1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef _FS_CEPH_MDS_CLIENT_H 3#define _FS_CEPH_MDS_CLIENT_H 4 5#include <linux/completion.h> 6#include <linux/kref.h> 7#include <linux/list.h> 8#include <linux/mutex.h> 9#include <linux/rbtree.h> 10#include <linux/spinlock.h> 11#include <linux/refcount.h> 12#include <linux/utsname.h> 13#include <linux/ktime.h> 14 15#include <linux/ceph/types.h> 16#include <linux/ceph/messenger.h> 17#include <linux/ceph/mdsmap.h> 18#include <linux/ceph/auth.h> 19 20#include "metric.h" 21#include "super.h" 22 23/* The first 8 bits are reserved for old ceph releases */ 24enum ceph_feature_type { 25 CEPHFS_FEATURE_MIMIC = 8, 26 CEPHFS_FEATURE_REPLY_ENCODING, 27 CEPHFS_FEATURE_RECLAIM_CLIENT, 28 CEPHFS_FEATURE_LAZY_CAP_WANTED, 29 CEPHFS_FEATURE_MULTI_RECONNECT, 30 CEPHFS_FEATURE_DELEG_INO, 31 CEPHFS_FEATURE_METRIC_COLLECT, 32 33 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT, 34}; 35 36#define CEPHFS_FEATURES_CLIENT_SUPPORTED { \ 37 0, 1, 2, 3, 4, 5, 6, 7, \ 38 CEPHFS_FEATURE_MIMIC, \ 39 CEPHFS_FEATURE_REPLY_ENCODING, \ 40 CEPHFS_FEATURE_LAZY_CAP_WANTED, \ 41 CEPHFS_FEATURE_MULTI_RECONNECT, \ 42 CEPHFS_FEATURE_DELEG_INO, \ 43 CEPHFS_FEATURE_METRIC_COLLECT, \ 44} 45#define CEPHFS_FEATURES_CLIENT_REQUIRED {} 46 47/* 48 * Some lock dependencies: 49 * 50 * session->s_mutex 51 * mdsc->mutex 52 * 53 * mdsc->snap_rwsem 54 * 55 * ci->i_ceph_lock 56 * mdsc->snap_flush_lock 57 * mdsc->cap_delay_lock 58 * 59 */ 60 61struct ceph_fs_client; 62struct ceph_cap; 63 64/* 65 * parsed info about a single inode. pointers are into the encoded 66 * on-wire structures within the mds reply message payload. 67 */ 68struct ceph_mds_reply_info_in { 69 struct ceph_mds_reply_inode *in; 70 struct ceph_dir_layout dir_layout; 71 u32 symlink_len; 72 char *symlink; 73 u32 xattr_len; 74 char *xattr_data; 75 u64 inline_version; 76 u32 inline_len; 77 char *inline_data; 78 u32 pool_ns_len; 79 char *pool_ns_data; 80 u64 max_bytes; 81 u64 max_files; 82 s32 dir_pin; 83 struct ceph_timespec btime; 84 struct ceph_timespec snap_btime; 85 u64 change_attr; 86}; 87 88struct ceph_mds_reply_dir_entry { 89 char *name; 90 u32 name_len; 91 struct ceph_mds_reply_lease *lease; 92 struct ceph_mds_reply_info_in inode; 93 loff_t offset; 94}; 95 96/* 97 * parsed info about an mds reply, including information about 98 * either: 1) the target inode and/or its parent directory and dentry, 99 * and directory contents (for readdir results), or 100 * 2) the file range lock info (for fcntl F_GETLK results). 101 */ 102struct ceph_mds_reply_info_parsed { 103 struct ceph_mds_reply_head *head; 104 105 /* trace */ 106 struct ceph_mds_reply_info_in diri, targeti; 107 struct ceph_mds_reply_dirfrag *dirfrag; 108 char *dname; 109 u32 dname_len; 110 struct ceph_mds_reply_lease *dlease; 111 112 /* extra */ 113 union { 114 /* for fcntl F_GETLK results */ 115 struct ceph_filelock *filelock_reply; 116 117 /* for readdir results */ 118 struct { 119 struct ceph_mds_reply_dirfrag *dir_dir; 120 size_t dir_buf_size; 121 int dir_nr; 122 bool dir_end; 123 bool dir_complete; 124 bool hash_order; 125 bool offset_hash; 126 struct ceph_mds_reply_dir_entry *dir_entries; 127 }; 128 129 /* for create results */ 130 struct { 131 bool has_create_ino; 132 u64 ino; 133 }; 134 }; 135 136 /* encoded blob describing snapshot contexts for certain 137 operations (e.g., open) */ 138 void *snapblob; 139 int snapblob_len; 140}; 141 142 143/* 144 * cap releases are batched and sent to the MDS en masse. 145 * 146 * Account for per-message overhead of mds_cap_release header 147 * and __le32 for osd epoch barrier trailing field. 148 */ 149#define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \ 150 sizeof(struct ceph_mds_cap_release)) / \ 151 sizeof(struct ceph_mds_cap_item)) 152 153 154/* 155 * state associated with each MDS<->client session 156 */ 157enum { 158 CEPH_MDS_SESSION_NEW = 1, 159 CEPH_MDS_SESSION_OPENING = 2, 160 CEPH_MDS_SESSION_OPEN = 3, 161 CEPH_MDS_SESSION_HUNG = 4, 162 CEPH_MDS_SESSION_RESTARTING = 5, 163 CEPH_MDS_SESSION_RECONNECTING = 6, 164 CEPH_MDS_SESSION_CLOSING = 7, 165 CEPH_MDS_SESSION_CLOSED = 8, 166 CEPH_MDS_SESSION_REJECTED = 9, 167}; 168 169struct ceph_mds_session { 170 struct ceph_mds_client *s_mdsc; 171 int s_mds; 172 int s_state; 173 unsigned long s_ttl; /* time until mds kills us */ 174 unsigned long s_features; 175 u64 s_seq; /* incoming msg seq # */ 176 struct mutex s_mutex; /* serialize session messages */ 177 178 struct ceph_connection s_con; 179 180 struct ceph_auth_handshake s_auth; 181 182 /* protected by s_gen_ttl_lock */ 183 spinlock_t s_gen_ttl_lock; 184 u32 s_cap_gen; /* inc each time we get mds stale msg */ 185 unsigned long s_cap_ttl; /* when session caps expire */ 186 187 /* protected by s_cap_lock */ 188 spinlock_t s_cap_lock; 189 refcount_t s_ref; 190 struct list_head s_caps; /* all caps issued by this session */ 191 struct ceph_cap *s_cap_iterator; 192 int s_nr_caps; 193 int s_num_cap_releases; 194 int s_cap_reconnect; 195 int s_readonly; 196 struct list_head s_cap_releases; /* waiting cap_release messages */ 197 struct work_struct s_cap_release_work; 198 199 /* See ceph_inode_info->i_dirty_item. */ 200 struct list_head s_cap_dirty; /* inodes w/ dirty caps */ 201 202 /* See ceph_inode_info->i_flushing_item. */ 203 struct list_head s_cap_flushing; /* inodes w/ flushing caps */ 204 205 unsigned long s_renew_requested; /* last time we sent a renew req */ 206 u64 s_renew_seq; 207 208 struct list_head s_waiting; /* waiting requests */ 209 struct list_head s_unsafe; /* unsafe requests */ 210 struct xarray s_delegated_inos; 211}; 212 213/* 214 * modes of choosing which MDS to send a request to 215 */ 216enum { 217 USE_ANY_MDS, 218 USE_RANDOM_MDS, 219 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */ 220}; 221 222struct ceph_mds_request; 223struct ceph_mds_client; 224 225/* 226 * request completion callback 227 */ 228typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc, 229 struct ceph_mds_request *req); 230/* 231 * wait for request completion callback 232 */ 233typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc, 234 struct ceph_mds_request *req); 235 236/* 237 * an in-flight mds request 238 */ 239struct ceph_mds_request { 240 u64 r_tid; /* transaction id */ 241 struct rb_node r_node; 242 struct ceph_mds_client *r_mdsc; 243 244 struct kref r_kref; 245 int r_op; /* mds op code */ 246 247 /* operation on what? */ 248 struct inode *r_inode; /* arg1 */ 249 struct dentry *r_dentry; /* arg1 */ 250 struct dentry *r_old_dentry; /* arg2: rename from or link from */ 251 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */ 252 char *r_path1, *r_path2; 253 struct ceph_vino r_ino1, r_ino2; 254 255 struct inode *r_parent; /* parent dir inode */ 256 struct inode *r_target_inode; /* resulting inode */ 257 258#define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */ 259#define CEPH_MDS_R_ABORTED (2) /* call was aborted */ 260#define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */ 261#define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */ 262#define CEPH_MDS_R_GOT_RESULT (5) /* got a result */ 263#define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */ 264#define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */ 265#define CEPH_MDS_R_ASYNC (8) /* async request */ 266 unsigned long r_req_flags; 267 268 struct mutex r_fill_mutex; 269 270 union ceph_mds_request_args r_args; 271 int r_fmode; /* file mode, if expecting cap */ 272 kuid_t r_uid; 273 kgid_t r_gid; 274 int r_request_release_offset; 275 struct timespec64 r_stamp; 276 277 /* for choosing which mds to send this request to */ 278 int r_direct_mode; 279 u32 r_direct_hash; /* choose dir frag based on this dentry hash */ 280 281 /* data payload is used for xattr ops */ 282 struct ceph_pagelist *r_pagelist; 283 284 /* what caps shall we drop? */ 285 int r_inode_drop, r_inode_unless; 286 int r_dentry_drop, r_dentry_unless; 287 int r_old_dentry_drop, r_old_dentry_unless; 288 struct inode *r_old_inode; 289 int r_old_inode_drop, r_old_inode_unless; 290 291 struct ceph_msg *r_request; /* original request */ 292 struct ceph_msg *r_reply; 293 struct ceph_mds_reply_info_parsed r_reply_info; 294 int r_err; 295 296 297 struct page *r_locked_page; 298 int r_dir_caps; 299 int r_num_caps; 300 u32 r_readdir_offset; 301 302 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */ 303 unsigned long r_started; /* start time to measure timeout against */ 304 unsigned long r_start_latency; /* start time to measure latency */ 305 unsigned long r_end_latency; /* finish time to measure latency */ 306 unsigned long r_request_started; /* start time for mds request only, 307 used to measure lease durations */ 308 309 /* link unsafe requests to parent directory, for fsync */ 310 struct inode *r_unsafe_dir; 311 struct list_head r_unsafe_dir_item; 312 313 /* unsafe requests that modify the target inode */ 314 struct list_head r_unsafe_target_item; 315 316 struct ceph_mds_session *r_session; 317 318 int r_attempts; /* resend attempts */ 319 int r_num_fwd; /* number of forward attempts */ 320 int r_resend_mds; /* mds to resend to next, if any*/ 321 u32 r_sent_on_mseq; /* cap mseq request was sent at*/ 322 u64 r_deleg_ino; 323 324 struct list_head r_wait; 325 struct completion r_completion; 326 struct completion r_safe_completion; 327 ceph_mds_request_callback_t r_callback; 328 ceph_mds_request_wait_callback_t r_wait_for_completion; 329 struct list_head r_unsafe_item; /* per-session unsafe list item */ 330 331 long long r_dir_release_cnt; 332 long long r_dir_ordered_cnt; 333 int r_readdir_cache_idx; 334 335 struct ceph_cap_reservation r_caps_reservation; 336}; 337 338struct ceph_pool_perm { 339 struct rb_node node; 340 int perm; 341 s64 pool; 342 size_t pool_ns_len; 343 char pool_ns[]; 344}; 345 346struct ceph_snapid_map { 347 struct rb_node node; 348 struct list_head lru; 349 atomic_t ref; 350 u64 snap; 351 dev_t dev; 352 unsigned long last_used; 353}; 354 355/* 356 * node for list of quotarealm inodes that are not visible from the filesystem 357 * mountpoint, but required to handle, e.g. quotas. 358 */ 359struct ceph_quotarealm_inode { 360 struct rb_node node; 361 u64 ino; 362 unsigned long timeout; /* last time a lookup failed for this inode */ 363 struct mutex mutex; 364 struct inode *inode; 365}; 366 367struct cap_wait { 368 struct list_head list; 369 u64 ino; 370 pid_t tgid; 371 int need; 372 int want; 373}; 374 375enum { 376 CEPH_MDSC_STOPPING_BEGIN = 1, 377 CEPH_MDSC_STOPPING_FLUSHED = 2, 378}; 379 380/* 381 * mds client state 382 */ 383struct ceph_mds_client { 384 struct ceph_fs_client *fsc; 385 struct mutex mutex; /* all nested structures */ 386 387 struct ceph_mdsmap *mdsmap; 388 struct completion safe_umount_waiters; 389 wait_queue_head_t session_close_wq; 390 struct list_head waiting_for_map; 391 int mdsmap_err; 392 393 struct ceph_mds_session **sessions; /* NULL for mds if no session */ 394 atomic_t num_sessions; 395 int max_sessions; /* len of sessions array */ 396 int stopping; /* true if shutting down */ 397 398 atomic64_t quotarealms_count; /* # realms with quota */ 399 /* 400 * We keep a list of inodes we don't see in the mountpoint but that we 401 * need to track quota realms. 402 */ 403 struct rb_root quotarealms_inodes; 404 struct mutex quotarealms_inodes_mutex; 405 406 /* 407 * snap_rwsem will cover cap linkage into snaprealms, and 408 * realm snap contexts. (later, we can do per-realm snap 409 * contexts locks..) the empty list contains realms with no 410 * references (implying they contain no inodes with caps) that 411 * should be destroyed. 412 */ 413 u64 last_snap_seq; 414 struct rw_semaphore snap_rwsem; 415 struct rb_root snap_realms; 416 struct list_head snap_empty; 417 int num_snap_realms; 418 spinlock_t snap_empty_lock; /* protect snap_empty */ 419 420 u64 last_tid; /* most recent mds request */ 421 u64 oldest_tid; /* oldest incomplete mds request, 422 excluding setfilelock requests */ 423 struct rb_root request_tree; /* pending mds requests */ 424 struct delayed_work delayed_work; /* delayed work */ 425 unsigned long last_renew_caps; /* last time we renewed our caps */ 426 struct list_head cap_delay_list; /* caps with delayed release */ 427 spinlock_t cap_delay_lock; /* protects cap_delay_list */ 428 struct list_head snap_flush_list; /* cap_snaps ready to flush */ 429 spinlock_t snap_flush_lock; 430 431 u64 last_cap_flush_tid; 432 struct list_head cap_flush_list; 433 struct list_head cap_dirty_migrating; /* ...that are migration... */ 434 int num_cap_flushing; /* # caps we are flushing */ 435 spinlock_t cap_dirty_lock; /* protects above items */ 436 wait_queue_head_t cap_flushing_wq; 437 438 struct work_struct cap_reclaim_work; 439 atomic_t cap_reclaim_pending; 440 441 /* 442 * Cap reservations 443 * 444 * Maintain a global pool of preallocated struct ceph_caps, referenced 445 * by struct ceph_caps_reservations. This ensures that we preallocate 446 * memory needed to successfully process an MDS response. (If an MDS 447 * sends us cap information and we fail to process it, we will have 448 * problems due to the client and MDS being out of sync.) 449 * 450 * Reservations are 'owned' by a ceph_cap_reservation context. 451 */ 452 spinlock_t caps_list_lock; 453 struct list_head caps_list; /* unused (reserved or 454 unreserved) */ 455 struct list_head cap_wait_list; 456 int caps_total_count; /* total caps allocated */ 457 int caps_use_count; /* in use */ 458 int caps_use_max; /* max used caps */ 459 int caps_reserve_count; /* unused, reserved */ 460 int caps_avail_count; /* unused, unreserved */ 461 int caps_min_count; /* keep at least this many 462 (unreserved) */ 463 spinlock_t dentry_list_lock; 464 struct list_head dentry_leases; /* fifo list */ 465 struct list_head dentry_dir_leases; /* lru list */ 466 467 struct ceph_client_metric metric; 468 469 spinlock_t snapid_map_lock; 470 struct rb_root snapid_map_tree; 471 struct list_head snapid_map_lru; 472 473 struct rw_semaphore pool_perm_rwsem; 474 struct rb_root pool_perm_tree; 475 476 char nodename[__NEW_UTS_LEN + 1]; 477}; 478 479extern const char *ceph_mds_op_name(int op); 480 481extern bool check_session_state(struct ceph_mds_session *s); 482void inc_session_sequence(struct ceph_mds_session *s); 483 484extern struct ceph_mds_session * 485__ceph_lookup_mds_session(struct ceph_mds_client *, int mds); 486 487extern const char *ceph_session_state_name(int s); 488 489extern struct ceph_mds_session * 490ceph_get_mds_session(struct ceph_mds_session *s); 491extern void ceph_put_mds_session(struct ceph_mds_session *s); 492 493extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc, 494 struct ceph_msg *msg, int mds); 495 496extern int ceph_mdsc_init(struct ceph_fs_client *fsc); 497extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc); 498extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc); 499extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc); 500 501extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc); 502 503extern void ceph_invalidate_dir_request(struct ceph_mds_request *req); 504extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req, 505 struct inode *dir); 506extern struct ceph_mds_request * 507ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode); 508extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, 509 struct inode *dir, 510 struct ceph_mds_request *req); 511extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc, 512 struct inode *dir, 513 struct ceph_mds_request *req); 514extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req); 515extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req); 516static inline void ceph_mdsc_get_request(struct ceph_mds_request *req) 517{ 518 kref_get(&req->r_kref); 519} 520extern void ceph_mdsc_release_request(struct kref *kref); 521static inline void ceph_mdsc_put_request(struct ceph_mds_request *req) 522{ 523 kref_put(&req->r_kref, ceph_mdsc_release_request); 524} 525 526extern void send_flush_mdlog(struct ceph_mds_session *s); 527extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc, 528 void (*cb)(struct ceph_mds_session *), 529 bool check_state); 530extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq); 531extern void __ceph_queue_cap_release(struct ceph_mds_session *session, 532 struct ceph_cap *cap); 533extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc, 534 struct ceph_mds_session *session); 535extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc); 536extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr); 537extern int ceph_iterate_session_caps(struct ceph_mds_session *session, 538 int (*cb)(struct inode *, 539 struct ceph_cap *, void *), 540 void *arg); 541extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc); 542 543static inline void ceph_mdsc_free_path(char *path, int len) 544{ 545 if (!IS_ERR_OR_NULL(path)) 546 __putname(path - (PATH_MAX - 1 - len)); 547} 548 549extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base, 550 int stop_on_nosnap); 551 552extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry); 553extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session, 554 struct dentry *dentry, char action, 555 u32 seq); 556 557extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, 558 struct ceph_msg *msg); 559extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, 560 struct ceph_msg *msg); 561 562extern struct ceph_mds_session * 563ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target); 564extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc, 565 struct ceph_mds_session *session); 566 567extern int ceph_trim_caps(struct ceph_mds_client *mdsc, 568 struct ceph_mds_session *session, 569 int max_caps); 570 571static inline int ceph_wait_on_async_create(struct inode *inode) 572{ 573 struct ceph_inode_info *ci = ceph_inode(inode); 574 575 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT, 576 TASK_INTERRUPTIBLE); 577} 578 579extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session); 580extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino); 581#endif 582