1 // SPDX-License-Identifier: GPL-2.0-only
2
3 #include <linux/ceph/ceph_debug.h>
4
5 #include <linux/backing-dev.h>
6 #include <linux/ctype.h>
7 #include <linux/fs.h>
8 #include <linux/inet.h>
9 #include <linux/in6.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/sched.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/statfs.h>
17 #include <linux/string.h>
18
19 #include "super.h"
20 #include "mds_client.h"
21 #include "cache.h"
22
23 #include <linux/ceph/ceph_features.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/mon_client.h>
26 #include <linux/ceph/auth.h>
27 #include <linux/ceph/debugfs.h>
28
29 /*
30 * Ceph superblock operations
31 *
32 * Handle the basics of mounting, unmounting.
33 */
34
35 /*
36 * super ops
37 */
ceph_put_super(struct super_block * s)38 static void ceph_put_super(struct super_block *s)
39 {
40 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
41
42 dout("put_super\n");
43 ceph_mdsc_close_sessions(fsc->mdsc);
44 }
45
ceph_statfs(struct dentry * dentry,struct kstatfs * buf)46 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
47 {
48 struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
49 struct ceph_mon_client *monc = &fsc->client->monc;
50 struct ceph_statfs st;
51 u64 fsid;
52 int err;
53 u64 data_pool;
54
55 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
56 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
57 } else {
58 data_pool = CEPH_NOPOOL;
59 }
60
61 dout("statfs\n");
62 err = ceph_monc_do_statfs(monc, data_pool, &st);
63 if (err < 0)
64 return err;
65
66 /* fill in kstatfs */
67 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
68
69 /*
70 * express utilization in terms of large blocks to avoid
71 * overflow on 32-bit machines.
72 *
73 * NOTE: for the time being, we make bsize == frsize to humor
74 * not-yet-ancient versions of glibc that are broken.
75 * Someday, we will probably want to report a real block
76 * size... whatever that may mean for a network file system!
77 */
78 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
79 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
80
81 /*
82 * By default use root quota for stats; fallback to overall filesystem
83 * usage if using 'noquotadf' mount option or if the root dir doesn't
84 * have max_bytes quota set.
85 */
86 if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
87 !ceph_quota_update_statfs(fsc, buf)) {
88 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
89 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
90 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
91 }
92
93 buf->f_files = le64_to_cpu(st.num_objects);
94 buf->f_ffree = -1;
95 buf->f_namelen = NAME_MAX;
96
97 /* Must convert the fsid, for consistent values across arches */
98 mutex_lock(&monc->mutex);
99 fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
100 le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
101 mutex_unlock(&monc->mutex);
102
103 buf->f_fsid.val[0] = fsid & 0xffffffff;
104 buf->f_fsid.val[1] = fsid >> 32;
105
106 return 0;
107 }
108
ceph_sync_fs(struct super_block * sb,int wait)109 static int ceph_sync_fs(struct super_block *sb, int wait)
110 {
111 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
112
113 if (!wait) {
114 dout("sync_fs (non-blocking)\n");
115 ceph_flush_dirty_caps(fsc->mdsc);
116 dout("sync_fs (non-blocking) done\n");
117 return 0;
118 }
119
120 dout("sync_fs (blocking)\n");
121 ceph_osdc_sync(&fsc->client->osdc);
122 ceph_mdsc_sync(fsc->mdsc);
123 dout("sync_fs (blocking) done\n");
124 return 0;
125 }
126
127 /*
128 * mount options
129 */
130 enum {
131 Opt_wsize,
132 Opt_rsize,
133 Opt_rasize,
134 Opt_caps_wanted_delay_min,
135 Opt_caps_wanted_delay_max,
136 Opt_caps_max,
137 Opt_readdir_max_entries,
138 Opt_readdir_max_bytes,
139 Opt_congestion_kb,
140 Opt_last_int,
141 /* int args above */
142 Opt_snapdirname,
143 Opt_mds_namespace,
144 Opt_fscache_uniq,
145 Opt_recover_session,
146 Opt_last_string,
147 /* string args above */
148 Opt_dirstat,
149 Opt_nodirstat,
150 Opt_rbytes,
151 Opt_norbytes,
152 Opt_asyncreaddir,
153 Opt_noasyncreaddir,
154 Opt_dcache,
155 Opt_nodcache,
156 Opt_ino32,
157 Opt_noino32,
158 Opt_fscache,
159 Opt_nofscache,
160 Opt_poolperm,
161 Opt_nopoolperm,
162 Opt_require_active_mds,
163 Opt_norequire_active_mds,
164 #ifdef CONFIG_CEPH_FS_POSIX_ACL
165 Opt_acl,
166 #endif
167 Opt_noacl,
168 Opt_quotadf,
169 Opt_noquotadf,
170 Opt_copyfrom,
171 Opt_nocopyfrom,
172 };
173
174 static match_table_t fsopt_tokens = {
175 {Opt_wsize, "wsize=%d"},
176 {Opt_rsize, "rsize=%d"},
177 {Opt_rasize, "rasize=%d"},
178 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
179 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
180 {Opt_caps_max, "caps_max=%d"},
181 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
182 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
183 {Opt_congestion_kb, "write_congestion_kb=%d"},
184 /* int args above */
185 {Opt_snapdirname, "snapdirname=%s"},
186 {Opt_mds_namespace, "mds_namespace=%s"},
187 {Opt_recover_session, "recover_session=%s"},
188 {Opt_fscache_uniq, "fsc=%s"},
189 /* string args above */
190 {Opt_dirstat, "dirstat"},
191 {Opt_nodirstat, "nodirstat"},
192 {Opt_rbytes, "rbytes"},
193 {Opt_norbytes, "norbytes"},
194 {Opt_asyncreaddir, "asyncreaddir"},
195 {Opt_noasyncreaddir, "noasyncreaddir"},
196 {Opt_dcache, "dcache"},
197 {Opt_nodcache, "nodcache"},
198 {Opt_ino32, "ino32"},
199 {Opt_noino32, "noino32"},
200 {Opt_fscache, "fsc"},
201 {Opt_nofscache, "nofsc"},
202 {Opt_poolperm, "poolperm"},
203 {Opt_nopoolperm, "nopoolperm"},
204 {Opt_require_active_mds, "require_active_mds"},
205 {Opt_norequire_active_mds, "norequire_active_mds"},
206 #ifdef CONFIG_CEPH_FS_POSIX_ACL
207 {Opt_acl, "acl"},
208 #endif
209 {Opt_noacl, "noacl"},
210 {Opt_quotadf, "quotadf"},
211 {Opt_noquotadf, "noquotadf"},
212 {Opt_copyfrom, "copyfrom"},
213 {Opt_nocopyfrom, "nocopyfrom"},
214 {-1, NULL}
215 };
216
217 /*
218 * Remove adjacent slashes and then the trailing slash, unless it is
219 * the only remaining character.
220 *
221 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
222 */
canonicalize_path(char * path)223 static void canonicalize_path(char *path)
224 {
225 int i, j = 0;
226
227 for (i = 0; path[i] != '\0'; i++) {
228 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
229 path[j++] = path[i];
230 }
231
232 if (j > 1 && path[j - 1] == '/')
233 j--;
234 path[j] = '\0';
235 }
236
parse_fsopt_token(char * c,void * private)237 static int parse_fsopt_token(char *c, void *private)
238 {
239 struct ceph_mount_options *fsopt = private;
240 substring_t argstr[MAX_OPT_ARGS];
241 int token, intval, ret;
242
243 token = match_token((char *)c, fsopt_tokens, argstr);
244 if (token < 0)
245 return -EINVAL;
246
247 if (token < Opt_last_int) {
248 ret = match_int(&argstr[0], &intval);
249 if (ret < 0) {
250 pr_err("bad option arg (not int) at '%s'\n", c);
251 return ret;
252 }
253 dout("got int token %d val %d\n", token, intval);
254 } else if (token > Opt_last_int && token < Opt_last_string) {
255 dout("got string token %d val %s\n", token,
256 argstr[0].from);
257 } else {
258 dout("got token %d\n", token);
259 }
260
261 switch (token) {
262 case Opt_snapdirname:
263 kfree(fsopt->snapdir_name);
264 fsopt->snapdir_name = kstrndup(argstr[0].from,
265 argstr[0].to-argstr[0].from,
266 GFP_KERNEL);
267 if (!fsopt->snapdir_name)
268 return -ENOMEM;
269 break;
270 case Opt_mds_namespace:
271 kfree(fsopt->mds_namespace);
272 fsopt->mds_namespace = kstrndup(argstr[0].from,
273 argstr[0].to-argstr[0].from,
274 GFP_KERNEL);
275 if (!fsopt->mds_namespace)
276 return -ENOMEM;
277 break;
278 case Opt_recover_session:
279 if (!strncmp(argstr[0].from, "no",
280 argstr[0].to - argstr[0].from)) {
281 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
282 } else if (!strncmp(argstr[0].from, "clean",
283 argstr[0].to - argstr[0].from)) {
284 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
285 } else {
286 return -EINVAL;
287 }
288 break;
289 case Opt_fscache_uniq:
290 #ifdef CONFIG_CEPH_FSCACHE
291 kfree(fsopt->fscache_uniq);
292 fsopt->fscache_uniq = kstrndup(argstr[0].from,
293 argstr[0].to-argstr[0].from,
294 GFP_KERNEL);
295 if (!fsopt->fscache_uniq)
296 return -ENOMEM;
297 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
298 break;
299 #else
300 pr_err("fscache support is disabled\n");
301 return -EINVAL;
302 #endif
303 case Opt_wsize:
304 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_WRITE_SIZE)
305 return -EINVAL;
306 fsopt->wsize = ALIGN(intval, PAGE_SIZE);
307 break;
308 case Opt_rsize:
309 if (intval < (int)PAGE_SIZE || intval > CEPH_MAX_READ_SIZE)
310 return -EINVAL;
311 fsopt->rsize = ALIGN(intval, PAGE_SIZE);
312 break;
313 case Opt_rasize:
314 if (intval < 0)
315 return -EINVAL;
316 fsopt->rasize = ALIGN(intval, PAGE_SIZE);
317 break;
318 case Opt_caps_wanted_delay_min:
319 if (intval < 1)
320 return -EINVAL;
321 fsopt->caps_wanted_delay_min = intval;
322 break;
323 case Opt_caps_wanted_delay_max:
324 if (intval < 1)
325 return -EINVAL;
326 fsopt->caps_wanted_delay_max = intval;
327 break;
328 case Opt_caps_max:
329 if (intval < 0)
330 return -EINVAL;
331 fsopt->caps_max = intval;
332 break;
333 case Opt_readdir_max_entries:
334 if (intval < 1)
335 return -EINVAL;
336 fsopt->max_readdir = intval;
337 break;
338 case Opt_readdir_max_bytes:
339 if (intval < (int)PAGE_SIZE && intval != 0)
340 return -EINVAL;
341 fsopt->max_readdir_bytes = intval;
342 break;
343 case Opt_congestion_kb:
344 if (intval < 1024) /* at least 1M */
345 return -EINVAL;
346 fsopt->congestion_kb = intval;
347 break;
348 case Opt_dirstat:
349 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
350 break;
351 case Opt_nodirstat:
352 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
353 break;
354 case Opt_rbytes:
355 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
356 break;
357 case Opt_norbytes:
358 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
359 break;
360 case Opt_asyncreaddir:
361 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
362 break;
363 case Opt_noasyncreaddir:
364 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
365 break;
366 case Opt_dcache:
367 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
368 break;
369 case Opt_nodcache:
370 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
371 break;
372 case Opt_ino32:
373 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
374 break;
375 case Opt_noino32:
376 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
377 break;
378 case Opt_fscache:
379 #ifdef CONFIG_CEPH_FSCACHE
380 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
381 kfree(fsopt->fscache_uniq);
382 fsopt->fscache_uniq = NULL;
383 break;
384 #else
385 pr_err("fscache support is disabled\n");
386 return -EINVAL;
387 #endif
388 case Opt_nofscache:
389 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
390 kfree(fsopt->fscache_uniq);
391 fsopt->fscache_uniq = NULL;
392 break;
393 case Opt_poolperm:
394 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
395 break;
396 case Opt_nopoolperm:
397 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
398 break;
399 case Opt_require_active_mds:
400 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
401 break;
402 case Opt_norequire_active_mds:
403 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
404 break;
405 case Opt_quotadf:
406 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
407 break;
408 case Opt_noquotadf:
409 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
410 break;
411 case Opt_copyfrom:
412 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
413 break;
414 case Opt_nocopyfrom:
415 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
416 break;
417 #ifdef CONFIG_CEPH_FS_POSIX_ACL
418 case Opt_acl:
419 fsopt->sb_flags |= SB_POSIXACL;
420 break;
421 #endif
422 case Opt_noacl:
423 fsopt->sb_flags &= ~SB_POSIXACL;
424 break;
425 default:
426 BUG_ON(token);
427 }
428 return 0;
429 }
430
destroy_mount_options(struct ceph_mount_options * args)431 static void destroy_mount_options(struct ceph_mount_options *args)
432 {
433 dout("destroy_mount_options %p\n", args);
434 kfree(args->snapdir_name);
435 kfree(args->mds_namespace);
436 kfree(args->server_path);
437 kfree(args->fscache_uniq);
438 kfree(args);
439 }
440
strcmp_null(const char * s1,const char * s2)441 static int strcmp_null(const char *s1, const char *s2)
442 {
443 if (!s1 && !s2)
444 return 0;
445 if (s1 && !s2)
446 return -1;
447 if (!s1 && s2)
448 return 1;
449 return strcmp(s1, s2);
450 }
451
compare_mount_options(struct ceph_mount_options * new_fsopt,struct ceph_options * new_opt,struct ceph_fs_client * fsc)452 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
453 struct ceph_options *new_opt,
454 struct ceph_fs_client *fsc)
455 {
456 struct ceph_mount_options *fsopt1 = new_fsopt;
457 struct ceph_mount_options *fsopt2 = fsc->mount_options;
458 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
459 int ret;
460
461 ret = memcmp(fsopt1, fsopt2, ofs);
462 if (ret)
463 return ret;
464
465 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
466 if (ret)
467 return ret;
468
469 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
470 if (ret)
471 return ret;
472
473 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
474 if (ret)
475 return ret;
476
477 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
478 if (ret)
479 return ret;
480
481 return ceph_compare_options(new_opt, fsc->client);
482 }
483
parse_mount_options(struct ceph_mount_options ** pfsopt,struct ceph_options ** popt,int flags,char * options,const char * dev_name)484 static int parse_mount_options(struct ceph_mount_options **pfsopt,
485 struct ceph_options **popt,
486 int flags, char *options,
487 const char *dev_name)
488 {
489 struct ceph_mount_options *fsopt;
490 const char *dev_name_end;
491 int err;
492
493 if (!dev_name || !*dev_name)
494 return -EINVAL;
495
496 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
497 if (!fsopt)
498 return -ENOMEM;
499
500 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
501
502 fsopt->sb_flags = flags;
503 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
504
505 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
506 fsopt->rsize = CEPH_MAX_READ_SIZE;
507 fsopt->rasize = CEPH_RASIZE_DEFAULT;
508 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
509 if (!fsopt->snapdir_name) {
510 err = -ENOMEM;
511 goto out;
512 }
513
514 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
515 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
516 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
517 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
518 fsopt->congestion_kb = default_congestion_kb();
519
520 /*
521 * Distinguish the server list from the path in "dev_name".
522 * Internally we do not include the leading '/' in the path.
523 *
524 * "dev_name" will look like:
525 * <server_spec>[,<server_spec>...]:[<path>]
526 * where
527 * <server_spec> is <ip>[:<port>]
528 * <path> is optional, but if present must begin with '/'
529 */
530 dev_name_end = strchr(dev_name, '/');
531 if (dev_name_end) {
532 /*
533 * The server_path will include the whole chars from userland
534 * including the leading '/'.
535 */
536 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
537 if (!fsopt->server_path) {
538 err = -ENOMEM;
539 goto out;
540 }
541
542 canonicalize_path(fsopt->server_path);
543 } else {
544 dev_name_end = dev_name + strlen(dev_name);
545 }
546 err = -EINVAL;
547 dev_name_end--; /* back up to ':' separator */
548 if (dev_name_end < dev_name || *dev_name_end != ':') {
549 pr_err("device name is missing path (no : separator in %s)\n",
550 dev_name);
551 goto out;
552 }
553 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
554 if (fsopt->server_path)
555 dout("server path '%s'\n", fsopt->server_path);
556
557 *popt = ceph_parse_options(options, dev_name, dev_name_end,
558 parse_fsopt_token, (void *)fsopt);
559 if (IS_ERR(*popt)) {
560 err = PTR_ERR(*popt);
561 goto out;
562 }
563
564 /* success */
565 *pfsopt = fsopt;
566 return 0;
567
568 out:
569 destroy_mount_options(fsopt);
570 return err;
571 }
572
573 /**
574 * ceph_show_options - Show mount options in /proc/mounts
575 * @m: seq_file to write to
576 * @root: root of that (sub)tree
577 */
ceph_show_options(struct seq_file * m,struct dentry * root)578 static int ceph_show_options(struct seq_file *m, struct dentry *root)
579 {
580 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
581 struct ceph_mount_options *fsopt = fsc->mount_options;
582 size_t pos;
583 int ret;
584
585 /* a comma between MNT/MS and client options */
586 seq_putc(m, ',');
587 pos = m->count;
588
589 ret = ceph_print_client_options(m, fsc->client, false);
590 if (ret)
591 return ret;
592
593 /* retract our comma if no client options */
594 if (m->count == pos)
595 m->count--;
596
597 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
598 seq_puts(m, ",dirstat");
599 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
600 seq_puts(m, ",rbytes");
601 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
602 seq_puts(m, ",noasyncreaddir");
603 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
604 seq_puts(m, ",nodcache");
605 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
606 seq_puts(m, ",ino32");
607 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
608 seq_show_option(m, "fsc", fsopt->fscache_uniq);
609 }
610 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
611 seq_puts(m, ",nopoolperm");
612 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
613 seq_puts(m, ",noquotadf");
614
615 #ifdef CONFIG_CEPH_FS_POSIX_ACL
616 if (fsopt->sb_flags & SB_POSIXACL)
617 seq_puts(m, ",acl");
618 else
619 seq_puts(m, ",noacl");
620 #endif
621
622 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
623 seq_puts(m, ",copyfrom");
624
625 if (fsopt->mds_namespace)
626 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
627
628 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
629 seq_show_option(m, "recover_session", "clean");
630
631 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
632 seq_printf(m, ",wsize=%d", fsopt->wsize);
633 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
634 seq_printf(m, ",rsize=%d", fsopt->rsize);
635 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
636 seq_printf(m, ",rasize=%d", fsopt->rasize);
637 if (fsopt->congestion_kb != default_congestion_kb())
638 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
639 if (fsopt->caps_max)
640 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
641 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
642 seq_printf(m, ",caps_wanted_delay_min=%d",
643 fsopt->caps_wanted_delay_min);
644 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
645 seq_printf(m, ",caps_wanted_delay_max=%d",
646 fsopt->caps_wanted_delay_max);
647 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
648 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
649 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
650 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
651 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
652 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
653
654 return 0;
655 }
656
657 /*
658 * handle any mon messages the standard library doesn't understand.
659 * return error if we don't either.
660 */
extra_mon_dispatch(struct ceph_client * client,struct ceph_msg * msg)661 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
662 {
663 struct ceph_fs_client *fsc = client->private;
664 int type = le16_to_cpu(msg->hdr.type);
665
666 switch (type) {
667 case CEPH_MSG_MDS_MAP:
668 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
669 return 0;
670 case CEPH_MSG_FS_MAP_USER:
671 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
672 return 0;
673 default:
674 return -1;
675 }
676 }
677
678 /*
679 * create a new fs client
680 *
681 * Success or not, this function consumes @fsopt and @opt.
682 */
create_fs_client(struct ceph_mount_options * fsopt,struct ceph_options * opt)683 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
684 struct ceph_options *opt)
685 {
686 struct ceph_fs_client *fsc;
687 int page_count;
688 size_t size;
689 int err;
690
691 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
692 if (!fsc) {
693 err = -ENOMEM;
694 goto fail;
695 }
696
697 fsc->client = ceph_create_client(opt, fsc);
698 if (IS_ERR(fsc->client)) {
699 err = PTR_ERR(fsc->client);
700 goto fail;
701 }
702 opt = NULL; /* fsc->client now owns this */
703
704 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
705 ceph_set_opt(fsc->client, ABORT_ON_FULL);
706
707 if (!fsopt->mds_namespace) {
708 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
709 0, true);
710 } else {
711 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
712 0, false);
713 }
714
715 fsc->mount_options = fsopt;
716
717 fsc->sb = NULL;
718 fsc->mount_state = CEPH_MOUNT_MOUNTING;
719 fsc->filp_gen = 1;
720
721 atomic_long_set(&fsc->writeback_count, 0);
722
723 err = -ENOMEM;
724 /*
725 * The number of concurrent works can be high but they don't need
726 * to be processed in parallel, limit concurrency.
727 */
728 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
729 if (!fsc->inode_wq)
730 goto fail_client;
731 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
732 if (!fsc->cap_wq)
733 goto fail_inode_wq;
734
735 /* set up mempools */
736 err = -ENOMEM;
737 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
738 size = sizeof (struct page *) * (page_count ? page_count : 1);
739 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
740 if (!fsc->wb_pagevec_pool)
741 goto fail_cap_wq;
742
743 return fsc;
744
745 fail_cap_wq:
746 destroy_workqueue(fsc->cap_wq);
747 fail_inode_wq:
748 destroy_workqueue(fsc->inode_wq);
749 fail_client:
750 ceph_destroy_client(fsc->client);
751 fail:
752 kfree(fsc);
753 if (opt)
754 ceph_destroy_options(opt);
755 destroy_mount_options(fsopt);
756 return ERR_PTR(err);
757 }
758
flush_fs_workqueues(struct ceph_fs_client * fsc)759 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
760 {
761 flush_workqueue(fsc->inode_wq);
762 flush_workqueue(fsc->cap_wq);
763 }
764
destroy_fs_client(struct ceph_fs_client * fsc)765 static void destroy_fs_client(struct ceph_fs_client *fsc)
766 {
767 dout("destroy_fs_client %p\n", fsc);
768
769 ceph_mdsc_destroy(fsc);
770 destroy_workqueue(fsc->inode_wq);
771 destroy_workqueue(fsc->cap_wq);
772
773 mempool_destroy(fsc->wb_pagevec_pool);
774
775 destroy_mount_options(fsc->mount_options);
776
777 ceph_destroy_client(fsc->client);
778
779 kfree(fsc);
780 dout("destroy_fs_client %p done\n", fsc);
781 }
782
783 /*
784 * caches
785 */
786 struct kmem_cache *ceph_inode_cachep;
787 struct kmem_cache *ceph_cap_cachep;
788 struct kmem_cache *ceph_cap_flush_cachep;
789 struct kmem_cache *ceph_dentry_cachep;
790 struct kmem_cache *ceph_file_cachep;
791 struct kmem_cache *ceph_dir_file_cachep;
792
ceph_inode_init_once(void * foo)793 static void ceph_inode_init_once(void *foo)
794 {
795 struct ceph_inode_info *ci = foo;
796 inode_init_once(&ci->vfs_inode);
797 }
798
init_caches(void)799 static int __init init_caches(void)
800 {
801 int error = -ENOMEM;
802
803 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
804 sizeof(struct ceph_inode_info),
805 __alignof__(struct ceph_inode_info),
806 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
807 SLAB_ACCOUNT, ceph_inode_init_once);
808 if (!ceph_inode_cachep)
809 return -ENOMEM;
810
811 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
812 if (!ceph_cap_cachep)
813 goto bad_cap;
814 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
815 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
816 if (!ceph_cap_flush_cachep)
817 goto bad_cap_flush;
818
819 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
820 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
821 if (!ceph_dentry_cachep)
822 goto bad_dentry;
823
824 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
825 if (!ceph_file_cachep)
826 goto bad_file;
827
828 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
829 if (!ceph_dir_file_cachep)
830 goto bad_dir_file;
831
832 error = ceph_fscache_register();
833 if (error)
834 goto bad_fscache;
835
836 return 0;
837
838 bad_fscache:
839 kmem_cache_destroy(ceph_dir_file_cachep);
840 bad_dir_file:
841 kmem_cache_destroy(ceph_file_cachep);
842 bad_file:
843 kmem_cache_destroy(ceph_dentry_cachep);
844 bad_dentry:
845 kmem_cache_destroy(ceph_cap_flush_cachep);
846 bad_cap_flush:
847 kmem_cache_destroy(ceph_cap_cachep);
848 bad_cap:
849 kmem_cache_destroy(ceph_inode_cachep);
850 return error;
851 }
852
destroy_caches(void)853 static void destroy_caches(void)
854 {
855 /*
856 * Make sure all delayed rcu free inodes are flushed before we
857 * destroy cache.
858 */
859 rcu_barrier();
860
861 kmem_cache_destroy(ceph_inode_cachep);
862 kmem_cache_destroy(ceph_cap_cachep);
863 kmem_cache_destroy(ceph_cap_flush_cachep);
864 kmem_cache_destroy(ceph_dentry_cachep);
865 kmem_cache_destroy(ceph_file_cachep);
866 kmem_cache_destroy(ceph_dir_file_cachep);
867
868 ceph_fscache_unregister();
869 }
870
871 /*
872 * ceph_umount_begin - initiate forced umount. Tear down down the
873 * mount, skipping steps that may hang while waiting for server(s).
874 */
ceph_umount_begin(struct super_block * sb)875 static void ceph_umount_begin(struct super_block *sb)
876 {
877 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
878
879 dout("ceph_umount_begin - starting forced umount\n");
880 if (!fsc)
881 return;
882 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
883 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
884 ceph_mdsc_force_umount(fsc->mdsc);
885 fsc->filp_gen++; // invalidate open files
886 }
887
ceph_remount(struct super_block * sb,int * flags,char * data)888 static int ceph_remount(struct super_block *sb, int *flags, char *data)
889 {
890 sync_filesystem(sb);
891 return 0;
892 }
893
894 static const struct super_operations ceph_super_ops = {
895 .alloc_inode = ceph_alloc_inode,
896 .free_inode = ceph_free_inode,
897 .write_inode = ceph_write_inode,
898 .drop_inode = generic_delete_inode,
899 .evict_inode = ceph_evict_inode,
900 .sync_fs = ceph_sync_fs,
901 .put_super = ceph_put_super,
902 .remount_fs = ceph_remount,
903 .show_options = ceph_show_options,
904 .statfs = ceph_statfs,
905 .umount_begin = ceph_umount_begin,
906 };
907
908 /*
909 * Bootstrap mount by opening the root directory. Note the mount
910 * @started time from caller, and time out if this takes too long.
911 */
open_root_dentry(struct ceph_fs_client * fsc,const char * path,unsigned long started)912 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
913 const char *path,
914 unsigned long started)
915 {
916 struct ceph_mds_client *mdsc = fsc->mdsc;
917 struct ceph_mds_request *req = NULL;
918 int err;
919 struct dentry *root;
920
921 /* open dir */
922 dout("open_root_inode opening '%s'\n", path);
923 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
924 if (IS_ERR(req))
925 return ERR_CAST(req);
926 req->r_path1 = kstrdup(path, GFP_NOFS);
927 if (!req->r_path1) {
928 root = ERR_PTR(-ENOMEM);
929 goto out;
930 }
931
932 req->r_ino1.ino = CEPH_INO_ROOT;
933 req->r_ino1.snap = CEPH_NOSNAP;
934 req->r_started = started;
935 req->r_timeout = fsc->client->options->mount_timeout;
936 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
937 req->r_num_caps = 2;
938 err = ceph_mdsc_do_request(mdsc, NULL, req);
939 if (err == 0) {
940 struct inode *inode = req->r_target_inode;
941 req->r_target_inode = NULL;
942 dout("open_root_inode success\n");
943 root = d_make_root(inode);
944 if (!root) {
945 root = ERR_PTR(-ENOMEM);
946 goto out;
947 }
948 dout("open_root_inode success, root dentry is %p\n", root);
949 } else {
950 root = ERR_PTR(err);
951 }
952 out:
953 ceph_mdsc_put_request(req);
954 return root;
955 }
956
957 /*
958 * mount: join the ceph cluster, and open root directory.
959 */
ceph_real_mount(struct ceph_fs_client * fsc)960 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc)
961 {
962 int err;
963 unsigned long started = jiffies; /* note the start time */
964 struct dentry *root;
965
966 dout("mount start %p\n", fsc);
967 mutex_lock(&fsc->client->mount_mutex);
968
969 if (!fsc->sb->s_root) {
970 const char *path = fsc->mount_options->server_path ?
971 fsc->mount_options->server_path + 1 : "";
972
973 err = __ceph_open_session(fsc->client, started);
974 if (err < 0)
975 goto out;
976
977 /* setup fscache */
978 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
979 err = ceph_fscache_register_fs(fsc);
980 if (err < 0)
981 goto out;
982 }
983
984 dout("mount opening path '%s'\n", path);
985
986 ceph_fs_debugfs_init(fsc);
987
988 root = open_root_dentry(fsc, path, started);
989 if (IS_ERR(root)) {
990 err = PTR_ERR(root);
991 goto out;
992 }
993 fsc->sb->s_root = dget(root);
994 } else {
995 root = dget(fsc->sb->s_root);
996 }
997
998 fsc->mount_state = CEPH_MOUNT_MOUNTED;
999 dout("mount success\n");
1000 mutex_unlock(&fsc->client->mount_mutex);
1001 return root;
1002
1003 out:
1004 mutex_unlock(&fsc->client->mount_mutex);
1005 return ERR_PTR(err);
1006 }
1007
ceph_set_super(struct super_block * s,void * data)1008 static int ceph_set_super(struct super_block *s, void *data)
1009 {
1010 struct ceph_fs_client *fsc = data;
1011 int ret;
1012
1013 dout("set_super %p data %p\n", s, data);
1014
1015 s->s_flags = fsc->mount_options->sb_flags;
1016 s->s_maxbytes = MAX_LFS_FILESIZE;
1017
1018 s->s_xattr = ceph_xattr_handlers;
1019 s->s_fs_info = fsc;
1020 fsc->sb = s;
1021 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1022
1023 s->s_op = &ceph_super_ops;
1024 s->s_d_op = &ceph_dentry_ops;
1025 s->s_export_op = &ceph_export_ops;
1026
1027 s->s_time_gran = 1;
1028 s->s_time_min = 0;
1029 s->s_time_max = U32_MAX;
1030
1031 ret = set_anon_super(s, NULL); /* what is that second arg for? */
1032 if (ret != 0)
1033 goto fail;
1034
1035 return ret;
1036
1037 fail:
1038 s->s_fs_info = NULL;
1039 fsc->sb = NULL;
1040 return ret;
1041 }
1042
1043 /*
1044 * share superblock if same fs AND options
1045 */
ceph_compare_super(struct super_block * sb,void * data)1046 static int ceph_compare_super(struct super_block *sb, void *data)
1047 {
1048 struct ceph_fs_client *new = data;
1049 struct ceph_mount_options *fsopt = new->mount_options;
1050 struct ceph_options *opt = new->client->options;
1051 struct ceph_fs_client *other = ceph_sb_to_client(sb);
1052
1053 dout("ceph_compare_super %p\n", sb);
1054
1055 if (compare_mount_options(fsopt, opt, other)) {
1056 dout("monitor(s)/mount options don't match\n");
1057 return 0;
1058 }
1059 if ((opt->flags & CEPH_OPT_FSID) &&
1060 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1061 dout("fsid doesn't match\n");
1062 return 0;
1063 }
1064 if (fsopt->sb_flags != other->mount_options->sb_flags) {
1065 dout("flags differ\n");
1066 return 0;
1067 }
1068 return 1;
1069 }
1070
1071 /*
1072 * construct our own bdi so we can control readahead, etc.
1073 */
1074 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1075
ceph_setup_bdi(struct super_block * sb,struct ceph_fs_client * fsc)1076 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1077 {
1078 int err;
1079
1080 err = super_setup_bdi_name(sb, "ceph-%ld",
1081 atomic_long_inc_return(&bdi_seq));
1082 if (err)
1083 return err;
1084
1085 /* set ra_pages based on rasize mount option? */
1086 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1087
1088 /* set io_pages based on max osd read size */
1089 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1090
1091 return 0;
1092 }
1093
ceph_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1094 static struct dentry *ceph_mount(struct file_system_type *fs_type,
1095 int flags, const char *dev_name, void *data)
1096 {
1097 struct super_block *sb;
1098 struct ceph_fs_client *fsc;
1099 struct dentry *res;
1100 int err;
1101 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
1102 struct ceph_mount_options *fsopt = NULL;
1103 struct ceph_options *opt = NULL;
1104
1105 dout("ceph_mount\n");
1106
1107 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1108 flags |= SB_POSIXACL;
1109 #endif
1110 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name);
1111 if (err < 0) {
1112 res = ERR_PTR(err);
1113 goto out_final;
1114 }
1115
1116 /* create client (which we may/may not use) */
1117 fsc = create_fs_client(fsopt, opt);
1118 if (IS_ERR(fsc)) {
1119 res = ERR_CAST(fsc);
1120 goto out_final;
1121 }
1122
1123 err = ceph_mdsc_init(fsc);
1124 if (err < 0) {
1125 res = ERR_PTR(err);
1126 goto out;
1127 }
1128
1129 if (ceph_test_opt(fsc->client, NOSHARE))
1130 compare_super = NULL;
1131 sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc);
1132 if (IS_ERR(sb)) {
1133 res = ERR_CAST(sb);
1134 goto out;
1135 }
1136
1137 if (ceph_sb_to_client(sb) != fsc) {
1138 destroy_fs_client(fsc);
1139 fsc = ceph_sb_to_client(sb);
1140 dout("get_sb got existing client %p\n", fsc);
1141 } else {
1142 dout("get_sb using new client %p\n", fsc);
1143 err = ceph_setup_bdi(sb, fsc);
1144 if (err < 0) {
1145 res = ERR_PTR(err);
1146 goto out_splat;
1147 }
1148 }
1149
1150 res = ceph_real_mount(fsc);
1151 if (IS_ERR(res))
1152 goto out_splat;
1153 dout("root %p inode %p ino %llx.%llx\n", res,
1154 d_inode(res), ceph_vinop(d_inode(res)));
1155 return res;
1156
1157 out_splat:
1158 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1159 pr_info("No mds server is up or the cluster is laggy\n");
1160 err = -EHOSTUNREACH;
1161 }
1162
1163 ceph_mdsc_close_sessions(fsc->mdsc);
1164 deactivate_locked_super(sb);
1165 goto out_final;
1166
1167 out:
1168 destroy_fs_client(fsc);
1169 out_final:
1170 dout("ceph_mount fail %ld\n", PTR_ERR(res));
1171 return res;
1172 }
1173
ceph_kill_sb(struct super_block * s)1174 static void ceph_kill_sb(struct super_block *s)
1175 {
1176 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1177
1178 dout("kill_sb %p\n", s);
1179
1180 ceph_mdsc_pre_umount(fsc->mdsc);
1181 flush_fs_workqueues(fsc);
1182
1183 /*
1184 * Though the kill_anon_super() will finally trigger the
1185 * sync_filesystem() anyway, we still need to do it here
1186 * and then bump the stage of shutdown to stop the work
1187 * queue as earlier as possible.
1188 */
1189 sync_filesystem(s);
1190
1191 fsc->mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1192
1193 kill_anon_super(s);
1194
1195 fsc->client->extra_mon_dispatch = NULL;
1196 ceph_fs_debugfs_cleanup(fsc);
1197
1198 ceph_fscache_unregister_fs(fsc);
1199
1200 destroy_fs_client(fsc);
1201 }
1202
1203 static struct file_system_type ceph_fs_type = {
1204 .owner = THIS_MODULE,
1205 .name = "ceph",
1206 .mount = ceph_mount,
1207 .kill_sb = ceph_kill_sb,
1208 .fs_flags = FS_RENAME_DOES_D_MOVE,
1209 };
1210 MODULE_ALIAS_FS("ceph");
1211
ceph_force_reconnect(struct super_block * sb)1212 int ceph_force_reconnect(struct super_block *sb)
1213 {
1214 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1215 int err = 0;
1216
1217 ceph_umount_begin(sb);
1218
1219 /* Make sure all page caches get invalidated.
1220 * see remove_session_caps_cb() */
1221 flush_workqueue(fsc->inode_wq);
1222
1223 /* In case that we were blacklisted. This also reset
1224 * all mon/osd connections */
1225 ceph_reset_client_addr(fsc->client);
1226
1227 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1228
1229 fsc->blacklisted = false;
1230 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1231
1232 if (sb->s_root) {
1233 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1234 CEPH_STAT_CAP_INODE, true);
1235 }
1236 return err;
1237 }
1238
init_ceph(void)1239 static int __init init_ceph(void)
1240 {
1241 int ret = init_caches();
1242 if (ret)
1243 goto out;
1244
1245 ceph_flock_init();
1246 ret = register_filesystem(&ceph_fs_type);
1247 if (ret)
1248 goto out_caches;
1249
1250 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1251
1252 return 0;
1253
1254 out_caches:
1255 destroy_caches();
1256 out:
1257 return ret;
1258 }
1259
exit_ceph(void)1260 static void __exit exit_ceph(void)
1261 {
1262 dout("exit_ceph\n");
1263 unregister_filesystem(&ceph_fs_type);
1264 destroy_caches();
1265 }
1266
1267 module_init(init_ceph);
1268 module_exit(exit_ceph);
1269
1270 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1271 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1272 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1273 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1274 MODULE_LICENSE("GPL");
1275