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/fs_context.h>
13 #include <linux/fs_parser.h>
14 #include <linux/sched.h>
15 #include <linux/seq_file.h>
16 #include <linux/slab.h>
17 #include <linux/statfs.h>
18 #include <linux/string.h>
19
20 #include "super.h"
21 #include "mds_client.h"
22 #include "cache.h"
23 #include "crypto.h"
24
25 #include <linux/ceph/ceph_features.h>
26 #include <linux/ceph/decode.h>
27 #include <linux/ceph/mon_client.h>
28 #include <linux/ceph/auth.h>
29 #include <linux/ceph/debugfs.h>
30
31 #include <uapi/linux/magic.h>
32
33 static DEFINE_SPINLOCK(ceph_fsc_lock);
34 static LIST_HEAD(ceph_fsc_list);
35
36 /*
37 * Ceph superblock operations
38 *
39 * Handle the basics of mounting, unmounting.
40 */
41
42 /*
43 * super ops
44 */
ceph_put_super(struct super_block * s)45 static void ceph_put_super(struct super_block *s)
46 {
47 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
48
49 doutc(fsc->client, "begin\n");
50 ceph_fscrypt_free_dummy_policy(fsc);
51 ceph_mdsc_close_sessions(fsc->mdsc);
52 doutc(fsc->client, "done\n");
53 }
54
ceph_statfs(struct dentry * dentry,struct kstatfs * buf)55 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
56 {
57 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(d_inode(dentry));
58 struct ceph_mon_client *monc = &fsc->client->monc;
59 struct ceph_statfs st;
60 int i, err;
61 u64 data_pool;
62
63 doutc(fsc->client, "begin\n");
64 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
65 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
66 } else {
67 data_pool = CEPH_NOPOOL;
68 }
69
70 err = ceph_monc_do_statfs(monc, data_pool, &st);
71 if (err < 0)
72 return err;
73
74 /* fill in kstatfs */
75 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
76
77 /*
78 * Express utilization in terms of large blocks to avoid
79 * overflow on 32-bit machines.
80 */
81 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
82
83 /*
84 * By default use root quota for stats; fallback to overall filesystem
85 * usage if using 'noquotadf' mount option or if the root dir doesn't
86 * have max_bytes quota set.
87 */
88 if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
89 !ceph_quota_update_statfs(fsc, buf)) {
90 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
91 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
92 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
93 }
94
95 /*
96 * NOTE: for the time being, we make bsize == frsize to humor
97 * not-yet-ancient versions of glibc that are broken.
98 * Someday, we will probably want to report a real block
99 * size... whatever that may mean for a network file system!
100 */
101 buf->f_bsize = buf->f_frsize;
102
103 buf->f_files = le64_to_cpu(st.num_objects);
104 buf->f_ffree = -1;
105 buf->f_namelen = NAME_MAX;
106
107 /* Must convert the fsid, for consistent values across arches */
108 buf->f_fsid.val[0] = 0;
109 mutex_lock(&monc->mutex);
110 for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
111 buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
112 mutex_unlock(&monc->mutex);
113
114 /* fold the fs_cluster_id into the upper bits */
115 buf->f_fsid.val[1] = monc->fs_cluster_id;
116
117 doutc(fsc->client, "done\n");
118 return 0;
119 }
120
ceph_sync_fs(struct super_block * sb,int wait)121 static int ceph_sync_fs(struct super_block *sb, int wait)
122 {
123 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
124 struct ceph_client *cl = fsc->client;
125
126 if (!wait) {
127 doutc(cl, "(non-blocking)\n");
128 ceph_flush_dirty_caps(fsc->mdsc);
129 ceph_flush_cap_releases(fsc->mdsc);
130 doutc(cl, "(non-blocking) done\n");
131 return 0;
132 }
133
134 doutc(cl, "(blocking)\n");
135 ceph_osdc_sync(&fsc->client->osdc);
136 ceph_mdsc_sync(fsc->mdsc);
137 doutc(cl, "(blocking) done\n");
138 return 0;
139 }
140
141 /*
142 * mount options
143 */
144 enum {
145 Opt_wsize,
146 Opt_rsize,
147 Opt_rasize,
148 Opt_caps_wanted_delay_min,
149 Opt_caps_wanted_delay_max,
150 Opt_caps_max,
151 Opt_readdir_max_entries,
152 Opt_readdir_max_bytes,
153 Opt_congestion_kb,
154 /* int args above */
155 Opt_snapdirname,
156 Opt_mds_namespace,
157 Opt_recover_session,
158 Opt_source,
159 Opt_mon_addr,
160 Opt_test_dummy_encryption,
161 /* string args above */
162 Opt_dirstat,
163 Opt_rbytes,
164 Opt_asyncreaddir,
165 Opt_dcache,
166 Opt_ino32,
167 Opt_fscache,
168 Opt_poolperm,
169 Opt_require_active_mds,
170 Opt_acl,
171 Opt_quotadf,
172 Opt_copyfrom,
173 Opt_wsync,
174 Opt_pagecache,
175 Opt_sparseread,
176 };
177
178 enum ceph_recover_session_mode {
179 ceph_recover_session_no,
180 ceph_recover_session_clean
181 };
182
183 static const struct constant_table ceph_param_recover[] = {
184 { "no", ceph_recover_session_no },
185 { "clean", ceph_recover_session_clean },
186 {}
187 };
188
189 static const struct fs_parameter_spec ceph_mount_parameters[] = {
190 fsparam_flag_no ("acl", Opt_acl),
191 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir),
192 fsparam_s32 ("caps_max", Opt_caps_max),
193 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max),
194 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min),
195 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb),
196 fsparam_flag_no ("copyfrom", Opt_copyfrom),
197 fsparam_flag_no ("dcache", Opt_dcache),
198 fsparam_flag_no ("dirstat", Opt_dirstat),
199 fsparam_flag_no ("fsc", Opt_fscache), // fsc|nofsc
200 fsparam_string ("fsc", Opt_fscache), // fsc=...
201 fsparam_flag_no ("ino32", Opt_ino32),
202 fsparam_string ("mds_namespace", Opt_mds_namespace),
203 fsparam_string ("mon_addr", Opt_mon_addr),
204 fsparam_flag_no ("poolperm", Opt_poolperm),
205 fsparam_flag_no ("quotadf", Opt_quotadf),
206 fsparam_u32 ("rasize", Opt_rasize),
207 fsparam_flag_no ("rbytes", Opt_rbytes),
208 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes),
209 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries),
210 fsparam_enum ("recover_session", Opt_recover_session, ceph_param_recover),
211 fsparam_flag_no ("require_active_mds", Opt_require_active_mds),
212 fsparam_u32 ("rsize", Opt_rsize),
213 fsparam_string ("snapdirname", Opt_snapdirname),
214 fsparam_string ("source", Opt_source),
215 fsparam_flag ("test_dummy_encryption", Opt_test_dummy_encryption),
216 fsparam_string ("test_dummy_encryption", Opt_test_dummy_encryption),
217 fsparam_u32 ("wsize", Opt_wsize),
218 fsparam_flag_no ("wsync", Opt_wsync),
219 fsparam_flag_no ("pagecache", Opt_pagecache),
220 fsparam_flag_no ("sparseread", Opt_sparseread),
221 {}
222 };
223
224 struct ceph_parse_opts_ctx {
225 struct ceph_options *copts;
226 struct ceph_mount_options *opts;
227 };
228
229 /*
230 * Remove adjacent slashes and then the trailing slash, unless it is
231 * the only remaining character.
232 *
233 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
234 */
canonicalize_path(char * path)235 static void canonicalize_path(char *path)
236 {
237 int i, j = 0;
238
239 for (i = 0; path[i] != '\0'; i++) {
240 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
241 path[j++] = path[i];
242 }
243
244 if (j > 1 && path[j - 1] == '/')
245 j--;
246 path[j] = '\0';
247 }
248
ceph_parse_old_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)249 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
250 struct fs_context *fc)
251 {
252 int r;
253 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
254 struct ceph_mount_options *fsopt = pctx->opts;
255
256 if (*dev_name_end != ':')
257 return invalfc(fc, "separator ':' missing in source");
258
259 r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
260 pctx->copts, fc->log.log, ',');
261 if (r)
262 return r;
263
264 fsopt->new_dev_syntax = false;
265 return 0;
266 }
267
ceph_parse_new_source(const char * dev_name,const char * dev_name_end,struct fs_context * fc)268 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
269 struct fs_context *fc)
270 {
271 size_t len;
272 struct ceph_fsid fsid;
273 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
274 struct ceph_options *opts = pctx->copts;
275 struct ceph_mount_options *fsopt = pctx->opts;
276 const char *name_start = dev_name;
277 const char *fsid_start, *fs_name_start;
278
279 if (*dev_name_end != '=') {
280 dout("separator '=' missing in source");
281 return -EINVAL;
282 }
283
284 fsid_start = strchr(dev_name, '@');
285 if (!fsid_start)
286 return invalfc(fc, "missing cluster fsid");
287 len = fsid_start - name_start;
288 kfree(opts->name);
289 opts->name = kstrndup(name_start, len, GFP_KERNEL);
290 if (!opts->name)
291 return -ENOMEM;
292 dout("using %s entity name", opts->name);
293
294 ++fsid_start; /* start of cluster fsid */
295 fs_name_start = strchr(fsid_start, '.');
296 if (!fs_name_start)
297 return invalfc(fc, "missing file system name");
298
299 if (ceph_parse_fsid(fsid_start, &fsid))
300 return invalfc(fc, "Invalid FSID");
301
302 ++fs_name_start; /* start of file system name */
303 len = dev_name_end - fs_name_start;
304
305 if (!namespace_equals(fsopt, fs_name_start, len))
306 return invalfc(fc, "Mismatching mds_namespace");
307 kfree(fsopt->mds_namespace);
308 fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
309 if (!fsopt->mds_namespace)
310 return -ENOMEM;
311 dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
312
313 fsopt->new_dev_syntax = true;
314 return 0;
315 }
316
317 /*
318 * Parse the source parameter for new device format. Distinguish the device
319 * spec from the path. Try parsing new device format and fallback to old
320 * format if needed.
321 *
322 * New device syntax will looks like:
323 * <device_spec>=/<path>
324 * where
325 * <device_spec> is name@fsid.fsname
326 * <path> is optional, but if present must begin with '/'
327 * (monitor addresses are passed via mount option)
328 *
329 * Old device syntax is:
330 * <server_spec>[,<server_spec>...]:[<path>]
331 * where
332 * <server_spec> is <ip>[:<port>]
333 * <path> is optional, but if present must begin with '/'
334 */
ceph_parse_source(struct fs_parameter * param,struct fs_context * fc)335 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
336 {
337 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
338 struct ceph_mount_options *fsopt = pctx->opts;
339 char *dev_name = param->string, *dev_name_end;
340 int ret;
341
342 dout("'%s'\n", dev_name);
343 if (!dev_name || !*dev_name)
344 return invalfc(fc, "Empty source");
345
346 dev_name_end = strchr(dev_name, '/');
347 if (dev_name_end) {
348 /*
349 * The server_path will include the whole chars from userland
350 * including the leading '/'.
351 */
352 kfree(fsopt->server_path);
353 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
354 if (!fsopt->server_path)
355 return -ENOMEM;
356
357 canonicalize_path(fsopt->server_path);
358 } else {
359 dev_name_end = dev_name + strlen(dev_name);
360 }
361
362 dev_name_end--; /* back up to separator */
363 if (dev_name_end < dev_name)
364 return invalfc(fc, "Path missing in source");
365
366 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
367 if (fsopt->server_path)
368 dout("server path '%s'\n", fsopt->server_path);
369
370 dout("trying new device syntax");
371 ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
372 if (ret) {
373 if (ret != -EINVAL)
374 return ret;
375 dout("trying old device syntax");
376 ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
377 if (ret)
378 return ret;
379 }
380
381 fc->source = param->string;
382 param->string = NULL;
383 return 0;
384 }
385
ceph_parse_mon_addr(struct fs_parameter * param,struct fs_context * fc)386 static int ceph_parse_mon_addr(struct fs_parameter *param,
387 struct fs_context *fc)
388 {
389 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
390 struct ceph_mount_options *fsopt = pctx->opts;
391
392 kfree(fsopt->mon_addr);
393 fsopt->mon_addr = param->string;
394 param->string = NULL;
395
396 return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
397 pctx->copts, fc->log.log, '/');
398 }
399
ceph_parse_mount_param(struct fs_context * fc,struct fs_parameter * param)400 static int ceph_parse_mount_param(struct fs_context *fc,
401 struct fs_parameter *param)
402 {
403 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
404 struct ceph_mount_options *fsopt = pctx->opts;
405 struct fs_parse_result result;
406 unsigned int mode;
407 int token, ret;
408
409 ret = ceph_parse_param(param, pctx->copts, fc->log.log);
410 if (ret != -ENOPARAM)
411 return ret;
412
413 token = fs_parse(fc, ceph_mount_parameters, param, &result);
414 dout("%s: fs_parse '%s' token %d\n",__func__, param->key, token);
415 if (token < 0)
416 return token;
417
418 switch (token) {
419 case Opt_snapdirname:
420 if (strlen(param->string) > NAME_MAX)
421 return invalfc(fc, "snapdirname too long");
422 kfree(fsopt->snapdir_name);
423 fsopt->snapdir_name = param->string;
424 param->string = NULL;
425 break;
426 case Opt_mds_namespace:
427 if (!namespace_equals(fsopt, param->string, strlen(param->string)))
428 return invalfc(fc, "Mismatching mds_namespace");
429 kfree(fsopt->mds_namespace);
430 fsopt->mds_namespace = param->string;
431 param->string = NULL;
432 break;
433 case Opt_recover_session:
434 mode = result.uint_32;
435 if (mode == ceph_recover_session_no)
436 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
437 else if (mode == ceph_recover_session_clean)
438 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
439 else
440 BUG();
441 break;
442 case Opt_source:
443 if (fc->source)
444 return invalfc(fc, "Multiple sources specified");
445 return ceph_parse_source(param, fc);
446 case Opt_mon_addr:
447 return ceph_parse_mon_addr(param, fc);
448 case Opt_wsize:
449 if (result.uint_32 < PAGE_SIZE ||
450 result.uint_32 > CEPH_MAX_WRITE_SIZE)
451 goto out_of_range;
452 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
453 break;
454 case Opt_rsize:
455 if (result.uint_32 < PAGE_SIZE ||
456 result.uint_32 > CEPH_MAX_READ_SIZE)
457 goto out_of_range;
458 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
459 break;
460 case Opt_rasize:
461 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
462 break;
463 case Opt_caps_wanted_delay_min:
464 if (result.uint_32 < 1)
465 goto out_of_range;
466 fsopt->caps_wanted_delay_min = result.uint_32;
467 break;
468 case Opt_caps_wanted_delay_max:
469 if (result.uint_32 < 1)
470 goto out_of_range;
471 fsopt->caps_wanted_delay_max = result.uint_32;
472 break;
473 case Opt_caps_max:
474 if (result.int_32 < 0)
475 goto out_of_range;
476 fsopt->caps_max = result.int_32;
477 break;
478 case Opt_readdir_max_entries:
479 if (result.uint_32 < 1)
480 goto out_of_range;
481 fsopt->max_readdir = result.uint_32;
482 break;
483 case Opt_readdir_max_bytes:
484 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
485 goto out_of_range;
486 fsopt->max_readdir_bytes = result.uint_32;
487 break;
488 case Opt_congestion_kb:
489 if (result.uint_32 < 1024) /* at least 1M */
490 goto out_of_range;
491 fsopt->congestion_kb = result.uint_32;
492 break;
493 case Opt_dirstat:
494 if (!result.negated)
495 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
496 else
497 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
498 break;
499 case Opt_rbytes:
500 if (!result.negated)
501 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
502 else
503 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
504 break;
505 case Opt_asyncreaddir:
506 if (!result.negated)
507 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
508 else
509 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
510 break;
511 case Opt_dcache:
512 if (!result.negated)
513 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
514 else
515 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
516 break;
517 case Opt_ino32:
518 if (!result.negated)
519 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
520 else
521 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
522 break;
523
524 case Opt_fscache:
525 #ifdef CONFIG_CEPH_FSCACHE
526 kfree(fsopt->fscache_uniq);
527 fsopt->fscache_uniq = NULL;
528 if (result.negated) {
529 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
530 } else {
531 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
532 fsopt->fscache_uniq = param->string;
533 param->string = NULL;
534 }
535 break;
536 #else
537 return invalfc(fc, "fscache support is disabled");
538 #endif
539 case Opt_poolperm:
540 if (!result.negated)
541 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
542 else
543 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
544 break;
545 case Opt_require_active_mds:
546 if (!result.negated)
547 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
548 else
549 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
550 break;
551 case Opt_quotadf:
552 if (!result.negated)
553 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
554 else
555 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
556 break;
557 case Opt_copyfrom:
558 if (!result.negated)
559 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
560 else
561 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
562 break;
563 case Opt_acl:
564 if (!result.negated) {
565 #ifdef CONFIG_CEPH_FS_POSIX_ACL
566 fc->sb_flags |= SB_POSIXACL;
567 #else
568 return invalfc(fc, "POSIX ACL support is disabled");
569 #endif
570 } else {
571 fc->sb_flags &= ~SB_POSIXACL;
572 }
573 break;
574 case Opt_wsync:
575 if (!result.negated)
576 fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
577 else
578 fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
579 break;
580 case Opt_pagecache:
581 if (result.negated)
582 fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
583 else
584 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
585 break;
586 case Opt_sparseread:
587 if (result.negated)
588 fsopt->flags &= ~CEPH_MOUNT_OPT_SPARSEREAD;
589 else
590 fsopt->flags |= CEPH_MOUNT_OPT_SPARSEREAD;
591 break;
592 case Opt_test_dummy_encryption:
593 #ifdef CONFIG_FS_ENCRYPTION
594 fscrypt_free_dummy_policy(&fsopt->dummy_enc_policy);
595 ret = fscrypt_parse_test_dummy_encryption(param,
596 &fsopt->dummy_enc_policy);
597 if (ret == -EINVAL) {
598 warnfc(fc, "Value of option \"%s\" is unrecognized",
599 param->key);
600 } else if (ret == -EEXIST) {
601 warnfc(fc, "Conflicting test_dummy_encryption options");
602 ret = -EINVAL;
603 }
604 #else
605 warnfc(fc,
606 "FS encryption not supported: test_dummy_encryption mount option ignored");
607 #endif
608 break;
609 default:
610 BUG();
611 }
612 return 0;
613
614 out_of_range:
615 return invalfc(fc, "%s out of range", param->key);
616 }
617
destroy_mount_options(struct ceph_mount_options * args)618 static void destroy_mount_options(struct ceph_mount_options *args)
619 {
620 dout("destroy_mount_options %p\n", args);
621 if (!args)
622 return;
623
624 kfree(args->snapdir_name);
625 kfree(args->mds_namespace);
626 kfree(args->server_path);
627 kfree(args->fscache_uniq);
628 kfree(args->mon_addr);
629 fscrypt_free_dummy_policy(&args->dummy_enc_policy);
630 kfree(args);
631 }
632
strcmp_null(const char * s1,const char * s2)633 static int strcmp_null(const char *s1, const char *s2)
634 {
635 if (!s1 && !s2)
636 return 0;
637 if (s1 && !s2)
638 return -1;
639 if (!s1 && s2)
640 return 1;
641 return strcmp(s1, s2);
642 }
643
compare_mount_options(struct ceph_mount_options * new_fsopt,struct ceph_options * new_opt,struct ceph_fs_client * fsc)644 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
645 struct ceph_options *new_opt,
646 struct ceph_fs_client *fsc)
647 {
648 struct ceph_mount_options *fsopt1 = new_fsopt;
649 struct ceph_mount_options *fsopt2 = fsc->mount_options;
650 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
651 int ret;
652
653 ret = memcmp(fsopt1, fsopt2, ofs);
654 if (ret)
655 return ret;
656
657 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
658 if (ret)
659 return ret;
660
661 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
662 if (ret)
663 return ret;
664
665 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
666 if (ret)
667 return ret;
668
669 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
670 if (ret)
671 return ret;
672
673 ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
674 if (ret)
675 return ret;
676
677 return ceph_compare_options(new_opt, fsc->client);
678 }
679
680 /**
681 * ceph_show_options - Show mount options in /proc/mounts
682 * @m: seq_file to write to
683 * @root: root of that (sub)tree
684 */
ceph_show_options(struct seq_file * m,struct dentry * root)685 static int ceph_show_options(struct seq_file *m, struct dentry *root)
686 {
687 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(root->d_sb);
688 struct ceph_mount_options *fsopt = fsc->mount_options;
689 size_t pos;
690 int ret;
691
692 /* a comma between MNT/MS and client options */
693 seq_putc(m, ',');
694 pos = m->count;
695
696 ret = ceph_print_client_options(m, fsc->client, false);
697 if (ret)
698 return ret;
699
700 /* retract our comma if no client options */
701 if (m->count == pos)
702 m->count--;
703
704 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
705 seq_puts(m, ",dirstat");
706 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
707 seq_puts(m, ",rbytes");
708 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
709 seq_puts(m, ",noasyncreaddir");
710 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
711 seq_puts(m, ",nodcache");
712 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
713 seq_puts(m, ",ino32");
714 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
715 seq_show_option(m, "fsc", fsopt->fscache_uniq);
716 }
717 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
718 seq_puts(m, ",nopoolperm");
719 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
720 seq_puts(m, ",noquotadf");
721
722 #ifdef CONFIG_CEPH_FS_POSIX_ACL
723 if (root->d_sb->s_flags & SB_POSIXACL)
724 seq_puts(m, ",acl");
725 else
726 seq_puts(m, ",noacl");
727 #endif
728
729 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
730 seq_puts(m, ",copyfrom");
731
732 /* dump mds_namespace when old device syntax is in use */
733 if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
734 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
735
736 if (fsopt->mon_addr)
737 seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
738
739 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
740 seq_show_option(m, "recover_session", "clean");
741
742 if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
743 seq_puts(m, ",wsync");
744 if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
745 seq_puts(m, ",nopagecache");
746 if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
747 seq_puts(m, ",sparseread");
748
749 fscrypt_show_test_dummy_encryption(m, ',', root->d_sb);
750
751 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
752 seq_printf(m, ",wsize=%u", fsopt->wsize);
753 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
754 seq_printf(m, ",rsize=%u", fsopt->rsize);
755 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
756 seq_printf(m, ",rasize=%u", fsopt->rasize);
757 if (fsopt->congestion_kb != default_congestion_kb())
758 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
759 if (fsopt->caps_max)
760 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
761 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
762 seq_printf(m, ",caps_wanted_delay_min=%u",
763 fsopt->caps_wanted_delay_min);
764 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
765 seq_printf(m, ",caps_wanted_delay_max=%u",
766 fsopt->caps_wanted_delay_max);
767 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
768 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
769 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
770 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
771 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
772 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
773
774 return 0;
775 }
776
777 /*
778 * handle any mon messages the standard library doesn't understand.
779 * return error if we don't either.
780 */
extra_mon_dispatch(struct ceph_client * client,struct ceph_msg * msg)781 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
782 {
783 struct ceph_fs_client *fsc = client->private;
784 int type = le16_to_cpu(msg->hdr.type);
785
786 switch (type) {
787 case CEPH_MSG_MDS_MAP:
788 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
789 return 0;
790 case CEPH_MSG_FS_MAP_USER:
791 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
792 return 0;
793 default:
794 return -1;
795 }
796 }
797
798 /*
799 * create a new fs client
800 *
801 * Success or not, this function consumes @fsopt and @opt.
802 */
create_fs_client(struct ceph_mount_options * fsopt,struct ceph_options * opt)803 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
804 struct ceph_options *opt)
805 {
806 struct ceph_fs_client *fsc;
807 int err;
808
809 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
810 if (!fsc) {
811 err = -ENOMEM;
812 goto fail;
813 }
814
815 fsc->client = ceph_create_client(opt, fsc);
816 if (IS_ERR(fsc->client)) {
817 err = PTR_ERR(fsc->client);
818 goto fail;
819 }
820 opt = NULL; /* fsc->client now owns this */
821
822 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
823 ceph_set_opt(fsc->client, ABORT_ON_FULL);
824
825 if (!fsopt->mds_namespace) {
826 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
827 0, true);
828 } else {
829 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
830 0, false);
831 }
832
833 fsc->mount_options = fsopt;
834
835 fsc->sb = NULL;
836 fsc->mount_state = CEPH_MOUNT_MOUNTING;
837 fsc->filp_gen = 1;
838 fsc->have_copy_from2 = true;
839
840 atomic_long_set(&fsc->writeback_count, 0);
841 fsc->write_congested = false;
842
843 err = -ENOMEM;
844 /*
845 * The number of concurrent works can be high but they don't need
846 * to be processed in parallel, limit concurrency.
847 */
848 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
849 if (!fsc->inode_wq)
850 goto fail_client;
851 fsc->cap_wq = alloc_workqueue("ceph-cap", WQ_PERCPU, 1);
852 if (!fsc->cap_wq)
853 goto fail_inode_wq;
854
855 hash_init(fsc->async_unlink_conflict);
856 spin_lock_init(&fsc->async_unlink_conflict_lock);
857
858 spin_lock(&ceph_fsc_lock);
859 list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
860 spin_unlock(&ceph_fsc_lock);
861
862 return fsc;
863
864 fail_inode_wq:
865 destroy_workqueue(fsc->inode_wq);
866 fail_client:
867 ceph_destroy_client(fsc->client);
868 fail:
869 kfree(fsc);
870 if (opt)
871 ceph_destroy_options(opt);
872 destroy_mount_options(fsopt);
873 return ERR_PTR(err);
874 }
875
flush_fs_workqueues(struct ceph_fs_client * fsc)876 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
877 {
878 flush_workqueue(fsc->inode_wq);
879 flush_workqueue(fsc->cap_wq);
880 }
881
destroy_fs_client(struct ceph_fs_client * fsc)882 static void destroy_fs_client(struct ceph_fs_client *fsc)
883 {
884 doutc(fsc->client, "%p\n", fsc);
885
886 spin_lock(&ceph_fsc_lock);
887 list_del(&fsc->metric_wakeup);
888 spin_unlock(&ceph_fsc_lock);
889
890 ceph_mdsc_destroy(fsc);
891 destroy_workqueue(fsc->inode_wq);
892 destroy_workqueue(fsc->cap_wq);
893
894 destroy_mount_options(fsc->mount_options);
895
896 ceph_destroy_client(fsc->client);
897
898 kfree(fsc);
899 dout("%s: %p done\n", __func__, fsc);
900 }
901
902 /*
903 * caches
904 */
905 struct kmem_cache *ceph_inode_cachep;
906 struct kmem_cache *ceph_cap_cachep;
907 struct kmem_cache *ceph_cap_snap_cachep;
908 struct kmem_cache *ceph_cap_flush_cachep;
909 struct kmem_cache *ceph_dentry_cachep;
910 struct kmem_cache *ceph_file_cachep;
911 struct kmem_cache *ceph_dir_file_cachep;
912 struct kmem_cache *ceph_mds_request_cachep;
913 mempool_t *ceph_wb_pagevec_pool;
914
ceph_inode_init_once(void * foo)915 static void ceph_inode_init_once(void *foo)
916 {
917 struct ceph_inode_info *ci = foo;
918 inode_init_once(&ci->netfs.inode);
919 }
920
init_caches(void)921 static int __init init_caches(void)
922 {
923 int error = -ENOMEM;
924
925 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
926 sizeof(struct ceph_inode_info),
927 __alignof__(struct ceph_inode_info),
928 SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
929 ceph_inode_init_once);
930 if (!ceph_inode_cachep)
931 return -ENOMEM;
932
933 ceph_cap_cachep = KMEM_CACHE(ceph_cap, 0);
934 if (!ceph_cap_cachep)
935 goto bad_cap;
936 ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, 0);
937 if (!ceph_cap_snap_cachep)
938 goto bad_cap_snap;
939 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
940 SLAB_RECLAIM_ACCOUNT);
941 if (!ceph_cap_flush_cachep)
942 goto bad_cap_flush;
943
944 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
945 SLAB_RECLAIM_ACCOUNT);
946 if (!ceph_dentry_cachep)
947 goto bad_dentry;
948
949 ceph_file_cachep = KMEM_CACHE(ceph_file_info, 0);
950 if (!ceph_file_cachep)
951 goto bad_file;
952
953 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, 0);
954 if (!ceph_dir_file_cachep)
955 goto bad_dir_file;
956
957 ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, 0);
958 if (!ceph_mds_request_cachep)
959 goto bad_mds_req;
960
961 ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10,
962 (CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT) * sizeof(struct page *));
963 if (!ceph_wb_pagevec_pool)
964 goto bad_pagevec_pool;
965
966 return 0;
967
968 bad_pagevec_pool:
969 kmem_cache_destroy(ceph_mds_request_cachep);
970 bad_mds_req:
971 kmem_cache_destroy(ceph_dir_file_cachep);
972 bad_dir_file:
973 kmem_cache_destroy(ceph_file_cachep);
974 bad_file:
975 kmem_cache_destroy(ceph_dentry_cachep);
976 bad_dentry:
977 kmem_cache_destroy(ceph_cap_flush_cachep);
978 bad_cap_flush:
979 kmem_cache_destroy(ceph_cap_snap_cachep);
980 bad_cap_snap:
981 kmem_cache_destroy(ceph_cap_cachep);
982 bad_cap:
983 kmem_cache_destroy(ceph_inode_cachep);
984 return error;
985 }
986
destroy_caches(void)987 static void destroy_caches(void)
988 {
989 /*
990 * Make sure all delayed rcu free inodes are flushed before we
991 * destroy cache.
992 */
993 rcu_barrier();
994
995 kmem_cache_destroy(ceph_inode_cachep);
996 kmem_cache_destroy(ceph_cap_cachep);
997 kmem_cache_destroy(ceph_cap_snap_cachep);
998 kmem_cache_destroy(ceph_cap_flush_cachep);
999 kmem_cache_destroy(ceph_dentry_cachep);
1000 kmem_cache_destroy(ceph_file_cachep);
1001 kmem_cache_destroy(ceph_dir_file_cachep);
1002 kmem_cache_destroy(ceph_mds_request_cachep);
1003 mempool_destroy(ceph_wb_pagevec_pool);
1004 }
1005
__ceph_umount_begin(struct ceph_fs_client * fsc)1006 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
1007 {
1008 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
1009 ceph_mdsc_force_umount(fsc->mdsc);
1010 fsc->filp_gen++; // invalidate open files
1011 }
1012
1013 /*
1014 * ceph_umount_begin - initiate forced umount. Tear down the
1015 * mount, skipping steps that may hang while waiting for server(s).
1016 */
ceph_umount_begin(struct super_block * sb)1017 void ceph_umount_begin(struct super_block *sb)
1018 {
1019 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1020
1021 doutc(fsc->client, "starting forced umount\n");
1022
1023 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
1024 __ceph_umount_begin(fsc);
1025 }
1026
1027 static const struct super_operations ceph_super_ops = {
1028 .alloc_inode = ceph_alloc_inode,
1029 .free_inode = ceph_free_inode,
1030 .write_inode = ceph_write_inode,
1031 .drop_inode = inode_just_drop,
1032 .evict_inode = ceph_evict_inode,
1033 .sync_fs = ceph_sync_fs,
1034 .put_super = ceph_put_super,
1035 .show_options = ceph_show_options,
1036 .statfs = ceph_statfs,
1037 .umount_begin = ceph_umount_begin,
1038 };
1039
1040 /*
1041 * Bootstrap mount by opening the root directory. Note the mount
1042 * @started time from caller, and time out if this takes too long.
1043 */
open_root_dentry(struct ceph_fs_client * fsc,const char * path,unsigned long started)1044 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1045 const char *path,
1046 unsigned long started)
1047 {
1048 struct ceph_client *cl = fsc->client;
1049 struct ceph_mds_client *mdsc = fsc->mdsc;
1050 struct ceph_mds_request *req = NULL;
1051 int err;
1052 struct dentry *root;
1053
1054 /* open dir */
1055 doutc(cl, "opening '%s'\n", path);
1056 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1057 if (IS_ERR(req))
1058 return ERR_CAST(req);
1059 req->r_path1 = kstrdup(path, GFP_NOFS);
1060 if (!req->r_path1) {
1061 root = ERR_PTR(-ENOMEM);
1062 goto out;
1063 }
1064
1065 req->r_ino1.ino = CEPH_INO_ROOT;
1066 req->r_ino1.snap = CEPH_NOSNAP;
1067 req->r_started = started;
1068 req->r_timeout = fsc->client->options->mount_timeout;
1069 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1070 req->r_num_caps = 2;
1071 err = ceph_mdsc_do_request(mdsc, NULL, req);
1072 if (err == 0) {
1073 struct inode *inode = req->r_target_inode;
1074 req->r_target_inode = NULL;
1075 doutc(cl, "success\n");
1076 root = d_make_root(inode);
1077 if (!root) {
1078 root = ERR_PTR(-ENOMEM);
1079 goto out;
1080 }
1081 doutc(cl, "success, root dentry is %p\n", root);
1082 } else {
1083 root = ERR_PTR(err);
1084 }
1085 out:
1086 ceph_mdsc_put_request(req);
1087 return root;
1088 }
1089
1090 #ifdef CONFIG_FS_ENCRYPTION
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1091 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1092 struct fs_context *fc,
1093 struct ceph_mount_options *fsopt)
1094 {
1095 struct ceph_fs_client *fsc = sb->s_fs_info;
1096
1097 if (!fscrypt_is_dummy_policy_set(&fsopt->dummy_enc_policy))
1098 return 0;
1099
1100 /* No changing encryption context on remount. */
1101 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE &&
1102 !fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1103 if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1104 &fsc->fsc_dummy_enc_policy))
1105 return 0;
1106 errorfc(fc, "Can't set test_dummy_encryption on remount");
1107 return -EINVAL;
1108 }
1109
1110 /* Also make sure fsopt doesn't contain a conflicting value. */
1111 if (fscrypt_is_dummy_policy_set(&fsc->fsc_dummy_enc_policy)) {
1112 if (fscrypt_dummy_policies_equal(&fsopt->dummy_enc_policy,
1113 &fsc->fsc_dummy_enc_policy))
1114 return 0;
1115 errorfc(fc, "Conflicting test_dummy_encryption options");
1116 return -EINVAL;
1117 }
1118
1119 fsc->fsc_dummy_enc_policy = fsopt->dummy_enc_policy;
1120 memset(&fsopt->dummy_enc_policy, 0, sizeof(fsopt->dummy_enc_policy));
1121
1122 warnfc(fc, "test_dummy_encryption mode enabled");
1123 return 0;
1124 }
1125 #else
ceph_apply_test_dummy_encryption(struct super_block * sb,struct fs_context * fc,struct ceph_mount_options * fsopt)1126 static int ceph_apply_test_dummy_encryption(struct super_block *sb,
1127 struct fs_context *fc,
1128 struct ceph_mount_options *fsopt)
1129 {
1130 return 0;
1131 }
1132 #endif
1133
1134 /*
1135 * mount: join the ceph cluster, and open root directory.
1136 */
ceph_real_mount(struct ceph_fs_client * fsc,struct fs_context * fc)1137 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1138 struct fs_context *fc)
1139 {
1140 struct ceph_client *cl = fsc->client;
1141 int err;
1142 unsigned long started = jiffies; /* note the start time */
1143 struct dentry *root;
1144
1145 doutc(cl, "mount start %p\n", fsc);
1146 mutex_lock(&fsc->client->mount_mutex);
1147
1148 if (!fsc->sb->s_root) {
1149 const char *path = fsc->mount_options->server_path ?
1150 fsc->mount_options->server_path + 1 : "";
1151
1152 err = __ceph_open_session(fsc->client, started);
1153 if (err < 0)
1154 goto out;
1155
1156 /* setup fscache */
1157 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1158 err = ceph_fscache_register_fs(fsc, fc);
1159 if (err < 0)
1160 goto out;
1161 }
1162
1163 err = ceph_apply_test_dummy_encryption(fsc->sb, fc,
1164 fsc->mount_options);
1165 if (err)
1166 goto out;
1167
1168 doutc(cl, "mount opening path '%s'\n", path);
1169
1170 ceph_fs_debugfs_init(fsc);
1171
1172 root = open_root_dentry(fsc, path, started);
1173 if (IS_ERR(root)) {
1174 err = PTR_ERR(root);
1175 goto out;
1176 }
1177 fsc->sb->s_root = dget(root);
1178 } else {
1179 root = dget(fsc->sb->s_root);
1180 }
1181
1182 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1183 doutc(cl, "mount success\n");
1184 mutex_unlock(&fsc->client->mount_mutex);
1185 return root;
1186
1187 out:
1188 mutex_unlock(&fsc->client->mount_mutex);
1189 ceph_fscrypt_free_dummy_policy(fsc);
1190 return ERR_PTR(err);
1191 }
1192
ceph_set_super(struct super_block * s,struct fs_context * fc)1193 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1194 {
1195 struct ceph_fs_client *fsc = s->s_fs_info;
1196 struct ceph_client *cl = fsc->client;
1197 int ret;
1198
1199 doutc(cl, "%p\n", s);
1200
1201 s->s_maxbytes = MAX_LFS_FILESIZE;
1202
1203 s->s_xattr = ceph_xattr_handlers;
1204 fsc->sb = s;
1205 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1206
1207 s->s_op = &ceph_super_ops;
1208 set_default_d_op(s, &ceph_dentry_ops);
1209 s->s_export_op = &ceph_export_ops;
1210
1211 s->s_time_gran = 1;
1212 s->s_time_min = 0;
1213 s->s_time_max = U32_MAX;
1214 s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1215 s->s_magic = CEPH_SUPER_MAGIC;
1216
1217 ceph_fscrypt_set_ops(s);
1218
1219 ret = set_anon_super_fc(s, fc);
1220 if (ret != 0)
1221 fsc->sb = NULL;
1222 return ret;
1223 }
1224
1225 /*
1226 * share superblock if same fs AND options
1227 */
ceph_compare_super(struct super_block * sb,struct fs_context * fc)1228 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1229 {
1230 struct ceph_fs_client *new = fc->s_fs_info;
1231 struct ceph_mount_options *fsopt = new->mount_options;
1232 struct ceph_options *opt = new->client->options;
1233 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1234 struct ceph_client *cl = fsc->client;
1235
1236 doutc(cl, "%p\n", sb);
1237
1238 if (compare_mount_options(fsopt, opt, fsc)) {
1239 doutc(cl, "monitor(s)/mount options don't match\n");
1240 return 0;
1241 }
1242 if ((opt->flags & CEPH_OPT_FSID) &&
1243 ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1244 doutc(cl, "fsid doesn't match\n");
1245 return 0;
1246 }
1247 if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1248 doutc(cl, "flags differ\n");
1249 return 0;
1250 }
1251
1252 if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1253 doutc(cl, "client is blocklisted (and CLEANRECOVER is not set)\n");
1254 return 0;
1255 }
1256
1257 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1258 doutc(cl, "client has been forcibly unmounted\n");
1259 return 0;
1260 }
1261
1262 return 1;
1263 }
1264
1265 /*
1266 * construct our own bdi so we can control readahead, etc.
1267 */
1268 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1269
ceph_setup_bdi(struct super_block * sb,struct ceph_fs_client * fsc)1270 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1271 {
1272 int err;
1273
1274 err = super_setup_bdi_name(sb, "ceph-%ld",
1275 atomic_long_inc_return(&bdi_seq));
1276 if (err)
1277 return err;
1278
1279 /* set ra_pages based on rasize mount option? */
1280 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1281
1282 /* set io_pages based on max osd read size */
1283 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1284
1285 return 0;
1286 }
1287
ceph_get_tree(struct fs_context * fc)1288 static int ceph_get_tree(struct fs_context *fc)
1289 {
1290 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1291 struct ceph_mount_options *fsopt = pctx->opts;
1292 struct super_block *sb;
1293 struct ceph_fs_client *fsc;
1294 struct dentry *res;
1295 int (*compare_super)(struct super_block *, struct fs_context *) =
1296 ceph_compare_super;
1297 int err;
1298
1299 dout("ceph_get_tree\n");
1300
1301 if (!fc->source)
1302 return invalfc(fc, "No source");
1303 if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1304 return invalfc(fc, "No monitor address");
1305
1306 /* create client (which we may/may not use) */
1307 fsc = create_fs_client(pctx->opts, pctx->copts);
1308 pctx->opts = NULL;
1309 pctx->copts = NULL;
1310 if (IS_ERR(fsc)) {
1311 err = PTR_ERR(fsc);
1312 goto out_final;
1313 }
1314
1315 err = ceph_mdsc_init(fsc);
1316 if (err < 0)
1317 goto out;
1318
1319 if (ceph_test_opt(fsc->client, NOSHARE))
1320 compare_super = NULL;
1321
1322 fc->s_fs_info = fsc;
1323 sb = sget_fc(fc, compare_super, ceph_set_super);
1324 fc->s_fs_info = NULL;
1325 if (IS_ERR(sb)) {
1326 err = PTR_ERR(sb);
1327 goto out;
1328 }
1329
1330 if (ceph_sb_to_fs_client(sb) != fsc) {
1331 destroy_fs_client(fsc);
1332 fsc = ceph_sb_to_fs_client(sb);
1333 dout("get_sb got existing client %p\n", fsc);
1334 } else {
1335 dout("get_sb using new client %p\n", fsc);
1336 err = ceph_setup_bdi(sb, fsc);
1337 if (err < 0)
1338 goto out_splat;
1339 }
1340
1341 res = ceph_real_mount(fsc, fc);
1342 if (IS_ERR(res)) {
1343 err = PTR_ERR(res);
1344 goto out_splat;
1345 }
1346
1347 doutc(fsc->client, "root %p inode %p ino %llx.%llx\n", res,
1348 d_inode(res), ceph_vinop(d_inode(res)));
1349 fc->root = fsc->sb->s_root;
1350 return 0;
1351
1352 out_splat:
1353 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1354 pr_info("No mds server is up or the cluster is laggy\n");
1355 err = -EHOSTUNREACH;
1356 }
1357
1358 ceph_mdsc_close_sessions(fsc->mdsc);
1359 deactivate_locked_super(sb);
1360 goto out_final;
1361
1362 out:
1363 destroy_fs_client(fsc);
1364 out_final:
1365 dout("ceph_get_tree fail %d\n", err);
1366 return err;
1367 }
1368
ceph_free_fc(struct fs_context * fc)1369 static void ceph_free_fc(struct fs_context *fc)
1370 {
1371 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1372
1373 if (pctx) {
1374 destroy_mount_options(pctx->opts);
1375 ceph_destroy_options(pctx->copts);
1376 kfree(pctx);
1377 }
1378 }
1379
ceph_reconfigure_fc(struct fs_context * fc)1380 static int ceph_reconfigure_fc(struct fs_context *fc)
1381 {
1382 int err;
1383 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1384 struct ceph_mount_options *fsopt = pctx->opts;
1385 struct super_block *sb = fc->root->d_sb;
1386 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1387
1388 err = ceph_apply_test_dummy_encryption(sb, fc, fsopt);
1389 if (err)
1390 return err;
1391
1392 if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1393 ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1394 else
1395 ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1396
1397 if (fsopt->flags & CEPH_MOUNT_OPT_SPARSEREAD)
1398 ceph_set_mount_opt(fsc, SPARSEREAD);
1399 else
1400 ceph_clear_mount_opt(fsc, SPARSEREAD);
1401
1402 if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1403 kfree(fsc->mount_options->mon_addr);
1404 fsc->mount_options->mon_addr = fsopt->mon_addr;
1405 fsopt->mon_addr = NULL;
1406 pr_notice_client(fsc->client,
1407 "monitor addresses recorded, but not used for reconnection");
1408 }
1409
1410 sync_filesystem(sb);
1411 return 0;
1412 }
1413
1414 static const struct fs_context_operations ceph_context_ops = {
1415 .free = ceph_free_fc,
1416 .parse_param = ceph_parse_mount_param,
1417 .get_tree = ceph_get_tree,
1418 .reconfigure = ceph_reconfigure_fc,
1419 };
1420
1421 /*
1422 * Set up the filesystem mount context.
1423 */
ceph_init_fs_context(struct fs_context * fc)1424 static int ceph_init_fs_context(struct fs_context *fc)
1425 {
1426 struct ceph_parse_opts_ctx *pctx;
1427 struct ceph_mount_options *fsopt;
1428
1429 pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1430 if (!pctx)
1431 return -ENOMEM;
1432
1433 pctx->copts = ceph_alloc_options();
1434 if (!pctx->copts)
1435 goto nomem;
1436
1437 pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1438 if (!pctx->opts)
1439 goto nomem;
1440
1441 fsopt = pctx->opts;
1442 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1443
1444 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1445 fsopt->rsize = CEPH_MAX_READ_SIZE;
1446 fsopt->rasize = CEPH_RASIZE_DEFAULT;
1447 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1448 if (!fsopt->snapdir_name)
1449 goto nomem;
1450
1451 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1452 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1453 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1454 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1455 fsopt->congestion_kb = default_congestion_kb();
1456
1457 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1458 fc->sb_flags |= SB_POSIXACL;
1459 #endif
1460
1461 fc->fs_private = pctx;
1462 fc->ops = &ceph_context_ops;
1463 return 0;
1464
1465 nomem:
1466 destroy_mount_options(pctx->opts);
1467 ceph_destroy_options(pctx->copts);
1468 kfree(pctx);
1469 return -ENOMEM;
1470 }
1471
1472 /*
1473 * Return true if it successfully increases the blocker counter,
1474 * or false if the mdsc is in stopping and flushed state.
1475 */
__inc_stopping_blocker(struct ceph_mds_client * mdsc)1476 static bool __inc_stopping_blocker(struct ceph_mds_client *mdsc)
1477 {
1478 spin_lock(&mdsc->stopping_lock);
1479 if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING) {
1480 spin_unlock(&mdsc->stopping_lock);
1481 return false;
1482 }
1483 atomic_inc(&mdsc->stopping_blockers);
1484 spin_unlock(&mdsc->stopping_lock);
1485 return true;
1486 }
1487
__dec_stopping_blocker(struct ceph_mds_client * mdsc)1488 static void __dec_stopping_blocker(struct ceph_mds_client *mdsc)
1489 {
1490 spin_lock(&mdsc->stopping_lock);
1491 if (!atomic_dec_return(&mdsc->stopping_blockers) &&
1492 mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHING)
1493 complete_all(&mdsc->stopping_waiter);
1494 spin_unlock(&mdsc->stopping_lock);
1495 }
1496
1497 /* For metadata IO requests */
ceph_inc_mds_stopping_blocker(struct ceph_mds_client * mdsc,struct ceph_mds_session * session)1498 bool ceph_inc_mds_stopping_blocker(struct ceph_mds_client *mdsc,
1499 struct ceph_mds_session *session)
1500 {
1501 mutex_lock(&session->s_mutex);
1502 inc_session_sequence(session);
1503 mutex_unlock(&session->s_mutex);
1504
1505 return __inc_stopping_blocker(mdsc);
1506 }
1507
ceph_dec_mds_stopping_blocker(struct ceph_mds_client * mdsc)1508 void ceph_dec_mds_stopping_blocker(struct ceph_mds_client *mdsc)
1509 {
1510 __dec_stopping_blocker(mdsc);
1511 }
1512
1513 /* For data IO requests */
ceph_inc_osd_stopping_blocker(struct ceph_mds_client * mdsc)1514 bool ceph_inc_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1515 {
1516 return __inc_stopping_blocker(mdsc);
1517 }
1518
ceph_dec_osd_stopping_blocker(struct ceph_mds_client * mdsc)1519 void ceph_dec_osd_stopping_blocker(struct ceph_mds_client *mdsc)
1520 {
1521 __dec_stopping_blocker(mdsc);
1522 }
1523
ceph_kill_sb(struct super_block * s)1524 static void ceph_kill_sb(struct super_block *s)
1525 {
1526 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(s);
1527 struct ceph_client *cl = fsc->client;
1528 struct ceph_mds_client *mdsc = fsc->mdsc;
1529 bool wait;
1530
1531 doutc(cl, "%p\n", s);
1532
1533 ceph_mdsc_pre_umount(mdsc);
1534 flush_fs_workqueues(fsc);
1535
1536 /*
1537 * Though the kill_anon_super() will finally trigger the
1538 * sync_filesystem() anyway, we still need to do it here and
1539 * then bump the stage of shutdown. This will allow us to
1540 * drop any further message, which will increase the inodes'
1541 * i_count reference counters but makes no sense any more,
1542 * from MDSs.
1543 *
1544 * Without this when evicting the inodes it may fail in the
1545 * kill_anon_super(), which will trigger a warning when
1546 * destroying the fscrypt keyring and then possibly trigger
1547 * a further crash in ceph module when the iput() tries to
1548 * evict the inodes later.
1549 */
1550 sync_filesystem(s);
1551
1552 if (atomic64_read(&mdsc->dirty_folios) > 0) {
1553 wait_queue_head_t *wq = &mdsc->flush_end_wq;
1554 long timeleft = wait_event_killable_timeout(*wq,
1555 atomic64_read(&mdsc->dirty_folios) <= 0,
1556 fsc->client->options->mount_timeout);
1557 if (!timeleft) /* timed out */
1558 pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1559 else if (timeleft < 0) /* killed */
1560 pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1561 }
1562
1563 spin_lock(&mdsc->stopping_lock);
1564 mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHING;
1565 wait = !!atomic_read(&mdsc->stopping_blockers);
1566 spin_unlock(&mdsc->stopping_lock);
1567
1568 if (wait && atomic_read(&mdsc->stopping_blockers)) {
1569 long timeleft = wait_for_completion_killable_timeout(
1570 &mdsc->stopping_waiter,
1571 fsc->client->options->mount_timeout);
1572 if (!timeleft) /* timed out */
1573 pr_warn_client(cl, "umount timed out, %ld\n", timeleft);
1574 else if (timeleft < 0) /* killed */
1575 pr_warn_client(cl, "umount was killed, %ld\n", timeleft);
1576 }
1577
1578 mdsc->stopping = CEPH_MDSC_STOPPING_FLUSHED;
1579 kill_anon_super(s);
1580
1581 fsc->client->extra_mon_dispatch = NULL;
1582 ceph_fs_debugfs_cleanup(fsc);
1583
1584 ceph_fscache_unregister_fs(fsc);
1585
1586 destroy_fs_client(fsc);
1587 }
1588
1589 static struct file_system_type ceph_fs_type = {
1590 .owner = THIS_MODULE,
1591 .name = "ceph",
1592 .init_fs_context = ceph_init_fs_context,
1593 .kill_sb = ceph_kill_sb,
1594 .fs_flags = FS_RENAME_DOES_D_MOVE | FS_ALLOW_IDMAP,
1595 };
1596 MODULE_ALIAS_FS("ceph");
1597
ceph_force_reconnect(struct super_block * sb)1598 int ceph_force_reconnect(struct super_block *sb)
1599 {
1600 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1601 int err = 0;
1602
1603 fsc->mount_state = CEPH_MOUNT_RECOVER;
1604 __ceph_umount_begin(fsc);
1605
1606 /* Make sure all page caches get invalidated.
1607 * see remove_session_caps_cb() */
1608 flush_workqueue(fsc->inode_wq);
1609
1610 /* In case that we were blocklisted. This also reset
1611 * all mon/osd connections */
1612 ceph_reset_client_addr(fsc->client);
1613
1614 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1615
1616 fsc->blocklisted = false;
1617 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1618
1619 if (sb->s_root) {
1620 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1621 CEPH_STAT_CAP_INODE, true);
1622 }
1623 return err;
1624 }
1625
init_ceph(void)1626 static int __init init_ceph(void)
1627 {
1628 int ret = init_caches();
1629 if (ret)
1630 goto out;
1631
1632 ceph_flock_init();
1633 ret = register_filesystem(&ceph_fs_type);
1634 if (ret)
1635 goto out_caches;
1636
1637 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1638
1639 return 0;
1640
1641 out_caches:
1642 destroy_caches();
1643 out:
1644 return ret;
1645 }
1646
exit_ceph(void)1647 static void __exit exit_ceph(void)
1648 {
1649 dout("exit_ceph\n");
1650 unregister_filesystem(&ceph_fs_type);
1651 destroy_caches();
1652 }
1653
param_set_metrics(const char * val,const struct kernel_param * kp)1654 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1655 {
1656 struct ceph_fs_client *fsc;
1657 int ret;
1658
1659 ret = param_set_bool(val, kp);
1660 if (ret) {
1661 pr_err("Failed to parse sending metrics switch value '%s'\n",
1662 val);
1663 return ret;
1664 } else if (!disable_send_metrics) {
1665 // wake up all the mds clients
1666 spin_lock(&ceph_fsc_lock);
1667 list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1668 metric_schedule_delayed(&fsc->mdsc->metric);
1669 }
1670 spin_unlock(&ceph_fsc_lock);
1671 }
1672
1673 return 0;
1674 }
1675
1676 static const struct kernel_param_ops param_ops_metrics = {
1677 .set = param_set_metrics,
1678 .get = param_get_bool,
1679 };
1680
1681 bool disable_send_metrics = false;
1682 module_param_cb(disable_send_metrics, ¶m_ops_metrics, &disable_send_metrics, 0644);
1683 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1684
1685 /* for both v1 and v2 syntax */
1686 static bool mount_support = true;
1687 static const struct kernel_param_ops param_ops_mount_syntax = {
1688 .get = param_get_bool,
1689 };
1690 module_param_cb(mount_syntax_v1, ¶m_ops_mount_syntax, &mount_support, 0444);
1691 module_param_cb(mount_syntax_v2, ¶m_ops_mount_syntax, &mount_support, 0444);
1692
1693 bool enable_unsafe_idmap = false;
1694 module_param(enable_unsafe_idmap, bool, 0644);
1695 MODULE_PARM_DESC(enable_unsafe_idmap,
1696 "Allow to use idmapped mounts with MDS without CEPHFS_FEATURE_HAS_OWNER_UIDGID");
1697
1698 module_init(init_ceph);
1699 module_exit(exit_ceph);
1700
1701 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1702 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1703 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1704 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1705 MODULE_LICENSE("GPL");
1706