[sheepdog] [PATCH v2 07/15] sheepfs: add 'volume' entry

MORITA Kazutaka morita.kazutaka at lab.ntt.co.jp
Sun May 20 12:03:27 CEST 2012


At Mon, 14 May 2012 17:47:32 +0800,
Liu Yuan wrote:
> 
> From: Liu Yuan <tailai.ly at taobao.com>
> 
> This is where we can access sheepdog's storage from well received 'file'
> abstraction. All the attatched volume will be seen as a file in volume
> directory.
> 
> To attatch a volume:
> echo test > sheepfs_dir/vdi/mount
> 
> Then you will see a file entry in sheepfs_dir/volume/test, which you can do
> the tricks you are fond of. For, e.g, we can boot an sheepdog image as normally
> as for an ordinary raw image:
> 
> $ qemu-system-x86_64 --enable-kvm -m 1024 -drive \
>   file=sheepfs_dir/volume/test,cache=writeback.
> 
> This file abstraction integrates well with kernel's other component such as
> pagecache.
> 
> Signed-off-by: Liu Yuan <tailai.ly at taobao.com>
> ---
>  sheepfs/Makefile.am |    2 +-
>  sheepfs/core.c      |    8 +-
>  sheepfs/sheepfs.h   |    8 ++
>  sheepfs/volume.c    |  299 +++++++++++++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 314 insertions(+), 3 deletions(-)
>  create mode 100644 sheepfs/volume.c
> 
> diff --git a/sheepfs/Makefile.am b/sheepfs/Makefile.am
> index f6201a6..c451b86 100644
> --- a/sheepfs/Makefile.am
> +++ b/sheepfs/Makefile.am
> @@ -23,7 +23,7 @@ INCLUDES		= -I$(top_builddir)/include -I$(top_srcdir)/include
>  
>  sbin_PROGRAMS		= sheepfs
>  
> -sheepfs_SOURCES		= core.c cluster.c vdi.c shadow_file.c
> +sheepfs_SOURCES		= core.c cluster.c vdi.c shadow_file.c volume.c
>  
>  sheepfs_LDADD	  	= ../lib/libsheepdog.a $(fuse_LIBS) $(LIBS)
>  sheepfs_DEPENDENCIES	= ../lib/libsheepdog.a
> diff --git a/sheepfs/core.c b/sheepfs/core.c
> index c042245..0f4f451 100644
> --- a/sheepfs/core.c
> +++ b/sheepfs/core.c
> @@ -48,10 +48,10 @@ static struct sheepfs_file_operation {
>  	size_t (*get_size)(const char *path);
>  } sheepfs_file_ops[] = {
>  	[OP_NULL        ] = { NULL, NULL, NULL },
> -	[OP_CLUSTER_INFO] = { cluster_info_read, NULL,
> -				cluster_info_get_size },
> +	[OP_CLUSTER_INFO] = { cluster_info_read, NULL, cluster_info_get_size },
>  	[OP_VDI_LIST]     = { vdi_list_read, NULL, vdi_list_get_size },
>  	[OP_VDI_MOUNT]    = { NULL, vdi_mount_write, NULL },
> +	[OP_VOLUME]       = { volume_read, volume_write, volume_get_size },
>  };
>  
>  int sheepfs_set_op(const char *path, unsigned opcode)
> @@ -184,6 +184,8 @@ static int sheepfs_main_loop(char *mountpoint)
>  	int ret = -1;
>  
>  	fuse_opt_add_arg(&args, "sheepfs"); /* placeholder for argv[0] */
> +	fuse_opt_add_arg(&args, "-oallow_root");
> +	fuse_opt_add_arg(&args, "-obig_writes");
>  	fuse_opt_add_arg(&args, "-ofsname=sheepfs");
>  	fuse_opt_add_arg(&args, mountpoint);
>  	if (sheepfs_debug)
> @@ -204,6 +206,8 @@ static int create_sheepfs_layout(void)
>  		return -1;
>  	if (create_vdi_layout() < 0)
>  		return -1;
> +	if (create_volume_layout() < 0)
> +		return -1;
>  
>  	return 0;
>  }
> diff --git a/sheepfs/sheepfs.h b/sheepfs/sheepfs.h
> index 48f3dbc..cfb1402 100644
> --- a/sheepfs/sheepfs.h
> +++ b/sheepfs/sheepfs.h
> @@ -6,6 +6,7 @@ enum sheepfs_opcode {
>  	OP_CLUSTER_INFO,
>  	OP_VDI_LIST,
>  	OP_VDI_MOUNT,
> +	OP_VOLUME,
>  };
>  
>  extern char sheepfs_shadow[];
> @@ -25,6 +26,13 @@ extern int shadow_file_getxattr(const char *path, const char *name,
>  extern int shadow_file_delete(const char *path);
>  extern int shadow_file_exsit(const char *path);
>  
> +/* volume.c */
> +extern int create_volume_layout(void);
> +extern int volume_read(const char *path, char *buf, size_t size, off_t offset);
> +extern int volume_write(const char *, const char *buf, size_t size, off_t);
> +extern size_t volume_get_size(const char *);
> +extern int volume_create_entry(const char *entry);
> +
>  /* cluster.c */
>  extern int cluster_info_read(const char *path, char *buf, size_t size, off_t);
>  extern size_t cluster_info_get_size(const char *path);
> diff --git a/sheepfs/volume.c b/sheepfs/volume.c
> new file mode 100644
> index 0000000..438e9a2
> --- /dev/null
> +++ b/sheepfs/volume.c
> @@ -0,0 +1,299 @@
> +/*
> + * Copyright (C) 2012 Taobao Inc.
> + *
> + * Liu Yuan <namei.unix at gmail.com>
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License version
> + * 2 as published by the Free Software Foundation.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program. If not, see <http://www.gnu.org/licenses/>.
> + */
> +#include <unistd.h>
> +#include <sys/types.h>
> +#include <sys/stat.h>
> +#include <fcntl.h>
> +#include <errno.h>
> +#include <stdlib.h>
> +#include <stdio.h>
> +#include <time.h>
> +#include <assert.h>
> +#include <syslog.h>
> +
> +#include "strbuf.h"
> +#include "sheepfs.h"
> +#include "net.h"
> +#include "rbtree.h"
> +
> +#define PATH_VOLUME	"/volume"
> +
> +#define SH_VID_NAME   "user.volume.vid"
> +#define SH_VID_SIZE   sizeof(uint32_t)
> +
> +#define SH_SIZE_NAME   "user.volume.size"
> +#define SH_SIZE_SIZE   sizeof(size_t)
> +
> +#define VOLUME_READ   0
> +#define VOLUME_WRITE  1
> +
> +struct vdi_inode {
> +	struct rb_node rb;
> +	uint32_t vid;
> +	struct sheepdog_inode *inode;
> +};
> +
> +static struct rb_root vdi_inode_tree = RB_ROOT;
> +
> +static struct vdi_inode *vdi_inode_tree_insert(struct vdi_inode *new)
> +{
> +	struct rb_node **p = &vdi_inode_tree.rb_node;
> +	struct rb_node *parent = NULL;
> +	struct vdi_inode *entry;
> +
> +	while (*p) {
> +		parent = *p;
> +		entry = rb_entry(parent, struct vdi_inode, rb);
> +
> +		if (new->vid < entry->vid)
> +			p = &(*p)->rb_left;
> +		else if (new->vid > entry->vid)
> +			p = &(*p)->rb_right;
> +		else
> +			return entry; /* already has this entry */
> +	}
> +	rb_link_node(&new->rb, parent, p);
> +	rb_insert_color(&new->rb, &vdi_inode_tree);
> +
> +	return NULL; /* insert successfully */
> +}
> +
> +static struct vdi_inode *vdi_inode_tree_search(uint32_t vid)
> +{
> +	struct rb_node *n = vdi_inode_tree.rb_node;
> +	struct vdi_inode *t;
> +
> +	while (n) {
> +		t = rb_entry(n, struct vdi_inode, rb);
> +
> +		if (vid < t->vid)
> +			n = n->rb_left;
> +		else if (vid > t->vid)
> +			n = n->rb_right;
> +		else
> +			return t; /* found it */
> +	}
> +
> +	return NULL;
> +}
> +
> +int create_volume_layout(void)
> +{
> +	if (shadow_dir_create(PATH_VOLUME) < 0)
> +		return -1;
> +	return 0;
> +}
> +
> +static int volume_rw_object(char *buf, uint64_t oid, size_t size,
> +			     off_t off, int rw)
> +{
> +	struct sd_obj_req hdr = { 0 };
> +	struct sd_obj_rsp *rsp = (struct sd_obj_rsp *)&hdr;
> +	int ret;
> +	unsigned wlen = 0, rlen = 0;
> +	int create = 0;
> +	uint32_t vid = oid_to_vid(oid);
> +	struct vdi_inode *vdi = vdi_inode_tree_search(vid);
> +	unsigned long idx = 0;
> +
> +	if (is_data_obj(oid)) {
> +		idx = data_oid_to_idx(oid);
> +		assert(vdi);
> +		if (!vdi->inode->data_vdi_id[idx]) {
> +			/* if object doesn't exist, we'er done */
> +			if (rw == VOLUME_READ) {
> +				memset(buf, 0, size);
> +				goto done;
> +			}
> +			create = 1;
> +		}
> +	}
> +
> +	if (rw == VOLUME_READ) {
> +		rlen = size;
> +		hdr.opcode = SD_OP_READ_OBJ;
> +	} else {
> +		wlen = size;
> +		hdr.opcode = create ?
> +			SD_OP_CREATE_AND_WRITE_OBJ : SD_OP_WRITE_OBJ;
> +		hdr.flags |= SD_FLAG_CMD_WRITE;
> +	}
> +
> +	hdr.oid = oid;
> +	hdr.data_length = size;
> +	hdr.offset = off;

If we add '-D' to the sheep command line options, hdr.data_length and
hdr.offset must be aligned to 512 bytes.  How about returning EINVAL
to users when off and size is not aligned?


> +	hdr.flags |= SD_FLAG_CMD_CACHE;
> +
> +	ret = exec_req(0, (struct sd_req *)&hdr, buf, &wlen, &rlen);
> +
> +	if (ret || rsp->result != SD_RES_SUCCESS) {
> +		syslog(LOG_ERR, "failed to %s object %" PRIx64 " ret %d, res %d\n",
> +			rw == VOLUME_READ ? "read" : "write", oid, ret,
> +			rsp->result);
> +		return -1;
> +	}
> +
> +	if (create) {
> +		vdi->inode->data_vdi_id[idx] = vid;
> +		/* writeback inode update */
> +		if (volume_rw_object((char *)&vid, vid_to_vdi_oid(vid),
> +				 sizeof(vid),
> +				 SD_INODE_HEADER_SIZE + sizeof(vid) * idx,
> +				 VOLUME_WRITE) < 0)
> +		return -1;

Wrong indentation level.


> +	}
> +done:
> +	return size;
> +}
> +
> +/* Do sync read/write */
> +static int volume_do_rw(const char *path, char *buf, size_t size,
> +			 off_t offset, int rw)
> +{
> +	uint32_t vid;
> +	uint64_t oid;
> +	unsigned long idx;
> +	off_t start;
> +	size_t len, ret;
> +
> +	if (shadow_file_getxattr(path, SH_VID_NAME, &vid, SH_VID_SIZE) < 0)
> +		return -1;
> +
> +	idx = offset / SD_DATA_OBJ_SIZE;
> +	oid = vid_to_data_oid(vid, idx);
> +	start = offset % SD_DATA_OBJ_SIZE;
> +
> +	len = SD_DATA_OBJ_SIZE - start;
> +	if (size < len)
> +		len = size;
> +
> +	do {
> +		syslog(LOG_ERR, "%s oid %"PRIx64", off %ju, len %zu,"
> +			" size %zu\n",
> +			rw == VOLUME_READ ? "read" : "write",
> +			oid, start, len, size);
> +		ret = volume_rw_object(buf, oid, len, start, rw);
> +
> +		if (ret != len)
> +			return -1;
> +
> +		oid++;
> +		size -= len;
> +		start = (start + len) % SD_DATA_OBJ_SIZE;
> +		buf += len;
> +		len = size > SD_DATA_OBJ_SIZE ? SD_DATA_OBJ_SIZE : size;
> +	} while (size > 0);
> +
> +	return 0;
> +}
> +
> +int volume_read(const char *path, char *buf, size_t size, off_t offset)
> +{
> +
> +	if (volume_do_rw(path, buf, size, offset, VOLUME_READ) < 0)
> +		return -EIO;
> +
> +	return size;
> +}
> +
> +int volume_write(const char *path, const char *buf, size_t size, off_t offset)
> +{
> +	if (volume_do_rw(path, (char *)buf, size, offset, VOLUME_WRITE) < 0)
> +		return -EIO;
> +
> +	return size;
> +}
> +
> +size_t volume_get_size(const char *path)
> +{
> +	size_t size = 0;
> +
> +	shadow_file_getxattr(path, SH_SIZE_NAME, &size, SH_SIZE_SIZE);
> +	return size;
> +}
> +
> +static int init_vdi_info(const char *entry, uint32_t *vid, size_t *size)
> +{
> +	struct strbuf *buf;
> +	void *inode_buf;
> +	struct vdi_inode *inode;
> +	char command[256] = { 0 };
> +	int ret = -1;
> +
> +	sprintf(command, "%s %s\n", "collie vdi list -r", entry);
> +	buf = sheepfs_run_cmd(command);
> +	if (!buf)
> +		return -1;
> +	if (sscanf(buf->buf, "%*s %*s %*d %zu %*s %*s %*s %"PRIx32,
> +	    size, vid) < 2) {
> +		syslog(LOG_ERR, "[%s] %m\n", __func__);

It seems that sscanf() doesn't set errno, so we shouldn't use %m here.


> +		goto out;
> +	}
> +
> +	inode_buf = malloc(SD_INODE_SIZE);
> +	if (!inode_buf) {
> +		syslog(LOG_ERR, "[%s] %m\n", __func__);
> +		goto out;
> +	}
> +
> +	if (volume_rw_object(inode_buf, vid_to_vdi_oid(*vid), SD_INODE_SIZE,
> +			     0, VOLUME_READ) < 0) {
> +		free(inode_buf);
> +		goto out;
> +	}
> +
> +	inode = xzalloc(sizeof(*inode));
> +	inode->vid = *vid;
> +	inode->inode = inode_buf;
> +	if (vdi_inode_tree_insert(inode)) {
> +		free(inode_buf);
> +		free(inode);
> +	}
> +	ret = 0;
> +out:
> +	strbuf_release(buf);
> +	return ret;
> +}
> +
> +int volume_create_entry(const char *entry)
> +{
> +	char path[PATH_MAX], *ch;
> +	uint32_t vid;
> +	size_t size;
> +
> +	ch = strchr(entry, '\n');
> +	if (ch != NULL)
> +		*ch = '\0';
> +
> +	sprintf(path, "%s/%s", PATH_VOLUME, entry);
> +	if (shadow_file_exsit(path))
> +		return 0;
> +
> +	if (shadow_file_create(path) < 0)
> +		return -1;

The requested vdi may not exist.  We should call init_vdi_info()
before creating a volume file.

Thanks,

Kazutaka

> +
> +	if (init_vdi_info(entry, &vid, &size) < 0)
> +		return -1;
> +	if (shadow_file_setxattr(path, SH_VID_NAME, &vid, SH_VID_SIZE) < 0) {
> +		shadow_file_delete(path);
> +		return -1;
> +	}
> +	if (shadow_file_setxattr(path, SH_SIZE_NAME, &size, SH_SIZE_SIZE) < 0) {
> +		shadow_file_delete(path);
> +		return -1;
> +	}
> +	if (sheepfs_set_op(path, OP_VOLUME) < 0)
> +		return -1;
> +
> +	return 0;
> +}
> -- 
> 1.7.8.2
> 



More information about the sheepdog mailing list