btrfs-progs/cmds-send.c

770 lines
17 KiB
C
Raw Normal View History

/*
* Copyright (C) 2012 Alexander Block. 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 v2 as published by the Free Software Foundation.
*
* 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 021110-1307, USA.
*/
#include "kerncompat.h"
#include <unistd.h>
#include <stdint.h>
#include <dirent.h>
#include <fcntl.h>
#include <pthread.h>
#include <math.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <libgen.h>
#include <mntent.h>
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
#include <assert.h>
#include <getopt.h>
#include <uuid/uuid.h>
#include <limits.h>
#include "ctree.h"
#include "ioctl.h"
#include "commands.h"
#include "list.h"
#include "utils.h"
#include "send.h"
#include "send-utils.h"
static int g_verbose = 0;
struct btrfs_send {
int send_fd;
int dump_fd;
int mnt_fd;
u64 *clone_sources;
u64 clone_sources_count;
char *root_path;
struct subvol_uuid_search sus;
};
static int get_root_id(struct btrfs_send *s, const char *path, u64 *root_id)
{
struct subvol_info *si;
si = subvol_uuid_search(&s->sus, 0, NULL, 0, path,
subvol_search_by_path);
if (!si)
return -ENOENT;
*root_id = si->root_id;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
free(si->path);
free(si);
return 0;
}
static struct subvol_info *get_parent(struct btrfs_send *s, u64 root_id)
{
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
struct subvol_info *si_tmp;
struct subvol_info *si;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
si_tmp = subvol_uuid_search(&s->sus, root_id, NULL, 0, NULL,
subvol_search_by_root_id);
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
if (!si_tmp)
return NULL;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
si = subvol_uuid_search(&s->sus, 0, si_tmp->parent_uuid, 0, NULL,
subvol_search_by_uuid);
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
free(si_tmp->path);
free(si_tmp);
return si;
}
static int find_good_parent(struct btrfs_send *s, u64 root_id, u64 *found)
{
int ret;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
struct subvol_info *parent = NULL;
struct subvol_info *parent2 = NULL;
struct subvol_info *best_parent = NULL;
__s64 tmp;
u64 best_diff = (u64)-1;
int i;
parent = get_parent(s, root_id);
if (!parent) {
ret = -ENOENT;
goto out;
}
for (i = 0; i < s->clone_sources_count; i++) {
if (s->clone_sources[i] == parent->root_id) {
best_parent = parent;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
parent = NULL;
goto out_found;
}
}
for (i = 0; i < s->clone_sources_count; i++) {
parent2 = get_parent(s, s->clone_sources[i]);
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
if (!parent2)
continue;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
if (parent2->root_id != parent->root_id) {
free(parent2->path);
free(parent2);
parent2 = NULL;
continue;
}
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
free(parent2->path);
free(parent2);
parent2 = subvol_uuid_search(&s->sus, s->clone_sources[i], NULL,
0, NULL, subvol_search_by_root_id);
if (!parent2) {
ret = -ENOENT;
goto out;
}
tmp = parent2->ctransid - parent->ctransid;
if (tmp < 0)
tmp *= -1;
if (tmp < best_diff) {
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
if (best_parent) {
free(best_parent->path);
free(best_parent);
}
best_parent = parent2;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
parent2 = NULL;
best_diff = tmp;
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
} else {
free(parent2->path);
free(parent2);
parent2 = NULL;
}
}
if (!best_parent) {
ret = -ENOENT;
goto out;
}
out_found:
*found = best_parent->root_id;
ret = 0;
out:
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
if (parent) {
free(parent->path);
free(parent);
}
if (best_parent) {
free(best_parent->path);
free(best_parent);
}
return ret;
}
static int add_clone_source(struct btrfs_send *s, u64 root_id)
{
void *tmp;
tmp = s->clone_sources;
s->clone_sources = realloc(s->clone_sources,
sizeof(*s->clone_sources) * (s->clone_sources_count + 1));
if (!s->clone_sources) {
free(tmp);
return -ENOMEM;
}
s->clone_sources[s->clone_sources_count++] = root_id;
return 0;
}
static int write_buf(int fd, const void *buf, int size)
{
int ret;
int pos = 0;
while (pos < size) {
ret = write(fd, (char*)buf + pos, size - pos);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to dump stream. %s\n",
strerror(-ret));
goto out;
}
if (!ret) {
ret = -EIO;
fprintf(stderr, "ERROR: failed to dump stream. %s\n",
strerror(-ret));
goto out;
}
pos += ret;
}
ret = 0;
out:
return ret;
}
static void *dump_thread(void *arg_)
{
int ret;
struct btrfs_send *s = (struct btrfs_send*)arg_;
char buf[4096];
int readed;
while (1) {
readed = read(s->send_fd, buf, sizeof(buf));
if (readed < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to read stream from "
"kernel. %s\n", strerror(-ret));
goto out;
}
if (!readed) {
ret = 0;
goto out;
}
ret = write_buf(s->dump_fd, buf, readed);
if (ret < 0)
goto out;
}
out:
if (ret < 0) {
exit(-ret);
}
return ERR_PTR(ret);
}
static int do_send(struct btrfs_send *send, u64 parent_root_id,
int is_first_subvol, int is_last_subvol, char *subvol,
u64 flags)
{
int ret;
pthread_t t_read;
struct btrfs_ioctl_send_args io_send;
void *t_err = NULL;
int subvol_fd = -1;
int pipefd[2] = {-1, -1};
subvol_fd = openat(send->mnt_fd, subvol, O_RDONLY | O_NOATIME);
if (subvol_fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: open %s failed. %s\n", subvol,
strerror(-ret));
goto out;
}
ret = pipe(pipefd);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: pipe failed. %s\n", strerror(-ret));
goto out;
}
memset(&io_send, 0, sizeof(io_send));
io_send.send_fd = pipefd[1];
send->send_fd = pipefd[0];
if (!ret)
ret = pthread_create(&t_read, NULL, dump_thread,
send);
if (ret) {
ret = -ret;
fprintf(stderr, "ERROR: thread setup failed: %s\n",
strerror(-ret));
goto out;
}
io_send.flags = flags;
io_send.clone_sources = (__u64*)send->clone_sources;
io_send.clone_sources_count = send->clone_sources_count;
io_send.parent_root = parent_root_id;
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
if (!is_first_subvol)
io_send.flags |= BTRFS_SEND_FLAG_OMIT_STREAM_HEADER;
if (!is_last_subvol)
io_send.flags |= BTRFS_SEND_FLAG_OMIT_END_CMD;
ret = ioctl(subvol_fd, BTRFS_IOC_SEND, &io_send);
if (ret) {
ret = -errno;
fprintf(stderr, "ERROR: send ioctl failed with %d: %s\n", ret,
strerror(-ret));
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
if (ret == -EINVAL && (!is_first_subvol || !is_last_subvol))
fprintf(stderr,
"Try upgrading your kernel or don't use -e.\n");
goto out;
}
if (g_verbose > 0)
fprintf(stderr, "BTRFS_IOC_SEND returned %d\n", ret);
if (g_verbose > 0)
fprintf(stderr, "joining genl thread\n");
close(pipefd[1]);
Btrfs-progs: use UUID tree for send/receive This commit changes the btrfs send/receive commands to use the UUID tree to map UUIDs to subvolumes, and to use the root tree to map subvolume IDs to paths. Now these tools start fast and are independent on the number of subvolules/snapshot that exist. Before this commit, mapping UUIDs to subvolume IDs was an operation with a high effort. The algorithm even had quadratic effort (based on the number of existing subvolumes). E.g. with 15,000 subvolumes it took much more than 5 minutes on a state of the art XEON CPU to start btrfs send or receive before these tools were able to send or receive the first byte). Even linear effort instead of the current quadratic effort would be too much since it would be a waste. And these data structures to allow mapping UUIDs to subvolume IDs had been created every time a btrfs send/receive instance was started. It is much more efficient to maintain a searchable persistent data structure in the filesystem, one that is updated whenever a subvolume/snapshot is created and deleted, and when the received subvolume UUID is set by the btrfs-receive tool. Therefore kernel code was added that is able to maintain data structures in the filesystem that allow to quickly search for a given UUID and to retrieve data that is assigned to this UUID, like which subvolume ID is related to this UUID. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-06-26 15:17:57 +00:00
pipefd[1] = -1;
ret = pthread_join(t_read, &t_err);
if (ret) {
ret = -ret;
fprintf(stderr, "ERROR: pthread_join failed: %s\n",
strerror(-ret));
goto out;
}
if (t_err) {
ret = (long int)t_err;
fprintf(stderr, "ERROR: failed to process send stream, ret=%ld "
"(%s)\n", (long int)t_err, strerror(-ret));
goto out;
}
ret = 0;
out:
if (subvol_fd != -1)
close(subvol_fd);
if (pipefd[0] != -1)
close(pipefd[0]);
if (pipefd[1] != -1)
close(pipefd[1]);
return ret;
}
char *get_subvol_name(char *mnt, char *full_path)
{
int len = strlen(mnt);
if (!len)
return full_path;
if (mnt[len - 1] != '/')
len += 1;
return full_path + len;
}
static int init_root_path(struct btrfs_send *s, const char *subvol)
{
int ret = 0;
if (s->root_path)
goto out;
ret = find_mount_root(subvol, &s->root_path);
if (ret < 0) {
fprintf(stderr,
"ERROR: failed to determine mount point for %s: %s\n",
subvol, strerror(-ret));
ret = -EINVAL;
goto out;
}
if (ret > 0) {
fprintf(stderr,
"ERROR: %s doesn't belong to btrfs mount point\n",
subvol);
ret = -EINVAL;
goto out;
}
s->mnt_fd = open(s->root_path, O_RDONLY | O_NOATIME);
if (s->mnt_fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: can't open '%s': %s\n", s->root_path,
strerror(-ret));
goto out;
}
ret = subvol_uuid_search_init(s->mnt_fd, &s->sus);
if (ret < 0) {
fprintf(stderr, "ERROR: failed to initialize subvol search. "
"%s\n", strerror(-ret));
goto out;
}
out:
return ret;
}
static int is_subvol_ro(struct btrfs_send *s, char *subvol)
{
int ret;
u64 flags;
int fd = -1;
fd = openat(s->mnt_fd, subvol, O_RDONLY | O_NOATIME);
if (fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to open %s. %s\n",
subvol, strerror(-ret));
goto out;
}
ret = ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to get flags for subvolume. "
"%s\n", strerror(-ret));
goto out;
}
if (flags & BTRFS_SUBVOL_RDONLY)
ret = 1;
else
ret = 0;
out:
if (fd != -1)
close(fd);
return ret;
}
int cmd_send(int argc, char **argv)
{
char *subvol = NULL;
int ret;
char outname[PATH_MAX];
struct btrfs_send send;
u32 i;
char *mount_root = NULL;
char *snapshot_parent = NULL;
u64 root_id = 0;
u64 parent_root_id = 0;
int full_send = 1;
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
int new_end_cmd_semantic = 0;
u64 send_flags = 0;
memset(&send, 0, sizeof(send));
send.dump_fd = fileno(stdout);
outname[0] = 0;
while (1) {
enum { GETOPT_VAL_SEND_NO_DATA = 256 };
static const struct option long_options[] = {
{ "no-data", no_argument, NULL, GETOPT_VAL_SEND_NO_DATA }
};
int c = getopt_long(argc, argv, "vec:f:i:p:", long_options, NULL);
if (c < 0)
break;
switch (c) {
case 'v':
g_verbose++;
break;
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
case 'e':
new_end_cmd_semantic = 1;
break;
case 'c':
subvol = realpath(optarg, NULL);
if (!subvol) {
ret = -errno;
fprintf(stderr, "ERROR: realpath %s failed. "
"%s\n", optarg, strerror(-ret));
goto out;
}
ret = init_root_path(&send, subvol);
if (ret < 0)
goto out;
ret = get_root_id(&send, get_subvol_name(send.root_path, subvol),
&root_id);
if (ret < 0) {
fprintf(stderr, "ERROR: could not resolve "
"root_id for %s\n", subvol);
goto out;
}
ret = is_subvol_ro(&send, subvol);
if (ret < 0)
goto out;
if (!ret) {
ret = -EINVAL;
fprintf(stderr,
"ERROR: cloned subvol %s is not read-only.\n",
subvol);
goto out;
}
ret = add_clone_source(&send, root_id);
if (ret < 0) {
fprintf(stderr, "ERROR: not enough memory\n");
goto out;
}
subvol_uuid_search_finit(&send.sus);
free(subvol);
subvol = NULL;
if (send.mnt_fd >= 0) {
close(send.mnt_fd);
send.mnt_fd = -1;
}
free(send.root_path);
send.root_path = NULL;
full_send = 0;
break;
case 'f':
if (arg_copy_path(outname, optarg, sizeof(outname))) {
fprintf(stderr,
"ERROR: output file path too long (%zu)\n",
strlen(optarg));
ret = 1;
goto out;
}
break;
case 'p':
if (snapshot_parent) {
fprintf(stderr, "ERROR: you cannot have more than one parent (-p)\n");
ret = 1;
goto out;
}
snapshot_parent = realpath(optarg, NULL);
if (!snapshot_parent) {
ret = -errno;
fprintf(stderr, "ERROR: realpath %s failed. "
"%s\n", optarg, strerror(-ret));
goto out;
}
ret = is_subvol_ro(&send, snapshot_parent);
if (ret < 0)
goto out;
if (!ret) {
ret = -EINVAL;
fprintf(stderr,
"ERROR: parent %s is not read-only.\n",
snapshot_parent);
goto out;
}
full_send = 0;
break;
case 'i':
fprintf(stderr,
"ERROR: -i was removed, use -c instead\n");
ret = 1;
goto out;
case GETOPT_VAL_SEND_NO_DATA:
send_flags |= BTRFS_SEND_FLAG_NO_FILE_DATA;
break;
case '?':
default:
fprintf(stderr, "ERROR: send args invalid.\n");
ret = 1;
goto out;
}
}
if (check_argc_min(argc - optind, 1))
usage(cmd_send_usage);
if (outname[0]) {
send.dump_fd = creat(outname, 0600);
if (send.dump_fd == -1) {
ret = -errno;
fprintf(stderr, "ERROR: can't create '%s': %s\n",
outname, strerror(-ret));
goto out;
}
}
if (isatty(send.dump_fd)) {
fprintf(stderr,
"ERROR: not dumping send stream into a terminal, "
"redirect it into a file\n");
ret = 1;
goto out;
}
/* use first send subvol to determine mount_root */
subvol = argv[optind];
subvol = realpath(argv[optind], NULL);
if (!subvol) {
ret = -errno;
fprintf(stderr, "ERROR: unable to resolve %s\n", argv[optind]);
goto out;
}
ret = init_root_path(&send, subvol);
if (ret < 0)
goto out;
if (snapshot_parent != NULL) {
ret = get_root_id(&send,
get_subvol_name(send.root_path, snapshot_parent),
&parent_root_id);
if (ret < 0) {
fprintf(stderr, "ERROR: could not resolve root_id "
"for %s\n", snapshot_parent);
goto out;
}
ret = add_clone_source(&send, parent_root_id);
if (ret < 0) {
fprintf(stderr, "ERROR: not enough memory\n");
goto out;
}
}
for (i = optind; i < argc; i++) {
free(subvol);
subvol = realpath(argv[i], NULL);
if (!subvol) {
ret = -errno;
fprintf(stderr, "ERROR: unable to resolve %s\n", argv[i]);
goto out;
}
ret = find_mount_root(subvol, &mount_root);
if (ret < 0) {
fprintf(stderr, "ERROR: find_mount_root failed on %s: "
"%s\n", subvol,
strerror(-ret));
goto out;
}
if (ret > 0) {
fprintf(stderr,
"ERROR: %s doesn't belong to btrfs mount point\n",
subvol);
ret = -EINVAL;
goto out;
}
if (strcmp(send.root_path, mount_root) != 0) {
ret = -EINVAL;
fprintf(stderr, "ERROR: all subvols must be from the "
"same fs.\n");
goto out;
}
free(mount_root);
ret = is_subvol_ro(&send, subvol);
if (ret < 0)
goto out;
if (!ret) {
ret = -EINVAL;
fprintf(stderr, "ERROR: %s is not read-only.\n",
subvol);
goto out;
}
}
if (send_flags & BTRFS_SEND_FLAG_NO_FILE_DATA)
printf("Mode NO_FILE_DATA enabled\n");
for (i = optind; i < argc; i++) {
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
int is_first_subvol;
int is_last_subvol;
free(subvol);
subvol = argv[i];
fprintf(stderr, "At subvol %s\n", subvol);
subvol = realpath(subvol, NULL);
if (!subvol) {
ret = -errno;
fprintf(stderr, "ERROR: realpath %s failed. "
"%s\n", argv[i], strerror(-ret));
goto out;
}
if (!full_send && !parent_root_id) {
ret = find_good_parent(&send, root_id, &parent_root_id);
if (ret < 0) {
fprintf(stderr, "ERROR: parent determination failed for %lld\n",
root_id);
goto out;
}
}
ret = is_subvol_ro(&send, subvol);
if (ret < 0)
goto out;
if (!ret) {
ret = -EINVAL;
fprintf(stderr, "ERROR: %s is not read-only.\n",
subvol);
goto out;
}
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
if (new_end_cmd_semantic) {
/* require new kernel */
is_first_subvol = (i == optind);
is_last_subvol = (i == argc - 1);
} else {
/* be compatible to old and new kernel */
is_first_subvol = 1;
is_last_subvol = 1;
}
ret = do_send(&send, parent_root_id, is_first_subvol,
is_last_subvol, subvol, send_flags);
if (ret < 0)
goto out;
/* done with this subvol, so add it to the clone sources */
ret = add_clone_source(&send, root_id);
if (ret < 0) {
fprintf(stderr, "ERROR: not enough memory\n");
goto out;
}
parent_root_id = 0;
full_send = 0;
}
ret = 0;
out:
free(subvol);
free(snapshot_parent);
free(send.clone_sources);
if (send.mnt_fd >= 0)
close(send.mnt_fd);
free(send.root_path);
subvol_uuid_search_finit(&send.sus);
return !!ret;
}
const char * const cmd_send_usage[] = {
"btrfs send [-ve] [-p <parent>] [-c <clone-src>] [-f <outfile>] <subvol> [<subvol>...]",
"Send the subvolume(s) to stdout.",
"Sends the subvolume(s) specified by <subvol> to stdout.",
"By default, this will send the whole subvolume. To do an incremental",
"send, use '-p <parent>'. If you want to allow btrfs to clone from",
"any additional local snapshots, use '-c <clone-src>' (multiple times",
"where applicable). You must not specify clone sources unless you",
"guarantee that these snapshots are exactly in the same state on both",
"sides, the sender and the receiver. It is allowed to omit the",
"'-p <parent>' option when '-c <clone-src>' options are given, in",
"which case 'btrfs send' will determine a suitable parent among the",
"clone sources itself.",
"\n",
"-v Enable verbose debug output. Each occurrence of",
" this option increases the verbose level more.",
Btrfs-progs: add send option for using new end-cmd semantic This commit adds a command line option to enable sending streams which make use of the new end-cmd semantic if multiple snapshots are sent back-to-back. The goal is to use the <end cmd> as an indication to stop reading the input stream. So far, the receiver could only use EOF to recognize the end. If the new command line option '-e' is set, this commit requires a kernel which is able to support the new flags in the send ioctl. New bits in the flags of the send ioctl will be set which cause EINVAL on old kernels. However, if the option '-e' is not set, it works with old and new kernels without any errors or any changed behavior. This used to be the encoding (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF The new format (if the two new flags are used) is this one: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> Note that the currently existing receivers treat <end cmd> only as an indication that a new <stream header> is following. This means, you can just skip the sequence <end cmd> <stream header> without loosing compatibility. As long as an EOF is following, the currently existing receivers handle the new format (if the two new flags are used) exactly as the old one. Also note that the kernel interface was changed in a way that is backward compatible to old btrfs-progs tools. You set one or two bits in the flags field of the ioctl to enable the new behavior. Old tools set these flags to zero, thus getting exactly the same as they got with older kernels. And this is exactly what happens if the new '-e' option is not set, the new bits in the flags are not set and thus old kernels and new kernels are both supported. So what is the benefit of this change? The goal is to be able to use a single stream (one TCP connection) to multiplex a request/response handshake plus Btrfs send streams, all in the same stream. In this case you cannot evaluate an EOF condition as an end of the Btrfs send stream. You need something else, and the <end cmd> is just perfect for this purpose. Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de> Signed-off-by: David Sterba <dsterba@suse.cz> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
2013-04-10 17:10:56 +00:00
"-e If sending multiple subvols at once, use the new",
" format and omit the end-cmd between the subvols.",
"-p <parent> Send an incremental stream from <parent> to",
" <subvol>.",
"-c <clone-src> Use this snapshot as a clone source for an ",
" incremental send (multiple allowed)",
"-f <outfile> Output is normally written to stdout. To write to",
" a file, use this option. An alternative would be to",
" use pipes.",
"--no-data send in NO_FILE_DATA mode, Note: the output stream",
" does not contain any file data and thus cannot be used",
" to transfer changes. This mode is faster and useful to",
" show the differences in metadata.",
NULL
};