ceph/src/os/filestore/ZFSFileStoreBackend.cc

   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
  29 #ifdef HAVE_LIBZFS
  30
  31 #include "ZFSFileStoreBackend.h"
  32
  33 #define dout_subsys ceph_subsys_filestore
  34 #undef dout_prefix
  35 #define dout_prefix *_dout << "zfsfilestorebackend(" << get_basedir_path() << ") "
  36
  37 ZFSFileStoreBackend::ZFSFileStoreBackend(FileStore *fs) :
  38   GenericFileStoreBackend(fs), base_zh(NULL), current_zh(NULL),
  39   m_filestore_zfs_snap(cct->_conf->filestore_zfs_snap)
  40 {
  41   int ret = zfs.init();
  42   if (ret < 0) {
  43     dout(0) << "ZFSFileStoreBackend: failed to init libzfs" << dendl;
  44     return;
  45   }
  46
  47   base_zh = zfs.path_to_zhandle(get_basedir_path().c_str(), ZFS::TYPE_FILESYSTEM);
  48   if (!base_zh) {
  49     dout(0) << "ZFSFileStoreBackend: failed to get zfs handler for basedir" << dendl;
  50     return;
  51   }
  52
  53   update_current_zh();
  54 }
  55
  56 ZFSFileStoreBackend::~ZFSFileStoreBackend()
  57 {
  58   if (base_zh)
  59     zfs.close(base_zh);
  60   if (current_zh)
  61     zfs.close(current_zh);
  62 }
  63
  64 int ZFSFileStoreBackend::update_current_zh()
  65 {
  66   char path[PATH_MAX];
  67   snprintf(path, sizeof(path), "%s/current", zfs.get_name(base_zh));
  68   ZFS::Handle *zh = zfs.open(path, ZFS::TYPE_FILESYSTEM);
  69   if (zh) {
  70     char *mnt;
  71     if (zfs.is_mounted(zh, &mnt)) {
  72       int ret = get_current_path() == mnt;
  73       free(mnt);
  74       if (ret) {
  75         current_zh = zh;
  76         return 0;
  77       }
  78     } else {
  79       int ret = zfs.mount(zh, NULL, 0);
  80       if (ret < 0) {
  81         ret = -errno;
  82         dout(0) << "update_current_zh: zfs_mount '" << zfs.get_name(zh)
  83                 << "' got " << cpp_strerror(ret) << dendl;
  84         return ret;
  85       }
  86     }
  87     zfs.close(zh);
  88   } else {
  89     dout(0) << "update_current_zh: zfs_open '" << path << "' got NULL" << dendl;
  90     return -ENOENT;
  91   }
  92
  93   zh = zfs.path_to_zhandle(get_current_path().c_str(), ZFS::TYPE_FILESYSTEM);
  94   if (zh) {
  95     if (strcmp(zfs.get_name(base_zh), zfs.get_name(zh))) {
  96       current_zh = zh;
  97       return 0;
  98     }
  99     zfs.close(zh);
 100     dout(0) << "update_current_zh: basedir and current/ on the same filesystem" << dendl;
 101   } else {
 102     dout(0) << "update_current_zh: current/ not exist" << dendl;
 103   }
 104   return -ENOENT;
 105 }
 106
 107 int ZFSFileStoreBackend::detect_features()
 108 {
 109   if (!current_zh)
 110     dout(0) << "detect_features: null zfs handle for current/" << dendl;
 111   return 0;
 112 }
 113
 114 bool ZFSFileStoreBackend::can_checkpoint()
 115 {
 116   return m_filestore_zfs_snap && current_zh != NULL;
 117 }
 118
 119 int ZFSFileStoreBackend::create_current()
 120 {
 121   struct stat st;
 122   int ret = ::stat(get_current_path().c_str(), &st);
 123   if (ret == 0) {
 124     // current/ exists
 125     if (!S_ISDIR(st.st_mode)) {
 126       dout(0) << "create_current: current/ exists but is not a directory" << dendl;
 127       return -ENOTDIR;
 128     }
 129     return 0;
 130   } else if (errno != ENOENT) {
 131     ret = -errno;
 132     dout(0) << "create_current: cannot stat current/ " << cpp_strerror(ret) << dendl;
 133     return ret;
 134   }
 135
 136   char path[PATH_MAX];
 137   snprintf(path, sizeof(path), "%s/current", zfs.get_name(base_zh));
 138   ret = zfs.create(path, ZFS::TYPE_FILESYSTEM);
 139   if (ret < 0 && errno != EEXIST) {
 140     ret = -errno;
 141     dout(0) << "create_current: zfs_create '" << path << "' got " << cpp_strerror(ret) << dendl;
 142     return ret;
 143   }
 144
 145   ret = update_current_zh();
 146   return ret;
 147 }
 148
 149 static int list_checkpoints_callback(ZFS::Handle *zh, void *data)
 150 {
 151   list<string> *ls = static_cast<list<string> *>(data);
 152   string str = ZFS::get_name(zh);
 153   size_t pos = str.find('@');
 154   assert(pos != string::npos && pos + 1 != str.length());
 155   ls->push_back(str.substr(pos + 1));
 156   return 0;
 157 }
 158
 159 int ZFSFileStoreBackend::list_checkpoints(list<string>& ls)
 160 {
 161   dout(10) << "list_checkpoints:" << dendl;
 162   if (!current_zh)
 163     return -EINVAL;
 164
 165   list<string> snaps;
 166   int ret = zfs.iter_snapshots_sorted(current_zh, list_checkpoints_callback, &snaps);
 167   if (ret < 0) {
 168     ret = -errno;
 169     dout(0) << "list_checkpoints: zfs_iter_snapshots_sorted got" << cpp_strerror(ret) << dendl;
 170     return ret;
 171   }
 172   ls.swap(snaps);
 173   return 0;
 174 }
 175
 176 int ZFSFileStoreBackend::create_checkpoint(const string& name, uint64_t *cid)
 177 {
 178   dout(10) << "create_checkpoint: '" << name << "'" << dendl;
 179   if (!current_zh)
 180     return -EINVAL;
 181
 182   // looks like zfsonlinux doesn't flush dirty data when taking snapshot
 183   int ret = sync_filesystem(get_current_fd());
 184   if (ret < 0) {
 185     ret = -errno;
 186     dout(0) << "create_checkpoint: sync_filesystem got" << cpp_strerror(ret) << dendl;
 187     return ret;
 188   }
 189
 190   char path[PATH_MAX];
 191   snprintf(path, sizeof(path), "%s@%s", zfs.get_name(current_zh), name.c_str());
 192   ret = zfs.snapshot(path, false);
 193   if (ret < 0) {
 194     ret = -errno;
 195     dout(0) << "create_checkpoint: zfs_snapshot '" << path << "' got" << cpp_strerror(ret) << dendl;
 196     return ret;
 197   }
 198   if (cid)
 199     *cid = 0;
 200   return 0;
 201 }
 202
 203 int ZFSFileStoreBackend::rollback_to(const string& name)
 204 {
 205   dout(10) << "rollback_to: '" << name << "'" << dendl;
 206   if (!current_zh)
 207     return -EINVAL;
 208
 209   // umount current to avoid triggering online rollback deadlock
 210   int ret;
 211   if (zfs.is_mounted(current_zh, NULL)) {
 212     ret = zfs.umount(current_zh, NULL, 0);
 213     if (ret < 0) {
 214       ret = -errno;
 215       dout(0) << "rollback_to: zfs_umount '" << zfs.get_name(current_zh) << "' got" << cpp_strerror(ret) << dendl;
 216     }
 217   }
 218
 219   char path[PATH_MAX];
 220   snprintf(path, sizeof(path), "%s@%s", zfs.get_name(current_zh), name.c_str());
 221
 222   ZFS::Handle *snap_zh = zfs.open(path, ZFS::TYPE_SNAPSHOT);
 223   if (!snap_zh) {
 224     dout(0) << "rollback_to: zfs_open '" << path << "' got NULL" << dendl;
 225     return -ENOENT;
 226   }
 227
 228   ret = zfs.rollback(current_zh, snap_zh, false);
 229   if (ret < 0) {
 230     ret = -errno;
 231     dout(0) << "rollback_to: zfs_rollback '" << zfs.get_name(snap_zh) << "' got" << cpp_strerror(ret) << dendl;
 232   }
 233
 234   if (!zfs.is_mounted(current_zh, NULL)) {
 235     int ret = zfs.mount(current_zh, NULL, 0);
 236     if (ret < 0) {
 237       ret = -errno;
 238       dout(0) << "update_current_zh: zfs_mount '" << zfs.get_name(current_zh) << "' got " << cpp_strerror(ret) << dendl;
 239       return ret;
 240     }
 241   }
 242
 243   zfs.close(snap_zh);
 244   return ret;
 245 }
 246
 247 int ZFSFileStoreBackend::destroy_checkpoint(const string& name)
 248 {
 249   dout(10) << "destroy_checkpoint: '" << name << "'" << dendl;
 250   if (!current_zh)
 251     return -EINVAL;
 252
 253   int ret = zfs.destroy_snaps(current_zh, name.c_str(), true);
 254   if (ret < 0) {
 255     ret = -errno;
 256     dout(0) << "destroy_checkpoint: zfs_destroy_snaps '" << name << "' got" << cpp_strerror(ret) << dendl;
 257   }
 258   return ret;
 259 }
 260 #endif