fs/ceph/mds_client.h

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