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 CEPHFS_FEATURE_OP_GETVXATTR,
35 CEPHFS_FEATURE_32BITS_RETRY_FWD,
36
37 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_32BITS_RETRY_FWD,
38 };
39
40 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
41 0, 1, 2, 3, 4, 5, 6, 7, \
42 CEPHFS_FEATURE_MIMIC, \
43 CEPHFS_FEATURE_REPLY_ENCODING, \
44 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
45 CEPHFS_FEATURE_MULTI_RECONNECT, \
46 CEPHFS_FEATURE_DELEG_INO, \
47 CEPHFS_FEATURE_METRIC_COLLECT, \
48 CEPHFS_FEATURE_ALTERNATE_NAME, \
49 CEPHFS_FEATURE_NOTIFY_SESSION_STATE, \
50 CEPHFS_FEATURE_OP_GETVXATTR, \
51 CEPHFS_FEATURE_32BITS_RETRY_FWD, \
52 }
53
54 /*
55 * Some lock dependencies:
56 *
57 * session->s_mutex
58 * mdsc->mutex
59 *
60 * mdsc->snap_rwsem
61 *
62 * ci->i_ceph_lock
63 * mdsc->snap_flush_lock
64 * mdsc->cap_delay_lock
65 *
66 */
67
68 struct ceph_fs_client;
69 struct ceph_cap;
70
71 /*
72 * parsed info about a single inode. pointers are into the encoded
73 * on-wire structures within the mds reply message payload.
74 */
75 struct ceph_mds_reply_info_in {
76 struct ceph_mds_reply_inode *in;
77 struct ceph_dir_layout dir_layout;
78 u32 symlink_len;
79 char *symlink;
80 u32 xattr_len;
81 char *xattr_data;
82 u64 inline_version;
83 u32 inline_len;
84 char *inline_data;
85 u32 pool_ns_len;
86 char *pool_ns_data;
87 u64 max_bytes;
88 u64 max_files;
89 s32 dir_pin;
90 struct ceph_timespec btime;
91 struct ceph_timespec snap_btime;
92 u8 *fscrypt_auth;
93 u8 *fscrypt_file;
94 u32 fscrypt_auth_len;
95 u32 fscrypt_file_len;
96 u64 rsnaps;
97 u64 change_attr;
98 };
99
100 struct ceph_mds_reply_dir_entry {
101 bool is_nokey;
102 char *name;
103 u32 name_len;
104 u32 raw_hash;
105 struct ceph_mds_reply_lease *lease;
106 struct ceph_mds_reply_info_in inode;
107 loff_t offset;
108 };
109
110 struct ceph_mds_reply_xattr {
111 char *xattr_value;
112 size_t xattr_value_len;
113 };
114
115 /*
116 * parsed info about an mds reply, including information about
117 * either: 1) the target inode and/or its parent directory and dentry,
118 * and directory contents (for readdir results), or
119 * 2) the file range lock info (for fcntl F_GETLK results).
120 */
121 struct ceph_mds_reply_info_parsed {
122 struct ceph_mds_reply_head *head;
123
124 /* trace */
125 struct ceph_mds_reply_info_in diri, targeti;
126 struct ceph_mds_reply_dirfrag *dirfrag;
127 char *dname;
128 u8 *altname;
129 u32 dname_len;
130 u32 altname_len;
131 struct ceph_mds_reply_lease *dlease;
132 struct ceph_mds_reply_xattr xattr_info;
133
134 /* extra */
135 union {
136 /* for fcntl F_GETLK results */
137 struct ceph_filelock *filelock_reply;
138
139 /* for readdir results */
140 struct {
141 struct ceph_mds_reply_dirfrag *dir_dir;
142 size_t dir_buf_size;
143 int dir_nr;
144 bool dir_end;
145 bool dir_complete;
146 bool hash_order;
147 bool offset_hash;
148 struct ceph_mds_reply_dir_entry *dir_entries;
149 };
150
151 /* for create results */
152 struct {
153 bool has_create_ino;
154 u64 ino;
155 };
156 };
157
158 /* encoded blob describing snapshot contexts for certain
159 operations (e.g., open) */
160 void *snapblob;
161 int snapblob_len;
162 };
163
164
165 /*
166 * cap releases are batched and sent to the MDS en masse.
167 *
168 * Account for per-message overhead of mds_cap_release header
169 * and __le32 for osd epoch barrier trailing field.
170 */
171 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
172 sizeof(struct ceph_mds_cap_release)) / \
173 sizeof(struct ceph_mds_cap_item))
174
175
176 /*
177 * state associated with each MDS<->client session
178 */
179 enum {
180 CEPH_MDS_SESSION_NEW = 1,
181 CEPH_MDS_SESSION_OPENING = 2,
182 CEPH_MDS_SESSION_OPEN = 3,
183 CEPH_MDS_SESSION_HUNG = 4,
184 CEPH_MDS_SESSION_RESTARTING = 5,
185 CEPH_MDS_SESSION_RECONNECTING = 6,
186 CEPH_MDS_SESSION_CLOSING = 7,
187 CEPH_MDS_SESSION_CLOSED = 8,
188 CEPH_MDS_SESSION_REJECTED = 9,
189 };
190
191 struct ceph_mds_session {
192 struct ceph_mds_client *s_mdsc;
193 int s_mds;
194 int s_state;
195 unsigned long s_ttl; /* time until mds kills us */
196 unsigned long s_features;
197 u64 s_seq; /* incoming msg seq # */
198 struct mutex s_mutex; /* serialize session messages */
199
200 struct ceph_connection s_con;
201
202 struct ceph_auth_handshake s_auth;
203
204 atomic_t s_cap_gen; /* inc each time we get mds stale msg */
205 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */
206
207 /* protected by s_cap_lock */
208 spinlock_t s_cap_lock;
209 refcount_t s_ref;
210 struct list_head s_caps; /* all caps issued by this session */
211 struct ceph_cap *s_cap_iterator;
212 int s_nr_caps;
213 int s_num_cap_releases;
214 int s_cap_reconnect;
215 int s_readonly;
216 struct list_head s_cap_releases; /* waiting cap_release messages */
217 struct work_struct s_cap_release_work;
218
219 /* See ceph_inode_info->i_dirty_item. */
220 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
221
222 /* See ceph_inode_info->i_flushing_item. */
223 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
224
225 unsigned long s_renew_requested; /* last time we sent a renew req */
226 u64 s_renew_seq;
227
228 struct list_head s_waiting; /* waiting requests */
229 struct list_head s_unsafe; /* unsafe requests */
230 struct xarray s_delegated_inos;
231 };
232
233 /*
234 * modes of choosing which MDS to send a request to
235 */
236 enum {
237 USE_ANY_MDS,
238 USE_RANDOM_MDS,
239 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
240 };
241
242 struct ceph_mds_request;
243 struct ceph_mds_client;
244
245 /*
246 * request completion callback
247 */
248 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
249 struct ceph_mds_request *req);
250 /*
251 * wait for request completion callback
252 */
253 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
254 struct ceph_mds_request *req);
255
256 /*
257 * an in-flight mds request
258 */
259 struct ceph_mds_request {
260 u64 r_tid; /* transaction id */
261 struct rb_node r_node;
262 struct ceph_mds_client *r_mdsc;
263
264 struct kref r_kref;
265 int r_op; /* mds op code */
266
267 /* operation on what? */
268 struct inode *r_inode; /* arg1 */
269 struct dentry *r_dentry; /* arg1 */
270 struct dentry *r_old_dentry; /* arg2: rename from or link from */
271 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
272 char *r_path1, *r_path2;
273 struct ceph_vino r_ino1, r_ino2;
274
275 struct inode *r_parent; /* parent dir inode */
276 struct inode *r_target_inode; /* resulting inode */
277 struct inode *r_new_inode; /* new inode (for creates) */
278
279 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
280 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
281 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
282 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
283 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
284 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
285 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
286 #define CEPH_MDS_R_ASYNC (8) /* async request */
287 #define CEPH_MDS_R_FSCRYPT_FILE (9) /* must marshal fscrypt_file field */
288 unsigned long r_req_flags;
289
290 struct mutex r_fill_mutex;
291
292 union ceph_mds_request_args r_args;
293
294 struct ceph_fscrypt_auth *r_fscrypt_auth;
295 u64 r_fscrypt_file;
296
297 u8 *r_altname; /* fscrypt binary crypttext for long filenames */
298 u32 r_altname_len; /* length of r_altname */
299
300 int r_fmode; /* file mode, if expecting cap */
301 int r_request_release_offset;
302 const struct cred *r_cred;
303 struct timespec64 r_stamp;
304
305 /* for choosing which mds to send this request to */
306 int r_direct_mode;
307 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
308
309 /* data payload is used for xattr ops */
310 struct ceph_pagelist *r_pagelist;
311
312 /* what caps shall we drop? */
313 int r_inode_drop, r_inode_unless;
314 int r_dentry_drop, r_dentry_unless;
315 int r_old_dentry_drop, r_old_dentry_unless;
316 struct inode *r_old_inode;
317 int r_old_inode_drop, r_old_inode_unless;
318
319 struct ceph_msg *r_request; /* original request */
320 struct ceph_msg *r_reply;
321 struct ceph_mds_reply_info_parsed r_reply_info;
322 int r_err;
323 u32 r_readdir_offset;
324
325 struct page *r_locked_page;
326 int r_dir_caps;
327 int r_num_caps;
328
329 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
330 unsigned long r_started; /* start time to measure timeout against */
331 unsigned long r_start_latency; /* start time to measure latency */
332 unsigned long r_end_latency; /* finish time to measure latency */
333 unsigned long r_request_started; /* start time for mds request only,
334 used to measure lease durations */
335
336 /* link unsafe requests to parent directory, for fsync */
337 struct inode *r_unsafe_dir;
338 struct list_head r_unsafe_dir_item;
339
340 /* unsafe requests that modify the target inode */
341 struct list_head r_unsafe_target_item;
342
343 struct ceph_mds_session *r_session;
344
345 int r_attempts; /* resend attempts */
346 int r_num_fwd; /* number of forward attempts */
347 int r_resend_mds; /* mds to resend to next, if any*/
348 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
349 u64 r_deleg_ino;
350
351 struct list_head r_wait;
352 struct completion r_completion;
353 struct completion r_safe_completion;
354 ceph_mds_request_callback_t r_callback;
355 struct list_head r_unsafe_item; /* per-session unsafe list item */
356
357 long long r_dir_release_cnt;
358 long long r_dir_ordered_cnt;
359 int r_readdir_cache_idx;
360
361 int r_feature_needed;
362
363 struct ceph_cap_reservation r_caps_reservation;
364 };
365
366 struct ceph_pool_perm {
367 struct rb_node node;
368 int perm;
369 s64 pool;
370 size_t pool_ns_len;
371 char pool_ns[];
372 };
373
374 struct ceph_snapid_map {
375 struct rb_node node;
376 struct list_head lru;
377 atomic_t ref;
378 dev_t dev;
379 u64 snap;
380 unsigned long last_used;
381 };
382
383 /*
384 * node for list of quotarealm inodes that are not visible from the filesystem
385 * mountpoint, but required to handle, e.g. quotas.
386 */
387 struct ceph_quotarealm_inode {
388 struct rb_node node;
389 u64 ino;
390 unsigned long timeout; /* last time a lookup failed for this inode */
391 struct mutex mutex;
392 struct inode *inode;
393 };
394
395 struct cap_wait {
396 struct list_head list;
397 u64 ino;
398 pid_t tgid;
399 int need;
400 int want;
401 };
402
403 enum {
404 CEPH_MDSC_STOPPING_BEGIN = 1,
405 CEPH_MDSC_STOPPING_FLUSHING = 2,
406 CEPH_MDSC_STOPPING_FLUSHED = 3,
407 };
408
409 /*
410 * mds client state
411 */
412 struct ceph_mds_client {
413 struct ceph_fs_client *fsc;
414 struct mutex mutex; /* all nested structures */
415
416 struct ceph_mdsmap *mdsmap;
417 struct completion safe_umount_waiters;
418 wait_queue_head_t session_close_wq;
419 struct list_head waiting_for_map;
420 int mdsmap_err;
421
422 struct ceph_mds_session **sessions; /* NULL for mds if no session */
423 atomic_t num_sessions;
424 int max_sessions; /* len of sessions array */
425
426 spinlock_t stopping_lock; /* protect snap_empty */
427 int stopping; /* the stage of shutting down */
428 atomic_t stopping_blockers;
429 struct completion stopping_waiter;
430
431 atomic64_t quotarealms_count; /* # realms with quota */
432 /*
433 * We keep a list of inodes we don't see in the mountpoint but that we
434 * need to track quota realms.
435 */
436 struct rb_root quotarealms_inodes;
437 struct mutex quotarealms_inodes_mutex;
438
439 /*
440 * snap_rwsem will cover cap linkage into snaprealms, and
441 * realm snap contexts. (later, we can do per-realm snap
442 * contexts locks..) the empty list contains realms with no
443 * references (implying they contain no inodes with caps) that
444 * should be destroyed.
445 */
446 u64 last_snap_seq;
447 struct rw_semaphore snap_rwsem;
448 struct rb_root snap_realms;
449 struct list_head snap_empty;
450 int num_snap_realms;
451 spinlock_t snap_empty_lock; /* protect snap_empty */
452
453 u64 last_tid; /* most recent mds request */
454 u64 oldest_tid; /* oldest incomplete mds request,
455 excluding setfilelock requests */
456 struct rb_root request_tree; /* pending mds requests */
457 struct delayed_work delayed_work; /* delayed work */
458 unsigned long last_renew_caps; /* last time we renewed our caps */
459 struct list_head cap_delay_list; /* caps with delayed release */
460 spinlock_t cap_delay_lock; /* protects cap_delay_list */
461 struct list_head snap_flush_list; /* cap_snaps ready to flush */
462 spinlock_t snap_flush_lock;
463
464 u64 last_cap_flush_tid;
465 struct list_head cap_flush_list;
466 struct list_head cap_dirty_migrating; /* ...that are migration... */
467 int num_cap_flushing; /* # caps we are flushing */
468 spinlock_t cap_dirty_lock; /* protects above items */
469 wait_queue_head_t cap_flushing_wq;
470
471 struct work_struct cap_reclaim_work;
472 atomic_t cap_reclaim_pending;
473
474 /*
475 * Cap reservations
476 *
477 * Maintain a global pool of preallocated struct ceph_caps, referenced
478 * by struct ceph_caps_reservations. This ensures that we preallocate
479 * memory needed to successfully process an MDS response. (If an MDS
480 * sends us cap information and we fail to process it, we will have
481 * problems due to the client and MDS being out of sync.)
482 *
483 * Reservations are 'owned' by a ceph_cap_reservation context.
484 */
485 spinlock_t caps_list_lock;
486 struct list_head caps_list; /* unused (reserved or
487 unreserved) */
488 struct list_head cap_wait_list;
489 int caps_total_count; /* total caps allocated */
490 int caps_use_count; /* in use */
491 int caps_use_max; /* max used caps */
492 int caps_reserve_count; /* unused, reserved */
493 int caps_avail_count; /* unused, unreserved */
494 int caps_min_count; /* keep at least this many
495 (unreserved) */
496 spinlock_t dentry_list_lock;
497 struct list_head dentry_leases; /* fifo list */
498 struct list_head dentry_dir_leases; /* lru list */
499
500 struct ceph_client_metric metric;
501
502 spinlock_t snapid_map_lock;
503 struct rb_root snapid_map_tree;
504 struct list_head snapid_map_lru;
505
506 struct rw_semaphore pool_perm_rwsem;
507 struct rb_root pool_perm_tree;
508
509 char nodename[__NEW_UTS_LEN + 1];
510 };
511
512 extern const char *ceph_mds_op_name(int op);
513
514 extern bool check_session_state(struct ceph_mds_session *s);
515 void inc_session_sequence(struct ceph_mds_session *s);
516
517 extern struct ceph_mds_session *
518 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
519
520 extern const char *ceph_session_state_name(int s);
521
522 extern struct ceph_mds_session *
523 ceph_get_mds_session(struct ceph_mds_session *s);
524 extern void ceph_put_mds_session(struct ceph_mds_session *s);
525
526 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
527 struct ceph_msg *msg, int mds);
528
529 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
530 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
531 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
532 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
533
534 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
535
536 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
537 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
538 struct inode *dir);
539 extern struct ceph_mds_request *
540 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
541 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
542 struct inode *dir,
543 struct ceph_mds_request *req);
544 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
545 struct ceph_mds_request *req,
546 ceph_mds_request_wait_callback_t wait_func);
547 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
548 struct inode *dir,
549 struct ceph_mds_request *req);
550 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
551 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
ceph_mdsc_get_request(struct ceph_mds_request * req)552 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
553 {
554 kref_get(&req->r_kref);
555 }
556 extern void ceph_mdsc_release_request(struct kref *kref);
ceph_mdsc_put_request(struct ceph_mds_request * req)557 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
558 {
559 kref_put(&req->r_kref, ceph_mdsc_release_request);
560 }
561
562 extern void send_flush_mdlog(struct ceph_mds_session *s);
563 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
564 void (*cb)(struct ceph_mds_session *),
565 bool check_state);
566 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
567 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
568 struct ceph_cap *cap);
569 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
570 struct ceph_mds_session *session);
571 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
572 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
573 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
574 int (*cb)(struct inode *, int mds, void *),
575 void *arg);
576 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
577
ceph_mdsc_free_path(char * path,int len)578 static inline void ceph_mdsc_free_path(char *path, int len)
579 {
580 if (!IS_ERR_OR_NULL(path))
581 __putname(path - (PATH_MAX - 1 - len));
582 }
583
584 extern char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc,
585 struct dentry *dentry, int *plen, u64 *base,
586 int for_wire);
587
588 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
589 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
590 struct dentry *dentry, char action,
591 u32 seq);
592
593 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
594 struct ceph_msg *msg);
595 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
596 struct ceph_msg *msg);
597
598 extern struct ceph_mds_session *
599 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
600 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
601 struct ceph_mds_session *session);
602
603 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
604 struct ceph_mds_session *session,
605 int max_caps);
606
ceph_wait_on_async_create(struct inode * inode)607 static inline int ceph_wait_on_async_create(struct inode *inode)
608 {
609 struct ceph_inode_info *ci = ceph_inode(inode);
610
611 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
612 TASK_KILLABLE);
613 }
614
615 extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
616 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
617 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
618 #endif
619