1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/ceph/ceph_debug.h>
5 #include <linux/backing-dev.h>
6 #include <linux/ctype.h>
8 #include <linux/inet.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>
21 #include "mds_client.h"
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>
31 * Ceph superblock operations
33 * Handle the basics of mounting, unmounting.
39 static void ceph_put_super(struct super_block *s)
41 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
44 ceph_mdsc_close_sessions(fsc->mdsc);
47 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
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;
56 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
57 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
59 data_pool = CEPH_NOPOOL;
63 err = ceph_monc_do_statfs(monc, data_pool, &st);
68 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
71 * express utilization in terms of large blocks to avoid
72 * overflow on 32-bit machines.
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!
79 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
80 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
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.
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);
94 buf->f_files = le64_to_cpu(st.num_objects);
96 buf->f_namelen = NAME_MAX;
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);
104 buf->f_fsid.val[0] = fsid & 0xffffffff;
105 buf->f_fsid.val[1] = fsid >> 32;
110 static int ceph_sync_fs(struct super_block *sb, int wait)
112 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
115 dout("sync_fs (non-blocking)\n");
116 ceph_flush_dirty_caps(fsc->mdsc);
117 dout("sync_fs (non-blocking) done\n");
121 dout("sync_fs (blocking)\n");
122 ceph_osdc_sync(&fsc->client->osdc);
123 ceph_mdsc_sync(fsc->mdsc);
124 dout("sync_fs (blocking) done\n");
135 Opt_caps_wanted_delay_min,
136 Opt_caps_wanted_delay_max,
138 Opt_readdir_max_entries,
139 Opt_readdir_max_bytes,
146 /* string args above */
154 Opt_require_active_mds,
160 enum ceph_recover_session_mode {
161 ceph_recover_session_no,
162 ceph_recover_session_clean
165 static const struct fs_parameter_enum ceph_mount_param_enums[] = {
166 { Opt_recover_session, "no", ceph_recover_session_no },
167 { Opt_recover_session, "clean", ceph_recover_session_clean },
171 static const struct fs_parameter_spec ceph_mount_param_specs[] = {
172 fsparam_flag_no ("acl", Opt_acl),
173 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir),
174 fsparam_s32 ("caps_max", Opt_caps_max),
175 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max),
176 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min),
177 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb),
178 fsparam_flag_no ("copyfrom", Opt_copyfrom),
179 fsparam_flag_no ("dcache", Opt_dcache),
180 fsparam_flag_no ("dirstat", Opt_dirstat),
181 __fsparam (fs_param_is_string, "fsc", Opt_fscache,
182 fs_param_neg_with_no | fs_param_v_optional),
183 fsparam_flag_no ("ino32", Opt_ino32),
184 fsparam_string ("mds_namespace", Opt_mds_namespace),
185 fsparam_flag_no ("poolperm", Opt_poolperm),
186 fsparam_flag_no ("quotadf", Opt_quotadf),
187 fsparam_u32 ("rasize", Opt_rasize),
188 fsparam_flag_no ("rbytes", Opt_rbytes),
189 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes),
190 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries),
191 fsparam_enum ("recover_session", Opt_recover_session),
192 fsparam_flag_no ("require_active_mds", Opt_require_active_mds),
193 fsparam_u32 ("rsize", Opt_rsize),
194 fsparam_string ("snapdirname", Opt_snapdirname),
195 fsparam_string ("source", Opt_source),
196 fsparam_u32 ("wsize", Opt_wsize),
200 static const struct fs_parameter_description ceph_mount_parameters = {
202 .specs = ceph_mount_param_specs,
203 .enums = ceph_mount_param_enums,
206 struct ceph_parse_opts_ctx {
207 struct ceph_options *copts;
208 struct ceph_mount_options *opts;
212 * Parse the source parameter. Distinguish the server list from the path.
214 * The source will look like:
215 * <server_spec>[,<server_spec>...]:[<path>]
217 * <server_spec> is <ip>[:<port>]
218 * <path> is optional, but if present must begin with '/'
220 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
222 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
223 struct ceph_mount_options *fsopt = pctx->opts;
224 char *dev_name = param->string, *dev_name_end;
227 dout("%s '%s'\n", __func__, dev_name);
228 if (!dev_name || !*dev_name)
229 return invalf(fc, "ceph: Empty source");
231 dev_name_end = strchr(dev_name, '/');
233 kfree(fsopt->server_path);
236 * The server_path will include the whole chars from userland
237 * including the leading '/'.
239 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
240 if (!fsopt->server_path)
243 dev_name_end = dev_name + strlen(dev_name);
246 dev_name_end--; /* back up to ':' separator */
247 if (dev_name_end < dev_name || *dev_name_end != ':')
248 return invalf(fc, "ceph: No path or : separator in source");
250 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
251 if (fsopt->server_path)
252 dout("server path '%s'\n", fsopt->server_path);
254 ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
259 fc->source = param->string;
260 param->string = NULL;
264 static int ceph_parse_mount_param(struct fs_context *fc,
265 struct fs_parameter *param)
267 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
268 struct ceph_mount_options *fsopt = pctx->opts;
269 struct fs_parse_result result;
273 ret = ceph_parse_param(param, pctx->copts, fc);
274 if (ret != -ENOPARAM)
277 token = fs_parse(fc, &ceph_mount_parameters, param, &result);
278 dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
283 case Opt_snapdirname:
284 kfree(fsopt->snapdir_name);
285 fsopt->snapdir_name = param->string;
286 param->string = NULL;
288 case Opt_mds_namespace:
289 kfree(fsopt->mds_namespace);
290 fsopt->mds_namespace = param->string;
291 param->string = NULL;
293 case Opt_recover_session:
294 mode = result.uint_32;
295 if (mode == ceph_recover_session_no)
296 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
297 else if (mode == ceph_recover_session_clean)
298 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
304 return invalf(fc, "ceph: Multiple sources specified");
305 return ceph_parse_source(param, fc);
307 if (result.uint_32 < PAGE_SIZE ||
308 result.uint_32 > CEPH_MAX_WRITE_SIZE)
310 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
313 if (result.uint_32 < PAGE_SIZE ||
314 result.uint_32 > CEPH_MAX_READ_SIZE)
316 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
319 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
321 case Opt_caps_wanted_delay_min:
322 if (result.uint_32 < 1)
324 fsopt->caps_wanted_delay_min = result.uint_32;
326 case Opt_caps_wanted_delay_max:
327 if (result.uint_32 < 1)
329 fsopt->caps_wanted_delay_max = result.uint_32;
332 if (result.int_32 < 0)
334 fsopt->caps_max = result.int_32;
336 case Opt_readdir_max_entries:
337 if (result.uint_32 < 1)
339 fsopt->max_readdir = result.uint_32;
341 case Opt_readdir_max_bytes:
342 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
344 fsopt->max_readdir_bytes = result.uint_32;
346 case Opt_congestion_kb:
347 if (result.uint_32 < 1024) /* at least 1M */
349 fsopt->congestion_kb = result.uint_32;
353 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
355 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
359 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
361 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
363 case Opt_asyncreaddir:
365 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
367 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
371 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
373 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
377 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
379 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
383 #ifdef CONFIG_CEPH_FSCACHE
384 kfree(fsopt->fscache_uniq);
385 fsopt->fscache_uniq = NULL;
386 if (result.negated) {
387 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
389 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
390 fsopt->fscache_uniq = param->string;
391 param->string = NULL;
395 return invalf(fc, "ceph: fscache support is disabled");
399 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
401 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
403 case Opt_require_active_mds:
405 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
407 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
411 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
413 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
417 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
419 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
422 if (!result.negated) {
423 #ifdef CONFIG_CEPH_FS_POSIX_ACL
424 fc->sb_flags |= SB_POSIXACL;
426 return invalf(fc, "ceph: POSIX ACL support is disabled");
429 fc->sb_flags &= ~SB_POSIXACL;
438 return invalf(fc, "ceph: %s out of range", param->key);
441 static void destroy_mount_options(struct ceph_mount_options *args)
443 dout("destroy_mount_options %p\n", args);
447 kfree(args->snapdir_name);
448 kfree(args->mds_namespace);
449 kfree(args->server_path);
450 kfree(args->fscache_uniq);
454 static int strcmp_null(const char *s1, const char *s2)
462 return strcmp(s1, s2);
466 * path_remove_extra_slash - Remove the extra slashes in the server path
467 * @server_path: the server path and could be NULL
469 * Return NULL if the path is NULL or only consists of "/", or a string
470 * without any extra slashes including the leading slash(es) and the
471 * slash(es) at the end of the server path, such as:
472 * "//dir1////dir2///" --> "dir1/dir2"
474 static char *path_remove_extra_slash(const char *server_path)
476 const char *path = server_path;
477 const char *cur, *end;
481 /* if the server path is omitted */
485 /* remove all the leading slashes */
489 /* if the server path only consists of slashes */
495 buf = kmalloc(len + 1, GFP_KERNEL);
497 return ERR_PTR(-ENOMEM);
502 cur = strchr(path, '/');
508 /* including one '/' */
512 memcpy(p, path, len);
515 while (cur <= end && *cur == '/')
518 } while (path < end);
523 * remove the last slash if there has and just to make sure that
524 * we will get something like "dir1/dir2"
532 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
533 struct ceph_options *new_opt,
534 struct ceph_fs_client *fsc)
536 struct ceph_mount_options *fsopt1 = new_fsopt;
537 struct ceph_mount_options *fsopt2 = fsc->mount_options;
538 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
542 ret = memcmp(fsopt1, fsopt2, ofs);
546 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
549 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
553 p1 = path_remove_extra_slash(fsopt1->server_path);
556 p2 = path_remove_extra_slash(fsopt2->server_path);
561 ret = strcmp_null(p1, p2);
567 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
571 return ceph_compare_options(new_opt, fsc->client);
575 * ceph_show_options - Show mount options in /proc/mounts
576 * @m: seq_file to write to
577 * @root: root of that (sub)tree
579 static int ceph_show_options(struct seq_file *m, struct dentry *root)
581 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
582 struct ceph_mount_options *fsopt = fsc->mount_options;
586 /* a comma between MNT/MS and client options */
590 ret = ceph_print_client_options(m, fsc->client, false);
594 /* retract our comma if no client options */
598 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
599 seq_puts(m, ",dirstat");
600 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
601 seq_puts(m, ",rbytes");
602 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
603 seq_puts(m, ",noasyncreaddir");
604 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
605 seq_puts(m, ",nodcache");
606 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
607 seq_puts(m, ",ino32");
608 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
609 seq_show_option(m, "fsc", fsopt->fscache_uniq);
611 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
612 seq_puts(m, ",nopoolperm");
613 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
614 seq_puts(m, ",noquotadf");
616 #ifdef CONFIG_CEPH_FS_POSIX_ACL
617 if (root->d_sb->s_flags & SB_POSIXACL)
620 seq_puts(m, ",noacl");
623 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
624 seq_puts(m, ",copyfrom");
626 if (fsopt->mds_namespace)
627 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
629 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
630 seq_show_option(m, "recover_session", "clean");
632 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
633 seq_printf(m, ",wsize=%u", fsopt->wsize);
634 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
635 seq_printf(m, ",rsize=%u", fsopt->rsize);
636 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
637 seq_printf(m, ",rasize=%u", fsopt->rasize);
638 if (fsopt->congestion_kb != default_congestion_kb())
639 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
641 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
642 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
643 seq_printf(m, ",caps_wanted_delay_min=%u",
644 fsopt->caps_wanted_delay_min);
645 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
646 seq_printf(m, ",caps_wanted_delay_max=%u",
647 fsopt->caps_wanted_delay_max);
648 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
649 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
650 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
651 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
652 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
653 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
659 * handle any mon messages the standard library doesn't understand.
660 * return error if we don't either.
662 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
664 struct ceph_fs_client *fsc = client->private;
665 int type = le16_to_cpu(msg->hdr.type);
668 case CEPH_MSG_MDS_MAP:
669 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
671 case CEPH_MSG_FS_MAP_USER:
672 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
680 * create a new fs client
682 * Success or not, this function consumes @fsopt and @opt.
684 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
685 struct ceph_options *opt)
687 struct ceph_fs_client *fsc;
692 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
698 fsc->client = ceph_create_client(opt, fsc);
699 if (IS_ERR(fsc->client)) {
700 err = PTR_ERR(fsc->client);
703 opt = NULL; /* fsc->client now owns this */
705 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
706 ceph_set_opt(fsc->client, ABORT_ON_FULL);
708 if (!fsopt->mds_namespace) {
709 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
712 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
716 fsc->mount_options = fsopt;
719 fsc->mount_state = CEPH_MOUNT_MOUNTING;
721 fsc->have_copy_from2 = true;
723 atomic_long_set(&fsc->writeback_count, 0);
727 * The number of concurrent works can be high but they don't need
728 * to be processed in parallel, limit concurrency.
730 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
733 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
737 /* set up mempools */
739 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
740 size = sizeof (struct page *) * (page_count ? page_count : 1);
741 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
742 if (!fsc->wb_pagevec_pool)
748 destroy_workqueue(fsc->cap_wq);
750 destroy_workqueue(fsc->inode_wq);
752 ceph_destroy_client(fsc->client);
756 ceph_destroy_options(opt);
757 destroy_mount_options(fsopt);
761 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
763 flush_workqueue(fsc->inode_wq);
764 flush_workqueue(fsc->cap_wq);
767 static void destroy_fs_client(struct ceph_fs_client *fsc)
769 dout("destroy_fs_client %p\n", fsc);
771 ceph_mdsc_destroy(fsc);
772 destroy_workqueue(fsc->inode_wq);
773 destroy_workqueue(fsc->cap_wq);
775 mempool_destroy(fsc->wb_pagevec_pool);
777 destroy_mount_options(fsc->mount_options);
779 ceph_destroy_client(fsc->client);
782 dout("destroy_fs_client %p done\n", fsc);
788 struct kmem_cache *ceph_inode_cachep;
789 struct kmem_cache *ceph_cap_cachep;
790 struct kmem_cache *ceph_cap_flush_cachep;
791 struct kmem_cache *ceph_dentry_cachep;
792 struct kmem_cache *ceph_file_cachep;
793 struct kmem_cache *ceph_dir_file_cachep;
795 static void ceph_inode_init_once(void *foo)
797 struct ceph_inode_info *ci = foo;
798 inode_init_once(&ci->vfs_inode);
801 static int __init init_caches(void)
805 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
806 sizeof(struct ceph_inode_info),
807 __alignof__(struct ceph_inode_info),
808 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
809 SLAB_ACCOUNT, ceph_inode_init_once);
810 if (!ceph_inode_cachep)
813 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
814 if (!ceph_cap_cachep)
816 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
817 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
818 if (!ceph_cap_flush_cachep)
821 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
822 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
823 if (!ceph_dentry_cachep)
826 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
827 if (!ceph_file_cachep)
830 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
831 if (!ceph_dir_file_cachep)
834 error = ceph_fscache_register();
841 kmem_cache_destroy(ceph_dir_file_cachep);
843 kmem_cache_destroy(ceph_file_cachep);
845 kmem_cache_destroy(ceph_dentry_cachep);
847 kmem_cache_destroy(ceph_cap_flush_cachep);
849 kmem_cache_destroy(ceph_cap_cachep);
851 kmem_cache_destroy(ceph_inode_cachep);
855 static void destroy_caches(void)
858 * Make sure all delayed rcu free inodes are flushed before we
863 kmem_cache_destroy(ceph_inode_cachep);
864 kmem_cache_destroy(ceph_cap_cachep);
865 kmem_cache_destroy(ceph_cap_flush_cachep);
866 kmem_cache_destroy(ceph_dentry_cachep);
867 kmem_cache_destroy(ceph_file_cachep);
868 kmem_cache_destroy(ceph_dir_file_cachep);
870 ceph_fscache_unregister();
874 * ceph_umount_begin - initiate forced umount. Tear down down the
875 * mount, skipping steps that may hang while waiting for server(s).
877 static void ceph_umount_begin(struct super_block *sb)
879 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
881 dout("ceph_umount_begin - starting forced umount\n");
884 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
885 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
886 ceph_mdsc_force_umount(fsc->mdsc);
887 fsc->filp_gen++; // invalidate open files
890 static const struct super_operations ceph_super_ops = {
891 .alloc_inode = ceph_alloc_inode,
892 .free_inode = ceph_free_inode,
893 .write_inode = ceph_write_inode,
894 .drop_inode = generic_delete_inode,
895 .evict_inode = ceph_evict_inode,
896 .sync_fs = ceph_sync_fs,
897 .put_super = ceph_put_super,
898 .show_options = ceph_show_options,
899 .statfs = ceph_statfs,
900 .umount_begin = ceph_umount_begin,
904 * Bootstrap mount by opening the root directory. Note the mount
905 * @started time from caller, and time out if this takes too long.
907 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
909 unsigned long started)
911 struct ceph_mds_client *mdsc = fsc->mdsc;
912 struct ceph_mds_request *req = NULL;
917 dout("open_root_inode opening '%s'\n", path);
918 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
920 return ERR_CAST(req);
921 req->r_path1 = kstrdup(path, GFP_NOFS);
923 root = ERR_PTR(-ENOMEM);
927 req->r_ino1.ino = CEPH_INO_ROOT;
928 req->r_ino1.snap = CEPH_NOSNAP;
929 req->r_started = started;
930 req->r_timeout = fsc->client->options->mount_timeout;
931 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
933 err = ceph_mdsc_do_request(mdsc, NULL, req);
935 struct inode *inode = req->r_target_inode;
936 req->r_target_inode = NULL;
937 dout("open_root_inode success\n");
938 root = d_make_root(inode);
940 root = ERR_PTR(-ENOMEM);
943 dout("open_root_inode success, root dentry is %p\n", root);
948 ceph_mdsc_put_request(req);
953 * mount: join the ceph cluster, and open root directory.
955 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
956 struct fs_context *fc)
959 unsigned long started = jiffies; /* note the start time */
962 dout("mount start %p\n", fsc);
963 mutex_lock(&fsc->client->mount_mutex);
965 if (!fsc->sb->s_root) {
966 const char *path, *p;
967 err = __ceph_open_session(fsc->client, started);
972 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
973 err = ceph_fscache_register_fs(fsc, fc);
978 p = path_remove_extra_slash(fsc->mount_options->server_path);
983 /* if the server path is omitted or just consists of '/' */
988 dout("mount opening path '%s'\n", path);
990 ceph_fs_debugfs_init(fsc);
992 root = open_root_dentry(fsc, path, started);
998 fsc->sb->s_root = dget(root);
1000 root = dget(fsc->sb->s_root);
1003 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1004 dout("mount success\n");
1005 mutex_unlock(&fsc->client->mount_mutex);
1009 mutex_unlock(&fsc->client->mount_mutex);
1010 return ERR_PTR(err);
1013 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1015 struct ceph_fs_client *fsc = s->s_fs_info;
1018 dout("set_super %p\n", s);
1020 s->s_maxbytes = MAX_LFS_FILESIZE;
1022 s->s_xattr = ceph_xattr_handlers;
1024 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1026 s->s_op = &ceph_super_ops;
1027 s->s_d_op = &ceph_dentry_ops;
1028 s->s_export_op = &ceph_export_ops;
1032 s->s_time_max = U32_MAX;
1034 ret = set_anon_super_fc(s, fc);
1041 * share superblock if same fs AND options
1043 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1045 struct ceph_fs_client *new = fc->s_fs_info;
1046 struct ceph_mount_options *fsopt = new->mount_options;
1047 struct ceph_options *opt = new->client->options;
1048 struct ceph_fs_client *other = ceph_sb_to_client(sb);
1050 dout("ceph_compare_super %p\n", sb);
1052 if (compare_mount_options(fsopt, opt, other)) {
1053 dout("monitor(s)/mount options don't match\n");
1056 if ((opt->flags & CEPH_OPT_FSID) &&
1057 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
1058 dout("fsid doesn't match\n");
1061 if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1062 dout("flags differ\n");
1069 * construct our own bdi so we can control readahead, etc.
1071 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1073 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1077 err = super_setup_bdi_name(sb, "ceph-%ld",
1078 atomic_long_inc_return(&bdi_seq));
1082 /* set ra_pages based on rasize mount option? */
1083 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1085 /* set io_pages based on max osd read size */
1086 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1091 static int ceph_get_tree(struct fs_context *fc)
1093 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1094 struct super_block *sb;
1095 struct ceph_fs_client *fsc;
1097 int (*compare_super)(struct super_block *, struct fs_context *) =
1101 dout("ceph_get_tree\n");
1104 return invalf(fc, "ceph: No source");
1106 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1107 fc->sb_flags |= SB_POSIXACL;
1110 /* create client (which we may/may not use) */
1111 fsc = create_fs_client(pctx->opts, pctx->copts);
1119 err = ceph_mdsc_init(fsc);
1123 if (ceph_test_opt(fsc->client, NOSHARE))
1124 compare_super = NULL;
1126 fc->s_fs_info = fsc;
1127 sb = sget_fc(fc, compare_super, ceph_set_super);
1128 fc->s_fs_info = NULL;
1134 if (ceph_sb_to_client(sb) != fsc) {
1135 destroy_fs_client(fsc);
1136 fsc = ceph_sb_to_client(sb);
1137 dout("get_sb got existing client %p\n", fsc);
1139 dout("get_sb using new client %p\n", fsc);
1140 err = ceph_setup_bdi(sb, fsc);
1145 res = ceph_real_mount(fsc, fc);
1150 dout("root %p inode %p ino %llx.%llx\n", res,
1151 d_inode(res), ceph_vinop(d_inode(res)));
1152 fc->root = fsc->sb->s_root;
1156 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1157 pr_info("No mds server is up or the cluster is laggy\n");
1158 err = -EHOSTUNREACH;
1161 ceph_mdsc_close_sessions(fsc->mdsc);
1162 deactivate_locked_super(sb);
1166 destroy_fs_client(fsc);
1168 dout("ceph_get_tree fail %d\n", err);
1172 static void ceph_free_fc(struct fs_context *fc)
1174 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1177 destroy_mount_options(pctx->opts);
1178 ceph_destroy_options(pctx->copts);
1183 static int ceph_reconfigure_fc(struct fs_context *fc)
1185 sync_filesystem(fc->root->d_sb);
1189 static const struct fs_context_operations ceph_context_ops = {
1190 .free = ceph_free_fc,
1191 .parse_param = ceph_parse_mount_param,
1192 .get_tree = ceph_get_tree,
1193 .reconfigure = ceph_reconfigure_fc,
1197 * Set up the filesystem mount context.
1199 static int ceph_init_fs_context(struct fs_context *fc)
1201 struct ceph_parse_opts_ctx *pctx;
1202 struct ceph_mount_options *fsopt;
1204 pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1208 pctx->copts = ceph_alloc_options();
1212 pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1217 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1219 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1220 fsopt->rsize = CEPH_MAX_READ_SIZE;
1221 fsopt->rasize = CEPH_RASIZE_DEFAULT;
1222 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1223 if (!fsopt->snapdir_name)
1226 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1227 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1228 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1229 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1230 fsopt->congestion_kb = default_congestion_kb();
1232 fc->fs_private = pctx;
1233 fc->ops = &ceph_context_ops;
1237 destroy_mount_options(pctx->opts);
1238 ceph_destroy_options(pctx->copts);
1243 static void ceph_kill_sb(struct super_block *s)
1245 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1246 dev_t dev = s->s_dev;
1248 dout("kill_sb %p\n", s);
1250 ceph_mdsc_pre_umount(fsc->mdsc);
1251 flush_fs_workqueues(fsc);
1253 generic_shutdown_super(s);
1255 fsc->client->extra_mon_dispatch = NULL;
1256 ceph_fs_debugfs_cleanup(fsc);
1258 ceph_fscache_unregister_fs(fsc);
1260 destroy_fs_client(fsc);
1261 free_anon_bdev(dev);
1264 static struct file_system_type ceph_fs_type = {
1265 .owner = THIS_MODULE,
1267 .init_fs_context = ceph_init_fs_context,
1268 .kill_sb = ceph_kill_sb,
1269 .fs_flags = FS_RENAME_DOES_D_MOVE,
1271 MODULE_ALIAS_FS("ceph");
1273 int ceph_force_reconnect(struct super_block *sb)
1275 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1278 ceph_umount_begin(sb);
1280 /* Make sure all page caches get invalidated.
1281 * see remove_session_caps_cb() */
1282 flush_workqueue(fsc->inode_wq);
1284 /* In case that we were blacklisted. This also reset
1285 * all mon/osd connections */
1286 ceph_reset_client_addr(fsc->client);
1288 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1290 fsc->blacklisted = false;
1291 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1294 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1295 CEPH_STAT_CAP_INODE, true);
1300 static int __init init_ceph(void)
1302 int ret = init_caches();
1307 ret = register_filesystem(&ceph_fs_type);
1311 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1321 static void __exit exit_ceph(void)
1323 dout("exit_ceph\n");
1324 unregister_filesystem(&ceph_fs_type);
1328 module_init(init_ceph);
1329 module_exit(exit_ceph);
1331 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1332 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1333 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1334 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1335 MODULE_LICENSE("GPL");