOV_EXCL_STOP */
	}
	swrite(state->hashseed, HASH_MAX, f);
	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
		goto bail;
		/* LCOV_EXCL_STOP */
	}

	/* previous hash is also needed by deallocated blocks after active rehash positions disappear */
	if (state->prevhash != HASH_UNDEFINED) {
		if (has_prevhash_data) {
			sputc('C', f);
			if (state->prevhash == HASH_MURMUR3) {
				sputc('u', f);
			} else if (state->prevhash == HASH_SPOOKY2) {
				sputc('k', f);
			} else {
				/* LCOV_EXCL_START */
				log_fatal(EINTERNAL, "Unexpected prevhash when writing the content file '%s'.\n", serrorfile(f));
				goto bail;
				/* LCOV_EXCL_STOP */
			}
			swrite(state->prevhashseed, HASH_MAX, f);
			if (serror(f)) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
				goto bail;
				/* LCOV_EXCL_STOP */
			}
		}
	}

	/* for each map */
	for (i = state->maplist; i != 0; i = i->next) {
		struct snapraid_map* map = i->data;
		struct snapraid_disk* disk;

		/* find the disk for this mapping */
		disk = find_disk_by_name(state, map->name);
		if (!disk) {
			/* LCOV_EXCL_START */
			log_fatal(EINTERNAL, "Internal inconsistency: Unmapped disk '%s'\n", map->name);
			os_abort();
			/* LCOV_EXCL_STOP */
		}

		/* save the mapping only if disk is mapped */
		if (disk->mapping_idx != -1) {
			const char* uuid;

			sputc('M', f);
			sputbs(map->name, f);
			sputb32(map->position, f);
			sputb64(map->total_blocks, f);
			sputb64(map->free_blocks, f);

			/*
			 * If the probed uuid is empty, but disk access was skipped (globally
			 * or individually), write the last known uuid. Otherwise, write the
			 * empty probed uuid.
			 */
			if (disk->uuid[0] == 0 && (state->opt.skip_disk_access || disk->skip_access)) {
				uuid = map->content_uuid;
			} else {
				uuid = disk->uuid;
			}
			sputbs(uuid, f);

			if (serror(f)) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
				goto bail;
				/* LCOV_EXCL_STOP */
			}
		}
	}

	/* for each parity */
	for (l = 0; l < state->level; ++l) {
		sputc('Q', f);
		sputb32(l, f);
		sputb64(state->parity[l].total_blocks, f);
		sputb64(state->parity[l].free_blocks, f);
		sputb32(state->parity[l].split_mac, f);
		for (s = 0; s < state->parity[l].split_mac; ++s) {
			const char* uuid;
			sputbs(state->parity[l].split_map[s].path, f);

			/*
			 * If the probed uuid is not empty, write it. Otherwise, if parity access was
			 * skipped (globally or individually), write the last known uuid read from the
			 * content file. Otherwise, write the empty probed uuid.
			 */
			if (state->parity[l].split_map[s].uuid[0] == 0
				&& (state->opt.skip_parity_access || state->parity[l].skip_access)) {
				uuid = state->parity[l].split_map[s].content_uuid;
			} else {
				uuid = state->parity[l].split_map[s].uuid;
			}
			sputbs(uuid, f);

			sputb64(state->parity[l].split_map[s].size, f);
		}
		if (serror(f)) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
			goto bail;
			/* LCOV_EXCL_STOP */
		}
	}

	/* for each disk */
	for (i = state->disklist; i != 0; i = i->next) {
		tommy_node* j;
		struct snapraid_disk* disk = i->data;

		/* if the disk is not mapped, skip it */
		if (disk->mapping_idx < 0)
			continue;

		/* for each file */
		for (j = disk->filelist; j != 0; j = j->next) {
			struct snapraid_file* file = j->data;
			size_t block_stride = block_sizeof();
			uint64_t size;
			uint64_t mtime_sec;
			int32_t mtime_nsec;
			uint64_t inode;

			size = file->size;
			mtime_sec = file->mtime_sec;
			mtime_nsec = file->mtime_nsec;
			inode = file->inode;

			sputc('f', f);
			sputb32(disk->mapping_idx, f);
			sputb64(size, f);
			sputb64(mtime_sec, f);
			/* encode STAT_NSEC_INVALID as 0 */
			if (mtime_nsec == STAT_NSEC_INVALID)
				sputb32(0, f);
			else
				sputb32(mtime_nsec + 1, f);
			sputb64(inode, f);
			sputbs(file->sub, f);
			if (serror(f)) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
				goto bail;
				/* LCOV_EXCL_STOP */
			}

			/* for all the blocks of the file */
			begin = 0;
			while (begin < file->blockmax) {
				struct snapraid_block* block = file_block(file, begin);
				unsigned char* end_block_ptr = (unsigned char*)block + block_stride;
				unsigned v_state = block_state_get(block);
				block_off_t extent_count;
				block_off_t v_pos = fs_file2par_get_run(disk, file, begin, &extent_count);
				block_off_t extent_end = begin + extent_count;
				block_off_t v_count;

				block_off_t end;

				/* find the end of run of blocks */
				end = begin + 1;
				while (end < file->blockmax) {
					block = (struct snapraid_block*)end_block_ptr;
					if (v_state != block_state_get(block))
						break;

					/* Resolve the mapping only when crossing an extent boundary. */
					if (end == extent_end) {
						block_off_t next_count;
						block_off_t next_pos = fs_file2par_get_run(disk, file, end, &next_count);

						if (v_pos + (end - begin) != next_pos)
							break;

						extent_end += next_count;
					}

					end_block_ptr += block_stride;
					++end;
				}

				switch (v_state) {
				case BLOCK_STATE_BLK :
					sputc('b', f);
					break;
				case BLOCK_STATE_REBUILD :
					sputc('r', f);
					break;
				case BLOCK_STATE_CHG :
					sputc('g', f);
					break;
				case BLOCK_STATE_REP :
					sputc('p', f);
					break;
				default :
					/* LCOV_EXCL_START */
					log_fatal(EINTERNAL, "Internal inconsistency: State for block %" PRIu64 " state %u\n", v_pos, v_state);
					os_abort();
					/* LCOV_EXCL_STOP */
				}

				sputb64(v_pos, f);

				v_count = end - begin;
				sputb64(v_count, f);

				/* write all hashes in the run */
				if (state_write_block_run(f, file, begin, v_count) != 0) {
					/* LCOV_EXCL_START */
					log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
					goto bail;
					/* LCOV_EXCL_STOP */
				}

				if (serror(f)) {
					/* LCOV_EXCL_START */
					log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
					goto bail;
					/* LCOV_EXCL_STOP */
				}

				/* next begin position */
				begin = end;
			}

			++count_file;
		}

		/* for each link */
		for (j = disk->linklist; j != 0; j = j->next) {
			struct snapraid_link* slink = j->data;

			switch (link_flag_get(slink, FILE_IS_LINK_MASK)) {
			case FILE_IS_HARDLINK :
				sputc('a', f);
				++count_hardlink;
				break;
			case FILE_IS_SYMLINK :
				sputc('s', f);
				++count_symlink;
				break;
			}

			sputb32(disk->mapping_idx, f);
			sputbs(slink->sub, f);
			sputbs(slink->linkto, f);
			if (serror(f)) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
				goto bail;
				/* LCOV_EXCL_STOP */
			}
		}

		/* for each dir */
		for (j = disk->dirlist; j != 0; j = j->next) {
			struct snapraid_dir* dir = j->data;

			sputc('r', f);
			sputb32(disk->mapping_idx, f);
			sputbs(dir->sub, f);
			if (serror(f)) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
				goto bail;
				/* LCOV_EXCL_STOP */
			}

			++count_dir;
		}

		/* deleted blocks of the disk */
		sputc('h', f);
		sputb32(disk->mapping_idx, f);
		if (serror(f)) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
			goto bail;
			/* LCOV_EXCL_STOP */
		}

		begin = 0;
		if (tommy_list_empty(&disk->deletedlist)) {
			/* the common case needs only one non-deleted run */
			if (blockmax) {
				sputb64(blockmax, f);
				sputc('O', f);
			}
		} else {
			while (begin < blockmax) {
				int is_deleted;
				block_off_t end;

				is_deleted = fs_is_block_deleted(disk, begin);

				/* find the end of run of blocks */
				end = begin + 1;
				while (end < blockmax
					&& is_deleted == fs_is_block_deleted(disk, end)
				) {
					++end;
				}

				sputb64(end - begin, f);

				if (is_deleted) {
					/* write the run of deleted blocks with hash */
					sputc('o', f);

					/* write all the hashes */
					while (begin < end) {
						struct snapraid_block* block = fs_par2block_get(disk, begin);

						swrite(block->hash, BLOCK_HASH_SIZE, f);

						++begin;
					}
				} else {
					/* used and empty blocks don't have a hash in this section */
					sputc('O', f);
					begin = end;
				}
			}
		}

		if (serror(f)) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
			goto bail;
			/* LCOV_EXCL_STOP */
		}

		/* deallocated files of the disk */
		if (!tommy_list_empty(&disk->dealloclist)) {
			sputc('d', f);
			sputb32(disk->mapping_idx, f);

			block_off_t v_count = tommy_list_count(&disk->dealloclist);

			sputb64(v_count, f);
			if (context->first)
				log_tag("content_info:dealloc:%s:%" PRIu64 "\n", disk->name, v_count);

			/* for each file */
			for (j = tommy_list_head(&disk->dealloclist); j != 0; j = j->next) {
				struct snapraid_dealloc* dealloc = j->data;
				sputbs(dealloc->sub, f);
				sputb64(dealloc->size, f);
				sputb64(dealloc->mtime_sec, f);

				/* encode STAT_NSEC_INVALID as 0 */
				if (dealloc->mtime_nsec == STAT_NSEC_INVALID)
					sputb32(0, f);
				else
					sputb32(dealloc->mtime_nsec + 1, f);

				if (context->first)
					log_tag("content_info:dealloc_entry:%s:%s:%" PRIu64 ":%" PRIu64 ":%u\n", disk->name, esc_tag(dealloc->sub), dealloc->size, dealloc->mtime_sec, dealloc->mtime_nsec);

				/* write all hashes with their generation */
				for (block_off_t k = 0; k < dealloc->blockmax; ++k) {
					struct snapraid_dealloc_block* block = &dealloc->block[k];

					sputc(block->prev ? 'p' : 'c', f);
					swrite(block->hash, BLOCK_HASH_SIZE, f);
				}
			}

			if (serror(f)) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
				goto bail;
				/* LCOV_EXCL_STOP */
			}
		} else {
			if (context->first)
				log_tag("content_info:dealloc:%s:0\n", disk->name);
		}
	}

	/* write the info for each block */
	sputc('i', f);
	/* ensure to write a 64 bit time */
	t64 = info_oldest;
	sputb64(t64, f);
	begin = 0;
	while (begin < blockmax) {
		snapraid_info info;
		block_off_t end;
		block_off_t count;
		time_t t;
		unsigned flag;

		/* find the complete run without resolving each array position */
		count = info_get_run(&state->infoarr, begin, blockmax - begin, &info);
		end = begin + count;
		sputb64(count, f);

		/* if there is info */
		if (info) {
			/* other flags */
			flag = 1; /* info is present */
			if (info_get_bad(info)) {
				flag |= 2;
				if (context->first)
					count_bad += count;
			}
			if (info_get_rehash(info)) {
				flag |= 4;
				if (context->first)
					count_rehash += count;
			}
			if (info_get_justsynced(info)) {
				flag |= 8;
				if (context->first)
					count_unscrubbed += count;
			}
			sputb32(flag, f);

			t = info_get_time(info);

			if (context->first)
				bucket_insert(&bucket_hash, t, count, flag & 8);

			/* truncate any time that is in the future */
			if (t > info_now)
				t = info_now;

			/* the oldest info is computed only on required blocks, so it may not be the absolute oldest */
			if (t < info_oldest)
				t = 0;
			else
				t -= info_oldest;

			/* ensure to write a 64 bit time */
			t64 = t;
			sputb64(t64, f);
		} else {
			/* write a special 0 flag to mark missing info */
			sputb32(0, f);
		}

		if (serror(f)) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
			goto bail;
			/* LCOV_EXCL_STOP */
		}

		/* next begin position */
		begin = end;
	}

	sputc('N', f);

	/* flush data written to the disk */
	if (sflush(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error writing the content file '%s' (in flush before crc). %s.\n", serrorfile(f), strerror(errno));
		goto bail;
		/* LCOV_EXCL_STOP */
	}

	/* get the file crc */
	crc = scrc(f);

	sputble32(crc, f);
	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error writing the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
		goto bail;
		/* LCOV_EXCL_STOP */
	}

	/* set output variables */
	context->crc = crc;
	context->count_file = count_file;
	context->count_hardlink = count_hardlink;
	context->count_symlink = count_symlink;
	context->count_dir = count_dir;
	context->count_bad = count_bad;
	context->count_rehash = count_rehash;
	context->count_unscrubbed = count_unscrubbed;

	/* store the bucket info list */
	if (context->first)
		bucket_to_list(&bucket_hash, &state->bucketlist, &state->bucketcount);

	/*
	 * Do not need to free the bucket entries in the success path because they
	 * are added only if context->first is true, and they are already moved
	 * to the list.
	 */

	tommy_hashdyn_done(&bucket_hash);
#if HAVE_MT_WRITE
	msg_progress("Saved %s in %" PRIu64 " ms\n", context->content, os_tick_ms() - start);
#endif
	return 0;

bail:
	tommy_hashdyn_foreach(&bucket_hash, bucket_free);
	tommy_hashdyn_done(&bucket_hash);
	return context;
}

struct state_write_info {
	uint32_t crc;
	block_off_t blockmax;
	block_off_t count_file;
	block_off_t count_hardlink;
	block_off_t count_symlink;
	block_off_t count_dir;
	block_off_t count_bad;
	block_off_t count_rehash;
	block_off_t count_unsynced;
	block_off_t count_unscrubbed;
};

static void state_write_content(struct snapraid_state* state, struct state_write_info* out_info)
{
#if HAVE_MT_WRITE
	int fail;
	int first;
#else
	STREAM* f;
	unsigned count_content;
	unsigned k;
	struct state_write_context* context;
	void* retval;
#endif
	tommy_node* i;
	block_off_t blockmax;
	time_t info_oldest;
	time_t info_now;
	time_t content_mtime;
	int info_has_rehash;
	int state_has_rebuild;
	int has_dealloc;
	int mapping_idx;
	block_off_t idx;
	uint32_t crc;
	block_off_t count_file;
	block_off_t count_hardlink;
	block_off_t count_symlink;
	block_off_t count_dir;
	block_off_t count_bad;
	block_off_t count_rehash;
	block_off_t count_unsynced;
	block_off_t count_unscrubbed;
#if !HAVE_MT_WRITE
	uint64_t start;
#endif

	/* blocks of all array */
	blockmax = parity_allocated_size(state);

	/* preparation is the only extent-tree user until serialization starts */
	fs_single_thread(state, 1);

	/* check the file-system on all disks */
	state_fscheck(state, "before write");

	/*
	 * Clear the info for unused blocks
	 * and get some other info
	 */
	info_oldest = 0; /* oldest time in info */
	info_now = time(0); /* get the present time */
	info_has_rehash = 0; /* if there is a rehash info */
	state_has_rebuild = 0;
	count_unsynced = 0;
	for (idx = 0; idx < blockmax; ++idx) {
		int is_unsynced;
		int has_deleted;
		int has_rebuild;

		/* if the position is used */
		if (fs_position_analyze(state, idx, &is_unsynced, &has_deleted, &has_rebuild)) {
			snapraid_info info = info_get(&state->infoarr, idx);

			if (has_rebuild)
				state_has_rebuild = 1;

			if (is_unsynced)
				++count_unsynced;

			/* only if there is some info to store */
			if (info) {
				time_t info_time = info_get_time(info);

				if (!info_oldest || info_time < info_oldest)
					info_oldest = info_time;

				if (info_get_rehash(info))
					info_has_rehash = 1;
			}
		} else {
			/* clear any previous info */
			info_set(&state->infoarr, idx, 0);

			/* and clear any deleted blocks, avoiding another scan in the common case */
			if (has_deleted)
				fs_position_clear_deleted(state, idx);
		}
	}

	if (info_has_rehash && state->prevhash == HASH_UNDEFINED) {
		/* LCOV_EXCL_START */
		log_fatal(EINTERNAL, "Internal inconsistency: Rehash blocks found but previous checksum is missing!\n");
		os_abort();
		/* LCOV_EXCL_STOP */
	}

	/* map disks */
	mapping_idx = 0;
	for (i = state->maplist; i != 0; i = i->next) {
		struct snapraid_map* map = i->data;
		struct snapraid_disk* disk;

		/* find the disk for this mapping */
		disk = find_disk_by_name(state, map->name);
		if (!disk) {
			/* LCOV_EXCL_START */
			log_fatal(EINTERNAL, "Internal inconsistency: Unmapped disk '%s'\n", map->name);
			os_abort();
			/* LCOV_EXCL_STOP */
		}

		/* save the mapping only for not empty disks */
		if (!fs_is_empty(disk, blockmax)) {
			/* assign the mapping index used to identify disks */
			disk->mapping_idx = mapping_idx;
			++mapping_idx;
		} else {
			/* mark the disk as without mapping */
			disk->mapping_idx = -1;
		}
	}

	int content_version = 3;
	has_dealloc = 0;

	/*
	 * Force at least version 3 as we want to always store the parity size.
	 * If there is a REBUILD block or a dealloc list, force version 4.
	 */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		if (!tommy_list_empty(&disk->dealloclist)) {
			has_dealloc = 1;
			break;
		}
	}
	if (state_has_rebuild || has_dealloc)
		content_version = 4;

	fs_single_thread(state, 0);

#if !HAVE_MT_WRITE
	start = os_tick_ms();
#endif

#if HAVE_MT_WRITE
	/* start all writing threads */
	first = 1;
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		struct state_write_context* context;
		char tmp[PATH_MAX];
		STREAM* f;

		msg_progress("Saving state to %s...\n", content->content);
		log_tag("content_write:%s\n", esc_tag(content->content));

		pathprint(tmp, sizeof(tmp), "%s.tmp", content->content);

		/* ensure to delete a previous stale file */
		if (remove(tmp) != 0) {
			if (errno != ENOENT) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error removing the stale content file '%s'. %s.\n", tmp, strerror(errno));
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}

		f = sopen_write(tmp, CONTENT_WRITE_ADVISE | STREAM_FLAGS_CRC);
		if (f == 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error opening the temporary content file '%s'. %s.\n", tmp, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		/* allocate the thread context */
		context = malloc_nofail(sizeof(struct state_write_context));
		content->context = context;

		/* initialize */
		context->state = state;
		context->content_version = content_version;
		context->blockmax = blockmax;
		context->info_oldest = info_oldest;
		context->info_now = info_now;
		context->has_prevhash_data = info_has_rehash || has_dealloc;
		context->count_unsynced = count_unsynced;
		context->f = f;
		context->first = first;
		context->content = content->content;
		first = 0;

		thread_create(&context->thread, state_write_thread, context);

		i = i->next;
	}

	/* join all threads before starting the durable finalization */
	fail = 0;
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		struct state_write_context* context = content->context;
		void* retval;

		thread_join(context->thread, &retval);

		if (retval) {
			/* LCOV_EXCL_START */
			fail = 1;
			/* LCOV_EXCL_STOP */
		}

		i = i->next;
	}

	/* abort on failure before finalizing any temporary content file */
	if (fail) {
		/* LCOV_EXCL_START */
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/*
	 * All temporary content files of this publication use the same modification time.
	 * SnapRAID guarantees that two state_write() publications cannot occur in the
	 * same second due to synchronization delays and the overall execution model.
	 */
	content_mtime = time(0);

	first = 1;
	crc = 0;
	count_file = 0;
	count_hardlink = 0;
	count_symlink = 0;
	count_dir = 0;
	count_bad = 0;
	count_rehash = 0;
	count_unscrubbed = 0;

	/* make all temporary content files durable only after all writers have terminated */
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		struct state_write_context* context = content->context;
		STREAM* f = context->f;

		/*
		 * Use the sequence fflush() -> fsync() -> fclose() -> rename() to ensure
		 * than even in a system crash event we have one valid copy of the file.
		 */
		if (sflush(f) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error writing the content file '%s', in flush(). %s.\n", serrorfile(f), strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		if (smtime(f, content_mtime, 0) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error setting the modification time of the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

#if HAVE_FSYNC
		if (ssync(f) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error writing the content file '%s' in sync(). %s.\n", serrorfile(f), strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
#endif

		if (sclose(f) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error closing the content file. %s.\n", strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		if (first) {
			first = 0;
			crc = context->crc;
			count_file = context->count_file;
			count_hardlink = context->count_hardlink;
			count_symlink = context->count_symlink;
			count_dir = context->count_dir;
			count_bad = context->count_bad;
			count_rehash = context->count_rehash;
			count_unsynced = context->count_unsynced;
			count_unscrubbed = context->count_unscrubbed;
		} else {
			if (crc != context->crc) {
				/* LCOV_EXCL_START */
				log_fatal(ECONTENT, "Different CRCs writing content streams.\n");
				log_fatal(ECONTENT, "DANGER! Your RAM memory is broken! DO NOT PROCEED UNTIL FIXED!\n");
				log_fatal(ECONTENT, "Try running a memory test like http://www.memtest86.com/\n");
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}

		free(context);

		i = i->next;
	}
#else
	/* count the content files */
	count_content = 0;
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		msg_progress("Saving state to %s...\n", content->content);
		++count_content;
		i = i->next;
	}

	/* open all the content files */
	f = sopen_multi_write(count_content, CONTENT_WRITE_ADVISE | STREAM_FLAGS_CRC);
	if (!f) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error opening the content files.\n");
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	k = 0;
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		char tmp[PATH_MAX];
		pathprint(tmp, sizeof(tmp), "%s.tmp", content->content);

		/* ensure to delete a previous stale file */
		if (remove(tmp) != 0) {
			if (errno != ENOENT) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error removing the stale content file '%s'. %s.\n", tmp, strerror(errno));
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}

		if (sopen_multi_file(f, k, tmp) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error opening the temporary content file '%s'. %s.\n", tmp, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		++k;
		i = i->next;
	}

	/* allocate the thread context */
	context = malloc_nofail(sizeof(struct state_write_context));

	/* initialize */
	context->state = state;
	context->content_version = content_version;
	context->blockmax = blockmax;
	context->info_oldest = info_oldest;
	context->info_now = info_now;
	context->has_prevhash_data = info_has_rehash || has_dealloc;
	context->count_unsynced = count_unsynced;
	context->f = f;
	context->first = 1;

	/* this serializer is the only extent-tree user until it completes */
	fs_single_thread(state, 1);
	retval = state_write_thread(context);
	fs_single_thread(state, 0);

	/* abort on failure */
	if (retval) {
		/* LCOV_EXCL_START */
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/*
	 * All temporary content files of this publication use the same modification time.
	 * SnapRAID guarantees that two state_write() publications cannot occur in the
	 * same second due to synchronization delays and the overall execution model.
	 */
	content_mtime = time(0);

	/*
	 * Use the sequence fflush() -> fsync() -> fclose() -> rename() to ensure
	 * than even in a system crash event we have one valid copy of the file.
	 */
	if (sflush(f) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error writing the content file '%s', in flush(). %s.\n", serrorfile(f), strerror(errno));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	if (smtime(f, content_mtime, 0) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error setting the modification time of the content file '%s'. %s.\n", serrorfile(f), strerror(errno));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

#if HAVE_FSYNC
	if (ssync(f) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error writing the content file '%s' in sync(). %s.\n", serrorfile(f), strerror(errno));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}
#endif

	if (sclose(f) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error closing the content file. %s.\n", strerror(errno));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	msg_progress("Saved state in %" PRIu64 " seconds\n", (os_tick_ms() - start) / 1000);

	crc = context->crc;
	count_file = context->count_file;
	count_hardlink = context->count_hardlink;
	count_symlink = context->count_symlink;
	count_dir = context->count_dir;
	count_bad = context->count_bad;
	count_rehash = context->count_rehash;
	count_unsynced = context->count_unsynced;
	count_unscrubbed = context->count_unscrubbed;

	free(context);
#endif

	out_info->crc = crc;
	out_info->blockmax = blockmax;
	out_info->count_file = count_file;
	out_info->count_hardlink = count_hardlink;
	out_info->count_symlink = count_symlink;
	out_info->count_dir = count_dir;
	out_info->count_bad = count_bad;
	out_info->count_rehash = count_rehash;
	out_info->count_unsynced = count_unsynced;
	out_info->count_unscrubbed = count_unscrubbed;
}

void state_probe(struct snapraid_state* state)
{
	char path[PATH_MAX];
	struct stat st;
	tommy_node* node;

	/* iterate over all the available content files and check the first one present */
	node = tommy_list_head(&state->contentlist);
	while (node) {
		struct snapraid_content* content = node->data;

		pathcpy(path, sizeof(path), content->content);

		if (stat(path, &st) == 0) {
			/* if found stop the search */
			break;
		}

		/* next content file */
		node = node->next;
	}

	/* if not found, assume empty */
	if (!node)
		return;

	if (!state->no_conf) {
		/* intentionally don't output the content path because it's not really read */
		log_tag("content_info:probe_unixtime:%" PRId64 "\n", (int64_t)st.st_mtime);
		log_flush();
	}
}

void state_read(struct snapraid_state* state)
{
	STREAM* f;
	char path[PATH_MAX];
	struct stat st;
	tommy_node* node;
	int ret;
	uint64_t start;
	struct state_read_context context;
	uint32_t map_idx;
	unsigned l, s;

	start = os_tick_ms();

	/*
	 * Iterate over all the available content files and load the first one present.
	 *
	 * Only missing files (ENOENT) trigger automatic failover to the next replica.
	 * If a file exists but is corrupted or damaged, reading will fail-fast with
	 * exit(EXIT_FAILURE) to alert the user to possible hardware or filesystem corruption.
	 * Once the user renames or removes the broken copy, the next run will fall back
	 * to the healthy replica and automatically re-create the missing one.
	 */
	f = 0;
	node = tommy_list_head(&state->contentlist);
	while (node) {
		struct snapraid_content* content = node->data;
		pathcpy(path, sizeof(path), content->content);

		msg_progress("Loading state from %s...\n", path);

		f = sopen_read(path, ADVISE_SEQUENTIAL | STREAM_FLAGS_CRC);
		if (f != 0) {
			/* if found stop the search */
			break;
		} else {
			/* if it's real error of an existing file, abort */
			if (errno != ENOENT) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error opening the content file '%s'. %s.\n", path, strerror(errno));
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* otherwise continue */
			if (node->next) {
				log_error(errno, "WARNING! Content file '%s' not found, attempting with another copy...\n", path);

				/* ensure to rewrite all the content files */
				state->need_write = 1;
			}
		}

		/* next content file */
		node = node->next;
	}

	/* if not found, assume empty */
	if (!f) {
		log_info(EUSER, "No content file found. Assuming empty.\n");

		/* create the initial mapping */
		state_map(state);
		return;
	}

	/* get the stat of the content file */
	ret = fstat(shandle(f), &st);
	if (ret != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error stating the content file '%s'. %s.\n", path, strerror(errno));
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}

	/* go further to check other content files */
	while (node) {
		char other_path[PATH_MAX];
		struct stat other_st;
		struct snapraid_content* content = node->data;
		pathcpy(other_path, sizeof(other_path), content->content);

		ret = stat(other_path, &other_st);
		if (ret != 0) {
			/* allow missing content files, but not any other kind of error */
			if (errno != ENOENT) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error stating the content file '%s'. %s.\n", other_path, strerror(errno));
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}

			/* ensure to rewrite all the content files */
			state->need_write = 1;
		} else {
			/* different size or modification time means that the replicas are not from the same publication */
			if (other_st.st_size != st.st_size || other_st.st_mtime != st.st_mtime) {
				log_error(ECONTENT, "WARNING! Content files '%s' and '%s' have a different size or modification time!\n", path, other_path);
				log_error(ECONTENT, "Likely the content replicas are from different publications or one of them is broken!\n");

				/* ensure to rewrite all the content files */
				state->need_write = 1;
			}
		}

		/* next content file */
		node = node->next;
	}

	/* previous hash, start with an undefined value */
	state->prevhash = HASH_UNDEFINED;

	/* intentionally not set the prevhashseed, if used valgrind will warn about it */

	/* parsing is the only user of the extent trees until the content is loaded */
	fs_single_thread(state, 1);
	state_read_content(state, path, f, &context);
	fs_single_thread(state, 0);

	sclose(f);

	msg_progress("Loaded state in %" PRIu64 " seconds\n", (os_tick_ms() - start) / 1000);

	if (state->unsynced_blocks)
		msg_progress("WARNING! The latest sync was interrupted!\n");

	state_content_check(state, path);

	/* update the mapping */
	state_map(state);

	/* for each data disk mapping read from the content file */
	map_idx = 0;
	for (node = tommy_list_head(&state->maplist); node != 0 && map_idx < context.count_map; node = node->next, ++map_idx) {
		struct snapraid_map* map = node->data;

		log_tag("content_data:%s:%" PRIi64 ":%" PRIi64 "\n",
			esc_tag(map->name),
			map->total_blocks * (uint64_t)state->block_size,
			map->free_blocks * (uint64_t)state->block_size);
		log_tag("content_data_split:%s:%s\n",
			esc_tag(map->name),
			map->content_uuid);
	}

	/* for each parity */
	for (l = 0; l < state->level; ++l) {
		log_tag("content_parity:%s:%" PRIi64 ":%" PRIi64 "\n",
			lev_config_name(l),
			state->parity[l].total_blocks * (uint64_t)state->block_size,
			state->parity[l].free_blocks * (uint64_t)state->block_size);

		for (s = 0; s < state->parity[l].split_mac; ++s) {
			char parity_name[64];

			if (s == 0)
				pathcpy(parity_name, sizeof(parity_name), lev_config_name(l));
			else
				pathprint(parity_name, sizeof(parity_name), "%s/%u", lev_config_name(l), s);

			log_tag("content_parity_split:%s:%s:%s:%" PRIi64 "\n",
				parity_name,
				state->parity[l].split_map[s].content_uuid,
				esc_tag(state->parity[l].split_map[s].path),
				state->parity[l].split_map[s].size);
		}
	}

	msg_verbose("%8" PRIu64 " files\n", context.count_file);
	msg_verbose("%8" PRIu64 " hardlinks\n", context.count_hardlink);
	msg_verbose("%8" PRIu64 " symlinks\n", context.count_symlink);
	msg_verbose("%8" PRIu64 " empty dirs\n", context.count_dir);

	log_tag("content_info:file:%" PRIu64 "\n", context.count_file);
	log_tag("content_info:hardlink:%" PRIu64 "\n", context.count_hardlink);
	log_tag("content_info:symlink:%" PRIu64 "\n", context.count_symlink);
	log_tag("content_info:dir_empty:%" PRIu64 "\n", context.count_dir);

	log_tag("content_info:block:%" PRIu64 "\n", context.blockmax);
	log_tag("content_info:block_bad:%" PRIu64 "\n", context.count_bad);
	log_tag("content_info:block_rehash:%" PRIu64 "\n", context.count_rehash);
	log_tag("content_info:block_unsynced:%" PRIu64 "\n", context.count_unsynced);
	log_tag("content_info:block_unscrubbed:%" PRIu64 "\n", context.count_unscrubbed);

	if (!state->no_conf) {
		log_tag("content:%s\n", esc_tag(path));
		log_tag("content_info:read_unixtime:%" PRId64 "\n", (int64_t)st.st_mtime);
	}
	log_flush();

	/* mark that we read the content file, and it passed all the checks */
	state->checked_read = 1;
}

struct state_verify_thread_context {
	struct snapraid_state* state;
	struct snapraid_content* content;
#if HAVE_MT_VERIFY
	thread_id_t thread;
#else
	void* retval;
#endif
	/* input */
	uint32_t crc;
	STREAM* f;
};

static void* state_verify_thread(void* arg)
{
	struct state_verify_thread_context* context = arg;
	struct snapraid_content* content = context->content;
	STREAM* f = context->f;
	unsigned char buf[4];
	uint32_t crc_stored;
	uint32_t crc_computed;
	uint64_t start;

	start = os_tick_ms();

	if (sdeplete(f, buf) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Failed to flush content file '%s'. %s.\n", serrorfile(f), strerror(errno));
		return context;
		/* LCOV_EXCL_STOP */
	}

	/* get the stored crc from the last four bytes */
	crc_stored = buf[0] | (uint32_t)buf[1] << 8 | (uint32_t)buf[2] << 16 | (uint32_t)buf[3] << 24;

	if (crc_stored != context->crc) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "DANGER! Wrong stored CRC in '%s'\n", serrorfile(f));
		return context;
		/* LCOV_EXCL_STOP */
	}

	/* get the computed crc */
	crc_computed = scrc(f);

	/* adjust the stored crc to include itself */
	crc_stored = crc32c(crc_stored, buf, 4);

	if (crc_computed != crc_stored) {
		/* LCOV_EXCL_START */
		log_fatal(ECONTENT, "DANGER! Wrong file CRC in '%s'\n", serrorfile(f));
		return context;
		/* LCOV_EXCL_STOP */
	}

	msg_progress("Verified %s in %" PRIu64 " seconds\n", content->content, (os_tick_ms() - start) / 1000);

	return 0;
}

static void state_verify_content(struct snapraid_state* state, uint32_t crc)
{
	tommy_node* i;
	int fail;

	msg_progress("Verifying...\n");

	/* start all reading threads */
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		struct state_verify_thread_context* context;
		char tmp[PATH_MAX];
		STREAM* f;

		pathprint(tmp, sizeof(tmp), "%s.tmp", content->content);
		f = sopen_read(tmp, ADVISE_SEQUENTIAL | STREAM_FLAGS_CRC);
		if (f == 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error reopening the temporary content file '%s'. %s.\n", tmp, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		/* allocate the thread context */
		context = malloc_nofail(sizeof(struct state_verify_thread_context));
		content->context = context;

		/* initialize */
		context->state = state;
		context->content = content;
		context->crc = crc;
		context->f = f;

#if HAVE_MT_VERIFY
		thread_create(&context->thread, state_verify_thread, context);
#else
		context->retval = state_verify_thread(context);
#endif

		i = i->next;
	}

	/* join all thread */
	fail = 0;
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		struct state_verify_thread_context* context = content->context;
		void* retval;

#if HAVE_MT_VERIFY
		thread_join(context->thread, &retval);
#else
		retval = context->retval;
#endif
		if (retval) {
			/* LCOV_EXCL_START */
			fail = 1;
			/* LCOV_EXCL_STOP */
		} else {
			STREAM* f = context->f;

			if (sclose(f) != 0) {
				/* LCOV_EXCL_START */
				log_fatal(errno, "Error closing the content file. %s.\n", strerror(errno));
				exit(EXIT_FAILURE);
				/* LCOV_EXCL_STOP */
			}
		}

		free(context);

		i = i->next;
	}

	/* abort on failure */
	if (fail) {
		/* LCOV_EXCL_START */
		exit(EXIT_FAILURE);
		/* LCOV_EXCL_STOP */
	}
}

static void state_rename_content(struct snapraid_state* state)
{
	tommy_node* i;

#ifndef _WIN32
	/*
	 * This sequence relies on POSIX file-system semantics.
	 *
	 * The temporary content file has already been flushed and fsync'ed in
	 * ssync() (using F_FULLFSYNC on macOS to force a physical media flush).
	 * rename()/renameat() atomically replaces the old content file, but the
	 * directory entry update is not necessarily durable when the rename
	 * returns.
	 *
	 * Open and fsync the parent directory after the rename to ensure that the
	 * new content file name is persisted before state_write() returns. This
	 * ordering is required for crash consistency because parity modifications
	 * may start only after the pre-sync content state has been durably
	 * published.
	 *
	 * On macOS, standard fsync() only pushes data from kernel buffers to the
	 * drive's volatile cache without requesting a physical media flush. F_FULLFSYNC
	 * forces the drive controller to commit all cached data to permanent storage.
	 * Because F_FULLFSYNC is not supported on directory descriptors, the subsequent
	 * directory fsync() only flushes to the drive cache and is not guaranteed to be
	 * instantly persistent across power loss. However, major database engines such
	 * as SQLite and PostgreSQL follow this exact same pattern on Darwin, making it
	 * the standard industry practice.
	 *
	 * See:
	 * https://github.com/sqlite/sqlite/blob/f3b9f74d81132426dee1ccc07a67fdad2ccfeaa9/src/os_unix.c#L3928-L3953
	 * https://github.com/postgres/postgres/blob/74c77052bc20017076100f0649ca0f44b1b98c29/src/backend/storage/file/fd.c#L835-L852
	 *
	 * When renameat() is available, perform the rename relative to the same
	 * directory descriptor that is subsequently fsync'ed. Use O_DIRECTORY
	 * when available, while retaining compatibility with POSIX systems where
	 * O_DIRECTORY is not defined.
	 */
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		char tmp[PATH_MAX];
		char dir[PATH_MAX];
		char* slash;
		int handle;

		pathcpy(dir, sizeof(dir), content->content);

		slash = strrchr(dir, '/');
		if (slash)
			slash[1] = 0;
		else
			pathcpy(dir, sizeof(dir), ".");

		int flags = O_RDONLY;
#ifdef O_DIRECTORY
		flags |= O_DIRECTORY;
#endif

		/* open the directory to get the handle */
		handle = open(dir, flags);
		if (handle < 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error opening the directory '%s'. %s.\n", dir, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		/* now rename the just written copy with the correct name */
		pathprint(tmp, sizeof(tmp), "%s.tmp", content->content);
#if HAVE_RENAMEAT
		const char* content_name = slash ? content->content + (slash - dir) + 1 : content->content;
		char tmp_name[PATH_MAX];

		pathprint(tmp_name, sizeof(tmp_name), "%s.tmp", content_name);
		if (renameat(handle, tmp_name, handle, content_name) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error renaming the content file '%s' to '%s'. %s.\n", tmp, content->content, strerror(errno));
			close(handle);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
#else
		if (rename(tmp, content->content) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error renaming the content file '%s' to '%s'. %s.\n", tmp, content->content, strerror(errno));
			close(handle);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}
#endif

		/* sync the directory */
		if (fsync(handle) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error syncing the directory '%s'. %s.\n", dir, strerror(errno));
			close(handle);
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		if (close(handle) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error closing the directory '%s'. %s.\n", dir, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		i = i->next;
	}
#else
	i = tommy_list_head(&state->contentlist);
	while (i) {
		struct snapraid_content* content = i->data;
		char tmp[PATH_MAX];

		/* now renames the just written copy with the correct name */
		pathprint(tmp, sizeof(tmp), "%s.tmp", content->content);
		if (rename(tmp, content->content) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(errno, "Error renaming the content file '%s' to '%s' in rename(). %s.\n", tmp, content->content, strerror(errno));
			exit(EXIT_FAILURE);
			/* LCOV_EXCL_STOP */
		}

		i = i->next;
	}
#endif
}

/*
 * Check if a modification time nanosecond value can be represented in the
 * content format.
 *
 * STAT_NSEC_INVALID is encoded as 0. Valid nanoseconds in the range
 * 0..999999999 are encoded as 1..1000000000.
 */
static int state_write_mtime_nsec_valid(int mtime_nsec)
{
	return mtime_nsec == STAT_NSEC_INVALID || (mtime_nsec >= 0 && mtime_nsec < 1000000000);
}

/*
 * Validate invariants required by the persistent content format.
 *
 * state_write() must never serialize a state that state_read() would reject.
 * In particular, some internal values use signed types or sentinel values but
 * are stored in the content file using unsigned encodings. If such values are
 * outside their valid domain, serializing them may produce a syntactically
 * valid content file that cannot be read back.
 *
 * Keep this check limited to serialization invariants. More complex consistency
 * checks on the filesystem and block mappings are performed separately by
 * state_fscheck().
 */
static void state_write_check(struct snapraid_state* state)
{
	tommy_node* i;
	unsigned l;
	unsigned s;
	block_off_t blockmax;

	if (state->block_size == 0) {
		log_fatal(EINTERNAL, "Internal inconsistency: Zero block size before state write!\n");
		os_abort();
	}

	if (state->level == 0 || state->level > LEV_MAX) {
		log_fatal(EINTERNAL, "Internal inconsistency: Invalid parity level count %u!\n", state->level);
		os_abort();
	}

	blockmax = parity_allocated_size(state);
	if (blockmax > BLOCK_MAX) {
		log_fatal(EINTERNAL, "Internal inconsistency: Parity size %" PRIu64 " is too large!\n", blockmax);
		os_abort();
	}

	for (l = 0; l < state->level; ++l) {
		data_off_t size_total = 0;

		if (state->parity[l].split_mac == 0 || state->parity[l].split_mac > SPLIT_MAX) {
			log_fatal(EINTERNAL, "Internal inconsistency: Invalid parity split count %u!\n", state->parity[l].split_mac);
			os_abort();
		}

		for (s = 0; s < state->parity[l].split_mac; ++s) {
			if (state->parity[l].split_map[s].size < 0) {
				log_fatal(EINTERNAL, "Internal inconsistency: Invalid parity split size %" PRIi64 "!\n", state->parity[l].split_map[s].size);
				os_abort();
			}

			/* keep the total within the bound enforced when reading the content file */
			if (state->parity[l].split_map[s].size > INT64_MAX - size_total) {
				log_fatal(EINTERNAL, "Internal inconsistency: Total parity split size is too large!\n");
				os_abort();
			}
			size_total += state->parity[l].split_map[s].size;
		}
	}

	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		tommy_node* j;

		for (j = disk->filelist; j != 0; j = j->next) {
			struct snapraid_file* file = j->data;

			if (file->size < 0) {
				log_fatal(EINTERNAL, "Internal inconsistency: Invalid file size %" PRIi64 "!\n", file->size);
				os_abort();
			}

			if (!state_write_mtime_nsec_valid(file->mtime_nsec)) {
				log_fatal(EINTERNAL, "Internal inconsistency: Invalid file mtime nanoseconds %d!\n", file->mtime_nsec);
				os_abort();
			}
		}

		for (j = disk->dealloclist; j != 0; j = j->next) {
			struct snapraid_dealloc* dealloc = j->data;

			if (dealloc->size < 0) {
				log_fatal(EINTERNAL, "Internal inconsistency: Invalid deallocated file size %" PRIi64 "!\n", dealloc->size);
				os_abort();
			}

			if (!state_write_mtime_nsec_valid(dealloc->mtime_nsec)) {
				log_fatal(EINTERNAL, "Internal inconsistency: Invalid deallocated file mtime nanoseconds %d!\n", dealloc->mtime_nsec);
				os_abort();
			}
		}
	}
}

/*
 * Emit structured mapping tags for data disks and parity splits after
 * content files have been durably published.
 */
static void state_write_log_mappings(struct snapraid_state* state)
{
	tommy_node* i;
	unsigned l, s;

	/* for each data disk mapping */
	for (i = state->maplist; i != 0; i = i->next) {
		struct snapraid_map* map = i->data;
		struct snapraid_disk* disk;
		const char* uuid;

		/* find the disk for this mapping */
		disk = find_disk_by_name(state, map->name);
		if (!disk || disk->mapping_idx == -1)
			continue;

		/*
		 * If the probed uuid is empty, but disk access was skipped (globally
		 * or individually), log the last known uuid. Otherwise, log the
		 * probed uuid.
		 */
		if (disk->uuid[0] == 0 && (state->opt.skip_disk_access || disk->skip_access)) {
			uuid = map->content_uuid;
		} else {
			uuid = disk->uuid;
		}

		log_tag("content_data:%s:%" PRIi64 ":%" PRIi64 "\n",
			esc_tag(map->name),
			map->total_blocks * (uint64_t)state->block_size,
			map->free_blocks * (uint64_t)state->block_size);
		log_tag("content_data_split:%s:%s\n",
			esc_tag(map->name),
			uuid);
	}

	/* for each parity */
	for (l = 0; l < state->level; ++l) {
		log_tag("content_parity:%s:%" PRIi64 ":%" PRIi64 "\n",
			lev_config_name(l),
			state->parity[l].total_blocks * (uint64_t)state->block_size,
			state->parity[l].free_blocks * (uint64_t)state->block_size);

		for (s = 0; s < state->parity[l].split_mac; ++s) {
			char parity_name[64];
			const char* uuid;

			/*
			 * If the probed uuid is not empty, log it. Otherwise, if parity access was
			 * skipped (globally or individually), log the last known uuid read from the
			 * content file. Otherwise, log the empty probed uuid.
			 */
			if (state->parity[l].split_map[s].uuid[0] == 0
				&& (state->opt.skip_parity_access || state->parity[l].skip_access)) {
				uuid = state->parity[l].split_map[s].content_uuid;
			} else {
				uuid = state->parity[l].split_map[s].uuid;
			}

			if (s == 0)
				pathcpy(parity_name, sizeof(parity_name), lev_config_name(l));
			else
				pathprint(parity_name, sizeof(parity_name), "%s/%u", lev_config_name(l), s);

			log_tag("content_parity_split:%s:%s:%s:%" PRIi64 "\n",
				parity_name,
				uuid,
				esc_tag(state->parity[l].split_map[s].path),
				state->parity[l].split_map[s].size);
		}
	}
}

void state_write(struct snapraid_state* state)
{
	struct state_write_info info;
	time_t now;

	/* verify that all persistent values are serializable */
	state_write_check(state);

	/* write all the content files */
	state_write_content(state, &info);

	/* verify the just written files */
	state_verify_content(state, info.crc);

	/* rename the new files, over the old ones */
	state_rename_content(state);

	/*
	 * Print verbose information, structured tags, and update block counters
	 * only after state_rename_content() succeeds, ensuring that the new content
	 * file is durable on disk before reporting or committing it.
	 */
	msg_verbose("%8" PRIu64 " files\n", info.count_file);
	msg_verbose("%8" PRIu64 " hardlinks\n", info.count_hardlink);
	msg_verbose("%8" PRIu64 " symlinks\n", info.count_symlink);
	msg_verbose("%8" PRIu64 " empty dirs\n", info.count_dir);

	state_write_log_mappings(state);

	log_tag("content_info:file:%" PRIu64 "\n", info.count_file);
	log_tag("content_info:hardlink:%" PRIu64 "\n", info.count_hardlink);
	log_tag("content_info:symlink:%" PRIu64 "\n", info.count_symlink);
	log_tag("content_info:dir_empty:%" PRIu64 "\n", info.count_dir);

	log_tag("content_info:block:%" PRIu64 "\n", info.blockmax);
	log_tag("content_info:block_bad:%" PRIu64 "\n", info.count_bad);
	log_tag("content_info:block_rehash:%" PRIu64 "\n", info.count_rehash);
	log_tag("content_info:block_unsynced:%" PRIu64 "\n", info.count_unsynced);
	log_tag("content_info:block_unscrubbed:%" PRIu64 "\n", info.count_unscrubbed);

	/* store the blocks counters */
	state->rehash_blocks = info.count_rehash;
	state->bad_blocks = info.count_bad;
	state->unsynced_blocks = info.count_unsynced;
	state->unscrubbed_blocks = info.count_unscrubbed;

	/* log the write time of the content file */
	now = time(0);
	log_tag("content_info:write_unixtime:%" PRId64 "\n", (int64_t)now);
	log_flush();

	state->need_write = 0; /* no write needed anymore */
	state->checked_read = 0; /* what we wrote is not checked in read */
	state->written = 1;
}

void state_commit(struct snapraid_state* state)
{
	tommy_node* i;

	/* for each disk */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		if (!tommy_list_empty(&disk->dealloclist)) {
			/* clear the dealloc list */
			tommy_list_foreach(&disk->dealloclist, dealloc_free);
			tommy_list_init(&disk->dealloclist);

			/* mark the state as needing write */
			state->need_write = 1;
		}
	}
}

void state_skip(struct snapraid_state* state)
{
	tommy_node* i;

	/* for each disk */
	for (i = state->disklist; i != 0; i = i->next) {
		tommy_node* j;
		struct snapraid_disk* disk = i->data;

		if (!disk->skip_access)
			continue;

		/* for each file */
		for (j = tommy_list_head(&disk->filelist); j != 0; j = j->next) {
			struct snapraid_file* file = j->data;
			file_flag_set(file, FILE_IS_EXCLUDED);
		}

		/* for each link */
		for (j = tommy_list_head(&disk->linklist); j != 0; j = j->next) {
			struct snapraid_link* slink = j->data;
			link_flag_set(slink, FILE_IS_EXCLUDED);
		}

		/* for each dir */
		for (j = tommy_list_head(&disk->dirlist); j != 0; j = j->next) {
			struct snapraid_dir* dir = j->data;
			dir_flag_set(dir, FILE_IS_EXCLUDED);
		}
	}
}

void state_filter(struct snapraid_state* state, tommy_list* filterlist_file, tommy_list* filterlist_disk, int filter_missing, int filter_error)
{
	tommy_node* i;
	unsigned l;

	/* if no filter, include all */
	if (!filter_missing && !filter_error && tommy_list_empty(filterlist_file) && tommy_list_empty(filterlist_disk))
		return;

	msg_progress("Selecting...\n");

	for (i = tommy_list_head(filterlist_disk); i != 0; i = i->next) {
		struct snapraid_filter* filter = i->data;
		msg_verbose("\t%s%s\n", filter->pattern, filter->is_disk ? "//" : "");
	}
	for (i = tommy_list_head(filterlist_file); i != 0; i = i->next) {
		struct snapraid_filter* filter = i->data;
		msg_verbose("\t%s%s\n", filter->pattern, filter->is_dir ? "/" : "");
	}
	if (filter_missing)
		msg_verbose("\t<missing>\n");
	if (filter_error)
		msg_verbose("\t<error>\n");

	/* for each disk */
	for (i = state->disklist; i != 0; i = i->next) {
		tommy_node* j;
		struct snapraid_disk* disk = i->data;

		/* if we filter for presence, we have to access the disk, so better to print something */
		if (filter_missing)
			msg_progress("Scanning disk %s...\n", disk->name);

		/* for each file */
		for (j = tommy_list_head(&disk->filelist); j != 0; j = j->next) {
			struct snapraid_file* file = j->data;

			if (filter_path(filterlist_disk, 0, disk->name, file->sub) != 0
				|| filter_path(filterlist_file, 0, disk->name, file->sub) != 0
				|| filter_existence(filter_missing, disk->dir, file->sub) != 0
				|| filter_correctness(filter_error, &state->infoarr, disk, file) != 0
			) {
				file_flag_set(file, FILE_IS_EXCLUDED);
			}
		}

		/* for each link */
		for (j = tommy_list_head(&disk->linklist); j != 0; j = j->next) {
			struct snapraid_link* slink = j->data;

			if (filter_error
				|| filter_path(filterlist_disk, 0, disk->name, slink->sub) != 0
				|| filter_path(filterlist_file, 0, disk->name, slink->sub) != 0
				|| filter_existence(filter_missing, disk->dir, slink->sub) != 0
			) {
				link_flag_set(slink, FILE_IS_EXCLUDED);
			}
		}

		/*
		 * Hardlink entries don't own data blocks; their data is represented by the
		 * target file. If a hardlink pathname is selected while its target file is
		 * excluded by a path filter, keep the target selected as well so check/fix
		 * can process the data backing the selected hardlink.
		 * With the missing filter, a present target must remain excluded because
		 * fix -m promises to leave existing files untouched. It can be used directly
		 * to recreate the missing hardlink without processing its data.
		 */
		for (j = tommy_list_head(&disk->linklist); j != 0; j = j->next) {
			struct snapraid_link* slink = j->data;

			if (link_flag_has(slink, FILE_IS_HARDLINK) && !link_flag_has(slink, FILE_IS_EXCLUDED)) {
				struct snapraid_file* file;

				file = tommy_hashdyn_search(&disk->pathset, file_path_compare_to_arg, slink->linkto, file_path_hash(slink->linkto));
				if (file && filter_existence(filter_missing, disk->dir, file->sub) == 0)
					file_flag_clear(file, FILE_IS_EXCLUDED);
			}
		}

		/* for each empty dir */
		for (j = tommy_list_head(&disk->dirlist); j != 0; j = j->next) {
			struct snapraid_dir* dir = j->data;

			if (filter_error
				|| filter_emptydir(filterlist_disk, 0, disk->name, dir->sub) != 0
				|| filter_emptydir(filterlist_file, 0, disk->name, dir->sub) != 0
				|| filter_existence(filter_missing, disk->dir, dir->sub) != 0
			) {
				dir_flag_set(dir, FILE_IS_EXCLUDED);
			}
		}
	}

	/* if we are filtering by disk, exclude any parity not explicitly included */
	if (!tommy_list_empty(filterlist_disk)) {
		/* for each parity disk */
		for (l = 0; l < state->level; ++l) {
			/* check if the parity is excluded by name */
			if (filter_path(filterlist_disk, 0, lev_config_name(l), 0) != 0) {
				/* excluded the parity from further operation */
				state->parity[l].is_excluded_by_filter = 1;
			}
		}
	} else {
		/* if we are filtering by file, exclude all parity */
		if (filter_missing || !tommy_list_empty(filterlist_file)) {
			/* for each parity disk */
			for (l = 0; l < state->level; ++l) {
				state->parity[l].is_excluded_by_filter = 1;
			}
		}
	}
}

int state_progress_begin(struct snapraid_state* state, block_off_t blockstart, block_off_t blockmax, block_off_t countmax)
{
	time_t now;

	if (state->opt.gui) {
		log_tag("run:begin:%" PRIu64 ":%" PRIu64 ":%" PRIu64 "\n", blockstart, blockmax, countmax);
		log_flush();
	}

	now = time(0);

	state->progress_whole_start = now;
	state->progress_tick = 0;
	state->progress_ptr = 0;
	state->progress_wasted = 0;

	/* start of thermal control */
	if (!state_thermal_begin(state, now))
		return -1;

	/* stop if requested */
	if (os_signal_interrupt()) {
		/* LCOV_EXCL_START */
		if (!state->opt.gui) {
			msg_status("Not starting due to interruption\n");
		}
		log_tag("sigint:0: SIGINT received\n");
		log_flush();
		return 1;
		/* LCOV_EXCL_STOP */
	}

	return 0;
}

void state_progress_end(struct snapraid_state* state, block_off_t countpos, block_off_t countmax, data_off_t countsize, const char* msg)
{
	if (state->opt.gui) {
		log_tag("run:end\n");
		log_flush();
	} else if (countmax == 0) {
		if (state->need_write || state->written) {
			msg_status("100%% completed\n");
		} else {
			msg_status("%s", msg);
		}
	} else {
		time_t now;
		time_t elapsed;

		data_off_t countsize_MB = (countsize + MEGA - 1) / MEGA;

		now = time(0);

		elapsed = now - state->progress_whole_start - state->progress_wasted;

		msg_bar("%u%% completed, %" PRIu64 " MB accessed", muldiv(countpos, 100, countmax), countsize_MB);

		msg_bar(" in %u:%02u", (unsigned)(elapsed / 3600), (unsigned)((elapsed % 3600) / 60));

		msg_bar("\n");
		msg_flush();
	}
}

void state_progress_stop(struct snapraid_state* state)
{
	time_t now;

	now = time(0);

	state->progress_interruption = now;
}

void state_progress_restart(struct snapraid_state* state)
{
	time_t now;

	now = time(0);

	/* reset the progress counter */
	state->progress_tick = 0;
	state->progress_ptr = 0;

	if (now >= state->progress_interruption) /* avoid degenerated cases when the clock is manually adjusted */
		state->progress_wasted += now - state->progress_interruption;
}

/**
 * Set the latest os_tick data in the progress vector.
 */
static void state_progress_latest(struct snapraid_state* state)
{
	tommy_node* i;
	unsigned l;

	state->progress_tick_misc[state->progress_ptr] = state->tick_misc;
	state->progress_tick_sched[state->progress_ptr] = state->tick_sched;
	state->progress_tick_raid[state->progress_ptr] = state->tick_raid;
	state->progress_tick_hash[state->progress_ptr] = state->tick_hash;
	state->progress_tick_io[state->progress_ptr] = state->tick_io;
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		disk->progress_tick[state->progress_ptr] = disk->tick;
	}
	for (l = 0; l < state->level; ++l)
		state->parity[l].progress_tick[state->progress_ptr] = state->parity[l].tick;
}

/**
 * Number of columns
 */
#define GRAPH_COLUMN 78

/**
 * Get the reference value, 0 if index is PROGRESS_MAX
 */
#define ref(map, index) (index < PROGRESS_MAX ? map[index] : 0)

static struct snapraid_thermal* state_thermal_find(struct snapraid_state* state, const char* name)
{
	tommy_node* i;
	struct snapraid_thermal* found = 0;

	for (i = tommy_list_head(&state->thermallist); i != 0; i = i->next) {
		struct snapraid_thermal* thermal = i->data;
		if (strcmp(thermal->name, name) == 0) {
			if (found == 0
			        /* if multiple matches, return the one with highest temperature */
				|| found->latest_temperature < thermal->latest_temperature
			        /* or, if the temperature is equal, the one with the higher r_squared */
				|| (found->latest_temperature == thermal->latest_temperature && found->params.r_squared < thermal->params.r_squared)
			) {
				found = thermal;
			}
		}
	}

	return found;
}

#define THERMAL_PAD "        "

static void state_progress_graph(struct snapraid_state* state, struct snapraid_io* io, unsigned current, unsigned oldest)
{
	uint64_t v;
	uint64_t tick_total;
	size_t bar;
	tommy_node* i;
	unsigned l;
	size_t pad;
	size_t pre;

	tick_total = 0;

	tick_total += state->progress_tick_misc[current] - ref(state->progress_tick_misc, oldest);
	tick_total += state->progress_tick_sched[current] - ref(state->progress_tick_sched, oldest);
	tick_total += state->progress_tick_raid[current] - ref(state->progress_tick_raid, oldest);
	tick_total += state->progress_tick_hash[current] - ref(state->progress_tick_hash, oldest);
	tick_total += state->progress_tick_io[current] - ref(state->progress_tick_io, oldest);
	if (!tick_total)
		return;

	pad = 4;
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;
		size_t len = strlen(disk->name);
		if (pad < len)
			pad = len;
	}
	for (l = 0; l < state->level; ++l) {
		size_t len = strlen(lev_config_name(l));
		if (pad < len)
			pad = len;
	}

	/* extra space */
	pad += 1;

	pre = 4;
	if (pad + pre + 30 < GRAPH_COLUMN)
		bar = GRAPH_COLUMN - pad - pre;
	else
		bar = 30; /* at least a bar of 30 */

	if (io) {
		const char* legend = "cached blocks (instant, more is better)";

		/* refresh the cached blocks info */
		io_refresh(io);

		printf("\n");

		for (i = state->disklist; i != 0; i = i->next) {
			struct snapraid_disk* disk = i->data;
			v = disk->cached_blocks;
			printr(disk->name, pad);
			printf("%4" PRIu64 " | ", v);

			if (disk->progress_file && disk->progress_file->sub)
				printf("%s", disk->progress_file->sub);
			else
				printf("-");

			printf("\n");
		}

		for (l = 0; l < state->level; ++l) {
			v = state->parity[l].cached_blocks;
			printr(lev_config_name(l), pad);
			printf("%4" PRIu64 " | ", v);
			printc('o', v * bar / io->io_max);
			printf("\n");
		}

		printc(' ', pad);
		printf("     |_");
		printc('_', bar);
		printf("\n");

		printc(' ', 5 + pad + 1 + bar / 2 - strlen(legend) / 2);
		printf("%s", legend);
		printf("\n");
	}

	printf("\n");

	pre = 4;
	if (state->thermal_temperature_limit != 0)
		pre += strlen(THERMAL_PAD);
	if (pad + pre + 30 < GRAPH_COLUMN)
		bar = GRAPH_COLUMN - pad - pre;
	else
		bar = 30; /* at least a bar of 30 */

	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		v = disk->progress_tick[current] - ref(disk->progress_tick, oldest);
		printr(disk->name, pad);

		if (state->thermal_temperature_limit != 0) {
			struct snapraid_thermal* thermal = state_thermal_find(state, disk->name);
			if (thermal) {
				if (thermal->params.r_squared >= THERMAL_R_SQUARED_LIMIT)
					printf(" %2u (%2u)", thermal->latest_temperature, (int)thermal->params.t_steady);
				else
					printf(" %2u     ", thermal->latest_temperature);
			} else {
				printf(THERMAL_PAD);
			}
		}

		printf("%3u%% | ", muldiv(v, 100, tick_total));
		printc('*', v * bar / tick_total);
		printf("\n");

		/* clear the file in progress */
		disk->progress_file = 0;
	}

	for (l = 0; l < state->level; ++l) {
		v = state->parity[l].progress_tick[current] - ref(state->parity[l].progress_tick, oldest);
		printr(lev_config_name(l), pad);

		if (state->thermal_temperature_limit != 0) {
			struct snapraid_thermal* thermal = state_thermal_find(state, lev_config_name(l));
			if (thermal) {
				if (thermal->params.r_squared >= THERMAL_R_SQUARED_LIMIT)
					printf(" %2u (%2u)", thermal->latest_temperature, (int)thermal->params.t_steady);
				else
					printf(" %2u     ", thermal->latest_temperature);
			} else {
				printf(THERMAL_PAD);
			}
		}

		printf("%3u%% | ", muldiv(v, 100, tick_total));
		printc('*', v * bar / tick_total);
		printf("\n");
	}

	v = state->progress_tick_raid[current] - ref(state->progress_tick_raid, oldest);
	printr("raid", pad);
	if (state->thermal_temperature_limit != 0)
		printf(THERMAL_PAD);
	printf("%3u%% | ", muldiv(v, 100, tick_total));
	printc('*', v * bar / tick_total);
	printf("\n");

	v = state->progress_tick_hash[current] - ref(state->progress_tick_hash, oldest);
	printr("hash", pad);
	if (state->thermal_temperature_limit != 0)
		printf(THERMAL_PAD);
	printf("%3u%% | ", muldiv(v, 100, tick_total));
	printc('*', v * bar / tick_total);
	printf("\n");

	v = state->progress_tick_sched[current] - ref(state->progress_tick_sched, oldest);
	printr("sched", pad);
	if (state->thermal_temperature_limit != 0)
		printf(THERMAL_PAD);
	printf("%3u%% | ", muldiv(v, 100, tick_total));
	printc('*', v * bar / tick_total);
	printf("\n");

	v = state->progress_tick_misc[current] - ref(state->progress_tick_misc, oldest);
	printr("misc", pad);
	if (state->thermal_temperature_limit != 0)
		printf(THERMAL_PAD);
	printf("%3u%% | ", muldiv(v, 100, tick_total));
	printc('*', v * bar / tick_total);
	printf("\n");

	printc(' ', pad);
	if (state->thermal_temperature_limit != 0)
		printf(THERMAL_PAD);
	printf("     |_");
	printc('_', bar);
	printf("\n");

	if (oldest == PROGRESS_MAX) {
		const char* legend = "wait time (total, less is better)";
		if (state->thermal_temperature_limit != 0)
			printf(THERMAL_PAD);
		printc(' ', 5 + pad + 1 + bar / 2 - strlen(legend) / 2);
		printf("%s", legend);
		printf("\n");
	} else {
		const char* legend_d = "wait time (last %d secs, less is better)";
		if (state->thermal_temperature_limit != 0)
			printf(THERMAL_PAD);
		printc(' ', 5 + pad + 1 + bar / 2 - strlen(legend_d) / 2);
		printf(legend_d, PROGRESS_MAX);
		printf("\n");
	}

	printf("\n");
}

int state_progress(struct snapraid_state* state, struct snapraid_io* io, block_off_t blockpos, block_off_t countpos, block_off_t countmax, data_off_t countsize)
{
	time_t now;
	uint64_t thermal_now;
	int pred;

	now = time(0);
	thermal_now = os_tick_sec();

	/* thermal measure */
	if (thermal_now > state->thermal_latest + THERMAL_PERIOD_SECONDS || state->opt.fake_device) {
		state->thermal_latest = thermal_now;
		state_thermal(state, now);
	}

	/* previous position */
	pred = state->progress_ptr + PROGRESS_MAX - 1;
	if (pred >= PROGRESS_MAX)
		pred -= PROGRESS_MAX;

	/* if the previous measure is different */
	if (state->progress_tick == 0
		|| state->progress_time[pred] != now
	) {
		time_t elapsed;
		unsigned out_perc = 0;
		unsigned out_size_speed = 0;
		unsigned out_block_speed = 0;
		unsigned out_cpu = 0;
		unsigned out_eta = 0;
		int out_temperature = 0;
		int out_steady = 0;
		int out_computed = 0;

		/* store the new measure */
		state->progress_time[state->progress_ptr] = now;
		state->progress_pos[state->progress_ptr] = countpos;
		state->progress_size[state->progress_ptr] = countsize;
		state_progress_latest(state);

		elapsed = now - state->progress_whole_start - state->progress_wasted;

		/* completion percentage */
		if (countmax)
			out_perc = muldiv(countpos, 100, countmax);

		/* if we have at least 5 measures */
		if (state->progress_tick >= 5
		        /* or if we are running in test mode, with at least one measure */
			|| (state->opt.force_progress && state->progress_tick >= 1)
		) {
			int oldest;
			int past;
			time_t delta_time;
			block_off_t delta_pos;
			data_off_t delta_size;
			uint64_t tick_cpu;
			uint64_t tick_total;
			uint64_t delta_tick_cpu;
			uint64_t delta_tick_total;
			uint64_t oldest_tick_cpu;
			uint64_t oldest_tick_total;
			tommy_node* i;

			/* number of past measures */
			past = state->progress_tick;

			/*
			 * Drop the oldest ones, to promptly
			 * skip the startup phase
			 */
			past -= past / 5;

			/* check how much we can go in the past */
			if (past >= PROGRESS_MAX - 1) {
				/*
				 * The vector is filled, so we are already in position
				 * to get the possible oldest one
				 */
				oldest = state->progress_ptr + 1;
			} else {
				/* go backward the number of positions selected */
				oldest = state->progress_ptr + PROGRESS_MAX - past;
			}
			if (oldest >= PROGRESS_MAX)
				oldest -= PROGRESS_MAX;

			tick_cpu = state->progress_tick_misc[state->progress_ptr]
				+ state->progress_tick_sched[state->progress_ptr]
				+ state->progress_tick_raid[state->progress_ptr]
				+ state->progress_tick_hash[state->progress_ptr];
			tick_total = tick_cpu + state->progress_tick_io[state->progress_ptr];

			oldest_tick_cpu = state->progress_tick_misc[oldest]
				+ state->progress_tick_sched[oldest]
				+ state->progress_tick_raid[oldest]
				+ state->progress_tick_hash[oldest];
			oldest_tick_total = oldest_tick_cpu + state->progress_tick_io[oldest];

			delta_time = now - state->progress_time[oldest];
			delta_pos = countpos - state->progress_pos[oldest];
			delta_size = countsize - state->progress_size[oldest];
			delta_tick_cpu = tick_cpu - oldest_tick_cpu;
			delta_tick_total = tick_total - oldest_tick_total;

			/* estimate the speed in MB/s */
			if (delta_time != 0)
				out_size_speed = (unsigned)(delta_size / MEGA / delta_time);

			/* estimate the speed in block/s */
			if (delta_time != 0)
				out_block_speed = (unsigned)(delta_pos / delta_time);

			/* estimate the cpu usage percentage */
			if (delta_tick_total != 0)
				out_cpu = muldiv(delta_tick_cpu, 100, delta_tick_total);

			/* estimate the remaining time in minutes */
			if (delta_pos != 0)
				out_eta = muldiv(countmax - countpos, delta_time, delta_pos);

			if (state->opt.force_stats) {
				os_clear();
				state_progress_graph(state, io, state->progress_ptr, oldest);
			}

			if (state->thermal_temperature_limit != 0) {
				/* get the max temperature */
				out_temperature = 0;
				for (i = tommy_list_head(&state->thermallist); i != 0; i = i->next) {
					struct snapraid_thermal* thermal = i->data;
					if (out_temperature < thermal->latest_temperature) {
						out_temperature = thermal->latest_temperature;
						if (thermal->params.r_squared >= THERMAL_R_SQUARED_LIMIT)
							out_steady = thermal->params.t_steady;
						else
							out_steady = 0;
					}
				}
			} else {
				out_temperature = 0;
				out_steady = 0;
			}

			/* we have the output value */
			out_computed = 1;
		}

		if (state->opt.gui) {
			char str_eta[32];
			char str_size_speed[32];
			char str_cpu[32];
			char str_temp[32];
			char str_steady[32];
			if (out_computed) {
				snprintf(str_eta, sizeof(str_eta), "%u", out_eta);
				snprintf(str_size_speed, sizeof(str_size_speed), "%u", out_size_speed);
				snprintf(str_cpu, sizeof(str_cpu), "%u", out_cpu);
			} else {
				str_eta[0] = 0;
				str_size_speed[0] = 0;
				str_cpu[0] = 0;
			}
			if (out_temperature) {
				snprintf(str_temp, sizeof(str_temp), "%u", out_temperature);
			} else {
				str_temp[0] = 0;
			}
			if (out_steady) {
				snprintf(str_steady, sizeof(str_steady), "%u", out_steady);
			} else {
				str_steady[0] = 0;
			}
			log_tag("run:pos:%" PRIu64 ":%" PRIu64 ":%" PRIu64 ":%u:%s:%s:%s:%" PRIu64 ":%s:%s\n", blockpos, countpos, countsize, out_perc, str_eta, str_size_speed, str_cpu, (uint64_t)elapsed, str_temp, str_steady);
			log_flush();
		} else {
			msg_bar("%u%%, %u MB", out_perc, (unsigned)(countsize / MEGA));
			if (out_computed) {
				msg_bar(", %u MB/s", out_size_speed);
				msg_bar(", %u stripe/s", out_block_speed);
				msg_bar(", CPU %u%%", out_cpu);
				if (out_temperature != 0) {
					if (out_steady != 0)
						msg_bar(", Tmax %u (%u)", out_temperature, out_steady);
					else
						msg_bar(", Tmax %u", out_temperature);
				}
				unsigned out_eta_minutes = out_eta / 60; /* minutes */
				msg_bar(", %u:%02u ETA", out_eta_minutes / 60, out_eta_minutes % 60);
			}
			msg_bar("\r");
			msg_flush();
		}

		/* next position to fill */
		++state->progress_ptr;
		if (state->progress_ptr >= PROGRESS_MAX)
			state->progress_ptr -= PROGRESS_MAX;

		/* one more measure */
		++state->progress_tick;
	}

	/* stop if requested */
	if (os_signal_interrupt()) {
		/* LCOV_EXCL_START */
		if (!state->opt.gui) {
			msg_bar("\n");
			msg_flush();
		}
		log_error(EUSER, "Stopping for interruption at block %" PRIu64 "\n", blockpos);
		log_tag("sigint:%" PRIu64 ": SIGINT received\n", blockpos);
		log_flush();
		return 1;
		/* LCOV_EXCL_STOP */
	}

	return 0;
}

void state_usage_waste(struct snapraid_state* state)
{
	uint64_t now = os_tick();

	state->tick_last = now;
}

void state_usage_misc(struct snapraid_state* state)
{
	uint64_t now = os_tick();
	uint64_t delta = now - state->tick_last;

	/* increment the time spent in computations */
	state->tick_misc += delta;

	state->tick_last = now;
}

void state_usage_sched(struct snapraid_state* state)
{
	uint64_t now = os_tick();
	uint64_t delta = now - state->tick_last;

	/* increment the time spent in computations */
	state->tick_sched += delta;

	state->tick_last = now;
}

void state_usage_raid(struct snapraid_state* state)
{
	uint64_t now = os_tick();
	uint64_t delta = now - state->tick_last;

	/* increment the time spent in computations */
	state->tick_raid += delta;

	state->tick_last = now;
}

void state_usage_hash(struct snapraid_state* state)
{
	uint64_t now = os_tick();
	uint64_t delta = now - state->tick_last;

	/* increment the time spent in computations */
	state->tick_hash += delta;

	state->tick_last = now;
}

void state_usage_file(struct snapraid_state* state, struct snapraid_disk* disk, struct snapraid_file* file)
{
	(void)state;

	disk->progress_file = file;
}

void state_usage_disk(struct snapraid_state* state, struct snapraid_handle* handle_map, unsigned* waiting_map, unsigned waiting_mac)
{
	uint64_t now = os_tick();
	uint64_t delta = now - state->tick_last;
	unsigned i;

	/* increment the time spent in the data disks */
	for (i = 0; i < waiting_mac; ++i) {
		struct snapraid_disk* disk = handle_map[waiting_map[i]].disk;

		if (!disk)
			continue;

		disk->tick += delta;
	}
	state->tick_io += delta;

	state->tick_last = now;
}

void state_usage_parity(struct snapraid_state* state, unsigned* waiting_map, unsigned waiting_mac)
{
	uint64_t now = os_tick();
	uint64_t delta = now - state->tick_last;
	unsigned i;

	/* increment the time spent in the parity disk */
	for (i = 0; i < waiting_mac; ++i)
		state->parity[waiting_map[i]].tick += delta;
	state->tick_io += delta;

	state->tick_last = now;
}

void state_usage_print(struct snapraid_state* state)
{
	/* set the latest data */
	state_progress_latest(state);

	if (msg_level < MSG_PROGRESS)
		return;

	/* print a graph for it */
	state_progress_graph(state, 0, state->progress_ptr, PROGRESS_MAX);
}

void state_fscheck(struct snapraid_state* state, const char* ope)
{
	tommy_node* i;

	/* check the file-system on all disks */
	for (i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		if (fs_check(disk) != 0) {
			/* LCOV_EXCL_START */
			log_fatal(EINTERNAL, "Internal inconsistency: File-system check for disk '%s' %s\n", disk->name, ope);
			os_abort();
			/* LCOV_EXCL_STOP */
		}
	}
}

void generate_configuration(const char* path)
{
	struct snapraid_state state;
	struct snapraid_content* content;
	unsigned l, s;
	tommy_node* j;

	state_init(&state);

	/* mark that we are without a configuration file */
	state.no_conf = 1;

	/* create the dummy content entry */
	content = content_alloc(path, -1);

	/* adds the content entry */
	tommy_list_insert_tail(&state.contentlist, &content->node, content);

	/* read the content file */
	state_read(&state);

	/* output a dummy configuration file */
	printf("# Configuration file generated from %s\n", path);
	printf("\n");
	printf("# Use this blocksize\n");
	printf("blocksize %u\n", state.block_size / KIBI);
	printf("\n");
	printf("# Use this hashsize\n");
	printf("hashsize %zu\n", BLOCK_HASH_SIZE);
	printf("\n");
	for (l = 0; l < state.level; ++l) {
		printf("# Set the correct path for the %s files\n", lev_name(l));
		printf("# You had %u of them:\n", state.parity[l].split_mac);
		for (s = 0; s < state.parity[l].split_mac; ++s) {
			printf("# %u:\n", s);
			printf("# PATH:");
			if (state.parity[l].split_map[s].path[0])
				printf("%s", state.parity[l].split_map[s].path);
			else
				printf("?");
			printf("\n");
			printf("# SIZE:%" PRIu64 "\n", state.parity[l].split_map[s].size);
			printf("# UUID:");
			if (state.parity[l].split_map[s].uuid[0])
				printf("%s", state.parity[l].split_map[s].uuid);
			else
				printf("?");
			printf("\n");
			printf("#\n");
		}
		printf("%s ENTER_HERE_THE_PARITY_FILES_COMMA_SEPARATED\n", lev_config_name(l));
		printf("\n");
	}
	printf("# Add any other content file\n");
	printf("content %s\n", path);
	printf("\n");
	for (j = state.maplist; j; j = j->next) {
		struct snapraid_map* map = j->data;
		struct snapraid_disk* disk;
		printf("# Set the correct dir for disk '%s'\n", map->name);
		if (map->content_uuid[0])
			printf("# Disk '%s' is the one with id '%s'\n", map->name, map->content_uuid);
		disk = find_disk_by_name(&state, map->name);
		if (disk && disk->filelist) {
			struct snapraid_file* file = disk->filelist->data;
			if (file) {
				printf("# and containing: %s\n", fmt_poll(disk, file->sub));
			}
		}
		printf("data %s ENTER_HERE_THE_DIR\n", map->name);
		printf("\n");
	}

	state_done(&state);
}

void state_read_barrier(struct snapraid_io* io)
{
	io_quiesce(io);
}

int state_write_barrier(struct snapraid_state* state, struct snapraid_io* io, struct snapraid_parity_handle* parity_handle, block_off_t blockcur)
{
	unsigned l;

	if (io) {
		int writer_error[IO_WRITER_ERROR_MAX];
		unsigned j;

		io_flush(io, writer_error);

		for (j = 0; j < IO_WRITER_ERROR_MAX; ++j) {
			if (writer_error[j]) {
				errno = EIO;
				return -1;
			}
		}
	}

	/* flush all parity handles to ensure data is written to disk */
	for (l = 0; l < state->level; ++l) {
		int ret = parity_sync(&parity_handle[l]);
		if (ret == -1) {
			/* LCOV_EXCL_START */
			log_tag("parity_error_io:%" PRIu64 ":%s: Sync error. %s.\n", blockcur, lev_config_name(l), strerror(errno));
			log_fatal(errno, "DANGER! Unexpected input/output error in disk %s. It isn't possible to continue.\n", lev_config_name(l));
			return -1;
			/* LCOV_EXCL_STOP */
		}
	}

	/* complete all read-ahead without consuming the results */
	if (io)
		state_read_barrier(io);

	return 0;
}

int state_load_ignore_file(tommy_list* filter_list, const char* path, const char* sub)
{
	STREAM* f;
	int line;
	int error = 0;

	f = sopen_read(path, STREAM_FLAGS_NOFOLLOW);
	if (!f) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error opening the ignore file '%s'. %s.\n", path, strerror(errno));
		return -1;
		/* LCOV_EXCL_STOP */
	}

	line = 1;
	while (1) {
		char buffer[PATH_MAX];
		ssize_t ret;
		int c;

		/* skip initial spaces */
		sgetspace(f);

		/* get the whole line */
		ret = sgetline(f, buffer, sizeof(buffer));
		if (ret < 0) {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Too long line in '%s' at line %u\n", path, line);
			error = 1;
			break;
			/* LCOV_EXCL_STOP */
		}

		if (buffer[0] == 0) {
			/* allow empty lines */
		} else if (buffer[0] == '#') {
			/* ignore comment lines */
		} else {
			struct snapraid_filter* filter;

			filter = filter_alloc_file(-1, sub, buffer);
			if (!filter) {
				/* LCOV_EXCL_START */
				log_fatal(EUSER, "Invalid ignore specification '%s' in '%s' at line %u\n", buffer, path, line);
				error = 1;
				break;
				/* LCOV_EXCL_STOP */
			}

			tommy_list_insert_tail(filter_list, &filter->node, filter);
		}

		/* next line */
		c = sgeteol(f);
		if (c == EOF) {
			break;
		}
		if (c != '\n') {
			/* LCOV_EXCL_START */
			log_fatal(EUSER, "Extra data in '%s' at line %u\n", path, line);
			error = 1;
			break;
			/* LCOV_EXCL_STOP */
		}
		++line;
	}

	if (serror(f)) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error reading the ignore file '%s' at line %u. %s.\n", path, line, strerror(errno));
		error = 1;
		/* LCOV_EXCL_STOP */
	}

	if (sclose(f) != 0) {
		/* LCOV_EXCL_START */
		log_fatal(errno, "Error closing the ignore file '%s'. %s.\n", path, strerror(errno));
		error = 1;
		/* LCOV_EXCL_STOP */
	}

	return error ? -1 : 0;
}

static int state_snapshot_dir(struct fssnapshot_struct* fss, const char* name, struct snapraid_disk* disk)
{
	struct stat st;

	/* fail if it doesn't exists at high level */
	if (fssnapshot_stat(fss, name, &st) != 0)
		return -1;

	if (disk) {
		char vol[PATH_MAX];

		if (fssnapshot_path(fss, name, vol, sizeof(vol)) != 0)
			return -1;

		pathcpy(disk->dir, sizeof(disk->dir), vol);
	}

	return 0;
}

int state_snapshot_scan(struct snapraid_state* state)
{
	if (!state->snapshot)
		return 0;

	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		/* check if it supports snapshot */
		int res = fssnapshot_mount(disk->mount_point, &disk->fss);
		if (res > 0)
			continue;
		if (res < 0) {
			log_fatal(EINTERNAL, "Failed to setup snapshot for disk '%s'.\n", disk->name);
			return -1;
		}

		msg_progress("Creating disk %s scan snapshot...\n", disk->name);

		/* delete a potential previous scan snapshot */
		if (state_snapshot_dir(&disk->fss, SNAPSHOT_SCAN, 0) == 0) {
			if (fssnapshot_delete(&disk->fss, SNAPSHOT_SCAN) != 0) {
				log_fatal(errno, "Failed to delete scan snapshot in '%s'. %s.\n", disk->fss.snapshot_dir, strerror(errno));
				return -1;
			}
		}

		/* create a new snapshot for scanning */
		if (fssnapshot_create(&disk->fss, SNAPSHOT_SCAN) != 0) {
			log_fatal(errno, "Failed to create scan snapshot '%s'. %s.\n", disk->fss.snapshot_dir, strerror(errno));
			return -1;
		}

		/* setup the snapshot in use */
		if (state_snapshot_dir(&disk->fss, SNAPSHOT_SCAN, disk) != 0) {
			log_error(errno, "Error stating scan snapshot '%s'. %s.\n", disk->fss.snapshot_dir, strerror(errno));
			return -1;
		}
	}

	return 0;
}

int state_snapshot_pending(struct snapraid_state* state)
{
	int error = 0;

	if (!state->snapshot)
		return 0;

	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		/* check if it supports snapshot */
		if (disk->fss.magic == 0)
			continue;

		msg_progress("Creating disk %s pending snapshot...\n", disk->name);

		/* publish the scan snapshot without dropping the previous recovery source first */
		if (fssnapshot_replace(&disk->fss, SNAPSHOT_SCAN, SNAPSHOT_PENDING) != 0) {
			log_fatal(errno, "Failed to replace pending snapshot in '%s'. %s.\n", disk->fss.snapshot_dir, strerror(errno));
			error = -1;
			continue;
		}

		/* setup the snapshot in use */
		if (state_snapshot_dir(&disk->fss, SNAPSHOT_PENDING, disk) != 0) {
			log_error(errno, "Error stating pending snapshot '%s'. %s.\n", disk->fss.snapshot_dir, strerror(errno));
			error = -1;
			continue;
		}
	}

	return error;
}

int state_snapshot_commit(struct snapraid_state* state)
{
	int error = 0;

	if (!state->snapshot)
		return 0;

	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		/* check if it supports snapshot */
		if (disk->fss.magic == 0)
			continue;

		msg_progress("Committing disk %s stable snapshot...\n", disk->name);

		/* publish pending without dropping the previous stable deallocation source first */
		if (fssnapshot_replace(&disk->fss, SNAPSHOT_PENDING, SNAPSHOT_STABLE) != 0) {
			log_fatal(errno, "Failed to replace stable snapshot in '%s'. %s.\n", disk->fss.snapshot_dir, strerror(errno));
			error = -1;
			continue;
		}
	}

	return error;
}

void state_snapshot_read(struct snapraid_state* state)
{
	if (!state->snapshot)
		return;

	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		/* check if it supports snapshot */
		int res = fssnapshot_mount(disk->mount_point, &disk->fss);
		if (res > 0)
			continue;
		if (res < 0) {
			log_error(EUSER, "WARNING! Disk %s snapshot setup or recovery failed, falling back to live filesystem.\n", disk->name);
			log_error(EUSER, "Recovery capability may be reduced if files have changed since the last successful sync.\n");
			continue;
		}

		if (state_snapshot_dir(&disk->fss, SNAPSHOT_PENDING, disk) == 0) {
			msg_progress("Using disk %s pending snapshot...\n", disk->name);
		} else {
			if (state_snapshot_dir(&disk->fss, SNAPSHOT_STABLE, disk) == 0) {
				msg_progress("Using disk %s stable snapshot...\n", disk->name);
			} else {
				/* fallback to standard mount point */
				log_error(EUSER, "WARNING! Disk %s snapshot missing, falling back to live filesystem.\n", disk->name);
				log_error(EUSER, "Recovery capability may be reduced if files have changed since the last successful sync.\n");
				msg_progress("Using disk %s live filesystem...\n", disk->name);
			}
		}
	}
}

void state_snapshot_write(struct snapraid_state* state, tommy_list* filterlist_disk)
{
	if (!state->snapshot)
		return;

	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		/* check if it supports snapshot */
		int res = fssnapshot_mount(disk->mount_point, &disk->fss);
		if (res > 0)
			continue;
		if (res < 0) {
			log_error(EUSER, "WARNING! Disk %s snapshot setup or recovery failed, falling back to live filesystem.\n", disk->name);
			log_error(EUSER, "Recovery capability may be reduced if files have changed since the last successful sync.\n");
			continue;
		}

		int writable = tommy_list_empty(filterlist_disk) || filter_path(filterlist_disk, 0, disk->name, 0) == 0;

		if (writable) {
			msg_progress("Using disk %s live filesystem...\n", disk->name);
		} else if (state_snapshot_dir(&disk->fss, SNAPSHOT_PENDING, disk) == 0) {
			msg_progress("Using disk %s pending snapshot...\n", disk->name);
		} else if (state_snapshot_dir(&disk->fss, SNAPSHOT_STABLE, disk) == 0) {
			msg_progress("Using disk %s stable snapshot...\n", disk->name);
		} else {
			/* fallback to standard mount point */
			log_error(EUSER, "WARNING! Disk %s snapshot missing, falling back to live filesystem.\n", disk->name);
			log_error(EUSER, "Recovery capability may be reduced if files have changed since the last successful sync.\n");
			msg_progress("Using disk %s live filesystem...\n", disk->name);
		}

		/*
		 * If sync was interrupted before updating parity, the content file
		 * already records BLOCK_STATE_DELETED and pending deallocations, while the
		 * deleted files remain physically present in the stable snapshot. Import them
		 * into the import index so OLD recovery hypotheses can resolve them without
		 * consuming RAID equations, even for disks that use the live filesystem for normal I/O.
		 */
		if (!tommy_list_empty(&disk->dealloclist) && state_snapshot_dir(&disk->fss, SNAPSHOT_STABLE, 0) == 0) {
			char vol[PATH_MAX];

			msg_progress("Importing disk %s stable snapshot %" PRIu64 " deallocated files...\n", disk->name, (uint64_t)tommy_list_count(&disk->dealloclist));

			if (fssnapshot_path(&disk->fss, SNAPSHOT_STABLE, vol, sizeof(vol)) == 0)
				state_dealloc(state, vol, &disk->dealloclist);
		}
	}
}

void state_snapshot_cleanup(struct snapraid_state* state)
{
	for (tommy_node* i = state->disklist; i != 0; i = i->next) {
		struct snapraid_disk* disk = i->data;

		/* check if it supports snapshot */
		if (disk->fss.magic == 0)
			continue;

		fssnapshot_unmount(&disk->fss);
	}
}