/*
 *  Copyright (C) 2008-2015, Parallels, Inc. All rights reserved.
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#include <stdio.h>
#include <stdlib.h>
#include <stddef.h>
#include <errno.h>
#include <fcntl.h>
#include <unistd.h>
#include <malloc.h>
#include <sys/stat.h>
#include <sys/mount.h>
#include <limits.h>
#include <getopt.h>
#include <linux/types.h>
#include <string.h>
#include <assert.h>

#include "ploop.h"

static int sync_cache(struct delta * delta)
{
	int skip = 0;

	if (!delta->l2_dirty)
		return 0;

	/* Sync data before we write out new index table. */
	if (fsync(delta->fd)) {
		ploop_err(errno, "fsync");
		return -1;
	}

	if (delta->l2_cache < 0) {
		ploop_err(0, "abort: delta->l2_cache < 0");
		return -1;
	}
	if (delta->l2_cache >= delta->l1_size) {
		ploop_err(0, "abort: delta->l2_cache >= delta->l1_size");
		return -1;
	}

	if (delta->l2_cache == 0)
		skip = sizeof(struct ploop_pvd_header);

	/* Write index table */
	if (PWRITE(delta, (__u8 *)delta->l2 + skip,
				S2B(delta->blocksize) - skip,
				(off_t)delta->l2_cache * S2B(delta->blocksize) + skip))
		return SYSEXIT_WRITE;

	/* Sync index table. We can delay this, but this does not
	 * improve performance
	 */
	if (fsync(delta->fd)) {
		ploop_err(errno, "fsync");
		return -1;
	}
	delta->l2_dirty = 0;
	return 0;
}

static int locate_l2_entry(struct delta_array *p, int level, int i, __u32 k, int *out)
{
	__u64 cluster;
	for (level++; level < p->delta_max; level++) {
		if (p->delta_arr[level].l2_cache == -1) {
			if (i >= p->delta_arr[level].l1_size)
				break; /* grow is monotonic! */
			cluster = S2B(p->delta_arr[level].blocksize);
			if (PREAD(&p->delta_arr[level], p->delta_arr[level].l2,
						cluster, (off_t)i * cluster))
				return SYSEXIT_READ;
			p->delta_arr[level].l2_cache = i;
		}
		if (p->delta_arr[level].l2[k]) {
			*out = level;
			return 0;
		}
	}
	*out = -1;
	return 0;
}

static int grow_lower_delta(const char *device, int top, int start_level, int end_level)
{
	off_t src_size = 0; /* bdsize of source delta to merge */
	off_t dst_size = 0; /* bdsize of destination delta for merge */
	char *src_image = NULL;
	char *dst_image = NULL;
	int i, devfd;
	struct ploop_pvd_header *vh;
	struct grow_maps grow_maps;
	char *fmt;
	int dst_is_raw = 0;
	void *buf = NULL;
	struct delta odelta = {};
	int ret;
	__u64 cluster;

	if (top) {
		if ((ret = ploop_get_size(device, &src_size)))
			return ret;
	} else {
		if (find_delta_names(device, end_level, end_level,
				     &src_image, &fmt)) {
			ploop_err(errno, "find_delta_names");
			ret = SYSEXIT_SYSFS;
			goto done;
		}

		if ((ret = read_size_from_image(src_image, 0, &src_size)))
			goto done;
	}

	if (find_delta_names(device, start_level, start_level,
			     &dst_image, &fmt)) {
		ploop_err(errno, "find_delta_names");
		ret = SYSEXIT_SYSFS;
		goto done;
	}

	if (strcmp(fmt, "raw") == 0)
		dst_is_raw = 1;

	if ((ret = read_size_from_image(dst_image, dst_is_raw, &dst_size)))
		goto done;

	if (src_size <= dst_size) {
		ret = 0;
		goto done;
	}

	if (dst_is_raw) {
		ret = grow_raw_delta(dst_image, S2B(src_size - dst_size));
		goto done;
	}

	/* Here we know for sure that destination delta is in ploop1 format */
	if (open_delta(&odelta, dst_image, O_RDWR, OD_NOFLAGS)) {
		ploop_err(errno, "open_delta");
		ret = SYSEXIT_OPEN;
		goto done;
	}

	if (dirty_delta(&odelta)) {
		ploop_err(errno, "dirty_delta");
		ret = SYSEXIT_WRITE;
		goto done;
	}
	cluster = S2B(odelta.blocksize);
	if (p_memalign(&buf, 4096, cluster)) {
		ret = SYSEXIT_MALLOC;
		goto done;
	}

	/* relocate blocks w/o nullifying them and changing on-disk header */
	if ((ret = grow_delta(&odelta, src_size, buf, &grow_maps)))
		goto done;

	if (clear_delta(&odelta)) {
		ploop_err(errno, "clear_delta");
		ret = SYSEXIT_WRITE;
		goto done;
	}

	devfd = open(device, O_RDONLY);
	if (devfd < 0) {
		ploop_err(errno, "open dev");
		ret = SYSEXIT_DEVICE;
		goto done;
	}

	/* update in-core map_node mappings for relocated blocks */
	grow_maps.ctl->level = start_level;
	ret = ioctl_device(devfd, PLOOP_IOC_UPDATE_INDEX, grow_maps.ctl);
	close(devfd);
	if (ret)
		goto done;

	/* nullify relocated blocks on disk */
	memset(buf, 0, cluster);
	for (i = 0; i < grow_maps.ctl->n_maps; i++)
		if (PWRITE(&odelta, buf, cluster,
			   (off_t)(grow_maps.zblks[i]) * cluster)) {
			ret = SYSEXIT_WRITE;
			goto done;
		}

	/* save new image header of destination delta on disk */
	vh = (struct ploop_pvd_header *)buf;
	generate_pvd_header(vh, src_size, odelta.blocksize, odelta.version);
	if (PREAD(&odelta, &vh->m_Flags, sizeof(vh->m_Flags),
		  offsetof(struct ploop_pvd_header, m_Flags))) {
		ret = SYSEXIT_READ;
		goto done;
	}
	if (PWRITE(&odelta, vh, sizeof(*vh), 0)) {
		ret = SYSEXIT_WRITE;
		goto done;
	}

	if (fsync(odelta.fd)) {
		ploop_err(errno, "fsync");
		ret = SYSEXIT_FSYNC;
	}

done:
	free(buf);
	free(src_image);
	free(dst_image);
	close_delta(&odelta);
	return ret;
}

int get_delta_info(const char *device, struct merge_info *info)
{
	char *fmt;

	if (ploop_get_attr(device, "top", &info->top_level)) {
		ploop_err(0, "Can't find top delta");
		return SYSEXIT_SYSFS;
	}

	if (info->top_level == 0) {
		ploop_err(0, "Single delta, nothing to merge");
		return SYSEXIT_PARAM;
	}

	if (info->end_level == 0)
		info->end_level = info->top_level;

	if (info->end_level > info->top_level ||
	    info->start_level > info->end_level)
	{
		ploop_err(0, "Illegal top level");
		return SYSEXIT_SYSFS;
	}

	if (info->end_level == info->top_level) {
		int running;

		if (ploop_get_attr(device, "running", &running)) {
			ploop_err(0, "Can't get running attr");
			return SYSEXIT_SYSFS;
		}

		if (running) {
			int ro;

			if (ploop_get_delta_attr(device, info->top_level, "ro", &ro)) {
				ploop_err(0, "Can't get ro attr");
				return SYSEXIT_SYSFS;
			}
			if (!ro)
				info->merge_top = 1;
		}
	}
	info->names = calloc(1, (info->end_level - info->start_level + 2) * sizeof(char*));
	if (info->names == NULL) {
		ploop_err(errno, "malloc");
		return SYSEXIT_MALLOC;
	}
	if (find_delta_names(device, info->start_level, info->end_level, info->names, &fmt))
		return SYSEXIT_SYSFS;
	if (strcmp(fmt, "raw") == 0)
		info->raw = 1;

	return 0;
}

/* Merge one or more deltas (specified by **images and start/end_level).
 * Merge can be online or offline, with or without top delta, to either
 * a lowest speicifed delta or to a new_image.
 *
 * Parameters:
 *  dev		ploop device (for online merge) or NULL (for offline merge)
 *  start_level	start (lower) delta level to merge (online only)
 *  end_level	end (higher) delta level to merge (online only)
 *  raw		set to 1 if merging to base delta which is raw delta
 *  merge_top	set to 1 is top rw delta is to be merged (online only)
 *  images	list of delta files to be merged, from top to bottom
 *  new_image	new image file name to merge to (NULL to merge to lowest delta)
 *
 * Returns: 0 or SYSEXIT_* error
 */
int merge_image(const char *device, int start_level, int end_level, int raw, int merge_top,
		      char **images, const char *new_image)
{
	int last_delta = 0;
	char **names = NULL;
	struct delta_array da = {};
	struct delta odelta = {};
	int i, i_end, ret = 0;
	__u32 k;
	__u32 allocated = 0;
	__u64 cluster;
	void *data_cache = NULL;
	__u32 blocksize = 0;
	__u32 prev_blocksize = 0;
	int version = PLOOP_FMT_UNDEFINED;
	const char *merged_image;

	if (new_image && access(new_image, F_OK) == 0) {
		ploop_err(EEXIST, "Can't merge to new image %s", new_image);
		return SYSEXIT_PARAM;
	}

	if (device) {
		/* Online merge */
		if (start_level >= end_level || start_level < 0) {
			ploop_err(0, "Invalid parameters: start_level %d end_level %d",
					start_level, end_level);
			return SYSEXIT_PARAM;
		}
		if (merge_top) {
			/* top delta is in running state merged
			by means of PLOOP_IOC_MERGE */
			end_level--;
			names = ++images;
			if (end_level <= start_level)
				end_level = 0;
		} else
			names = images;

		if (!new_image)
			if ((ret = grow_lower_delta(device, merge_top,
						start_level, end_level)))
				return ret;

		if (end_level == 0) {
			int lfd;

			if (new_image) {
				/* Special case: only one delta below top one:
				 * copy it to new_image and do ploop_replace()
				 */
				ret = copy_delta(names[0], new_image);
				if (ret)
					return ret;

				ret = replace_delta(device, start_level, new_image);
				if (ret)
					goto rm_delta;
			}
			ploop_log(0, "Merging top delta");
			lfd = open(device, O_RDONLY);
			if (lfd < 0) {
				ploop_err(errno, "open dev %s", device);
				ret = SYSEXIT_DEVICE;
				goto rm_delta;
			}

			ret = ioctl_device(lfd, PLOOP_IOC_MERGE, 0);
			close(lfd);

rm_delta:
			if (ret && new_image)
				unlink(new_image);

			return ret;
		}
		last_delta = end_level - start_level;
	} else {
		last_delta = get_list_size(images) - 1;
		names = images;
	}

	if (new_image) {
		last_delta++;
		merged_image = new_image;
	}
	else {
		merged_image = names[last_delta];
	}

	init_delta_array(&da);

	for (i = 0; i < last_delta; i++) {
		// FIXME: add check for blocksize
		ret = extend_delta_array(&da, names[i],
					device ? O_RDONLY|O_DIRECT : O_RDONLY,
					device ? OD_NOFLAGS : OD_OFFLINE);
		if (ret)
			goto merge_done2;

		blocksize = da.delta_arr[i].blocksize;
		if (i != 0 && blocksize != prev_blocksize) {
			ploop_err(errno, "Wrong blocksize %s bs=%d [prev bs=%d]",
					names[i], blocksize, prev_blocksize);
			ret = SYSEXIT_PLOOPFMT;
			goto merge_done2;
		}
		prev_blocksize = blocksize;

		if (i != 0 && version != da.delta_arr[i].version) {
			ploop_err(errno, "Wrong version %s %d [prev %d]",
					names[i], da.delta_arr[i].version, version);
			ret = SYSEXIT_PLOOPFMT;
			goto merge_done2;
		}
		version = da.delta_arr[i].version;
	}
	if (blocksize == 0) {
		ploop_err(errno, "Wrong blocksize 0");
		ret = SYSEXIT_PLOOPFMT;
		goto merge_done2;

	}
	cluster = S2B(blocksize);

	if (new_image) { /* Create it */
		struct ploop_pvd_header *vh;
		off_t size;
		int mode = (raw) ? PLOOP_RAW_MODE : PLOOP_EXPANDED_MODE;

		vh = (struct ploop_pvd_header *)da.delta_arr[0].hdr0;
		size = get_SizeInSectors(vh);

		ret = create_image(new_image, blocksize, size, mode, version);
		if (ret)
			goto merge_done2;
	}

	if (!raw) {
		if (open_delta(&odelta, merged_image, O_RDWR,
			       device ? OD_NOFLAGS : OD_OFFLINE)) {
			ploop_err(errno, "open_delta");
			ret = SYSEXIT_OPEN;
			goto merge_done2;
		}
		if (dirty_delta(&odelta)) {
			ploop_err(errno, "dirty_delta");
			ret = SYSEXIT_WRITE;
			goto merge_done2;
		}
	} else {
		if (open_delta_simple(&odelta, merged_image, O_RDWR,
				      device ? 0 : OD_OFFLINE)) {
			ret = SYSEXIT_WRITE;
			goto merge_done2;
		}
	}
	if (p_memalign(&data_cache, 4096, cluster)) {
		ret = SYSEXIT_MALLOC;
		goto merge_done2;
	}

	if (!device && !new_image) {
		struct ploop_pvd_header *vh;
		vh = (struct ploop_pvd_header *)da.delta_arr[0].hdr0;

		if (!raw) {
			if ((ret = grow_delta(&odelta, get_SizeInSectors(vh),
				   data_cache, NULL)))
				goto merge_done;
		} else {
			off_t src_size = get_SizeInSectors(vh);
			off_t dst_size;

			ret = read_size_from_image(merged_image, 1, &dst_size);
			if (ret)
				goto merge_done;

			if (src_size > dst_size) {
				ret = grow_raw_delta(merged_image,
					       S2B(src_size - dst_size));
				if (ret)
					goto merge_done;
			}
		}
	}

	i_end = (da.delta_arr[0].l2_size + PLOOP_MAP_OFFSET + cluster/4 - 1) /
		(cluster/4);
	for (i = 0; i < i_end; i++) {
		int k_start = 0;
		int k_end   = cluster/4;

		/* Load L2 table */
		if (PREAD(&da.delta_arr[0], da.delta_arr[0].l2,
			  cluster, (off_t)i * cluster)) {
			ret = SYSEXIT_READ;
			goto merge_done;
		}

		/* Announce L2 cache valid. This information is not used. */
		da.delta_arr[0].l2_cache = i;

		/* And invalidate L2 cache for lower delta, they will
		 * be fetched on demand.
		 */
		for (k = 1; k < last_delta; k++)
			da.delta_arr[k].l2_cache = -1;

		/* Iterate over all L2 entries */
		if (i == 0)
			k_start = PLOOP_MAP_OFFSET;
		if (i == i_end - 1)
			k_end   = da.delta_arr[0].l2_size + PLOOP_MAP_OFFSET -
				  i * cluster/4;

		for (k = k_start; k < k_end; k++) {
			int level2 = 0;

			/* If entry is not present in base level,
			 * lookup lower deltas.
			 */
			if (da.delta_arr[0].l2[k] == 0) {
				ret = locate_l2_entry(&da, 0, i, k, &level2);
				if (ret)
					goto merge_done;
				if (level2 < 0)
					continue;
			}

			if (PREAD(&da.delta_arr[level2], data_cache, cluster,
						S2B(ploop_ioff_to_sec(da.delta_arr[level2].l2[k],
								blocksize, version)))) {
				ret = SYSEXIT_READ;
				goto merge_done;
			}

			if (raw) {
				off_t opos;
				opos = i * (cluster/4) + k - PLOOP_MAP_OFFSET;
				if (PWRITE(&odelta, data_cache, cluster,
					   opos*cluster)) {
					ret = SYSEXIT_WRITE;
					goto merge_done;
				}
				continue;
			}

			if (i != odelta.l2_cache) {
				if (odelta.l2_cache >= 0)
					if ((ret = sync_cache(&odelta)))
						goto merge_done;

				odelta.l2_cache = i;
				if (PREAD(&odelta, odelta.l2, cluster,
					  (off_t)(i * cluster))) {
					ret = SYSEXIT_READ;
					goto merge_done;
				}
				odelta.l2_dirty = 0;
			}

			if (odelta.l2[k] == 0) {
				odelta.l2[k] = ploop_sec_to_ioff((off_t)odelta.alloc_head++ * B2S(cluster),
							blocksize, version);
				if (odelta.l2[k] == 0) {
					ploop_err(0, "abort: odelta.l2[k] == 0");
					ret = SYSEXIT_ABORT;
					goto merge_done;
				}
				odelta.l2_dirty = 1;
				allocated++;
			}
			if (PWRITE(&odelta, data_cache, cluster,
						S2B(ploop_ioff_to_sec(odelta.l2[k],
								blocksize, version)))) {
				ret = SYSEXIT_WRITE;
				goto merge_done;
			}
		}
	}

	if (fsync(odelta.fd)) {
		ploop_err(errno, "fsync");
		ret = SYSEXIT_FSYNC;
		goto merge_done;
	}

	if (odelta.l2_dirty) {
		int skip = 0;

		if (odelta.l2_cache < 0) {
			ploop_err(0, "abort: odelta.l2_cache < 0");
			ret = SYSEXIT_ABORT;
			goto merge_done;
		}
		if (odelta.l2_cache >= odelta.l1_size) {
			ploop_err(0, "abort: odelta.l2_cache >= odelta.l1_size");
			ret = SYSEXIT_ABORT;
			goto merge_done;
		}

		if (odelta.l2_cache == 0)
			skip = sizeof(struct ploop_pvd_header);

		if (PWRITE(&odelta, (__u8 *)odelta.l2 + skip,
					cluster - skip,
					(off_t)odelta.l2_cache * cluster + skip)) {
			ret = SYSEXIT_WRITE;
			goto merge_done;
		}
		if (fsync(odelta.fd)) {
			ploop_err(errno, "fsync");
			ret = SYSEXIT_FSYNC;
			goto merge_done;
		}
	}

	if (!raw && clear_delta(&odelta)) {
		ploop_err(errno, "clear_delta");
		ret = SYSEXIT_WRITE;
	} else {
		if (fsync(odelta.fd)) {
			ploop_err(errno, "fsync");
			ret = SYSEXIT_FSYNC;
		}
	}

merge_done:
	close_delta(&odelta);

	if (device && !ret) {
		int lfd;
		__u32 level;

		lfd = open(device, O_RDONLY);
		if (lfd < 0) {
			ploop_err(errno, "open dev");
			ret = SYSEXIT_DEVICE;
			goto merge_done2;
		}

		if (new_image) {
			int fd;

			fd = open(new_image, O_DIRECT | O_RDONLY);
			if (fd < 0) {
				ploop_err(errno, "Can't open %s", new_image);
				ret = SYSEXIT_OPEN;
				goto close_lfd;
			}
			ret = do_replace_delta(lfd, start_level, fd, blocksize, new_image);
			close(fd);
			if (ret)
				goto close_lfd;
		}

		level = start_level + 1;

		for (i = start_level + 1; i <= end_level; i++) {
			ret = ioctl_device(lfd, PLOOP_IOC_DEL_DELTA, &level);
			if (ret) {
				close(lfd);
				goto merge_done2;
			}
		}

		if (merge_top) {
			ploop_log(0, "Merging top delta");
			ret = ioctl_device(lfd, PLOOP_IOC_MERGE, 0);
		}
close_lfd:
		close(lfd);
	}
merge_done2:
	free(data_cache);
	deinit_delta_array(&da);
	return ret;
}

/* Only call after ploop_di_can_merge() returned 0 */
int check_snapshot_mounts(struct ploop_disk_images_data *di,
		const char *ancestor_guid, const char *descendant_guid)
{
	const char *guid = descendant_guid;

	while (1) {
		int idx, temp, ret;
		const char *fname;

		idx = find_snapshot_by_guid(di, guid);
		if (idx == -1) {
			ploop_err(0, "Can't find snapshot by guid %s",
					guid);
			return SYSEXIT_DISKDESCR;
		}

		fname = find_image_by_guid(di, guid);
		if (!fname) {
			ploop_err(0, "Can't find image by guid %s",
					guid);
			return SYSEXIT_DISKDESCR;
		}

		temp = di->snapshots[idx]->temporary;

		ret = check_snapshot_mount(di, guid, fname, temp);
		if (ret)
			return ret;

		if (!guidcmp(guid, ancestor_guid))
			break;

		guid = di->snapshots[idx]->parent_guid;
		if (!strcmp(guid, NONE_UUID)) {
			/* can't happen */
			ploop_err(0, "Unexpectedly found %s to be base", guid);
			return SYSEXIT_DISKDESCR;
		}
	}

	return 0;
}

/* Logs a line showing deltas to merge and the merge destination,
 * that looks something like these:
 *
 *	Merging image: delta_file -> parent_delta_file
 *
 *	Merging images: delta_file_1 file_2 file_3 -> (new) new_delta
 */
static void log_merge_images_info(struct ploop_disk_images_data *di,
		char **names, const char *new_delta)
{
	char basedir[PATH_MAX];
	char imglist[LOG_BUF_SIZE];
	char merged_image[PATH_MAX];
	int i, pos = 0, nimg;

	get_basedir(di->runtime->xml_fname, basedir, sizeof(basedir));

	nimg = get_list_size(names) - 1;
	if (new_delta) {
		nimg++;
		// merged_image = new_delta;
		normalize_image_name(basedir, new_delta,
				merged_image, sizeof(merged_image));
	}
	else {
		// merged_image = names[nimg];
		normalize_image_name(basedir, names[nimg],
				merged_image, sizeof(merged_image));
	}

	for (i = 0; i < nimg; i++) {
		int n;
		char img[PATH_MAX];

		normalize_image_name(basedir, names[i], img, sizeof(img));
		n = snprintf(imglist + pos, sizeof(imglist) - pos, "%s ", img);
		if (n <= 0)
			// error?
			break;
		pos += n;
		if (pos >= sizeof(imglist))
			// output truncated!
			break;
	}

	ploop_log(0, "Merging image%s: %s-> %s%s",
			nimg > 1 ? "s" : "",
			imglist,
			new_delta ? "(new) " : "", merged_image);
}

int ploop_merge_snapshot_by_guid(struct ploop_disk_images_data *di,
		const char *guid, const char *new_delta)
{
	char conf[PATH_MAX-4];
	char conf_tmp[PATH_MAX];
	char dev[64];
	char *device = NULL;
	char *parent_fname = NULL;
	char *child_fname = NULL;
	char *delete_fname = NULL;
	const char *child_guid = NULL;
	const char *parent_guid = NULL;
	char *names[3] = {};
	int ret;
	int start_level = -1;
	int end_level = -1;
	int merge_top = 0;
	int raw = 0;
	int online = 0;
	int parent_idx, child_idx; /* parent and child snapshot ids */
	struct merge_info info = {};
	int i, nelem;

	ret = SYSEXIT_PARAM;

	child_guid = guid;
	parent_guid = ploop_find_parent_by_guid(di, guid);
	if (!parent_guid) {
		ploop_err(0, "Can't find parent of uuid %s "
				"(is that a base image?)", guid);
		goto err;
	}

	child_fname = find_image_by_guid(di, child_guid);
	if (child_fname == NULL) {
		ploop_err(0, "Can't find image by uuid %s",
				child_guid);
		goto err;
	}
	parent_fname = find_image_by_guid(di, parent_guid);
	if (parent_fname == NULL) {
		ploop_err(0, "Can't find image by uuid %s",
				parent_guid);
		goto err;
	}
	parent_idx = find_snapshot_by_guid(di, parent_guid);
	if (parent_idx == -1) {
		ploop_err(0, "Can't find snapshot by uuid %s",
				parent_guid);

		goto err;
	}
	child_idx = find_snapshot_by_guid(di, child_guid);
	if (child_idx == -1) {
		ploop_err(0, "Can't find snapshot by uuid %s",
				child_guid);

		goto err;
	}

	nelem = ploop_get_child_count_by_uuid(di, parent_guid);
	if (nelem > 1) {
		ploop_err(0, "Can't merge to snapshot %s: it has %d children",
				parent_guid, nelem);
		ret = SYSEXIT_PARAM;
		goto err;
	}

	ret = ploop_find_dev_by_dd(di, dev, sizeof(dev));
	if (ret == -1) {
		ret = SYSEXIT_SYS;
		goto err;
	}
	else if (ret == 0)
		online = 1;

	if (online) {
		ret = complete_running_operation(di, dev);
		if (ret)
			goto err;
		if ((ret = get_delta_info(dev, &info)))
			goto err;
		nelem = get_list_size(info.names);
		for (i = 0; info.names[i] != NULL; i++) {
			ret = ploop_fname_cmp(info.names[i], child_fname);
			if (ret == -1) {
				goto err;
			} else if (ret == 0) {
				child_fname = info.names[i];
				end_level = nelem - i - 1;
				continue;
			}
			ret = ploop_fname_cmp(info.names[i], parent_fname);
			if (ret == -1)
				goto err;
			else if (ret == 0) {
				parent_fname = info.names[i];
				start_level = nelem - i - 1;
			}
		}

		if (end_level != -1 && start_level != -1) {
			if (end_level != start_level + 1) {
				ploop_err(0, "Inconsistency detected %s [%d] %s [%d]",
						parent_fname, end_level, child_fname, start_level);
				ret = SYSEXIT_PARAM;
				goto err;
			}
			device = dev;
			if (start_level == 0)
				raw = info.raw;
			merge_top = (info.top_level == end_level);
		} else if (end_level == -1 && start_level == -1) {
			online = 0;
		} else {
			ploop_err(0, "Inconsistency detected %s [%d] %s [%d]",
					parent_fname, end_level, child_fname, start_level);
			ret = SYSEXIT_PARAM;
			goto err;
		}
	}

	if (!online) {
		start_level = 0;
		end_level = 1;
		/* Only base image could be in RAW format */
		if (di->mode == PLOOP_RAW_MODE &&
				!guidcmp(di->snapshots[parent_idx]->parent_guid, NONE_UUID))
			raw = 1;
	}

	names[0] = delete_fname = strdup(child_fname);
	if (names[0] == NULL) {
		ret = SYSEXIT_MALLOC;
		goto err;
	}
	names[1] = strdup(parent_fname);
	if (names[1] == NULL) {
		ret = SYSEXIT_MALLOC;
		goto err;
	}
	names[2] = NULL;

	ret = check_snapshot_mount(di, parent_guid, parent_fname,
			di->snapshots[parent_idx]->temporary);
	if (ret)
		goto err;
	ret = check_snapshot_mount(di, child_guid, child_fname,
			di->snapshots[child_idx]->temporary);
	if (ret)
		goto err;

	ploop_log(0, "%sline %s merge %s -> %s%s",
			online ? "On": "Off",
			get_snap_str(di->snapshots[parent_idx]->temporary),
			child_guid, parent_guid,
			raw ? " (raw)" : "");

	log_merge_images_info(di, names, new_delta);

	/* make validation before real merge */
	ret = ploop_di_merge_image(di, child_guid);
	if (ret)
		goto err;
	/* We can't use parent_guid and child_guid below, since one of them
	 * is already free()'d in ploop_di_merge_image().
	 * Hint the compiler/static analyser to error out if they are used.
	 */
	parent_guid = child_guid = NULL;

	get_disk_descriptor_fname(di, conf, sizeof(conf));
	snprintf(conf_tmp, sizeof(conf_tmp), "%s.tmp", conf);
	ret = ploop_store_diskdescriptor(conf_tmp, di);
	if (ret)
		goto err;

	ret = merge_image(device, start_level, end_level, raw, merge_top, names, new_delta);
	if (ret)
		goto err;

	if (new_delta) {
		/* Write new delta name to dd.xml, and remove the old file.
		 * Note we can only write new delta now after merge_image()
		 * as the file is created and we can use realpath() on it.
		 */
		int idx;
		char *oldimg, *newimg;

		newimg = realpath(new_delta, NULL);
		if (!newimg) {
			ploop_err(errno, "Error in realpath(%s)", new_delta);
			ret = SYSEXIT_PARAM;
			goto err;
		}

		idx = find_image_idx_by_guid(di, child_guid);
		if (idx == -1) {
			ploop_err(0, "Unable to find image by uuid %s",
					child_guid ? child_guid : "NULL");
			ret = SYSEXIT_PARAM;
			goto err;
		}

		oldimg = di->images[idx]->file;
		di->images[idx]->file = newimg;

		ret = ploop_store_diskdescriptor(conf_tmp, di);
		if (ret)
			goto err;

		ploop_log(0, "Removing %s", oldimg);
		ret = unlink(oldimg);
		if (ret) {
			ploop_err(errno, "unlink %s", oldimg);
			ret = SYSEXIT_UNLINK;
		}

		free(oldimg);
	}

	if (rename(conf_tmp, conf)) {
		ploop_err(errno, "Can't rename %s %s",
				conf_tmp, conf);
		ret = SYSEXIT_RENAME;
		goto err;
	}

	ploop_log(0, "Removing %s", delete_fname);
	if (unlink(delete_fname)) {
		ploop_err(errno, "unlink %s", delete_fname);
		ret = SYSEXIT_UNLINK;
	}

	if (ret == 0)
		ploop_log(0, "ploop snapshot has been successfully merged");
	else
		ploop_err(0, "failed to merge ploop snapshot");

err:
	for (i = 0; names[i] != NULL; i++)
		free(names[i]);

	ploop_free_array(info.names);

	return ret;
}

/* Given a device and any two deltas, figure out if a merge between these
 * is online or offline. For online case, fill in the merge parameters.
 *
 * Parameters:
 *   dev	ploop device
 *   anc_fname	ancestor (grand parent) image file name
 *   dsc_fname	descentant (grand child) image file name
 *
 * Returns:
 *   0 or SYSEXIT_* error
 *
 * Returned data:
 *   *online		1 if this is online merge and the following data:
 *   mi->start_level	start level to merge
 *   mi->end_level	end level to merge
 *   mi->raw		whether base delta is raw
 *   mi->top_level	device top level
 *   mi->merge_top	whether top level is affected
 *   mi->names		list of deltas to merge, from descendant to ancestor
 */
static int prepare_online_merge(const char *dev,
		const char *anc_fname, const char *dsc_fname,
		int *online, struct merge_info *mi)
{
	struct merge_info info = {};
	int i;
	int anc_idx = -1, dsc_idx = -1; /* indexes in info */
	int ret;

	ret = get_delta_info(dev, &info);
	if (ret)
		return ret;

	ret = SYSEXIT_FSTAT;
	/* Try to find both images among deltas in running ploop */
	for (i = 0; info.names[i] != NULL; i++) {
		int r;

		if (dsc_idx == -1) {
			r = ploop_fname_cmp(info.names[i], dsc_fname);
			if (r == -1)
				goto out;
			else if (r == 0) {
				dsc_idx = i;
				continue;
			}
		}
		if (anc_idx == -1) {
			r = ploop_fname_cmp(info.names[i], anc_fname);
			if (r == -1)
				goto out;
			else if (r == 0) {
				anc_idx = i;
				continue;
			}
		}
	}

	/* Figure out if it's online, offline, or mixed case */
	if (anc_idx != -1 && dsc_idx != -1) {
		/* found both deltas: ONLINE */
		int nelem, i, idx;

		*online = 1;
		nelem = get_list_size(info.names);
		mi->start_level = nelem - anc_idx - 1;
		mi->end_level = nelem - dsc_idx - 1;
		if (mi->end_level <= mi->start_level) {
			ploop_err(0, "Inconsistency: end_level <= start_level"
					" (%d <= %d)",
					mi->end_level, mi->start_level);
			ret = SYSEXIT_PARAM;
			goto out;
		}

		mi->names = calloc(mi->end_level - mi->start_level + 2,
				sizeof(char *));
		if (!mi->names) {
			ploop_err(errno, "Can't malloc");
			ret = SYSEXIT_MALLOC;
			goto out;
		}
		/* Copy part of names we are interested in */
		for (i = 0, idx = dsc_idx; idx <= anc_idx; idx++, i++)
			mi->names[i] = strdup(info.names[idx]);

		/* set the needed flags */
		mi->raw = (mi->start_level == 0) ? info.raw : 0;
		mi->top_level = info.top_level;
		mi->merge_top = (info.top_level == mi->end_level);

		ret = 0;
	}
	else if (anc_idx == -1 && dsc_idx == -1) {
		/* no used deltas found: OFFLINE */
		*online = 0;
		ret = 0;
	}
	else {
		/* Some deltas found: mixed case */
		ploop_err(0, "Can't do mixed merge (some deltas "
				"are online, some are offline)");
		ret = SYSEXIT_PARAM;
	}
out:
	ploop_free_array(info.names);

	return ret;
}

/* Given two snapshot guids, fill in the merge info (for offline case)
 *
 * Parameters:
 *   di			disk descriptor info
 *   ancestor_guid	ancestor (grand parent, lower) guid
 *   descendant_guid	descentant (grand child, higher) guid
 *
 * Returns:
 *   0 or SYSEXIT_* error
 *
 * Returned data:
 *   mi->raw		whether base delta is raw
 *   mi->names		list of deltas to merge, from descentant to ancestor
 */
static int prepare_offline_merge(struct ploop_disk_images_data *di,
		const char *ancestor_guid, const char *descendant_guid,
		struct merge_info *mi)
{
	int ret = 0;

	mi->names = make_images_list_by_guids(di,
			descendant_guid, ancestor_guid, 1);
	if (!mi->names)
		/* error is printed by make_images_list_by_guids */
		return SYSEXIT_SYS;

	if (di->mode == PLOOP_RAW_MODE) {
		int i;

		/* We should set raw=1 if merging down to base image */
		i = find_snapshot_by_guid(di, ancestor_guid);
		if (i == -1) { /* should not ever happen */
			ploop_err(0, "Can't find snapshot by guid %s",
					ancestor_guid);
			ret = SYSEXIT_PARAM;
			goto out;
		}
		if (strcmp(di->snapshots[i]->parent_guid, NONE_UUID) == 0)
			mi->raw = 1;
	}

out:
	if (ret && mi->names)
		ploop_free_array(mi->names);

	return ret;
}

/* Merge multiple snapshots by a range of GUIDs */
int ploop_merge_snapshots_by_guids(struct ploop_disk_images_data *di,
		const char *descendant_guid, const char *ancestor_guid,
		const char *new_delta)
{
	int i, ret;
	int online;
	char dev[64] = "";
	struct merge_info info = {};
	char conf[PATH_MAX-4];
	char conf_tmp[PATH_MAX];

	/* Can we merge these? */
	ret = ploop_di_can_merge_images(di, ancestor_guid, descendant_guid);
	if (ret)
		return ret;

	/* Is it online or offline merge? */
	ret = ploop_find_dev_by_dd(di, dev, sizeof(dev));
	if (ret == -1)
		return SYSEXIT_SYS;
	online = !ret;
	if (online) {
		const char *anc_fname, *dsc_fname;

		anc_fname = find_image_by_guid(di, ancestor_guid);
		dsc_fname = find_image_by_guid(di, descendant_guid);
		/* Can be online, offline, or mixed */
		ret = prepare_online_merge(dev, anc_fname, dsc_fname,
				&online, &info);
		if (ret)
			return ret;
	}

	if (!online) {
		ret = prepare_offline_merge(di,
				ancestor_guid, descendant_guid, &info);
		if (ret)
			goto out;
	}

	/* FIXME: check if any snapshots are mounted */
	ret = check_snapshot_mounts(di,
			ancestor_guid, descendant_guid);
	if (ret)
		goto out;

	i = get_list_size(info.names);
	ploop_log(0, "%s merge of %d snapshot%s",
			online ? "Online" : "Offline",
			i - 1, i < 3 ? "" : "s");
	log_merge_images_info(di, info.names, new_delta);

	/* First, merge in dd.xml (and do some extra checks */
	ret = ploop_di_merge_images(di, descendant_guid, i - 1);
	if (ret)
		goto out;

	get_disk_descriptor_fname(di, conf, sizeof(conf));
	snprintf(conf_tmp, sizeof(conf_tmp), "%s.tmp", conf);
	ret = ploop_store_diskdescriptor(conf_tmp, di);
	if (ret)
		goto out;

	/* Do the actual merge */
	ret = merge_image(online ? dev : NULL,
			info.start_level, info.end_level,
			info.raw, info.merge_top, info.names, new_delta);
	if (ret)
		goto out;

	if (new_delta) {
		/* Write new delta name to dd.xml, and remove the old file.
		 * Note we can only write new delta now after merge_image()
		 * as the file is created and we can use realpath() on it.
		 */
		int idx;
		char *oldimg, *newimg;

		newimg = realpath(new_delta, NULL);
		if (!newimg) {
			ploop_err(errno, "Error in realpath(%s)", new_delta);
			ret = SYSEXIT_PARAM;
			goto out;
		}

		idx = find_image_idx_by_guid(di, descendant_guid);
		if (idx == -1) {
			ploop_err(0, "Can't find image by uuid %s",
					descendant_guid);
			ret = SYSEXIT_PARAM;
			goto out;
		}

		oldimg = di->images[idx]->file;
		di->images[idx]->file = newimg;

		ret = ploop_store_diskdescriptor(conf_tmp, di);
		if (ret)
			goto out;

		ploop_log(0, "Removing %s", oldimg);
		ret = unlink(oldimg);
		if (ret) {
			ploop_err(errno, "Can't remove %s", oldimg);
			ret = SYSEXIT_UNLINK;
		}

		free(oldimg);
	}

	/* Put a new dd.xml */
	if (rename(conf_tmp, conf)) {
		ploop_err(errno, "Can't rename %s %s", conf_tmp, conf);
		ret = SYSEXIT_RENAME;
		goto out;
	}

	/* Remove images which were merged, i.e. all but the lowest one */
	for (i = 0; info.names[i+1]; i++) {
		const char *file = info.names[i];

		ploop_log(0, "Removing %s", file);
		if (unlink(file)) {
			ploop_err(errno, "Can't remove %s", file);
			ret = SYSEXIT_UNLINK;
		}
	}

	if (ret == 0)
		ploop_log(0, "Snapshots successfully merged");

out:
	ploop_free_array(info.names);

	return ret;
}

int ploop_merge_snapshot(struct ploop_disk_images_data *di, struct ploop_merge_param *param)
{
	int ret = SYSEXIT_PARAM;
	const char *guid = NULL;

	if (ploop_lock_dd(di))
		return SYSEXIT_LOCK;

	if (param->guid2 != NULL) {
		/* merge multiple snapshots at once */
		ret = ploop_merge_snapshots_by_guids(di, param->guid, param->guid2, param->new_delta);
		goto unlock;
	}
	else if (param->guid != NULL)
		guid = param->guid;
	else if (!param->merge_all)
		guid = di->top_guid;

	if (guid != NULL) {
		ret = ploop_merge_snapshot_by_guid(di, guid, param->new_delta);
	} else {
		/* merge all from top to base */
		guid = get_base_delta_uuid(di);
		ret =  ploop_merge_snapshots_by_guids(di, di->top_guid, guid, param->new_delta);
	}
unlock:
	ploop_unlock_dd(di);

	return ret;
}