[ceph.git] / ceph / src / os / filestore / ZFSFileStoreBackend.cc

// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab

#include "include/int_types.h"
#include "include/types.h"

#include <unistd.h>
#include <fcntl.h>
#include <errno.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/ioctl.h>

#include "include/compat.h"
#include "include/linux_fiemap.h"
#include "include/color.h"
#include "include/buffer.h"
#include "include/assert.h"

#include <iostream>
#include <fstream>
#include <sstream>

#include "common/errno.h"
#include "common/config.h"
#include "common/sync_filesystem.h"

#include "ZFSFileStoreBackend.h"

#define dout_context cct()
#define dout_subsys ceph_subsys_filestore
#undef dout_prefix
#define dout_prefix *_dout << "zfsfilestorebackend(" << get_basedir_path() << ") "

ZFSFileStoreBackend::ZFSFileStoreBackend(FileStore *fs) :
  GenericFileStoreBackend(fs), base_zh(NULL), current_zh(NULL),
  m_filestore_zfs_snap(cct()->_conf->filestore_zfs_snap)
{
  int ret = zfs.init();
  if (ret < 0) {
    dout(0) << "ZFSFileStoreBackend: failed to init libzfs" << dendl;
    return;
  }

  base_zh = zfs.path_to_zhandle(get_basedir_path().c_str(), ZFS::TYPE_FILESYSTEM);
  if (!base_zh) {
    dout(0) << "ZFSFileStoreBackend: failed to get zfs handler for basedir" << dendl;
    return;
  }

  update_current_zh();
}

ZFSFileStoreBackend::~ZFSFileStoreBackend()
{
  if (base_zh)
    zfs.close(base_zh);
  if (current_zh)
    zfs.close(current_zh);
}

int ZFSFileStoreBackend::update_current_zh()
{
  char path[PATH_MAX];
  snprintf(path, sizeof(path), "%s/current", zfs.get_name(base_zh));
  ZFS::Handle *zh = zfs.open(path, ZFS::TYPE_FILESYSTEM);
  if (zh) {
    char *mnt;
    if (zfs.is_mounted(zh, &mnt)) {
      int ret = get_current_path() == mnt;
      free(mnt);
      if (ret) {
	current_zh = zh;
	return 0;
      }
    } else {
      int ret = zfs.mount(zh, NULL, 0);
      if (ret < 0) {
	ret = -errno;
	dout(0) << "update_current_zh: zfs_mount '" << zfs.get_name(zh)
		<< "' got " << cpp_strerror(ret) << dendl;
	return ret;
      }
    }
    zfs.close(zh);
  } else {
    dout(0) << "update_current_zh: zfs_open '" << path << "' got NULL" << dendl;
    return -ENOENT;
  }

  zh = zfs.path_to_zhandle(get_current_path().c_str(), ZFS::TYPE_FILESYSTEM);
  if (zh) {
    if (strcmp(zfs.get_name(base_zh), zfs.get_name(zh))) {
      current_zh = zh;
      return 0;
    }
    zfs.close(zh);
    dout(0) << "update_current_zh: basedir and current/ on the same filesystem" << dendl;
  } else {
    dout(0) << "update_current_zh: current/ not exist" << dendl;
  }
  return -ENOENT;
}

int ZFSFileStoreBackend::detect_features()
{
  if (!current_zh)
    dout(0) << "detect_features: null zfs handle for current/" << dendl;
  return 0;
}

bool ZFSFileStoreBackend::can_checkpoint()
{
  return m_filestore_zfs_snap && current_zh != NULL;
}

int ZFSFileStoreBackend::create_current()
{
  struct stat st;
  int ret = ::stat(get_current_path().c_str(), &st);
  if (ret == 0) {
    // current/ exists
    if (!S_ISDIR(st.st_mode)) {
      dout(0) << "create_current: current/ exists but is not a directory" << dendl;
      return -ENOTDIR;
    }
    return 0;
  } else if (errno != ENOENT) {
    ret = -errno;
    dout(0) << "create_current: cannot stat current/ " << cpp_strerror(ret) << dendl;
    return ret;
  }

  char path[PATH_MAX];
  snprintf(path, sizeof(path), "%s/current", zfs.get_name(base_zh));
  ret = zfs.create(path, ZFS::TYPE_FILESYSTEM);
  if (ret < 0 && errno != EEXIST) {
    ret = -errno;
    dout(0) << "create_current: zfs_create '" << path << "' got " << cpp_strerror(ret) << dendl;
    return ret;
  }

  ret = update_current_zh();
  return ret;
}

static int list_checkpoints_callback(ZFS::Handle *zh, void *data)
{
  list<string> *ls = static_cast<list<string> *>(data);
  string str = ZFS::get_name(zh);
  size_t pos = str.find('@');
  assert(pos != string::npos && pos + 1 != str.length());
  ls->push_back(str.substr(pos + 1));
  return 0;
}

int ZFSFileStoreBackend::list_checkpoints(list<string>& ls)
{
  dout(10) << "list_checkpoints:" << dendl;
  if (!current_zh)
    return -EINVAL;

  list<string> snaps;
  int ret = zfs.iter_snapshots_sorted(current_zh, list_checkpoints_callback, &snaps);
  if (ret < 0) {
    ret = -errno;
    dout(0) << "list_checkpoints: zfs_iter_snapshots_sorted got" << cpp_strerror(ret) << dendl;
    return ret;
  }
  ls.swap(snaps);
  return 0;
}

int ZFSFileStoreBackend::create_checkpoint(const string& name, uint64_t *cid)
{
  dout(10) << "create_checkpoint: '" << name << "'" << dendl;
  if (!current_zh)
    return -EINVAL;

  // looks like zfsonlinux doesn't flush dirty data when taking snapshot
  int ret = sync_filesystem(get_current_fd());
  if (ret < 0) {
    ret = -errno;
    dout(0) << "create_checkpoint: sync_filesystem got" << cpp_strerror(ret) << dendl;
    return ret;
  }

  char path[PATH_MAX];
  snprintf(path, sizeof(path), "%s@%s", zfs.get_name(current_zh), name.c_str());
  ret = zfs.snapshot(path, false);
  if (ret < 0) {
    ret = -errno;
    dout(0) << "create_checkpoint: zfs_snapshot '" << path << "' got" << cpp_strerror(ret) << dendl;
    return ret;
  }
  if (cid)
    *cid = 0;
  return 0;
}

int ZFSFileStoreBackend::rollback_to(const string& name)
{
  dout(10) << "rollback_to: '" << name << "'" << dendl;
  if (!current_zh)
    return -EINVAL;

  // umount current to avoid triggering online rollback deadlock
  int ret;
  if (zfs.is_mounted(current_zh, NULL)) {
    ret = zfs.umount(current_zh, NULL, 0);
    if (ret < 0) {
      ret = -errno;
      dout(0) << "rollback_to: zfs_umount '" << zfs.get_name(current_zh) << "' got" << cpp_strerror(ret) << dendl;
    }
  }

  char path[PATH_MAX];
  snprintf(path, sizeof(path), "%s@%s", zfs.get_name(current_zh), name.c_str());

  ZFS::Handle *snap_zh = zfs.open(path, ZFS::TYPE_SNAPSHOT);
  if (!snap_zh) {
    dout(0) << "rollback_to: zfs_open '" << path << "' got NULL" << dendl;
    return -ENOENT;
  }

  ret = zfs.rollback(current_zh, snap_zh, false);
  if (ret < 0) {
    ret = -errno;
    dout(0) << "rollback_to: zfs_rollback '" << zfs.get_name(snap_zh) << "' got" << cpp_strerror(ret) << dendl;
  }

  if (!zfs.is_mounted(current_zh, NULL)) {
    int ret = zfs.mount(current_zh, NULL, 0);
    if (ret < 0) {
      ret = -errno;
      dout(0) << "update_current_zh: zfs_mount '" << zfs.get_name(current_zh) << "' got " << cpp_strerror(ret) << dendl;
      return ret;
    }
  }

  zfs.close(snap_zh);
  return ret;
}

int ZFSFileStoreBackend::destroy_checkpoint(const string& name)
{
  dout(10) << "destroy_checkpoint: '" << name << "'" << dendl;
  if (!current_zh)
    return -EINVAL;

  int ret = zfs.destroy_snaps(current_zh, name.c_str(), true);
  if (ret < 0) {
    ret = -errno;
    dout(0) << "destroy_checkpoint: zfs_destroy_snaps '" << name << "' got" << cpp_strerror(ret) << dendl;
  }
  return ret;
}
Commit	Line	Data
7c673cae FG	1	// -- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t --
	2	// vim: ts=8 sw=2 smarttab
	3
	4	#include "include/int_types.h"
	5	#include "include/types.h"
	6
	7	#include <unistd.h>
	8	#include <fcntl.h>
	9	#include <errno.h>
	10	#include <stdlib.h>
	11	#include <sys/types.h>
	12	#include <sys/stat.h>
	13	#include <sys/ioctl.h>
	14
	15	#include "include/compat.h"
	16	#include "include/linux_fiemap.h"
	17	#include "include/color.h"
	18	#include "include/buffer.h"
	19	#include "include/assert.h"
	20
	21	#include <iostream>
	22	#include <fstream>
	23	#include <sstream>
	24
	25	#include "common/errno.h"
	26	#include "common/config.h"
	27	#include "common/sync_filesystem.h"
	28
7c673cae FG	29	#include "ZFSFileStoreBackend.h"
7c673cae FG	30
224ce89b	31	#define dout_context cct()
7c673cae FG	32	#define dout_subsys ceph_subsys_filestore
	33	#undef dout_prefix
	34	#define dout_prefix *_dout << "zfsfilestorebackend(" << get_basedir_path() << ") "
	35
	36	ZFSFileStoreBackend::ZFSFileStoreBackend(FileStore *fs) :
	37	GenericFileStoreBackend(fs), base_zh(NULL), current_zh(NULL),
224ce89b	38	m_filestore_zfs_snap(cct()->_conf->filestore_zfs_snap)
7c673cae FG	39	{
	40	int ret = zfs.init();
	41	if (ret < 0) {
	42	dout(0) << "ZFSFileStoreBackend: failed to init libzfs" << dendl;
	43	return;
	44	}
	45
	46	base_zh = zfs.path_to_zhandle(get_basedir_path().c_str(), ZFS::TYPE_FILESYSTEM);
	47	if (!base_zh) {
	48	dout(0) << "ZFSFileStoreBackend: failed to get zfs handler for basedir" << dendl;
	49	return;
	50	}
	51
	52	update_current_zh();
	53	}
	54
	55	ZFSFileStoreBackend::~ZFSFileStoreBackend()
	56	{
	57	if (base_zh)
	58	zfs.close(base_zh);
	59	if (current_zh)
	60	zfs.close(current_zh);
	61	}
	62
	63	int ZFSFileStoreBackend::update_current_zh()
	64	{
	65	char path[PATH_MAX];
	66	snprintf(path, sizeof(path), "%s/current", zfs.get_name(base_zh));
	67	ZFS::Handle *zh = zfs.open(path, ZFS::TYPE_FILESYSTEM);
	68	if (zh) {
	69	char *mnt;
	70	if (zfs.is_mounted(zh, &mnt)) {
	71	int ret = get_current_path() == mnt;
	72	free(mnt);
	73	if (ret) {
	74	current_zh = zh;
	75	return 0;
	76	}
	77	} else {
	78	int ret = zfs.mount(zh, NULL, 0);
	79	if (ret < 0) {
	80	ret = -errno;
	81	dout(0) << "update_current_zh: zfs_mount '" << zfs.get_name(zh)
	82	<< "' got " << cpp_strerror(ret) << dendl;
	83	return ret;
	84	}
	85	}
	86	zfs.close(zh);
	87	} else {
	88	dout(0) << "update_current_zh: zfs_open '" << path << "' got NULL" << dendl;
	89	return -ENOENT;
	90	}
	91
	92	zh = zfs.path_to_zhandle(get_current_path().c_str(), ZFS::TYPE_FILESYSTEM);
	93	if (zh) {
	94	if (strcmp(zfs.get_name(base_zh), zfs.get_name(zh))) {
	95	current_zh = zh;
	96	return 0;
	97	}
	98	zfs.close(zh);
	99	dout(0) << "update_current_zh: basedir and current/ on the same filesystem" << dendl;
	100	} else {
	101	dout(0) << "update_current_zh: current/ not exist" << dendl;
	102	}
103	return -ENOENT;
104	}
105
106	int ZFSFileStoreBackend::detect_features()
107	{
108	if (!current_zh)
109	dout(0) << "detect_features: null zfs handle for current/" << dendl;
110	return 0;
111	}
112
113	bool ZFSFileStoreBackend::can_checkpoint()
114	{
115	return m_filestore_zfs_snap && current_zh != NULL;
116	}
117
118	int ZFSFileStoreBackend::create_current()
119	{
120	struct stat st;
121	int ret = ::stat(get_current_path().c_str(), &st);
122	if (ret == 0) {
123	// current/ exists
124	if (!S_ISDIR(st.st_mode)) {
125	dout(0) << "create_current: current/ exists but is not a directory" << dendl;
126	return -ENOTDIR;
127	}
128	return 0;
129	} else if (errno != ENOENT) {
130	ret = -errno;
131	dout(0) << "create_current: cannot stat current/ " << cpp_strerror(ret) << dendl;
132	return ret;
133	}
134
135	char path[PATH_MAX];
136	snprintf(path, sizeof(path), "%s/current", zfs.get_name(base_zh));
137	ret = zfs.create(path, ZFS::TYPE_FILESYSTEM);
138	if (ret < 0 && errno != EEXIST) {
139	ret = -errno;
140	dout(0) << "create_current: zfs_create '" << path << "' got " << cpp_strerror(ret) << dendl;
141	return ret;
142	}
143
144	ret = update_current_zh();
145	return ret;
146	}
147
148	static int list_checkpoints_callback(ZFS::Handle zh, void data)
149	{
150	list<string> ls = static_cast<list<string> >(data);
151	string str = ZFS::get_name(zh);
152	size_t pos = str.find('@');
153	assert(pos != string::npos && pos + 1 != str.length());
154	ls->push_back(str.substr(pos + 1));
155	return 0;
156	}
157
158	int ZFSFileStoreBackend::list_checkpoints(list<string>& ls)
159	{
160	dout(10) << "list_checkpoints:" << dendl;
161	if (!current_zh)
162	return -EINVAL;
163
164	list<string> snaps;
165	int ret = zfs.iter_snapshots_sorted(current_zh, list_checkpoints_callback, &snaps);
166	if (ret < 0) {
167	ret = -errno;
168	dout(0) << "list_checkpoints: zfs_iter_snapshots_sorted got" << cpp_strerror(ret) << dendl;
169	return ret;
170	}
171	ls.swap(snaps);
172	return 0;
173	}
174
175	int ZFSFileStoreBackend::create_checkpoint(const string& name, uint64_t *cid)
176	{
177	dout(10) << "create_checkpoint: '" << name << "'" << dendl;
178	if (!current_zh)
179	return -EINVAL;
180
181	// looks like zfsonlinux doesn't flush dirty data when taking snapshot
182	int ret = sync_filesystem(get_current_fd());
183	if (ret < 0) {
184	ret = -errno;
185	dout(0) << "create_checkpoint: sync_filesystem got" << cpp_strerror(ret) << dendl;
186	return ret;
187	}
188
189	char path[PATH_MAX];
190	snprintf(path, sizeof(path), "%s@%s", zfs.get_name(current_zh), name.c_str());
191	ret = zfs.snapshot(path, false);
192	if (ret < 0) {
193	ret = -errno;
194	dout(0) << "create_checkpoint: zfs_snapshot '" << path << "' got" << cpp_strerror(ret) << dendl;
195	return ret;
196	}
197	if (cid)
198	*cid = 0;
199	return 0;
200	}
201
202	int ZFSFileStoreBackend::rollback_to(const string& name)
203	{
204	dout(10) << "rollback_to: '" << name << "'" << dendl;
205	if (!current_zh)
206	return -EINVAL;
207
208	// umount current to avoid triggering online rollback deadlock
209	int ret;
210	if (zfs.is_mounted(current_zh, NULL)) {
211	ret = zfs.umount(current_zh, NULL, 0);
212	if (ret < 0) {
213	ret = -errno;
214	dout(0) << "rollback_to: zfs_umount '" << zfs.get_name(current_zh) << "' got" << cpp_strerror(ret) << dendl;
215	}
216	}
217
218	char path[PATH_MAX];
219	snprintf(path, sizeof(path), "%s@%s", zfs.get_name(current_zh), name.c_str());
220
221	ZFS::Handle *snap_zh = zfs.open(path, ZFS::TYPE_SNAPSHOT);
222	if (!snap_zh) {
223	dout(0) << "rollback_to: zfs_open '" << path << "' got NULL" << dendl;
224	return -ENOENT;
225	}
226
227	ret = zfs.rollback(current_zh, snap_zh, false);
228	if (ret < 0) {
229	ret = -errno;
230	dout(0) << "rollback_to: zfs_rollback '" << zfs.get_name(snap_zh) << "' got" << cpp_strerror(ret) << dendl;
231	}
232
233	if (!zfs.is_mounted(current_zh, NULL)) {
234	int ret = zfs.mount(current_zh, NULL, 0);
235	if (ret < 0) {
236	ret = -errno;
237	dout(0) << "update_current_zh: zfs_mount '" << zfs.get_name(current_zh) << "' got " << cpp_strerror(ret) << dendl;
238	return ret;
239	}
240	}
241
242	zfs.close(snap_zh);
243	return ret;
244	}
245
246	int ZFSFileStoreBackend::destroy_checkpoint(const string& name)
247	{
248	dout(10) << "destroy_checkpoint: '" << name << "'" << dendl;
249	if (!current_zh)
250	return -EINVAL;
251
252	int ret = zfs.destroy_snaps(current_zh, name.c_str(), true);
253	if (ret < 0) {
254	ret = -errno;
255	dout(0) << "destroy_checkpoint: zfs_destroy_snaps '" << name << "' got" << cpp_strerror(ret) << dendl;
256	}
257	return ret;
258	}