btrfs-progs/cmds-receive.c

1037 lines
24 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 <ftw.h>
#include <wait.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 <sys/stat.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/xattr.h>
#include <uuid/uuid.h>
#include "ctree.h"
#include "ioctl.h"
#include "commands.h"
#include "utils.h"
#include "list.h"
#include "btrfs-list.h"
#include "send.h"
#include "send-stream.h"
#include "send-utils.h"
static int g_verbose = 0;
struct btrfs_receive
{
int mnt_fd;
int dest_dir_fd;
int write_fd;
char *write_path;
char *root_path;
char *dest_dir_path; /* relative to root_path */
char *full_subvol_path;
int dest_dir_chroot;
struct subvol_info *cur_subvol;
struct subvol_uuid_search sus;
Btrfs-progs: btrfs-receive optionally honors the end-cmd A new option is added to btrfs-receive to change the behavior when an <end cmd> is received in the Btrfs send stream. The traditional behavior (which still is the default) is to continue to read the stream until an EOF condition is encountered. If an <end cmd> is received, afterwards either an EOF or a new <stream header> is expected. The new behavior (if the -e option is set on the command line) is to terminate after an <end cmd> is read without the need for an EOF. This allows the stream (e.g. a single TCP stream) to carry additional data or even multiple Btrfs send streams. Old btrfs-send tools used to encode multiple snapshots like this (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF If the new -e option is set, the expected format is like this: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> The btrfs-send tool is changed in a seperate commit to always use the new format, i.e. to send an <end cmd> only at the end. 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. The goal of changing the semantic of <end cmd> 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>
2013-04-09 17:08:40 +00:00
int honor_end_cmd;
};
static int finish_subvol(struct btrfs_receive *r)
{
int ret;
int subvol_fd = -1;
struct btrfs_ioctl_received_subvol_args rs_args;
char uuid_str[BTRFS_UUID_UNPARSED_SIZE];
u64 flags;
if (r->cur_subvol == NULL)
return 0;
subvol_fd = openat(r->mnt_fd, r->cur_subvol->path,
O_RDONLY | O_NOATIME);
if (subvol_fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: open %s failed. %s\n",
r->cur_subvol->path, strerror(-ret));
goto out;
}
memset(&rs_args, 0, sizeof(rs_args));
memcpy(rs_args.uuid, r->cur_subvol->received_uuid, BTRFS_UUID_SIZE);
rs_args.stransid = r->cur_subvol->stransid;
if (g_verbose >= 1) {
uuid_unparse((u8*)rs_args.uuid, uuid_str);
fprintf(stderr, "BTRFS_IOC_SET_RECEIVED_SUBVOL uuid=%s, "
"stransid=%llu\n", uuid_str, rs_args.stransid);
}
ret = ioctl(subvol_fd, BTRFS_IOC_SET_RECEIVED_SUBVOL, &rs_args);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: BTRFS_IOC_SET_RECEIVED_SUBVOL failed. %s\n",
strerror(-ret));
goto out;
}
r->cur_subvol->rtransid = rs_args.rtransid;
ret = ioctl(subvol_fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: BTRFS_IOC_SUBVOL_GETFLAGS failed. %s\n",
strerror(-ret));
goto out;
}
flags |= BTRFS_SUBVOL_RDONLY;
ret = ioctl(subvol_fd, BTRFS_IOC_SUBVOL_SETFLAGS, &flags);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to make subvolume read only. "
"%s\n", strerror(-ret));
goto out;
}
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 (r->cur_subvol) {
free(r->cur_subvol->path);
free(r->cur_subvol);
r->cur_subvol = NULL;
}
if (subvol_fd != -1)
close(subvol_fd);
return ret;
}
static int process_subvol(const char *path, const u8 *uuid, u64 ctransid,
void *user)
{
int ret;
struct btrfs_receive *r = user;
struct btrfs_ioctl_vol_args args_v1;
char uuid_str[BTRFS_UUID_UNPARSED_SIZE];
ret = finish_subvol(r);
if (ret < 0)
goto out;
r->cur_subvol = calloc(1, sizeof(*r->cur_subvol));
if (strlen(r->dest_dir_path) == 0)
r->cur_subvol->path = strdup(path);
else
r->cur_subvol->path = path_cat(r->dest_dir_path, path);
free(r->full_subvol_path);
r->full_subvol_path = path_cat3(r->root_path, r->dest_dir_path, path);
fprintf(stderr, "At subvol %s\n", path);
memcpy(r->cur_subvol->received_uuid, uuid, BTRFS_UUID_SIZE);
r->cur_subvol->stransid = ctransid;
if (g_verbose) {
uuid_unparse((u8*)r->cur_subvol->received_uuid, uuid_str);
fprintf(stderr, "receiving subvol %s uuid=%s, stransid=%llu\n",
path, uuid_str,
r->cur_subvol->stransid);
}
memset(&args_v1, 0, sizeof(args_v1));
strncpy_null(args_v1.name, path);
ret = ioctl(r->dest_dir_fd, BTRFS_IOC_SUBVOL_CREATE, &args_v1);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: creating subvolume %s failed. "
"%s\n", path, strerror(-ret));
goto out;
}
out:
return ret;
}
static int process_snapshot(const char *path, const u8 *uuid, u64 ctransid,
const u8 *parent_uuid, u64 parent_ctransid,
void *user)
{
int ret;
struct btrfs_receive *r = user;
char uuid_str[BTRFS_UUID_UNPARSED_SIZE];
struct btrfs_ioctl_vol_args_v2 args_v2;
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_subvol = NULL;
ret = finish_subvol(r);
if (ret < 0)
goto out;
r->cur_subvol = calloc(1, sizeof(*r->cur_subvol));
if (strlen(r->dest_dir_path) == 0)
r->cur_subvol->path = strdup(path);
else
r->cur_subvol->path = path_cat(r->dest_dir_path, path);
free(r->full_subvol_path);
r->full_subvol_path = path_cat3(r->root_path, r->dest_dir_path, path);
fprintf(stdout, "At snapshot %s\n", path);
memcpy(r->cur_subvol->received_uuid, uuid, BTRFS_UUID_SIZE);
r->cur_subvol->stransid = ctransid;
if (g_verbose) {
uuid_unparse((u8*)r->cur_subvol->received_uuid, uuid_str);
fprintf(stderr, "receiving snapshot %s uuid=%s, "
"ctransid=%llu ", path, uuid_str,
r->cur_subvol->stransid);
uuid_unparse(parent_uuid, uuid_str);
fprintf(stderr, "parent_uuid=%s, parent_ctransid=%llu\n",
uuid_str, parent_ctransid);
}
memset(&args_v2, 0, sizeof(args_v2));
strncpy_null(args_v2.name, path);
parent_subvol = subvol_uuid_search(&r->sus, 0, parent_uuid,
parent_ctransid, NULL, subvol_search_by_received_uuid);
if (!parent_subvol) {
parent_subvol = subvol_uuid_search(&r->sus, 0, parent_uuid,
parent_ctransid, NULL, subvol_search_by_uuid);
}
if (!parent_subvol) {
ret = -ENOENT;
fprintf(stderr, "ERROR: could not find parent subvolume\n");
goto out;
}
/*if (rs_args.ctransid > rs_args.rtransid) {
if (!r->force) {
ret = -EINVAL;
fprintf(stderr, "ERROR: subvolume %s was modified after it was received.\n", r->subvol_parent_name);
goto out;
} else {
fprintf(stderr, "WARNING: subvolume %s was modified after it was received.\n", r->subvol_parent_name);
}
}*/
args_v2.fd = openat(r->mnt_fd, parent_subvol->path,
O_RDONLY | O_NOATIME);
if (args_v2.fd < 0) {
ret = -errno;
if (errno != ENOENT)
fprintf(stderr, "ERROR: open %s failed. %s\n",
parent_subvol->path, strerror(-ret));
else
fprintf(stderr,
"It seems that you have changed your default "
"subvolume or you specify other subvolume to\n"
"mount btrfs, try to remount this btrfs filesystem "
"with fs tree, and run btrfs receive again!\n");
goto out;
}
ret = ioctl(r->dest_dir_fd, BTRFS_IOC_SNAP_CREATE_V2, &args_v2);
close(args_v2.fd);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: creating snapshot %s -> %s "
"failed. %s\n", parent_subvol->path,
path, strerror(-ret));
goto out;
}
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_subvol) {
free(parent_subvol->path);
free(parent_subvol);
}
return ret;
}
static int process_mkfile(const char *path, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "mkfile %s\n", path);
ret = creat(full_path, 0600);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: mkfile %s failed. %s\n", path,
strerror(-ret));
goto out;
}
close(ret);
ret = 0;
out:
free(full_path);
return ret;
}
static int process_mkdir(const char *path, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "mkdir %s\n", path);
ret = mkdir(full_path, 0700);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: mkdir %s failed. %s\n", path,
strerror(-ret));
}
free(full_path);
return ret;
}
static int process_mknod(const char *path, u64 mode, u64 dev, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "mknod %s mode=%llu, dev=%llu\n",
path, mode, dev);
ret = mknod(full_path, mode & S_IFMT, dev);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: mknod %s failed. %s\n", path,
strerror(-ret));
}
free(full_path);
return ret;
}
static int process_mkfifo(const char *path, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "mkfifo %s\n", path);
ret = mkfifo(full_path, 0600);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: mkfifo %s failed. %s\n", path,
strerror(-ret));
}
free(full_path);
return ret;
}
static int process_mksock(const char *path, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "mksock %s\n", path);
ret = mknod(full_path, 0600 | S_IFSOCK, 0);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: mknod %s failed. %s\n", path,
strerror(-ret));
}
free(full_path);
return ret;
}
static int process_symlink(const char *path, const char *lnk, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "symlink %s -> %s\n", path, lnk);
ret = symlink(lnk, full_path);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: symlink %s -> %s failed. %s\n", path,
lnk, strerror(-ret));
}
free(full_path);
return ret;
}
static int process_rename(const char *from, const char *to, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_from = path_cat(r->full_subvol_path, from);
char *full_to = path_cat(r->full_subvol_path, to);
if (g_verbose >= 2)
fprintf(stderr, "rename %s -> %s\n", from, to);
ret = rename(full_from, full_to);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: rename %s -> %s failed. %s\n", from,
to, strerror(-ret));
}
free(full_from);
free(full_to);
return ret;
}
static int process_link(const char *path, const char *lnk, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
char *full_link_path = path_cat(r->full_subvol_path, lnk);
if (g_verbose >= 2)
fprintf(stderr, "link %s -> %s\n", path, lnk);
ret = link(full_link_path, full_path);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: link %s -> %s failed. %s\n", path,
lnk, strerror(-ret));
}
free(full_path);
free(full_link_path);
return ret;
}
static int process_unlink(const char *path, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "unlink %s\n", path);
ret = unlink(full_path);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: unlink %s failed. %s\n", path,
strerror(-ret));
}
free(full_path);
return ret;
}
static int process_rmdir(const char *path, void *user)
{
int ret;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "rmdir %s\n", path);
ret = rmdir(full_path);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: rmdir %s failed. %s\n", path,
strerror(-ret));
}
free(full_path);
return ret;
}
static int open_inode_for_write(struct btrfs_receive *r, const char *path)
{
int ret = 0;
if (r->write_fd != -1) {
if (strcmp(r->write_path, path) == 0)
goto out;
close(r->write_fd);
r->write_fd = -1;
}
r->write_fd = open(path, O_RDWR);
if (r->write_fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: open %s failed. %s\n", path,
strerror(-ret));
goto out;
}
free(r->write_path);
r->write_path = strdup(path);
out:
return ret;
}
static void close_inode_for_write(struct btrfs_receive *r)
{
if(r->write_fd == -1)
return;
close(r->write_fd);
r->write_fd = -1;
r->write_path[0] = 0;
}
static int process_write(const char *path, const void *data, u64 offset,
u64 len, void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
u64 pos = 0;
int w;
ret = open_inode_for_write(r, full_path);
if (ret < 0)
goto out;
while (pos < len) {
w = pwrite(r->write_fd, (char*)data + pos, len - pos,
offset + pos);
if (w < 0) {
ret = -errno;
fprintf(stderr, "ERROR: writing to %s failed. %s\n",
path, strerror(-ret));
goto out;
}
pos += w;
}
out:
free(full_path);
return ret;
}
static int process_clone(const char *path, u64 offset, u64 len,
const u8 *clone_uuid, u64 clone_ctransid,
const char *clone_path, u64 clone_offset,
void *user)
{
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
int ret;
struct btrfs_receive *r = user;
struct btrfs_ioctl_clone_range_args clone_args;
struct subvol_info *si = NULL;
char *full_path = path_cat(r->full_subvol_path, path);
char *subvol_path = NULL;
char *full_clone_path = NULL;
int clone_fd = -1;
ret = open_inode_for_write(r, full_path);
if (ret < 0)
goto out;
si = subvol_uuid_search(&r->sus, 0, clone_uuid, clone_ctransid, NULL,
subvol_search_by_received_uuid);
if (!si) {
if (memcmp(clone_uuid, r->cur_subvol->received_uuid,
BTRFS_UUID_SIZE) == 0) {
/* TODO check generation of extent */
subvol_path = strdup(r->cur_subvol->path);
} else {
ret = -ENOENT;
fprintf(stderr, "ERROR: did not find source subvol.\n");
goto out;
}
} else {
/*if (rs_args.ctransid > rs_args.rtransid) {
if (!r->force) {
ret = -EINVAL;
fprintf(stderr, "ERROR: subvolume %s was "
"modified after it was "
"received.\n",
r->subvol_parent_name);
goto out;
} else {
fprintf(stderr, "WARNING: subvolume %s was "
"modified after it was "
"received.\n",
r->subvol_parent_name);
}
}*/
subvol_path = strdup(si->path);
}
full_clone_path = path_cat3(r->root_path, subvol_path, clone_path);
clone_fd = open(full_clone_path, O_RDONLY | O_NOATIME);
if (clone_fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to open %s. %s\n",
full_clone_path, strerror(-ret));
goto out;
}
clone_args.src_fd = clone_fd;
clone_args.src_offset = clone_offset;
clone_args.src_length = len;
clone_args.dest_offset = offset;
ret = ioctl(r->write_fd, BTRFS_IOC_CLONE_RANGE, &clone_args);
if (ret) {
ret = -errno;
fprintf(stderr, "ERROR: failed to clone extents to %s\n%s\n",
path, strerror(-ret));
goto out;
}
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 (si) {
free(si->path);
free(si);
}
free(full_path);
free(full_clone_path);
free(subvol_path);
if (clone_fd != -1)
close(clone_fd);
return ret;
}
static int process_set_xattr(const char *path, const char *name,
const void *data, int len, void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2) {
fprintf(stderr, "set_xattr %s - name=%s data_len=%d "
"data=%.*s\n", path, name, len,
len, (char*)data);
}
ret = lsetxattr(full_path, name, data, len, 0);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: lsetxattr %s %s=%.*s failed. %s\n",
path, name, len, (char*)data, strerror(-ret));
goto out;
}
out:
free(full_path);
return ret;
}
static int process_remove_xattr(const char *path, const char *name, void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2) {
fprintf(stderr, "remove_xattr %s - name=%s\n",
path, name);
}
ret = lremovexattr(full_path, name);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: lremovexattr %s %s failed. %s\n",
path, name, strerror(-ret));
goto out;
}
out:
free(full_path);
return ret;
}
static int process_truncate(const char *path, u64 size, void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "truncate %s size=%llu\n", path, size);
ret = truncate(full_path, size);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: truncate %s failed. %s\n",
path, strerror(-ret));
goto out;
}
out:
free(full_path);
return ret;
}
static int process_chmod(const char *path, u64 mode, void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "chmod %s - mode=0%o\n", path, (int)mode);
ret = chmod(full_path, mode);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: chmod %s failed. %s\n",
path, strerror(-ret));
goto out;
}
out:
free(full_path);
return ret;
}
static int process_chown(const char *path, u64 uid, u64 gid, void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
if (g_verbose >= 2)
fprintf(stderr, "chown %s - uid=%llu, gid=%llu\n", path,
uid, gid);
ret = lchown(full_path, uid, gid);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: chown %s failed. %s\n",
path, strerror(-ret));
goto out;
}
out:
free(full_path);
return ret;
}
static int process_utimes(const char *path, struct timespec *at,
struct timespec *mt, struct timespec *ct,
void *user)
{
int ret = 0;
struct btrfs_receive *r = user;
char *full_path = path_cat(r->full_subvol_path, path);
struct timespec tv[2];
if (g_verbose >= 2)
fprintf(stderr, "utimes %s\n", path);
tv[0] = *at;
tv[1] = *mt;
ret = utimensat(AT_FDCWD, full_path, tv, AT_SYMLINK_NOFOLLOW);
if (ret < 0) {
ret = -errno;
fprintf(stderr, "ERROR: utimes %s failed. %s\n",
path, strerror(-ret));
goto out;
}
out:
free(full_path);
return ret;
}
static struct btrfs_send_ops send_ops = {
.subvol = process_subvol,
.snapshot = process_snapshot,
.mkfile = process_mkfile,
.mkdir = process_mkdir,
.mknod = process_mknod,
.mkfifo = process_mkfifo,
.mksock = process_mksock,
.symlink = process_symlink,
.rename = process_rename,
.link = process_link,
.unlink = process_unlink,
.rmdir = process_rmdir,
.write = process_write,
.clone = process_clone,
.set_xattr = process_set_xattr,
.remove_xattr = process_remove_xattr,
.truncate = process_truncate,
.chmod = process_chmod,
.chown = process_chown,
.utimes = process_utimes,
};
static int do_receive(struct btrfs_receive *r, const char *tomnt, int r_fd,
u64 max_errors)
{
int ret;
char *dest_dir_full_path;
int end = 0;
dest_dir_full_path = realpath(tomnt, NULL);
if (!dest_dir_full_path) {
ret = -errno;
fprintf(stderr, "ERROR: realpath(%s) failed. %s\n", tomnt,
strerror(-ret));
goto out;
}
r->dest_dir_fd = open(dest_dir_full_path, O_RDONLY | O_NOATIME);
if (r->dest_dir_fd < 0) {
ret = -errno;
fprintf(stderr,
"ERROR: failed to open destination directory %s. %s\n",
dest_dir_full_path, strerror(-ret));
goto out;
}
ret = find_mount_root(dest_dir_full_path, &r->root_path);
if (ret < 0) {
fprintf(stderr,
"ERROR: failed to determine mount point for %s: %s\n",
dest_dir_full_path, strerror(-ret));
ret = -EINVAL;
goto out;
}
if (ret > 0) {
fprintf(stderr,
"ERROR: %s doesn't belong to btrfs mount point\n",
dest_dir_full_path);
ret = -EINVAL;
goto out;
}
r->mnt_fd = open(r->root_path, O_RDONLY | O_NOATIME);
if (r->mnt_fd < 0) {
ret = -errno;
fprintf(stderr, "ERROR: failed to open %s. %s\n", r->root_path,
strerror(-ret));
goto out;
}
if (r->dest_dir_chroot) {
if (chroot(dest_dir_full_path)) {
ret = -errno;
fprintf(stderr,
"ERROR: failed to chroot to %s, %s\n",
dest_dir_full_path,
strerror(-ret));
goto out;
}
if (chdir("/")) {
ret = -errno;
fprintf(stderr,
"ERROR: failed to chdir to /, %s\n",
strerror(-ret));
goto out;
}
fprintf(stderr, "Chroot to %s\n", dest_dir_full_path);
r->root_path = strdup("/");
r->dest_dir_path = r->root_path;
} else {
/*
* find_mount_root returns a root_path that is a subpath of
* dest_dir_full_path. Now get the other part of root_path,
* which is the destination dir relative to root_path.
*/
r->dest_dir_path = dest_dir_full_path + strlen(r->root_path);
while (r->dest_dir_path[0] == '/')
r->dest_dir_path++;
}
ret = subvol_uuid_search_init(r->mnt_fd, &r->sus);
if (ret < 0)
goto out;
while (!end) {
Btrfs-progs: btrfs-receive optionally honors the end-cmd A new option is added to btrfs-receive to change the behavior when an <end cmd> is received in the Btrfs send stream. The traditional behavior (which still is the default) is to continue to read the stream until an EOF condition is encountered. If an <end cmd> is received, afterwards either an EOF or a new <stream header> is expected. The new behavior (if the -e option is set on the command line) is to terminate after an <end cmd> is read without the need for an EOF. This allows the stream (e.g. a single TCP stream) to carry additional data or even multiple Btrfs send streams. Old btrfs-send tools used to encode multiple snapshots like this (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF If the new -e option is set, the expected format is like this: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> The btrfs-send tool is changed in a seperate commit to always use the new format, i.e. to send an <end cmd> only at the end. 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. The goal of changing the semantic of <end cmd> 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>
2013-04-09 17:08:40 +00:00
ret = btrfs_read_and_process_send_stream(r_fd, &send_ops, r,
r->honor_end_cmd,
max_errors);
if (ret < 0)
goto out;
if (ret)
end = 1;
close_inode_for_write(r);
ret = finish_subvol(r);
if (ret < 0)
goto out;
}
ret = 0;
out:
if (r->write_fd != -1) {
close(r->write_fd);
r->write_fd = -1;
}
free(r->root_path);
r->root_path = NULL;
free(r->write_path);
r->write_path = NULL;
free(r->full_subvol_path);
r->full_subvol_path = NULL;
r->dest_dir_path = NULL;
free(dest_dir_full_path);
if (r->cur_subvol) {
free(r->cur_subvol->path);
free(r->cur_subvol);
r->cur_subvol = NULL;
}
subvol_uuid_search_finit(&r->sus);
if (r->mnt_fd != -1) {
close(r->mnt_fd);
r->mnt_fd = -1;
}
if (r->dest_dir_fd != -1) {
close(r->dest_dir_fd);
r->dest_dir_fd = -1;
}
return ret;
}
int cmd_receive(int argc, char **argv)
{
char *tomnt = NULL;
char *fromfile = NULL;
struct btrfs_receive r;
int receive_fd = fileno(stdin);
u64 max_errors = 1;
int ret;
memset(&r, 0, sizeof(r));
r.mnt_fd = -1;
r.write_fd = -1;
r.dest_dir_fd = -1;
r.dest_dir_chroot = 0;
while (1) {
int c;
static const struct option long_opts[] = {
{ "max-errors", required_argument, NULL, 'E' },
{ "chroot", no_argument, NULL, 'C' },
{ NULL, 0, NULL, 0 }
};
c = getopt_long(argc, argv, "Cevf:", long_opts, NULL);
if (c < 0)
break;
switch (c) {
case 'v':
g_verbose++;
break;
case 'f':
fromfile = optarg;
break;
Btrfs-progs: btrfs-receive optionally honors the end-cmd A new option is added to btrfs-receive to change the behavior when an <end cmd> is received in the Btrfs send stream. The traditional behavior (which still is the default) is to continue to read the stream until an EOF condition is encountered. If an <end cmd> is received, afterwards either an EOF or a new <stream header> is expected. The new behavior (if the -e option is set on the command line) is to terminate after an <end cmd> is read without the need for an EOF. This allows the stream (e.g. a single TCP stream) to carry additional data or even multiple Btrfs send streams. Old btrfs-send tools used to encode multiple snapshots like this (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF If the new -e option is set, the expected format is like this: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> The btrfs-send tool is changed in a seperate commit to always use the new format, i.e. to send an <end cmd> only at the end. 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. The goal of changing the semantic of <end cmd> 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>
2013-04-09 17:08:40 +00:00
case 'e':
r.honor_end_cmd = 1;
break;
case 'C':
r.dest_dir_chroot = 1;
break;
case 'E':
max_errors = arg_strtou64(optarg);
break;
case '?':
default:
fprintf(stderr, "ERROR: receive args invalid.\n");
return 1;
}
}
if (check_argc_exact(argc - optind, 1))
usage(cmd_receive_usage);
tomnt = argv[optind];
if (fromfile) {
receive_fd = open(fromfile, O_RDONLY | O_NOATIME);
if (receive_fd < 0) {
fprintf(stderr, "ERROR: failed to open %s\n", fromfile);
return 1;
}
}
ret = do_receive(&r, tomnt, receive_fd, max_errors);
return !!ret;
}
const char * const cmd_receive_usage[] = {
"btrfs receive [-ve] [-f <infile>] [--max-errors <N>] <mount>",
"Receive subvolumes from stdin.",
"Receives one or more subvolumes that were previously",
"sent with btrfs send. The received subvolumes are stored",
"into <mount>.",
"btrfs receive will fail in case a receiving subvolume",
"already exists. It will also fail in case a previously",
"received subvolume was changed after it was received.",
"After receiving a subvolume, it is immediately set to",
"read only.\n",
"-v Enable verbose debug output. Each",
" occurrence of this option increases the",
" verbose level more.",
"-f <infile> By default, btrfs receive uses stdin",
" to receive the subvolumes. Use this",
" option to specify a file to use instead.",
Btrfs-progs: btrfs-receive optionally honors the end-cmd A new option is added to btrfs-receive to change the behavior when an <end cmd> is received in the Btrfs send stream. The traditional behavior (which still is the default) is to continue to read the stream until an EOF condition is encountered. If an <end cmd> is received, afterwards either an EOF or a new <stream header> is expected. The new behavior (if the -e option is set on the command line) is to terminate after an <end cmd> is read without the need for an EOF. This allows the stream (e.g. a single TCP stream) to carry additional data or even multiple Btrfs send streams. Old btrfs-send tools used to encode multiple snapshots like this (with 2 snapshots in this example): <stream header> + <sequence of commands> + <end cmd> + <stream header> + <sequence of commands> + <end cmd> + EOF If the new -e option is set, the expected format is like this: <stream header> + <sequence of commands> + <sequence of commands> + <end cmd> The btrfs-send tool is changed in a seperate commit to always use the new format, i.e. to send an <end cmd> only at the end. 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. The goal of changing the semantic of <end cmd> 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>
2013-04-09 17:08:40 +00:00
"-e Terminate after receiving an <end cmd>",
" in the data stream. Without this option,",
" the receiver terminates only if an error",
" is recognized or on EOF.",
"-C|--chroot confine the process to <mount> using chroot",
"--max-errors <N> Terminate as soon as N errors happened while",
" processing commands from the send stream.",
" Default value is 1. A value of 0 means no limit.",
NULL
};