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