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>
30 #include <uapi/linux/magic.h>
32 static DEFINE_SPINLOCK(ceph_fsc_lock);
33 static LIST_HEAD(ceph_fsc_list);
36 * Ceph superblock operations
38 * Handle the basics of mounting, unmounting.
44 static void ceph_put_super(struct super_block *s)
46 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
49 ceph_mdsc_close_sessions(fsc->mdsc);
52 static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf)
54 struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry));
55 struct ceph_mon_client *monc = &fsc->client->monc;
56 struct ceph_statfs st;
60 if (fsc->mdsc->mdsmap->m_num_data_pg_pools == 1) {
61 data_pool = fsc->mdsc->mdsmap->m_data_pg_pools[0];
63 data_pool = CEPH_NOPOOL;
67 err = ceph_monc_do_statfs(monc, data_pool, &st);
72 buf->f_type = CEPH_SUPER_MAGIC; /* ?? */
75 * express utilization in terms of large blocks to avoid
76 * overflow on 32-bit machines.
78 * NOTE: for the time being, we make bsize == frsize to humor
79 * not-yet-ancient versions of glibc that are broken.
80 * Someday, we will probably want to report a real block
81 * size... whatever that may mean for a network file system!
83 buf->f_bsize = 1 << CEPH_BLOCK_SHIFT;
84 buf->f_frsize = 1 << CEPH_BLOCK_SHIFT;
87 * By default use root quota for stats; fallback to overall filesystem
88 * usage if using 'noquotadf' mount option or if the root dir doesn't
89 * have max_bytes quota set.
91 if (ceph_test_mount_opt(fsc, NOQUOTADF) ||
92 !ceph_quota_update_statfs(fsc, buf)) {
93 buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10);
94 buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
95 buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10);
98 buf->f_files = le64_to_cpu(st.num_objects);
100 buf->f_namelen = NAME_MAX;
102 /* Must convert the fsid, for consistent values across arches */
103 buf->f_fsid.val[0] = 0;
104 mutex_lock(&monc->mutex);
105 for (i = 0 ; i < sizeof(monc->monmap->fsid) / sizeof(__le32) ; ++i)
106 buf->f_fsid.val[0] ^= le32_to_cpu(((__le32 *)&monc->monmap->fsid)[i]);
107 mutex_unlock(&monc->mutex);
109 /* fold the fs_cluster_id into the upper bits */
110 buf->f_fsid.val[1] = monc->fs_cluster_id;
115 static int ceph_sync_fs(struct super_block *sb, int wait)
117 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
120 dout("sync_fs (non-blocking)\n");
121 ceph_flush_dirty_caps(fsc->mdsc);
122 dout("sync_fs (non-blocking) done\n");
126 dout("sync_fs (blocking)\n");
127 ceph_osdc_sync(&fsc->client->osdc);
128 ceph_mdsc_sync(fsc->mdsc);
129 dout("sync_fs (blocking) done\n");
140 Opt_caps_wanted_delay_min,
141 Opt_caps_wanted_delay_max,
143 Opt_readdir_max_entries,
144 Opt_readdir_max_bytes,
152 /* string args above */
160 Opt_require_active_mds,
168 enum ceph_recover_session_mode {
169 ceph_recover_session_no,
170 ceph_recover_session_clean
173 static const struct constant_table ceph_param_recover[] = {
174 { "no", ceph_recover_session_no },
175 { "clean", ceph_recover_session_clean },
179 static const struct fs_parameter_spec ceph_mount_parameters[] = {
180 fsparam_flag_no ("acl", Opt_acl),
181 fsparam_flag_no ("asyncreaddir", Opt_asyncreaddir),
182 fsparam_s32 ("caps_max", Opt_caps_max),
183 fsparam_u32 ("caps_wanted_delay_max", Opt_caps_wanted_delay_max),
184 fsparam_u32 ("caps_wanted_delay_min", Opt_caps_wanted_delay_min),
185 fsparam_u32 ("write_congestion_kb", Opt_congestion_kb),
186 fsparam_flag_no ("copyfrom", Opt_copyfrom),
187 fsparam_flag_no ("dcache", Opt_dcache),
188 fsparam_flag_no ("dirstat", Opt_dirstat),
189 fsparam_flag_no ("fsc", Opt_fscache), // fsc|nofsc
190 fsparam_string ("fsc", Opt_fscache), // fsc=...
191 fsparam_flag_no ("ino32", Opt_ino32),
192 fsparam_string ("mds_namespace", Opt_mds_namespace),
193 fsparam_flag_no ("poolperm", Opt_poolperm),
194 fsparam_flag_no ("quotadf", Opt_quotadf),
195 fsparam_u32 ("rasize", Opt_rasize),
196 fsparam_flag_no ("rbytes", Opt_rbytes),
197 fsparam_u32 ("readdir_max_bytes", Opt_readdir_max_bytes),
198 fsparam_u32 ("readdir_max_entries", Opt_readdir_max_entries),
199 fsparam_enum ("recover_session", Opt_recover_session, ceph_param_recover),
200 fsparam_flag_no ("require_active_mds", Opt_require_active_mds),
201 fsparam_u32 ("rsize", Opt_rsize),
202 fsparam_string ("snapdirname", Opt_snapdirname),
203 fsparam_string ("source", Opt_source),
204 fsparam_string ("mon_addr", Opt_mon_addr),
205 fsparam_u32 ("wsize", Opt_wsize),
206 fsparam_flag_no ("wsync", Opt_wsync),
207 fsparam_flag_no ("pagecache", Opt_pagecache),
211 struct ceph_parse_opts_ctx {
212 struct ceph_options *copts;
213 struct ceph_mount_options *opts;
217 * Remove adjacent slashes and then the trailing slash, unless it is
218 * the only remaining character.
220 * E.g. "//dir1////dir2///" --> "/dir1/dir2", "///" --> "/".
222 static void canonicalize_path(char *path)
226 for (i = 0; path[i] != '\0'; i++) {
227 if (path[i] != '/' || j < 1 || path[j - 1] != '/')
231 if (j > 1 && path[j - 1] == '/')
237 * Check if the mds namespace in ceph_mount_options matches
238 * the passed in namespace string. First time match (when
239 * ->mds_namespace is NULL) is treated specially, since
240 * ->mds_namespace needs to be initialized by the caller.
242 static int namespace_equals(struct ceph_mount_options *fsopt,
243 const char *namespace, size_t len)
245 return !(fsopt->mds_namespace &&
246 (strlen(fsopt->mds_namespace) != len ||
247 strncmp(fsopt->mds_namespace, namespace, len)));
250 static int ceph_parse_old_source(const char *dev_name, const char *dev_name_end,
251 struct fs_context *fc)
254 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
255 struct ceph_mount_options *fsopt = pctx->opts;
257 if (*dev_name_end != ':')
258 return invalfc(fc, "separator ':' missing in source");
260 r = ceph_parse_mon_ips(dev_name, dev_name_end - dev_name,
261 pctx->copts, fc->log.log, ',');
265 fsopt->new_dev_syntax = false;
269 static int ceph_parse_new_source(const char *dev_name, const char *dev_name_end,
270 struct fs_context *fc)
273 struct ceph_fsid fsid;
274 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
275 struct ceph_mount_options *fsopt = pctx->opts;
276 char *fsid_start, *fs_name_start;
278 if (*dev_name_end != '=') {
279 dout("separator '=' missing in source");
283 fsid_start = strchr(dev_name, '@');
285 return invalfc(fc, "missing cluster fsid");
286 ++fsid_start; /* start of cluster fsid */
288 fs_name_start = strchr(fsid_start, '.');
290 return invalfc(fc, "missing file system name");
292 if (ceph_parse_fsid(fsid_start, &fsid))
293 return invalfc(fc, "Invalid FSID");
295 ++fs_name_start; /* start of file system name */
296 len = dev_name_end - fs_name_start;
298 if (!namespace_equals(fsopt, fs_name_start, len))
299 return invalfc(fc, "Mismatching mds_namespace");
300 kfree(fsopt->mds_namespace);
301 fsopt->mds_namespace = kstrndup(fs_name_start, len, GFP_KERNEL);
302 if (!fsopt->mds_namespace)
304 dout("file system (mds namespace) '%s'\n", fsopt->mds_namespace);
306 fsopt->new_dev_syntax = true;
311 * Parse the source parameter for new device format. Distinguish the device
312 * spec from the path. Try parsing new device format and fallback to old
315 * New device syntax will looks like:
316 * <device_spec>=/<path>
318 * <device_spec> is name@fsid.fsname
319 * <path> is optional, but if present must begin with '/'
320 * (monitor addresses are passed via mount option)
322 * Old device syntax is:
323 * <server_spec>[,<server_spec>...]:[<path>]
325 * <server_spec> is <ip>[:<port>]
326 * <path> is optional, but if present must begin with '/'
328 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
330 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
331 struct ceph_mount_options *fsopt = pctx->opts;
332 char *dev_name = param->string, *dev_name_end;
335 dout("%s '%s'\n", __func__, dev_name);
336 if (!dev_name || !*dev_name)
337 return invalfc(fc, "Empty source");
339 dev_name_end = strchr(dev_name, '/');
342 * The server_path will include the whole chars from userland
343 * including the leading '/'.
345 kfree(fsopt->server_path);
346 fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
347 if (!fsopt->server_path)
350 canonicalize_path(fsopt->server_path);
352 dev_name_end = dev_name + strlen(dev_name);
355 dev_name_end--; /* back up to separator */
356 if (dev_name_end < dev_name)
357 return invalfc(fc, "Path missing in source");
359 dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
360 if (fsopt->server_path)
361 dout("server path '%s'\n", fsopt->server_path);
363 dout("trying new device syntax");
364 ret = ceph_parse_new_source(dev_name, dev_name_end, fc);
368 dout("trying old device syntax");
369 ret = ceph_parse_old_source(dev_name, dev_name_end, fc);
374 fc->source = param->string;
375 param->string = NULL;
379 static int ceph_parse_mon_addr(struct fs_parameter *param,
380 struct fs_context *fc)
382 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
383 struct ceph_mount_options *fsopt = pctx->opts;
385 kfree(fsopt->mon_addr);
386 fsopt->mon_addr = param->string;
387 param->string = NULL;
389 return ceph_parse_mon_ips(fsopt->mon_addr, strlen(fsopt->mon_addr),
390 pctx->copts, fc->log.log, '/');
393 static int ceph_parse_mount_param(struct fs_context *fc,
394 struct fs_parameter *param)
396 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
397 struct ceph_mount_options *fsopt = pctx->opts;
398 struct fs_parse_result result;
402 ret = ceph_parse_param(param, pctx->copts, fc->log.log);
403 if (ret != -ENOPARAM)
406 token = fs_parse(fc, ceph_mount_parameters, param, &result);
407 dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
412 case Opt_snapdirname:
413 kfree(fsopt->snapdir_name);
414 fsopt->snapdir_name = param->string;
415 param->string = NULL;
417 case Opt_mds_namespace:
418 if (!namespace_equals(fsopt, param->string, strlen(param->string)))
419 return invalfc(fc, "Mismatching mds_namespace");
420 kfree(fsopt->mds_namespace);
421 fsopt->mds_namespace = param->string;
422 param->string = NULL;
424 case Opt_recover_session:
425 mode = result.uint_32;
426 if (mode == ceph_recover_session_no)
427 fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
428 else if (mode == ceph_recover_session_clean)
429 fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
435 return invalfc(fc, "Multiple sources specified");
436 return ceph_parse_source(param, fc);
438 return ceph_parse_mon_addr(param, fc);
440 if (result.uint_32 < PAGE_SIZE ||
441 result.uint_32 > CEPH_MAX_WRITE_SIZE)
443 fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
446 if (result.uint_32 < PAGE_SIZE ||
447 result.uint_32 > CEPH_MAX_READ_SIZE)
449 fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
452 fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
454 case Opt_caps_wanted_delay_min:
455 if (result.uint_32 < 1)
457 fsopt->caps_wanted_delay_min = result.uint_32;
459 case Opt_caps_wanted_delay_max:
460 if (result.uint_32 < 1)
462 fsopt->caps_wanted_delay_max = result.uint_32;
465 if (result.int_32 < 0)
467 fsopt->caps_max = result.int_32;
469 case Opt_readdir_max_entries:
470 if (result.uint_32 < 1)
472 fsopt->max_readdir = result.uint_32;
474 case Opt_readdir_max_bytes:
475 if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
477 fsopt->max_readdir_bytes = result.uint_32;
479 case Opt_congestion_kb:
480 if (result.uint_32 < 1024) /* at least 1M */
482 fsopt->congestion_kb = result.uint_32;
486 fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
488 fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
492 fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
494 fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
496 case Opt_asyncreaddir:
498 fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
500 fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
504 fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
506 fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
510 fsopt->flags |= CEPH_MOUNT_OPT_INO32;
512 fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
516 #ifdef CONFIG_CEPH_FSCACHE
517 kfree(fsopt->fscache_uniq);
518 fsopt->fscache_uniq = NULL;
519 if (result.negated) {
520 fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
522 fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
523 fsopt->fscache_uniq = param->string;
524 param->string = NULL;
528 return invalfc(fc, "fscache support is disabled");
532 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
534 fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
536 case Opt_require_active_mds:
538 fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
540 fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
544 fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
546 fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
550 fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
552 fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
555 if (!result.negated) {
556 #ifdef CONFIG_CEPH_FS_POSIX_ACL
557 fc->sb_flags |= SB_POSIXACL;
559 return invalfc(fc, "POSIX ACL support is disabled");
562 fc->sb_flags &= ~SB_POSIXACL;
567 fsopt->flags &= ~CEPH_MOUNT_OPT_ASYNC_DIROPS;
569 fsopt->flags |= CEPH_MOUNT_OPT_ASYNC_DIROPS;
573 fsopt->flags |= CEPH_MOUNT_OPT_NOPAGECACHE;
575 fsopt->flags &= ~CEPH_MOUNT_OPT_NOPAGECACHE;
583 return invalfc(fc, "%s out of range", param->key);
586 static void destroy_mount_options(struct ceph_mount_options *args)
588 dout("destroy_mount_options %p\n", args);
592 kfree(args->snapdir_name);
593 kfree(args->mds_namespace);
594 kfree(args->server_path);
595 kfree(args->fscache_uniq);
596 kfree(args->mon_addr);
600 static int strcmp_null(const char *s1, const char *s2)
608 return strcmp(s1, s2);
611 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
612 struct ceph_options *new_opt,
613 struct ceph_fs_client *fsc)
615 struct ceph_mount_options *fsopt1 = new_fsopt;
616 struct ceph_mount_options *fsopt2 = fsc->mount_options;
617 int ofs = offsetof(struct ceph_mount_options, snapdir_name);
620 ret = memcmp(fsopt1, fsopt2, ofs);
624 ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
628 ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
632 ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
636 ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
640 ret = strcmp_null(fsopt1->mon_addr, fsopt2->mon_addr);
644 return ceph_compare_options(new_opt, fsc->client);
648 * ceph_show_options - Show mount options in /proc/mounts
649 * @m: seq_file to write to
650 * @root: root of that (sub)tree
652 static int ceph_show_options(struct seq_file *m, struct dentry *root)
654 struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
655 struct ceph_mount_options *fsopt = fsc->mount_options;
659 /* a comma between MNT/MS and client options */
663 ret = ceph_print_client_options(m, fsc->client, false);
667 /* retract our comma if no client options */
671 if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
672 seq_puts(m, ",dirstat");
673 if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
674 seq_puts(m, ",rbytes");
675 if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
676 seq_puts(m, ",noasyncreaddir");
677 if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
678 seq_puts(m, ",nodcache");
679 if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
680 seq_puts(m, ",ino32");
681 if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
682 seq_show_option(m, "fsc", fsopt->fscache_uniq);
684 if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
685 seq_puts(m, ",nopoolperm");
686 if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
687 seq_puts(m, ",noquotadf");
689 #ifdef CONFIG_CEPH_FS_POSIX_ACL
690 if (root->d_sb->s_flags & SB_POSIXACL)
693 seq_puts(m, ",noacl");
696 if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
697 seq_puts(m, ",copyfrom");
699 /* dump mds_namespace when old device syntax is in use */
700 if (fsopt->mds_namespace && !fsopt->new_dev_syntax)
701 seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
704 seq_printf(m, ",mon_addr=%s", fsopt->mon_addr);
706 if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
707 seq_show_option(m, "recover_session", "clean");
709 if (!(fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS))
710 seq_puts(m, ",wsync");
712 if (fsopt->flags & CEPH_MOUNT_OPT_NOPAGECACHE)
713 seq_puts(m, ",nopagecache");
715 if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
716 seq_printf(m, ",wsize=%u", fsopt->wsize);
717 if (fsopt->rsize != CEPH_MAX_READ_SIZE)
718 seq_printf(m, ",rsize=%u", fsopt->rsize);
719 if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
720 seq_printf(m, ",rasize=%u", fsopt->rasize);
721 if (fsopt->congestion_kb != default_congestion_kb())
722 seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
724 seq_printf(m, ",caps_max=%d", fsopt->caps_max);
725 if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
726 seq_printf(m, ",caps_wanted_delay_min=%u",
727 fsopt->caps_wanted_delay_min);
728 if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
729 seq_printf(m, ",caps_wanted_delay_max=%u",
730 fsopt->caps_wanted_delay_max);
731 if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
732 seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
733 if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
734 seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
735 if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
736 seq_show_option(m, "snapdirname", fsopt->snapdir_name);
742 * handle any mon messages the standard library doesn't understand.
743 * return error if we don't either.
745 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
747 struct ceph_fs_client *fsc = client->private;
748 int type = le16_to_cpu(msg->hdr.type);
751 case CEPH_MSG_MDS_MAP:
752 ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
754 case CEPH_MSG_FS_MAP_USER:
755 ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
763 * create a new fs client
765 * Success or not, this function consumes @fsopt and @opt.
767 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
768 struct ceph_options *opt)
770 struct ceph_fs_client *fsc;
773 fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
779 fsc->client = ceph_create_client(opt, fsc);
780 if (IS_ERR(fsc->client)) {
781 err = PTR_ERR(fsc->client);
784 opt = NULL; /* fsc->client now owns this */
786 fsc->client->extra_mon_dispatch = extra_mon_dispatch;
787 ceph_set_opt(fsc->client, ABORT_ON_FULL);
789 if (!fsopt->mds_namespace) {
790 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
793 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
797 fsc->mount_options = fsopt;
800 fsc->mount_state = CEPH_MOUNT_MOUNTING;
802 fsc->have_copy_from2 = true;
804 atomic_long_set(&fsc->writeback_count, 0);
805 fsc->write_congested = false;
809 * The number of concurrent works can be high but they don't need
810 * to be processed in parallel, limit concurrency.
812 fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
815 fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
819 hash_init(fsc->async_unlink_conflict);
820 spin_lock_init(&fsc->async_unlink_conflict_lock);
822 spin_lock(&ceph_fsc_lock);
823 list_add_tail(&fsc->metric_wakeup, &ceph_fsc_list);
824 spin_unlock(&ceph_fsc_lock);
829 destroy_workqueue(fsc->inode_wq);
831 ceph_destroy_client(fsc->client);
835 ceph_destroy_options(opt);
836 destroy_mount_options(fsopt);
840 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
842 flush_workqueue(fsc->inode_wq);
843 flush_workqueue(fsc->cap_wq);
846 static void destroy_fs_client(struct ceph_fs_client *fsc)
848 dout("destroy_fs_client %p\n", fsc);
850 spin_lock(&ceph_fsc_lock);
851 list_del(&fsc->metric_wakeup);
852 spin_unlock(&ceph_fsc_lock);
854 ceph_mdsc_destroy(fsc);
855 destroy_workqueue(fsc->inode_wq);
856 destroy_workqueue(fsc->cap_wq);
858 destroy_mount_options(fsc->mount_options);
860 ceph_destroy_client(fsc->client);
863 dout("destroy_fs_client %p done\n", fsc);
869 struct kmem_cache *ceph_inode_cachep;
870 struct kmem_cache *ceph_cap_cachep;
871 struct kmem_cache *ceph_cap_snap_cachep;
872 struct kmem_cache *ceph_cap_flush_cachep;
873 struct kmem_cache *ceph_dentry_cachep;
874 struct kmem_cache *ceph_file_cachep;
875 struct kmem_cache *ceph_dir_file_cachep;
876 struct kmem_cache *ceph_mds_request_cachep;
877 mempool_t *ceph_wb_pagevec_pool;
879 static void ceph_inode_init_once(void *foo)
881 struct ceph_inode_info *ci = foo;
882 inode_init_once(&ci->netfs.inode);
885 static int __init init_caches(void)
889 ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
890 sizeof(struct ceph_inode_info),
891 __alignof__(struct ceph_inode_info),
892 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
893 SLAB_ACCOUNT, ceph_inode_init_once);
894 if (!ceph_inode_cachep)
897 ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
898 if (!ceph_cap_cachep)
900 ceph_cap_snap_cachep = KMEM_CACHE(ceph_cap_snap, SLAB_MEM_SPREAD);
901 if (!ceph_cap_snap_cachep)
903 ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
904 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
905 if (!ceph_cap_flush_cachep)
908 ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
909 SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
910 if (!ceph_dentry_cachep)
913 ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
914 if (!ceph_file_cachep)
917 ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
918 if (!ceph_dir_file_cachep)
921 ceph_mds_request_cachep = KMEM_CACHE(ceph_mds_request, SLAB_MEM_SPREAD);
922 if (!ceph_mds_request_cachep)
925 ceph_wb_pagevec_pool = mempool_create_kmalloc_pool(10, CEPH_MAX_WRITE_SIZE >> PAGE_SHIFT);
926 if (!ceph_wb_pagevec_pool)
927 goto bad_pagevec_pool;
932 kmem_cache_destroy(ceph_mds_request_cachep);
934 kmem_cache_destroy(ceph_dir_file_cachep);
936 kmem_cache_destroy(ceph_file_cachep);
938 kmem_cache_destroy(ceph_dentry_cachep);
940 kmem_cache_destroy(ceph_cap_flush_cachep);
942 kmem_cache_destroy(ceph_cap_snap_cachep);
944 kmem_cache_destroy(ceph_cap_cachep);
946 kmem_cache_destroy(ceph_inode_cachep);
950 static void destroy_caches(void)
953 * Make sure all delayed rcu free inodes are flushed before we
958 kmem_cache_destroy(ceph_inode_cachep);
959 kmem_cache_destroy(ceph_cap_cachep);
960 kmem_cache_destroy(ceph_cap_snap_cachep);
961 kmem_cache_destroy(ceph_cap_flush_cachep);
962 kmem_cache_destroy(ceph_dentry_cachep);
963 kmem_cache_destroy(ceph_file_cachep);
964 kmem_cache_destroy(ceph_dir_file_cachep);
965 kmem_cache_destroy(ceph_mds_request_cachep);
966 mempool_destroy(ceph_wb_pagevec_pool);
969 static void __ceph_umount_begin(struct ceph_fs_client *fsc)
971 ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
972 ceph_mdsc_force_umount(fsc->mdsc);
973 fsc->filp_gen++; // invalidate open files
977 * ceph_umount_begin - initiate forced umount. Tear down the
978 * mount, skipping steps that may hang while waiting for server(s).
980 void ceph_umount_begin(struct super_block *sb)
982 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
984 dout("ceph_umount_begin - starting forced umount\n");
987 fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
988 __ceph_umount_begin(fsc);
991 static const struct super_operations ceph_super_ops = {
992 .alloc_inode = ceph_alloc_inode,
993 .free_inode = ceph_free_inode,
994 .write_inode = ceph_write_inode,
995 .drop_inode = generic_delete_inode,
996 .evict_inode = ceph_evict_inode,
997 .sync_fs = ceph_sync_fs,
998 .put_super = ceph_put_super,
999 .show_options = ceph_show_options,
1000 .statfs = ceph_statfs,
1001 .umount_begin = ceph_umount_begin,
1005 * Bootstrap mount by opening the root directory. Note the mount
1006 * @started time from caller, and time out if this takes too long.
1008 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
1010 unsigned long started)
1012 struct ceph_mds_client *mdsc = fsc->mdsc;
1013 struct ceph_mds_request *req = NULL;
1015 struct dentry *root;
1018 dout("open_root_inode opening '%s'\n", path);
1019 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1021 return ERR_CAST(req);
1022 req->r_path1 = kstrdup(path, GFP_NOFS);
1023 if (!req->r_path1) {
1024 root = ERR_PTR(-ENOMEM);
1028 req->r_ino1.ino = CEPH_INO_ROOT;
1029 req->r_ino1.snap = CEPH_NOSNAP;
1030 req->r_started = started;
1031 req->r_timeout = fsc->client->options->mount_timeout;
1032 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
1033 req->r_num_caps = 2;
1034 err = ceph_mdsc_do_request(mdsc, NULL, req);
1036 struct inode *inode = req->r_target_inode;
1037 req->r_target_inode = NULL;
1038 dout("open_root_inode success\n");
1039 root = d_make_root(inode);
1041 root = ERR_PTR(-ENOMEM);
1044 dout("open_root_inode success, root dentry is %p\n", root);
1046 root = ERR_PTR(err);
1049 ceph_mdsc_put_request(req);
1054 * mount: join the ceph cluster, and open root directory.
1056 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
1057 struct fs_context *fc)
1060 unsigned long started = jiffies; /* note the start time */
1061 struct dentry *root;
1063 dout("mount start %p\n", fsc);
1064 mutex_lock(&fsc->client->mount_mutex);
1066 if (!fsc->sb->s_root) {
1067 const char *path = fsc->mount_options->server_path ?
1068 fsc->mount_options->server_path + 1 : "";
1070 err = __ceph_open_session(fsc->client, started);
1075 if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
1076 err = ceph_fscache_register_fs(fsc, fc);
1081 dout("mount opening path '%s'\n", path);
1083 ceph_fs_debugfs_init(fsc);
1085 root = open_root_dentry(fsc, path, started);
1087 err = PTR_ERR(root);
1090 fsc->sb->s_root = dget(root);
1092 root = dget(fsc->sb->s_root);
1095 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1096 dout("mount success\n");
1097 mutex_unlock(&fsc->client->mount_mutex);
1101 mutex_unlock(&fsc->client->mount_mutex);
1102 return ERR_PTR(err);
1105 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
1107 struct ceph_fs_client *fsc = s->s_fs_info;
1110 dout("set_super %p\n", s);
1112 s->s_maxbytes = MAX_LFS_FILESIZE;
1114 s->s_xattr = ceph_xattr_handlers;
1116 fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
1118 s->s_op = &ceph_super_ops;
1119 s->s_d_op = &ceph_dentry_ops;
1120 s->s_export_op = &ceph_export_ops;
1124 s->s_time_max = U32_MAX;
1125 s->s_flags |= SB_NODIRATIME | SB_NOATIME;
1127 ret = set_anon_super_fc(s, fc);
1134 * share superblock if same fs AND options
1136 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
1138 struct ceph_fs_client *new = fc->s_fs_info;
1139 struct ceph_mount_options *fsopt = new->mount_options;
1140 struct ceph_options *opt = new->client->options;
1141 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1143 dout("ceph_compare_super %p\n", sb);
1145 if (compare_mount_options(fsopt, opt, fsc)) {
1146 dout("monitor(s)/mount options don't match\n");
1149 if ((opt->flags & CEPH_OPT_FSID) &&
1150 ceph_fsid_compare(&opt->fsid, &fsc->client->fsid)) {
1151 dout("fsid doesn't match\n");
1154 if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
1155 dout("flags differ\n");
1159 if (fsc->blocklisted && !ceph_test_mount_opt(fsc, CLEANRECOVER)) {
1160 dout("client is blocklisted (and CLEANRECOVER is not set)\n");
1164 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
1165 dout("client has been forcibly unmounted\n");
1173 * construct our own bdi so we can control readahead, etc.
1175 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1177 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
1181 err = super_setup_bdi_name(sb, "ceph-%ld",
1182 atomic_long_inc_return(&bdi_seq));
1186 /* set ra_pages based on rasize mount option? */
1187 sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1189 /* set io_pages based on max osd read size */
1190 sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1195 static int ceph_get_tree(struct fs_context *fc)
1197 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1198 struct ceph_mount_options *fsopt = pctx->opts;
1199 struct super_block *sb;
1200 struct ceph_fs_client *fsc;
1202 int (*compare_super)(struct super_block *, struct fs_context *) =
1206 dout("ceph_get_tree\n");
1209 return invalfc(fc, "No source");
1210 if (fsopt->new_dev_syntax && !fsopt->mon_addr)
1211 return invalfc(fc, "No monitor address");
1213 /* create client (which we may/may not use) */
1214 fsc = create_fs_client(pctx->opts, pctx->copts);
1222 err = ceph_mdsc_init(fsc);
1226 if (ceph_test_opt(fsc->client, NOSHARE))
1227 compare_super = NULL;
1229 fc->s_fs_info = fsc;
1230 sb = sget_fc(fc, compare_super, ceph_set_super);
1231 fc->s_fs_info = NULL;
1237 if (ceph_sb_to_client(sb) != fsc) {
1238 destroy_fs_client(fsc);
1239 fsc = ceph_sb_to_client(sb);
1240 dout("get_sb got existing client %p\n", fsc);
1242 dout("get_sb using new client %p\n", fsc);
1243 err = ceph_setup_bdi(sb, fsc);
1248 res = ceph_real_mount(fsc, fc);
1253 dout("root %p inode %p ino %llx.%llx\n", res,
1254 d_inode(res), ceph_vinop(d_inode(res)));
1255 fc->root = fsc->sb->s_root;
1259 if (!ceph_mdsmap_is_cluster_available(fsc->mdsc->mdsmap)) {
1260 pr_info("No mds server is up or the cluster is laggy\n");
1261 err = -EHOSTUNREACH;
1264 ceph_mdsc_close_sessions(fsc->mdsc);
1265 deactivate_locked_super(sb);
1269 destroy_fs_client(fsc);
1271 dout("ceph_get_tree fail %d\n", err);
1275 static void ceph_free_fc(struct fs_context *fc)
1277 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1280 destroy_mount_options(pctx->opts);
1281 ceph_destroy_options(pctx->copts);
1286 static int ceph_reconfigure_fc(struct fs_context *fc)
1288 struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1289 struct ceph_mount_options *fsopt = pctx->opts;
1290 struct ceph_fs_client *fsc = ceph_sb_to_client(fc->root->d_sb);
1292 if (fsopt->flags & CEPH_MOUNT_OPT_ASYNC_DIROPS)
1293 ceph_set_mount_opt(fsc, ASYNC_DIROPS);
1295 ceph_clear_mount_opt(fsc, ASYNC_DIROPS);
1297 if (strcmp_null(fsc->mount_options->mon_addr, fsopt->mon_addr)) {
1298 kfree(fsc->mount_options->mon_addr);
1299 fsc->mount_options->mon_addr = fsopt->mon_addr;
1300 fsopt->mon_addr = NULL;
1301 pr_notice("ceph: monitor addresses recorded, but not used for reconnection");
1304 sync_filesystem(fc->root->d_sb);
1308 static const struct fs_context_operations ceph_context_ops = {
1309 .free = ceph_free_fc,
1310 .parse_param = ceph_parse_mount_param,
1311 .get_tree = ceph_get_tree,
1312 .reconfigure = ceph_reconfigure_fc,
1316 * Set up the filesystem mount context.
1318 static int ceph_init_fs_context(struct fs_context *fc)
1320 struct ceph_parse_opts_ctx *pctx;
1321 struct ceph_mount_options *fsopt;
1323 pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1327 pctx->copts = ceph_alloc_options();
1331 pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1336 fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1338 fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1339 fsopt->rsize = CEPH_MAX_READ_SIZE;
1340 fsopt->rasize = CEPH_RASIZE_DEFAULT;
1341 fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1342 if (!fsopt->snapdir_name)
1345 fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1346 fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1347 fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1348 fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1349 fsopt->congestion_kb = default_congestion_kb();
1351 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1352 fc->sb_flags |= SB_POSIXACL;
1355 fc->fs_private = pctx;
1356 fc->ops = &ceph_context_ops;
1360 destroy_mount_options(pctx->opts);
1361 ceph_destroy_options(pctx->copts);
1366 static void ceph_kill_sb(struct super_block *s)
1368 struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1370 dout("kill_sb %p\n", s);
1372 ceph_mdsc_pre_umount(fsc->mdsc);
1373 flush_fs_workqueues(fsc);
1377 fsc->client->extra_mon_dispatch = NULL;
1378 ceph_fs_debugfs_cleanup(fsc);
1380 ceph_fscache_unregister_fs(fsc);
1382 destroy_fs_client(fsc);
1385 static struct file_system_type ceph_fs_type = {
1386 .owner = THIS_MODULE,
1388 .init_fs_context = ceph_init_fs_context,
1389 .kill_sb = ceph_kill_sb,
1390 .fs_flags = FS_RENAME_DOES_D_MOVE,
1392 MODULE_ALIAS_FS("ceph");
1394 int ceph_force_reconnect(struct super_block *sb)
1396 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1399 fsc->mount_state = CEPH_MOUNT_RECOVER;
1400 __ceph_umount_begin(fsc);
1402 /* Make sure all page caches get invalidated.
1403 * see remove_session_caps_cb() */
1404 flush_workqueue(fsc->inode_wq);
1406 /* In case that we were blocklisted. This also reset
1407 * all mon/osd connections */
1408 ceph_reset_client_addr(fsc->client);
1410 ceph_osdc_clear_abort_err(&fsc->client->osdc);
1412 fsc->blocklisted = false;
1413 fsc->mount_state = CEPH_MOUNT_MOUNTED;
1416 err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1417 CEPH_STAT_CAP_INODE, true);
1422 static int __init init_ceph(void)
1424 int ret = init_caches();
1429 ret = register_filesystem(&ceph_fs_type);
1433 pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1443 static void __exit exit_ceph(void)
1445 dout("exit_ceph\n");
1446 unregister_filesystem(&ceph_fs_type);
1450 static int param_set_metrics(const char *val, const struct kernel_param *kp)
1452 struct ceph_fs_client *fsc;
1455 ret = param_set_bool(val, kp);
1457 pr_err("Failed to parse sending metrics switch value '%s'\n",
1460 } else if (!disable_send_metrics) {
1461 // wake up all the mds clients
1462 spin_lock(&ceph_fsc_lock);
1463 list_for_each_entry(fsc, &ceph_fsc_list, metric_wakeup) {
1464 metric_schedule_delayed(&fsc->mdsc->metric);
1466 spin_unlock(&ceph_fsc_lock);
1472 static const struct kernel_param_ops param_ops_metrics = {
1473 .set = param_set_metrics,
1474 .get = param_get_bool,
1477 bool disable_send_metrics = false;
1478 module_param_cb(disable_send_metrics, ¶m_ops_metrics, &disable_send_metrics, 0644);
1479 MODULE_PARM_DESC(disable_send_metrics, "Enable sending perf metrics to ceph cluster (default: on)");
1481 /* for both v1 and v2 syntax */
1482 static bool mount_support = true;
1483 static const struct kernel_param_ops param_ops_mount_syntax = {
1484 .get = param_get_bool,
1486 module_param_cb(mount_syntax_v1, ¶m_ops_mount_syntax, &mount_support, 0444);
1487 module_param_cb(mount_syntax_v2, ¶m_ops_mount_syntax, &mount_support, 0444);
1489 module_init(init_ceph);
1490 module_exit(exit_ceph);
1492 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1493 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1494 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1495 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1496 MODULE_LICENSE("GPL");