mds_client.h 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577
  1. /* SPDX-License-Identifier: GPL-2.0 */
  2. #ifndef _FS_CEPH_MDS_CLIENT_H
  3. #define _FS_CEPH_MDS_CLIENT_H
  4. #include <linux/completion.h>
  5. #include <linux/kref.h>
  6. #include <linux/list.h>
  7. #include <linux/mutex.h>
  8. #include <linux/rbtree.h>
  9. #include <linux/spinlock.h>
  10. #include <linux/refcount.h>
  11. #include <linux/utsname.h>
  12. #include <linux/ktime.h>
  13. #include <linux/ceph/types.h>
  14. #include <linux/ceph/messenger.h>
  15. #include <linux/ceph/mdsmap.h>
  16. #include <linux/ceph/auth.h>
  17. #include "metric.h"
  18. #include "super.h"
  19. /* The first 8 bits are reserved for old ceph releases */
  20. enum ceph_feature_type {
  21. CEPHFS_FEATURE_MIMIC = 8,
  22. CEPHFS_FEATURE_REPLY_ENCODING,
  23. CEPHFS_FEATURE_RECLAIM_CLIENT,
  24. CEPHFS_FEATURE_LAZY_CAP_WANTED,
  25. CEPHFS_FEATURE_MULTI_RECONNECT,
  26. CEPHFS_FEATURE_DELEG_INO,
  27. CEPHFS_FEATURE_METRIC_COLLECT,
  28. CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT,
  29. };
  30. /*
  31. * This will always have the highest feature bit value
  32. * as the last element of the array.
  33. */
  34. #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
  35. 0, 1, 2, 3, 4, 5, 6, 7, \
  36. CEPHFS_FEATURE_MIMIC, \
  37. CEPHFS_FEATURE_REPLY_ENCODING, \
  38. CEPHFS_FEATURE_LAZY_CAP_WANTED, \
  39. CEPHFS_FEATURE_MULTI_RECONNECT, \
  40. CEPHFS_FEATURE_DELEG_INO, \
  41. CEPHFS_FEATURE_METRIC_COLLECT, \
  42. \
  43. CEPHFS_FEATURE_MAX, \
  44. }
  45. #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
  46. /*
  47. * Some lock dependencies:
  48. *
  49. * session->s_mutex
  50. * mdsc->mutex
  51. *
  52. * mdsc->snap_rwsem
  53. *
  54. * ci->i_ceph_lock
  55. * mdsc->snap_flush_lock
  56. * mdsc->cap_delay_lock
  57. *
  58. */
  59. struct ceph_fs_client;
  60. struct ceph_cap;
  61. /*
  62. * parsed info about a single inode. pointers are into the encoded
  63. * on-wire structures within the mds reply message payload.
  64. */
  65. struct ceph_mds_reply_info_in {
  66. struct ceph_mds_reply_inode *in;
  67. struct ceph_dir_layout dir_layout;
  68. u32 symlink_len;
  69. char *symlink;
  70. u32 xattr_len;
  71. char *xattr_data;
  72. u64 inline_version;
  73. u32 inline_len;
  74. char *inline_data;
  75. u32 pool_ns_len;
  76. char *pool_ns_data;
  77. u64 max_bytes;
  78. u64 max_files;
  79. s32 dir_pin;
  80. struct ceph_timespec btime;
  81. struct ceph_timespec snap_btime;
  82. u64 change_attr;
  83. };
  84. struct ceph_mds_reply_dir_entry {
  85. char *name;
  86. u32 name_len;
  87. struct ceph_mds_reply_lease *lease;
  88. struct ceph_mds_reply_info_in inode;
  89. loff_t offset;
  90. };
  91. /*
  92. * parsed info about an mds reply, including information about
  93. * either: 1) the target inode and/or its parent directory and dentry,
  94. * and directory contents (for readdir results), or
  95. * 2) the file range lock info (for fcntl F_GETLK results).
  96. */
  97. struct ceph_mds_reply_info_parsed {
  98. struct ceph_mds_reply_head *head;
  99. /* trace */
  100. struct ceph_mds_reply_info_in diri, targeti;
  101. struct ceph_mds_reply_dirfrag *dirfrag;
  102. char *dname;
  103. u32 dname_len;
  104. struct ceph_mds_reply_lease *dlease;
  105. /* extra */
  106. union {
  107. /* for fcntl F_GETLK results */
  108. struct ceph_filelock *filelock_reply;
  109. /* for readdir results */
  110. struct {
  111. struct ceph_mds_reply_dirfrag *dir_dir;
  112. size_t dir_buf_size;
  113. int dir_nr;
  114. bool dir_end;
  115. bool dir_complete;
  116. bool hash_order;
  117. bool offset_hash;
  118. struct ceph_mds_reply_dir_entry *dir_entries;
  119. };
  120. /* for create results */
  121. struct {
  122. bool has_create_ino;
  123. u64 ino;
  124. };
  125. };
  126. /* encoded blob describing snapshot contexts for certain
  127. operations (e.g., open) */
  128. void *snapblob;
  129. int snapblob_len;
  130. };
  131. /*
  132. * cap releases are batched and sent to the MDS en masse.
  133. *
  134. * Account for per-message overhead of mds_cap_release header
  135. * and __le32 for osd epoch barrier trailing field.
  136. */
  137. #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
  138. sizeof(struct ceph_mds_cap_release)) / \
  139. sizeof(struct ceph_mds_cap_item))
  140. /*
  141. * state associated with each MDS<->client session
  142. */
  143. enum {
  144. CEPH_MDS_SESSION_NEW = 1,
  145. CEPH_MDS_SESSION_OPENING = 2,
  146. CEPH_MDS_SESSION_OPEN = 3,
  147. CEPH_MDS_SESSION_HUNG = 4,
  148. CEPH_MDS_SESSION_RESTARTING = 5,
  149. CEPH_MDS_SESSION_RECONNECTING = 6,
  150. CEPH_MDS_SESSION_CLOSING = 7,
  151. CEPH_MDS_SESSION_CLOSED = 8,
  152. CEPH_MDS_SESSION_REJECTED = 9,
  153. };
  154. struct ceph_mds_session {
  155. struct ceph_mds_client *s_mdsc;
  156. int s_mds;
  157. int s_state;
  158. unsigned long s_ttl; /* time until mds kills us */
  159. unsigned long s_features;
  160. u64 s_seq; /* incoming msg seq # */
  161. struct mutex s_mutex; /* serialize session messages */
  162. struct ceph_connection s_con;
  163. struct ceph_auth_handshake s_auth;
  164. /* protected by s_gen_ttl_lock */
  165. spinlock_t s_gen_ttl_lock;
  166. u32 s_cap_gen; /* inc each time we get mds stale msg */
  167. unsigned long s_cap_ttl; /* when session caps expire */
  168. /* protected by s_cap_lock */
  169. spinlock_t s_cap_lock;
  170. refcount_t s_ref;
  171. struct list_head s_caps; /* all caps issued by this session */
  172. struct ceph_cap *s_cap_iterator;
  173. int s_nr_caps;
  174. int s_num_cap_releases;
  175. int s_cap_reconnect;
  176. int s_readonly;
  177. struct list_head s_cap_releases; /* waiting cap_release messages */
  178. struct work_struct s_cap_release_work;
  179. /* See ceph_inode_info->i_dirty_item. */
  180. struct list_head s_cap_dirty; /* inodes w/ dirty caps */
  181. /* See ceph_inode_info->i_flushing_item. */
  182. struct list_head s_cap_flushing; /* inodes w/ flushing caps */
  183. unsigned long s_renew_requested; /* last time we sent a renew req */
  184. u64 s_renew_seq;
  185. struct list_head s_waiting; /* waiting requests */
  186. struct list_head s_unsafe; /* unsafe requests */
  187. struct xarray s_delegated_inos;
  188. };
  189. /*
  190. * modes of choosing which MDS to send a request to
  191. */
  192. enum {
  193. USE_ANY_MDS,
  194. USE_RANDOM_MDS,
  195. USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
  196. };
  197. struct ceph_mds_request;
  198. struct ceph_mds_client;
  199. /*
  200. * request completion callback
  201. */
  202. typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
  203. struct ceph_mds_request *req);
  204. /*
  205. * wait for request completion callback
  206. */
  207. typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
  208. struct ceph_mds_request *req);
  209. /*
  210. * an in-flight mds request
  211. */
  212. struct ceph_mds_request {
  213. u64 r_tid; /* transaction id */
  214. struct rb_node r_node;
  215. struct ceph_mds_client *r_mdsc;
  216. struct kref r_kref;
  217. int r_op; /* mds op code */
  218. /* operation on what? */
  219. struct inode *r_inode; /* arg1 */
  220. struct dentry *r_dentry; /* arg1 */
  221. struct dentry *r_old_dentry; /* arg2: rename from or link from */
  222. struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
  223. char *r_path1, *r_path2;
  224. struct ceph_vino r_ino1, r_ino2;
  225. struct inode *r_parent; /* parent dir inode */
  226. struct inode *r_target_inode; /* resulting inode */
  227. #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
  228. #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
  229. #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
  230. #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
  231. #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
  232. #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
  233. #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
  234. #define CEPH_MDS_R_ASYNC (8) /* async request */
  235. unsigned long r_req_flags;
  236. struct mutex r_fill_mutex;
  237. union ceph_mds_request_args r_args;
  238. int r_fmode; /* file mode, if expecting cap */
  239. kuid_t r_uid;
  240. kgid_t r_gid;
  241. int r_request_release_offset;
  242. struct timespec64 r_stamp;
  243. /* for choosing which mds to send this request to */
  244. int r_direct_mode;
  245. u32 r_direct_hash; /* choose dir frag based on this dentry hash */
  246. /* data payload is used for xattr ops */
  247. struct ceph_pagelist *r_pagelist;
  248. /* what caps shall we drop? */
  249. int r_inode_drop, r_inode_unless;
  250. int r_dentry_drop, r_dentry_unless;
  251. int r_old_dentry_drop, r_old_dentry_unless;
  252. struct inode *r_old_inode;
  253. int r_old_inode_drop, r_old_inode_unless;
  254. struct ceph_msg *r_request; /* original request */
  255. struct ceph_msg *r_reply;
  256. struct ceph_mds_reply_info_parsed r_reply_info;
  257. int r_err;
  258. struct page *r_locked_page;
  259. int r_dir_caps;
  260. int r_num_caps;
  261. u32 r_readdir_offset;
  262. unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
  263. unsigned long r_started; /* start time to measure timeout against */
  264. unsigned long r_start_latency; /* start time to measure latency */
  265. unsigned long r_end_latency; /* finish time to measure latency */
  266. unsigned long r_request_started; /* start time for mds request only,
  267. used to measure lease durations */
  268. /* link unsafe requests to parent directory, for fsync */
  269. struct inode *r_unsafe_dir;
  270. struct list_head r_unsafe_dir_item;
  271. /* unsafe requests that modify the target inode */
  272. struct list_head r_unsafe_target_item;
  273. struct ceph_mds_session *r_session;
  274. int r_attempts; /* resend attempts */
  275. int r_num_fwd; /* number of forward attempts */
  276. int r_resend_mds; /* mds to resend to next, if any*/
  277. u32 r_sent_on_mseq; /* cap mseq request was sent at*/
  278. u64 r_deleg_ino;
  279. struct list_head r_wait;
  280. struct completion r_completion;
  281. struct completion r_safe_completion;
  282. ceph_mds_request_callback_t r_callback;
  283. ceph_mds_request_wait_callback_t r_wait_for_completion;
  284. struct list_head r_unsafe_item; /* per-session unsafe list item */
  285. long long r_dir_release_cnt;
  286. long long r_dir_ordered_cnt;
  287. int r_readdir_cache_idx;
  288. struct ceph_cap_reservation r_caps_reservation;
  289. };
  290. struct ceph_pool_perm {
  291. struct rb_node node;
  292. int perm;
  293. s64 pool;
  294. size_t pool_ns_len;
  295. char pool_ns[];
  296. };
  297. struct ceph_snapid_map {
  298. struct rb_node node;
  299. struct list_head lru;
  300. atomic_t ref;
  301. u64 snap;
  302. dev_t dev;
  303. unsigned long last_used;
  304. };
  305. /*
  306. * node for list of quotarealm inodes that are not visible from the filesystem
  307. * mountpoint, but required to handle, e.g. quotas.
  308. */
  309. struct ceph_quotarealm_inode {
  310. struct rb_node node;
  311. u64 ino;
  312. unsigned long timeout; /* last time a lookup failed for this inode */
  313. struct mutex mutex;
  314. struct inode *inode;
  315. };
  316. struct cap_wait {
  317. struct list_head list;
  318. u64 ino;
  319. pid_t tgid;
  320. int need;
  321. int want;
  322. };
  323. /*
  324. * mds client state
  325. */
  326. struct ceph_mds_client {
  327. struct ceph_fs_client *fsc;
  328. struct mutex mutex; /* all nested structures */
  329. struct ceph_mdsmap *mdsmap;
  330. struct completion safe_umount_waiters;
  331. wait_queue_head_t session_close_wq;
  332. struct list_head waiting_for_map;
  333. int mdsmap_err;
  334. struct ceph_mds_session **sessions; /* NULL for mds if no session */
  335. atomic_t num_sessions;
  336. int max_sessions; /* len of sessions array */
  337. int stopping; /* true if shutting down */
  338. atomic64_t quotarealms_count; /* # realms with quota */
  339. /*
  340. * We keep a list of inodes we don't see in the mountpoint but that we
  341. * need to track quota realms.
  342. */
  343. struct rb_root quotarealms_inodes;
  344. struct mutex quotarealms_inodes_mutex;
  345. /*
  346. * snap_rwsem will cover cap linkage into snaprealms, and
  347. * realm snap contexts. (later, we can do per-realm snap
  348. * contexts locks..) the empty list contains realms with no
  349. * references (implying they contain no inodes with caps) that
  350. * should be destroyed.
  351. */
  352. u64 last_snap_seq;
  353. struct rw_semaphore snap_rwsem;
  354. struct rb_root snap_realms;
  355. struct list_head snap_empty;
  356. int num_snap_realms;
  357. spinlock_t snap_empty_lock; /* protect snap_empty */
  358. u64 last_tid; /* most recent mds request */
  359. u64 oldest_tid; /* oldest incomplete mds request,
  360. excluding setfilelock requests */
  361. struct rb_root request_tree; /* pending mds requests */
  362. struct delayed_work delayed_work; /* delayed work */
  363. unsigned long last_renew_caps; /* last time we renewed our caps */
  364. struct list_head cap_delay_list; /* caps with delayed release */
  365. spinlock_t cap_delay_lock; /* protects cap_delay_list */
  366. struct list_head snap_flush_list; /* cap_snaps ready to flush */
  367. spinlock_t snap_flush_lock;
  368. u64 last_cap_flush_tid;
  369. struct list_head cap_flush_list;
  370. struct list_head cap_dirty_migrating; /* ...that are migration... */
  371. int num_cap_flushing; /* # caps we are flushing */
  372. spinlock_t cap_dirty_lock; /* protects above items */
  373. wait_queue_head_t cap_flushing_wq;
  374. struct work_struct cap_reclaim_work;
  375. atomic_t cap_reclaim_pending;
  376. /*
  377. * Cap reservations
  378. *
  379. * Maintain a global pool of preallocated struct ceph_caps, referenced
  380. * by struct ceph_caps_reservations. This ensures that we preallocate
  381. * memory needed to successfully process an MDS response. (If an MDS
  382. * sends us cap information and we fail to process it, we will have
  383. * problems due to the client and MDS being out of sync.)
  384. *
  385. * Reservations are 'owned' by a ceph_cap_reservation context.
  386. */
  387. spinlock_t caps_list_lock;
  388. struct list_head caps_list; /* unused (reserved or
  389. unreserved) */
  390. struct list_head cap_wait_list;
  391. int caps_total_count; /* total caps allocated */
  392. int caps_use_count; /* in use */
  393. int caps_use_max; /* max used caps */
  394. int caps_reserve_count; /* unused, reserved */
  395. int caps_avail_count; /* unused, unreserved */
  396. int caps_min_count; /* keep at least this many
  397. (unreserved) */
  398. spinlock_t dentry_list_lock;
  399. struct list_head dentry_leases; /* fifo list */
  400. struct list_head dentry_dir_leases; /* lru list */
  401. struct ceph_client_metric metric;
  402. spinlock_t snapid_map_lock;
  403. struct rb_root snapid_map_tree;
  404. struct list_head snapid_map_lru;
  405. struct rw_semaphore pool_perm_rwsem;
  406. struct rb_root pool_perm_tree;
  407. char nodename[__NEW_UTS_LEN + 1];
  408. };
  409. extern const char *ceph_mds_op_name(int op);
  410. extern bool check_session_state(struct ceph_mds_session *s);
  411. void inc_session_sequence(struct ceph_mds_session *s);
  412. extern struct ceph_mds_session *
  413. __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
  414. extern const char *ceph_session_state_name(int s);
  415. extern struct ceph_mds_session *
  416. ceph_get_mds_session(struct ceph_mds_session *s);
  417. extern void ceph_put_mds_session(struct ceph_mds_session *s);
  418. extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
  419. struct ceph_msg *msg, int mds);
  420. extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
  421. extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
  422. extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
  423. extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
  424. extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
  425. extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
  426. extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
  427. struct inode *dir);
  428. extern struct ceph_mds_request *
  429. ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
  430. extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
  431. struct inode *dir,
  432. struct ceph_mds_request *req);
  433. extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
  434. struct inode *dir,
  435. struct ceph_mds_request *req);
  436. extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
  437. extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
  438. static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
  439. {
  440. kref_get(&req->r_kref);
  441. }
  442. extern void ceph_mdsc_release_request(struct kref *kref);
  443. static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
  444. {
  445. kref_put(&req->r_kref, ceph_mdsc_release_request);
  446. }
  447. extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
  448. struct ceph_cap *cap);
  449. extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
  450. struct ceph_mds_session *session);
  451. extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
  452. extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
  453. extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
  454. int (*cb)(struct inode *,
  455. struct ceph_cap *, void *),
  456. void *arg);
  457. extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
  458. static inline void ceph_mdsc_free_path(char *path, int len)
  459. {
  460. if (!IS_ERR_OR_NULL(path))
  461. __putname(path - (PATH_MAX - 1 - len));
  462. }
  463. extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
  464. int stop_on_nosnap);
  465. extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
  466. extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
  467. struct dentry *dentry, char action,
  468. u32 seq);
  469. extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
  470. struct ceph_msg *msg);
  471. extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
  472. struct ceph_msg *msg);
  473. extern struct ceph_mds_session *
  474. ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
  475. extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
  476. struct ceph_mds_session *session);
  477. extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
  478. struct ceph_mds_session *session,
  479. int max_caps);
  480. static inline int ceph_wait_on_async_create(struct inode *inode)
  481. {
  482. struct ceph_inode_info *ci = ceph_inode(inode);
  483. return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
  484. TASK_INTERRUPTIBLE);
  485. }
  486. extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
  487. extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
  488. #endif