mdadm.go 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434
  1. // Copyright 2019 Yunion
  2. //
  3. // Licensed under the Apache License, Version 2.0 (the "License");
  4. // you may not use this file except in compliance with the License.
  5. // You may obtain a copy of the License at
  6. //
  7. // http://www.apache.org/licenses/LICENSE-2.0
  8. //
  9. // Unless required by applicable law or agreed to in writing, software
  10. // distributed under the License is distributed on an "AS IS" BASIS,
  11. // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12. // See the License for the specific language governing permissions and
  13. // limitations under the License.
  14. package mdadm
  15. import (
  16. "fmt"
  17. "path"
  18. "regexp"
  19. "strconv"
  20. "strings"
  21. "time"
  22. "yunion.io/x/log"
  23. "yunion.io/x/pkg/errors"
  24. api "yunion.io/x/onecloud/pkg/apis/compute"
  25. "yunion.io/x/onecloud/pkg/baremetal/utils/raid"
  26. "yunion.io/x/onecloud/pkg/compute/baremetal"
  27. "yunion.io/x/onecloud/pkg/util/ssh"
  28. "yunion.io/x/onecloud/pkg/util/sysutils"
  29. )
  30. const (
  31. MDADM_BIN = "/sbin/mdadm"
  32. )
  33. func init() {
  34. raid.RegisterDriver(baremetal.DISK_DRIVER_LINUX, NewMdadmRaidLinux)
  35. raid.RegisterDriver(baremetal.DISK_DRIVER_PCIE, NewMdadmRaidPcie)
  36. }
  37. type MdadmRaid struct {
  38. term raid.IExecTerm
  39. adapter *MdadmRaidAdapter
  40. driverName string
  41. }
  42. func NewMdadmRaidLinux(term raid.IExecTerm) raid.IRaidDriver {
  43. return &MdadmRaid{
  44. term: term,
  45. driverName: baremetal.DISK_DRIVER_LINUX,
  46. }
  47. }
  48. func NewMdadmRaidPcie(term raid.IExecTerm) raid.IRaidDriver {
  49. return &MdadmRaid{
  50. term: term,
  51. driverName: baremetal.DISK_DRIVER_PCIE,
  52. }
  53. }
  54. func (r *MdadmRaid) GetName() string {
  55. return r.driverName
  56. }
  57. func (r *MdadmRaid) ParsePhyDevs() error {
  58. if r.adapter == nil {
  59. r.adapter = &MdadmRaidAdapter{
  60. raid: r,
  61. term: r.term,
  62. index: 0,
  63. }
  64. }
  65. return nil
  66. }
  67. func (r *MdadmRaid) SetDevicesForAdapter(adapterIdx int, devs []*baremetal.BaremetalStorage) {
  68. r.adapter.setDevices(devs)
  69. for i := range devs {
  70. devPath := path.Join("/dev", devs[i].Dev)
  71. cmd := fmt.Sprintf("%s --examine %s | grep UUID", MDADM_BIN, devPath)
  72. output, err := r.term.Run(cmd)
  73. if err == nil && len(output) > 0 {
  74. for _, line := range output {
  75. segs := strings.SplitN(strings.TrimSpace(line), ":", 2)
  76. if len(segs) == 2 {
  77. uuid := strings.TrimSpace(segs[1])
  78. cmd = fmt.Sprintf("%s --assemble --scan --uuid=%s", MDADM_BIN, uuid)
  79. output, err := r.term.Run(cmd)
  80. if err != nil {
  81. log.Errorf("faield assemble mdadm %s: %s", uuid, output)
  82. }
  83. }
  84. }
  85. }
  86. }
  87. }
  88. func (r *MdadmRaid) GetAdapters() []raid.IRaidAdapter {
  89. return []raid.IRaidAdapter{r.adapter}
  90. }
  91. func (r *MdadmRaid) PreBuildRaid(confs []*api.BaremetalDiskConfig, adapterIdx int) error {
  92. return nil
  93. }
  94. func deviceHasRaid(devPath string, term *ssh.Client) bool {
  95. cmd := fmt.Sprintf("%s --examine %s 2>/dev/null || true", MDADM_BIN, devPath)
  96. output, err := term.Run(cmd)
  97. if err != nil {
  98. log.Errorf("examine device %s: %s", devPath, err)
  99. return false
  100. }
  101. for _, line := range output {
  102. if strings.Contains(line, "mdadm") || strings.Contains(line, "ARRAY") {
  103. return true
  104. }
  105. }
  106. return false
  107. }
  108. func (r *MdadmRaid) CleanRaid() error {
  109. return nil
  110. }
  111. func CleanMdadmPartitions(term *ssh.Client) {
  112. out, err := term.Run("ls -1 /dev/md/")
  113. if err != nil {
  114. log.Errorf("failed get md devices %s, %s", out, err)
  115. return
  116. }
  117. // destory mdadm soft raid
  118. for _, line := range out {
  119. dev := strings.TrimSpace(line)
  120. if !strings.HasPrefix(dev, "md") {
  121. continue
  122. }
  123. out, err = term.Run(fmt.Sprintf("dd if=/dev/zero of=/dev/md/%s bs=512 count=34", dev))
  124. if err != nil {
  125. log.Errorf("failed clean mdadm partitions %s %s", out, err)
  126. }
  127. out, err = term.Run(fmt.Sprintf("dd if=/dev/zero of=/dev/md/%s bs=512 count=34 seek=$(( $(blockdev --getsz /dev/md/%s) - 34 ))", dev, dev))
  128. if err != nil {
  129. log.Errorf("failed clean mdadm partitions %s %s", out, err)
  130. }
  131. out, err = term.Run(fmt.Sprintf("hdparm -z /dev/md/%s", dev))
  132. if err != nil {
  133. log.Errorf("failed clean mdadm partitions %s %s", out, err)
  134. }
  135. }
  136. }
  137. func CleanRaid(term *ssh.Client) error {
  138. CleanMdadmPartitions(term)
  139. // stop md devices
  140. cmd := fmt.Sprintf("%s --stop --scan", MDADM_BIN)
  141. _, err := term.Run(cmd)
  142. if err != nil {
  143. log.Warningf("Stop md devices: %s", err)
  144. }
  145. pcieRet, err := term.Run("/lib/mos/lsdisk --pcie")
  146. if err != nil {
  147. log.Warningf("Fail to retrieve PCIE DISK info %s", err)
  148. } else {
  149. pcieDiskInfo := sysutils.ParsePCIEDiskInfo(pcieRet)
  150. for i := range pcieDiskInfo {
  151. devPath := path.Join("/dev", pcieDiskInfo[i].Dev)
  152. if deviceHasRaid(devPath, term) {
  153. cmd := fmt.Sprintf("%s --zero-superblock --force %s", MDADM_BIN, devPath)
  154. out, err := term.Run(cmd)
  155. if err != nil {
  156. return errors.Wrapf(err, "zero superblock on %s: %s", devPath, out)
  157. }
  158. }
  159. }
  160. }
  161. nonraidRet, err := term.Run("/lib/mos/lsdisk --nonraid")
  162. if err != nil {
  163. log.Warningf("Fail to retrieve SCSI DISK info %s", err)
  164. } else {
  165. nonraidDiskInfo := sysutils.ParseSCSIDiskInfo(nonraidRet)
  166. for i := range nonraidDiskInfo {
  167. devPath := path.Join("/dev", nonraidDiskInfo[i].Dev)
  168. if deviceHasRaid(devPath, term) {
  169. cmd := fmt.Sprintf("%s --zero-superblock --force %s", MDADM_BIN, devPath)
  170. out, err := term.Run(cmd)
  171. if err != nil {
  172. return errors.Wrapf(err, "zero superblock on %s: %s", devPath, out)
  173. }
  174. }
  175. }
  176. }
  177. out, err := term.Run("rm /dev/md/*")
  178. if err != nil {
  179. log.Warningf("failed soft link at /dev/md %s", out)
  180. }
  181. return nil
  182. }
  183. type MdadmRaidAdapter struct {
  184. raid *MdadmRaid
  185. term raid.IExecTerm
  186. index int
  187. devs []*baremetal.BaremetalStorage
  188. }
  189. func (a *MdadmRaidAdapter) GetIndex() int {
  190. return a.index
  191. }
  192. func (a *MdadmRaidAdapter) PreBuildRaid(confs []*api.BaremetalDiskConfig) error {
  193. return nil
  194. }
  195. func (a *MdadmRaidAdapter) GetLogicVolumes() ([]*raid.RaidLogicalVolume, error) {
  196. lvs := make([]*raid.RaidLogicalVolume, 0)
  197. cmd := "ls -1 /dev/md/* 2>/dev/null || true"
  198. output, err := a.term.Run(cmd)
  199. if err != nil {
  200. return lvs, nil
  201. }
  202. for _, line := range output {
  203. line = strings.TrimSpace(line)
  204. if strings.HasPrefix(line, "/dev/md/md") {
  205. mdPath := line
  206. numStr := strings.TrimPrefix(line, "/dev/md/md")
  207. if strings.HasSuffix(numStr, "_0") {
  208. numStr = strings.TrimSuffix(numStr, "_0")
  209. }
  210. if num, err := strconv.Atoi(numStr); err == nil {
  211. res, err := a.term.Run(fmt.Sprintf("readlink -f %s", line))
  212. if err == nil && len(res) > 0 {
  213. mdPath = strings.TrimSpace(res[0])
  214. lv := &raid.RaidLogicalVolume{
  215. Index: num,
  216. Adapter: a.index,
  217. BlockDev: mdPath,
  218. }
  219. lvs = append(lvs, lv)
  220. }
  221. }
  222. }
  223. }
  224. return lvs, nil
  225. }
  226. func (a *MdadmRaidAdapter) RemoveLogicVolumes() error {
  227. //cmd := fmt.Sprintf("%s --stop --scan", MDADM_BIN)
  228. //_, err := a.term.Run(cmd)
  229. //if err != nil {
  230. // log.Warningf("Stop md devices: %v", err)
  231. //}
  232. return nil
  233. }
  234. func (a *MdadmRaidAdapter) GetDevices() []*baremetal.BaremetalStorage {
  235. return a.devs
  236. }
  237. func (a *MdadmRaidAdapter) setDevices(devs []*baremetal.BaremetalStorage) {
  238. a.devs = devs
  239. }
  240. func (a *MdadmRaidAdapter) BuildRaid0(devs []*baremetal.BaremetalStorage, conf *api.BaremetalDiskConfig) error {
  241. return a.buildRaid("0", devs, conf)
  242. }
  243. func (a *MdadmRaidAdapter) BuildRaid1(devs []*baremetal.BaremetalStorage, conf *api.BaremetalDiskConfig) error {
  244. return a.buildRaid("1", devs, conf)
  245. }
  246. func (a *MdadmRaidAdapter) BuildRaid5(devs []*baremetal.BaremetalStorage, conf *api.BaremetalDiskConfig) error {
  247. return a.buildRaid("5", devs, conf)
  248. }
  249. func (a *MdadmRaidAdapter) BuildRaid10(devs []*baremetal.BaremetalStorage, conf *api.BaremetalDiskConfig) error {
  250. return a.buildRaid("10", devs, conf)
  251. }
  252. func (a *MdadmRaidAdapter) BuildNoneRaid(devs []*baremetal.BaremetalStorage) error {
  253. return nil
  254. }
  255. func (a *MdadmRaidAdapter) PostBuildRaid() error {
  256. return nil
  257. }
  258. func (a *MdadmRaidAdapter) buildRaid(level string, devs []*baremetal.BaremetalStorage, conf *api.BaremetalDiskConfig) error {
  259. if len(devs) == 0 {
  260. return fmt.Errorf("no devices provided for RAID %s", level)
  261. }
  262. var mdNum int
  263. var err error
  264. if conf.SoftRaidIdx != nil {
  265. mdNum = *conf.SoftRaidIdx
  266. } else {
  267. mdNum, err = a.getNextMdNum()
  268. if err != nil {
  269. return errors.Wrap(err, "get next md number")
  270. }
  271. }
  272. devPaths := make([]string, 0, len(devs))
  273. for _, dev := range devs {
  274. if dev.Dev == "" {
  275. return fmt.Errorf("device path is empty for storage")
  276. }
  277. devPaths = append(devPaths, path.Join("/dev", dev.Dev))
  278. }
  279. for _, dev := range devPaths {
  280. if err := a.ensureDeviceClean(dev); err != nil {
  281. return errors.Wrapf(err, "clean device %s", dev)
  282. }
  283. }
  284. mdDev := fmt.Sprintf("/dev/md/md%d", mdNum)
  285. imsmDev := fmt.Sprintf("/dev/md/imsm%d", mdNum)
  286. cmdImsm := fmt.Sprintf("%s --create %s --metadata=imsm --raid-devices=%d --run --force %s", MDADM_BIN, imsmDev, len(devs), strings.Join(devPaths, " "))
  287. output, err := a.term.Run(cmdImsm)
  288. if err != nil {
  289. log.Errorf("mdadm create imsm raid %s failed, output: %v %s", level, output, err)
  290. } else {
  291. a.term.Run(fmt.Sprintf("%s --wait %s", MDADM_BIN, imsmDev))
  292. time.Sleep(time.Second * 3)
  293. }
  294. args := []string{
  295. "--create",
  296. mdDev,
  297. fmt.Sprintf("--level=%s", level),
  298. fmt.Sprintf("--raid-devices=%d", len(devs)),
  299. "--force",
  300. "--run",
  301. }
  302. for _, dev := range devPaths {
  303. args = append(args, dev)
  304. }
  305. args = append(args, "--assume-clean")
  306. cmd := fmt.Sprintf("%s %s", MDADM_BIN, strings.Join(args, " "))
  307. log.Infof("Building software RAID %s: %s", level, cmd)
  308. output, err = a.term.Run(cmd)
  309. if err != nil {
  310. return errors.Wrapf(err, "mdadm create raid %s failed, output: %v", level, output)
  311. }
  312. cmd = fmt.Sprintf("%s --wait %s", MDADM_BIN, mdDev)
  313. output, err = a.term.Run(cmd)
  314. if err != nil {
  315. log.Errorf("mdadm wait raid %s failed: %s", mdDev, output)
  316. //return errors.Wrapf(err, "mdadm wait raid %s failed, output: %v", mdDev, output)
  317. }
  318. log.Infof("Successfully created software RAID %s: /dev/md/md%d, start sync block devs", level, mdNum)
  319. for i := range devPaths {
  320. flushCmd := fmt.Sprintf("blockdev --flushbufs %s", devPaths[i])
  321. output, err = a.term.Run(flushCmd)
  322. if err != nil {
  323. return errors.Wrapf(err, "mdadm blockdev flushbufs %s failed, output: %v", devPaths[i], output)
  324. }
  325. }
  326. output, err = a.term.Run("sync")
  327. if err != nil {
  328. return errors.Wrapf(err, "mdadm %s sync failed, output: %v", mdDev, output)
  329. }
  330. return nil
  331. }
  332. func (a *MdadmRaidAdapter) getNextMdNum() (int, error) {
  333. cmd := "ls -1 /dev/md/ 2>/dev/null | grep -E '/dev/md/md[0-9]+$' || true"
  334. output, err := a.term.Run(cmd)
  335. if err != nil {
  336. return 0, errors.Wrap(err, "list md devices")
  337. }
  338. usedNums := make(map[int]bool)
  339. mdNumRe := regexp.MustCompile(`/dev/md/md(\d+)`)
  340. for _, line := range output {
  341. matches := mdNumRe.FindStringSubmatch(line)
  342. if len(matches) > 1 {
  343. if num, err := strconv.Atoi(matches[1]); err == nil {
  344. usedNums[num] = true
  345. }
  346. }
  347. }
  348. for i := 0; i < 256; i++ {
  349. if !usedNums[i] {
  350. return i, nil
  351. }
  352. }
  353. return 0, fmt.Errorf("no available md device number")
  354. }
  355. func (a *MdadmRaidAdapter) ensureDeviceClean(dev string) error {
  356. cmd := fmt.Sprintf("%s --examine %s 2>/dev/null || true", MDADM_BIN, dev)
  357. output, err := a.term.Run(cmd)
  358. if err != nil {
  359. return errors.Wrapf(err, "examine device %s", dev)
  360. }
  361. for _, line := range output {
  362. if strings.Contains(line, "mdadm") || strings.Contains(line, "ARRAY") {
  363. cmd := fmt.Sprintf("%s --zero-superblock --force %s", MDADM_BIN, dev)
  364. _, err := a.term.Run(cmd)
  365. if err != nil {
  366. return errors.Wrapf(err, "zero superblock on %s", dev)
  367. }
  368. break
  369. }
  370. }
  371. return nil
  372. }