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;
111 static int ceph_sync_fs(struct super_block *sb, int wait)
113 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
116 dout("sync_fs (non-blocking)\n");
117 ceph_flush_dirty_caps(fsc->mdsc);
118 dout("sync_fs (non-blocking) done\n");
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");
136 Opt_caps_wanted_delay_min,
137 Opt_caps_wanted_delay_max,
139 Opt_readdir_max_entries,
140 Opt_readdir_max_bytes,
147 /* string args above */
155 Opt_require_active_mds,
161 enum ceph_recover_session_mode {
162 ceph_recover_session_no,
163 ceph_recover_session_clean
166 static const struct constant_table ceph_param_recover[] = {
167 { "no", ceph_recover_session_no },
168 { "clean", ceph_recover_session_clean },
172 static const struct fs_parameter_spec ceph_mount_param_specs[] = {
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 (fs_param_is_string, "fsc", Opt_fscache,
183 fs_param_neg_with_no | fs_param_v_optional, NULL),
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),
201 static const struct fs_parameter_description ceph_mount_parameters = {
203 .specs = ceph_mount_param_specs,
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.
213 * Internally we do not include the leading '/' in the path.
215 * The source will look like:
216 * <server_spec>[,<server_spec>...]:[<path>]
218 * <server_spec> is <ip>[:<port>]
219 * <path> is optional, but if present must begin with '/'
221 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
223 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
224 struct ceph_mount_options *fsopt = pctx->opts;
225 char *dev_name = param->string, *dev_name_end;
228 dout("%s '%s'\n", __func__, dev_name);
229 if (!dev_name || !*dev_name)
230 return invalf(fc, "ceph: Empty source");
232 dev_name_end = strchr(dev_name, '/');
234 if (strlen(dev_name_end) > 1) {
235 kfree(fsopt->server_path);
236 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
237 if (!fsopt->server_path)
241 dev_name_end = dev_name + strlen(dev_name);
244 dev_name_end--; /* back up to ':' separator */
245 if (dev_name_end < dev_name || *dev_name_end != ':')
246 return invalf(fc, "ceph: No path or : separator in source");
248 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
249 if (fsopt->server_path)
250 dout("server path '%s'\n", fsopt->server_path);
252 ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
253 pctx->copts, fc->log.log);
257 fc->source = param->string;
258 param->string = NULL;
262 static int ceph_parse_mount_param(struct fs_context *fc,
263 struct fs_parameter *param)
265 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
266 struct ceph_mount_options *fsopt = pctx->opts;
267 struct fs_parse_result result;
271 ret = ceph_parse_param(param, pctx->copts, fc->log.log);
272 if (ret != -ENOPARAM)
275 token = fs_parse(fc, &ceph_mount_parameters, param, &result);
276 dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
281 case Opt_snapdirname:
282 kfree(fsopt->snapdir_name);
283 fsopt->snapdir_name = param->string;
284 param->string = NULL;
286 case Opt_mds_namespace:
287 kfree(fsopt->mds_namespace);
288 fsopt->mds_namespace = param->string;
289 param->string = NULL;
291 case Opt_recover_session:
292 mode = result.uint_32;
293 if (mode == ceph_recover_session_no)
294 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
295 else if (mode == ceph_recover_session_clean)
296 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
302 return invalf(fc, "ceph: Multiple sources specified");
303 return ceph_parse_source(param, fc);
305 if (result.uint_32 < PAGE_SIZE ||
306 result.uint_32 > CEPH_MAX_WRITE_SIZE)
308 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
311 if (result.uint_32 < PAGE_SIZE ||
312 result.uint_32 > CEPH_MAX_READ_SIZE)
314 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
317 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
319 case Opt_caps_wanted_delay_min:
320 if (result.uint_32 < 1)
322 fsopt->caps_wanted_delay_min = result.uint_32;
324 case Opt_caps_wanted_delay_max:
325 if (result.uint_32 < 1)
327 fsopt->caps_wanted_delay_max = result.uint_32;
330 if (result.int_32 < 0)
332 fsopt->caps_max = result.int_32;
334 case Opt_readdir_max_entries:
335 if (result.uint_32 < 1)
337 fsopt->max_readdir = result.uint_32;
339 case Opt_readdir_max_bytes:
340 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
342 fsopt->max_readdir_bytes = result.uint_32;
344 case Opt_congestion_kb:
345 if (result.uint_32 < 1024) /* at least 1M */
347 fsopt->congestion_kb = result.uint_32;
351 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
353 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
357 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
359 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
361 case Opt_asyncreaddir:
363 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
365 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
369 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
371 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
375 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
377 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
381 #ifdef CONFIG_CEPH_FSCACHE
382 kfree(fsopt->fscache_uniq);
383 fsopt->fscache_uniq = NULL;
384 if (result.negated) {
385 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
387 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
388 fsopt->fscache_uniq = param->string;
389 param->string = NULL;
393 return invalf(fc, "ceph: fscache support is disabled");
397 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
399 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
401 case Opt_require_active_mds:
403 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
405 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
409 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
411 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
415 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
417 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
420 if (!result.negated) {
421 #ifdef CONFIG_CEPH_FS_POSIX_ACL
422 fc->sb_flags |= SB_POSIXACL;
424 return invalf(fc, "ceph: POSIX ACL support is disabled");
427 fc->sb_flags &= ~SB_POSIXACL;
436 return invalf(fc, "ceph: %s out of range", param->key);
439 static void destroy_mount_options(struct ceph_mount_options *args)
441 dout("destroy_mount_options %p\n", args);
445 kfree(args->snapdir_name);
446 kfree(args->mds_namespace);
447 kfree(args->server_path);
448 kfree(args->fscache_uniq);
452 static int strcmp_null(const char *s1, const char *s2)
460 return strcmp(s1, s2);
463 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
464 struct ceph_options *new_opt,
465 struct ceph_fs_client *fsc)
467 struct ceph_mount_options *fsopt1 = new_fsopt;
468 struct ceph_mount_options *fsopt2 = fsc->mount_options;
469 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
472 ret = memcmp(fsopt1, fsopt2, ofs);
476 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
479 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
482 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
485 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
489 return ceph_compare_options(new_opt, fsc->client);
493 * ceph_show_options - Show mount options in /proc/mounts
494 * @m: seq_file to write to
495 * @root: root of that (sub)tree
497 static int ceph_show_options(struct seq_file *m, struct dentry *root)
499 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
500 struct ceph_mount_options *fsopt = fsc->mount_options;
504 /* a comma between MNT/MS and client options */
508 ret = ceph_print_client_options(m, fsc->client, false);
512 /* retract our comma if no client options */
516 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
517 seq_puts(m, ",dirstat");
518 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
519 seq_puts(m, ",rbytes");
520 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
521 seq_puts(m, ",noasyncreaddir");
522 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
523 seq_puts(m, ",nodcache");
524 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
525 seq_puts(m, ",ino32");
526 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
527 seq_show_option(m, "fsc", fsopt->fscache_uniq);
529 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
530 seq_puts(m, ",nopoolperm");
531 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
532 seq_puts(m, ",noquotadf");
534 #ifdef CONFIG_CEPH_FS_POSIX_ACL
535 if (root->d_sb->s_flags & SB_POSIXACL)
538 seq_puts(m, ",noacl");
541 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
542 seq_puts(m, ",copyfrom");
544 if (fsopt->mds_namespace)
545 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
547 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
548 seq_show_option(m, "recover_session", "clean");
550 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
551 seq_printf(m, ",wsize=%u", fsopt->wsize);
552 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
553 seq_printf(m, ",rsize=%u", fsopt->rsize);
554 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
555 seq_printf(m, ",rasize=%u", fsopt->rasize);
556 if (fsopt->congestion_kb != default_congestion_kb())
557 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
559 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
560 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
561 seq_printf(m, ",caps_wanted_delay_min=%u",
562 fsopt->caps_wanted_delay_min);
563 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
564 seq_printf(m, ",caps_wanted_delay_max=%u",
565 fsopt->caps_wanted_delay_max);
566 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
567 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
568 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
569 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
570 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
571 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
577 * handle any mon messages the standard library doesn't understand.
578 * return error if we don't either.
580 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
582 struct ceph_fs_client *fsc = client->private;
583 int type = le16_to_cpu(msg->hdr.type);
586 case CEPH_MSG_MDS_MAP:
587 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
589 case CEPH_MSG_FS_MAP_USER:
590 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
598 * create a new fs client
600 * Success or not, this function consumes @fsopt and @opt.
602 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
603 struct ceph_options *opt)
605 struct ceph_fs_client *fsc;
610 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
616 fsc->client = ceph_create_client(opt, fsc);
617 if (IS_ERR(fsc->client)) {
618 err = PTR_ERR(fsc->client);
621 opt = NULL; /* fsc->client now owns this */
623 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
624 ceph_set_opt(fsc->client, ABORT_ON_FULL);
626 if (!fsopt->mds_namespace) {
627 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
630 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
634 fsc->mount_options = fsopt;
637 fsc->mount_state = CEPH_MOUNT_MOUNTING;
640 atomic_long_set(&fsc->writeback_count, 0);
644 * The number of concurrent works can be high but they don't need
645 * to be processed in parallel, limit concurrency.
647 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
650 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
654 /* set up mempools */
656 page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
657 size = sizeof (struct page *) * (page_count ? page_count : 1);
658 fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
659 if (!fsc->wb_pagevec_pool)
665 destroy_workqueue(fsc->cap_wq);
667 destroy_workqueue(fsc->inode_wq);
669 ceph_destroy_client(fsc->client);
673 ceph_destroy_options(opt);
674 destroy_mount_options(fsopt);
678 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
680 flush_workqueue(fsc->inode_wq);
681 flush_workqueue(fsc->cap_wq);
684 static void destroy_fs_client(struct ceph_fs_client *fsc)
686 dout("destroy_fs_client %p\n", fsc);
688 ceph_mdsc_destroy(fsc);
689 destroy_workqueue(fsc->inode_wq);
690 destroy_workqueue(fsc->cap_wq);
692 mempool_destroy(fsc->wb_pagevec_pool);
694 destroy_mount_options(fsc->mount_options);
696 ceph_destroy_client(fsc->client);
699 dout("destroy_fs_client %p done\n", fsc);
705 struct kmem_cache *ceph_inode_cachep;
706 struct kmem_cache *ceph_cap_cachep;
707 struct kmem_cache *ceph_cap_flush_cachep;
708 struct kmem_cache *ceph_dentry_cachep;
709 struct kmem_cache *ceph_file_cachep;
710 struct kmem_cache *ceph_dir_file_cachep;
712 static void ceph_inode_init_once(void *foo)
714 struct ceph_inode_info *ci = foo;
715 inode_init_once(&ci->vfs_inode);
718 static int __init init_caches(void)
722 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
723 sizeof(struct ceph_inode_info),
724 __alignof__(struct ceph_inode_info),
725 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
726 SLAB_ACCOUNT, ceph_inode_init_once);
727 if (!ceph_inode_cachep)
730 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
731 if (!ceph_cap_cachep)
733 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
734 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
735 if (!ceph_cap_flush_cachep)
738 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
739 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
740 if (!ceph_dentry_cachep)
743 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
744 if (!ceph_file_cachep)
747 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
748 if (!ceph_dir_file_cachep)
751 error = ceph_fscache_register();
758 kmem_cache_destroy(ceph_dir_file_cachep);
760 kmem_cache_destroy(ceph_file_cachep);
762 kmem_cache_destroy(ceph_dentry_cachep);
764 kmem_cache_destroy(ceph_cap_flush_cachep);
766 kmem_cache_destroy(ceph_cap_cachep);
768 kmem_cache_destroy(ceph_inode_cachep);
772 static void destroy_caches(void)
775 * Make sure all delayed rcu free inodes are flushed before we
780 kmem_cache_destroy(ceph_inode_cachep);
781 kmem_cache_destroy(ceph_cap_cachep);
782 kmem_cache_destroy(ceph_cap_flush_cachep);
783 kmem_cache_destroy(ceph_dentry_cachep);
784 kmem_cache_destroy(ceph_file_cachep);
785 kmem_cache_destroy(ceph_dir_file_cachep);
787 ceph_fscache_unregister();
792 * ceph_umount_begin - initiate forced umount. Tear down down the
793 * mount, skipping steps that may hang while waiting for server(s).
795 static void ceph_umount_begin(struct super_block *sb)
797 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
799 dout("ceph_umount_begin - starting forced umount\n");
802 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
803 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
804 ceph_mdsc_force_umount(fsc->mdsc);
805 fsc->filp_gen++; // invalidate open files
808 static const struct super_operations ceph_super_ops = {
809 .alloc_inode = ceph_alloc_inode,
810 .free_inode = ceph_free_inode,
811 .write_inode = ceph_write_inode,
812 .drop_inode = generic_delete_inode,
813 .evict_inode = ceph_evict_inode,
814 .sync_fs = ceph_sync_fs,
815 .put_super = ceph_put_super,
816 .show_options = ceph_show_options,
817 .statfs = ceph_statfs,
818 .umount_begin = ceph_umount_begin,
822 * Bootstrap mount by opening the root directory. Note the mount
823 * @started time from caller, and time out if this takes too long.
825 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
827 unsigned long started)
829 struct ceph_mds_client *mdsc = fsc->mdsc;
830 struct ceph_mds_request *req = NULL;
835 dout("open_root_inode opening '%s'\n", path);
836 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
838 return ERR_CAST(req);
839 req->r_path1 = kstrdup(path, GFP_NOFS);
841 root = ERR_PTR(-ENOMEM);
845 req->r_ino1.ino = CEPH_INO_ROOT;
846 req->r_ino1.snap = CEPH_NOSNAP;
847 req->r_started = started;
848 req->r_timeout = fsc->client->options->mount_timeout;
849 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
851 err = ceph_mdsc_do_request(mdsc, NULL, req);
853 struct inode *inode = req->r_target_inode;
854 req->r_target_inode = NULL;
855 dout("open_root_inode success\n");
856 root = d_make_root(inode);
858 root = ERR_PTR(-ENOMEM);
861 dout("open_root_inode success, root dentry is %p\n", root);
866 ceph_mdsc_put_request(req);
874 * mount: join the ceph cluster, and open root directory.
876 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
877 struct fs_context *fc)
880 unsigned long started = jiffies; /* note the start time */
883 dout("mount start %p\n", fsc);
884 mutex_lock(&fsc->client->mount_mutex);
886 if (!fsc->sb->s_root) {
888 err = __ceph_open_session(fsc->client, started);
893 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
894 err = ceph_fscache_register_fs(fsc, fc);
899 if (!fsc->mount_options->server_path) {
901 dout("mount opening path \\t\n");
903 path = fsc->mount_options->server_path + 1;
904 dout("mount opening path %s\n", path);
907 ceph_fs_debugfs_init(fsc);
909 root = open_root_dentry(fsc, path, started);
914 fsc->sb->s_root = dget(root);
916 root = dget(fsc->sb->s_root);
919 fsc->mount_state = CEPH_MOUNT_MOUNTED;
920 dout("mount success\n");
921 mutex_unlock(&fsc->client->mount_mutex);
925 mutex_unlock(&fsc->client->mount_mutex);
929 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
931 struct ceph_fs_client *fsc = s->s_fs_info;
934 dout("set_super %p\n", s);
936 s->s_maxbytes = MAX_LFS_FILESIZE;
938 s->s_xattr = ceph_xattr_handlers;
940 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
942 s->s_op = &ceph_super_ops;
943 s->s_d_op = &ceph_dentry_ops;
944 s->s_export_op = &ceph_export_ops;
948 s->s_time_max = U32_MAX;
950 ret = set_anon_super_fc(s, fc);
957 * share superblock if same fs AND options
959 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
961 struct ceph_fs_client *new = fc->s_fs_info;
962 struct ceph_mount_options *fsopt = new->mount_options;
963 struct ceph_options *opt = new->client->options;
964 struct ceph_fs_client *other = ceph_sb_to_client(sb);
966 dout("ceph_compare_super %p\n", sb);
968 if (compare_mount_options(fsopt, opt, other)) {
969 dout("monitor(s)/mount options don't match\n");
972 if ((opt->flags & CEPH_OPT_FSID) &&
973 ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
974 dout("fsid doesn't match\n");
977 if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
978 dout("flags differ\n");
985 * construct our own bdi so we can control readahead, etc.
987 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
989 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
993 err = super_setup_bdi_name(sb, "ceph-%ld",
994 atomic_long_inc_return(&bdi_seq));
998 /* set ra_pages based on rasize mount option? */
999 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1001 /* set io_pages based on max osd read size */
1002 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1007 static int ceph_get_tree(struct fs_context *fc)
1009 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1010 struct super_block *sb;
1011 struct ceph_fs_client *fsc;
1013 int (*compare_super)(struct super_block *, struct fs_context *) =
1017 dout("ceph_get_tree\n");
1020 return invalf(fc, "ceph: No source");
1022 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1023 fc->sb_flags |= SB_POSIXACL;
1026 /* create client (which we may/may not use) */
1027 fsc = create_fs_client(pctx->opts, pctx->copts);
1035 err = ceph_mdsc_init(fsc);
1039 if (ceph_test_opt(fsc->client, NOSHARE))
1040 compare_super = NULL;
1042 fc->s_fs_info = fsc;
1043 sb = sget_fc(fc, compare_super, ceph_set_super);
1044 fc->s_fs_info = NULL;
1050 if (ceph_sb_to_client(sb) != fsc) {
1051 destroy_fs_client(fsc);
1052 fsc = ceph_sb_to_client(sb);
1053 dout("get_sb got existing client %p\n", fsc);
1055 dout("get_sb using new client %p\n", fsc);
1056 err = ceph_setup_bdi(sb, fsc);
1061 res = ceph_real_mount(fsc, fc);
1066 dout("root %p inode %p ino %llx.%llx\n", res,
1067 d_inode(res), ceph_vinop(d_inode(res)));
1068 fc->root = fsc->sb->s_root;
1072 ceph_mdsc_close_sessions(fsc->mdsc);
1073 deactivate_locked_super(sb);
1077 destroy_fs_client(fsc);
1079 dout("ceph_get_tree fail %d\n", err);
1083 static void ceph_free_fc(struct fs_context *fc)
1085 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1088 destroy_mount_options(pctx->opts);
1089 ceph_destroy_options(pctx->copts);
1094 static int ceph_reconfigure_fc(struct fs_context *fc)
1096 sync_filesystem(fc->root->d_sb);
1100 static const struct fs_context_operations ceph_context_ops = {
1101 .free = ceph_free_fc,
1102 .parse_param = ceph_parse_mount_param,
1103 .get_tree = ceph_get_tree,
1104 .reconfigure = ceph_reconfigure_fc,
1108 * Set up the filesystem mount context.
1110 static int ceph_init_fs_context(struct fs_context *fc)
1112 struct ceph_parse_opts_ctx *pctx;
1113 struct ceph_mount_options *fsopt;
1115 pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1119 pctx->copts = ceph_alloc_options();
1123 pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1128 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1130 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1131 fsopt->rsize = CEPH_MAX_READ_SIZE;
1132 fsopt->rasize = CEPH_RASIZE_DEFAULT;
1133 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1134 if (!fsopt->snapdir_name)
1137 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1138 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1139 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1140 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1141 fsopt->congestion_kb = default_congestion_kb();
1143 fc->fs_private = pctx;
1144 fc->ops = &ceph_context_ops;
1148 destroy_mount_options(pctx->opts);
1149 ceph_destroy_options(pctx->copts);
1154 static void ceph_kill_sb(struct super_block *s)
1156 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1157 dev_t dev = s->s_dev;
1159 dout("kill_sb %p\n", s);
1161 ceph_mdsc_pre_umount(fsc->mdsc);
1162 flush_fs_workqueues(fsc);
1164 generic_shutdown_super(s);
1166 fsc->client->extra_mon_dispatch = NULL;
1167 ceph_fs_debugfs_cleanup(fsc);
1169 ceph_fscache_unregister_fs(fsc);
1171 destroy_fs_client(fsc);
1172 free_anon_bdev(dev);
1175 static struct file_system_type ceph_fs_type = {
1176 .owner = THIS_MODULE,
1178 .init_fs_context = ceph_init_fs_context,
1179 .kill_sb = ceph_kill_sb,
1180 .fs_flags = FS_RENAME_DOES_D_MOVE,
1182 MODULE_ALIAS_FS("ceph");
1184 int ceph_force_reconnect(struct super_block *sb)
1186 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1189 ceph_umount_begin(sb);
1191 /* Make sure all page caches get invalidated.
1192 * see remove_session_caps_cb() */
1193 flush_workqueue(fsc->inode_wq);
1195 /* In case that we were blacklisted. This also reset
1196 * all mon/osd connections */
1197 ceph_reset_client_addr(fsc->client);
1199 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1201 fsc->blacklisted = false;
1202 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1205 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1206 CEPH_STAT_CAP_INODE, true);
1211 static int __init init_ceph(void)
1213 int ret = init_caches();
1218 ret = register_filesystem(&ceph_fs_type);
1222 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1232 static void __exit exit_ceph(void)
1234 dout("exit_ceph\n");
1235 unregister_filesystem(&ceph_fs_type);
1239 module_init(init_ceph);
1240 module_exit(exit_ceph);
1242 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1243 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1244 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1245 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1246 MODULE_LICENSE("GPL");