658 lines · c
1/* SPDX-License-Identifier: GPL-2.0 */2#ifndef _FS_CEPH_MDS_CLIENT_H3#define _FS_CEPH_MDS_CLIENT_H4 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/auth.h>18 19#include "mdsmap.h"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 CEPHFS_FEATURE_ALTERNATE_NAME,33 CEPHFS_FEATURE_NOTIFY_SESSION_STATE,34 CEPHFS_FEATURE_OP_GETVXATTR,35 CEPHFS_FEATURE_32BITS_RETRY_FWD,36 CEPHFS_FEATURE_NEW_SNAPREALM_INFO,37 CEPHFS_FEATURE_HAS_OWNER_UIDGID,38 CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK,39 40 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK,41};42 43#define CEPHFS_FEATURES_CLIENT_SUPPORTED { \44 0, 1, 2, 3, 4, 5, 6, 7, \45 CEPHFS_FEATURE_MIMIC, \46 CEPHFS_FEATURE_REPLY_ENCODING, \47 CEPHFS_FEATURE_LAZY_CAP_WANTED, \48 CEPHFS_FEATURE_MULTI_RECONNECT, \49 CEPHFS_FEATURE_DELEG_INO, \50 CEPHFS_FEATURE_METRIC_COLLECT, \51 CEPHFS_FEATURE_ALTERNATE_NAME, \52 CEPHFS_FEATURE_NOTIFY_SESSION_STATE, \53 CEPHFS_FEATURE_OP_GETVXATTR, \54 CEPHFS_FEATURE_32BITS_RETRY_FWD, \55 CEPHFS_FEATURE_HAS_OWNER_UIDGID, \56 CEPHFS_FEATURE_MDS_AUTH_CAPS_CHECK, \57}58 59/*60 * Some lock dependencies:61 *62 * session->s_mutex63 * mdsc->mutex64 *65 * mdsc->snap_rwsem66 *67 * ci->i_ceph_lock68 * mdsc->snap_flush_lock69 * mdsc->cap_delay_lock70 *71 */72 73struct ceph_fs_client;74struct ceph_cap;75 76#define MDS_AUTH_UID_ANY -177 78struct ceph_mds_cap_match {79 s64 uid; /* default to MDS_AUTH_UID_ANY */80 u32 num_gids;81 u32 *gids; /* use these GIDs */82 char *path; /* require path to be child of this83 (may be "" or "/" for any) */84 char *fs_name;85 bool root_squash; /* default to false */86};87 88struct ceph_mds_cap_auth {89 struct ceph_mds_cap_match match;90 bool readable;91 bool writeable;92};93 94/*95 * parsed info about a single inode. pointers are into the encoded96 * on-wire structures within the mds reply message payload.97 */98struct ceph_mds_reply_info_in {99 struct ceph_mds_reply_inode *in;100 struct ceph_dir_layout dir_layout;101 u32 symlink_len;102 char *symlink;103 u32 xattr_len;104 char *xattr_data;105 u64 inline_version;106 u32 inline_len;107 char *inline_data;108 u32 pool_ns_len;109 char *pool_ns_data;110 u64 max_bytes;111 u64 max_files;112 s32 dir_pin;113 struct ceph_timespec btime;114 struct ceph_timespec snap_btime;115 u8 *fscrypt_auth;116 u8 *fscrypt_file;117 u32 fscrypt_auth_len;118 u32 fscrypt_file_len;119 u64 rsnaps;120 u64 change_attr;121};122 123struct ceph_mds_reply_dir_entry {124 bool is_nokey;125 char *name;126 u32 name_len;127 u32 raw_hash;128 struct ceph_mds_reply_lease *lease;129 struct ceph_mds_reply_info_in inode;130 loff_t offset;131};132 133struct ceph_mds_reply_xattr {134 char *xattr_value;135 size_t xattr_value_len;136};137 138/*139 * parsed info about an mds reply, including information about140 * either: 1) the target inode and/or its parent directory and dentry,141 * and directory contents (for readdir results), or142 * 2) the file range lock info (for fcntl F_GETLK results).143 */144struct ceph_mds_reply_info_parsed {145 struct ceph_mds_reply_head *head;146 147 /* trace */148 struct ceph_mds_reply_info_in diri, targeti;149 struct ceph_mds_reply_dirfrag *dirfrag;150 char *dname;151 u8 *altname;152 u32 dname_len;153 u32 altname_len;154 struct ceph_mds_reply_lease *dlease;155 struct ceph_mds_reply_xattr xattr_info;156 157 /* extra */158 union {159 /* for fcntl F_GETLK results */160 struct ceph_filelock *filelock_reply;161 162 /* for readdir results */163 struct {164 struct ceph_mds_reply_dirfrag *dir_dir;165 size_t dir_buf_size;166 int dir_nr;167 bool dir_end;168 bool dir_complete;169 bool hash_order;170 bool offset_hash;171 struct ceph_mds_reply_dir_entry *dir_entries;172 };173 174 /* for create results */175 struct {176 bool has_create_ino;177 u64 ino;178 };179 };180 181 /* encoded blob describing snapshot contexts for certain182 operations (e.g., open) */183 void *snapblob;184 int snapblob_len;185};186 187 188/*189 * cap releases are batched and sent to the MDS en masse.190 *191 * Account for per-message overhead of mds_cap_release header192 * and __le32 for osd epoch barrier trailing field.193 */194#define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \195 sizeof(struct ceph_mds_cap_release)) / \196 sizeof(struct ceph_mds_cap_item))197 198 199/*200 * state associated with each MDS<->client session201 */202enum {203 CEPH_MDS_SESSION_NEW = 1,204 CEPH_MDS_SESSION_OPENING = 2,205 CEPH_MDS_SESSION_OPEN = 3,206 CEPH_MDS_SESSION_HUNG = 4,207 CEPH_MDS_SESSION_RESTARTING = 5,208 CEPH_MDS_SESSION_RECONNECTING = 6,209 CEPH_MDS_SESSION_CLOSING = 7,210 CEPH_MDS_SESSION_CLOSED = 8,211 CEPH_MDS_SESSION_REJECTED = 9,212};213 214struct ceph_mds_session {215 struct ceph_mds_client *s_mdsc;216 int s_mds;217 int s_state;218 unsigned long s_ttl; /* time until mds kills us */219 unsigned long s_features;220 u64 s_seq; /* incoming msg seq # */221 struct mutex s_mutex; /* serialize session messages */222 223 struct ceph_connection s_con;224 225 struct ceph_auth_handshake s_auth;226 227 atomic_t s_cap_gen; /* inc each time we get mds stale msg */228 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */229 230 /* protected by s_cap_lock */231 spinlock_t s_cap_lock;232 refcount_t s_ref;233 struct list_head s_caps; /* all caps issued by this session */234 struct ceph_cap *s_cap_iterator;235 int s_nr_caps;236 int s_num_cap_releases;237 int s_cap_reconnect;238 int s_readonly;239 struct list_head s_cap_releases; /* waiting cap_release messages */240 struct work_struct s_cap_release_work;241 242 /* See ceph_inode_info->i_dirty_item. */243 struct list_head s_cap_dirty; /* inodes w/ dirty caps */244 245 /* See ceph_inode_info->i_flushing_item. */246 struct list_head s_cap_flushing; /* inodes w/ flushing caps */247 248 unsigned long s_renew_requested; /* last time we sent a renew req */249 u64 s_renew_seq;250 251 struct list_head s_waiting; /* waiting requests */252 struct list_head s_unsafe; /* unsafe requests */253 struct xarray s_delegated_inos;254};255 256/*257 * modes of choosing which MDS to send a request to258 */259enum {260 USE_ANY_MDS,261 USE_RANDOM_MDS,262 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */263};264 265struct ceph_mds_request;266struct ceph_mds_client;267 268/*269 * request completion callback270 */271typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,272 struct ceph_mds_request *req);273/*274 * wait for request completion callback275 */276typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,277 struct ceph_mds_request *req);278 279/*280 * an in-flight mds request281 */282struct ceph_mds_request {283 u64 r_tid; /* transaction id */284 struct rb_node r_node;285 struct ceph_mds_client *r_mdsc;286 287 struct kref r_kref;288 int r_op; /* mds op code */289 290 /* operation on what? */291 struct inode *r_inode; /* arg1 */292 struct dentry *r_dentry; /* arg1 */293 struct dentry *r_old_dentry; /* arg2: rename from or link from */294 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */295 char *r_path1, *r_path2;296 struct ceph_vino r_ino1, r_ino2;297 298 struct inode *r_parent; /* parent dir inode */299 struct inode *r_target_inode; /* resulting inode */300 struct inode *r_new_inode; /* new inode (for creates) */301 302#define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */303#define CEPH_MDS_R_ABORTED (2) /* call was aborted */304#define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */305#define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */306#define CEPH_MDS_R_GOT_RESULT (5) /* got a result */307#define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */308#define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */309#define CEPH_MDS_R_ASYNC (8) /* async request */310#define CEPH_MDS_R_FSCRYPT_FILE (9) /* must marshal fscrypt_file field */311 unsigned long r_req_flags;312 313 struct mutex r_fill_mutex;314 315 union ceph_mds_request_args r_args;316 317 struct ceph_fscrypt_auth *r_fscrypt_auth;318 u64 r_fscrypt_file;319 320 u8 *r_altname; /* fscrypt binary crypttext for long filenames */321 u32 r_altname_len; /* length of r_altname */322 323 int r_fmode; /* file mode, if expecting cap */324 int r_request_release_offset;325 const struct cred *r_cred;326 struct mnt_idmap *r_mnt_idmap;327 struct timespec64 r_stamp;328 329 /* for choosing which mds to send this request to */330 int r_direct_mode;331 u32 r_direct_hash; /* choose dir frag based on this dentry hash */332 333 /* data payload is used for xattr ops */334 struct ceph_pagelist *r_pagelist;335 336 /* what caps shall we drop? */337 int r_inode_drop, r_inode_unless;338 int r_dentry_drop, r_dentry_unless;339 int r_old_dentry_drop, r_old_dentry_unless;340 struct inode *r_old_inode;341 int r_old_inode_drop, r_old_inode_unless;342 343 struct ceph_msg *r_request; /* original request */344 struct ceph_msg *r_reply;345 struct ceph_mds_reply_info_parsed r_reply_info;346 int r_err;347 u32 r_readdir_offset;348 349 struct page *r_locked_page;350 int r_dir_caps;351 int r_num_caps;352 353 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */354 unsigned long r_started; /* start time to measure timeout against */355 unsigned long r_start_latency; /* start time to measure latency */356 unsigned long r_end_latency; /* finish time to measure latency */357 unsigned long r_request_started; /* start time for mds request only,358 used to measure lease durations */359 360 /* link unsafe requests to parent directory, for fsync */361 struct inode *r_unsafe_dir;362 struct list_head r_unsafe_dir_item;363 364 /* unsafe requests that modify the target inode */365 struct list_head r_unsafe_target_item;366 367 struct ceph_mds_session *r_session;368 369 int r_attempts; /* resend attempts */370 int r_num_fwd; /* number of forward attempts */371 int r_resend_mds; /* mds to resend to next, if any*/372 u32 r_sent_on_mseq; /* cap mseq request was sent at*/373 u64 r_deleg_ino;374 375 struct list_head r_wait;376 struct completion r_completion;377 struct completion r_safe_completion;378 ceph_mds_request_callback_t r_callback;379 struct list_head r_unsafe_item; /* per-session unsafe list item */380 381 long long r_dir_release_cnt;382 long long r_dir_ordered_cnt;383 int r_readdir_cache_idx;384 385 int r_feature_needed;386 387 struct ceph_cap_reservation r_caps_reservation;388};389 390struct ceph_pool_perm {391 struct rb_node node;392 int perm;393 s64 pool;394 size_t pool_ns_len;395 char pool_ns[];396};397 398struct ceph_snapid_map {399 struct rb_node node;400 struct list_head lru;401 atomic_t ref;402 dev_t dev;403 u64 snap;404 unsigned long last_used;405};406 407/*408 * node for list of quotarealm inodes that are not visible from the filesystem409 * mountpoint, but required to handle, e.g. quotas.410 */411struct ceph_quotarealm_inode {412 struct rb_node node;413 u64 ino;414 unsigned long timeout; /* last time a lookup failed for this inode */415 struct mutex mutex;416 struct inode *inode;417};418 419#ifdef CONFIG_DEBUG_FS420 421struct cap_wait {422 struct list_head list;423 u64 ino;424 pid_t tgid;425 int need;426 int want;427};428 429#endif430 431enum {432 CEPH_MDSC_STOPPING_BEGIN = 1,433 CEPH_MDSC_STOPPING_FLUSHING = 2,434 CEPH_MDSC_STOPPING_FLUSHED = 3,435};436 437/*438 * mds client state439 */440struct ceph_mds_client {441 struct ceph_fs_client *fsc;442 struct mutex mutex; /* all nested structures */443 444 struct ceph_mdsmap *mdsmap;445 struct completion safe_umount_waiters;446 wait_queue_head_t session_close_wq;447 struct list_head waiting_for_map;448 int mdsmap_err;449 450 struct ceph_mds_session **sessions; /* NULL for mds if no session */451 atomic_t num_sessions;452 int max_sessions; /* len of sessions array */453 454 spinlock_t stopping_lock; /* protect snap_empty */455 int stopping; /* the stage of shutting down */456 atomic_t stopping_blockers;457 struct completion stopping_waiter;458 459 atomic64_t quotarealms_count; /* # realms with quota */460 /*461 * We keep a list of inodes we don't see in the mountpoint but that we462 * need to track quota realms.463 */464 struct rb_root quotarealms_inodes;465 struct mutex quotarealms_inodes_mutex;466 467 /*468 * snap_rwsem will cover cap linkage into snaprealms, and469 * realm snap contexts. (later, we can do per-realm snap470 * contexts locks..) the empty list contains realms with no471 * references (implying they contain no inodes with caps) that472 * should be destroyed.473 */474 u64 last_snap_seq;475 struct rw_semaphore snap_rwsem;476 struct rb_root snap_realms;477 struct list_head snap_empty;478 int num_snap_realms;479 spinlock_t snap_empty_lock; /* protect snap_empty */480 481 u64 last_tid; /* most recent mds request */482 u64 oldest_tid; /* oldest incomplete mds request,483 excluding setfilelock requests */484 struct rb_root request_tree; /* pending mds requests */485 struct delayed_work delayed_work; /* delayed work */486 unsigned long last_renew_caps; /* last time we renewed our caps */487 struct list_head cap_delay_list; /* caps with delayed release */488 struct list_head cap_unlink_delay_list; /* caps with delayed release for unlink */489 spinlock_t cap_delay_lock; /* protects cap_delay_list and cap_unlink_delay_list */490 struct list_head snap_flush_list; /* cap_snaps ready to flush */491 spinlock_t snap_flush_lock;492 493 u64 last_cap_flush_tid;494 struct list_head cap_flush_list;495 struct list_head cap_dirty_migrating; /* ...that are migration... */496 int num_cap_flushing; /* # caps we are flushing */497 spinlock_t cap_dirty_lock; /* protects above items */498 wait_queue_head_t cap_flushing_wq;499 500 struct work_struct cap_reclaim_work;501 atomic_t cap_reclaim_pending;502 503 struct work_struct cap_unlink_work;504 505 /*506 * Cap reservations507 *508 * Maintain a global pool of preallocated struct ceph_caps, referenced509 * by struct ceph_caps_reservations. This ensures that we preallocate510 * memory needed to successfully process an MDS response. (If an MDS511 * sends us cap information and we fail to process it, we will have512 * problems due to the client and MDS being out of sync.)513 *514 * Reservations are 'owned' by a ceph_cap_reservation context.515 */516 spinlock_t caps_list_lock;517 struct list_head caps_list; /* unused (reserved or518 unreserved) */519#ifdef CONFIG_DEBUG_FS520 struct list_head cap_wait_list;521#endif522 int caps_total_count; /* total caps allocated */523 int caps_use_count; /* in use */524 int caps_use_max; /* max used caps */525 int caps_reserve_count; /* unused, reserved */526 int caps_avail_count; /* unused, unreserved */527 int caps_min_count; /* keep at least this many528 (unreserved) */529 spinlock_t dentry_list_lock;530 struct list_head dentry_leases; /* fifo list */531 struct list_head dentry_dir_leases; /* lru list */532 533 struct ceph_client_metric metric;534 535 spinlock_t snapid_map_lock;536 struct rb_root snapid_map_tree;537 struct list_head snapid_map_lru;538 539 struct rw_semaphore pool_perm_rwsem;540 struct rb_root pool_perm_tree;541 542 u32 s_cap_auths_num;543 struct ceph_mds_cap_auth *s_cap_auths;544 545 char nodename[__NEW_UTS_LEN + 1];546};547 548extern const char *ceph_mds_op_name(int op);549 550extern bool check_session_state(struct ceph_mds_session *s);551void inc_session_sequence(struct ceph_mds_session *s);552 553extern struct ceph_mds_session *554__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);555 556extern const char *ceph_session_state_name(int s);557 558extern struct ceph_mds_session *559ceph_get_mds_session(struct ceph_mds_session *s);560extern void ceph_put_mds_session(struct ceph_mds_session *s);561 562extern int ceph_mdsc_init(struct ceph_fs_client *fsc);563extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);564extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);565extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);566 567extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);568 569extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);570extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,571 struct inode *dir);572extern struct ceph_mds_request *573ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);574extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,575 struct inode *dir,576 struct ceph_mds_request *req);577int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,578 struct ceph_mds_request *req,579 ceph_mds_request_wait_callback_t wait_func);580extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,581 struct inode *dir,582 struct ceph_mds_request *req);583extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);584extern void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req);585static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)586{587 kref_get(&req->r_kref);588}589extern void ceph_mdsc_release_request(struct kref *kref);590static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)591{592 kref_put(&req->r_kref, ceph_mdsc_release_request);593}594 595extern void send_flush_mdlog(struct ceph_mds_session *s);596extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,597 void (*cb)(struct ceph_mds_session *),598 bool check_state);599extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);600extern void __ceph_queue_cap_release(struct ceph_mds_session *session,601 struct ceph_cap *cap);602extern void ceph_flush_session_cap_releases(struct ceph_mds_client *mdsc,603 struct ceph_mds_session *session);604extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);605extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);606extern void ceph_queue_cap_unlink_work(struct ceph_mds_client *mdsc);607extern int ceph_iterate_session_caps(struct ceph_mds_session *session,608 int (*cb)(struct inode *, int mds, void *),609 void *arg);610extern int ceph_mds_check_access(struct ceph_mds_client *mdsc, char *tpath,611 int mask);612 613extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);614 615static inline void ceph_mdsc_free_path(char *path, int len)616{617 if (!IS_ERR_OR_NULL(path))618 __putname(path - (PATH_MAX - 1 - len));619}620 621extern char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc,622 struct dentry *dentry, int *plen, u64 *base,623 int for_wire);624 625extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);626extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,627 struct dentry *dentry, char action,628 u32 seq);629 630extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,631 struct ceph_msg *msg);632extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,633 struct ceph_msg *msg);634 635extern struct ceph_mds_session *636ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);637extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,638 struct ceph_mds_session *session);639 640extern int ceph_trim_caps(struct ceph_mds_client *mdsc,641 struct ceph_mds_session *session,642 int max_caps);643 644static inline int ceph_wait_on_async_create(struct inode *inode)645{646 struct ceph_inode_info *ci = ceph_inode(inode);647 648 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,649 TASK_KILLABLE);650}651 652extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);653extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);654extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);655 656extern bool enable_unsafe_idmap;657#endif658