OSDN Git Service

fs_parse: handle optional arguments sanely
[tomoyo/tomoyo-test1.git] / fs / ceph / super.c
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
30 /*
31  * Ceph superblock operations
32  *
33  * Handle the basics of mounting, unmounting.
34  */
35
36 /*
37  * super ops
38  */
39 static void ceph_put_super(struct super_block *s)
40 {
41         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
42
43         dout("put_super\n");
44         ceph_mdsc_close_sessions(fsc->mdsc);
45 }
46
47 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
48 {
49         struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
50         struct ceph_mon_client *monc = &fsc->client->monc;
51         struct ceph_statfs st;
52         u64 fsid;
53         int err;
54         u64 data_pool;
55
56         if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
57                 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
58         } else {
59                 data_pool = CEPH_NOPOOL;
60         }
61
62         dout("statfs\n");
63         err = ceph_monc_do_statfs(monc, data_pool, &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          * NOTE: for the time being, we make bsize == frsize to humor
75          * not-yet-ancient versions of glibc that are broken.
76          * Someday, we will probably want to report a real block
77          * size...  whatever that may mean for a network file system!
78          */
79         buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
80         buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
81
82         /*
83          * By default use root quota for stats; fallback to overall filesystem
84          * usage if using 'noquotadf' mount option or if the root dir doesn't
85          * have max_bytes quota set.
86          */
87         if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
88             !ceph_quota_update_statfs(fsc, buf)) {
89                 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
90                 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
91                 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
92         }
93
94         buf->f_files = le64_to_cpu(st.num_objects);
95         buf->f_ffree = -1;
96         buf->f_namelen = NAME_MAX;
97
98         /* Must convert the fsid, for consistent values across arches */
99         mutex_lock(&monc->mutex);
100         fsid = le64_to_cpu(*(__le64 *)(&monc->monmap->fsid)) ^
101                le64_to_cpu(*((__le64 *)&monc->monmap->fsid + 1));
102         mutex_unlock(&monc->mutex);
103
104         buf->f_fsid.val[0] = fsid & 0xffffffff;
105         buf->f_fsid.val[1] = fsid >> 32;
106
107         return 0;
108 }
109
110
111 static int ceph_sync_fs(struct super_block *sb, int wait)
112 {
113         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
114
115         if (!wait) {
116                 dout("sync_fs (non-blocking)\n");
117                 ceph_flush_dirty_caps(fsc->mdsc);
118                 dout("sync_fs (non-blocking) done\n");
119                 return 0;
120         }
121
122         dout("sync_fs (blocking)\n");
123         ceph_osdc_sync(&fsc->client->osdc);
124         ceph_mdsc_sync(fsc->mdsc);
125         dout("sync_fs (blocking) done\n");
126         return 0;
127 }
128
129 /*
130  * mount options
131  */
132 enum {
133         Opt_wsize,
134         Opt_rsize,
135         Opt_rasize,
136         Opt_caps_wanted_delay_min,
137         Opt_caps_wanted_delay_max,
138         Opt_caps_max,
139         Opt_readdir_max_entries,
140         Opt_readdir_max_bytes,
141         Opt_congestion_kb,
142         /* int args above */
143         Opt_snapdirname,
144         Opt_mds_namespace,
145         Opt_recover_session,
146         Opt_source,
147         /* string args above */
148         Opt_dirstat,
149         Opt_rbytes,
150         Opt_asyncreaddir,
151         Opt_dcache,
152         Opt_ino32,
153         Opt_fscache,
154         Opt_poolperm,
155         Opt_require_active_mds,
156         Opt_acl,
157         Opt_quotadf,
158         Opt_copyfrom,
159 };
160
161 enum ceph_recover_session_mode {
162         ceph_recover_session_no,
163         ceph_recover_session_clean
164 };
165
166 static const struct constant_table ceph_param_recover[] = {
167         { "no",         ceph_recover_session_no },
168         { "clean",      ceph_recover_session_clean },
169         {}
170 };
171
172 static const struct fs_parameter_spec ceph_mount_parameters[] = {
173         fsparam_flag_no ("acl",                         Opt_acl),
174         fsparam_flag_no ("asyncreaddir",                Opt_asyncreaddir),
175         fsparam_s32     ("caps_max",                    Opt_caps_max),
176         fsparam_u32     ("caps_wanted_delay_max",       Opt_caps_wanted_delay_max),
177         fsparam_u32     ("caps_wanted_delay_min",       Opt_caps_wanted_delay_min),
178         fsparam_u32     ("write_congestion_kb",         Opt_congestion_kb),
179         fsparam_flag_no ("copyfrom",                    Opt_copyfrom),
180         fsparam_flag_no ("dcache",                      Opt_dcache),
181         fsparam_flag_no ("dirstat",                     Opt_dirstat),
182         fsparam_flag_no ("fsc",                         Opt_fscache), // fsc|nofsc
183         fsparam_string  ("fsc",                         Opt_fscache), // fsc=...
184         fsparam_flag_no ("ino32",                       Opt_ino32),
185         fsparam_string  ("mds_namespace",               Opt_mds_namespace),
186         fsparam_flag_no ("poolperm",                    Opt_poolperm),
187         fsparam_flag_no ("quotadf",                     Opt_quotadf),
188         fsparam_u32     ("rasize",                      Opt_rasize),
189         fsparam_flag_no ("rbytes",                      Opt_rbytes),
190         fsparam_u32     ("readdir_max_bytes",           Opt_readdir_max_bytes),
191         fsparam_u32     ("readdir_max_entries",         Opt_readdir_max_entries),
192         fsparam_enum    ("recover_session",             Opt_recover_session, ceph_param_recover),
193         fsparam_flag_no ("require_active_mds",          Opt_require_active_mds),
194         fsparam_u32     ("rsize",                       Opt_rsize),
195         fsparam_string  ("snapdirname",                 Opt_snapdirname),
196         fsparam_string  ("source",                      Opt_source),
197         fsparam_u32     ("wsize",                       Opt_wsize),
198         {}
199 };
200
201 struct ceph_parse_opts_ctx {
202         struct ceph_options             *copts;
203         struct ceph_mount_options       *opts;
204 };
205
206 /*
207  * Parse the source parameter.  Distinguish the server list from the path.
208  * Internally we do not include the leading '/' in the path.
209  *
210  * The source will look like:
211  *     <server_spec>[,<server_spec>...]:[<path>]
212  * where
213  *     <server_spec> is <ip>[:<port>]
214  *     <path> is optional, but if present must begin with '/'
215  */
216 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
217 {
218         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
219         struct ceph_mount_options *fsopt = pctx->opts;
220         char *dev_name = param->string, *dev_name_end;
221         int ret;
222
223         dout("%s '%s'\n", __func__, dev_name);
224         if (!dev_name || !*dev_name)
225                 return invalf(fc, "ceph: Empty source");
226
227         dev_name_end = strchr(dev_name, '/');
228         if (dev_name_end) {
229                 if (strlen(dev_name_end) > 1) {
230                         kfree(fsopt->server_path);
231                         fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
232                         if (!fsopt->server_path)
233                                 return -ENOMEM;
234                 }
235         } else {
236                 dev_name_end = dev_name + strlen(dev_name);
237         }
238
239         dev_name_end--;         /* back up to ':' separator */
240         if (dev_name_end < dev_name || *dev_name_end != ':')
241                 return invalf(fc, "ceph: No path or : separator in source");
242
243         dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
244         if (fsopt->server_path)
245                 dout("server path '%s'\n", fsopt->server_path);
246
247         ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
248                                  pctx->copts, fc->log.log);
249         if (ret)
250                 return ret;
251
252         fc->source = param->string;
253         param->string = NULL;
254         return 0;
255 }
256
257 static int ceph_parse_mount_param(struct fs_context *fc,
258                                   struct fs_parameter *param)
259 {
260         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
261         struct ceph_mount_options *fsopt = pctx->opts;
262         struct fs_parse_result result;
263         unsigned int mode;
264         int token, ret;
265
266         ret = ceph_parse_param(param, pctx->copts, fc->log.log);
267         if (ret != -ENOPARAM)
268                 return ret;
269
270         token = fs_parse(fc, ceph_mount_parameters, param, &result);
271         dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
272         if (token < 0)
273                 return token;
274
275         switch (token) {
276         case Opt_snapdirname:
277                 kfree(fsopt->snapdir_name);
278                 fsopt->snapdir_name = param->string;
279                 param->string = NULL;
280                 break;
281         case Opt_mds_namespace:
282                 kfree(fsopt->mds_namespace);
283                 fsopt->mds_namespace = param->string;
284                 param->string = NULL;
285                 break;
286         case Opt_recover_session:
287                 mode = result.uint_32;
288                 if (mode == ceph_recover_session_no)
289                         fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
290                 else if (mode == ceph_recover_session_clean)
291                         fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
292                 else
293                         BUG();
294                 break;
295         case Opt_source:
296                 if (fc->source)
297                         return invalf(fc, "ceph: Multiple sources specified");
298                 return ceph_parse_source(param, fc);
299         case Opt_wsize:
300                 if (result.uint_32 < PAGE_SIZE ||
301                     result.uint_32 > CEPH_MAX_WRITE_SIZE)
302                         goto out_of_range;
303                 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
304                 break;
305         case Opt_rsize:
306                 if (result.uint_32 < PAGE_SIZE ||
307                     result.uint_32 > CEPH_MAX_READ_SIZE)
308                         goto out_of_range;
309                 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
310                 break;
311         case Opt_rasize:
312                 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
313                 break;
314         case Opt_caps_wanted_delay_min:
315                 if (result.uint_32 < 1)
316                         goto out_of_range;
317                 fsopt->caps_wanted_delay_min = result.uint_32;
318                 break;
319         case Opt_caps_wanted_delay_max:
320                 if (result.uint_32 < 1)
321                         goto out_of_range;
322                 fsopt->caps_wanted_delay_max = result.uint_32;
323                 break;
324         case Opt_caps_max:
325                 if (result.int_32 < 0)
326                         goto out_of_range;
327                 fsopt->caps_max = result.int_32;
328                 break;
329         case Opt_readdir_max_entries:
330                 if (result.uint_32 < 1)
331                         goto out_of_range;
332                 fsopt->max_readdir = result.uint_32;
333                 break;
334         case Opt_readdir_max_bytes:
335                 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
336                         goto out_of_range;
337                 fsopt->max_readdir_bytes = result.uint_32;
338                 break;
339         case Opt_congestion_kb:
340                 if (result.uint_32 < 1024) /* at least 1M */
341                         goto out_of_range;
342                 fsopt->congestion_kb = result.uint_32;
343                 break;
344         case Opt_dirstat:
345                 if (!result.negated)
346                         fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
347                 else
348                         fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
349                 break;
350         case Opt_rbytes:
351                 if (!result.negated)
352                         fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
353                 else
354                         fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
355                 break;
356         case Opt_asyncreaddir:
357                 if (!result.negated)
358                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
359                 else
360                         fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
361                 break;
362         case Opt_dcache:
363                 if (!result.negated)
364                         fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
365                 else
366                         fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
367                 break;
368         case Opt_ino32:
369                 if (!result.negated)
370                         fsopt->flags |= CEPH_MOUNT_OPT_INO32;
371                 else
372                         fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
373                 break;
374
375         case Opt_fscache:
376 #ifdef CONFIG_CEPH_FSCACHE
377                 kfree(fsopt->fscache_uniq);
378                 fsopt->fscache_uniq = NULL;
379                 if (result.negated) {
380                         fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
381                 } else {
382                         fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
383                         fsopt->fscache_uniq = param->string;
384                         param->string = NULL;
385                 }
386                 break;
387 #else
388                 return invalf(fc, "ceph: fscache support is disabled");
389 #endif
390         case Opt_poolperm:
391                 if (!result.negated)
392                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
393                 else
394                         fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
395                 break;
396         case Opt_require_active_mds:
397                 if (!result.negated)
398                         fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
399                 else
400                         fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
401                 break;
402         case Opt_quotadf:
403                 if (!result.negated)
404                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
405                 else
406                         fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
407                 break;
408         case Opt_copyfrom:
409                 if (!result.negated)
410                         fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
411                 else
412                         fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
413                 break;
414         case Opt_acl:
415                 if (!result.negated) {
416 #ifdef CONFIG_CEPH_FS_POSIX_ACL
417                         fc->sb_flags |= SB_POSIXACL;
418 #else
419                         return invalf(fc, "ceph: POSIX ACL support is disabled");
420 #endif
421                 } else {
422                         fc->sb_flags &= ~SB_POSIXACL;
423                 }
424                 break;
425         default:
426                 BUG();
427         }
428         return 0;
429
430 out_of_range:
431         return invalf(fc, "ceph: %s out of range", param->key);
432 }
433
434 static void destroy_mount_options(struct ceph_mount_options *args)
435 {
436         dout("destroy_mount_options %p\n", args);
437         if (!args)
438                 return;
439
440         kfree(args->snapdir_name);
441         kfree(args->mds_namespace);
442         kfree(args->server_path);
443         kfree(args->fscache_uniq);
444         kfree(args);
445 }
446
447 static int strcmp_null(const char *s1, const char *s2)
448 {
449         if (!s1 && !s2)
450                 return 0;
451         if (s1 && !s2)
452                 return -1;
453         if (!s1 && s2)
454                 return 1;
455         return strcmp(s1, s2);
456 }
457
458 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
459                                  struct ceph_options *new_opt,
460                                  struct ceph_fs_client *fsc)
461 {
462         struct ceph_mount_options *fsopt1 = new_fsopt;
463         struct ceph_mount_options *fsopt2 = fsc->mount_options;
464         int ofs = offsetof(struct ceph_mount_options, snapdir_name);
465         int ret;
466
467         ret = memcmp(fsopt1, fsopt2, ofs);
468         if (ret)
469                 return ret;
470
471         ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
472         if (ret)
473                 return ret;
474         ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
475         if (ret)
476                 return ret;
477         ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
478         if (ret)
479                 return ret;
480         ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
481         if (ret)
482                 return ret;
483
484         return ceph_compare_options(new_opt, fsc->client);
485 }
486
487 /**
488  * ceph_show_options - Show mount options in /proc/mounts
489  * @m: seq_file to write to
490  * @root: root of that (sub)tree
491  */
492 static int ceph_show_options(struct seq_file *m, struct dentry *root)
493 {
494         struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
495         struct ceph_mount_options *fsopt = fsc->mount_options;
496         size_t pos;
497         int ret;
498
499         /* a comma between MNT/MS and client options */
500         seq_putc(m, ',');
501         pos = m->count;
502
503         ret = ceph_print_client_options(m, fsc->client, false);
504         if (ret)
505                 return ret;
506
507         /* retract our comma if no client options */
508         if (m->count == pos)
509                 m->count--;
510
511         if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
512                 seq_puts(m, ",dirstat");
513         if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
514                 seq_puts(m, ",rbytes");
515         if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
516                 seq_puts(m, ",noasyncreaddir");
517         if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
518                 seq_puts(m, ",nodcache");
519         if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
520                 seq_puts(m, ",ino32");
521         if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
522                 seq_show_option(m, "fsc", fsopt->fscache_uniq);
523         }
524         if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
525                 seq_puts(m, ",nopoolperm");
526         if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
527                 seq_puts(m, ",noquotadf");
528
529 #ifdef CONFIG_CEPH_FS_POSIX_ACL
530         if (root->d_sb->s_flags & SB_POSIXACL)
531                 seq_puts(m, ",acl");
532         else
533                 seq_puts(m, ",noacl");
534 #endif
535
536         if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
537                 seq_puts(m, ",copyfrom");
538
539         if (fsopt->mds_namespace)
540                 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
541
542         if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
543                 seq_show_option(m, "recover_session", "clean");
544
545         if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
546                 seq_printf(m, ",wsize=%u", fsopt->wsize);
547         if (fsopt->rsize != CEPH_MAX_READ_SIZE)
548                 seq_printf(m, ",rsize=%u", fsopt->rsize);
549         if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
550                 seq_printf(m, ",rasize=%u", fsopt->rasize);
551         if (fsopt->congestion_kb != default_congestion_kb())
552                 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
553         if (fsopt->caps_max)
554                 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
555         if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
556                 seq_printf(m, ",caps_wanted_delay_min=%u",
557                          fsopt->caps_wanted_delay_min);
558         if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
559                 seq_printf(m, ",caps_wanted_delay_max=%u",
560                            fsopt->caps_wanted_delay_max);
561         if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
562                 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
563         if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
564                 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
565         if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
566                 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
567
568         return 0;
569 }
570
571 /*
572  * handle any mon messages the standard library doesn't understand.
573  * return error if we don't either.
574  */
575 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
576 {
577         struct ceph_fs_client *fsc = client->private;
578         int type = le16_to_cpu(msg->hdr.type);
579
580         switch (type) {
581         case CEPH_MSG_MDS_MAP:
582                 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
583                 return 0;
584         case CEPH_MSG_FS_MAP_USER:
585                 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
586                 return 0;
587         default:
588                 return -1;
589         }
590 }
591
592 /*
593  * create a new fs client
594  *
595  * Success or not, this function consumes @fsopt and @opt.
596  */
597 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
598                                         struct ceph_options *opt)
599 {
600         struct ceph_fs_client *fsc;
601         int page_count;
602         size_t size;
603         int err;
604
605         fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
606         if (!fsc) {
607                 err = -ENOMEM;
608                 goto fail;
609         }
610
611         fsc->client = ceph_create_client(opt, fsc);
612         if (IS_ERR(fsc->client)) {
613                 err = PTR_ERR(fsc->client);
614                 goto fail;
615         }
616         opt = NULL; /* fsc->client now owns this */
617
618         fsc->client->extra_mon_dispatch = extra_mon_dispatch;
619         ceph_set_opt(fsc->client, ABORT_ON_FULL);
620
621         if (!fsopt->mds_namespace) {
622                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
623                                    0, true);
624         } else {
625                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
626                                    0, false);
627         }
628
629         fsc->mount_options = fsopt;
630
631         fsc->sb = NULL;
632         fsc->mount_state = CEPH_MOUNT_MOUNTING;
633         fsc->filp_gen = 1;
634
635         atomic_long_set(&fsc->writeback_count, 0);
636
637         err = -ENOMEM;
638         /*
639          * The number of concurrent works can be high but they don't need
640          * to be processed in parallel, limit concurrency.
641          */
642         fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
643         if (!fsc->inode_wq)
644                 goto fail_client;
645         fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
646         if (!fsc->cap_wq)
647                 goto fail_inode_wq;
648
649         /* set up mempools */
650         err = -ENOMEM;
651         page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
652         size = sizeof (struct page *) * (page_count ? page_count : 1);
653         fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
654         if (!fsc->wb_pagevec_pool)
655                 goto fail_cap_wq;
656
657         return fsc;
658
659 fail_cap_wq:
660         destroy_workqueue(fsc->cap_wq);
661 fail_inode_wq:
662         destroy_workqueue(fsc->inode_wq);
663 fail_client:
664         ceph_destroy_client(fsc->client);
665 fail:
666         kfree(fsc);
667         if (opt)
668                 ceph_destroy_options(opt);
669         destroy_mount_options(fsopt);
670         return ERR_PTR(err);
671 }
672
673 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
674 {
675         flush_workqueue(fsc->inode_wq);
676         flush_workqueue(fsc->cap_wq);
677 }
678
679 static void destroy_fs_client(struct ceph_fs_client *fsc)
680 {
681         dout("destroy_fs_client %p\n", fsc);
682
683         ceph_mdsc_destroy(fsc);
684         destroy_workqueue(fsc->inode_wq);
685         destroy_workqueue(fsc->cap_wq);
686
687         mempool_destroy(fsc->wb_pagevec_pool);
688
689         destroy_mount_options(fsc->mount_options);
690
691         ceph_destroy_client(fsc->client);
692
693         kfree(fsc);
694         dout("destroy_fs_client %p done\n", fsc);
695 }
696
697 /*
698  * caches
699  */
700 struct kmem_cache *ceph_inode_cachep;
701 struct kmem_cache *ceph_cap_cachep;
702 struct kmem_cache *ceph_cap_flush_cachep;
703 struct kmem_cache *ceph_dentry_cachep;
704 struct kmem_cache *ceph_file_cachep;
705 struct kmem_cache *ceph_dir_file_cachep;
706
707 static void ceph_inode_init_once(void *foo)
708 {
709         struct ceph_inode_info *ci = foo;
710         inode_init_once(&ci->vfs_inode);
711 }
712
713 static int __init init_caches(void)
714 {
715         int error = -ENOMEM;
716
717         ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
718                                       sizeof(struct ceph_inode_info),
719                                       __alignof__(struct ceph_inode_info),
720                                       SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
721                                       SLAB_ACCOUNT, ceph_inode_init_once);
722         if (!ceph_inode_cachep)
723                 return -ENOMEM;
724
725         ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
726         if (!ceph_cap_cachep)
727                 goto bad_cap;
728         ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
729                                            SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
730         if (!ceph_cap_flush_cachep)
731                 goto bad_cap_flush;
732
733         ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
734                                         SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
735         if (!ceph_dentry_cachep)
736                 goto bad_dentry;
737
738         ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
739         if (!ceph_file_cachep)
740                 goto bad_file;
741
742         ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
743         if (!ceph_dir_file_cachep)
744                 goto bad_dir_file;
745
746         error = ceph_fscache_register();
747         if (error)
748                 goto bad_fscache;
749
750         return 0;
751
752 bad_fscache:
753         kmem_cache_destroy(ceph_dir_file_cachep);
754 bad_dir_file:
755         kmem_cache_destroy(ceph_file_cachep);
756 bad_file:
757         kmem_cache_destroy(ceph_dentry_cachep);
758 bad_dentry:
759         kmem_cache_destroy(ceph_cap_flush_cachep);
760 bad_cap_flush:
761         kmem_cache_destroy(ceph_cap_cachep);
762 bad_cap:
763         kmem_cache_destroy(ceph_inode_cachep);
764         return error;
765 }
766
767 static void destroy_caches(void)
768 {
769         /*
770          * Make sure all delayed rcu free inodes are flushed before we
771          * destroy cache.
772          */
773         rcu_barrier();
774
775         kmem_cache_destroy(ceph_inode_cachep);
776         kmem_cache_destroy(ceph_cap_cachep);
777         kmem_cache_destroy(ceph_cap_flush_cachep);
778         kmem_cache_destroy(ceph_dentry_cachep);
779         kmem_cache_destroy(ceph_file_cachep);
780         kmem_cache_destroy(ceph_dir_file_cachep);
781
782         ceph_fscache_unregister();
783 }
784
785
786 /*
787  * ceph_umount_begin - initiate forced umount.  Tear down down the
788  * mount, skipping steps that may hang while waiting for server(s).
789  */
790 static void ceph_umount_begin(struct super_block *sb)
791 {
792         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
793
794         dout("ceph_umount_begin - starting forced umount\n");
795         if (!fsc)
796                 return;
797         fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
798         ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
799         ceph_mdsc_force_umount(fsc->mdsc);
800         fsc->filp_gen++; // invalidate open files
801 }
802
803 static const struct super_operations ceph_super_ops = {
804         .alloc_inode    = ceph_alloc_inode,
805         .free_inode     = ceph_free_inode,
806         .write_inode    = ceph_write_inode,
807         .drop_inode     = generic_delete_inode,
808         .evict_inode    = ceph_evict_inode,
809         .sync_fs        = ceph_sync_fs,
810         .put_super      = ceph_put_super,
811         .show_options   = ceph_show_options,
812         .statfs         = ceph_statfs,
813         .umount_begin   = ceph_umount_begin,
814 };
815
816 /*
817  * Bootstrap mount by opening the root directory.  Note the mount
818  * @started time from caller, and time out if this takes too long.
819  */
820 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
821                                        const char *path,
822                                        unsigned long started)
823 {
824         struct ceph_mds_client *mdsc = fsc->mdsc;
825         struct ceph_mds_request *req = NULL;
826         int err;
827         struct dentry *root;
828
829         /* open dir */
830         dout("open_root_inode opening '%s'\n", path);
831         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
832         if (IS_ERR(req))
833                 return ERR_CAST(req);
834         req->r_path1 = kstrdup(path, GFP_NOFS);
835         if (!req->r_path1) {
836                 root = ERR_PTR(-ENOMEM);
837                 goto out;
838         }
839
840         req->r_ino1.ino = CEPH_INO_ROOT;
841         req->r_ino1.snap = CEPH_NOSNAP;
842         req->r_started = started;
843         req->r_timeout = fsc->client->options->mount_timeout;
844         req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
845         req->r_num_caps = 2;
846         err = ceph_mdsc_do_request(mdsc, NULL, req);
847         if (err == 0) {
848                 struct inode *inode = req->r_target_inode;
849                 req->r_target_inode = NULL;
850                 dout("open_root_inode success\n");
851                 root = d_make_root(inode);
852                 if (!root) {
853                         root = ERR_PTR(-ENOMEM);
854                         goto out;
855                 }
856                 dout("open_root_inode success, root dentry is %p\n", root);
857         } else {
858                 root = ERR_PTR(err);
859         }
860 out:
861         ceph_mdsc_put_request(req);
862         return root;
863 }
864
865
866
867
868 /*
869  * mount: join the ceph cluster, and open root directory.
870  */
871 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
872                                       struct fs_context *fc)
873 {
874         int err;
875         unsigned long started = jiffies;  /* note the start time */
876         struct dentry *root;
877
878         dout("mount start %p\n", fsc);
879         mutex_lock(&fsc->client->mount_mutex);
880
881         if (!fsc->sb->s_root) {
882                 const char *path;
883                 err = __ceph_open_session(fsc->client, started);
884                 if (err < 0)
885                         goto out;
886
887                 /* setup fscache */
888                 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
889                         err = ceph_fscache_register_fs(fsc, fc);
890                         if (err < 0)
891                                 goto out;
892                 }
893
894                 if (!fsc->mount_options->server_path) {
895                         path = "";
896                         dout("mount opening path \\t\n");
897                 } else {
898                         path = fsc->mount_options->server_path + 1;
899                         dout("mount opening path %s\n", path);
900                 }
901
902                 ceph_fs_debugfs_init(fsc);
903
904                 root = open_root_dentry(fsc, path, started);
905                 if (IS_ERR(root)) {
906                         err = PTR_ERR(root);
907                         goto out;
908                 }
909                 fsc->sb->s_root = dget(root);
910         } else {
911                 root = dget(fsc->sb->s_root);
912         }
913
914         fsc->mount_state = CEPH_MOUNT_MOUNTED;
915         dout("mount success\n");
916         mutex_unlock(&fsc->client->mount_mutex);
917         return root;
918
919 out:
920         mutex_unlock(&fsc->client->mount_mutex);
921         return ERR_PTR(err);
922 }
923
924 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
925 {
926         struct ceph_fs_client *fsc = s->s_fs_info;
927         int ret;
928
929         dout("set_super %p\n", s);
930
931         s->s_maxbytes = MAX_LFS_FILESIZE;
932
933         s->s_xattr = ceph_xattr_handlers;
934         fsc->sb = s;
935         fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
936
937         s->s_op = &ceph_super_ops;
938         s->s_d_op = &ceph_dentry_ops;
939         s->s_export_op = &ceph_export_ops;
940
941         s->s_time_gran = 1;
942         s->s_time_min = 0;
943         s->s_time_max = U32_MAX;
944
945         ret = set_anon_super_fc(s, fc);
946         if (ret != 0)
947                 fsc->sb = NULL;
948         return ret;
949 }
950
951 /*
952  * share superblock if same fs AND options
953  */
954 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
955 {
956         struct ceph_fs_client *new = fc->s_fs_info;
957         struct ceph_mount_options *fsopt = new->mount_options;
958         struct ceph_options *opt = new->client->options;
959         struct ceph_fs_client *other = ceph_sb_to_client(sb);
960
961         dout("ceph_compare_super %p\n", sb);
962
963         if (compare_mount_options(fsopt, opt, other)) {
964                 dout("monitor(s)/mount options don't match\n");
965                 return 0;
966         }
967         if ((opt->flags & CEPH_OPT_FSID) &&
968             ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
969                 dout("fsid doesn't match\n");
970                 return 0;
971         }
972         if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
973                 dout("flags differ\n");
974                 return 0;
975         }
976         return 1;
977 }
978
979 /*
980  * construct our own bdi so we can control readahead, etc.
981  */
982 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
983
984 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
985 {
986         int err;
987
988         err = super_setup_bdi_name(sb, "ceph-%ld",
989                                    atomic_long_inc_return(&bdi_seq));
990         if (err)
991                 return err;
992
993         /* set ra_pages based on rasize mount option? */
994         sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
995
996         /* set io_pages based on max osd read size */
997         sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
998
999         return 0;
1000 }
1001
1002 static int ceph_get_tree(struct fs_context *fc)
1003 {
1004         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1005         struct super_block *sb;
1006         struct ceph_fs_client *fsc;
1007         struct dentry *res;
1008         int (*compare_super)(struct super_block *, struct fs_context *) =
1009                 ceph_compare_super;
1010         int err;
1011
1012         dout("ceph_get_tree\n");
1013
1014         if (!fc->source)
1015                 return invalf(fc, "ceph: No source");
1016
1017 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1018         fc->sb_flags |= SB_POSIXACL;
1019 #endif
1020
1021         /* create client (which we may/may not use) */
1022         fsc = create_fs_client(pctx->opts, pctx->copts);
1023         pctx->opts = NULL;
1024         pctx->copts = NULL;
1025         if (IS_ERR(fsc)) {
1026                 err = PTR_ERR(fsc);
1027                 goto out_final;
1028         }
1029
1030         err = ceph_mdsc_init(fsc);
1031         if (err < 0)
1032                 goto out;
1033
1034         if (ceph_test_opt(fsc->client, NOSHARE))
1035                 compare_super = NULL;
1036
1037         fc->s_fs_info = fsc;
1038         sb = sget_fc(fc, compare_super, ceph_set_super);
1039         fc->s_fs_info = NULL;
1040         if (IS_ERR(sb)) {
1041                 err = PTR_ERR(sb);
1042                 goto out;
1043         }
1044
1045         if (ceph_sb_to_client(sb) != fsc) {
1046                 destroy_fs_client(fsc);
1047                 fsc = ceph_sb_to_client(sb);
1048                 dout("get_sb got existing client %p\n", fsc);
1049         } else {
1050                 dout("get_sb using new client %p\n", fsc);
1051                 err = ceph_setup_bdi(sb, fsc);
1052                 if (err < 0)
1053                         goto out_splat;
1054         }
1055
1056         res = ceph_real_mount(fsc, fc);
1057         if (IS_ERR(res)) {
1058                 err = PTR_ERR(res);
1059                 goto out_splat;
1060         }
1061         dout("root %p inode %p ino %llx.%llx\n", res,
1062              d_inode(res), ceph_vinop(d_inode(res)));
1063         fc->root = fsc->sb->s_root;
1064         return 0;
1065
1066 out_splat:
1067         ceph_mdsc_close_sessions(fsc->mdsc);
1068         deactivate_locked_super(sb);
1069         goto out_final;
1070
1071 out:
1072         destroy_fs_client(fsc);
1073 out_final:
1074         dout("ceph_get_tree fail %d\n", err);
1075         return err;
1076 }
1077
1078 static void ceph_free_fc(struct fs_context *fc)
1079 {
1080         struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1081
1082         if (pctx) {
1083                 destroy_mount_options(pctx->opts);
1084                 ceph_destroy_options(pctx->copts);
1085                 kfree(pctx);
1086         }
1087 }
1088
1089 static int ceph_reconfigure_fc(struct fs_context *fc)
1090 {
1091         sync_filesystem(fc->root->d_sb);
1092         return 0;
1093 }
1094
1095 static const struct fs_context_operations ceph_context_ops = {
1096         .free           = ceph_free_fc,
1097         .parse_param    = ceph_parse_mount_param,
1098         .get_tree       = ceph_get_tree,
1099         .reconfigure    = ceph_reconfigure_fc,
1100 };
1101
1102 /*
1103  * Set up the filesystem mount context.
1104  */
1105 static int ceph_init_fs_context(struct fs_context *fc)
1106 {
1107         struct ceph_parse_opts_ctx *pctx;
1108         struct ceph_mount_options *fsopt;
1109
1110         pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1111         if (!pctx)
1112                 return -ENOMEM;
1113
1114         pctx->copts = ceph_alloc_options();
1115         if (!pctx->copts)
1116                 goto nomem;
1117
1118         pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1119         if (!pctx->opts)
1120                 goto nomem;
1121
1122         fsopt = pctx->opts;
1123         fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1124
1125         fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1126         fsopt->rsize = CEPH_MAX_READ_SIZE;
1127         fsopt->rasize = CEPH_RASIZE_DEFAULT;
1128         fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1129         if (!fsopt->snapdir_name)
1130                 goto nomem;
1131
1132         fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1133         fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1134         fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1135         fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1136         fsopt->congestion_kb = default_congestion_kb();
1137
1138         fc->fs_private = pctx;
1139         fc->ops = &ceph_context_ops;
1140         return 0;
1141
1142 nomem:
1143         destroy_mount_options(pctx->opts);
1144         ceph_destroy_options(pctx->copts);
1145         kfree(pctx);
1146         return -ENOMEM;
1147 }
1148
1149 static void ceph_kill_sb(struct super_block *s)
1150 {
1151         struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1152         dev_t dev = s->s_dev;
1153
1154         dout("kill_sb %p\n", s);
1155
1156         ceph_mdsc_pre_umount(fsc->mdsc);
1157         flush_fs_workqueues(fsc);
1158
1159         generic_shutdown_super(s);
1160
1161         fsc->client->extra_mon_dispatch = NULL;
1162         ceph_fs_debugfs_cleanup(fsc);
1163
1164         ceph_fscache_unregister_fs(fsc);
1165
1166         destroy_fs_client(fsc);
1167         free_anon_bdev(dev);
1168 }
1169
1170 static struct file_system_type ceph_fs_type = {
1171         .owner          = THIS_MODULE,
1172         .name           = "ceph",
1173         .init_fs_context = ceph_init_fs_context,
1174         .kill_sb        = ceph_kill_sb,
1175         .fs_flags       = FS_RENAME_DOES_D_MOVE,
1176 };
1177 MODULE_ALIAS_FS("ceph");
1178
1179 int ceph_force_reconnect(struct super_block *sb)
1180 {
1181         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1182         int err = 0;
1183
1184         ceph_umount_begin(sb);
1185
1186         /* Make sure all page caches get invalidated.
1187          * see remove_session_caps_cb() */
1188         flush_workqueue(fsc->inode_wq);
1189
1190         /* In case that we were blacklisted. This also reset
1191          * all mon/osd connections */
1192         ceph_reset_client_addr(fsc->client);
1193
1194         ceph_osdc_clear_abort_err(&fsc->client->osdc);
1195
1196         fsc->blacklisted = false;
1197         fsc->mount_state = CEPH_MOUNT_MOUNTED;
1198
1199         if (sb->s_root) {
1200                 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1201                                         CEPH_STAT_CAP_INODE, true);
1202         }
1203         return err;
1204 }
1205
1206 static int __init init_ceph(void)
1207 {
1208         int ret = init_caches();
1209         if (ret)
1210                 goto out;
1211
1212         ceph_flock_init();
1213         ret = register_filesystem(&ceph_fs_type);
1214         if (ret)
1215                 goto out_caches;
1216
1217         pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1218
1219         return 0;
1220
1221 out_caches:
1222         destroy_caches();
1223 out:
1224         return ret;
1225 }
1226
1227 static void __exit exit_ceph(void)
1228 {
1229         dout("exit_ceph\n");
1230         unregister_filesystem(&ceph_fs_type);
1231         destroy_caches();
1232 }
1233
1234 module_init(init_ceph);
1235 module_exit(exit_ceph);
1236
1237 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1238 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1239 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1240 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1241 MODULE_LICENSE("GPL");