Loading...
1#ifndef _FS_CEPH_SUPER_H
2#define _FS_CEPH_SUPER_H
3
4#include <linux/ceph/ceph_debug.h>
5
6#include <asm/unaligned.h>
7#include <linux/backing-dev.h>
8#include <linux/completion.h>
9#include <linux/exportfs.h>
10#include <linux/fs.h>
11#include <linux/mempool.h>
12#include <linux/pagemap.h>
13#include <linux/wait.h>
14#include <linux/writeback.h>
15#include <linux/slab.h>
16
17#include <linux/ceph/libceph.h>
18
19/* f_type in struct statfs */
20#define CEPH_SUPER_MAGIC 0x00c36400
21
22/* large granularity for statfs utilization stats to facilitate
23 * large volume sizes on 32-bit machines. */
24#define CEPH_BLOCK_SHIFT 20 /* 1 MB */
25#define CEPH_BLOCK (1 << CEPH_BLOCK_SHIFT)
26
27#define CEPH_MOUNT_OPT_DIRSTAT (1<<4) /* `cat dirname` for stats */
28#define CEPH_MOUNT_OPT_RBYTES (1<<5) /* dir st_bytes = rbytes */
29#define CEPH_MOUNT_OPT_NOASYNCREADDIR (1<<7) /* no dcache readdir */
30#define CEPH_MOUNT_OPT_INO32 (1<<8) /* 32 bit inos */
31#define CEPH_MOUNT_OPT_DCACHE (1<<9) /* use dcache for readdir etc */
32
33#define CEPH_MOUNT_OPT_DEFAULT (CEPH_MOUNT_OPT_RBYTES)
34
35#define ceph_set_mount_opt(fsc, opt) \
36 (fsc)->mount_options->flags |= CEPH_MOUNT_OPT_##opt;
37#define ceph_test_mount_opt(fsc, opt) \
38 (!!((fsc)->mount_options->flags & CEPH_MOUNT_OPT_##opt))
39
40#define CEPH_RSIZE_DEFAULT 0 /* max read size */
41#define CEPH_RASIZE_DEFAULT (8192*1024) /* readahead */
42#define CEPH_MAX_READDIR_DEFAULT 1024
43#define CEPH_MAX_READDIR_BYTES_DEFAULT (512*1024)
44#define CEPH_SNAPDIRNAME_DEFAULT ".snap"
45
46struct ceph_mount_options {
47 int flags;
48 int sb_flags;
49
50 int wsize; /* max write size */
51 int rsize; /* max read size */
52 int rasize; /* max readahead */
53 int congestion_kb; /* max writeback in flight */
54 int caps_wanted_delay_min, caps_wanted_delay_max;
55 int cap_release_safety;
56 int max_readdir; /* max readdir result (entires) */
57 int max_readdir_bytes; /* max readdir result (bytes) */
58
59 /*
60 * everything above this point can be memcmp'd; everything below
61 * is handled in compare_mount_options()
62 */
63
64 char *snapdir_name; /* default ".snap" */
65};
66
67struct ceph_fs_client {
68 struct super_block *sb;
69
70 struct ceph_mount_options *mount_options;
71 struct ceph_client *client;
72
73 unsigned long mount_state;
74 int min_caps; /* min caps i added */
75
76 struct ceph_mds_client *mdsc;
77
78 /* writeback */
79 mempool_t *wb_pagevec_pool;
80 struct workqueue_struct *wb_wq;
81 struct workqueue_struct *pg_inv_wq;
82 struct workqueue_struct *trunc_wq;
83 atomic_long_t writeback_count;
84
85 struct backing_dev_info backing_dev_info;
86
87#ifdef CONFIG_DEBUG_FS
88 struct dentry *debugfs_dentry_lru, *debugfs_caps;
89 struct dentry *debugfs_congestion_kb;
90 struct dentry *debugfs_bdi;
91 struct dentry *debugfs_mdsc, *debugfs_mdsmap;
92#endif
93};
94
95
96/*
97 * File i/o capability. This tracks shared state with the metadata
98 * server that allows us to cache or writeback attributes or to read
99 * and write data. For any given inode, we should have one or more
100 * capabilities, one issued by each metadata server, and our
101 * cumulative access is the OR of all issued capabilities.
102 *
103 * Each cap is referenced by the inode's i_caps rbtree and by per-mds
104 * session capability lists.
105 */
106struct ceph_cap {
107 struct ceph_inode_info *ci;
108 struct rb_node ci_node; /* per-ci cap tree */
109 struct ceph_mds_session *session;
110 struct list_head session_caps; /* per-session caplist */
111 int mds;
112 u64 cap_id; /* unique cap id (mds provided) */
113 int issued; /* latest, from the mds */
114 int implemented; /* implemented superset of issued (for revocation) */
115 int mds_wanted;
116 u32 seq, issue_seq, mseq;
117 u32 cap_gen; /* active/stale cycle */
118 unsigned long last_used;
119 struct list_head caps_item;
120};
121
122#define CHECK_CAPS_NODELAY 1 /* do not delay any further */
123#define CHECK_CAPS_AUTHONLY 2 /* only check auth cap */
124#define CHECK_CAPS_FLUSH 4 /* flush any dirty caps */
125
126/*
127 * Snapped cap state that is pending flush to mds. When a snapshot occurs,
128 * we first complete any in-process sync writes and writeback any dirty
129 * data before flushing the snapped state (tracked here) back to the MDS.
130 */
131struct ceph_cap_snap {
132 atomic_t nref;
133 struct ceph_inode_info *ci;
134 struct list_head ci_item, flushing_item;
135
136 u64 follows, flush_tid;
137 int issued, dirty;
138 struct ceph_snap_context *context;
139
140 umode_t mode;
141 uid_t uid;
142 gid_t gid;
143
144 struct ceph_buffer *xattr_blob;
145 u64 xattr_version;
146
147 u64 size;
148 struct timespec mtime, atime, ctime;
149 u64 time_warp_seq;
150 int writing; /* a sync write is still in progress */
151 int dirty_pages; /* dirty pages awaiting writeback */
152};
153
154static inline void ceph_put_cap_snap(struct ceph_cap_snap *capsnap)
155{
156 if (atomic_dec_and_test(&capsnap->nref)) {
157 if (capsnap->xattr_blob)
158 ceph_buffer_put(capsnap->xattr_blob);
159 kfree(capsnap);
160 }
161}
162
163/*
164 * The frag tree describes how a directory is fragmented, potentially across
165 * multiple metadata servers. It is also used to indicate points where
166 * metadata authority is delegated, and whether/where metadata is replicated.
167 *
168 * A _leaf_ frag will be present in the i_fragtree IFF there is
169 * delegation info. That is, if mds >= 0 || ndist > 0.
170 */
171#define CEPH_MAX_DIRFRAG_REP 4
172
173struct ceph_inode_frag {
174 struct rb_node node;
175
176 /* fragtree state */
177 u32 frag;
178 int split_by; /* i.e. 2^(split_by) children */
179
180 /* delegation and replication info */
181 int mds; /* -1 if same authority as parent */
182 int ndist; /* >0 if replicated */
183 int dist[CEPH_MAX_DIRFRAG_REP];
184};
185
186/*
187 * We cache inode xattrs as an encoded blob until they are first used,
188 * at which point we parse them into an rbtree.
189 */
190struct ceph_inode_xattr {
191 struct rb_node node;
192
193 const char *name;
194 int name_len;
195 const char *val;
196 int val_len;
197 int dirty;
198
199 int should_free_name;
200 int should_free_val;
201};
202
203/*
204 * Ceph dentry state
205 */
206struct ceph_dentry_info {
207 unsigned long flags;
208 struct ceph_mds_session *lease_session;
209 u32 lease_gen, lease_shared_gen;
210 u32 lease_seq;
211 unsigned long lease_renew_after, lease_renew_from;
212 struct list_head lru;
213 struct dentry *dentry;
214 u64 time;
215 u64 offset;
216};
217
218/*
219 * dentry flags
220 *
221 * The locking for D_COMPLETE is a bit odd:
222 * - we can clear it at almost any time (see ceph_d_prune)
223 * - it is only meaningful if:
224 * - we hold dir inode i_ceph_lock
225 * - we hold dir FILE_SHARED caps
226 * - the dentry D_COMPLETE is set
227 */
228#define CEPH_D_COMPLETE 1 /* if set, d_u.d_subdirs is complete directory */
229
230struct ceph_inode_xattrs_info {
231 /*
232 * (still encoded) xattr blob. we avoid the overhead of parsing
233 * this until someone actually calls getxattr, etc.
234 *
235 * blob->vec.iov_len == 4 implies there are no xattrs; blob ==
236 * NULL means we don't know.
237 */
238 struct ceph_buffer *blob, *prealloc_blob;
239
240 struct rb_root index;
241 bool dirty;
242 int count;
243 int names_size;
244 int vals_size;
245 u64 version, index_version;
246};
247
248/*
249 * Ceph inode.
250 */
251struct ceph_inode_info {
252 struct ceph_vino i_vino; /* ceph ino + snap */
253
254 spinlock_t i_ceph_lock;
255
256 u64 i_version;
257 u32 i_time_warp_seq;
258
259 unsigned i_ceph_flags;
260 unsigned long i_release_count;
261
262 struct ceph_dir_layout i_dir_layout;
263 struct ceph_file_layout i_layout;
264 char *i_symlink;
265
266 /* for dirs */
267 struct timespec i_rctime;
268 u64 i_rbytes, i_rfiles, i_rsubdirs;
269 u64 i_files, i_subdirs;
270 u64 i_max_offset; /* largest readdir offset, set with D_COMPLETE */
271
272 struct rb_root i_fragtree;
273 struct mutex i_fragtree_mutex;
274
275 struct ceph_inode_xattrs_info i_xattrs;
276
277 /* capabilities. protected _both_ by i_ceph_lock and cap->session's
278 * s_mutex. */
279 struct rb_root i_caps; /* cap list */
280 struct ceph_cap *i_auth_cap; /* authoritative cap, if any */
281 unsigned i_dirty_caps, i_flushing_caps; /* mask of dirtied fields */
282 struct list_head i_dirty_item, i_flushing_item;
283 u64 i_cap_flush_seq;
284 /* we need to track cap writeback on a per-cap-bit basis, to allow
285 * overlapping, pipelined cap flushes to the mds. we can probably
286 * reduce the tid to 8 bits if we're concerned about inode size. */
287 u16 i_cap_flush_last_tid, i_cap_flush_tid[CEPH_CAP_BITS];
288 wait_queue_head_t i_cap_wq; /* threads waiting on a capability */
289 unsigned long i_hold_caps_min; /* jiffies */
290 unsigned long i_hold_caps_max; /* jiffies */
291 struct list_head i_cap_delay_list; /* for delayed cap release to mds */
292 int i_cap_exporting_mds; /* to handle cap migration between */
293 unsigned i_cap_exporting_mseq; /* mds's. */
294 unsigned i_cap_exporting_issued;
295 struct ceph_cap_reservation i_cap_migration_resv;
296 struct list_head i_cap_snaps; /* snapped state pending flush to mds */
297 struct ceph_snap_context *i_head_snapc; /* set if wr_buffer_head > 0 or
298 dirty|flushing caps */
299 unsigned i_snap_caps; /* cap bits for snapped files */
300
301 int i_nr_by_mode[CEPH_FILE_MODE_NUM]; /* open file counts */
302
303 u32 i_truncate_seq; /* last truncate to smaller size */
304 u64 i_truncate_size; /* and the size we last truncated down to */
305 int i_truncate_pending; /* still need to call vmtruncate */
306
307 u64 i_max_size; /* max file size authorized by mds */
308 u64 i_reported_size; /* (max_)size reported to or requested of mds */
309 u64 i_wanted_max_size; /* offset we'd like to write too */
310 u64 i_requested_max_size; /* max_size we've requested */
311
312 /* held references to caps */
313 int i_pin_ref;
314 int i_rd_ref, i_rdcache_ref, i_wr_ref, i_wb_ref;
315 int i_wrbuffer_ref, i_wrbuffer_ref_head;
316 u32 i_shared_gen; /* increment each time we get FILE_SHARED */
317 u32 i_rdcache_gen; /* incremented each time we get FILE_CACHE. */
318 u32 i_rdcache_revoking; /* RDCACHE gen to async invalidate, if any */
319
320 struct list_head i_unsafe_writes; /* uncommitted sync writes */
321 struct list_head i_unsafe_dirops; /* uncommitted mds dir ops */
322 spinlock_t i_unsafe_lock;
323
324 struct ceph_snap_realm *i_snap_realm; /* snap realm (if caps) */
325 int i_snap_realm_counter; /* snap realm (if caps) */
326 struct list_head i_snap_realm_item;
327 struct list_head i_snap_flush_item;
328
329 struct work_struct i_wb_work; /* writeback work */
330 struct work_struct i_pg_inv_work; /* page invalidation work */
331
332 struct work_struct i_vmtruncate_work;
333
334 struct inode vfs_inode; /* at end */
335};
336
337static inline struct ceph_inode_info *ceph_inode(struct inode *inode)
338{
339 return container_of(inode, struct ceph_inode_info, vfs_inode);
340}
341
342static inline struct ceph_fs_client *ceph_inode_to_client(struct inode *inode)
343{
344 return (struct ceph_fs_client *)inode->i_sb->s_fs_info;
345}
346
347static inline struct ceph_fs_client *ceph_sb_to_client(struct super_block *sb)
348{
349 return (struct ceph_fs_client *)sb->s_fs_info;
350}
351
352static inline struct ceph_vino ceph_vino(struct inode *inode)
353{
354 return ceph_inode(inode)->i_vino;
355}
356
357/*
358 * ino_t is <64 bits on many architectures, blech.
359 *
360 * i_ino (kernel inode) st_ino (userspace)
361 * i386 32 32
362 * x86_64+ino32 64 32
363 * x86_64 64 64
364 */
365static inline u32 ceph_ino_to_ino32(__u64 vino)
366{
367 u32 ino = vino & 0xffffffff;
368 ino ^= vino >> 32;
369 if (!ino)
370 ino = 2;
371 return ino;
372}
373
374/*
375 * kernel i_ino value
376 */
377static inline ino_t ceph_vino_to_ino(struct ceph_vino vino)
378{
379#if BITS_PER_LONG == 32
380 return ceph_ino_to_ino32(vino.ino);
381#else
382 return (ino_t)vino.ino;
383#endif
384}
385
386/*
387 * user-visible ino (stat, filldir)
388 */
389#if BITS_PER_LONG == 32
390static inline ino_t ceph_translate_ino(struct super_block *sb, ino_t ino)
391{
392 return ino;
393}
394#else
395static inline ino_t ceph_translate_ino(struct super_block *sb, ino_t ino)
396{
397 if (ceph_test_mount_opt(ceph_sb_to_client(sb), INO32))
398 ino = ceph_ino_to_ino32(ino);
399 return ino;
400}
401#endif
402
403
404/* for printf-style formatting */
405#define ceph_vinop(i) ceph_inode(i)->i_vino.ino, ceph_inode(i)->i_vino.snap
406
407static inline u64 ceph_ino(struct inode *inode)
408{
409 return ceph_inode(inode)->i_vino.ino;
410}
411static inline u64 ceph_snap(struct inode *inode)
412{
413 return ceph_inode(inode)->i_vino.snap;
414}
415
416static inline int ceph_ino_compare(struct inode *inode, void *data)
417{
418 struct ceph_vino *pvino = (struct ceph_vino *)data;
419 struct ceph_inode_info *ci = ceph_inode(inode);
420 return ci->i_vino.ino == pvino->ino &&
421 ci->i_vino.snap == pvino->snap;
422}
423
424static inline struct inode *ceph_find_inode(struct super_block *sb,
425 struct ceph_vino vino)
426{
427 ino_t t = ceph_vino_to_ino(vino);
428 return ilookup5(sb, t, ceph_ino_compare, &vino);
429}
430
431
432/*
433 * Ceph inode.
434 */
435#define CEPH_I_NODELAY 4 /* do not delay cap release */
436#define CEPH_I_FLUSH 8 /* do not delay flush of dirty metadata */
437#define CEPH_I_NOFLUSH 16 /* do not flush dirty caps */
438
439static inline void ceph_i_clear(struct inode *inode, unsigned mask)
440{
441 struct ceph_inode_info *ci = ceph_inode(inode);
442
443 spin_lock(&ci->i_ceph_lock);
444 ci->i_ceph_flags &= ~mask;
445 spin_unlock(&ci->i_ceph_lock);
446}
447
448static inline void ceph_i_set(struct inode *inode, unsigned mask)
449{
450 struct ceph_inode_info *ci = ceph_inode(inode);
451
452 spin_lock(&ci->i_ceph_lock);
453 ci->i_ceph_flags |= mask;
454 spin_unlock(&ci->i_ceph_lock);
455}
456
457static inline bool ceph_i_test(struct inode *inode, unsigned mask)
458{
459 struct ceph_inode_info *ci = ceph_inode(inode);
460 bool r;
461
462 spin_lock(&ci->i_ceph_lock);
463 r = (ci->i_ceph_flags & mask) == mask;
464 spin_unlock(&ci->i_ceph_lock);
465 return r;
466}
467
468
469/* find a specific frag @f */
470extern struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci,
471 u32 f);
472
473/*
474 * choose fragment for value @v. copy frag content to pfrag, if leaf
475 * exists
476 */
477extern u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
478 struct ceph_inode_frag *pfrag,
479 int *found);
480
481static inline struct ceph_dentry_info *ceph_dentry(struct dentry *dentry)
482{
483 return (struct ceph_dentry_info *)dentry->d_fsdata;
484}
485
486static inline loff_t ceph_make_fpos(unsigned frag, unsigned off)
487{
488 return ((loff_t)frag << 32) | (loff_t)off;
489}
490
491/*
492 * set/clear directory D_COMPLETE flag
493 */
494void ceph_dir_set_complete(struct inode *inode);
495void ceph_dir_clear_complete(struct inode *inode);
496bool ceph_dir_test_complete(struct inode *inode);
497
498/*
499 * caps helpers
500 */
501static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)
502{
503 return !RB_EMPTY_ROOT(&ci->i_caps);
504}
505
506extern int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented);
507extern int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int t);
508extern int __ceph_caps_issued_other(struct ceph_inode_info *ci,
509 struct ceph_cap *cap);
510
511static inline int ceph_caps_issued(struct ceph_inode_info *ci)
512{
513 int issued;
514 spin_lock(&ci->i_ceph_lock);
515 issued = __ceph_caps_issued(ci, NULL);
516 spin_unlock(&ci->i_ceph_lock);
517 return issued;
518}
519
520static inline int ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask,
521 int touch)
522{
523 int r;
524 spin_lock(&ci->i_ceph_lock);
525 r = __ceph_caps_issued_mask(ci, mask, touch);
526 spin_unlock(&ci->i_ceph_lock);
527 return r;
528}
529
530static inline int __ceph_caps_dirty(struct ceph_inode_info *ci)
531{
532 return ci->i_dirty_caps | ci->i_flushing_caps;
533}
534extern int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask);
535
536extern int ceph_caps_revoking(struct ceph_inode_info *ci, int mask);
537extern int __ceph_caps_used(struct ceph_inode_info *ci);
538
539extern int __ceph_caps_file_wanted(struct ceph_inode_info *ci);
540
541/*
542 * wanted, by virtue of open file modes AND cap refs (buffered/cached data)
543 */
544static inline int __ceph_caps_wanted(struct ceph_inode_info *ci)
545{
546 int w = __ceph_caps_file_wanted(ci) | __ceph_caps_used(ci);
547 if (w & CEPH_CAP_FILE_BUFFER)
548 w |= CEPH_CAP_FILE_EXCL; /* we want EXCL if dirty data */
549 return w;
550}
551
552/* what the mds thinks we want */
553extern int __ceph_caps_mds_wanted(struct ceph_inode_info *ci);
554
555extern void ceph_caps_init(struct ceph_mds_client *mdsc);
556extern void ceph_caps_finalize(struct ceph_mds_client *mdsc);
557extern void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta);
558extern int ceph_reserve_caps(struct ceph_mds_client *mdsc,
559 struct ceph_cap_reservation *ctx, int need);
560extern int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
561 struct ceph_cap_reservation *ctx);
562extern void ceph_reservation_status(struct ceph_fs_client *client,
563 int *total, int *avail, int *used,
564 int *reserved, int *min);
565
566
567
568/*
569 * we keep buffered readdir results attached to file->private_data
570 */
571#define CEPH_F_SYNC 1
572#define CEPH_F_ATEND 2
573
574struct ceph_file_info {
575 short fmode; /* initialized on open */
576 short flags; /* CEPH_F_* */
577
578 /* readdir: position within the dir */
579 u32 frag;
580 struct ceph_mds_request *last_readdir;
581
582 /* readdir: position within a frag */
583 unsigned offset; /* offset of last chunk, adjusted for . and .. */
584 u64 next_offset; /* offset of next chunk (last_name's + 1) */
585 char *last_name; /* last entry in previous chunk */
586 struct dentry *dentry; /* next dentry (for dcache readdir) */
587 unsigned long dir_release_count;
588
589 /* used for -o dirstat read() on directory thing */
590 char *dir_info;
591 int dir_info_len;
592};
593
594
595
596/*
597 * A "snap realm" describes a subset of the file hierarchy sharing
598 * the same set of snapshots that apply to it. The realms themselves
599 * are organized into a hierarchy, such that children inherit (some of)
600 * the snapshots of their parents.
601 *
602 * All inodes within the realm that have capabilities are linked into a
603 * per-realm list.
604 */
605struct ceph_snap_realm {
606 u64 ino;
607 atomic_t nref;
608 struct rb_node node;
609
610 u64 created, seq;
611 u64 parent_ino;
612 u64 parent_since; /* snapid when our current parent became so */
613
614 u64 *prior_parent_snaps; /* snaps inherited from any parents we */
615 int num_prior_parent_snaps; /* had prior to parent_since */
616 u64 *snaps; /* snaps specific to this realm */
617 int num_snaps;
618
619 struct ceph_snap_realm *parent;
620 struct list_head children; /* list of child realms */
621 struct list_head child_item;
622
623 struct list_head empty_item; /* if i have ref==0 */
624
625 struct list_head dirty_item; /* if realm needs new context */
626
627 /* the current set of snaps for this realm */
628 struct ceph_snap_context *cached_context;
629
630 struct list_head inodes_with_caps;
631 spinlock_t inodes_with_caps_lock;
632};
633
634static inline int default_congestion_kb(void)
635{
636 int congestion_kb;
637
638 /*
639 * Copied from NFS
640 *
641 * congestion size, scale with available memory.
642 *
643 * 64MB: 8192k
644 * 128MB: 11585k
645 * 256MB: 16384k
646 * 512MB: 23170k
647 * 1GB: 32768k
648 * 2GB: 46340k
649 * 4GB: 65536k
650 * 8GB: 92681k
651 * 16GB: 131072k
652 *
653 * This allows larger machines to have larger/more transfers.
654 * Limit the default to 256M
655 */
656 congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
657 if (congestion_kb > 256*1024)
658 congestion_kb = 256*1024;
659
660 return congestion_kb;
661}
662
663
664
665/* snap.c */
666struct ceph_snap_realm *ceph_lookup_snap_realm(struct ceph_mds_client *mdsc,
667 u64 ino);
668extern void ceph_get_snap_realm(struct ceph_mds_client *mdsc,
669 struct ceph_snap_realm *realm);
670extern void ceph_put_snap_realm(struct ceph_mds_client *mdsc,
671 struct ceph_snap_realm *realm);
672extern int ceph_update_snap_trace(struct ceph_mds_client *m,
673 void *p, void *e, bool deletion);
674extern void ceph_handle_snap(struct ceph_mds_client *mdsc,
675 struct ceph_mds_session *session,
676 struct ceph_msg *msg);
677extern void ceph_queue_cap_snap(struct ceph_inode_info *ci);
678extern int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
679 struct ceph_cap_snap *capsnap);
680extern void ceph_cleanup_empty_realms(struct ceph_mds_client *mdsc);
681
682/*
683 * a cap_snap is "pending" if it is still awaiting an in-progress
684 * sync write (that may/may not still update size, mtime, etc.).
685 */
686static inline bool __ceph_have_pending_cap_snap(struct ceph_inode_info *ci)
687{
688 return !list_empty(&ci->i_cap_snaps) &&
689 list_entry(ci->i_cap_snaps.prev, struct ceph_cap_snap,
690 ci_item)->writing;
691}
692
693/* inode.c */
694extern const struct inode_operations ceph_file_iops;
695
696extern struct inode *ceph_alloc_inode(struct super_block *sb);
697extern void ceph_destroy_inode(struct inode *inode);
698
699extern struct inode *ceph_get_inode(struct super_block *sb,
700 struct ceph_vino vino);
701extern struct inode *ceph_get_snapdir(struct inode *parent);
702extern int ceph_fill_file_size(struct inode *inode, int issued,
703 u32 truncate_seq, u64 truncate_size, u64 size);
704extern void ceph_fill_file_time(struct inode *inode, int issued,
705 u64 time_warp_seq, struct timespec *ctime,
706 struct timespec *mtime, struct timespec *atime);
707extern int ceph_fill_trace(struct super_block *sb,
708 struct ceph_mds_request *req,
709 struct ceph_mds_session *session);
710extern int ceph_readdir_prepopulate(struct ceph_mds_request *req,
711 struct ceph_mds_session *session);
712
713extern int ceph_inode_holds_cap(struct inode *inode, int mask);
714
715extern int ceph_inode_set_size(struct inode *inode, loff_t size);
716extern void __ceph_do_pending_vmtruncate(struct inode *inode);
717extern void ceph_queue_vmtruncate(struct inode *inode);
718
719extern void ceph_queue_invalidate(struct inode *inode);
720extern void ceph_queue_writeback(struct inode *inode);
721
722extern int ceph_do_getattr(struct inode *inode, int mask);
723extern int ceph_permission(struct inode *inode, int mask);
724extern int ceph_setattr(struct dentry *dentry, struct iattr *attr);
725extern int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
726 struct kstat *stat);
727
728/* xattr.c */
729extern int ceph_setxattr(struct dentry *, const char *, const void *,
730 size_t, int);
731extern ssize_t ceph_getxattr(struct dentry *, const char *, void *, size_t);
732extern ssize_t ceph_listxattr(struct dentry *, char *, size_t);
733extern int ceph_removexattr(struct dentry *, const char *);
734extern void __ceph_build_xattrs_blob(struct ceph_inode_info *ci);
735extern void __ceph_destroy_xattrs(struct ceph_inode_info *ci);
736extern void __init ceph_xattr_init(void);
737extern void ceph_xattr_exit(void);
738
739/* caps.c */
740extern const char *ceph_cap_string(int c);
741extern void ceph_handle_caps(struct ceph_mds_session *session,
742 struct ceph_msg *msg);
743extern int ceph_add_cap(struct inode *inode,
744 struct ceph_mds_session *session, u64 cap_id,
745 int fmode, unsigned issued, unsigned wanted,
746 unsigned cap, unsigned seq, u64 realmino, int flags,
747 struct ceph_cap_reservation *caps_reservation);
748extern void __ceph_remove_cap(struct ceph_cap *cap);
749static inline void ceph_remove_cap(struct ceph_cap *cap)
750{
751 spin_lock(&cap->ci->i_ceph_lock);
752 __ceph_remove_cap(cap);
753 spin_unlock(&cap->ci->i_ceph_lock);
754}
755extern void ceph_put_cap(struct ceph_mds_client *mdsc,
756 struct ceph_cap *cap);
757
758extern void ceph_queue_caps_release(struct inode *inode);
759extern int ceph_write_inode(struct inode *inode, struct writeback_control *wbc);
760extern int ceph_fsync(struct file *file, loff_t start, loff_t end,
761 int datasync);
762extern void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
763 struct ceph_mds_session *session);
764extern struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci,
765 int mds);
766extern int ceph_get_cap_mds(struct inode *inode);
767extern void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps);
768extern void ceph_put_cap_refs(struct ceph_inode_info *ci, int had);
769extern void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
770 struct ceph_snap_context *snapc);
771extern void __ceph_flush_snaps(struct ceph_inode_info *ci,
772 struct ceph_mds_session **psession,
773 int again);
774extern void ceph_check_caps(struct ceph_inode_info *ci, int flags,
775 struct ceph_mds_session *session);
776extern void ceph_check_delayed_caps(struct ceph_mds_client *mdsc);
777extern void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc);
778
779extern int ceph_encode_inode_release(void **p, struct inode *inode,
780 int mds, int drop, int unless, int force);
781extern int ceph_encode_dentry_release(void **p, struct dentry *dn,
782 int mds, int drop, int unless);
783
784extern int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
785 int *got, loff_t endoff);
786
787/* for counting open files by mode */
788static inline void __ceph_get_fmode(struct ceph_inode_info *ci, int mode)
789{
790 ci->i_nr_by_mode[mode]++;
791}
792extern void ceph_put_fmode(struct ceph_inode_info *ci, int mode);
793
794/* addr.c */
795extern const struct address_space_operations ceph_aops;
796extern int ceph_mmap(struct file *file, struct vm_area_struct *vma);
797
798/* file.c */
799extern const struct file_operations ceph_file_fops;
800extern const struct address_space_operations ceph_aops;
801extern int ceph_copy_to_page_vector(struct page **pages,
802 const char *data,
803 loff_t off, size_t len);
804extern int ceph_copy_from_page_vector(struct page **pages,
805 char *data,
806 loff_t off, size_t len);
807extern struct page **ceph_alloc_page_vector(int num_pages, gfp_t flags);
808extern int ceph_open(struct inode *inode, struct file *file);
809extern struct dentry *ceph_lookup_open(struct inode *dir, struct dentry *dentry,
810 struct nameidata *nd, int mode,
811 int locked_dir);
812extern int ceph_release(struct inode *inode, struct file *filp);
813
814/* dir.c */
815extern const struct file_operations ceph_dir_fops;
816extern const struct inode_operations ceph_dir_iops;
817extern const struct dentry_operations ceph_dentry_ops, ceph_snap_dentry_ops,
818 ceph_snapdir_dentry_ops;
819
820extern int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry);
821extern int ceph_handle_snapdir(struct ceph_mds_request *req,
822 struct dentry *dentry, int err);
823extern struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
824 struct dentry *dentry, int err);
825
826extern void ceph_dentry_lru_add(struct dentry *dn);
827extern void ceph_dentry_lru_touch(struct dentry *dn);
828extern void ceph_dentry_lru_del(struct dentry *dn);
829extern void ceph_invalidate_dentry_lease(struct dentry *dentry);
830extern unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn);
831extern struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry);
832
833/*
834 * our d_ops vary depending on whether the inode is live,
835 * snapshotted (read-only), or a virtual ".snap" directory.
836 */
837int ceph_init_dentry(struct dentry *dentry);
838
839
840/* ioctl.c */
841extern long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
842
843/* export.c */
844extern const struct export_operations ceph_export_ops;
845
846/* locks.c */
847extern int ceph_lock(struct file *file, int cmd, struct file_lock *fl);
848extern int ceph_flock(struct file *file, int cmd, struct file_lock *fl);
849extern void ceph_count_locks(struct inode *inode, int *p_num, int *f_num);
850extern int ceph_encode_locks(struct inode *i, struct ceph_pagelist *p,
851 int p_locks, int f_locks);
852extern int lock_to_ceph_filelock(struct file_lock *fl, struct ceph_filelock *c);
853
854/* debugfs.c */
855extern int ceph_fs_debugfs_init(struct ceph_fs_client *client);
856extern void ceph_fs_debugfs_cleanup(struct ceph_fs_client *client);
857
858#endif /* _FS_CEPH_SUPER_H */
1#ifndef _FS_CEPH_SUPER_H
2#define _FS_CEPH_SUPER_H
3
4#include <linux/ceph/ceph_debug.h>
5
6#include <asm/unaligned.h>
7#include <linux/backing-dev.h>
8#include <linux/completion.h>
9#include <linux/exportfs.h>
10#include <linux/fs.h>
11#include <linux/mempool.h>
12#include <linux/pagemap.h>
13#include <linux/wait.h>
14#include <linux/writeback.h>
15#include <linux/slab.h>
16#include <linux/posix_acl.h>
17
18#include <linux/ceph/libceph.h>
19
20#ifdef CONFIG_CEPH_FSCACHE
21#include <linux/fscache.h>
22#endif
23
24/* f_type in struct statfs */
25#define CEPH_SUPER_MAGIC 0x00c36400
26
27/* large granularity for statfs utilization stats to facilitate
28 * large volume sizes on 32-bit machines. */
29#define CEPH_BLOCK_SHIFT 22 /* 4 MB */
30#define CEPH_BLOCK (1 << CEPH_BLOCK_SHIFT)
31
32#define CEPH_MOUNT_OPT_DIRSTAT (1<<4) /* `cat dirname` for stats */
33#define CEPH_MOUNT_OPT_RBYTES (1<<5) /* dir st_bytes = rbytes */
34#define CEPH_MOUNT_OPT_NOASYNCREADDIR (1<<7) /* no dcache readdir */
35#define CEPH_MOUNT_OPT_INO32 (1<<8) /* 32 bit inos */
36#define CEPH_MOUNT_OPT_DCACHE (1<<9) /* use dcache for readdir etc */
37#define CEPH_MOUNT_OPT_FSCACHE (1<<10) /* use fscache */
38#define CEPH_MOUNT_OPT_NOPOOLPERM (1<<11) /* no pool permission check */
39
40#define CEPH_MOUNT_OPT_DEFAULT CEPH_MOUNT_OPT_DCACHE
41
42#define ceph_set_mount_opt(fsc, opt) \
43 (fsc)->mount_options->flags |= CEPH_MOUNT_OPT_##opt;
44#define ceph_test_mount_opt(fsc, opt) \
45 (!!((fsc)->mount_options->flags & CEPH_MOUNT_OPT_##opt))
46
47#define CEPH_RSIZE_DEFAULT 0 /* max read size */
48#define CEPH_RASIZE_DEFAULT (8192*1024) /* readahead */
49#define CEPH_MAX_READDIR_DEFAULT 1024
50#define CEPH_MAX_READDIR_BYTES_DEFAULT (512*1024)
51#define CEPH_SNAPDIRNAME_DEFAULT ".snap"
52
53struct ceph_mount_options {
54 int flags;
55 int sb_flags;
56
57 int wsize; /* max write size */
58 int rsize; /* max read size */
59 int rasize; /* max readahead */
60 int congestion_kb; /* max writeback in flight */
61 int caps_wanted_delay_min, caps_wanted_delay_max;
62 int cap_release_safety;
63 int max_readdir; /* max readdir result (entires) */
64 int max_readdir_bytes; /* max readdir result (bytes) */
65
66 /*
67 * everything above this point can be memcmp'd; everything below
68 * is handled in compare_mount_options()
69 */
70
71 char *snapdir_name; /* default ".snap" */
72};
73
74struct ceph_fs_client {
75 struct super_block *sb;
76
77 struct ceph_mount_options *mount_options;
78 struct ceph_client *client;
79
80 unsigned long mount_state;
81 int min_caps; /* min caps i added */
82
83 struct ceph_mds_client *mdsc;
84
85 /* writeback */
86 mempool_t *wb_pagevec_pool;
87 struct workqueue_struct *wb_wq;
88 struct workqueue_struct *pg_inv_wq;
89 struct workqueue_struct *trunc_wq;
90 atomic_long_t writeback_count;
91
92 struct backing_dev_info backing_dev_info;
93
94#ifdef CONFIG_DEBUG_FS
95 struct dentry *debugfs_dentry_lru, *debugfs_caps;
96 struct dentry *debugfs_congestion_kb;
97 struct dentry *debugfs_bdi;
98 struct dentry *debugfs_mdsc, *debugfs_mdsmap;
99 struct dentry *debugfs_mds_sessions;
100#endif
101
102#ifdef CONFIG_CEPH_FSCACHE
103 struct fscache_cookie *fscache;
104 struct workqueue_struct *revalidate_wq;
105#endif
106};
107
108
109/*
110 * File i/o capability. This tracks shared state with the metadata
111 * server that allows us to cache or writeback attributes or to read
112 * and write data. For any given inode, we should have one or more
113 * capabilities, one issued by each metadata server, and our
114 * cumulative access is the OR of all issued capabilities.
115 *
116 * Each cap is referenced by the inode's i_caps rbtree and by per-mds
117 * session capability lists.
118 */
119struct ceph_cap {
120 struct ceph_inode_info *ci;
121 struct rb_node ci_node; /* per-ci cap tree */
122 struct ceph_mds_session *session;
123 struct list_head session_caps; /* per-session caplist */
124 u64 cap_id; /* unique cap id (mds provided) */
125 union {
126 /* in-use caps */
127 struct {
128 int issued; /* latest, from the mds */
129 int implemented; /* implemented superset of
130 issued (for revocation) */
131 int mds, mds_wanted;
132 };
133 /* caps to release */
134 struct {
135 u64 cap_ino;
136 int queue_release;
137 };
138 };
139 u32 seq, issue_seq, mseq;
140 u32 cap_gen; /* active/stale cycle */
141 unsigned long last_used;
142 struct list_head caps_item;
143};
144
145#define CHECK_CAPS_NODELAY 1 /* do not delay any further */
146#define CHECK_CAPS_AUTHONLY 2 /* only check auth cap */
147#define CHECK_CAPS_FLUSH 4 /* flush any dirty caps */
148
149/*
150 * Snapped cap state that is pending flush to mds. When a snapshot occurs,
151 * we first complete any in-process sync writes and writeback any dirty
152 * data before flushing the snapped state (tracked here) back to the MDS.
153 */
154struct ceph_cap_snap {
155 atomic_t nref;
156 struct ceph_inode_info *ci;
157 struct list_head ci_item, flushing_item;
158
159 u64 follows, flush_tid;
160 int issued, dirty;
161 struct ceph_snap_context *context;
162
163 umode_t mode;
164 kuid_t uid;
165 kgid_t gid;
166
167 struct ceph_buffer *xattr_blob;
168 u64 xattr_version;
169
170 u64 size;
171 struct timespec mtime, atime, ctime;
172 u64 time_warp_seq;
173 int writing; /* a sync write is still in progress */
174 int dirty_pages; /* dirty pages awaiting writeback */
175 bool inline_data;
176 bool need_flush;
177};
178
179static inline void ceph_put_cap_snap(struct ceph_cap_snap *capsnap)
180{
181 if (atomic_dec_and_test(&capsnap->nref)) {
182 if (capsnap->xattr_blob)
183 ceph_buffer_put(capsnap->xattr_blob);
184 kfree(capsnap);
185 }
186}
187
188struct ceph_cap_flush {
189 u64 tid;
190 int caps;
191 struct rb_node g_node; // global
192 union {
193 struct rb_node i_node; // inode
194 struct list_head list;
195 };
196};
197
198/*
199 * The frag tree describes how a directory is fragmented, potentially across
200 * multiple metadata servers. It is also used to indicate points where
201 * metadata authority is delegated, and whether/where metadata is replicated.
202 *
203 * A _leaf_ frag will be present in the i_fragtree IFF there is
204 * delegation info. That is, if mds >= 0 || ndist > 0.
205 */
206#define CEPH_MAX_DIRFRAG_REP 4
207
208struct ceph_inode_frag {
209 struct rb_node node;
210
211 /* fragtree state */
212 u32 frag;
213 int split_by; /* i.e. 2^(split_by) children */
214
215 /* delegation and replication info */
216 int mds; /* -1 if same authority as parent */
217 int ndist; /* >0 if replicated */
218 int dist[CEPH_MAX_DIRFRAG_REP];
219};
220
221/*
222 * We cache inode xattrs as an encoded blob until they are first used,
223 * at which point we parse them into an rbtree.
224 */
225struct ceph_inode_xattr {
226 struct rb_node node;
227
228 const char *name;
229 int name_len;
230 const char *val;
231 int val_len;
232 int dirty;
233
234 int should_free_name;
235 int should_free_val;
236};
237
238/*
239 * Ceph dentry state
240 */
241struct ceph_dentry_info {
242 struct ceph_mds_session *lease_session;
243 u32 lease_gen, lease_shared_gen;
244 u32 lease_seq;
245 unsigned long lease_renew_after, lease_renew_from;
246 struct list_head lru;
247 struct dentry *dentry;
248 u64 time;
249 u64 offset;
250};
251
252struct ceph_inode_xattrs_info {
253 /*
254 * (still encoded) xattr blob. we avoid the overhead of parsing
255 * this until someone actually calls getxattr, etc.
256 *
257 * blob->vec.iov_len == 4 implies there are no xattrs; blob ==
258 * NULL means we don't know.
259 */
260 struct ceph_buffer *blob, *prealloc_blob;
261
262 struct rb_root index;
263 bool dirty;
264 int count;
265 int names_size;
266 int vals_size;
267 u64 version, index_version;
268};
269
270/*
271 * Ceph inode.
272 */
273struct ceph_inode_info {
274 struct ceph_vino i_vino; /* ceph ino + snap */
275
276 spinlock_t i_ceph_lock;
277
278 u64 i_version;
279 u64 i_inline_version;
280 u32 i_time_warp_seq;
281
282 unsigned i_ceph_flags;
283 atomic64_t i_release_count;
284 atomic64_t i_ordered_count;
285 atomic64_t i_complete_seq[2];
286
287 struct ceph_dir_layout i_dir_layout;
288 struct ceph_file_layout i_layout;
289 size_t i_pool_ns_len;
290 char *i_symlink;
291
292 /* for dirs */
293 struct timespec i_rctime;
294 u64 i_rbytes, i_rfiles, i_rsubdirs;
295 u64 i_files, i_subdirs;
296
297 struct rb_root i_fragtree;
298 struct mutex i_fragtree_mutex;
299
300 struct ceph_inode_xattrs_info i_xattrs;
301
302 /* capabilities. protected _both_ by i_ceph_lock and cap->session's
303 * s_mutex. */
304 struct rb_root i_caps; /* cap list */
305 struct ceph_cap *i_auth_cap; /* authoritative cap, if any */
306 unsigned i_dirty_caps, i_flushing_caps; /* mask of dirtied fields */
307 struct list_head i_dirty_item, i_flushing_item;
308 /* we need to track cap writeback on a per-cap-bit basis, to allow
309 * overlapping, pipelined cap flushes to the mds. we can probably
310 * reduce the tid to 8 bits if we're concerned about inode size. */
311 struct ceph_cap_flush *i_prealloc_cap_flush;
312 struct rb_root i_cap_flush_tree;
313 wait_queue_head_t i_cap_wq; /* threads waiting on a capability */
314 unsigned long i_hold_caps_min; /* jiffies */
315 unsigned long i_hold_caps_max; /* jiffies */
316 struct list_head i_cap_delay_list; /* for delayed cap release to mds */
317 struct ceph_cap_reservation i_cap_migration_resv;
318 struct list_head i_cap_snaps; /* snapped state pending flush to mds */
319 struct ceph_snap_context *i_head_snapc; /* set if wr_buffer_head > 0 or
320 dirty|flushing caps */
321 unsigned i_snap_caps; /* cap bits for snapped files */
322
323 int i_nr_by_mode[CEPH_FILE_MODE_NUM]; /* open file counts */
324
325 struct mutex i_truncate_mutex;
326 u32 i_truncate_seq; /* last truncate to smaller size */
327 u64 i_truncate_size; /* and the size we last truncated down to */
328 int i_truncate_pending; /* still need to call vmtruncate */
329
330 u64 i_max_size; /* max file size authorized by mds */
331 u64 i_reported_size; /* (max_)size reported to or requested of mds */
332 u64 i_wanted_max_size; /* offset we'd like to write too */
333 u64 i_requested_max_size; /* max_size we've requested */
334
335 /* held references to caps */
336 int i_pin_ref;
337 int i_rd_ref, i_rdcache_ref, i_wr_ref, i_wb_ref;
338 int i_wrbuffer_ref, i_wrbuffer_ref_head;
339 u32 i_shared_gen; /* increment each time we get FILE_SHARED */
340 u32 i_rdcache_gen; /* incremented each time we get FILE_CACHE. */
341 u32 i_rdcache_revoking; /* RDCACHE gen to async invalidate, if any */
342
343 struct list_head i_unsafe_writes; /* uncommitted sync writes */
344 struct list_head i_unsafe_dirops; /* uncommitted mds dir ops */
345 struct list_head i_unsafe_iops; /* uncommitted mds inode ops */
346 spinlock_t i_unsafe_lock;
347
348 struct ceph_snap_realm *i_snap_realm; /* snap realm (if caps) */
349 int i_snap_realm_counter; /* snap realm (if caps) */
350 struct list_head i_snap_realm_item;
351 struct list_head i_snap_flush_item;
352
353 struct work_struct i_wb_work; /* writeback work */
354 struct work_struct i_pg_inv_work; /* page invalidation work */
355
356 struct work_struct i_vmtruncate_work;
357
358#ifdef CONFIG_CEPH_FSCACHE
359 struct fscache_cookie *fscache;
360 u32 i_fscache_gen; /* sequence, for delayed fscache validate */
361 struct work_struct i_revalidate_work;
362#endif
363 struct inode vfs_inode; /* at end */
364};
365
366static inline struct ceph_inode_info *ceph_inode(struct inode *inode)
367{
368 return container_of(inode, struct ceph_inode_info, vfs_inode);
369}
370
371static inline struct ceph_fs_client *ceph_inode_to_client(struct inode *inode)
372{
373 return (struct ceph_fs_client *)inode->i_sb->s_fs_info;
374}
375
376static inline struct ceph_fs_client *ceph_sb_to_client(struct super_block *sb)
377{
378 return (struct ceph_fs_client *)sb->s_fs_info;
379}
380
381static inline struct ceph_vino ceph_vino(struct inode *inode)
382{
383 return ceph_inode(inode)->i_vino;
384}
385
386/*
387 * ino_t is <64 bits on many architectures, blech.
388 *
389 * i_ino (kernel inode) st_ino (userspace)
390 * i386 32 32
391 * x86_64+ino32 64 32
392 * x86_64 64 64
393 */
394static inline u32 ceph_ino_to_ino32(__u64 vino)
395{
396 u32 ino = vino & 0xffffffff;
397 ino ^= vino >> 32;
398 if (!ino)
399 ino = 2;
400 return ino;
401}
402
403/*
404 * kernel i_ino value
405 */
406static inline ino_t ceph_vino_to_ino(struct ceph_vino vino)
407{
408#if BITS_PER_LONG == 32
409 return ceph_ino_to_ino32(vino.ino);
410#else
411 return (ino_t)vino.ino;
412#endif
413}
414
415/*
416 * user-visible ino (stat, filldir)
417 */
418#if BITS_PER_LONG == 32
419static inline ino_t ceph_translate_ino(struct super_block *sb, ino_t ino)
420{
421 return ino;
422}
423#else
424static inline ino_t ceph_translate_ino(struct super_block *sb, ino_t ino)
425{
426 if (ceph_test_mount_opt(ceph_sb_to_client(sb), INO32))
427 ino = ceph_ino_to_ino32(ino);
428 return ino;
429}
430#endif
431
432
433/* for printf-style formatting */
434#define ceph_vinop(i) ceph_inode(i)->i_vino.ino, ceph_inode(i)->i_vino.snap
435
436static inline u64 ceph_ino(struct inode *inode)
437{
438 return ceph_inode(inode)->i_vino.ino;
439}
440static inline u64 ceph_snap(struct inode *inode)
441{
442 return ceph_inode(inode)->i_vino.snap;
443}
444
445static inline int ceph_ino_compare(struct inode *inode, void *data)
446{
447 struct ceph_vino *pvino = (struct ceph_vino *)data;
448 struct ceph_inode_info *ci = ceph_inode(inode);
449 return ci->i_vino.ino == pvino->ino &&
450 ci->i_vino.snap == pvino->snap;
451}
452
453static inline struct inode *ceph_find_inode(struct super_block *sb,
454 struct ceph_vino vino)
455{
456 ino_t t = ceph_vino_to_ino(vino);
457 return ilookup5(sb, t, ceph_ino_compare, &vino);
458}
459
460
461/*
462 * Ceph inode.
463 */
464#define CEPH_I_DIR_ORDERED (1 << 0) /* dentries in dir are ordered */
465#define CEPH_I_NODELAY (1 << 1) /* do not delay cap release */
466#define CEPH_I_FLUSH (1 << 2) /* do not delay flush of dirty metadata */
467#define CEPH_I_NOFLUSH (1 << 3) /* do not flush dirty caps */
468#define CEPH_I_POOL_PERM (1 << 4) /* pool rd/wr bits are valid */
469#define CEPH_I_POOL_RD (1 << 5) /* can read from pool */
470#define CEPH_I_POOL_WR (1 << 6) /* can write to pool */
471#define CEPH_I_SEC_INITED (1 << 7) /* security initialized */
472
473static inline void __ceph_dir_set_complete(struct ceph_inode_info *ci,
474 long long release_count,
475 long long ordered_count)
476{
477 smp_mb__before_atomic();
478 atomic64_set(&ci->i_complete_seq[0], release_count);
479 atomic64_set(&ci->i_complete_seq[1], ordered_count);
480}
481
482static inline void __ceph_dir_clear_complete(struct ceph_inode_info *ci)
483{
484 atomic64_inc(&ci->i_release_count);
485}
486
487static inline void __ceph_dir_clear_ordered(struct ceph_inode_info *ci)
488{
489 atomic64_inc(&ci->i_ordered_count);
490}
491
492static inline bool __ceph_dir_is_complete(struct ceph_inode_info *ci)
493{
494 return atomic64_read(&ci->i_complete_seq[0]) ==
495 atomic64_read(&ci->i_release_count);
496}
497
498static inline bool __ceph_dir_is_complete_ordered(struct ceph_inode_info *ci)
499{
500 return atomic64_read(&ci->i_complete_seq[0]) ==
501 atomic64_read(&ci->i_release_count) &&
502 atomic64_read(&ci->i_complete_seq[1]) ==
503 atomic64_read(&ci->i_ordered_count);
504}
505
506static inline void ceph_dir_clear_complete(struct inode *inode)
507{
508 __ceph_dir_clear_complete(ceph_inode(inode));
509}
510
511static inline void ceph_dir_clear_ordered(struct inode *inode)
512{
513 __ceph_dir_clear_ordered(ceph_inode(inode));
514}
515
516static inline bool ceph_dir_is_complete_ordered(struct inode *inode)
517{
518 bool ret = __ceph_dir_is_complete_ordered(ceph_inode(inode));
519 smp_rmb();
520 return ret;
521}
522
523/* find a specific frag @f */
524extern struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci,
525 u32 f);
526
527/*
528 * choose fragment for value @v. copy frag content to pfrag, if leaf
529 * exists
530 */
531extern u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
532 struct ceph_inode_frag *pfrag,
533 int *found);
534
535static inline struct ceph_dentry_info *ceph_dentry(struct dentry *dentry)
536{
537 return (struct ceph_dentry_info *)dentry->d_fsdata;
538}
539
540static inline loff_t ceph_make_fpos(unsigned frag, unsigned off)
541{
542 return ((loff_t)frag << 32) | (loff_t)off;
543}
544
545/*
546 * caps helpers
547 */
548static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)
549{
550 return !RB_EMPTY_ROOT(&ci->i_caps);
551}
552
553extern int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented);
554extern int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int t);
555extern int __ceph_caps_issued_other(struct ceph_inode_info *ci,
556 struct ceph_cap *cap);
557
558static inline int ceph_caps_issued(struct ceph_inode_info *ci)
559{
560 int issued;
561 spin_lock(&ci->i_ceph_lock);
562 issued = __ceph_caps_issued(ci, NULL);
563 spin_unlock(&ci->i_ceph_lock);
564 return issued;
565}
566
567static inline int ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask,
568 int touch)
569{
570 int r;
571 spin_lock(&ci->i_ceph_lock);
572 r = __ceph_caps_issued_mask(ci, mask, touch);
573 spin_unlock(&ci->i_ceph_lock);
574 return r;
575}
576
577static inline int __ceph_caps_dirty(struct ceph_inode_info *ci)
578{
579 return ci->i_dirty_caps | ci->i_flushing_caps;
580}
581extern struct ceph_cap_flush *ceph_alloc_cap_flush(void);
582extern void ceph_free_cap_flush(struct ceph_cap_flush *cf);
583extern int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
584 struct ceph_cap_flush **pcf);
585
586extern int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
587 struct ceph_cap *ocap, int mask);
588extern int ceph_caps_revoking(struct ceph_inode_info *ci, int mask);
589extern int __ceph_caps_used(struct ceph_inode_info *ci);
590
591extern int __ceph_caps_file_wanted(struct ceph_inode_info *ci);
592
593/*
594 * wanted, by virtue of open file modes AND cap refs (buffered/cached data)
595 */
596static inline int __ceph_caps_wanted(struct ceph_inode_info *ci)
597{
598 int w = __ceph_caps_file_wanted(ci) | __ceph_caps_used(ci);
599 if (w & CEPH_CAP_FILE_BUFFER)
600 w |= CEPH_CAP_FILE_EXCL; /* we want EXCL if dirty data */
601 return w;
602}
603
604/* what the mds thinks we want */
605extern int __ceph_caps_mds_wanted(struct ceph_inode_info *ci);
606
607extern void ceph_caps_init(struct ceph_mds_client *mdsc);
608extern void ceph_caps_finalize(struct ceph_mds_client *mdsc);
609extern void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta);
610extern void ceph_reserve_caps(struct ceph_mds_client *mdsc,
611 struct ceph_cap_reservation *ctx, int need);
612extern int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
613 struct ceph_cap_reservation *ctx);
614extern void ceph_reservation_status(struct ceph_fs_client *client,
615 int *total, int *avail, int *used,
616 int *reserved, int *min);
617
618
619
620/*
621 * we keep buffered readdir results attached to file->private_data
622 */
623#define CEPH_F_SYNC 1
624#define CEPH_F_ATEND 2
625
626struct ceph_file_info {
627 short fmode; /* initialized on open */
628 short flags; /* CEPH_F_* */
629
630 /* readdir: position within the dir */
631 u32 frag;
632 struct ceph_mds_request *last_readdir;
633
634 /* readdir: position within a frag */
635 unsigned offset; /* offset of last chunk, adjusted for . and .. */
636 unsigned next_offset; /* offset of next chunk (last_name's + 1) */
637 char *last_name; /* last entry in previous chunk */
638 long long dir_release_count;
639 long long dir_ordered_count;
640 int readdir_cache_idx;
641
642 /* used for -o dirstat read() on directory thing */
643 char *dir_info;
644 int dir_info_len;
645};
646
647struct ceph_readdir_cache_control {
648 struct page *page;
649 struct dentry **dentries;
650 int index;
651};
652
653/*
654 * A "snap realm" describes a subset of the file hierarchy sharing
655 * the same set of snapshots that apply to it. The realms themselves
656 * are organized into a hierarchy, such that children inherit (some of)
657 * the snapshots of their parents.
658 *
659 * All inodes within the realm that have capabilities are linked into a
660 * per-realm list.
661 */
662struct ceph_snap_realm {
663 u64 ino;
664 atomic_t nref;
665 struct rb_node node;
666
667 u64 created, seq;
668 u64 parent_ino;
669 u64 parent_since; /* snapid when our current parent became so */
670
671 u64 *prior_parent_snaps; /* snaps inherited from any parents we */
672 u32 num_prior_parent_snaps; /* had prior to parent_since */
673 u64 *snaps; /* snaps specific to this realm */
674 u32 num_snaps;
675
676 struct ceph_snap_realm *parent;
677 struct list_head children; /* list of child realms */
678 struct list_head child_item;
679
680 struct list_head empty_item; /* if i have ref==0 */
681
682 struct list_head dirty_item; /* if realm needs new context */
683
684 /* the current set of snaps for this realm */
685 struct ceph_snap_context *cached_context;
686
687 struct list_head inodes_with_caps;
688 spinlock_t inodes_with_caps_lock;
689};
690
691static inline int default_congestion_kb(void)
692{
693 int congestion_kb;
694
695 /*
696 * Copied from NFS
697 *
698 * congestion size, scale with available memory.
699 *
700 * 64MB: 8192k
701 * 128MB: 11585k
702 * 256MB: 16384k
703 * 512MB: 23170k
704 * 1GB: 32768k
705 * 2GB: 46340k
706 * 4GB: 65536k
707 * 8GB: 92681k
708 * 16GB: 131072k
709 *
710 * This allows larger machines to have larger/more transfers.
711 * Limit the default to 256M
712 */
713 congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
714 if (congestion_kb > 256*1024)
715 congestion_kb = 256*1024;
716
717 return congestion_kb;
718}
719
720
721
722/* snap.c */
723struct ceph_snap_realm *ceph_lookup_snap_realm(struct ceph_mds_client *mdsc,
724 u64 ino);
725extern void ceph_get_snap_realm(struct ceph_mds_client *mdsc,
726 struct ceph_snap_realm *realm);
727extern void ceph_put_snap_realm(struct ceph_mds_client *mdsc,
728 struct ceph_snap_realm *realm);
729extern int ceph_update_snap_trace(struct ceph_mds_client *m,
730 void *p, void *e, bool deletion,
731 struct ceph_snap_realm **realm_ret);
732extern void ceph_handle_snap(struct ceph_mds_client *mdsc,
733 struct ceph_mds_session *session,
734 struct ceph_msg *msg);
735extern void ceph_queue_cap_snap(struct ceph_inode_info *ci);
736extern int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
737 struct ceph_cap_snap *capsnap);
738extern void ceph_cleanup_empty_realms(struct ceph_mds_client *mdsc);
739
740/*
741 * a cap_snap is "pending" if it is still awaiting an in-progress
742 * sync write (that may/may not still update size, mtime, etc.).
743 */
744static inline bool __ceph_have_pending_cap_snap(struct ceph_inode_info *ci)
745{
746 return !list_empty(&ci->i_cap_snaps) &&
747 list_last_entry(&ci->i_cap_snaps, struct ceph_cap_snap,
748 ci_item)->writing;
749}
750
751/* inode.c */
752extern const struct inode_operations ceph_file_iops;
753
754extern struct inode *ceph_alloc_inode(struct super_block *sb);
755extern void ceph_destroy_inode(struct inode *inode);
756extern int ceph_drop_inode(struct inode *inode);
757
758extern struct inode *ceph_get_inode(struct super_block *sb,
759 struct ceph_vino vino);
760extern struct inode *ceph_get_snapdir(struct inode *parent);
761extern int ceph_fill_file_size(struct inode *inode, int issued,
762 u32 truncate_seq, u64 truncate_size, u64 size);
763extern void ceph_fill_file_time(struct inode *inode, int issued,
764 u64 time_warp_seq, struct timespec *ctime,
765 struct timespec *mtime, struct timespec *atime);
766extern int ceph_fill_trace(struct super_block *sb,
767 struct ceph_mds_request *req,
768 struct ceph_mds_session *session);
769extern int ceph_readdir_prepopulate(struct ceph_mds_request *req,
770 struct ceph_mds_session *session);
771
772extern int ceph_inode_holds_cap(struct inode *inode, int mask);
773
774extern int ceph_inode_set_size(struct inode *inode, loff_t size);
775extern void __ceph_do_pending_vmtruncate(struct inode *inode);
776extern void ceph_queue_vmtruncate(struct inode *inode);
777
778extern void ceph_queue_invalidate(struct inode *inode);
779extern void ceph_queue_writeback(struct inode *inode);
780
781extern int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
782 int mask, bool force);
783static inline int ceph_do_getattr(struct inode *inode, int mask, bool force)
784{
785 return __ceph_do_getattr(inode, NULL, mask, force);
786}
787extern int ceph_permission(struct inode *inode, int mask);
788extern int ceph_setattr(struct dentry *dentry, struct iattr *attr);
789extern int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
790 struct kstat *stat);
791
792/* xattr.c */
793extern int ceph_setxattr(struct dentry *, const char *, const void *,
794 size_t, int);
795int __ceph_setxattr(struct dentry *, const char *, const void *, size_t, int);
796ssize_t __ceph_getxattr(struct inode *, const char *, void *, size_t);
797int __ceph_removexattr(struct dentry *, const char *);
798extern ssize_t ceph_getxattr(struct dentry *, const char *, void *, size_t);
799extern ssize_t ceph_listxattr(struct dentry *, char *, size_t);
800extern int ceph_removexattr(struct dentry *, const char *);
801extern void __ceph_build_xattrs_blob(struct ceph_inode_info *ci);
802extern void __ceph_destroy_xattrs(struct ceph_inode_info *ci);
803extern void __init ceph_xattr_init(void);
804extern void ceph_xattr_exit(void);
805extern const struct xattr_handler *ceph_xattr_handlers[];
806
807#ifdef CONFIG_SECURITY
808extern bool ceph_security_xattr_deadlock(struct inode *in);
809extern bool ceph_security_xattr_wanted(struct inode *in);
810#else
811static inline bool ceph_security_xattr_deadlock(struct inode *in)
812{
813 return false;
814}
815static inline bool ceph_security_xattr_wanted(struct inode *in)
816{
817 return false;
818}
819#endif
820
821/* acl.c */
822struct ceph_acls_info {
823 void *default_acl;
824 void *acl;
825 struct ceph_pagelist *pagelist;
826};
827
828#ifdef CONFIG_CEPH_FS_POSIX_ACL
829
830struct posix_acl *ceph_get_acl(struct inode *, int);
831int ceph_set_acl(struct inode *inode, struct posix_acl *acl, int type);
832int ceph_pre_init_acls(struct inode *dir, umode_t *mode,
833 struct ceph_acls_info *info);
834void ceph_init_inode_acls(struct inode *inode, struct ceph_acls_info *info);
835void ceph_release_acls_info(struct ceph_acls_info *info);
836
837static inline void ceph_forget_all_cached_acls(struct inode *inode)
838{
839 forget_all_cached_acls(inode);
840}
841
842#else
843
844#define ceph_get_acl NULL
845#define ceph_set_acl NULL
846
847static inline int ceph_pre_init_acls(struct inode *dir, umode_t *mode,
848 struct ceph_acls_info *info)
849{
850 return 0;
851}
852static inline void ceph_init_inode_acls(struct inode *inode,
853 struct ceph_acls_info *info)
854{
855}
856static inline void ceph_release_acls_info(struct ceph_acls_info *info)
857{
858}
859static inline int ceph_acl_chmod(struct dentry *dentry, struct inode *inode)
860{
861 return 0;
862}
863
864static inline void ceph_forget_all_cached_acls(struct inode *inode)
865{
866}
867
868#endif
869
870/* caps.c */
871extern const char *ceph_cap_string(int c);
872extern void ceph_handle_caps(struct ceph_mds_session *session,
873 struct ceph_msg *msg);
874extern struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
875 struct ceph_cap_reservation *ctx);
876extern void ceph_add_cap(struct inode *inode,
877 struct ceph_mds_session *session, u64 cap_id,
878 int fmode, unsigned issued, unsigned wanted,
879 unsigned cap, unsigned seq, u64 realmino, int flags,
880 struct ceph_cap **new_cap);
881extern void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release);
882extern void ceph_put_cap(struct ceph_mds_client *mdsc,
883 struct ceph_cap *cap);
884extern int ceph_is_any_caps(struct inode *inode);
885
886extern void ceph_queue_caps_release(struct inode *inode);
887extern int ceph_write_inode(struct inode *inode, struct writeback_control *wbc);
888extern int ceph_fsync(struct file *file, loff_t start, loff_t end,
889 int datasync);
890extern void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc,
891 struct ceph_mds_session *session);
892extern void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
893 struct ceph_mds_session *session);
894extern struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci,
895 int mds);
896extern int ceph_get_cap_mds(struct inode *inode);
897extern void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps);
898extern void ceph_put_cap_refs(struct ceph_inode_info *ci, int had);
899extern void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
900 struct ceph_snap_context *snapc);
901extern void __ceph_flush_snaps(struct ceph_inode_info *ci,
902 struct ceph_mds_session **psession,
903 int again);
904extern void ceph_check_caps(struct ceph_inode_info *ci, int flags,
905 struct ceph_mds_session *session);
906extern void ceph_check_delayed_caps(struct ceph_mds_client *mdsc);
907extern void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc);
908
909extern int ceph_encode_inode_release(void **p, struct inode *inode,
910 int mds, int drop, int unless, int force);
911extern int ceph_encode_dentry_release(void **p, struct dentry *dn,
912 int mds, int drop, int unless);
913
914extern int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
915 loff_t endoff, int *got, struct page **pinned_page);
916
917/* for counting open files by mode */
918static inline void __ceph_get_fmode(struct ceph_inode_info *ci, int mode)
919{
920 ci->i_nr_by_mode[mode]++;
921}
922extern void ceph_put_fmode(struct ceph_inode_info *ci, int mode);
923
924/* addr.c */
925extern const struct address_space_operations ceph_aops;
926extern int ceph_mmap(struct file *file, struct vm_area_struct *vma);
927extern int ceph_uninline_data(struct file *filp, struct page *locked_page);
928extern int ceph_pool_perm_check(struct ceph_inode_info *ci, int need);
929extern void ceph_pool_perm_destroy(struct ceph_mds_client* mdsc);
930
931/* file.c */
932extern const struct file_operations ceph_file_fops;
933
934extern int ceph_open(struct inode *inode, struct file *file);
935extern int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
936 struct file *file, unsigned flags, umode_t mode,
937 int *opened);
938extern int ceph_release(struct inode *inode, struct file *filp);
939extern void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
940 char *data, size_t len);
941/* dir.c */
942extern const struct file_operations ceph_dir_fops;
943extern const struct file_operations ceph_snapdir_fops;
944extern const struct inode_operations ceph_dir_iops;
945extern const struct inode_operations ceph_snapdir_iops;
946extern const struct dentry_operations ceph_dentry_ops, ceph_snap_dentry_ops,
947 ceph_snapdir_dentry_ops;
948
949extern int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry);
950extern int ceph_handle_snapdir(struct ceph_mds_request *req,
951 struct dentry *dentry, int err);
952extern struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
953 struct dentry *dentry, int err);
954
955extern void ceph_dentry_lru_add(struct dentry *dn);
956extern void ceph_dentry_lru_touch(struct dentry *dn);
957extern void ceph_dentry_lru_del(struct dentry *dn);
958extern void ceph_invalidate_dentry_lease(struct dentry *dentry);
959extern unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn);
960extern void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl);
961
962/*
963 * our d_ops vary depending on whether the inode is live,
964 * snapshotted (read-only), or a virtual ".snap" directory.
965 */
966int ceph_init_dentry(struct dentry *dentry);
967
968
969/* ioctl.c */
970extern long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
971
972/* export.c */
973extern const struct export_operations ceph_export_ops;
974
975/* locks.c */
976extern __init void ceph_flock_init(void);
977extern int ceph_lock(struct file *file, int cmd, struct file_lock *fl);
978extern int ceph_flock(struct file *file, int cmd, struct file_lock *fl);
979extern void ceph_count_locks(struct inode *inode, int *p_num, int *f_num);
980extern int ceph_encode_locks_to_buffer(struct inode *inode,
981 struct ceph_filelock *flocks,
982 int num_fcntl_locks,
983 int num_flock_locks);
984extern int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
985 struct ceph_pagelist *pagelist,
986 int num_fcntl_locks, int num_flock_locks);
987extern int lock_to_ceph_filelock(struct file_lock *fl, struct ceph_filelock *c);
988
989/* debugfs.c */
990extern int ceph_fs_debugfs_init(struct ceph_fs_client *client);
991extern void ceph_fs_debugfs_cleanup(struct ceph_fs_client *client);
992
993#endif /* _FS_CEPH_SUPER_H */