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