Loading...
1
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/backing-dev.h>
5#include <linux/ctype.h>
6#include <linux/fs.h>
7#include <linux/inet.h>
8#include <linux/in6.h>
9#include <linux/module.h>
10#include <linux/mount.h>
11#include <linux/parser.h>
12#include <linux/sched.h>
13#include <linux/seq_file.h>
14#include <linux/slab.h>
15#include <linux/statfs.h>
16#include <linux/string.h>
17
18#include "super.h"
19#include "mds_client.h"
20
21#include <linux/ceph/decode.h>
22#include <linux/ceph/mon_client.h>
23#include <linux/ceph/auth.h>
24#include <linux/ceph/debugfs.h>
25
26/*
27 * Ceph superblock operations
28 *
29 * Handle the basics of mounting, unmounting.
30 */
31
32/*
33 * super ops
34 */
35static void ceph_put_super(struct super_block *s)
36{
37 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
38
39 dout("put_super\n");
40 ceph_mdsc_close_sessions(fsc->mdsc);
41
42 /*
43 * ensure we release the bdi before put_anon_super releases
44 * the device name.
45 */
46 if (s->s_bdi == &fsc->backing_dev_info) {
47 bdi_unregister(&fsc->backing_dev_info);
48 s->s_bdi = NULL;
49 }
50
51 return;
52}
53
54static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
55{
56 struct ceph_fs_client *fsc = ceph_inode_to_client(dentry->d_inode);
57 struct ceph_monmap *monmap = fsc->client->monc.monmap;
58 struct ceph_statfs st;
59 u64 fsid;
60 int err;
61
62 dout("statfs\n");
63 err = ceph_monc_do_statfs(&fsc->client->monc, &st);
64 if (err < 0)
65 return err;
66
67 /* fill in kstatfs */
68 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
69
70 /*
71 * express utilization in terms of large blocks to avoid
72 * overflow on 32-bit machines.
73 */
74 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
75 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
76 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
77 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
78
79 buf->f_files = le64_to_cpu(st.num_objects);
80 buf->f_ffree = -1;
81 buf->f_namelen = NAME_MAX;
82 buf->f_frsize = PAGE_CACHE_SIZE;
83
84 /* leave fsid little-endian, regardless of host endianness */
85 fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1);
86 buf->f_fsid.val[0] = fsid & 0xffffffff;
87 buf->f_fsid.val[1] = fsid >> 32;
88
89 return 0;
90}
91
92
93static int ceph_sync_fs(struct super_block *sb, int wait)
94{
95 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
96
97 if (!wait) {
98 dout("sync_fs (non-blocking)\n");
99 ceph_flush_dirty_caps(fsc->mdsc);
100 dout("sync_fs (non-blocking) done\n");
101 return 0;
102 }
103
104 dout("sync_fs (blocking)\n");
105 ceph_osdc_sync(&fsc->client->osdc);
106 ceph_mdsc_sync(fsc->mdsc);
107 dout("sync_fs (blocking) done\n");
108 return 0;
109}
110
111/*
112 * mount options
113 */
114enum {
115 Opt_wsize,
116 Opt_rsize,
117 Opt_caps_wanted_delay_min,
118 Opt_caps_wanted_delay_max,
119 Opt_cap_release_safety,
120 Opt_readdir_max_entries,
121 Opt_readdir_max_bytes,
122 Opt_congestion_kb,
123 Opt_last_int,
124 /* int args above */
125 Opt_snapdirname,
126 Opt_last_string,
127 /* string args above */
128 Opt_dirstat,
129 Opt_nodirstat,
130 Opt_rbytes,
131 Opt_norbytes,
132 Opt_noasyncreaddir,
133 Opt_ino32,
134};
135
136static match_table_t fsopt_tokens = {
137 {Opt_wsize, "wsize=%d"},
138 {Opt_rsize, "rsize=%d"},
139 {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"},
140 {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"},
141 {Opt_cap_release_safety, "cap_release_safety=%d"},
142 {Opt_readdir_max_entries, "readdir_max_entries=%d"},
143 {Opt_readdir_max_bytes, "readdir_max_bytes=%d"},
144 {Opt_congestion_kb, "write_congestion_kb=%d"},
145 /* int args above */
146 {Opt_snapdirname, "snapdirname=%s"},
147 /* string args above */
148 {Opt_dirstat, "dirstat"},
149 {Opt_nodirstat, "nodirstat"},
150 {Opt_rbytes, "rbytes"},
151 {Opt_norbytes, "norbytes"},
152 {Opt_noasyncreaddir, "noasyncreaddir"},
153 {Opt_ino32, "ino32"},
154 {-1, NULL}
155};
156
157static int parse_fsopt_token(char *c, void *private)
158{
159 struct ceph_mount_options *fsopt = private;
160 substring_t argstr[MAX_OPT_ARGS];
161 int token, intval, ret;
162
163 token = match_token((char *)c, fsopt_tokens, argstr);
164 if (token < 0)
165 return -EINVAL;
166
167 if (token < Opt_last_int) {
168 ret = match_int(&argstr[0], &intval);
169 if (ret < 0) {
170 pr_err("bad mount option arg (not int) "
171 "at '%s'\n", c);
172 return ret;
173 }
174 dout("got int token %d val %d\n", token, intval);
175 } else if (token > Opt_last_int && token < Opt_last_string) {
176 dout("got string token %d val %s\n", token,
177 argstr[0].from);
178 } else {
179 dout("got token %d\n", token);
180 }
181
182 switch (token) {
183 case Opt_snapdirname:
184 kfree(fsopt->snapdir_name);
185 fsopt->snapdir_name = kstrndup(argstr[0].from,
186 argstr[0].to-argstr[0].from,
187 GFP_KERNEL);
188 if (!fsopt->snapdir_name)
189 return -ENOMEM;
190 break;
191
192 /* misc */
193 case Opt_wsize:
194 fsopt->wsize = intval;
195 break;
196 case Opt_rsize:
197 fsopt->rsize = intval;
198 break;
199 case Opt_caps_wanted_delay_min:
200 fsopt->caps_wanted_delay_min = intval;
201 break;
202 case Opt_caps_wanted_delay_max:
203 fsopt->caps_wanted_delay_max = intval;
204 break;
205 case Opt_readdir_max_entries:
206 fsopt->max_readdir = intval;
207 break;
208 case Opt_readdir_max_bytes:
209 fsopt->max_readdir_bytes = intval;
210 break;
211 case Opt_congestion_kb:
212 fsopt->congestion_kb = intval;
213 break;
214 case Opt_dirstat:
215 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
216 break;
217 case Opt_nodirstat:
218 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
219 break;
220 case Opt_rbytes:
221 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
222 break;
223 case Opt_norbytes:
224 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
225 break;
226 case Opt_noasyncreaddir:
227 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
228 break;
229 case Opt_ino32:
230 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
231 break;
232 default:
233 BUG_ON(token);
234 }
235 return 0;
236}
237
238static void destroy_mount_options(struct ceph_mount_options *args)
239{
240 dout("destroy_mount_options %p\n", args);
241 kfree(args->snapdir_name);
242 kfree(args);
243}
244
245static int strcmp_null(const char *s1, const char *s2)
246{
247 if (!s1 && !s2)
248 return 0;
249 if (s1 && !s2)
250 return -1;
251 if (!s1 && s2)
252 return 1;
253 return strcmp(s1, s2);
254}
255
256static int compare_mount_options(struct ceph_mount_options *new_fsopt,
257 struct ceph_options *new_opt,
258 struct ceph_fs_client *fsc)
259{
260 struct ceph_mount_options *fsopt1 = new_fsopt;
261 struct ceph_mount_options *fsopt2 = fsc->mount_options;
262 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
263 int ret;
264
265 ret = memcmp(fsopt1, fsopt2, ofs);
266 if (ret)
267 return ret;
268
269 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
270 if (ret)
271 return ret;
272
273 return ceph_compare_options(new_opt, fsc->client);
274}
275
276static int parse_mount_options(struct ceph_mount_options **pfsopt,
277 struct ceph_options **popt,
278 int flags, char *options,
279 const char *dev_name,
280 const char **path)
281{
282 struct ceph_mount_options *fsopt;
283 const char *dev_name_end;
284 int err = -ENOMEM;
285
286 fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL);
287 if (!fsopt)
288 return -ENOMEM;
289
290 dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name);
291
292 fsopt->sb_flags = flags;
293 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
294
295 fsopt->rsize = CEPH_RSIZE_DEFAULT;
296 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
297 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
298 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
299 fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT;
300 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
301 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
302 fsopt->congestion_kb = default_congestion_kb();
303
304 /* ip1[:port1][,ip2[:port2]...]:/subdir/in/fs */
305 err = -EINVAL;
306 if (!dev_name)
307 goto out;
308 *path = strstr(dev_name, ":/");
309 if (*path == NULL) {
310 pr_err("device name is missing path (no :/ in %s)\n",
311 dev_name);
312 goto out;
313 }
314 dev_name_end = *path;
315 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
316
317 /* path on server */
318 *path += 2;
319 dout("server path '%s'\n", *path);
320
321 err = ceph_parse_options(popt, options, dev_name, dev_name_end,
322 parse_fsopt_token, (void *)fsopt);
323 if (err)
324 goto out;
325
326 /* success */
327 *pfsopt = fsopt;
328 return 0;
329
330out:
331 destroy_mount_options(fsopt);
332 return err;
333}
334
335/**
336 * ceph_show_options - Show mount options in /proc/mounts
337 * @m: seq_file to write to
338 * @mnt: mount descriptor
339 */
340static int ceph_show_options(struct seq_file *m, struct vfsmount *mnt)
341{
342 struct ceph_fs_client *fsc = ceph_sb_to_client(mnt->mnt_sb);
343 struct ceph_mount_options *fsopt = fsc->mount_options;
344 struct ceph_options *opt = fsc->client->options;
345
346 if (opt->flags & CEPH_OPT_FSID)
347 seq_printf(m, ",fsid=%pU", &opt->fsid);
348 if (opt->flags & CEPH_OPT_NOSHARE)
349 seq_puts(m, ",noshare");
350 if (opt->flags & CEPH_OPT_NOCRC)
351 seq_puts(m, ",nocrc");
352
353 if (opt->name)
354 seq_printf(m, ",name=%s", opt->name);
355 if (opt->key)
356 seq_puts(m, ",secret=<hidden>");
357
358 if (opt->mount_timeout != CEPH_MOUNT_TIMEOUT_DEFAULT)
359 seq_printf(m, ",mount_timeout=%d", opt->mount_timeout);
360 if (opt->osd_idle_ttl != CEPH_OSD_IDLE_TTL_DEFAULT)
361 seq_printf(m, ",osd_idle_ttl=%d", opt->osd_idle_ttl);
362 if (opt->osd_timeout != CEPH_OSD_TIMEOUT_DEFAULT)
363 seq_printf(m, ",osdtimeout=%d", opt->osd_timeout);
364 if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
365 seq_printf(m, ",osdkeepalivetimeout=%d",
366 opt->osd_keepalive_timeout);
367
368 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
369 seq_puts(m, ",dirstat");
370 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES) == 0)
371 seq_puts(m, ",norbytes");
372 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
373 seq_puts(m, ",noasyncreaddir");
374
375 if (fsopt->wsize)
376 seq_printf(m, ",wsize=%d", fsopt->wsize);
377 if (fsopt->rsize != CEPH_RSIZE_DEFAULT)
378 seq_printf(m, ",rsize=%d", fsopt->rsize);
379 if (fsopt->congestion_kb != default_congestion_kb())
380 seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb);
381 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
382 seq_printf(m, ",caps_wanted_delay_min=%d",
383 fsopt->caps_wanted_delay_min);
384 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
385 seq_printf(m, ",caps_wanted_delay_max=%d",
386 fsopt->caps_wanted_delay_max);
387 if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT)
388 seq_printf(m, ",cap_release_safety=%d",
389 fsopt->cap_release_safety);
390 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
391 seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir);
392 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
393 seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes);
394 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
395 seq_printf(m, ",snapdirname=%s", fsopt->snapdir_name);
396 return 0;
397}
398
399/*
400 * handle any mon messages the standard library doesn't understand.
401 * return error if we don't either.
402 */
403static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
404{
405 struct ceph_fs_client *fsc = client->private;
406 int type = le16_to_cpu(msg->hdr.type);
407
408 switch (type) {
409 case CEPH_MSG_MDS_MAP:
410 ceph_mdsc_handle_map(fsc->mdsc, msg);
411 return 0;
412
413 default:
414 return -1;
415 }
416}
417
418/*
419 * create a new fs client
420 */
421struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
422 struct ceph_options *opt)
423{
424 struct ceph_fs_client *fsc;
425 int err = -ENOMEM;
426
427 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
428 if (!fsc)
429 return ERR_PTR(-ENOMEM);
430
431 fsc->client = ceph_create_client(opt, fsc);
432 if (IS_ERR(fsc->client)) {
433 err = PTR_ERR(fsc->client);
434 goto fail;
435 }
436 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
437 fsc->client->supported_features |= CEPH_FEATURE_FLOCK |
438 CEPH_FEATURE_DIRLAYOUTHASH;
439 fsc->client->monc.want_mdsmap = 1;
440
441 fsc->mount_options = fsopt;
442
443 fsc->sb = NULL;
444 fsc->mount_state = CEPH_MOUNT_MOUNTING;
445
446 atomic_long_set(&fsc->writeback_count, 0);
447
448 err = bdi_init(&fsc->backing_dev_info);
449 if (err < 0)
450 goto fail_client;
451
452 err = -ENOMEM;
453 /*
454 * The number of concurrent works can be high but they don't need
455 * to be processed in parallel, limit concurrency.
456 */
457 fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1);
458 if (fsc->wb_wq == NULL)
459 goto fail_bdi;
460 fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1);
461 if (fsc->pg_inv_wq == NULL)
462 goto fail_wb_wq;
463 fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1);
464 if (fsc->trunc_wq == NULL)
465 goto fail_pg_inv_wq;
466
467 /* set up mempools */
468 err = -ENOMEM;
469 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10,
470 fsc->mount_options->wsize >> PAGE_CACHE_SHIFT);
471 if (!fsc->wb_pagevec_pool)
472 goto fail_trunc_wq;
473
474 /* caps */
475 fsc->min_caps = fsopt->max_readdir;
476
477 return fsc;
478
479fail_trunc_wq:
480 destroy_workqueue(fsc->trunc_wq);
481fail_pg_inv_wq:
482 destroy_workqueue(fsc->pg_inv_wq);
483fail_wb_wq:
484 destroy_workqueue(fsc->wb_wq);
485fail_bdi:
486 bdi_destroy(&fsc->backing_dev_info);
487fail_client:
488 ceph_destroy_client(fsc->client);
489fail:
490 kfree(fsc);
491 return ERR_PTR(err);
492}
493
494void destroy_fs_client(struct ceph_fs_client *fsc)
495{
496 dout("destroy_fs_client %p\n", fsc);
497
498 destroy_workqueue(fsc->wb_wq);
499 destroy_workqueue(fsc->pg_inv_wq);
500 destroy_workqueue(fsc->trunc_wq);
501
502 bdi_destroy(&fsc->backing_dev_info);
503
504 mempool_destroy(fsc->wb_pagevec_pool);
505
506 destroy_mount_options(fsc->mount_options);
507
508 ceph_fs_debugfs_cleanup(fsc);
509
510 ceph_destroy_client(fsc->client);
511
512 kfree(fsc);
513 dout("destroy_fs_client %p done\n", fsc);
514}
515
516/*
517 * caches
518 */
519struct kmem_cache *ceph_inode_cachep;
520struct kmem_cache *ceph_cap_cachep;
521struct kmem_cache *ceph_dentry_cachep;
522struct kmem_cache *ceph_file_cachep;
523
524static void ceph_inode_init_once(void *foo)
525{
526 struct ceph_inode_info *ci = foo;
527 inode_init_once(&ci->vfs_inode);
528}
529
530static int __init init_caches(void)
531{
532 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
533 sizeof(struct ceph_inode_info),
534 __alignof__(struct ceph_inode_info),
535 (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
536 ceph_inode_init_once);
537 if (ceph_inode_cachep == NULL)
538 return -ENOMEM;
539
540 ceph_cap_cachep = KMEM_CACHE(ceph_cap,
541 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
542 if (ceph_cap_cachep == NULL)
543 goto bad_cap;
544
545 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
546 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
547 if (ceph_dentry_cachep == NULL)
548 goto bad_dentry;
549
550 ceph_file_cachep = KMEM_CACHE(ceph_file_info,
551 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
552 if (ceph_file_cachep == NULL)
553 goto bad_file;
554
555 return 0;
556
557bad_file:
558 kmem_cache_destroy(ceph_dentry_cachep);
559bad_dentry:
560 kmem_cache_destroy(ceph_cap_cachep);
561bad_cap:
562 kmem_cache_destroy(ceph_inode_cachep);
563 return -ENOMEM;
564}
565
566static void destroy_caches(void)
567{
568 kmem_cache_destroy(ceph_inode_cachep);
569 kmem_cache_destroy(ceph_cap_cachep);
570 kmem_cache_destroy(ceph_dentry_cachep);
571 kmem_cache_destroy(ceph_file_cachep);
572}
573
574
575/*
576 * ceph_umount_begin - initiate forced umount. Tear down down the
577 * mount, skipping steps that may hang while waiting for server(s).
578 */
579static void ceph_umount_begin(struct super_block *sb)
580{
581 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
582
583 dout("ceph_umount_begin - starting forced umount\n");
584 if (!fsc)
585 return;
586 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
587 return;
588}
589
590static const struct super_operations ceph_super_ops = {
591 .alloc_inode = ceph_alloc_inode,
592 .destroy_inode = ceph_destroy_inode,
593 .write_inode = ceph_write_inode,
594 .sync_fs = ceph_sync_fs,
595 .put_super = ceph_put_super,
596 .show_options = ceph_show_options,
597 .statfs = ceph_statfs,
598 .umount_begin = ceph_umount_begin,
599};
600
601/*
602 * Bootstrap mount by opening the root directory. Note the mount
603 * @started time from caller, and time out if this takes too long.
604 */
605static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
606 const char *path,
607 unsigned long started)
608{
609 struct ceph_mds_client *mdsc = fsc->mdsc;
610 struct ceph_mds_request *req = NULL;
611 int err;
612 struct dentry *root;
613
614 /* open dir */
615 dout("open_root_inode opening '%s'\n", path);
616 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
617 if (IS_ERR(req))
618 return ERR_CAST(req);
619 req->r_path1 = kstrdup(path, GFP_NOFS);
620 req->r_ino1.ino = CEPH_INO_ROOT;
621 req->r_ino1.snap = CEPH_NOSNAP;
622 req->r_started = started;
623 req->r_timeout = fsc->client->options->mount_timeout * HZ;
624 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
625 req->r_num_caps = 2;
626 err = ceph_mdsc_do_request(mdsc, NULL, req);
627 if (err == 0) {
628 dout("open_root_inode success\n");
629 if (ceph_ino(req->r_target_inode) == CEPH_INO_ROOT &&
630 fsc->sb->s_root == NULL)
631 root = d_alloc_root(req->r_target_inode);
632 else
633 root = d_obtain_alias(req->r_target_inode);
634 req->r_target_inode = NULL;
635 dout("open_root_inode success, root dentry is %p\n", root);
636 } else {
637 root = ERR_PTR(err);
638 }
639 ceph_mdsc_put_request(req);
640 return root;
641}
642
643
644
645
646/*
647 * mount: join the ceph cluster, and open root directory.
648 */
649static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
650 const char *path)
651{
652 int err;
653 unsigned long started = jiffies; /* note the start time */
654 struct dentry *root;
655 int first = 0; /* first vfsmount for this super_block */
656
657 dout("mount start\n");
658 mutex_lock(&fsc->client->mount_mutex);
659
660 err = __ceph_open_session(fsc->client, started);
661 if (err < 0)
662 goto out;
663
664 dout("mount opening root\n");
665 root = open_root_dentry(fsc, "", started);
666 if (IS_ERR(root)) {
667 err = PTR_ERR(root);
668 goto out;
669 }
670 if (fsc->sb->s_root) {
671 dput(root);
672 } else {
673 fsc->sb->s_root = root;
674 first = 1;
675
676 err = ceph_fs_debugfs_init(fsc);
677 if (err < 0)
678 goto fail;
679 }
680
681 if (path[0] == 0) {
682 dget(root);
683 } else {
684 dout("mount opening base mountpoint\n");
685 root = open_root_dentry(fsc, path, started);
686 if (IS_ERR(root)) {
687 err = PTR_ERR(root);
688 goto fail;
689 }
690 }
691
692 fsc->mount_state = CEPH_MOUNT_MOUNTED;
693 dout("mount success\n");
694 mutex_unlock(&fsc->client->mount_mutex);
695 return root;
696
697out:
698 mutex_unlock(&fsc->client->mount_mutex);
699 return ERR_PTR(err);
700
701fail:
702 if (first) {
703 dput(fsc->sb->s_root);
704 fsc->sb->s_root = NULL;
705 }
706 goto out;
707}
708
709static int ceph_set_super(struct super_block *s, void *data)
710{
711 struct ceph_fs_client *fsc = data;
712 int ret;
713
714 dout("set_super %p data %p\n", s, data);
715
716 s->s_flags = fsc->mount_options->sb_flags;
717 s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */
718
719 s->s_fs_info = fsc;
720 fsc->sb = s;
721
722 s->s_op = &ceph_super_ops;
723 s->s_export_op = &ceph_export_ops;
724
725 s->s_time_gran = 1000; /* 1000 ns == 1 us */
726
727 ret = set_anon_super(s, NULL); /* what is that second arg for? */
728 if (ret != 0)
729 goto fail;
730
731 return ret;
732
733fail:
734 s->s_fs_info = NULL;
735 fsc->sb = NULL;
736 return ret;
737}
738
739/*
740 * share superblock if same fs AND options
741 */
742static int ceph_compare_super(struct super_block *sb, void *data)
743{
744 struct ceph_fs_client *new = data;
745 struct ceph_mount_options *fsopt = new->mount_options;
746 struct ceph_options *opt = new->client->options;
747 struct ceph_fs_client *other = ceph_sb_to_client(sb);
748
749 dout("ceph_compare_super %p\n", sb);
750
751 if (compare_mount_options(fsopt, opt, other)) {
752 dout("monitor(s)/mount options don't match\n");
753 return 0;
754 }
755 if ((opt->flags & CEPH_OPT_FSID) &&
756 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
757 dout("fsid doesn't match\n");
758 return 0;
759 }
760 if (fsopt->sb_flags != other->mount_options->sb_flags) {
761 dout("flags differ\n");
762 return 0;
763 }
764 return 1;
765}
766
767/*
768 * construct our own bdi so we can control readahead, etc.
769 */
770static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
771
772static int ceph_register_bdi(struct super_block *sb,
773 struct ceph_fs_client *fsc)
774{
775 int err;
776
777 /* set ra_pages based on rsize mount option? */
778 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
779 fsc->backing_dev_info.ra_pages =
780 (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
781 >> PAGE_SHIFT;
782 else
783 fsc->backing_dev_info.ra_pages =
784 default_backing_dev_info.ra_pages;
785
786 err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%d",
787 atomic_long_inc_return(&bdi_seq));
788 if (!err)
789 sb->s_bdi = &fsc->backing_dev_info;
790 return err;
791}
792
793static struct dentry *ceph_mount(struct file_system_type *fs_type,
794 int flags, const char *dev_name, void *data)
795{
796 struct super_block *sb;
797 struct ceph_fs_client *fsc;
798 struct dentry *res;
799 int err;
800 int (*compare_super)(struct super_block *, void *) = ceph_compare_super;
801 const char *path = NULL;
802 struct ceph_mount_options *fsopt = NULL;
803 struct ceph_options *opt = NULL;
804
805 dout("ceph_mount\n");
806 err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path);
807 if (err < 0) {
808 res = ERR_PTR(err);
809 goto out_final;
810 }
811
812 /* create client (which we may/may not use) */
813 fsc = create_fs_client(fsopt, opt);
814 if (IS_ERR(fsc)) {
815 res = ERR_CAST(fsc);
816 destroy_mount_options(fsopt);
817 ceph_destroy_options(opt);
818 goto out_final;
819 }
820
821 err = ceph_mdsc_init(fsc);
822 if (err < 0) {
823 res = ERR_PTR(err);
824 goto out;
825 }
826
827 if (ceph_test_opt(fsc->client, NOSHARE))
828 compare_super = NULL;
829 sb = sget(fs_type, compare_super, ceph_set_super, fsc);
830 if (IS_ERR(sb)) {
831 res = ERR_CAST(sb);
832 goto out;
833 }
834
835 if (ceph_sb_to_client(sb) != fsc) {
836 ceph_mdsc_destroy(fsc);
837 destroy_fs_client(fsc);
838 fsc = ceph_sb_to_client(sb);
839 dout("get_sb got existing client %p\n", fsc);
840 } else {
841 dout("get_sb using new client %p\n", fsc);
842 err = ceph_register_bdi(sb, fsc);
843 if (err < 0) {
844 res = ERR_PTR(err);
845 goto out_splat;
846 }
847 }
848
849 res = ceph_real_mount(fsc, path);
850 if (IS_ERR(res))
851 goto out_splat;
852 dout("root %p inode %p ino %llx.%llx\n", res,
853 res->d_inode, ceph_vinop(res->d_inode));
854 return res;
855
856out_splat:
857 ceph_mdsc_close_sessions(fsc->mdsc);
858 deactivate_locked_super(sb);
859 goto out_final;
860
861out:
862 ceph_mdsc_destroy(fsc);
863 destroy_fs_client(fsc);
864out_final:
865 dout("ceph_mount fail %ld\n", PTR_ERR(res));
866 return res;
867}
868
869static void ceph_kill_sb(struct super_block *s)
870{
871 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
872 dout("kill_sb %p\n", s);
873 ceph_mdsc_pre_umount(fsc->mdsc);
874 kill_anon_super(s); /* will call put_super after sb is r/o */
875 ceph_mdsc_destroy(fsc);
876 destroy_fs_client(fsc);
877}
878
879static struct file_system_type ceph_fs_type = {
880 .owner = THIS_MODULE,
881 .name = "ceph",
882 .mount = ceph_mount,
883 .kill_sb = ceph_kill_sb,
884 .fs_flags = FS_RENAME_DOES_D_MOVE,
885};
886
887#define _STRINGIFY(x) #x
888#define STRINGIFY(x) _STRINGIFY(x)
889
890static int __init init_ceph(void)
891{
892 int ret = init_caches();
893 if (ret)
894 goto out;
895
896 ret = register_filesystem(&ceph_fs_type);
897 if (ret)
898 goto out_icache;
899
900 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
901
902 return 0;
903
904out_icache:
905 destroy_caches();
906out:
907 return ret;
908}
909
910static void __exit exit_ceph(void)
911{
912 dout("exit_ceph\n");
913 unregister_filesystem(&ceph_fs_type);
914 destroy_caches();
915}
916
917module_init(init_ceph);
918module_exit(exit_ceph);
919
920MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
921MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
922MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
923MODULE_DESCRIPTION("Ceph filesystem for Linux");
924MODULE_LICENSE("GPL");
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
24#include <linux/ceph/ceph_features.h>
25#include <linux/ceph/decode.h>
26#include <linux/ceph/mon_client.h>
27#include <linux/ceph/auth.h>
28#include <linux/ceph/debugfs.h>
29
30static DEFINE_SPINLOCK(ceph_fsc_lock);
31static LIST_HEAD(ceph_fsc_list);
32
33/*
34 * Ceph superblock operations
35 *
36 * Handle the basics of mounting, unmounting.
37 */
38
39/*
40 * super ops
41 */
42static void ceph_put_super(struct super_block *s)
43{
44 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
45
46 dout("put_super\n");
47 ceph_mdsc_close_sessions(fsc->mdsc);
48}
49
50static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
51{
52 struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
53 struct ceph_mon_client *monc = &fsc->client->monc;
54 struct ceph_statfs st;
55 u64 fsid;
56 int err;
57 u64 data_pool;
58
59 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
60 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
61 } else {
62 data_pool = CEPH_NOPOOL;
63 }
64
65 dout("statfs\n");
66 err = ceph_monc_do_statfs(monc, data_pool, &st);
67 if (err < 0)
68 return err;
69
70 /* fill in kstatfs */
71 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
72
73 /*
74 * express utilization in terms of large blocks to avoid
75 * overflow on 32-bit machines.
76 *
77 * NOTE: for the time being, we make bsize == frsize to humor
78 * not-yet-ancient versions of glibc that are broken.
79 * Someday, we will probably want to report a real block
80 * size... whatever that may mean for a network file system!
81 */
82 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
83 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
84
85 /*
86 * By default use root quota for stats; fallback to overall filesystem
87 * usage if using 'noquotadf' mount option or if the root dir doesn't
88 * have max_bytes quota set.
89 */
90 if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
91 !ceph_quota_update_statfs(fsc, buf)) {
92 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
93 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
94 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
95 }
96
97 buf->f_files = le64_to_cpu(st.num_objects);
98 buf->f_ffree = -1;
99 buf->f_namelen = NAME_MAX;
100
101 /* Must convert the fsid, for consistent values across arches */
102 mutex_lock(&monc->mutex);
103 fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
104 le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
105 mutex_unlock(&monc->mutex);
106
107 buf->f_fsid = u64_to_fsid(fsid);
108
109 return 0;
110}
111
112static int ceph_sync_fs(struct super_block *sb, int wait)
113{
114 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
115
116 if (!wait) {
117 dout("sync_fs (non-blocking)\n");
118 ceph_flush_dirty_caps(fsc->mdsc);
119 dout("sync_fs (non-blocking) done\n");
120 return 0;
121 }
122
123 dout("sync_fs (blocking)\n");
124 ceph_osdc_sync(&fsc->client->osdc);
125 ceph_mdsc_sync(fsc->mdsc);
126 dout("sync_fs (blocking) done\n");
127 return 0;
128}
129
130/*
131 * mount options
132 */
133enum {
134 Opt_wsize,
135 Opt_rsize,
136 Opt_rasize,
137 Opt_caps_wanted_delay_min,
138 Opt_caps_wanted_delay_max,
139 Opt_caps_max,
140 Opt_readdir_max_entries,
141 Opt_readdir_max_bytes,
142 Opt_congestion_kb,
143 /* int args above */
144 Opt_snapdirname,
145 Opt_mds_namespace,
146 Opt_recover_session,
147 Opt_source,
148 /* string args above */
149 Opt_dirstat,
150 Opt_rbytes,
151 Opt_asyncreaddir,
152 Opt_dcache,
153 Opt_ino32,
154 Opt_fscache,
155 Opt_poolperm,
156 Opt_require_active_mds,
157 Opt_acl,
158 Opt_quotadf,
159 Opt_copyfrom,
160 Opt_wsync,
161};
162
163enum ceph_recover_session_mode {
164 ceph_recover_session_no,
165 ceph_recover_session_clean
166};
167
168static const struct constant_table ceph_param_recover[] = {
169 { "no", ceph_recover_session_no },
170 { "clean", ceph_recover_session_clean },
171 {}
172};
173
174static const struct fs_parameter_spec ceph_mount_parameters[] = {
175 fsparam_flag_no ("acl", Opt_acl),
176 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir),
177 fsparam_s32 ("caps_max", Opt_caps_max),
178 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max),
179 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min),
180 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb),
181 fsparam_flag_no ("copyfrom", Opt_copyfrom),
182 fsparam_flag_no ("dcache", Opt_dcache),
183 fsparam_flag_no ("dirstat", Opt_dirstat),
184 fsparam_flag_no ("fsc", Opt_fscache), // fsc|nofsc
185 fsparam_string ("fsc", Opt_fscache), // fsc=...
186 fsparam_flag_no ("ino32", Opt_ino32),
187 fsparam_string ("mds_namespace", Opt_mds_namespace),
188 fsparam_flag_no ("poolperm", Opt_poolperm),
189 fsparam_flag_no ("quotadf", Opt_quotadf),
190 fsparam_u32 ("rasize", Opt_rasize),
191 fsparam_flag_no ("rbytes", Opt_rbytes),
192 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes),
193 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries),
194 fsparam_enum ("recover_session", Opt_recover_session, ceph_param_recover),
195 fsparam_flag_no ("require_active_mds", Opt_require_active_mds),
196 fsparam_u32 ("rsize", Opt_rsize),
197 fsparam_string ("snapdirname", Opt_snapdirname),
198 fsparam_string ("source", Opt_source),
199 fsparam_u32 ("wsize", Opt_wsize),
200 fsparam_flag_no ("wsync", Opt_wsync),
201 {}
202};
203
204struct ceph_parse_opts_ctx {
205 struct ceph_options *copts;
206 struct ceph_mount_options *opts;
207};
208
209/*
210 * Remove adjacent slashes and then the trailing slash, unless it is
211 * the only remaining character.
212 *
213 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
214 */
215static void canonicalize_path(char *path)
216{
217 int i, j = 0;
218
219 for (i = 0; path[i] != '\0'; i++) {
220 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
221 path[j++] = path[i];
222 }
223
224 if (j > 1 && path[j - 1] == '/')
225 j--;
226 path[j] = '\0';
227}
228
229/*
230 * Parse the source parameter. Distinguish the server list from the path.
231 *
232 * The source will look like:
233 * <server_spec>[,<server_spec>...]:[<path>]
234 * where
235 * <server_spec> is <ip>[:<port>]
236 * <path> is optional, but if present must begin with '/'
237 */
238static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
239{
240 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
241 struct ceph_mount_options *fsopt = pctx->opts;
242 char *dev_name = param->string, *dev_name_end;
243 int ret;
244
245 dout("%s '%s'\n", __func__, dev_name);
246 if (!dev_name || !*dev_name)
247 return invalfc(fc, "Empty source");
248
249 dev_name_end = strchr(dev_name, '/');
250 if (dev_name_end) {
251 /*
252 * The server_path will include the whole chars from userland
253 * including the leading '/'.
254 */
255 kfree(fsopt->server_path);
256 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
257 if (!fsopt->server_path)
258 return -ENOMEM;
259
260 canonicalize_path(fsopt->server_path);
261 } else {
262 dev_name_end = dev_name + strlen(dev_name);
263 }
264
265 dev_name_end--; /* back up to ':' separator */
266 if (dev_name_end < dev_name || *dev_name_end != ':')
267 return invalfc(fc, "No path or : separator in source");
268
269 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
270 if (fsopt->server_path)
271 dout("server path '%s'\n", fsopt->server_path);
272
273 ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
274 pctx->copts, fc->log.log);
275 if (ret)
276 return ret;
277
278 fc->source = param->string;
279 param->string = NULL;
280 return 0;
281}
282
283static int ceph_parse_mount_param(struct fs_context *fc,
284 struct fs_parameter *param)
285{
286 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
287 struct ceph_mount_options *fsopt = pctx->opts;
288 struct fs_parse_result result;
289 unsigned int mode;
290 int token, ret;
291
292 ret = ceph_parse_param(param, pctx->copts, fc->log.log);
293 if (ret != -ENOPARAM)
294 return ret;
295
296 token = fs_parse(fc, ceph_mount_parameters, param, &result);
297 dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
298 if (token < 0)
299 return token;
300
301 switch (token) {
302 case Opt_snapdirname:
303 kfree(fsopt->snapdir_name);
304 fsopt->snapdir_name = param->string;
305 param->string = NULL;
306 break;
307 case Opt_mds_namespace:
308 kfree(fsopt->mds_namespace);
309 fsopt->mds_namespace = param->string;
310 param->string = NULL;
311 break;
312 case Opt_recover_session:
313 mode = result.uint_32;
314 if (mode == ceph_recover_session_no)
315 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
316 else if (mode == ceph_recover_session_clean)
317 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
318 else
319 BUG();
320 break;
321 case Opt_source:
322 if (fc->source)
323 return invalfc(fc, "Multiple sources specified");
324 return ceph_parse_source(param, fc);
325 case Opt_wsize:
326 if (result.uint_32 < PAGE_SIZE ||
327 result.uint_32 > CEPH_MAX_WRITE_SIZE)
328 goto out_of_range;
329 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
330 break;
331 case Opt_rsize:
332 if (result.uint_32 < PAGE_SIZE ||
333 result.uint_32 > CEPH_MAX_READ_SIZE)
334 goto out_of_range;
335 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
336 break;
337 case Opt_rasize:
338 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
339 break;
340 case Opt_caps_wanted_delay_min:
341 if (result.uint_32 < 1)
342 goto out_of_range;
343 fsopt->caps_wanted_delay_min = result.uint_32;
344 break;
345 case Opt_caps_wanted_delay_max:
346 if (result.uint_32 < 1)
347 goto out_of_range;
348 fsopt->caps_wanted_delay_max = result.uint_32;
349 break;
350 case Opt_caps_max:
351 if (result.int_32 < 0)
352 goto out_of_range;
353 fsopt->caps_max = result.int_32;
354 break;
355 case Opt_readdir_max_entries:
356 if (result.uint_32 < 1)
357 goto out_of_range;
358 fsopt->max_readdir = result.uint_32;
359 break;
360 case Opt_readdir_max_bytes:
361 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
362 goto out_of_range;
363 fsopt->max_readdir_bytes = result.uint_32;
364 break;
365 case Opt_congestion_kb:
366 if (result.uint_32 < 1024) /* at least 1M */
367 goto out_of_range;
368 fsopt->congestion_kb = result.uint_32;
369 break;
370 case Opt_dirstat:
371 if (!result.negated)
372 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
373 else
374 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
375 break;
376 case Opt_rbytes:
377 if (!result.negated)
378 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
379 else
380 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
381 break;
382 case Opt_asyncreaddir:
383 if (!result.negated)
384 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
385 else
386 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
387 break;
388 case Opt_dcache:
389 if (!result.negated)
390 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
391 else
392 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
393 break;
394 case Opt_ino32:
395 if (!result.negated)
396 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
397 else
398 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
399 break;
400
401 case Opt_fscache:
402#ifdef CONFIG_CEPH_FSCACHE
403 kfree(fsopt->fscache_uniq);
404 fsopt->fscache_uniq = NULL;
405 if (result.negated) {
406 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
407 } else {
408 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
409 fsopt->fscache_uniq = param->string;
410 param->string = NULL;
411 }
412 break;
413#else
414 return invalfc(fc, "fscache support is disabled");
415#endif
416 case Opt_poolperm:
417 if (!result.negated)
418 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
419 else
420 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
421 break;
422 case Opt_require_active_mds:
423 if (!result.negated)
424 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
425 else
426 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
427 break;
428 case Opt_quotadf:
429 if (!result.negated)
430 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
431 else
432 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
433 break;
434 case Opt_copyfrom:
435 if (!result.negated)
436 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
437 else
438 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
439 break;
440 case Opt_acl:
441 if (!result.negated) {
442#ifdef CONFIG_CEPH_FS_POSIX_ACL
443 fc->sb_flags |= SB_POSIXACL;
444#else
445 return invalfc(fc, "POSIX ACL support is disabled");
446#endif
447 } else {
448 fc->sb_flags &= ~SB_POSIXACL;
449 }
450 break;
451 case Opt_wsync:
452 if (!result.negated)
453 fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
454 else
455 fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
456 break;
457 default:
458 BUG();
459 }
460 return 0;
461
462out_of_range:
463 return invalfc(fc, "%s out of range", param->key);
464}
465
466static void destroy_mount_options(struct ceph_mount_options *args)
467{
468 dout("destroy_mount_options %p\n", args);
469 if (!args)
470 return;
471
472 kfree(args->snapdir_name);
473 kfree(args->mds_namespace);
474 kfree(args->server_path);
475 kfree(args->fscache_uniq);
476 kfree(args);
477}
478
479static int strcmp_null(const char *s1, const char *s2)
480{
481 if (!s1 && !s2)
482 return 0;
483 if (s1 && !s2)
484 return -1;
485 if (!s1 && s2)
486 return 1;
487 return strcmp(s1, s2);
488}
489
490static int compare_mount_options(struct ceph_mount_options *new_fsopt,
491 struct ceph_options *new_opt,
492 struct ceph_fs_client *fsc)
493{
494 struct ceph_mount_options *fsopt1 = new_fsopt;
495 struct ceph_mount_options *fsopt2 = fsc->mount_options;
496 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
497 int ret;
498
499 ret = memcmp(fsopt1, fsopt2, ofs);
500 if (ret)
501 return ret;
502
503 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
504 if (ret)
505 return ret;
506
507 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
508 if (ret)
509 return ret;
510
511 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
512 if (ret)
513 return ret;
514
515 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
516 if (ret)
517 return ret;
518
519 return ceph_compare_options(new_opt, fsc->client);
520}
521
522/**
523 * ceph_show_options - Show mount options in /proc/mounts
524 * @m: seq_file to write to
525 * @root: root of that (sub)tree
526 */
527static int ceph_show_options(struct seq_file *m, struct dentry *root)
528{
529 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
530 struct ceph_mount_options *fsopt = fsc->mount_options;
531 size_t pos;
532 int ret;
533
534 /* a comma between MNT/MS and client options */
535 seq_putc(m, ',');
536 pos = m->count;
537
538 ret = ceph_print_client_options(m, fsc->client, false);
539 if (ret)
540 return ret;
541
542 /* retract our comma if no client options */
543 if (m->count == pos)
544 m->count--;
545
546 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
547 seq_puts(m, ",dirstat");
548 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
549 seq_puts(m, ",rbytes");
550 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
551 seq_puts(m, ",noasyncreaddir");
552 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
553 seq_puts(m, ",nodcache");
554 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
555 seq_puts(m, ",ino32");
556 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
557 seq_show_option(m, "fsc", fsopt->fscache_uniq);
558 }
559 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
560 seq_puts(m, ",nopoolperm");
561 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
562 seq_puts(m, ",noquotadf");
563
564#ifdef CONFIG_CEPH_FS_POSIX_ACL
565 if (root->d_sb->s_flags & SB_POSIXACL)
566 seq_puts(m, ",acl");
567 else
568 seq_puts(m, ",noacl");
569#endif
570
571 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
572 seq_puts(m, ",copyfrom");
573
574 if (fsopt->mds_namespace)
575 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
576
577 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
578 seq_show_option(m, "recover_session", "clean");
579
580 if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
581 seq_puts(m, ",nowsync");
582
583 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
584 seq_printf(m, ",wsize=%u", fsopt->wsize);
585 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
586 seq_printf(m, ",rsize=%u", fsopt->rsize);
587 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
588 seq_printf(m, ",rasize=%u", fsopt->rasize);
589 if (fsopt->congestion_kb != default_congestion_kb())
590 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
591 if (fsopt->caps_max)
592 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
593 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
594 seq_printf(m, ",caps_wanted_delay_min=%u",
595 fsopt->caps_wanted_delay_min);
596 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
597 seq_printf(m, ",caps_wanted_delay_max=%u",
598 fsopt->caps_wanted_delay_max);
599 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
600 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
601 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
602 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
603 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
604 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
605
606 return 0;
607}
608
609/*
610 * handle any mon messages the standard library doesn't understand.
611 * return error if we don't either.
612 */
613static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
614{
615 struct ceph_fs_client *fsc = client->private;
616 int type = le16_to_cpu(msg->hdr.type);
617
618 switch (type) {
619 case CEPH_MSG_MDS_MAP:
620 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
621 return 0;
622 case CEPH_MSG_FS_MAP_USER:
623 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
624 return 0;
625 default:
626 return -1;
627 }
628}
629
630/*
631 * create a new fs client
632 *
633 * Success or not, this function consumes @fsopt and @opt.
634 */
635static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
636 struct ceph_options *opt)
637{
638 struct ceph_fs_client *fsc;
639 int err;
640
641 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
642 if (!fsc) {
643 err = -ENOMEM;
644 goto fail;
645 }
646
647 fsc->client = ceph_create_client(opt, fsc);
648 if (IS_ERR(fsc->client)) {
649 err = PTR_ERR(fsc->client);
650 goto fail;
651 }
652 opt = NULL; /* fsc->client now owns this */
653
654 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
655 ceph_set_opt(fsc->client, ABORT_ON_FULL);
656
657 if (!fsopt->mds_namespace) {
658 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
659 0, true);
660 } else {
661 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
662 0, false);
663 }
664
665 fsc->mount_options = fsopt;
666
667 fsc->sb = NULL;
668 fsc->mount_state = CEPH_MOUNT_MOUNTING;
669 fsc->filp_gen = 1;
670 fsc->have_copy_from2 = true;
671
672 atomic_long_set(&fsc->writeback_count, 0);
673
674 err = -ENOMEM;
675 /*
676 * The number of concurrent works can be high but they don't need
677 * to be processed in parallel, limit concurrency.
678 */
679 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
680 if (!fsc->inode_wq)
681 goto fail_client;
682 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
683 if (!fsc->cap_wq)
684 goto fail_inode_wq;
685
686 spin_lock(&ceph_fsc_lock);
687 list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
688 spin_unlock(&ceph_fsc_lock);
689
690 return fsc;
691
692fail_inode_wq:
693 destroy_workqueue(fsc->inode_wq);
694fail_client:
695 ceph_destroy_client(fsc->client);
696fail:
697 kfree(fsc);
698 if (opt)
699 ceph_destroy_options(opt);
700 destroy_mount_options(fsopt);
701 return ERR_PTR(err);
702}
703
704static void flush_fs_workqueues(struct ceph_fs_client *fsc)
705{
706 flush_workqueue(fsc->inode_wq);
707 flush_workqueue(fsc->cap_wq);
708}
709
710static void destroy_fs_client(struct ceph_fs_client *fsc)
711{
712 dout("destroy_fs_client %p\n", fsc);
713
714 spin_lock(&ceph_fsc_lock);
715 list_del(&fsc->metric_wakeup);
716 spin_unlock(&ceph_fsc_lock);
717
718 ceph_mdsc_destroy(fsc);
719 destroy_workqueue(fsc->inode_wq);
720 destroy_workqueue(fsc->cap_wq);
721
722 destroy_mount_options(fsc->mount_options);
723
724 ceph_destroy_client(fsc->client);
725
726 kfree(fsc);
727 dout("destroy_fs_client %p done\n", fsc);
728}
729
730/*
731 * caches
732 */
733struct kmem_cache *ceph_inode_cachep;
734struct kmem_cache *ceph_cap_cachep;
735struct kmem_cache *ceph_cap_flush_cachep;
736struct kmem_cache *ceph_dentry_cachep;
737struct kmem_cache *ceph_file_cachep;
738struct kmem_cache *ceph_dir_file_cachep;
739struct kmem_cache *ceph_mds_request_cachep;
740mempool_t *ceph_wb_pagevec_pool;
741
742static void ceph_inode_init_once(void *foo)
743{
744 struct ceph_inode_info *ci = foo;
745 inode_init_once(&ci->vfs_inode);
746}
747
748static int __init init_caches(void)
749{
750 int error = -ENOMEM;
751
752 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
753 sizeof(struct ceph_inode_info),
754 __alignof__(struct ceph_inode_info),
755 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
756 SLAB_ACCOUNT, ceph_inode_init_once);
757 if (!ceph_inode_cachep)
758 return -ENOMEM;
759
760 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
761 if (!ceph_cap_cachep)
762 goto bad_cap;
763 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
764 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
765 if (!ceph_cap_flush_cachep)
766 goto bad_cap_flush;
767
768 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
769 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
770 if (!ceph_dentry_cachep)
771 goto bad_dentry;
772
773 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
774 if (!ceph_file_cachep)
775 goto bad_file;
776
777 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
778 if (!ceph_dir_file_cachep)
779 goto bad_dir_file;
780
781 ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
782 if (!ceph_mds_request_cachep)
783 goto bad_mds_req;
784
785 ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
786 if (!ceph_wb_pagevec_pool)
787 goto bad_pagevec_pool;
788
789 error = ceph_fscache_register();
790 if (error)
791 goto bad_fscache;
792
793 return 0;
794
795bad_fscache:
796 kmem_cache_destroy(ceph_mds_request_cachep);
797bad_pagevec_pool:
798 mempool_destroy(ceph_wb_pagevec_pool);
799bad_mds_req:
800 kmem_cache_destroy(ceph_dir_file_cachep);
801bad_dir_file:
802 kmem_cache_destroy(ceph_file_cachep);
803bad_file:
804 kmem_cache_destroy(ceph_dentry_cachep);
805bad_dentry:
806 kmem_cache_destroy(ceph_cap_flush_cachep);
807bad_cap_flush:
808 kmem_cache_destroy(ceph_cap_cachep);
809bad_cap:
810 kmem_cache_destroy(ceph_inode_cachep);
811 return error;
812}
813
814static void destroy_caches(void)
815{
816 /*
817 * Make sure all delayed rcu free inodes are flushed before we
818 * destroy cache.
819 */
820 rcu_barrier();
821
822 kmem_cache_destroy(ceph_inode_cachep);
823 kmem_cache_destroy(ceph_cap_cachep);
824 kmem_cache_destroy(ceph_cap_flush_cachep);
825 kmem_cache_destroy(ceph_dentry_cachep);
826 kmem_cache_destroy(ceph_file_cachep);
827 kmem_cache_destroy(ceph_dir_file_cachep);
828 kmem_cache_destroy(ceph_mds_request_cachep);
829 mempool_destroy(ceph_wb_pagevec_pool);
830
831 ceph_fscache_unregister();
832}
833
834static void __ceph_umount_begin(struct ceph_fs_client *fsc)
835{
836 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
837 ceph_mdsc_force_umount(fsc->mdsc);
838 fsc->filp_gen++; // invalidate open files
839}
840
841/*
842 * ceph_umount_begin - initiate forced umount. Tear down the
843 * mount, skipping steps that may hang while waiting for server(s).
844 */
845static void ceph_umount_begin(struct super_block *sb)
846{
847 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
848
849 dout("ceph_umount_begin - starting forced umount\n");
850 if (!fsc)
851 return;
852 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
853 __ceph_umount_begin(fsc);
854}
855
856static const struct super_operations ceph_super_ops = {
857 .alloc_inode = ceph_alloc_inode,
858 .free_inode = ceph_free_inode,
859 .write_inode = ceph_write_inode,
860 .drop_inode = generic_delete_inode,
861 .evict_inode = ceph_evict_inode,
862 .sync_fs = ceph_sync_fs,
863 .put_super = ceph_put_super,
864 .show_options = ceph_show_options,
865 .statfs = ceph_statfs,
866 .umount_begin = ceph_umount_begin,
867};
868
869/*
870 * Bootstrap mount by opening the root directory. Note the mount
871 * @started time from caller, and time out if this takes too long.
872 */
873static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
874 const char *path,
875 unsigned long started)
876{
877 struct ceph_mds_client *mdsc = fsc->mdsc;
878 struct ceph_mds_request *req = NULL;
879 int err;
880 struct dentry *root;
881
882 /* open dir */
883 dout("open_root_inode opening '%s'\n", path);
884 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
885 if (IS_ERR(req))
886 return ERR_CAST(req);
887 req->r_path1 = kstrdup(path, GFP_NOFS);
888 if (!req->r_path1) {
889 root = ERR_PTR(-ENOMEM);
890 goto out;
891 }
892
893 req->r_ino1.ino = CEPH_INO_ROOT;
894 req->r_ino1.snap = CEPH_NOSNAP;
895 req->r_started = started;
896 req->r_timeout = fsc->client->options->mount_timeout;
897 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
898 req->r_num_caps = 2;
899 err = ceph_mdsc_do_request(mdsc, NULL, req);
900 if (err == 0) {
901 struct inode *inode = req->r_target_inode;
902 req->r_target_inode = NULL;
903 dout("open_root_inode success\n");
904 root = d_make_root(inode);
905 if (!root) {
906 root = ERR_PTR(-ENOMEM);
907 goto out;
908 }
909 dout("open_root_inode success, root dentry is %p\n", root);
910 } else {
911 root = ERR_PTR(err);
912 }
913out:
914 ceph_mdsc_put_request(req);
915 return root;
916}
917
918/*
919 * mount: join the ceph cluster, and open root directory.
920 */
921static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
922 struct fs_context *fc)
923{
924 int err;
925 unsigned long started = jiffies; /* note the start time */
926 struct dentry *root;
927
928 dout("mount start %p\n", fsc);
929 mutex_lock(&fsc->client->mount_mutex);
930
931 if (!fsc->sb->s_root) {
932 const char *path = fsc->mount_options->server_path ?
933 fsc->mount_options->server_path + 1 : "";
934
935 err = __ceph_open_session(fsc->client, started);
936 if (err < 0)
937 goto out;
938
939 /* setup fscache */
940 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
941 err = ceph_fscache_register_fs(fsc, fc);
942 if (err < 0)
943 goto out;
944 }
945
946 dout("mount opening path '%s'\n", path);
947
948 ceph_fs_debugfs_init(fsc);
949
950 root = open_root_dentry(fsc, path, started);
951 if (IS_ERR(root)) {
952 err = PTR_ERR(root);
953 goto out;
954 }
955 fsc->sb->s_root = dget(root);
956 } else {
957 root = dget(fsc->sb->s_root);
958 }
959
960 fsc->mount_state = CEPH_MOUNT_MOUNTED;
961 dout("mount success\n");
962 mutex_unlock(&fsc->client->mount_mutex);
963 return root;
964
965out:
966 mutex_unlock(&fsc->client->mount_mutex);
967 return ERR_PTR(err);
968}
969
970static int ceph_set_super(struct super_block *s, struct fs_context *fc)
971{
972 struct ceph_fs_client *fsc = s->s_fs_info;
973 int ret;
974
975 dout("set_super %p\n", s);
976
977 s->s_maxbytes = MAX_LFS_FILESIZE;
978
979 s->s_xattr = ceph_xattr_handlers;
980 fsc->sb = s;
981 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
982
983 s->s_op = &ceph_super_ops;
984 s->s_d_op = &ceph_dentry_ops;
985 s->s_export_op = &ceph_export_ops;
986
987 s->s_time_gran = 1;
988 s->s_time_min = 0;
989 s->s_time_max = U32_MAX;
990
991 ret = set_anon_super_fc(s, fc);
992 if (ret != 0)
993 fsc->sb = NULL;
994 return ret;
995}
996
997/*
998 * share superblock if same fs AND options
999 */
1000static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1001{
1002 struct ceph_fs_client *new = fc->s_fs_info;
1003 struct ceph_mount_options *fsopt = new->mount_options;
1004 struct ceph_options *opt = new->client->options;
1005 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1006
1007 dout("ceph_compare_super %p\n", sb);
1008
1009 if (compare_mount_options(fsopt, opt, fsc)) {
1010 dout("monitor(s)/mount options don't match\n");
1011 return 0;
1012 }
1013 if ((opt->flags & CEPH_OPT_FSID) &&
1014 ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1015 dout("fsid doesn't match\n");
1016 return 0;
1017 }
1018 if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1019 dout("flags differ\n");
1020 return 0;
1021 }
1022
1023 if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1024 dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1025 return 0;
1026 }
1027
1028 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1029 dout("client has been forcibly unmounted\n");
1030 return 0;
1031 }
1032
1033 return 1;
1034}
1035
1036/*
1037 * construct our own bdi so we can control readahead, etc.
1038 */
1039static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1040
1041static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1042{
1043 int err;
1044
1045 err = super_setup_bdi_name(sb, "ceph-%ld",
1046 atomic_long_inc_return(&bdi_seq));
1047 if (err)
1048 return err;
1049
1050 /* set ra_pages based on rasize mount option? */
1051 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1052
1053 /* set io_pages based on max osd read size */
1054 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1055
1056 return 0;
1057}
1058
1059static int ceph_get_tree(struct fs_context *fc)
1060{
1061 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1062 struct super_block *sb;
1063 struct ceph_fs_client *fsc;
1064 struct dentry *res;
1065 int (*compare_super)(struct super_block *, struct fs_context *) =
1066 ceph_compare_super;
1067 int err;
1068
1069 dout("ceph_get_tree\n");
1070
1071 if (!fc->source)
1072 return invalfc(fc, "No source");
1073
1074 /* create client (which we may/may not use) */
1075 fsc = create_fs_client(pctx->opts, pctx->copts);
1076 pctx->opts = NULL;
1077 pctx->copts = NULL;
1078 if (IS_ERR(fsc)) {
1079 err = PTR_ERR(fsc);
1080 goto out_final;
1081 }
1082
1083 err = ceph_mdsc_init(fsc);
1084 if (err < 0)
1085 goto out;
1086
1087 if (ceph_test_opt(fsc->client, NOSHARE))
1088 compare_super = NULL;
1089
1090 fc->s_fs_info = fsc;
1091 sb = sget_fc(fc, compare_super, ceph_set_super);
1092 fc->s_fs_info = NULL;
1093 if (IS_ERR(sb)) {
1094 err = PTR_ERR(sb);
1095 goto out;
1096 }
1097
1098 if (ceph_sb_to_client(sb) != fsc) {
1099 destroy_fs_client(fsc);
1100 fsc = ceph_sb_to_client(sb);
1101 dout("get_sb got existing client %p\n", fsc);
1102 } else {
1103 dout("get_sb using new client %p\n", fsc);
1104 err = ceph_setup_bdi(sb, fsc);
1105 if (err < 0)
1106 goto out_splat;
1107 }
1108
1109 res = ceph_real_mount(fsc, fc);
1110 if (IS_ERR(res)) {
1111 err = PTR_ERR(res);
1112 goto out_splat;
1113 }
1114 dout("root %p inode %p ino %llx.%llx\n", res,
1115 d_inode(res), ceph_vinop(d_inode(res)));
1116 fc->root = fsc->sb->s_root;
1117 return 0;
1118
1119out_splat:
1120 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1121 pr_info("No mds server is up or the cluster is laggy\n");
1122 err = -EHOSTUNREACH;
1123 }
1124
1125 ceph_mdsc_close_sessions(fsc->mdsc);
1126 deactivate_locked_super(sb);
1127 goto out_final;
1128
1129out:
1130 destroy_fs_client(fsc);
1131out_final:
1132 dout("ceph_get_tree fail %d\n", err);
1133 return err;
1134}
1135
1136static void ceph_free_fc(struct fs_context *fc)
1137{
1138 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1139
1140 if (pctx) {
1141 destroy_mount_options(pctx->opts);
1142 ceph_destroy_options(pctx->copts);
1143 kfree(pctx);
1144 }
1145}
1146
1147static int ceph_reconfigure_fc(struct fs_context *fc)
1148{
1149 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1150 struct ceph_mount_options *fsopt = pctx->opts;
1151 struct ceph_fs_client *fsc = ceph_sb_to_client(fc->root->d_sb);
1152
1153 if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1154 ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1155 else
1156 ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1157
1158 sync_filesystem(fc->root->d_sb);
1159 return 0;
1160}
1161
1162static const struct fs_context_operations ceph_context_ops = {
1163 .free = ceph_free_fc,
1164 .parse_param = ceph_parse_mount_param,
1165 .get_tree = ceph_get_tree,
1166 .reconfigure = ceph_reconfigure_fc,
1167};
1168
1169/*
1170 * Set up the filesystem mount context.
1171 */
1172static int ceph_init_fs_context(struct fs_context *fc)
1173{
1174 struct ceph_parse_opts_ctx *pctx;
1175 struct ceph_mount_options *fsopt;
1176
1177 pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1178 if (!pctx)
1179 return -ENOMEM;
1180
1181 pctx->copts = ceph_alloc_options();
1182 if (!pctx->copts)
1183 goto nomem;
1184
1185 pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1186 if (!pctx->opts)
1187 goto nomem;
1188
1189 fsopt = pctx->opts;
1190 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1191
1192 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1193 fsopt->rsize = CEPH_MAX_READ_SIZE;
1194 fsopt->rasize = CEPH_RASIZE_DEFAULT;
1195 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1196 if (!fsopt->snapdir_name)
1197 goto nomem;
1198
1199 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1200 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1201 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1202 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1203 fsopt->congestion_kb = default_congestion_kb();
1204
1205#ifdef CONFIG_CEPH_FS_POSIX_ACL
1206 fc->sb_flags |= SB_POSIXACL;
1207#endif
1208
1209 fc->fs_private = pctx;
1210 fc->ops = &ceph_context_ops;
1211 return 0;
1212
1213nomem:
1214 destroy_mount_options(pctx->opts);
1215 ceph_destroy_options(pctx->copts);
1216 kfree(pctx);
1217 return -ENOMEM;
1218}
1219
1220static void ceph_kill_sb(struct super_block *s)
1221{
1222 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1223
1224 dout("kill_sb %p\n", s);
1225
1226 ceph_mdsc_pre_umount(fsc->mdsc);
1227 flush_fs_workqueues(fsc);
1228
1229 kill_anon_super(s);
1230
1231 fsc->client->extra_mon_dispatch = NULL;
1232 ceph_fs_debugfs_cleanup(fsc);
1233
1234 ceph_fscache_unregister_fs(fsc);
1235
1236 destroy_fs_client(fsc);
1237}
1238
1239static struct file_system_type ceph_fs_type = {
1240 .owner = THIS_MODULE,
1241 .name = "ceph",
1242 .init_fs_context = ceph_init_fs_context,
1243 .kill_sb = ceph_kill_sb,
1244 .fs_flags = FS_RENAME_DOES_D_MOVE,
1245};
1246MODULE_ALIAS_FS("ceph");
1247
1248int ceph_force_reconnect(struct super_block *sb)
1249{
1250 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1251 int err = 0;
1252
1253 fsc->mount_state = CEPH_MOUNT_RECOVER;
1254 __ceph_umount_begin(fsc);
1255
1256 /* Make sure all page caches get invalidated.
1257 * see remove_session_caps_cb() */
1258 flush_workqueue(fsc->inode_wq);
1259
1260 /* In case that we were blocklisted. This also reset
1261 * all mon/osd connections */
1262 ceph_reset_client_addr(fsc->client);
1263
1264 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1265
1266 fsc->blocklisted = false;
1267 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1268
1269 if (sb->s_root) {
1270 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1271 CEPH_STAT_CAP_INODE, true);
1272 }
1273 return err;
1274}
1275
1276static int __init init_ceph(void)
1277{
1278 int ret = init_caches();
1279 if (ret)
1280 goto out;
1281
1282 ceph_flock_init();
1283 ret = register_filesystem(&ceph_fs_type);
1284 if (ret)
1285 goto out_caches;
1286
1287 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1288
1289 return 0;
1290
1291out_caches:
1292 destroy_caches();
1293out:
1294 return ret;
1295}
1296
1297static void __exit exit_ceph(void)
1298{
1299 dout("exit_ceph\n");
1300 unregister_filesystem(&ceph_fs_type);
1301 destroy_caches();
1302}
1303
1304static int param_set_metrics(const char *val, const struct kernel_param *kp)
1305{
1306 struct ceph_fs_client *fsc;
1307 int ret;
1308
1309 ret = param_set_bool(val, kp);
1310 if (ret) {
1311 pr_err("Failed to parse sending metrics switch value '%s'\n",
1312 val);
1313 return ret;
1314 } else if (!disable_send_metrics) {
1315 // wake up all the mds clients
1316 spin_lock(&ceph_fsc_lock);
1317 list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1318 metric_schedule_delayed(&fsc->mdsc->metric);
1319 }
1320 spin_unlock(&ceph_fsc_lock);
1321 }
1322
1323 return 0;
1324}
1325
1326static const struct kernel_param_ops param_ops_metrics = {
1327 .set = param_set_metrics,
1328 .get = param_get_bool,
1329};
1330
1331bool disable_send_metrics = false;
1332module_param_cb(disable_send_metrics, ¶m_ops_metrics, &disable_send_metrics, 0644);
1333MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1334
1335module_init(init_ceph);
1336module_exit(exit_ceph);
1337
1338MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1339MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1340MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1341MODULE_DESCRIPTION("Ceph filesystem for Linux");
1342MODULE_LICENSE("GPL");