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#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 */
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _FS_CEPH_SUPER_H
3#define _FS_CEPH_SUPER_H
4
5#include <linux/ceph/ceph_debug.h>
6#include <linux/ceph/osd_client.h>
7
8#include <asm/unaligned.h>
9#include <linux/backing-dev.h>
10#include <linux/completion.h>
11#include <linux/exportfs.h>
12#include <linux/fs.h>
13#include <linux/mempool.h>
14#include <linux/pagemap.h>
15#include <linux/wait.h>
16#include <linux/writeback.h>
17#include <linux/slab.h>
18#include <linux/posix_acl.h>
19#include <linux/refcount.h>
20#include <linux/security.h>
21#include <linux/netfs.h>
22#include <linux/fscache.h>
23#include <linux/hashtable.h>
24
25#include <linux/ceph/libceph.h>
26#include "crypto.h"
27
28/* large granularity for statfs utilization stats to facilitate
29 * large volume sizes on 32-bit machines. */
30#define CEPH_BLOCK_SHIFT 22 /* 4 MB */
31#define CEPH_BLOCK (1 << CEPH_BLOCK_SHIFT)
32#define CEPH_4K_BLOCK_SHIFT 12 /* 4 KB */
33
34#define CEPH_MOUNT_OPT_CLEANRECOVER (1<<1) /* auto reonnect (clean mode) after blocklisted */
35#define CEPH_MOUNT_OPT_DIRSTAT (1<<4) /* `cat dirname` for stats */
36#define CEPH_MOUNT_OPT_RBYTES (1<<5) /* dir st_bytes = rbytes */
37#define CEPH_MOUNT_OPT_NOASYNCREADDIR (1<<7) /* no dcache readdir */
38#define CEPH_MOUNT_OPT_INO32 (1<<8) /* 32 bit inos */
39#define CEPH_MOUNT_OPT_DCACHE (1<<9) /* use dcache for readdir etc */
40#define CEPH_MOUNT_OPT_FSCACHE (1<<10) /* use fscache */
41#define CEPH_MOUNT_OPT_NOPOOLPERM (1<<11) /* no pool permission check */
42#define CEPH_MOUNT_OPT_MOUNTWAIT (1<<12) /* mount waits if no mds is up */
43#define CEPH_MOUNT_OPT_NOQUOTADF (1<<13) /* no root dir quota in statfs */
44#define CEPH_MOUNT_OPT_NOCOPYFROM (1<<14) /* don't use RADOS 'copy-from' op */
45#define CEPH_MOUNT_OPT_ASYNC_DIROPS (1<<15) /* allow async directory ops */
46#define CEPH_MOUNT_OPT_NOPAGECACHE (1<<16) /* bypass pagecache altogether */
47#define CEPH_MOUNT_OPT_SPARSEREAD (1<<17) /* always do sparse reads */
48
49#define CEPH_MOUNT_OPT_DEFAULT \
50 (CEPH_MOUNT_OPT_DCACHE | \
51 CEPH_MOUNT_OPT_NOCOPYFROM | \
52 CEPH_MOUNT_OPT_ASYNC_DIROPS)
53
54#define ceph_set_mount_opt(fsc, opt) \
55 (fsc)->mount_options->flags |= CEPH_MOUNT_OPT_##opt
56#define ceph_clear_mount_opt(fsc, opt) \
57 (fsc)->mount_options->flags &= ~CEPH_MOUNT_OPT_##opt
58#define ceph_test_mount_opt(fsc, opt) \
59 (!!((fsc)->mount_options->flags & CEPH_MOUNT_OPT_##opt))
60
61/* max size of osd read request, limited by libceph */
62#define CEPH_MAX_READ_SIZE CEPH_MSG_MAX_DATA_LEN
63/* osd has a configurable limitaion of max write size.
64 * CEPH_MSG_MAX_DATA_LEN should be small enough. */
65#define CEPH_MAX_WRITE_SIZE CEPH_MSG_MAX_DATA_LEN
66#define CEPH_RASIZE_DEFAULT (8192*1024) /* max readahead */
67#define CEPH_MAX_READDIR_DEFAULT 1024
68#define CEPH_MAX_READDIR_BYTES_DEFAULT (512*1024)
69#define CEPH_SNAPDIRNAME_DEFAULT ".snap"
70
71/*
72 * Delay telling the MDS we no longer want caps, in case we reopen
73 * the file. Delay a minimum amount of time, even if we send a cap
74 * message for some other reason. Otherwise, take the oppotunity to
75 * update the mds to avoid sending another message later.
76 */
77#define CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT 5 /* cap release delay */
78#define CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT 60 /* cap release delay */
79
80struct ceph_mount_options {
81 unsigned int flags;
82
83 unsigned int wsize; /* max write size */
84 unsigned int rsize; /* max read size */
85 unsigned int rasize; /* max readahead */
86 unsigned int congestion_kb; /* max writeback in flight */
87 unsigned int caps_wanted_delay_min, caps_wanted_delay_max;
88 int caps_max;
89 unsigned int max_readdir; /* max readdir result (entries) */
90 unsigned int max_readdir_bytes; /* max readdir result (bytes) */
91
92 bool new_dev_syntax;
93
94 /*
95 * everything above this point can be memcmp'd; everything below
96 * is handled in compare_mount_options()
97 */
98
99 char *snapdir_name; /* default ".snap" */
100 char *mds_namespace; /* default NULL */
101 char *server_path; /* default NULL (means "/") */
102 char *fscache_uniq; /* default NULL */
103 char *mon_addr;
104 struct fscrypt_dummy_policy dummy_enc_policy;
105};
106
107/* mount state */
108enum {
109 CEPH_MOUNT_MOUNTING,
110 CEPH_MOUNT_MOUNTED,
111 CEPH_MOUNT_UNMOUNTING,
112 CEPH_MOUNT_UNMOUNTED,
113 CEPH_MOUNT_SHUTDOWN,
114 CEPH_MOUNT_RECOVER,
115 CEPH_MOUNT_FENCE_IO,
116};
117
118#define CEPH_ASYNC_CREATE_CONFLICT_BITS 8
119
120struct ceph_fs_client {
121 struct super_block *sb;
122
123 struct list_head metric_wakeup;
124
125 struct ceph_mount_options *mount_options;
126 struct ceph_client *client;
127
128 int mount_state;
129
130 bool blocklisted;
131
132 bool have_copy_from2;
133
134 u32 filp_gen;
135 loff_t max_file_size;
136
137 struct ceph_mds_client *mdsc;
138
139 atomic_long_t writeback_count;
140 bool write_congested;
141
142 struct workqueue_struct *inode_wq;
143 struct workqueue_struct *cap_wq;
144
145 DECLARE_HASHTABLE(async_unlink_conflict, CEPH_ASYNC_CREATE_CONFLICT_BITS);
146 spinlock_t async_unlink_conflict_lock;
147
148#ifdef CONFIG_DEBUG_FS
149 struct dentry *debugfs_dentry_lru, *debugfs_caps;
150 struct dentry *debugfs_congestion_kb;
151 struct dentry *debugfs_bdi;
152 struct dentry *debugfs_mdsc, *debugfs_mdsmap;
153 struct dentry *debugfs_status;
154 struct dentry *debugfs_mds_sessions;
155 struct dentry *debugfs_metrics_dir;
156#endif
157
158#ifdef CONFIG_CEPH_FSCACHE
159 struct fscache_volume *fscache;
160#endif
161#ifdef CONFIG_FS_ENCRYPTION
162 struct fscrypt_dummy_policy fsc_dummy_enc_policy;
163#endif
164};
165
166/*
167 * File i/o capability. This tracks shared state with the metadata
168 * server that allows us to cache or writeback attributes or to read
169 * and write data. For any given inode, we should have one or more
170 * capabilities, one issued by each metadata server, and our
171 * cumulative access is the OR of all issued capabilities.
172 *
173 * Each cap is referenced by the inode's i_caps rbtree and by per-mds
174 * session capability lists.
175 */
176struct ceph_cap {
177 struct ceph_inode_info *ci;
178 struct rb_node ci_node; /* per-ci cap tree */
179 struct ceph_mds_session *session;
180 struct list_head session_caps; /* per-session caplist */
181 u64 cap_id; /* unique cap id (mds provided) */
182 union {
183 /* in-use caps */
184 struct {
185 int issued; /* latest, from the mds */
186 int implemented; /* implemented superset of
187 issued (for revocation) */
188 int mds; /* mds index for this cap */
189 int mds_wanted; /* caps wanted from this mds */
190 };
191 /* caps to release */
192 struct {
193 u64 cap_ino;
194 int queue_release;
195 };
196 };
197 u32 seq, issue_seq, mseq;
198 u32 cap_gen; /* active/stale cycle */
199 unsigned long last_used;
200 struct list_head caps_item;
201};
202
203#define CHECK_CAPS_AUTHONLY 1 /* only check auth cap */
204#define CHECK_CAPS_FLUSH 2 /* flush any dirty caps */
205#define CHECK_CAPS_NOINVAL 4 /* don't invalidate pagecache */
206
207struct ceph_cap_flush {
208 u64 tid;
209 int caps;
210 bool wake; /* wake up flush waiters when finish ? */
211 bool is_capsnap; /* true means capsnap */
212 struct list_head g_list; // global
213 struct list_head i_list; // per inode
214};
215
216/*
217 * Snapped cap state that is pending flush to mds. When a snapshot occurs,
218 * we first complete any in-process sync writes and writeback any dirty
219 * data before flushing the snapped state (tracked here) back to the MDS.
220 */
221struct ceph_cap_snap {
222 refcount_t nref;
223 struct list_head ci_item;
224
225 struct ceph_cap_flush cap_flush;
226
227 u64 follows;
228 int issued, dirty;
229 struct ceph_snap_context *context;
230
231 umode_t mode;
232 kuid_t uid;
233 kgid_t gid;
234
235 struct ceph_buffer *xattr_blob;
236 u64 xattr_version;
237
238 u64 size;
239 u64 change_attr;
240 struct timespec64 mtime, atime, ctime, btime;
241 u64 time_warp_seq;
242 u64 truncate_size;
243 u32 truncate_seq;
244 int writing; /* a sync write is still in progress */
245 int dirty_pages; /* dirty pages awaiting writeback */
246 bool inline_data;
247 bool need_flush;
248};
249
250static inline void ceph_put_cap_snap(struct ceph_cap_snap *capsnap)
251{
252 if (refcount_dec_and_test(&capsnap->nref)) {
253 if (capsnap->xattr_blob)
254 ceph_buffer_put(capsnap->xattr_blob);
255 kmem_cache_free(ceph_cap_snap_cachep, capsnap);
256 }
257}
258
259/*
260 * The frag tree describes how a directory is fragmented, potentially across
261 * multiple metadata servers. It is also used to indicate points where
262 * metadata authority is delegated, and whether/where metadata is replicated.
263 *
264 * A _leaf_ frag will be present in the i_fragtree IFF there is
265 * delegation info. That is, if mds >= 0 || ndist > 0.
266 */
267#define CEPH_MAX_DIRFRAG_REP 4
268
269struct ceph_inode_frag {
270 struct rb_node node;
271
272 /* fragtree state */
273 u32 frag;
274 int split_by; /* i.e. 2^(split_by) children */
275
276 /* delegation and replication info */
277 int mds; /* -1 if same authority as parent */
278 int ndist; /* >0 if replicated */
279 int dist[CEPH_MAX_DIRFRAG_REP];
280};
281
282/*
283 * We cache inode xattrs as an encoded blob until they are first used,
284 * at which point we parse them into an rbtree.
285 */
286struct ceph_inode_xattr {
287 struct rb_node node;
288
289 const char *name;
290 int name_len;
291 const char *val;
292 int val_len;
293 int dirty;
294
295 int should_free_name;
296 int should_free_val;
297};
298
299/*
300 * Ceph dentry state
301 */
302struct ceph_dentry_info {
303 struct dentry *dentry;
304 struct ceph_mds_session *lease_session;
305 struct list_head lease_list;
306 struct hlist_node hnode;
307 unsigned long flags;
308 int lease_shared_gen;
309 u32 lease_gen;
310 u32 lease_seq;
311 unsigned long lease_renew_after, lease_renew_from;
312 unsigned long time;
313 u64 offset;
314};
315
316#define CEPH_DENTRY_REFERENCED (1 << 0)
317#define CEPH_DENTRY_LEASE_LIST (1 << 1)
318#define CEPH_DENTRY_SHRINK_LIST (1 << 2)
319#define CEPH_DENTRY_PRIMARY_LINK (1 << 3)
320#define CEPH_DENTRY_ASYNC_UNLINK_BIT (4)
321#define CEPH_DENTRY_ASYNC_UNLINK (1 << CEPH_DENTRY_ASYNC_UNLINK_BIT)
322#define CEPH_DENTRY_ASYNC_CREATE_BIT (5)
323#define CEPH_DENTRY_ASYNC_CREATE (1 << CEPH_DENTRY_ASYNC_CREATE_BIT)
324
325struct ceph_inode_xattrs_info {
326 /*
327 * (still encoded) xattr blob. we avoid the overhead of parsing
328 * this until someone actually calls getxattr, etc.
329 *
330 * blob->vec.iov_len == 4 implies there are no xattrs; blob ==
331 * NULL means we don't know.
332 */
333 struct ceph_buffer *blob, *prealloc_blob;
334
335 struct rb_root index;
336 bool dirty;
337 int count;
338 int names_size;
339 int vals_size;
340 u64 version, index_version;
341};
342
343/*
344 * Ceph inode.
345 */
346struct ceph_inode_info {
347 struct netfs_inode netfs; /* Netfslib context and vfs inode */
348 struct ceph_vino i_vino; /* ceph ino + snap */
349
350 spinlock_t i_ceph_lock;
351
352 u64 i_version;
353 u64 i_inline_version;
354 u32 i_time_warp_seq;
355
356 unsigned long i_ceph_flags;
357 atomic64_t i_release_count;
358 atomic64_t i_ordered_count;
359 atomic64_t i_complete_seq[2];
360
361 struct ceph_dir_layout i_dir_layout;
362 struct ceph_file_layout i_layout;
363 struct ceph_file_layout i_cached_layout; // for async creates
364 char *i_symlink;
365
366 /* for dirs */
367 struct timespec64 i_rctime;
368 u64 i_rbytes, i_rfiles, i_rsubdirs, i_rsnaps;
369 u64 i_files, i_subdirs;
370
371 /* quotas */
372 u64 i_max_bytes, i_max_files;
373
374 s32 i_dir_pin;
375
376 struct rb_root i_fragtree;
377 int i_fragtree_nsplits;
378 struct mutex i_fragtree_mutex;
379
380 struct ceph_inode_xattrs_info i_xattrs;
381
382 /* capabilities. protected _both_ by i_ceph_lock and cap->session's
383 * s_mutex. */
384 struct rb_root i_caps; /* cap list */
385 struct ceph_cap *i_auth_cap; /* authoritative cap, if any */
386 unsigned i_dirty_caps, i_flushing_caps; /* mask of dirtied fields */
387
388 /*
389 * Link to the auth cap's session's s_cap_dirty list. s_cap_dirty
390 * is protected by the mdsc->cap_dirty_lock, but each individual item
391 * is also protected by the inode's i_ceph_lock. Walking s_cap_dirty
392 * requires the mdsc->cap_dirty_lock. List presence for an item can
393 * be tested under the i_ceph_lock. Changing anything requires both.
394 */
395 struct list_head i_dirty_item;
396
397 /*
398 * Link to session's s_cap_flushing list. Protected in a similar
399 * fashion to i_dirty_item, but also by the s_mutex for changes. The
400 * s_cap_flushing list can be walked while holding either the s_mutex
401 * or msdc->cap_dirty_lock. List presence can also be checked while
402 * holding the i_ceph_lock for this inode.
403 */
404 struct list_head i_flushing_item;
405
406 /* we need to track cap writeback on a per-cap-bit basis, to allow
407 * overlapping, pipelined cap flushes to the mds. we can probably
408 * reduce the tid to 8 bits if we're concerned about inode size. */
409 struct ceph_cap_flush *i_prealloc_cap_flush;
410 struct list_head i_cap_flush_list;
411 wait_queue_head_t i_cap_wq; /* threads waiting on a capability */
412 unsigned long i_hold_caps_max; /* jiffies */
413 struct list_head i_cap_delay_list; /* for delayed cap release to mds */
414 struct ceph_cap_reservation i_cap_migration_resv;
415 struct list_head i_cap_snaps; /* snapped state pending flush to mds */
416 struct ceph_snap_context *i_head_snapc; /* set if wr_buffer_head > 0 or
417 dirty|flushing caps */
418 unsigned i_snap_caps; /* cap bits for snapped files */
419
420 unsigned long i_last_rd;
421 unsigned long i_last_wr;
422 int i_nr_by_mode[CEPH_FILE_MODE_BITS]; /* open file counts */
423
424 struct mutex i_truncate_mutex;
425 u32 i_truncate_seq; /* last truncate to smaller size */
426 u64 i_truncate_size; /* and the size we last truncated down to */
427 int i_truncate_pending; /* still need to call vmtruncate */
428 /*
429 * For none fscrypt case it equals to i_truncate_size or it will
430 * equals to fscrypt_file_size
431 */
432 u64 i_truncate_pagecache_size;
433
434 u64 i_max_size; /* max file size authorized by mds */
435 u64 i_reported_size; /* (max_)size reported to or requested of mds */
436 u64 i_wanted_max_size; /* offset we'd like to write too */
437 u64 i_requested_max_size; /* max_size we've requested */
438
439 /* held references to caps */
440 int i_pin_ref;
441 int i_rd_ref, i_rdcache_ref, i_wr_ref, i_wb_ref, i_fx_ref;
442 int i_wrbuffer_ref, i_wrbuffer_ref_head;
443 atomic_t i_filelock_ref;
444 atomic_t i_shared_gen; /* increment each time we get FILE_SHARED */
445 u32 i_rdcache_gen; /* incremented each time we get FILE_CACHE. */
446 u32 i_rdcache_revoking; /* RDCACHE gen to async invalidate, if any */
447
448 struct list_head i_unsafe_dirops; /* uncommitted mds dir ops */
449 struct list_head i_unsafe_iops; /* uncommitted mds inode ops */
450 spinlock_t i_unsafe_lock;
451
452 union {
453 struct ceph_snap_realm *i_snap_realm; /* snap realm (if caps) */
454 struct ceph_snapid_map *i_snapid_map; /* snapid -> dev_t */
455 };
456 struct list_head i_snap_realm_item;
457 struct list_head i_snap_flush_item;
458 struct timespec64 i_btime;
459 struct timespec64 i_snap_btime;
460
461 struct work_struct i_work;
462 unsigned long i_work_mask;
463
464#ifdef CONFIG_FS_ENCRYPTION
465 u32 fscrypt_auth_len;
466 u32 fscrypt_file_len;
467 u8 *fscrypt_auth;
468 u8 *fscrypt_file;
469#endif
470};
471
472struct ceph_netfs_request_data {
473 int caps;
474
475 /*
476 * Maximum size of a file readahead request.
477 * The fadvise could update the bdi's default ra_pages.
478 */
479 unsigned int file_ra_pages;
480
481 /* Set it if fadvise disables file readahead entirely */
482 bool file_ra_disabled;
483};
484
485static inline struct ceph_inode_info *
486ceph_inode(const struct inode *inode)
487{
488 return container_of(inode, struct ceph_inode_info, netfs.inode);
489}
490
491static inline struct ceph_fs_client *
492ceph_inode_to_fs_client(const struct inode *inode)
493{
494 return (struct ceph_fs_client *)inode->i_sb->s_fs_info;
495}
496
497static inline struct ceph_fs_client *
498ceph_sb_to_fs_client(const struct super_block *sb)
499{
500 return (struct ceph_fs_client *)sb->s_fs_info;
501}
502
503static inline struct ceph_mds_client *
504ceph_sb_to_mdsc(const struct super_block *sb)
505{
506 return (struct ceph_mds_client *)ceph_sb_to_fs_client(sb)->mdsc;
507}
508
509static inline struct ceph_client *
510ceph_inode_to_client(const struct inode *inode)
511{
512 return (struct ceph_client *)ceph_inode_to_fs_client(inode)->client;
513}
514
515static inline struct ceph_vino
516ceph_vino(const struct inode *inode)
517{
518 return ceph_inode(inode)->i_vino;
519}
520
521static inline u32 ceph_ino_to_ino32(u64 vino)
522{
523 u32 ino = vino & 0xffffffff;
524 ino ^= vino >> 32;
525 if (!ino)
526 ino = 2;
527 return ino;
528}
529
530/*
531 * Inode numbers in cephfs are 64 bits, but inode->i_ino is 32-bits on
532 * some arches. We generally do not use this value inside the ceph driver, but
533 * we do want to set it to something, so that generic vfs code has an
534 * appropriate value for tracepoints and the like.
535 */
536static inline ino_t ceph_vino_to_ino_t(struct ceph_vino vino)
537{
538 if (sizeof(ino_t) == sizeof(u32))
539 return ceph_ino_to_ino32(vino.ino);
540 return (ino_t)vino.ino;
541}
542
543/* for printf-style formatting */
544#define ceph_vinop(i) ceph_inode(i)->i_vino.ino, ceph_inode(i)->i_vino.snap
545
546static inline u64 ceph_ino(struct inode *inode)
547{
548 return ceph_inode(inode)->i_vino.ino;
549}
550
551static inline u64 ceph_snap(struct inode *inode)
552{
553 return ceph_inode(inode)->i_vino.snap;
554}
555
556/**
557 * ceph_present_ino - format an inode number for presentation to userland
558 * @sb: superblock where the inode lives
559 * @ino: inode number to (possibly) convert
560 *
561 * If the user mounted with the ino32 option, then the 64-bit value needs
562 * to be converted to something that can fit inside 32 bits. Note that
563 * internal kernel code never uses this value, so this is entirely for
564 * userland consumption.
565 */
566static inline u64 ceph_present_ino(struct super_block *sb, u64 ino)
567{
568 if (unlikely(ceph_test_mount_opt(ceph_sb_to_fs_client(sb), INO32)))
569 return ceph_ino_to_ino32(ino);
570 return ino;
571}
572
573static inline u64 ceph_present_inode(struct inode *inode)
574{
575 return ceph_present_ino(inode->i_sb, ceph_ino(inode));
576}
577
578static inline int ceph_ino_compare(struct inode *inode, void *data)
579{
580 struct ceph_vino *pvino = (struct ceph_vino *)data;
581 struct ceph_inode_info *ci = ceph_inode(inode);
582 return ci->i_vino.ino == pvino->ino &&
583 ci->i_vino.snap == pvino->snap;
584}
585
586/*
587 * The MDS reserves a set of inodes for its own usage. These should never
588 * be accessible by clients, and so the MDS has no reason to ever hand these
589 * out. The range is CEPH_MDS_INO_MDSDIR_OFFSET..CEPH_INO_SYSTEM_BASE.
590 *
591 * These come from src/mds/mdstypes.h in the ceph sources.
592 */
593#define CEPH_MAX_MDS 0x100
594#define CEPH_NUM_STRAY 10
595#define CEPH_MDS_INO_MDSDIR_OFFSET (1 * CEPH_MAX_MDS)
596#define CEPH_MDS_INO_LOG_OFFSET (2 * CEPH_MAX_MDS)
597#define CEPH_INO_SYSTEM_BASE ((6*CEPH_MAX_MDS) + (CEPH_MAX_MDS * CEPH_NUM_STRAY))
598
599static inline bool ceph_vino_is_reserved(const struct ceph_vino vino)
600{
601 if (vino.ino >= CEPH_INO_SYSTEM_BASE ||
602 vino.ino < CEPH_MDS_INO_MDSDIR_OFFSET)
603 return false;
604
605 /* Don't warn on mdsdirs */
606 WARN_RATELIMIT(vino.ino >= CEPH_MDS_INO_LOG_OFFSET,
607 "Attempt to access reserved inode number 0x%llx",
608 vino.ino);
609 return true;
610}
611
612static inline struct inode *ceph_find_inode(struct super_block *sb,
613 struct ceph_vino vino)
614{
615 if (ceph_vino_is_reserved(vino))
616 return NULL;
617
618 /*
619 * NB: The hashval will be run through the fs/inode.c hash function
620 * anyway, so there is no need to squash the inode number down to
621 * 32-bits first. Just use low-order bits on arches with 32-bit long.
622 */
623 return ilookup5(sb, (unsigned long)vino.ino, ceph_ino_compare, &vino);
624}
625
626
627/*
628 * Ceph inode.
629 */
630#define CEPH_I_DIR_ORDERED (1 << 0) /* dentries in dir are ordered */
631#define CEPH_I_FLUSH (1 << 2) /* do not delay flush of dirty metadata */
632#define CEPH_I_POOL_PERM (1 << 3) /* pool rd/wr bits are valid */
633#define CEPH_I_POOL_RD (1 << 4) /* can read from pool */
634#define CEPH_I_POOL_WR (1 << 5) /* can write to pool */
635#define CEPH_I_SEC_INITED (1 << 6) /* security initialized */
636#define CEPH_I_KICK_FLUSH (1 << 7) /* kick flushing caps */
637#define CEPH_I_FLUSH_SNAPS (1 << 8) /* need flush snapss */
638#define CEPH_I_ERROR_WRITE (1 << 9) /* have seen write errors */
639#define CEPH_I_ERROR_FILELOCK (1 << 10) /* have seen file lock errors */
640#define CEPH_I_ODIRECT (1 << 11) /* inode in direct I/O mode */
641#define CEPH_ASYNC_CREATE_BIT (12) /* async create in flight for this */
642#define CEPH_I_ASYNC_CREATE (1 << CEPH_ASYNC_CREATE_BIT)
643#define CEPH_I_SHUTDOWN (1 << 13) /* inode is no longer usable */
644#define CEPH_I_ASYNC_CHECK_CAPS (1 << 14) /* check caps immediately after async
645 creating finishes */
646
647/*
648 * Masks of ceph inode work.
649 */
650#define CEPH_I_WORK_WRITEBACK 0
651#define CEPH_I_WORK_INVALIDATE_PAGES 1
652#define CEPH_I_WORK_VMTRUNCATE 2
653#define CEPH_I_WORK_CHECK_CAPS 3
654#define CEPH_I_WORK_FLUSH_SNAPS 4
655
656/*
657 * We set the ERROR_WRITE bit when we start seeing write errors on an inode
658 * and then clear it when they start succeeding. Note that we do a lockless
659 * check first, and only take the lock if it looks like it needs to be changed.
660 * The write submission code just takes this as a hint, so we're not too
661 * worried if a few slip through in either direction.
662 */
663static inline void ceph_set_error_write(struct ceph_inode_info *ci)
664{
665 if (!(READ_ONCE(ci->i_ceph_flags) & CEPH_I_ERROR_WRITE)) {
666 spin_lock(&ci->i_ceph_lock);
667 ci->i_ceph_flags |= CEPH_I_ERROR_WRITE;
668 spin_unlock(&ci->i_ceph_lock);
669 }
670}
671
672static inline void ceph_clear_error_write(struct ceph_inode_info *ci)
673{
674 if (READ_ONCE(ci->i_ceph_flags) & CEPH_I_ERROR_WRITE) {
675 spin_lock(&ci->i_ceph_lock);
676 ci->i_ceph_flags &= ~CEPH_I_ERROR_WRITE;
677 spin_unlock(&ci->i_ceph_lock);
678 }
679}
680
681static inline void __ceph_dir_set_complete(struct ceph_inode_info *ci,
682 long long release_count,
683 long long ordered_count)
684{
685 /*
686 * Makes sure operations that setup readdir cache (update page
687 * cache and i_size) are strongly ordered w.r.t. the following
688 * atomic64_set() operations.
689 */
690 smp_mb();
691 atomic64_set(&ci->i_complete_seq[0], release_count);
692 atomic64_set(&ci->i_complete_seq[1], ordered_count);
693}
694
695static inline void __ceph_dir_clear_complete(struct ceph_inode_info *ci)
696{
697 atomic64_inc(&ci->i_release_count);
698}
699
700static inline void __ceph_dir_clear_ordered(struct ceph_inode_info *ci)
701{
702 atomic64_inc(&ci->i_ordered_count);
703}
704
705static inline bool __ceph_dir_is_complete(struct ceph_inode_info *ci)
706{
707 return atomic64_read(&ci->i_complete_seq[0]) ==
708 atomic64_read(&ci->i_release_count);
709}
710
711static inline bool __ceph_dir_is_complete_ordered(struct ceph_inode_info *ci)
712{
713 return atomic64_read(&ci->i_complete_seq[0]) ==
714 atomic64_read(&ci->i_release_count) &&
715 atomic64_read(&ci->i_complete_seq[1]) ==
716 atomic64_read(&ci->i_ordered_count);
717}
718
719static inline void ceph_dir_clear_complete(struct inode *inode)
720{
721 __ceph_dir_clear_complete(ceph_inode(inode));
722}
723
724static inline void ceph_dir_clear_ordered(struct inode *inode)
725{
726 __ceph_dir_clear_ordered(ceph_inode(inode));
727}
728
729static inline bool ceph_dir_is_complete_ordered(struct inode *inode)
730{
731 bool ret = __ceph_dir_is_complete_ordered(ceph_inode(inode));
732 smp_rmb();
733 return ret;
734}
735
736/* find a specific frag @f */
737extern struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci,
738 u32 f);
739
740/*
741 * choose fragment for value @v. copy frag content to pfrag, if leaf
742 * exists
743 */
744extern u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
745 struct ceph_inode_frag *pfrag,
746 int *found);
747
748static inline struct ceph_dentry_info *ceph_dentry(const struct dentry *dentry)
749{
750 return (struct ceph_dentry_info *)dentry->d_fsdata;
751}
752
753/*
754 * caps helpers
755 */
756static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)
757{
758 return !RB_EMPTY_ROOT(&ci->i_caps);
759}
760
761extern int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented);
762extern int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int t);
763extern int __ceph_caps_issued_mask_metric(struct ceph_inode_info *ci, int mask,
764 int t);
765extern int __ceph_caps_issued_other(struct ceph_inode_info *ci,
766 struct ceph_cap *cap);
767
768static inline int ceph_caps_issued(struct ceph_inode_info *ci)
769{
770 int issued;
771 spin_lock(&ci->i_ceph_lock);
772 issued = __ceph_caps_issued(ci, NULL);
773 spin_unlock(&ci->i_ceph_lock);
774 return issued;
775}
776
777static inline int ceph_caps_issued_mask_metric(struct ceph_inode_info *ci,
778 int mask, int touch)
779{
780 int r;
781 spin_lock(&ci->i_ceph_lock);
782 r = __ceph_caps_issued_mask_metric(ci, mask, touch);
783 spin_unlock(&ci->i_ceph_lock);
784 return r;
785}
786
787static inline int __ceph_caps_dirty(struct ceph_inode_info *ci)
788{
789 return ci->i_dirty_caps | ci->i_flushing_caps;
790}
791extern struct ceph_cap_flush *ceph_alloc_cap_flush(void);
792extern void ceph_free_cap_flush(struct ceph_cap_flush *cf);
793extern int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
794 struct ceph_cap_flush **pcf);
795
796extern int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
797 struct ceph_cap *ocap, int mask);
798extern int ceph_caps_revoking(struct ceph_inode_info *ci, int mask);
799extern int __ceph_caps_used(struct ceph_inode_info *ci);
800
801static inline bool __ceph_is_file_opened(struct ceph_inode_info *ci)
802{
803 return ci->i_nr_by_mode[0];
804}
805extern int __ceph_caps_file_wanted(struct ceph_inode_info *ci);
806extern int __ceph_caps_wanted(struct ceph_inode_info *ci);
807
808/* what the mds thinks we want */
809extern int __ceph_caps_mds_wanted(struct ceph_inode_info *ci, bool check);
810
811extern void ceph_caps_init(struct ceph_mds_client *mdsc);
812extern void ceph_caps_finalize(struct ceph_mds_client *mdsc);
813extern void ceph_adjust_caps_max_min(struct ceph_mds_client *mdsc,
814 struct ceph_mount_options *fsopt);
815extern int ceph_reserve_caps(struct ceph_mds_client *mdsc,
816 struct ceph_cap_reservation *ctx, int need);
817extern void ceph_unreserve_caps(struct ceph_mds_client *mdsc,
818 struct ceph_cap_reservation *ctx);
819extern void ceph_reservation_status(struct ceph_fs_client *client,
820 int *total, int *avail, int *used,
821 int *reserved, int *min);
822extern void change_auth_cap_ses(struct ceph_inode_info *ci,
823 struct ceph_mds_session *session);
824
825
826
827/*
828 * we keep buffered readdir results attached to file->private_data
829 */
830#define CEPH_F_SYNC 1
831#define CEPH_F_ATEND 2
832
833struct ceph_file_info {
834 short fmode; /* initialized on open */
835 short flags; /* CEPH_F_* */
836
837 spinlock_t rw_contexts_lock;
838 struct list_head rw_contexts;
839
840 u32 filp_gen;
841};
842
843struct ceph_dir_file_info {
844 struct ceph_file_info file_info;
845
846 /* readdir: position within the dir */
847 u32 frag;
848 struct ceph_mds_request *last_readdir;
849
850 /* readdir: position within a frag */
851 unsigned next_offset; /* offset of next chunk (last_name's + 1) */
852 char *last_name; /* last entry in previous chunk */
853 long long dir_release_count;
854 long long dir_ordered_count;
855 int readdir_cache_idx;
856
857 /* used for -o dirstat read() on directory thing */
858 char *dir_info;
859 int dir_info_len;
860};
861
862struct ceph_rw_context {
863 struct list_head list;
864 struct task_struct *thread;
865 int caps;
866};
867
868#define CEPH_DEFINE_RW_CONTEXT(_name, _caps) \
869 struct ceph_rw_context _name = { \
870 .thread = current, \
871 .caps = _caps, \
872 }
873
874static inline void ceph_add_rw_context(struct ceph_file_info *cf,
875 struct ceph_rw_context *ctx)
876{
877 spin_lock(&cf->rw_contexts_lock);
878 list_add(&ctx->list, &cf->rw_contexts);
879 spin_unlock(&cf->rw_contexts_lock);
880}
881
882static inline void ceph_del_rw_context(struct ceph_file_info *cf,
883 struct ceph_rw_context *ctx)
884{
885 spin_lock(&cf->rw_contexts_lock);
886 list_del(&ctx->list);
887 spin_unlock(&cf->rw_contexts_lock);
888}
889
890static inline struct ceph_rw_context*
891ceph_find_rw_context(struct ceph_file_info *cf)
892{
893 struct ceph_rw_context *ctx, *found = NULL;
894 spin_lock(&cf->rw_contexts_lock);
895 list_for_each_entry(ctx, &cf->rw_contexts, list) {
896 if (ctx->thread == current) {
897 found = ctx;
898 break;
899 }
900 }
901 spin_unlock(&cf->rw_contexts_lock);
902 return found;
903}
904
905struct ceph_readdir_cache_control {
906 struct page *page;
907 struct dentry **dentries;
908 int index;
909};
910
911/*
912 * A "snap realm" describes a subset of the file hierarchy sharing
913 * the same set of snapshots that apply to it. The realms themselves
914 * are organized into a hierarchy, such that children inherit (some of)
915 * the snapshots of their parents.
916 *
917 * All inodes within the realm that have capabilities are linked into a
918 * per-realm list.
919 */
920struct ceph_snap_realm {
921 u64 ino;
922 struct inode *inode;
923 atomic_t nref;
924 struct rb_node node;
925
926 u64 created, seq;
927 u64 parent_ino;
928 u64 parent_since; /* snapid when our current parent became so */
929
930 u64 *prior_parent_snaps; /* snaps inherited from any parents we */
931 u32 num_prior_parent_snaps; /* had prior to parent_since */
932 u64 *snaps; /* snaps specific to this realm */
933 u32 num_snaps;
934
935 struct ceph_snap_realm *parent;
936 struct list_head children; /* list of child realms */
937 struct list_head child_item;
938
939 struct list_head empty_item; /* if i have ref==0 */
940
941 struct list_head dirty_item; /* if realm needs new context */
942
943 struct list_head rebuild_item; /* rebuild snap realms _downward_ in hierarchy */
944
945 /* the current set of snaps for this realm */
946 struct ceph_snap_context *cached_context;
947
948 struct list_head inodes_with_caps;
949 spinlock_t inodes_with_caps_lock;
950};
951
952static inline int default_congestion_kb(void)
953{
954 int congestion_kb;
955
956 /*
957 * Copied from NFS
958 *
959 * congestion size, scale with available memory.
960 *
961 * 64MB: 8192k
962 * 128MB: 11585k
963 * 256MB: 16384k
964 * 512MB: 23170k
965 * 1GB: 32768k
966 * 2GB: 46340k
967 * 4GB: 65536k
968 * 8GB: 92681k
969 * 16GB: 131072k
970 *
971 * This allows larger machines to have larger/more transfers.
972 * Limit the default to 256M
973 */
974 congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10);
975 if (congestion_kb > 256*1024)
976 congestion_kb = 256*1024;
977
978 return congestion_kb;
979}
980
981
982/* super.c */
983extern int ceph_force_reconnect(struct super_block *sb);
984/* snap.c */
985struct ceph_snap_realm *ceph_lookup_snap_realm(struct ceph_mds_client *mdsc,
986 u64 ino);
987extern void ceph_get_snap_realm(struct ceph_mds_client *mdsc,
988 struct ceph_snap_realm *realm);
989extern void ceph_put_snap_realm(struct ceph_mds_client *mdsc,
990 struct ceph_snap_realm *realm);
991extern int ceph_update_snap_trace(struct ceph_mds_client *m,
992 void *p, void *e, bool deletion,
993 struct ceph_snap_realm **realm_ret);
994void ceph_change_snap_realm(struct inode *inode, struct ceph_snap_realm *realm);
995extern void ceph_handle_snap(struct ceph_mds_client *mdsc,
996 struct ceph_mds_session *session,
997 struct ceph_msg *msg);
998extern int __ceph_finish_cap_snap(struct ceph_inode_info *ci,
999 struct ceph_cap_snap *capsnap);
1000extern void ceph_cleanup_global_and_empty_realms(struct ceph_mds_client *mdsc);
1001
1002extern struct ceph_snapid_map *ceph_get_snapid_map(struct ceph_mds_client *mdsc,
1003 u64 snap);
1004extern void ceph_put_snapid_map(struct ceph_mds_client* mdsc,
1005 struct ceph_snapid_map *sm);
1006extern void ceph_trim_snapid_map(struct ceph_mds_client *mdsc);
1007extern void ceph_cleanup_snapid_map(struct ceph_mds_client *mdsc);
1008void ceph_umount_begin(struct super_block *sb);
1009
1010
1011/*
1012 * a cap_snap is "pending" if it is still awaiting an in-progress
1013 * sync write (that may/may not still update size, mtime, etc.).
1014 */
1015static inline bool __ceph_have_pending_cap_snap(struct ceph_inode_info *ci)
1016{
1017 return !list_empty(&ci->i_cap_snaps) &&
1018 list_last_entry(&ci->i_cap_snaps, struct ceph_cap_snap,
1019 ci_item)->writing;
1020}
1021
1022/* inode.c */
1023struct ceph_mds_reply_info_in;
1024struct ceph_mds_reply_dirfrag;
1025struct ceph_acl_sec_ctx;
1026
1027extern const struct inode_operations ceph_file_iops;
1028
1029extern struct inode *ceph_alloc_inode(struct super_block *sb);
1030extern void ceph_evict_inode(struct inode *inode);
1031extern void ceph_free_inode(struct inode *inode);
1032
1033struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
1034 umode_t *mode, struct ceph_acl_sec_ctx *as_ctx);
1035void ceph_as_ctx_to_req(struct ceph_mds_request *req,
1036 struct ceph_acl_sec_ctx *as_ctx);
1037
1038extern struct inode *ceph_get_inode(struct super_block *sb,
1039 struct ceph_vino vino,
1040 struct inode *newino);
1041extern struct inode *ceph_get_snapdir(struct inode *parent);
1042extern int ceph_fill_file_size(struct inode *inode, int issued,
1043 u32 truncate_seq, u64 truncate_size, u64 size);
1044extern void ceph_fill_file_time(struct inode *inode, int issued,
1045 u64 time_warp_seq, struct timespec64 *ctime,
1046 struct timespec64 *mtime,
1047 struct timespec64 *atime);
1048extern int ceph_fill_inode(struct inode *inode, struct page *locked_page,
1049 struct ceph_mds_reply_info_in *iinfo,
1050 struct ceph_mds_reply_dirfrag *dirinfo,
1051 struct ceph_mds_session *session, int cap_fmode,
1052 struct ceph_cap_reservation *caps_reservation);
1053extern int ceph_fill_trace(struct super_block *sb,
1054 struct ceph_mds_request *req);
1055extern int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1056 struct ceph_mds_session *session);
1057
1058extern int ceph_inode_holds_cap(struct inode *inode, int mask);
1059
1060extern bool ceph_inode_set_size(struct inode *inode, loff_t size);
1061extern void __ceph_do_pending_vmtruncate(struct inode *inode);
1062
1063void ceph_queue_inode_work(struct inode *inode, int work_bit);
1064
1065static inline void ceph_queue_vmtruncate(struct inode *inode)
1066{
1067 ceph_queue_inode_work(inode, CEPH_I_WORK_VMTRUNCATE);
1068}
1069
1070static inline void ceph_queue_invalidate(struct inode *inode)
1071{
1072 ceph_queue_inode_work(inode, CEPH_I_WORK_INVALIDATE_PAGES);
1073}
1074
1075static inline void ceph_queue_writeback(struct inode *inode)
1076{
1077 ceph_queue_inode_work(inode, CEPH_I_WORK_WRITEBACK);
1078}
1079
1080static inline void ceph_queue_check_caps(struct inode *inode)
1081{
1082 ceph_queue_inode_work(inode, CEPH_I_WORK_CHECK_CAPS);
1083}
1084
1085static inline void ceph_queue_flush_snaps(struct inode *inode)
1086{
1087 ceph_queue_inode_work(inode, CEPH_I_WORK_FLUSH_SNAPS);
1088}
1089
1090extern int ceph_try_to_choose_auth_mds(struct inode *inode, int mask);
1091extern int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1092 int mask, bool force);
1093static inline int ceph_do_getattr(struct inode *inode, int mask, bool force)
1094{
1095 return __ceph_do_getattr(inode, NULL, mask, force);
1096}
1097extern int ceph_permission(struct mnt_idmap *idmap,
1098 struct inode *inode, int mask);
1099
1100struct ceph_iattr {
1101 struct ceph_fscrypt_auth *fscrypt_auth;
1102};
1103
1104extern int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode,
1105 struct iattr *attr, struct ceph_iattr *cia);
1106extern int ceph_setattr(struct mnt_idmap *idmap,
1107 struct dentry *dentry, struct iattr *attr);
1108extern int ceph_getattr(struct mnt_idmap *idmap,
1109 const struct path *path, struct kstat *stat,
1110 u32 request_mask, unsigned int flags);
1111void ceph_inode_shutdown(struct inode *inode);
1112
1113static inline bool ceph_inode_is_shutdown(struct inode *inode)
1114{
1115 unsigned long flags = READ_ONCE(ceph_inode(inode)->i_ceph_flags);
1116 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
1117 int state = READ_ONCE(fsc->mount_state);
1118
1119 return (flags & CEPH_I_SHUTDOWN) || state >= CEPH_MOUNT_SHUTDOWN;
1120}
1121
1122/* xattr.c */
1123int __ceph_setxattr(struct inode *, const char *, const void *, size_t, int);
1124int ceph_do_getvxattr(struct inode *inode, const char *name, void *value, size_t size);
1125ssize_t __ceph_getxattr(struct inode *, const char *, void *, size_t);
1126extern ssize_t ceph_listxattr(struct dentry *, char *, size_t);
1127extern struct ceph_buffer *__ceph_build_xattrs_blob(struct ceph_inode_info *ci);
1128extern void __ceph_destroy_xattrs(struct ceph_inode_info *ci);
1129extern const struct xattr_handler * const ceph_xattr_handlers[];
1130
1131struct ceph_acl_sec_ctx {
1132#ifdef CONFIG_CEPH_FS_POSIX_ACL
1133 void *default_acl;
1134 void *acl;
1135#endif
1136#ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1137 void *sec_ctx;
1138 u32 sec_ctxlen;
1139#endif
1140#ifdef CONFIG_FS_ENCRYPTION
1141 struct ceph_fscrypt_auth *fscrypt_auth;
1142#endif
1143 struct ceph_pagelist *pagelist;
1144};
1145
1146#ifdef CONFIG_SECURITY
1147extern bool ceph_security_xattr_deadlock(struct inode *in);
1148extern bool ceph_security_xattr_wanted(struct inode *in);
1149#else
1150static inline bool ceph_security_xattr_deadlock(struct inode *in)
1151{
1152 return false;
1153}
1154static inline bool ceph_security_xattr_wanted(struct inode *in)
1155{
1156 return false;
1157}
1158#endif
1159
1160#ifdef CONFIG_CEPH_FS_SECURITY_LABEL
1161extern int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1162 struct ceph_acl_sec_ctx *ctx);
1163static inline void ceph_security_invalidate_secctx(struct inode *inode)
1164{
1165 security_inode_invalidate_secctx(inode);
1166}
1167#else
1168static inline int ceph_security_init_secctx(struct dentry *dentry, umode_t mode,
1169 struct ceph_acl_sec_ctx *ctx)
1170{
1171 return 0;
1172}
1173static inline void ceph_security_invalidate_secctx(struct inode *inode)
1174{
1175}
1176#endif
1177
1178void ceph_release_acl_sec_ctx(struct ceph_acl_sec_ctx *as_ctx);
1179
1180/* acl.c */
1181#ifdef CONFIG_CEPH_FS_POSIX_ACL
1182
1183struct posix_acl *ceph_get_acl(struct inode *, int, bool);
1184int ceph_set_acl(struct mnt_idmap *idmap,
1185 struct dentry *dentry, struct posix_acl *acl, int type);
1186int ceph_pre_init_acls(struct inode *dir, umode_t *mode,
1187 struct ceph_acl_sec_ctx *as_ctx);
1188void ceph_init_inode_acls(struct inode *inode,
1189 struct ceph_acl_sec_ctx *as_ctx);
1190
1191static inline void ceph_forget_all_cached_acls(struct inode *inode)
1192{
1193 forget_all_cached_acls(inode);
1194}
1195
1196#else
1197
1198#define ceph_get_acl NULL
1199#define ceph_set_acl NULL
1200
1201static inline int ceph_pre_init_acls(struct inode *dir, umode_t *mode,
1202 struct ceph_acl_sec_ctx *as_ctx)
1203{
1204 return 0;
1205}
1206static inline void ceph_init_inode_acls(struct inode *inode,
1207 struct ceph_acl_sec_ctx *as_ctx)
1208{
1209}
1210static inline int ceph_acl_chmod(struct dentry *dentry, struct inode *inode)
1211{
1212 return 0;
1213}
1214
1215static inline void ceph_forget_all_cached_acls(struct inode *inode)
1216{
1217}
1218
1219#endif
1220
1221/* caps.c */
1222extern const char *ceph_cap_string(int c);
1223extern void ceph_handle_caps(struct ceph_mds_session *session,
1224 struct ceph_msg *msg);
1225extern struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
1226 struct ceph_cap_reservation *ctx);
1227extern void ceph_add_cap(struct inode *inode,
1228 struct ceph_mds_session *session, u64 cap_id,
1229 unsigned issued, unsigned wanted,
1230 unsigned cap, unsigned seq, u64 realmino, int flags,
1231 struct ceph_cap **new_cap);
1232extern void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release);
1233extern void ceph_remove_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1234 bool queue_release);
1235extern void __ceph_remove_caps(struct ceph_inode_info *ci);
1236extern void ceph_put_cap(struct ceph_mds_client *mdsc,
1237 struct ceph_cap *cap);
1238extern int ceph_is_any_caps(struct inode *inode);
1239
1240extern int ceph_write_inode(struct inode *inode, struct writeback_control *wbc);
1241extern int ceph_fsync(struct file *file, loff_t start, loff_t end,
1242 int datasync);
1243extern void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc,
1244 struct ceph_mds_session *session);
1245extern void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1246 struct ceph_mds_session *session);
1247void ceph_kick_flushing_inode_caps(struct ceph_mds_session *session,
1248 struct ceph_inode_info *ci);
1249extern struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci,
1250 int mds);
1251extern struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci,
1252 int mds);
1253extern void ceph_take_cap_refs(struct ceph_inode_info *ci, int caps,
1254 bool snap_rwsem_locked);
1255extern void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps);
1256extern void ceph_put_cap_refs(struct ceph_inode_info *ci, int had);
1257extern void ceph_put_cap_refs_async(struct ceph_inode_info *ci, int had);
1258extern void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
1259 struct ceph_snap_context *snapc);
1260extern void __ceph_remove_capsnap(struct inode *inode,
1261 struct ceph_cap_snap *capsnap,
1262 bool *wake_ci, bool *wake_mdsc);
1263extern void ceph_remove_capsnap(struct inode *inode,
1264 struct ceph_cap_snap *capsnap,
1265 bool *wake_ci, bool *wake_mdsc);
1266extern void ceph_flush_snaps(struct ceph_inode_info *ci,
1267 struct ceph_mds_session **psession);
1268extern bool __ceph_should_report_size(struct ceph_inode_info *ci);
1269extern void ceph_check_caps(struct ceph_inode_info *ci, int flags);
1270extern unsigned long ceph_check_delayed_caps(struct ceph_mds_client *mdsc);
1271extern void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc);
1272extern int ceph_drop_caps_for_unlink(struct inode *inode);
1273extern int ceph_encode_inode_release(void **p, struct inode *inode,
1274 int mds, int drop, int unless, int force);
1275extern int ceph_encode_dentry_release(void **p, struct dentry *dn,
1276 struct inode *dir,
1277 int mds, int drop, int unless);
1278
1279extern int __ceph_get_caps(struct inode *inode, struct ceph_file_info *fi,
1280 int need, int want, loff_t endoff, int *got);
1281extern int ceph_get_caps(struct file *filp, int need, int want,
1282 loff_t endoff, int *got);
1283extern int ceph_try_get_caps(struct inode *inode,
1284 int need, int want, bool nonblock, int *got);
1285
1286/* for counting open files by mode */
1287extern void ceph_get_fmode(struct ceph_inode_info *ci, int mode, int count);
1288extern void ceph_put_fmode(struct ceph_inode_info *ci, int mode, int count);
1289extern void __ceph_touch_fmode(struct ceph_inode_info *ci,
1290 struct ceph_mds_client *mdsc, int fmode);
1291
1292/* addr.c */
1293extern const struct address_space_operations ceph_aops;
1294extern const struct netfs_request_ops ceph_netfs_ops;
1295extern int ceph_mmap(struct file *file, struct vm_area_struct *vma);
1296extern int ceph_uninline_data(struct file *file);
1297extern int ceph_pool_perm_check(struct inode *inode, int need);
1298extern void ceph_pool_perm_destroy(struct ceph_mds_client* mdsc);
1299int ceph_purge_inode_cap(struct inode *inode, struct ceph_cap *cap, bool *invalidate);
1300
1301static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
1302{
1303 if (ci->i_inline_version == CEPH_INLINE_NONE ||
1304 ci->i_inline_version == 1) /* initial version, no data */
1305 return false;
1306 return true;
1307}
1308
1309/* file.c */
1310extern const struct file_operations ceph_file_fops;
1311
1312extern int ceph_renew_caps(struct inode *inode, int fmode);
1313extern int ceph_open(struct inode *inode, struct file *file);
1314extern int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
1315 struct file *file, unsigned flags, umode_t mode);
1316extern ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
1317 struct iov_iter *to, int *retry_op,
1318 u64 *last_objver);
1319extern int ceph_release(struct inode *inode, struct file *filp);
1320extern void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1321 char *data, size_t len);
1322
1323/* dir.c */
1324extern const struct file_operations ceph_dir_fops;
1325extern const struct file_operations ceph_snapdir_fops;
1326extern const struct inode_operations ceph_dir_iops;
1327extern const struct inode_operations ceph_snapdir_iops;
1328extern const struct dentry_operations ceph_dentry_ops;
1329
1330extern loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order);
1331extern int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry);
1332extern struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req,
1333 struct dentry *dentry);
1334extern struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
1335 struct dentry *dentry, int err);
1336
1337extern void __ceph_dentry_lease_touch(struct ceph_dentry_info *di);
1338extern void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di);
1339extern void ceph_invalidate_dentry_lease(struct dentry *dentry);
1340extern int ceph_trim_dentries(struct ceph_mds_client *mdsc);
1341extern unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn);
1342extern void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl);
1343
1344/* ioctl.c */
1345extern long ceph_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
1346
1347/* export.c */
1348extern const struct export_operations ceph_export_ops;
1349struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino);
1350
1351/* locks.c */
1352extern __init void ceph_flock_init(void);
1353extern int ceph_lock(struct file *file, int cmd, struct file_lock *fl);
1354extern int ceph_flock(struct file *file, int cmd, struct file_lock *fl);
1355extern void ceph_count_locks(struct inode *inode, int *p_num, int *f_num);
1356extern int ceph_encode_locks_to_buffer(struct inode *inode,
1357 struct ceph_filelock *flocks,
1358 int num_fcntl_locks,
1359 int num_flock_locks);
1360extern int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
1361 struct ceph_pagelist *pagelist,
1362 int num_fcntl_locks, int num_flock_locks);
1363
1364/* debugfs.c */
1365extern void ceph_fs_debugfs_init(struct ceph_fs_client *client);
1366extern void ceph_fs_debugfs_cleanup(struct ceph_fs_client *client);
1367
1368/* quota.c */
1369
1370enum quota_get_realm {
1371 QUOTA_GET_MAX_FILES,
1372 QUOTA_GET_MAX_BYTES,
1373 QUOTA_GET_ANY
1374};
1375
1376static inline bool __ceph_has_quota(struct ceph_inode_info *ci,
1377 enum quota_get_realm which)
1378{
1379 bool has_quota = false;
1380
1381 switch (which) {
1382 case QUOTA_GET_MAX_BYTES:
1383 has_quota = !!ci->i_max_bytes;
1384 break;
1385 case QUOTA_GET_MAX_FILES:
1386 has_quota = !!ci->i_max_files;
1387 break;
1388 default:
1389 has_quota = !!(ci->i_max_files || ci->i_max_bytes);
1390 }
1391 return has_quota;
1392}
1393
1394extern void ceph_adjust_quota_realms_count(struct inode *inode, bool inc);
1395
1396static inline void __ceph_update_quota(struct ceph_inode_info *ci,
1397 u64 max_bytes, u64 max_files)
1398{
1399 bool had_quota, has_quota;
1400 had_quota = __ceph_has_quota(ci, QUOTA_GET_ANY);
1401 ci->i_max_bytes = max_bytes;
1402 ci->i_max_files = max_files;
1403 has_quota = __ceph_has_quota(ci, QUOTA_GET_ANY);
1404
1405 if (had_quota != has_quota)
1406 ceph_adjust_quota_realms_count(&ci->netfs.inode, has_quota);
1407}
1408
1409static inline int __ceph_sparse_read_ext_count(struct inode *inode, u64 len)
1410{
1411 int cnt = 0;
1412
1413 if (IS_ENCRYPTED(inode)) {
1414 cnt = len >> CEPH_FSCRYPT_BLOCK_SHIFT;
1415 if (cnt > CEPH_SPARSE_EXT_ARRAY_INITIAL)
1416 cnt = 0;
1417 }
1418
1419 return cnt;
1420}
1421
1422extern void ceph_handle_quota(struct ceph_mds_client *mdsc,
1423 struct ceph_mds_session *session,
1424 struct ceph_msg *msg);
1425extern bool ceph_quota_is_max_files_exceeded(struct inode *inode);
1426extern bool ceph_quota_is_same_realm(struct inode *old, struct inode *new);
1427extern bool ceph_quota_is_max_bytes_exceeded(struct inode *inode,
1428 loff_t newlen);
1429extern bool ceph_quota_is_max_bytes_approaching(struct inode *inode,
1430 loff_t newlen);
1431extern bool ceph_quota_update_statfs(struct ceph_fs_client *fsc,
1432 struct kstatfs *buf);
1433extern void ceph_cleanup_quotarealms_inodes(struct ceph_mds_client *mdsc);
1434
1435bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1436 struct ceph_mds_session *session);
1437void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc);
1438bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc);
1439void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc);
1440#endif /* _FS_CEPH_SUPER_H */