]>
Commit | Line | Data |
---|---|---|
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- | |
2 | // vim: ts=8 sw=2 smarttab | |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net> | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
15 | ||
16 | #ifndef CEPH_CLIENT_H | |
17 | #define CEPH_CLIENT_H | |
18 | ||
19 | #include "common/CommandTable.h" | |
20 | #include "common/Finisher.h" | |
21 | #include "common/Timer.h" | |
22 | #include "common/ceph_mutex.h" | |
23 | #include "common/cmdparse.h" | |
24 | #include "common/compiler_extensions.h" | |
25 | #include "include/common_fwd.h" | |
26 | #include "include/cephfs/ceph_ll_client.h" | |
27 | #include "include/filepath.h" | |
28 | #include "include/interval_set.h" | |
29 | #include "include/lru.h" | |
30 | #include "include/types.h" | |
31 | #include "include/unordered_map.h" | |
32 | #include "include/unordered_set.h" | |
33 | #include "mds/mdstypes.h" | |
34 | #include "msg/Dispatcher.h" | |
35 | #include "msg/MessageRef.h" | |
36 | #include "msg/Messenger.h" | |
37 | #include "osdc/ObjectCacher.h" | |
38 | ||
39 | #include "InodeRef.h" | |
40 | #include "MetaSession.h" | |
41 | #include "UserPerm.h" | |
42 | ||
43 | #include <fstream> | |
44 | #include <map> | |
45 | #include <memory> | |
46 | #include <set> | |
47 | #include <string> | |
48 | ||
49 | using std::set; | |
50 | using std::map; | |
51 | using std::fstream; | |
52 | ||
53 | class FSMap; | |
54 | class FSMapUser; | |
55 | class MonClient; | |
56 | ||
57 | ||
58 | struct DirStat; | |
59 | struct LeaseStat; | |
60 | struct InodeStat; | |
61 | ||
62 | class Filer; | |
63 | class Objecter; | |
64 | class WritebackHandler; | |
65 | ||
66 | class MDSMap; | |
67 | class Message; | |
68 | ||
69 | enum { | |
70 | l_c_first = 20000, | |
71 | l_c_reply, | |
72 | l_c_lat, | |
73 | l_c_wrlat, | |
74 | l_c_read, | |
75 | l_c_fsync, | |
76 | l_c_last, | |
77 | }; | |
78 | ||
79 | ||
80 | class MDSCommandOp : public CommandOp | |
81 | { | |
82 | public: | |
83 | mds_gid_t mds_gid; | |
84 | ||
85 | explicit MDSCommandOp(ceph_tid_t t) : CommandOp(t) {} | |
86 | }; | |
87 | ||
88 | /* error code for ceph_fuse */ | |
89 | #define CEPH_FUSE_NO_MDS_UP -((1<<16)+0) /* no mds up deteced in ceph_fuse */ | |
90 | #define CEPH_FUSE_LAST -((1<<16)+1) /* (unused) */ | |
91 | ||
92 | // ============================================ | |
93 | // types for my local metadata cache | |
94 | /* basic structure: | |
95 | ||
96 | - Dentries live in an LRU loop. they get expired based on last access. | |
97 | see include/lru.h. items can be bumped to "mid" or "top" of list, etc. | |
98 | - Inode has ref count for each Fh, Dir, or Dentry that points to it. | |
99 | - when Inode ref goes to 0, it's expired. | |
100 | - when Dir is empty, it's removed (and it's Inode ref--) | |
101 | ||
102 | */ | |
103 | ||
104 | /* getdir result */ | |
105 | struct DirEntry { | |
106 | explicit DirEntry(const string &s) : d_name(s), stmask(0) {} | |
107 | DirEntry(const string &n, struct stat& s, int stm) : d_name(n), st(s), stmask(stm) {} | |
108 | ||
109 | string d_name; | |
110 | struct stat st; | |
111 | int stmask; | |
112 | }; | |
113 | ||
114 | struct Cap; | |
115 | class Dir; | |
116 | class Dentry; | |
117 | struct SnapRealm; | |
118 | struct Fh; | |
119 | struct CapSnap; | |
120 | ||
121 | struct MetaRequest; | |
122 | class ceph_lock_state_t; | |
123 | ||
124 | // ======================================================== | |
125 | // client interface | |
126 | ||
127 | struct dir_result_t { | |
128 | static const int SHIFT = 28; | |
129 | static const int64_t MASK = (1 << SHIFT) - 1; | |
130 | static const int64_t HASH = 0xFFULL << (SHIFT + 24); // impossible frag bits | |
131 | static const loff_t END = 1ULL << (SHIFT + 32); | |
132 | ||
133 | struct dentry { | |
134 | int64_t offset; | |
135 | string name; | |
136 | InodeRef inode; | |
137 | explicit dentry(int64_t o) : offset(o) {} | |
138 | dentry(int64_t o, const string& n, const InodeRef& in) : | |
139 | offset(o), name(n), inode(in) {} | |
140 | }; | |
141 | struct dentry_off_lt { | |
142 | bool operator()(const dentry& d, int64_t off) const { | |
143 | return dir_result_t::fpos_cmp(d.offset, off) < 0; | |
144 | } | |
145 | }; | |
146 | ||
147 | ||
148 | explicit dir_result_t(Inode *in, const UserPerm& perms); | |
149 | ||
150 | ||
151 | static uint64_t make_fpos(unsigned h, unsigned l, bool hash) { | |
152 | uint64_t v = ((uint64_t)h<< SHIFT) | (uint64_t)l; | |
153 | if (hash) | |
154 | v |= HASH; | |
155 | else | |
156 | ceph_assert((v & HASH) != HASH); | |
157 | return v; | |
158 | } | |
159 | static unsigned fpos_high(uint64_t p) { | |
160 | unsigned v = (p & (END-1)) >> SHIFT; | |
161 | if ((p & HASH) == HASH) | |
162 | return ceph_frag_value(v); | |
163 | return v; | |
164 | } | |
165 | static unsigned fpos_low(uint64_t p) { | |
166 | return p & MASK; | |
167 | } | |
168 | static int fpos_cmp(uint64_t l, uint64_t r) { | |
169 | int c = ceph_frag_compare(fpos_high(l), fpos_high(r)); | |
170 | if (c) | |
171 | return c; | |
172 | if (fpos_low(l) == fpos_low(r)) | |
173 | return 0; | |
174 | return fpos_low(l) < fpos_low(r) ? -1 : 1; | |
175 | } | |
176 | ||
177 | unsigned offset_high() { return fpos_high(offset); } | |
178 | unsigned offset_low() { return fpos_low(offset); } | |
179 | ||
180 | void set_end() { offset |= END; } | |
181 | bool at_end() { return (offset & END); } | |
182 | ||
183 | void set_hash_order() { offset |= HASH; } | |
184 | bool hash_order() { return (offset & HASH) == HASH; } | |
185 | ||
186 | bool is_cached() { | |
187 | if (buffer.empty()) | |
188 | return false; | |
189 | if (hash_order()) { | |
190 | return buffer_frag.contains(offset_high()); | |
191 | } else { | |
192 | return buffer_frag == frag_t(offset_high()); | |
193 | } | |
194 | } | |
195 | ||
196 | void reset() { | |
197 | last_name.clear(); | |
198 | next_offset = 2; | |
199 | offset = 0; | |
200 | ordered_count = 0; | |
201 | cache_index = 0; | |
202 | buffer.clear(); | |
203 | } | |
204 | ||
205 | InodeRef inode; | |
206 | int64_t offset; // hash order: | |
207 | // (0xff << 52) | ((24 bits hash) << 28) | | |
208 | // (the nth entry has hash collision); | |
209 | // frag+name order; | |
210 | // ((frag value) << 28) | (the nth entry in frag); | |
211 | ||
212 | unsigned next_offset; // offset of next chunk (last_name's + 1) | |
213 | string last_name; // last entry in previous chunk | |
214 | ||
215 | uint64_t release_count; | |
216 | uint64_t ordered_count; | |
217 | unsigned cache_index; | |
218 | int start_shared_gen; // dir shared_gen at start of readdir | |
219 | UserPerm perms; | |
220 | ||
221 | frag_t buffer_frag; | |
222 | ||
223 | vector<dentry> buffer; | |
224 | }; | |
225 | ||
226 | class Client : public Dispatcher, public md_config_obs_t { | |
227 | public: | |
228 | friend class C_Block_Sync; // Calls block map and protected helpers | |
229 | friend class C_Client_CacheInvalidate; // calls ino_invalidate_cb | |
230 | friend class C_Client_DentryInvalidate; // calls dentry_invalidate_cb | |
231 | friend class C_Client_FlushComplete; // calls put_inode() | |
232 | friend class C_Client_Remount; | |
233 | friend class C_Client_RequestInterrupt; | |
234 | friend class C_Deleg_Timeout; // Asserts on client_lock, called when a delegation is unreturned | |
235 | friend class C_Client_CacheRelease; // Asserts on client_lock | |
236 | friend class SyntheticClient; | |
237 | friend void intrusive_ptr_release(Inode *in); | |
238 | ||
239 | using Dispatcher::cct; | |
240 | ||
241 | typedef int (*add_dirent_cb_t)(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in); | |
242 | ||
243 | class CommandHook : public AdminSocketHook { | |
244 | public: | |
245 | explicit CommandHook(Client *client); | |
246 | int call(std::string_view command, const cmdmap_t& cmdmap, | |
247 | Formatter *f, | |
248 | std::ostream& errss, | |
249 | bufferlist& out) override; | |
250 | private: | |
251 | Client *m_client; | |
252 | }; | |
253 | ||
254 | Client(Messenger *m, MonClient *mc, Objecter *objecter_); | |
255 | Client(const Client&) = delete; | |
256 | Client(const Client&&) = delete; | |
257 | virtual ~Client() override; | |
258 | ||
259 | static UserPerm pick_my_perms(CephContext *c) { | |
260 | uid_t uid = c->_conf->client_mount_uid >= 0 ? c->_conf->client_mount_uid : -1; | |
261 | gid_t gid = c->_conf->client_mount_gid >= 0 ? c->_conf->client_mount_gid : -1; | |
262 | return UserPerm(uid, gid); | |
263 | } | |
264 | UserPerm pick_my_perms() { | |
265 | uid_t uid = user_id >= 0 ? user_id : -1; | |
266 | gid_t gid = group_id >= 0 ? group_id : -1; | |
267 | return UserPerm(uid, gid); | |
268 | } | |
269 | ||
270 | int mount(const std::string &mount_root, const UserPerm& perms, | |
271 | bool require_mds=false, const std::string &fs_name=""); | |
272 | void unmount(); | |
273 | void abort_conn(); | |
274 | ||
275 | void set_uuid(const std::string& uuid); | |
276 | void set_session_timeout(unsigned timeout); | |
277 | int start_reclaim(const std::string& uuid, unsigned flags, | |
278 | const std::string& fs_name); | |
279 | void finish_reclaim(); | |
280 | ||
281 | fs_cluster_id_t get_fs_cid() { | |
282 | return fscid; | |
283 | } | |
284 | ||
285 | int mds_command( | |
286 | const std::string &mds_spec, | |
287 | const std::vector<std::string>& cmd, | |
288 | const bufferlist& inbl, | |
289 | bufferlist *poutbl, std::string *prs, Context *onfinish); | |
290 | ||
291 | // these should (more or less) mirror the actual system calls. | |
292 | int statfs(const char *path, struct statvfs *stbuf, const UserPerm& perms); | |
293 | ||
294 | // crap | |
295 | int chdir(const char *s, std::string &new_cwd, const UserPerm& perms); | |
296 | void _getcwd(std::string& cwd, const UserPerm& perms); | |
297 | void getcwd(std::string& cwd, const UserPerm& perms); | |
298 | ||
299 | // namespace ops | |
300 | int opendir(const char *name, dir_result_t **dirpp, const UserPerm& perms); | |
301 | int closedir(dir_result_t *dirp); | |
302 | ||
303 | /** | |
304 | * Fill a directory listing from dirp, invoking cb for each entry | |
305 | * with the given pointer, the dirent, the struct stat, the stmask, | |
306 | * and the offset. | |
307 | * | |
308 | * Returns 0 if it reached the end of the directory. | |
309 | * If @a cb returns a negative error code, stop and return that. | |
310 | */ | |
311 | int readdir_r_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
312 | unsigned want=0, unsigned flags=AT_NO_ATTR_SYNC, | |
313 | bool getref=false); | |
314 | ||
315 | struct dirent * readdir(dir_result_t *d); | |
316 | int readdir_r(dir_result_t *dirp, struct dirent *de); | |
317 | int readdirplus_r(dir_result_t *dirp, struct dirent *de, struct ceph_statx *stx, unsigned want, unsigned flags, Inode **out); | |
318 | ||
319 | int getdir(const char *relpath, list<string>& names, | |
320 | const UserPerm& perms); // get the whole dir at once. | |
321 | ||
322 | /** | |
323 | * Returns the length of the buffer that got filled in, or -errno. | |
324 | * If it returns -ERANGE you just need to increase the size of the | |
325 | * buffer and try again. | |
326 | */ | |
327 | int _getdents(dir_result_t *dirp, char *buf, int buflen, bool ful); // get a bunch of dentries at once | |
328 | int getdents(dir_result_t *dirp, char *buf, int buflen) { | |
329 | return _getdents(dirp, buf, buflen, true); | |
330 | } | |
331 | int getdnames(dir_result_t *dirp, char *buf, int buflen) { | |
332 | return _getdents(dirp, buf, buflen, false); | |
333 | } | |
334 | ||
335 | void rewinddir(dir_result_t *dirp); | |
336 | loff_t telldir(dir_result_t *dirp); | |
337 | void seekdir(dir_result_t *dirp, loff_t offset); | |
338 | ||
339 | int link(const char *existing, const char *newname, const UserPerm& perm); | |
340 | int unlink(const char *path, const UserPerm& perm); | |
341 | int rename(const char *from, const char *to, const UserPerm& perm); | |
342 | ||
343 | // dirs | |
344 | int mkdir(const char *path, mode_t mode, const UserPerm& perm); | |
345 | int mkdirs(const char *path, mode_t mode, const UserPerm& perms); | |
346 | int rmdir(const char *path, const UserPerm& perms); | |
347 | ||
348 | // symlinks | |
349 | int readlink(const char *path, char *buf, loff_t size, const UserPerm& perms); | |
350 | ||
351 | int symlink(const char *existing, const char *newname, const UserPerm& perms); | |
352 | ||
353 | // inode stuff | |
354 | unsigned statx_to_mask(unsigned int flags, unsigned int want); | |
355 | int stat(const char *path, struct stat *stbuf, const UserPerm& perms, | |
356 | frag_info_t *dirstat=0, int mask=CEPH_STAT_CAP_INODE_ALL); | |
357 | int statx(const char *path, struct ceph_statx *stx, | |
358 | const UserPerm& perms, | |
359 | unsigned int want, unsigned int flags); | |
360 | int lstat(const char *path, struct stat *stbuf, const UserPerm& perms, | |
361 | frag_info_t *dirstat=0, int mask=CEPH_STAT_CAP_INODE_ALL); | |
362 | ||
363 | int setattr(const char *relpath, struct stat *attr, int mask, | |
364 | const UserPerm& perms); | |
365 | int setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
366 | const UserPerm& perms, int flags=0); | |
367 | int fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms); | |
368 | int fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms); | |
369 | int chmod(const char *path, mode_t mode, const UserPerm& perms); | |
370 | int fchmod(int fd, mode_t mode, const UserPerm& perms); | |
371 | int lchmod(const char *path, mode_t mode, const UserPerm& perms); | |
372 | int chown(const char *path, uid_t new_uid, gid_t new_gid, | |
373 | const UserPerm& perms); | |
374 | int fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms); | |
375 | int lchown(const char *path, uid_t new_uid, gid_t new_gid, | |
376 | const UserPerm& perms); | |
377 | int utime(const char *path, struct utimbuf *buf, const UserPerm& perms); | |
378 | int lutime(const char *path, struct utimbuf *buf, const UserPerm& perms); | |
379 | int futime(int fd, struct utimbuf *buf, const UserPerm& perms); | |
380 | int utimes(const char *relpath, struct timeval times[2], const UserPerm& perms); | |
381 | int lutimes(const char *relpath, struct timeval times[2], const UserPerm& perms); | |
382 | int futimes(int fd, struct timeval times[2], const UserPerm& perms); | |
383 | int futimens(int fd, struct timespec times[2], const UserPerm& perms); | |
384 | int flock(int fd, int operation, uint64_t owner); | |
385 | int truncate(const char *path, loff_t size, const UserPerm& perms); | |
386 | ||
387 | // file ops | |
388 | int mknod(const char *path, mode_t mode, const UserPerm& perms, dev_t rdev=0); | |
389 | int open(const char *path, int flags, const UserPerm& perms, mode_t mode=0); | |
390 | int open(const char *path, int flags, const UserPerm& perms, | |
391 | mode_t mode, int stripe_unit, int stripe_count, int object_size, | |
392 | const char *data_pool); | |
393 | int lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, | |
394 | const UserPerm& perms); | |
395 | int lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode=NULL); | |
396 | int lookup_name(Inode *in, Inode *parent, const UserPerm& perms); | |
397 | int close(int fd); | |
398 | loff_t lseek(int fd, loff_t offset, int whence); | |
399 | int read(int fd, char *buf, loff_t size, loff_t offset=-1); | |
400 | int preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset=-1); | |
401 | int write(int fd, const char *buf, loff_t size, loff_t offset=-1); | |
402 | int pwritev(int fd, const struct iovec *iov, int iovcnt, loff_t offset=-1); | |
403 | int fake_write_size(int fd, loff_t size); | |
404 | int ftruncate(int fd, loff_t size, const UserPerm& perms); | |
405 | int fsync(int fd, bool syncdataonly); | |
406 | int fstat(int fd, struct stat *stbuf, const UserPerm& perms, | |
407 | int mask=CEPH_STAT_CAP_INODE_ALL); | |
408 | int fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
409 | unsigned int want, unsigned int flags); | |
410 | int fallocate(int fd, int mode, loff_t offset, loff_t length); | |
411 | ||
412 | // full path xattr ops | |
413 | int getxattr(const char *path, const char *name, void *value, size_t size, | |
414 | const UserPerm& perms); | |
415 | int lgetxattr(const char *path, const char *name, void *value, size_t size, | |
416 | const UserPerm& perms); | |
417 | int fgetxattr(int fd, const char *name, void *value, size_t size, | |
418 | const UserPerm& perms); | |
419 | int listxattr(const char *path, char *list, size_t size, const UserPerm& perms); | |
420 | int llistxattr(const char *path, char *list, size_t size, const UserPerm& perms); | |
421 | int flistxattr(int fd, char *list, size_t size, const UserPerm& perms); | |
422 | int removexattr(const char *path, const char *name, const UserPerm& perms); | |
423 | int lremovexattr(const char *path, const char *name, const UserPerm& perms); | |
424 | int fremovexattr(int fd, const char *name, const UserPerm& perms); | |
425 | int setxattr(const char *path, const char *name, const void *value, | |
426 | size_t size, int flags, const UserPerm& perms); | |
427 | int lsetxattr(const char *path, const char *name, const void *value, | |
428 | size_t size, int flags, const UserPerm& perms); | |
429 | int fsetxattr(int fd, const char *name, const void *value, size_t size, | |
430 | int flags, const UserPerm& perms); | |
431 | ||
432 | int sync_fs(); | |
433 | int64_t drop_caches(); | |
434 | ||
435 | // hpc lazyio | |
436 | int lazyio(int fd, int enable); | |
437 | int lazyio_propagate(int fd, loff_t offset, size_t count); | |
438 | int lazyio_synchronize(int fd, loff_t offset, size_t count); | |
439 | ||
440 | // expose file layout | |
441 | int describe_layout(const char *path, file_layout_t* layout, | |
442 | const UserPerm& perms); | |
443 | int fdescribe_layout(int fd, file_layout_t* layout); | |
444 | int get_file_stripe_address(int fd, loff_t offset, vector<entity_addr_t>& address); | |
445 | int get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds); | |
446 | int get_osd_addr(int osd, entity_addr_t& addr); | |
447 | ||
448 | // expose mdsmap | |
449 | int64_t get_default_pool_id(); | |
450 | ||
451 | // expose osdmap | |
452 | int get_local_osd(); | |
453 | int get_pool_replication(int64_t pool); | |
454 | int64_t get_pool_id(const char *pool_name); | |
455 | string get_pool_name(int64_t pool); | |
456 | int get_osd_crush_location(int id, vector<pair<string, string> >& path); | |
457 | ||
458 | int enumerate_layout(int fd, vector<ObjectExtent>& result, | |
459 | loff_t length, loff_t offset); | |
460 | ||
461 | int mksnap(const char *path, const char *name, const UserPerm& perm); | |
462 | int rmsnap(const char *path, const char *name, const UserPerm& perm); | |
463 | ||
464 | // Inode permission checking | |
465 | int inode_permission(Inode *in, const UserPerm& perms, unsigned want); | |
466 | ||
467 | // expose caps | |
468 | int get_caps_issued(int fd); | |
469 | int get_caps_issued(const char *path, const UserPerm& perms); | |
470 | ||
471 | snapid_t ll_get_snapid(Inode *in); | |
472 | vinodeno_t ll_get_vino(Inode *in) { | |
473 | std::lock_guard lock(client_lock); | |
474 | return _get_vino(in); | |
475 | } | |
476 | // get inode from faked ino | |
477 | Inode *ll_get_inode(ino_t ino); | |
478 | Inode *ll_get_inode(vinodeno_t vino); | |
479 | int ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
480 | Inode **out, const UserPerm& perms); | |
481 | int ll_lookup_inode(struct inodeno_t ino, const UserPerm& perms, Inode **inode); | |
482 | int ll_lookupx(Inode *parent, const char *name, Inode **out, | |
483 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
484 | const UserPerm& perms); | |
485 | bool ll_forget(Inode *in, uint64_t count); | |
486 | bool ll_put(Inode *in); | |
487 | int ll_get_snap_ref(snapid_t snap); | |
488 | ||
489 | int ll_getattr(Inode *in, struct stat *st, const UserPerm& perms); | |
490 | int ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
491 | unsigned int flags, const UserPerm& perms); | |
492 | int ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
493 | const UserPerm& perms); | |
494 | int ll_setattr(Inode *in, struct stat *st, int mask, | |
495 | const UserPerm& perms); | |
496 | int ll_getxattr(Inode *in, const char *name, void *value, size_t size, | |
497 | const UserPerm& perms); | |
498 | int ll_setxattr(Inode *in, const char *name, const void *value, size_t size, | |
499 | int flags, const UserPerm& perms); | |
500 | int ll_removexattr(Inode *in, const char *name, const UserPerm& perms); | |
501 | int ll_listxattr(Inode *in, char *list, size_t size, const UserPerm& perms); | |
502 | int ll_opendir(Inode *in, int flags, dir_result_t **dirpp, | |
503 | const UserPerm& perms); | |
504 | int ll_releasedir(dir_result_t* dirp); | |
505 | int ll_fsyncdir(dir_result_t* dirp); | |
506 | int ll_readlink(Inode *in, char *buf, size_t bufsize, const UserPerm& perms); | |
507 | int ll_mknod(Inode *in, const char *name, mode_t mode, dev_t rdev, | |
508 | struct stat *attr, Inode **out, const UserPerm& perms); | |
509 | int ll_mknodx(Inode *parent, const char *name, mode_t mode, dev_t rdev, | |
510 | Inode **out, struct ceph_statx *stx, unsigned want, | |
511 | unsigned flags, const UserPerm& perms); | |
512 | int ll_mkdir(Inode *in, const char *name, mode_t mode, struct stat *attr, | |
513 | Inode **out, const UserPerm& perm); | |
514 | int ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
515 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
516 | const UserPerm& perms); | |
517 | int ll_symlink(Inode *in, const char *name, const char *value, | |
518 | struct stat *attr, Inode **out, const UserPerm& perms); | |
519 | int ll_symlinkx(Inode *parent, const char *name, const char *value, | |
520 | Inode **out, struct ceph_statx *stx, unsigned want, | |
521 | unsigned flags, const UserPerm& perms); | |
522 | int ll_unlink(Inode *in, const char *name, const UserPerm& perm); | |
523 | int ll_rmdir(Inode *in, const char *name, const UserPerm& perms); | |
524 | int ll_rename(Inode *parent, const char *name, Inode *newparent, | |
525 | const char *newname, const UserPerm& perm); | |
526 | int ll_link(Inode *in, Inode *newparent, const char *newname, | |
527 | const UserPerm& perm); | |
528 | int ll_open(Inode *in, int flags, Fh **fh, const UserPerm& perms); | |
529 | int _ll_create(Inode *parent, const char *name, mode_t mode, | |
530 | int flags, InodeRef *in, int caps, Fh **fhp, | |
531 | const UserPerm& perms); | |
532 | int ll_create(Inode *parent, const char *name, mode_t mode, int flags, | |
533 | struct stat *attr, Inode **out, Fh **fhp, | |
534 | const UserPerm& perms); | |
535 | int ll_createx(Inode *parent, const char *name, mode_t mode, | |
536 | int oflags, Inode **outp, Fh **fhp, | |
537 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
538 | const UserPerm& perms); | |
539 | int ll_read_block(Inode *in, uint64_t blockid, char *buf, uint64_t offset, | |
540 | uint64_t length, file_layout_t* layout); | |
541 | ||
542 | int ll_write_block(Inode *in, uint64_t blockid, | |
543 | char* buf, uint64_t offset, | |
544 | uint64_t length, file_layout_t* layout, | |
545 | uint64_t snapseq, uint32_t sync); | |
546 | int ll_commit_blocks(Inode *in, uint64_t offset, uint64_t length); | |
547 | ||
548 | int ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms); | |
549 | int ll_walk(const char* name, Inode **i, struct ceph_statx *stx, | |
550 | unsigned int want, unsigned int flags, const UserPerm& perms); | |
551 | uint32_t ll_stripe_unit(Inode *in); | |
552 | int ll_file_layout(Inode *in, file_layout_t *layout); | |
553 | uint64_t ll_snap_seq(Inode *in); | |
554 | ||
555 | int ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl); | |
556 | int ll_write(Fh *fh, loff_t off, loff_t len, const char *data); | |
557 | int64_t ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off); | |
558 | int64_t ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off); | |
559 | loff_t ll_lseek(Fh *fh, loff_t offset, int whence); | |
560 | int ll_flush(Fh *fh); | |
561 | int ll_fsync(Fh *fh, bool syncdataonly); | |
562 | int ll_sync_inode(Inode *in, bool syncdataonly); | |
563 | int ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length); | |
564 | int ll_release(Fh *fh); | |
565 | int ll_getlk(Fh *fh, struct flock *fl, uint64_t owner); | |
566 | int ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep); | |
567 | int ll_flock(Fh *fh, int cmd, uint64_t owner); | |
568 | int ll_lazyio(Fh *fh, int enable); | |
569 | int ll_file_layout(Fh *fh, file_layout_t *layout); | |
570 | void ll_interrupt(void *d); | |
571 | bool ll_handle_umask() { | |
572 | return acl_type != NO_ACL; | |
573 | } | |
574 | ||
575 | int ll_get_stripe_osd(struct Inode *in, uint64_t blockno, | |
576 | file_layout_t* layout); | |
577 | uint64_t ll_get_internal_offset(struct Inode *in, uint64_t blockno); | |
578 | ||
579 | int ll_num_osds(void); | |
580 | int ll_osdaddr(int osd, uint32_t *addr); | |
581 | int ll_osdaddr(int osd, char* buf, size_t size); | |
582 | ||
583 | void ll_register_callbacks(struct ceph_client_callback_args *args); | |
584 | int test_dentry_handling(bool can_invalidate); | |
585 | ||
586 | const char** get_tracked_conf_keys() const override; | |
587 | void handle_conf_change(const ConfigProxy& conf, | |
588 | const std::set <std::string> &changed) override; | |
589 | uint32_t get_deleg_timeout() { return deleg_timeout; } | |
590 | int set_deleg_timeout(uint32_t timeout); | |
591 | int ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv); | |
592 | ||
593 | entity_name_t get_myname() { return messenger->get_myname(); } | |
594 | void wait_on_list(std::list<ceph::condition_variable*>& ls); | |
595 | void signal_cond_list(std::list<ceph::condition_variable*>& ls); | |
596 | ||
597 | void set_filer_flags(int flags); | |
598 | void clear_filer_flags(int flags); | |
599 | ||
600 | void tear_down_cache(); | |
601 | ||
602 | void update_metadata(std::string const &k, std::string const &v); | |
603 | ||
604 | client_t get_nodeid() { return whoami; } | |
605 | ||
606 | inodeno_t get_root_ino(); | |
607 | Inode *get_root(); | |
608 | ||
609 | virtual int init(); | |
610 | virtual void shutdown(); | |
611 | ||
612 | // messaging | |
613 | void handle_mds_map(const MConstRef<MMDSMap>& m); | |
614 | void handle_fs_map(const MConstRef<MFSMap>& m); | |
615 | void handle_fs_map_user(const MConstRef<MFSMapUser>& m); | |
616 | void handle_osd_map(const MConstRef<MOSDMap>& m); | |
617 | ||
618 | void handle_lease(const MConstRef<MClientLease>& m); | |
619 | ||
620 | // inline data | |
621 | int uninline_data(Inode *in, Context *onfinish); | |
622 | ||
623 | // file caps | |
624 | void check_cap_issue(Inode *in, unsigned issued); | |
625 | void add_update_cap(Inode *in, MetaSession *session, uint64_t cap_id, | |
626 | unsigned issued, unsigned wanted, unsigned seq, unsigned mseq, | |
627 | inodeno_t realm, int flags, const UserPerm& perms); | |
628 | void remove_cap(Cap *cap, bool queue_release); | |
629 | void remove_all_caps(Inode *in); | |
630 | void remove_session_caps(MetaSession *session); | |
631 | int mark_caps_flushing(Inode *in, ceph_tid_t *ptid); | |
632 | void adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s); | |
633 | void flush_caps_sync(); | |
634 | void kick_flushing_caps(Inode *in, MetaSession *session); | |
635 | void kick_flushing_caps(MetaSession *session); | |
636 | void early_kick_flushing_caps(MetaSession *session); | |
637 | int get_caps(Inode *in, int need, int want, int *have, loff_t endoff); | |
638 | int get_caps_used(Inode *in); | |
639 | ||
640 | void maybe_update_snaprealm(SnapRealm *realm, snapid_t snap_created, snapid_t snap_highwater, | |
641 | vector<snapid_t>& snaps); | |
642 | ||
643 | void handle_quota(const MConstRef<MClientQuota>& m); | |
644 | void handle_snap(const MConstRef<MClientSnap>& m); | |
645 | void handle_caps(const MConstRef<MClientCaps>& m); | |
646 | void handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m); | |
647 | void handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m); | |
648 | void handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m); | |
649 | void handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m); | |
650 | void handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m); | |
651 | void handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m); | |
652 | void cap_delay_requeue(Inode *in); | |
653 | ||
654 | void send_cap(Inode *in, MetaSession *session, Cap *cap, int flags, | |
655 | int used, int want, int retain, int flush, | |
656 | ceph_tid_t flush_tid); | |
657 | ||
658 | void send_flush_snap(Inode *in, MetaSession *session, snapid_t follows, CapSnap& capsnap); | |
659 | ||
660 | void flush_snaps(Inode *in); | |
661 | void get_cap_ref(Inode *in, int cap); | |
662 | void put_cap_ref(Inode *in, int cap); | |
663 | void wait_sync_caps(Inode *in, ceph_tid_t want); | |
664 | void wait_sync_caps(ceph_tid_t want); | |
665 | void queue_cap_snap(Inode *in, SnapContext &old_snapc); | |
666 | void finish_cap_snap(Inode *in, CapSnap &capsnap, int used); | |
667 | void _flushed_cap_snap(Inode *in, snapid_t seq); | |
668 | ||
669 | void _schedule_invalidate_dentry_callback(Dentry *dn, bool del); | |
670 | void _async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name); | |
671 | void _try_to_trim_inode(Inode *in, bool sched_inval); | |
672 | ||
673 | void _schedule_invalidate_callback(Inode *in, int64_t off, int64_t len); | |
674 | void _invalidate_inode_cache(Inode *in); | |
675 | void _invalidate_inode_cache(Inode *in, int64_t off, int64_t len); | |
676 | void _async_invalidate(vinodeno_t ino, int64_t off, int64_t len); | |
677 | ||
678 | void _schedule_ino_release_callback(Inode *in); | |
679 | void _async_inode_release(vinodeno_t ino); | |
680 | ||
681 | bool _release(Inode *in); | |
682 | ||
683 | /** | |
684 | * Initiate a flush of the data associated with the given inode. | |
685 | * If you specify a Context, you are responsible for holding an inode | |
686 | * reference for the duration of the flush. If not, _flush() will | |
687 | * take the reference for you. | |
688 | * @param in The Inode whose data you wish to flush. | |
689 | * @param c The Context you wish us to complete once the data is | |
690 | * flushed. If already flushed, this will be called in-line. | |
691 | * | |
692 | * @returns true if the data was already flushed, false otherwise. | |
693 | */ | |
694 | bool _flush(Inode *in, Context *c); | |
695 | void _flush_range(Inode *in, int64_t off, uint64_t size); | |
696 | void _flushed(Inode *in); | |
697 | void flush_set_callback(ObjectCacher::ObjectSet *oset); | |
698 | ||
699 | void close_release(Inode *in); | |
700 | void close_safe(Inode *in); | |
701 | ||
702 | void lock_fh_pos(Fh *f); | |
703 | void unlock_fh_pos(Fh *f); | |
704 | ||
705 | // metadata cache | |
706 | void update_dir_dist(Inode *in, DirStat *st); | |
707 | ||
708 | void clear_dir_complete_and_ordered(Inode *diri, bool complete); | |
709 | void insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri); | |
710 | Inode* insert_trace(MetaRequest *request, MetaSession *session); | |
711 | void update_inode_file_size(Inode *in, int issued, uint64_t size, | |
712 | uint64_t truncate_seq, uint64_t truncate_size); | |
713 | void update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq, | |
714 | utime_t ctime, utime_t mtime, utime_t atime); | |
715 | ||
716 | Inode *add_update_inode(InodeStat *st, utime_t ttl, MetaSession *session, | |
717 | const UserPerm& request_perms); | |
718 | Dentry *insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
719 | Inode *in, utime_t from, MetaSession *session, | |
720 | Dentry *old_dentry = NULL); | |
721 | void update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session); | |
722 | ||
723 | bool use_faked_inos() { return _use_faked_inos; } | |
724 | vinodeno_t map_faked_ino(ino_t ino); | |
725 | ||
726 | //notify the mds to flush the mdlog | |
727 | void flush_mdlog_sync(); | |
728 | void flush_mdlog(MetaSession *session); | |
729 | ||
730 | void renew_caps(); | |
731 | void renew_caps(MetaSession *session); | |
732 | void flush_cap_releases(); | |
733 | void tick(); | |
734 | ||
735 | xlist<Inode*> &get_dirty_list() { return dirty_list; } | |
736 | ||
737 | SafeTimer timer; | |
738 | ||
739 | std::unique_ptr<PerfCounters> logger; | |
740 | std::unique_ptr<MDSMap> mdsmap; | |
741 | ||
742 | bool fuse_default_permissions; | |
743 | ||
744 | protected: | |
745 | /* Flags for check_caps() */ | |
746 | static const unsigned CHECK_CAPS_NODELAY = 0x1; | |
747 | static const unsigned CHECK_CAPS_SYNCHRONOUS = 0x2; | |
748 | ||
749 | ||
750 | bool is_initialized() const { return initialized; } | |
751 | ||
752 | void check_caps(Inode *in, unsigned flags); | |
753 | ||
754 | void set_cap_epoch_barrier(epoch_t e); | |
755 | ||
756 | void handle_command_reply(const MConstRef<MCommandReply>& m); | |
757 | int fetch_fsmap(bool user); | |
758 | int resolve_mds( | |
759 | const std::string &mds_spec, | |
760 | std::vector<mds_gid_t> *targets); | |
761 | ||
762 | void get_session_metadata(std::map<std::string, std::string> *meta) const; | |
763 | bool have_open_session(mds_rank_t mds); | |
764 | void got_mds_push(MetaSession *s); | |
765 | MetaSession *_get_mds_session(mds_rank_t mds, Connection *con); ///< return session for mds *and* con; null otherwise | |
766 | MetaSession *_get_or_open_mds_session(mds_rank_t mds); | |
767 | MetaSession *_open_mds_session(mds_rank_t mds); | |
768 | void _close_mds_session(MetaSession *s); | |
769 | void _closed_mds_session(MetaSession *s); | |
770 | bool _any_stale_sessions() const; | |
771 | void _kick_stale_sessions(); | |
772 | void handle_client_session(const MConstRef<MClientSession>& m); | |
773 | void send_reconnect(MetaSession *s); | |
774 | void resend_unsafe_requests(MetaSession *s); | |
775 | void wait_unsafe_requests(); | |
776 | ||
777 | void _sync_write_commit(Inode *in); | |
778 | ||
779 | void dump_mds_requests(Formatter *f); | |
780 | void dump_mds_sessions(Formatter *f); | |
781 | ||
782 | int make_request(MetaRequest *req, const UserPerm& perms, | |
783 | InodeRef *ptarget = 0, bool *pcreated = 0, | |
784 | mds_rank_t use_mds=-1, bufferlist *pdirbl=0); | |
785 | void put_request(MetaRequest *request); | |
786 | void unregister_request(MetaRequest *request); | |
787 | ||
788 | int verify_reply_trace(int r, MetaSession *session, MetaRequest *request, | |
789 | const MConstRef<MClientReply>& reply, | |
790 | InodeRef *ptarget, bool *pcreated, | |
791 | const UserPerm& perms); | |
792 | void encode_cap_releases(MetaRequest *request, mds_rank_t mds); | |
793 | int encode_inode_release(Inode *in, MetaRequest *req, | |
794 | mds_rank_t mds, int drop, | |
795 | int unless,int force=0); | |
796 | void encode_dentry_release(Dentry *dn, MetaRequest *req, | |
797 | mds_rank_t mds, int drop, int unless); | |
798 | mds_rank_t choose_target_mds(MetaRequest *req, Inode** phash_diri=NULL); | |
799 | void connect_mds_targets(mds_rank_t mds); | |
800 | void send_request(MetaRequest *request, MetaSession *session, | |
801 | bool drop_cap_releases=false); | |
802 | MRef<MClientRequest> build_client_request(MetaRequest *request); | |
803 | void kick_requests(MetaSession *session); | |
804 | void kick_requests_closed(MetaSession *session); | |
805 | void handle_client_request_forward(const MConstRef<MClientRequestForward>& reply); | |
806 | void handle_client_reply(const MConstRef<MClientReply>& reply); | |
807 | bool is_dir_operation(MetaRequest *request); | |
808 | ||
809 | // fake inode number for 32-bits ino_t | |
810 | void _assign_faked_ino(Inode *in); | |
811 | void _assign_faked_root(Inode *in); | |
812 | void _release_faked_ino(Inode *in); | |
813 | void _reset_faked_inos(); | |
814 | vinodeno_t _map_faked_ino(ino_t ino); | |
815 | ||
816 | // Optional extra metadata about me to send to the MDS | |
817 | void populate_metadata(const std::string &mount_root); | |
818 | ||
819 | SnapRealm *get_snap_realm(inodeno_t r); | |
820 | SnapRealm *get_snap_realm_maybe(inodeno_t r); | |
821 | void put_snap_realm(SnapRealm *realm); | |
822 | bool adjust_realm_parent(SnapRealm *realm, inodeno_t parent); | |
823 | void update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool must_flush=true); | |
824 | void invalidate_snaprealm_and_children(SnapRealm *realm); | |
825 | ||
826 | Inode *open_snapdir(Inode *diri); | |
827 | ||
828 | int get_fd() { | |
829 | int fd = free_fd_set.range_start(); | |
830 | free_fd_set.erase(fd, 1); | |
831 | return fd; | |
832 | } | |
833 | void put_fd(int fd) { | |
834 | free_fd_set.insert(fd, 1); | |
835 | } | |
836 | ||
837 | /* | |
838 | * Resolve file descriptor, or return NULL. | |
839 | */ | |
840 | Fh *get_filehandle(int fd) { | |
841 | ceph::unordered_map<int, Fh*>::iterator p = fd_map.find(fd); | |
842 | if (p == fd_map.end()) | |
843 | return NULL; | |
844 | return p->second; | |
845 | } | |
846 | ||
847 | // helpers | |
848 | void wake_up_session_caps(MetaSession *s, bool reconnect); | |
849 | ||
850 | void wait_on_context_list(list<Context*>& ls); | |
851 | void signal_context_list(list<Context*>& ls); | |
852 | ||
853 | // -- metadata cache stuff | |
854 | ||
855 | // decrease inode ref. delete if dangling. | |
856 | void put_inode(Inode *in, int n=1); | |
857 | void close_dir(Dir *dir); | |
858 | ||
859 | int subscribe_mdsmap(const std::string &fs_name=""); | |
860 | ||
861 | void _abort_mds_sessions(int err); | |
862 | ||
863 | // same as unmount() but for when the client_lock is already held | |
864 | void _unmount(bool abort); | |
865 | ||
866 | //int get_cache_size() { return lru.lru_get_size(); } | |
867 | ||
868 | /** | |
869 | * Don't call this with in==NULL, use get_or_create for that | |
870 | * leave dn set to default NULL unless you're trying to add | |
871 | * a new inode to a pre-created Dentry | |
872 | */ | |
873 | Dentry* link(Dir *dir, const string& name, Inode *in, Dentry *dn); | |
874 | void unlink(Dentry *dn, bool keepdir, bool keepdentry); | |
875 | ||
876 | // path traversal for high-level interface | |
877 | int path_walk(const filepath& fp, InodeRef *end, const UserPerm& perms, | |
878 | bool followsym=true, int mask=0); | |
879 | ||
880 | int fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat=0, nest_info_t *rstat=0); | |
881 | int fill_stat(InodeRef& in, struct stat *st, frag_info_t *dirstat=0, nest_info_t *rstat=0) { | |
882 | return fill_stat(in.get(), st, dirstat, rstat); | |
883 | } | |
884 | ||
885 | void fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx); | |
886 | void fill_statx(InodeRef& in, unsigned int mask, struct ceph_statx *stx) { | |
887 | return fill_statx(in.get(), mask, stx); | |
888 | } | |
889 | ||
890 | void touch_dn(Dentry *dn); | |
891 | ||
892 | // trim cache. | |
893 | void trim_cache(bool trim_kernel_dcache=false); | |
894 | void trim_cache_for_reconnect(MetaSession *s); | |
895 | void trim_dentry(Dentry *dn); | |
896 | void trim_caps(MetaSession *s, uint64_t max); | |
897 | void _invalidate_kernel_dcache(); | |
898 | void _trim_negative_child_dentries(InodeRef& in); | |
899 | ||
900 | void dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected); | |
901 | void dump_cache(Formatter *f); // debug | |
902 | ||
903 | // force read-only | |
904 | void force_session_readonly(MetaSession *s); | |
905 | ||
906 | void dump_status(Formatter *f); // debug | |
907 | ||
908 | bool ms_dispatch2(const MessageRef& m) override; | |
909 | ||
910 | void ms_handle_connect(Connection *con) override; | |
911 | bool ms_handle_reset(Connection *con) override; | |
912 | void ms_handle_remote_reset(Connection *con) override; | |
913 | bool ms_handle_refused(Connection *con) override; | |
914 | ||
915 | int authenticate(); | |
916 | ||
917 | Inode* get_quota_root(Inode *in, const UserPerm& perms); | |
918 | bool check_quota_condition(Inode *in, const UserPerm& perms, | |
919 | std::function<bool (const Inode &)> test); | |
920 | bool is_quota_files_exceeded(Inode *in, const UserPerm& perms); | |
921 | bool is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
922 | const UserPerm& perms); | |
923 | bool is_quota_bytes_approaching(Inode *in, const UserPerm& perms); | |
924 | ||
925 | int check_pool_perm(Inode *in, int need); | |
926 | ||
927 | void handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply); | |
928 | ||
929 | /** | |
930 | * Call this when an OSDMap is seen with a full flag (global or per pool) | |
931 | * set. | |
932 | * | |
933 | * @param pool the pool ID affected, or -1 if all. | |
934 | */ | |
935 | void _handle_full_flag(int64_t pool); | |
936 | ||
937 | void _close_sessions(); | |
938 | ||
939 | void _pre_init(); | |
940 | ||
941 | /** | |
942 | * The basic housekeeping parts of init (perf counters, admin socket) | |
943 | * that is independent of how objecters/monclient/messengers are | |
944 | * being set up. | |
945 | */ | |
946 | void _finish_init(); | |
947 | ||
948 | // global client lock | |
949 | // - protects Client and buffer cache both! | |
950 | ceph::mutex client_lock = ceph::make_mutex("Client::client_lock"); | |
951 | ; | |
952 | ||
953 | std::map<snapid_t, int> ll_snap_ref; | |
954 | ||
955 | Inode* root = nullptr; | |
956 | map<Inode*, InodeRef> root_parents; | |
957 | Inode* root_ancestor = nullptr; | |
958 | LRU lru; // lru list of Dentry's in our local metadata cache. | |
959 | ||
960 | InodeRef cwd; | |
961 | ||
962 | std::unique_ptr<Filer> filer; | |
963 | std::unique_ptr<ObjectCacher> objectcacher; | |
964 | std::unique_ptr<WritebackHandler> writeback_handler; | |
965 | ||
966 | Messenger *messenger; | |
967 | MonClient *monclient; | |
968 | Objecter *objecter; | |
969 | ||
970 | client_t whoami; | |
971 | ||
972 | ||
973 | private: | |
974 | struct C_Readahead : public Context { | |
975 | C_Readahead(Client *c, Fh *f); | |
976 | ~C_Readahead() override; | |
977 | void finish(int r) override; | |
978 | ||
979 | Client *client; | |
980 | Fh *f; | |
981 | }; | |
982 | ||
983 | /* | |
984 | * These define virtual xattrs exposing the recursive directory | |
985 | * statistics and layout metadata. | |
986 | */ | |
987 | struct VXattr { | |
988 | const string name; | |
989 | size_t (Client::*getxattr_cb)(Inode *in, char *val, size_t size); | |
990 | bool readonly; | |
991 | bool (Client::*exists_cb)(Inode *in); | |
992 | unsigned int flags; | |
993 | }; | |
994 | ||
995 | enum { | |
996 | NO_ACL = 0, | |
997 | POSIX_ACL, | |
998 | }; | |
999 | ||
1000 | enum { | |
1001 | MAY_EXEC = 1, | |
1002 | MAY_WRITE = 2, | |
1003 | MAY_READ = 4, | |
1004 | }; | |
1005 | ||
1006 | ||
1007 | /* Flags for VXattr */ | |
1008 | static const unsigned VXATTR_RSTAT = 0x1; | |
1009 | ||
1010 | static const VXattr _dir_vxattrs[]; | |
1011 | static const VXattr _file_vxattrs[]; | |
1012 | ||
1013 | ||
1014 | ||
1015 | void fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off); | |
1016 | ||
1017 | int _opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms); | |
1018 | void _readdir_drop_dirp_buffer(dir_result_t *dirp); | |
1019 | bool _readdir_have_frag(dir_result_t *dirp); | |
1020 | void _readdir_next_frag(dir_result_t *dirp); | |
1021 | void _readdir_rechoose_frag(dir_result_t *dirp); | |
1022 | int _readdir_get_frag(dir_result_t *dirp); | |
1023 | int _readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, int caps, bool getref); | |
1024 | void _closedir(dir_result_t *dirp); | |
1025 | ||
1026 | // other helpers | |
1027 | void _fragmap_remove_non_leaves(Inode *in); | |
1028 | void _fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds); | |
1029 | ||
1030 | void _ll_get(Inode *in); | |
1031 | int _ll_put(Inode *in, uint64_t num); | |
1032 | void _ll_drop_pins(); | |
1033 | ||
1034 | Fh *_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms); | |
1035 | int _release_fh(Fh *fh); | |
1036 | void _put_fh(Fh *fh); | |
1037 | ||
1038 | int _do_remount(bool retry_on_error); | |
1039 | ||
1040 | int _read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, bool *checkeof); | |
1041 | int _read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl); | |
1042 | ||
1043 | // internal interface | |
1044 | // call these with client_lock held! | |
1045 | int _do_lookup(Inode *dir, const string& name, int mask, InodeRef *target, | |
1046 | const UserPerm& perms); | |
1047 | ||
1048 | int _lookup(Inode *dir, const string& dname, int mask, InodeRef *target, | |
1049 | const UserPerm& perm); | |
1050 | ||
1051 | int _link(Inode *in, Inode *dir, const char *name, const UserPerm& perm, | |
1052 | InodeRef *inp = 0); | |
1053 | int _unlink(Inode *dir, const char *name, const UserPerm& perm); | |
1054 | int _rename(Inode *olddir, const char *oname, Inode *ndir, const char *nname, const UserPerm& perm); | |
1055 | int _mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, | |
1056 | InodeRef *inp = 0); | |
1057 | int _rmdir(Inode *dir, const char *name, const UserPerm& perms); | |
1058 | int _symlink(Inode *dir, const char *name, const char *target, | |
1059 | const UserPerm& perms, InodeRef *inp = 0); | |
1060 | int _mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
1061 | const UserPerm& perms, InodeRef *inp = 0); | |
1062 | int _do_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
1063 | const UserPerm& perms, InodeRef *inp); | |
1064 | void stat_to_statx(struct stat *st, struct ceph_statx *stx); | |
1065 | int __setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
1066 | const UserPerm& perms, InodeRef *inp = 0); | |
1067 | int _setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
1068 | const UserPerm& perms); | |
1069 | int _setattr(InodeRef &in, struct stat *attr, int mask, | |
1070 | const UserPerm& perms); | |
1071 | int _ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
1072 | const UserPerm& perms, InodeRef *inp = 0); | |
1073 | int _getattr(Inode *in, int mask, const UserPerm& perms, bool force=false); | |
1074 | int _getattr(InodeRef &in, int mask, const UserPerm& perms, bool force=false) { | |
1075 | return _getattr(in.get(), mask, perms, force); | |
1076 | } | |
1077 | int _readlink(Inode *in, char *buf, size_t size); | |
1078 | int _getxattr(Inode *in, const char *name, void *value, size_t len, | |
1079 | const UserPerm& perms); | |
1080 | int _getxattr(InodeRef &in, const char *name, void *value, size_t len, | |
1081 | const UserPerm& perms); | |
1082 | int _listxattr(Inode *in, char *names, size_t len, const UserPerm& perms); | |
1083 | int _do_setxattr(Inode *in, const char *name, const void *value, size_t len, | |
1084 | int flags, const UserPerm& perms); | |
1085 | int _setxattr(Inode *in, const char *name, const void *value, size_t len, | |
1086 | int flags, const UserPerm& perms); | |
1087 | int _setxattr(InodeRef &in, const char *name, const void *value, size_t len, | |
1088 | int flags, const UserPerm& perms); | |
1089 | int _setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap); | |
1090 | void _setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t len); | |
1091 | int _removexattr(Inode *in, const char *nm, const UserPerm& perms); | |
1092 | int _removexattr(InodeRef &in, const char *nm, const UserPerm& perms); | |
1093 | int _open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
1094 | const UserPerm& perms); | |
1095 | int _renew_caps(Inode *in); | |
1096 | int _create(Inode *in, const char *name, int flags, mode_t mode, InodeRef *inp, | |
1097 | Fh **fhp, int stripe_unit, int stripe_count, int object_size, | |
1098 | const char *data_pool, bool *created, const UserPerm &perms); | |
1099 | ||
1100 | loff_t _lseek(Fh *fh, loff_t offset, int whence); | |
1101 | int64_t _read(Fh *fh, int64_t offset, uint64_t size, bufferlist *bl); | |
1102 | int64_t _write(Fh *fh, int64_t offset, uint64_t size, const char *buf, | |
1103 | const struct iovec *iov, int iovcnt); | |
1104 | int64_t _preadv_pwritev_locked(Fh *f, const struct iovec *iov, | |
1105 | unsigned iovcnt, int64_t offset, bool write, bool clamp_to_int); | |
1106 | int _preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write); | |
1107 | int _flush(Fh *fh); | |
1108 | int _fsync(Fh *fh, bool syncdataonly); | |
1109 | int _fsync(Inode *in, bool syncdataonly); | |
1110 | int _sync_fs(); | |
1111 | int _fallocate(Fh *fh, int mode, int64_t offset, int64_t length); | |
1112 | int _getlk(Fh *fh, struct flock *fl, uint64_t owner); | |
1113 | int _setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep); | |
1114 | int _flock(Fh *fh, int cmd, uint64_t owner); | |
1115 | int _lazyio(Fh *fh, int enable); | |
1116 | ||
1117 | int get_or_create(Inode *dir, const char* name, | |
1118 | Dentry **pdn, bool expect_null=false); | |
1119 | ||
1120 | int xattr_permission(Inode *in, const char *name, unsigned want, | |
1121 | const UserPerm& perms); | |
1122 | int may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
1123 | const UserPerm& perms); | |
1124 | int may_open(Inode *in, int flags, const UserPerm& perms); | |
1125 | int may_lookup(Inode *dir, const UserPerm& perms); | |
1126 | int may_create(Inode *dir, const UserPerm& perms); | |
1127 | int may_delete(Inode *dir, const char *name, const UserPerm& perms); | |
1128 | int may_hardlink(Inode *in, const UserPerm& perms); | |
1129 | ||
1130 | int _getattr_for_perm(Inode *in, const UserPerm& perms); | |
1131 | ||
1132 | vinodeno_t _get_vino(Inode *in); | |
1133 | ||
1134 | bool _vxattrcb_quota_exists(Inode *in); | |
1135 | size_t _vxattrcb_quota(Inode *in, char *val, size_t size); | |
1136 | size_t _vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size); | |
1137 | size_t _vxattrcb_quota_max_files(Inode *in, char *val, size_t size); | |
1138 | ||
1139 | bool _vxattrcb_layout_exists(Inode *in); | |
1140 | size_t _vxattrcb_layout(Inode *in, char *val, size_t size); | |
1141 | size_t _vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size); | |
1142 | size_t _vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size); | |
1143 | size_t _vxattrcb_layout_object_size(Inode *in, char *val, size_t size); | |
1144 | size_t _vxattrcb_layout_pool(Inode *in, char *val, size_t size); | |
1145 | size_t _vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size); | |
1146 | size_t _vxattrcb_dir_entries(Inode *in, char *val, size_t size); | |
1147 | size_t _vxattrcb_dir_files(Inode *in, char *val, size_t size); | |
1148 | size_t _vxattrcb_dir_subdirs(Inode *in, char *val, size_t size); | |
1149 | size_t _vxattrcb_dir_rentries(Inode *in, char *val, size_t size); | |
1150 | size_t _vxattrcb_dir_rfiles(Inode *in, char *val, size_t size); | |
1151 | size_t _vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size); | |
1152 | size_t _vxattrcb_dir_rbytes(Inode *in, char *val, size_t size); | |
1153 | size_t _vxattrcb_dir_rctime(Inode *in, char *val, size_t size); | |
1154 | ||
1155 | bool _vxattrcb_dir_pin_exists(Inode *in); | |
1156 | size_t _vxattrcb_dir_pin(Inode *in, char *val, size_t size); | |
1157 | ||
1158 | bool _vxattrcb_snap_btime_exists(Inode *in); | |
1159 | size_t _vxattrcb_snap_btime(Inode *in, char *val, size_t size); | |
1160 | ||
1161 | static const VXattr *_get_vxattrs(Inode *in); | |
1162 | static const VXattr *_match_vxattr(Inode *in, const char *name); | |
1163 | ||
1164 | int _do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
1165 | struct flock *fl, uint64_t owner, bool removing=false); | |
1166 | int _interrupt_filelock(MetaRequest *req); | |
1167 | void _encode_filelocks(Inode *in, bufferlist& bl); | |
1168 | void _release_filelocks(Fh *fh); | |
1169 | void _update_lock_state(struct flock *fl, uint64_t owner, ceph_lock_state_t *lock_state); | |
1170 | ||
1171 | int _posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
1172 | const UserPerm& perms); | |
1173 | int _posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms); | |
1174 | int _posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want); | |
1175 | ||
1176 | mds_rank_t _get_random_up_mds() const; | |
1177 | ||
1178 | int _ll_getattr(Inode *in, int caps, const UserPerm& perms); | |
1179 | int _lookup_parent(Inode *in, const UserPerm& perms, Inode **parent=NULL); | |
1180 | int _lookup_name(Inode *in, Inode *parent, const UserPerm& perms); | |
1181 | int _lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode=NULL); | |
1182 | bool _ll_forget(Inode *in, uint64_t count); | |
1183 | ||
1184 | ||
1185 | uint32_t deleg_timeout = 0; | |
1186 | ||
1187 | client_switch_interrupt_callback_t switch_interrupt_cb = nullptr; | |
1188 | client_remount_callback_t remount_cb = nullptr; | |
1189 | client_ino_callback_t ino_invalidate_cb = nullptr; | |
1190 | client_dentry_callback_t dentry_invalidate_cb = nullptr; | |
1191 | client_umask_callback_t umask_cb = nullptr; | |
1192 | client_ino_release_t ino_release_cb = nullptr; | |
1193 | void *callback_handle = nullptr; | |
1194 | bool can_invalidate_dentries = false; | |
1195 | ||
1196 | Finisher async_ino_invalidator; | |
1197 | Finisher async_dentry_invalidator; | |
1198 | Finisher interrupt_finisher; | |
1199 | Finisher remount_finisher; | |
1200 | Finisher async_ino_releasor; | |
1201 | Finisher objecter_finisher; | |
1202 | ||
1203 | Context *tick_event = nullptr; | |
1204 | utime_t last_cap_renew; | |
1205 | ||
1206 | CommandHook m_command_hook; | |
1207 | ||
1208 | int user_id, group_id; | |
1209 | int acl_type = NO_ACL; | |
1210 | ||
1211 | epoch_t cap_epoch_barrier = 0; | |
1212 | ||
1213 | // mds sessions | |
1214 | map<mds_rank_t, MetaSession> mds_sessions; // mds -> push seq | |
1215 | std::list<ceph::condition_variable*> waiting_for_mdsmap; | |
1216 | ||
1217 | // FSMap, for when using mds_command | |
1218 | std::list<ceph::condition_variable*> waiting_for_fsmap; | |
1219 | std::unique_ptr<FSMap> fsmap; | |
1220 | std::unique_ptr<FSMapUser> fsmap_user; | |
1221 | ||
1222 | // MDS command state | |
1223 | CommandTable<MDSCommandOp> command_table; | |
1224 | ||
1225 | bool _use_faked_inos; | |
1226 | ||
1227 | // Cluster fsid | |
1228 | fs_cluster_id_t fscid; | |
1229 | ||
1230 | // file handles, etc. | |
1231 | interval_set<int> free_fd_set; // unused fds | |
1232 | ceph::unordered_map<int, Fh*> fd_map; | |
1233 | set<Fh*> ll_unclosed_fh_set; | |
1234 | ceph::unordered_set<dir_result_t*> opened_dirs; | |
1235 | ||
1236 | bool initialized = false; | |
1237 | bool mounted = false; | |
1238 | bool unmounting = false; | |
1239 | bool blacklisted = false; | |
1240 | ||
1241 | ceph::unordered_map<vinodeno_t, Inode*> inode_map; | |
1242 | ceph::unordered_map<ino_t, vinodeno_t> faked_ino_map; | |
1243 | interval_set<ino_t> free_faked_inos; | |
1244 | ino_t last_used_faked_ino; | |
1245 | ino_t last_used_faked_root; | |
1246 | ||
1247 | // When an MDS has sent us a REJECT, remember that and don't | |
1248 | // contact it again. Remember which inst rejected us, so that | |
1249 | // when we talk to another inst with the same rank we can | |
1250 | // try again. | |
1251 | std::map<mds_rank_t, entity_addrvec_t> rejected_by_mds; | |
1252 | ||
1253 | int local_osd = -ENXIO; | |
1254 | epoch_t local_osd_epoch = 0; | |
1255 | ||
1256 | int unsafe_sync_write = 0; | |
1257 | ||
1258 | // mds requests | |
1259 | ceph_tid_t last_tid = 0; | |
1260 | ceph_tid_t oldest_tid = 0; // oldest incomplete mds request, excluding setfilelock requests | |
1261 | map<ceph_tid_t, MetaRequest*> mds_requests; | |
1262 | ||
1263 | // cap flushing | |
1264 | ceph_tid_t last_flush_tid = 1; | |
1265 | ||
1266 | // dirty_list keeps all the dirty inodes before flushing. | |
1267 | xlist<Inode*> delayed_list, dirty_list; | |
1268 | int num_flushing_caps = 0; | |
1269 | ceph::unordered_map<inodeno_t,SnapRealm*> snap_realms; | |
1270 | std::map<std::string, std::string> metadata; | |
1271 | ||
1272 | // trace generation | |
1273 | ofstream traceout; | |
1274 | ||
1275 | ceph::condition_variable mount_cond, sync_cond; | |
1276 | ||
1277 | std::map<std::pair<int64_t,std::string>, int> pool_perms; | |
1278 | std::list<ceph::condition_variable*> waiting_for_pool_perm; | |
1279 | ||
1280 | uint64_t retries_on_invalidate = 0; | |
1281 | ||
1282 | // state reclaim | |
1283 | std::list<ceph::condition_variable*> waiting_for_reclaim; | |
1284 | int reclaim_errno = 0; | |
1285 | epoch_t reclaim_osd_epoch = 0; | |
1286 | entity_addrvec_t reclaim_target_addrs; | |
1287 | }; | |
1288 | ||
1289 | /** | |
1290 | * Specialization of Client that manages its own Objecter instance | |
1291 | * and handles init/shutdown of messenger/monclient | |
1292 | */ | |
1293 | class StandaloneClient : public Client | |
1294 | { | |
1295 | public: | |
1296 | StandaloneClient(Messenger *m, MonClient *mc); | |
1297 | ||
1298 | ~StandaloneClient() override; | |
1299 | ||
1300 | int init() override; | |
1301 | void shutdown() override; | |
1302 | }; | |
1303 | ||
1304 | #endif |