xref: /openbmc/linux/fs/ceph/super.c (revision f519cd13)
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 fs_parameter_enum ceph_mount_param_enums[] = {
167 	{ Opt_recover_session,	"no",		ceph_recover_session_no },
168 	{ Opt_recover_session,	"clean",	ceph_recover_session_clean },
169 	{}
170 };
171 
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),
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),
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 static const struct fs_parameter_description ceph_mount_parameters = {
202 	.name           = "ceph",
203 	.specs          = ceph_mount_param_specs,
204 	.enums		= ceph_mount_param_enums,
205 };
206 
207 struct ceph_parse_opts_ctx {
208 	struct ceph_options		*copts;
209 	struct ceph_mount_options	*opts;
210 };
211 
212 /*
213  * Parse the source parameter.  Distinguish the server list from the path.
214  * Internally we do not include the leading '/' in the path.
215  *
216  * The source will look like:
217  *     <server_spec>[,<server_spec>...]:[<path>]
218  * where
219  *     <server_spec> is <ip>[:<port>]
220  *     <path> is optional, but if present must begin with '/'
221  */
222 static int ceph_parse_source(struct fs_parameter *param, struct fs_context *fc)
223 {
224 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
225 	struct ceph_mount_options *fsopt = pctx->opts;
226 	char *dev_name = param->string, *dev_name_end;
227 	int ret;
228 
229 	dout("%s '%s'\n", __func__, dev_name);
230 	if (!dev_name || !*dev_name)
231 		return invalf(fc, "ceph: Empty source");
232 
233 	dev_name_end = strchr(dev_name, '/');
234 	if (dev_name_end) {
235 		if (strlen(dev_name_end) > 1) {
236 			kfree(fsopt->server_path);
237 			fsopt->server_path = kstrdup(dev_name_end, GFP_KERNEL);
238 			if (!fsopt->server_path)
239 				return -ENOMEM;
240 		}
241 	} else {
242 		dev_name_end = dev_name + strlen(dev_name);
243 	}
244 
245 	dev_name_end--;		/* back up to ':' separator */
246 	if (dev_name_end < dev_name || *dev_name_end != ':')
247 		return invalf(fc, "ceph: No path or : separator in source");
248 
249 	dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name);
250 	if (fsopt->server_path)
251 		dout("server path '%s'\n", fsopt->server_path);
252 
253 	ret = ceph_parse_mon_ips(param->string, dev_name_end - dev_name,
254 				 pctx->copts, fc);
255 	if (ret)
256 		return ret;
257 
258 	fc->source = param->string;
259 	param->string = NULL;
260 	return 0;
261 }
262 
263 static int ceph_parse_mount_param(struct fs_context *fc,
264 				  struct fs_parameter *param)
265 {
266 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
267 	struct ceph_mount_options *fsopt = pctx->opts;
268 	struct fs_parse_result result;
269 	unsigned int mode;
270 	int token, ret;
271 
272 	ret = ceph_parse_param(param, pctx->copts, fc);
273 	if (ret != -ENOPARAM)
274 		return ret;
275 
276 	token = fs_parse(fc, &ceph_mount_parameters, param, &result);
277 	dout("%s fs_parse '%s' token %d\n", __func__, param->key, token);
278 	if (token < 0)
279 		return token;
280 
281 	switch (token) {
282 	case Opt_snapdirname:
283 		kfree(fsopt->snapdir_name);
284 		fsopt->snapdir_name = param->string;
285 		param->string = NULL;
286 		break;
287 	case Opt_mds_namespace:
288 		kfree(fsopt->mds_namespace);
289 		fsopt->mds_namespace = param->string;
290 		param->string = NULL;
291 		break;
292 	case Opt_recover_session:
293 		mode = result.uint_32;
294 		if (mode == ceph_recover_session_no)
295 			fsopt->flags &= ~CEPH_MOUNT_OPT_CLEANRECOVER;
296 		else if (mode == ceph_recover_session_clean)
297 			fsopt->flags |= CEPH_MOUNT_OPT_CLEANRECOVER;
298 		else
299 			BUG();
300 		break;
301 	case Opt_source:
302 		if (fc->source)
303 			return invalf(fc, "ceph: Multiple sources specified");
304 		return ceph_parse_source(param, fc);
305 	case Opt_wsize:
306 		if (result.uint_32 < PAGE_SIZE ||
307 		    result.uint_32 > CEPH_MAX_WRITE_SIZE)
308 			goto out_of_range;
309 		fsopt->wsize = ALIGN(result.uint_32, PAGE_SIZE);
310 		break;
311 	case Opt_rsize:
312 		if (result.uint_32 < PAGE_SIZE ||
313 		    result.uint_32 > CEPH_MAX_READ_SIZE)
314 			goto out_of_range;
315 		fsopt->rsize = ALIGN(result.uint_32, PAGE_SIZE);
316 		break;
317 	case Opt_rasize:
318 		fsopt->rasize = ALIGN(result.uint_32, PAGE_SIZE);
319 		break;
320 	case Opt_caps_wanted_delay_min:
321 		if (result.uint_32 < 1)
322 			goto out_of_range;
323 		fsopt->caps_wanted_delay_min = result.uint_32;
324 		break;
325 	case Opt_caps_wanted_delay_max:
326 		if (result.uint_32 < 1)
327 			goto out_of_range;
328 		fsopt->caps_wanted_delay_max = result.uint_32;
329 		break;
330 	case Opt_caps_max:
331 		if (result.int_32 < 0)
332 			goto out_of_range;
333 		fsopt->caps_max = result.int_32;
334 		break;
335 	case Opt_readdir_max_entries:
336 		if (result.uint_32 < 1)
337 			goto out_of_range;
338 		fsopt->max_readdir = result.uint_32;
339 		break;
340 	case Opt_readdir_max_bytes:
341 		if (result.uint_32 < PAGE_SIZE && result.uint_32 != 0)
342 			goto out_of_range;
343 		fsopt->max_readdir_bytes = result.uint_32;
344 		break;
345 	case Opt_congestion_kb:
346 		if (result.uint_32 < 1024) /* at least 1M */
347 			goto out_of_range;
348 		fsopt->congestion_kb = result.uint_32;
349 		break;
350 	case Opt_dirstat:
351 		if (!result.negated)
352 			fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT;
353 		else
354 			fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT;
355 		break;
356 	case Opt_rbytes:
357 		if (!result.negated)
358 			fsopt->flags |= CEPH_MOUNT_OPT_RBYTES;
359 		else
360 			fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES;
361 		break;
362 	case Opt_asyncreaddir:
363 		if (!result.negated)
364 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR;
365 		else
366 			fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR;
367 		break;
368 	case Opt_dcache:
369 		if (!result.negated)
370 			fsopt->flags |= CEPH_MOUNT_OPT_DCACHE;
371 		else
372 			fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE;
373 		break;
374 	case Opt_ino32:
375 		if (!result.negated)
376 			fsopt->flags |= CEPH_MOUNT_OPT_INO32;
377 		else
378 			fsopt->flags &= ~CEPH_MOUNT_OPT_INO32;
379 		break;
380 
381 	case Opt_fscache:
382 #ifdef CONFIG_CEPH_FSCACHE
383 		kfree(fsopt->fscache_uniq);
384 		fsopt->fscache_uniq = NULL;
385 		if (result.negated) {
386 			fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE;
387 		} else {
388 			fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE;
389 			fsopt->fscache_uniq = param->string;
390 			param->string = NULL;
391 		}
392 		break;
393 #else
394 		return invalf(fc, "ceph: fscache support is disabled");
395 #endif
396 	case Opt_poolperm:
397 		if (!result.negated)
398 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM;
399 		else
400 			fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM;
401 		break;
402 	case Opt_require_active_mds:
403 		if (!result.negated)
404 			fsopt->flags &= ~CEPH_MOUNT_OPT_MOUNTWAIT;
405 		else
406 			fsopt->flags |= CEPH_MOUNT_OPT_MOUNTWAIT;
407 		break;
408 	case Opt_quotadf:
409 		if (!result.negated)
410 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOQUOTADF;
411 		else
412 			fsopt->flags |= CEPH_MOUNT_OPT_NOQUOTADF;
413 		break;
414 	case Opt_copyfrom:
415 		if (!result.negated)
416 			fsopt->flags &= ~CEPH_MOUNT_OPT_NOCOPYFROM;
417 		else
418 			fsopt->flags |= CEPH_MOUNT_OPT_NOCOPYFROM;
419 		break;
420 	case Opt_acl:
421 		if (!result.negated) {
422 #ifdef CONFIG_CEPH_FS_POSIX_ACL
423 			fc->sb_flags |= SB_POSIXACL;
424 #else
425 			return invalf(fc, "ceph: POSIX ACL support is disabled");
426 #endif
427 		} else {
428 			fc->sb_flags &= ~SB_POSIXACL;
429 		}
430 		break;
431 	default:
432 		BUG();
433 	}
434 	return 0;
435 
436 out_of_range:
437 	return invalf(fc, "ceph: %s out of range", param->key);
438 }
439 
440 static void destroy_mount_options(struct ceph_mount_options *args)
441 {
442 	dout("destroy_mount_options %p\n", args);
443 	if (!args)
444 		return;
445 
446 	kfree(args->snapdir_name);
447 	kfree(args->mds_namespace);
448 	kfree(args->server_path);
449 	kfree(args->fscache_uniq);
450 	kfree(args);
451 }
452 
453 static int strcmp_null(const char *s1, const char *s2)
454 {
455 	if (!s1 && !s2)
456 		return 0;
457 	if (s1 && !s2)
458 		return -1;
459 	if (!s1 && s2)
460 		return 1;
461 	return strcmp(s1, s2);
462 }
463 
464 static int compare_mount_options(struct ceph_mount_options *new_fsopt,
465 				 struct ceph_options *new_opt,
466 				 struct ceph_fs_client *fsc)
467 {
468 	struct ceph_mount_options *fsopt1 = new_fsopt;
469 	struct ceph_mount_options *fsopt2 = fsc->mount_options;
470 	int ofs = offsetof(struct ceph_mount_options, snapdir_name);
471 	int ret;
472 
473 	ret = memcmp(fsopt1, fsopt2, ofs);
474 	if (ret)
475 		return ret;
476 
477 	ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name);
478 	if (ret)
479 		return ret;
480 	ret = strcmp_null(fsopt1->mds_namespace, fsopt2->mds_namespace);
481 	if (ret)
482 		return ret;
483 	ret = strcmp_null(fsopt1->server_path, fsopt2->server_path);
484 	if (ret)
485 		return ret;
486 	ret = strcmp_null(fsopt1->fscache_uniq, fsopt2->fscache_uniq);
487 	if (ret)
488 		return ret;
489 
490 	return ceph_compare_options(new_opt, fsc->client);
491 }
492 
493 /**
494  * ceph_show_options - Show mount options in /proc/mounts
495  * @m: seq_file to write to
496  * @root: root of that (sub)tree
497  */
498 static int ceph_show_options(struct seq_file *m, struct dentry *root)
499 {
500 	struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb);
501 	struct ceph_mount_options *fsopt = fsc->mount_options;
502 	size_t pos;
503 	int ret;
504 
505 	/* a comma between MNT/MS and client options */
506 	seq_putc(m, ',');
507 	pos = m->count;
508 
509 	ret = ceph_print_client_options(m, fsc->client, false);
510 	if (ret)
511 		return ret;
512 
513 	/* retract our comma if no client options */
514 	if (m->count == pos)
515 		m->count--;
516 
517 	if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT)
518 		seq_puts(m, ",dirstat");
519 	if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES))
520 		seq_puts(m, ",rbytes");
521 	if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR)
522 		seq_puts(m, ",noasyncreaddir");
523 	if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0)
524 		seq_puts(m, ",nodcache");
525 	if (fsopt->flags & CEPH_MOUNT_OPT_INO32)
526 		seq_puts(m, ",ino32");
527 	if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) {
528 		seq_show_option(m, "fsc", fsopt->fscache_uniq);
529 	}
530 	if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM)
531 		seq_puts(m, ",nopoolperm");
532 	if (fsopt->flags & CEPH_MOUNT_OPT_NOQUOTADF)
533 		seq_puts(m, ",noquotadf");
534 
535 #ifdef CONFIG_CEPH_FS_POSIX_ACL
536 	if (root->d_sb->s_flags & SB_POSIXACL)
537 		seq_puts(m, ",acl");
538 	else
539 		seq_puts(m, ",noacl");
540 #endif
541 
542 	if ((fsopt->flags & CEPH_MOUNT_OPT_NOCOPYFROM) == 0)
543 		seq_puts(m, ",copyfrom");
544 
545 	if (fsopt->mds_namespace)
546 		seq_show_option(m, "mds_namespace", fsopt->mds_namespace);
547 
548 	if (fsopt->flags & CEPH_MOUNT_OPT_CLEANRECOVER)
549 		seq_show_option(m, "recover_session", "clean");
550 
551 	if (fsopt->wsize != CEPH_MAX_WRITE_SIZE)
552 		seq_printf(m, ",wsize=%u", fsopt->wsize);
553 	if (fsopt->rsize != CEPH_MAX_READ_SIZE)
554 		seq_printf(m, ",rsize=%u", fsopt->rsize);
555 	if (fsopt->rasize != CEPH_RASIZE_DEFAULT)
556 		seq_printf(m, ",rasize=%u", fsopt->rasize);
557 	if (fsopt->congestion_kb != default_congestion_kb())
558 		seq_printf(m, ",write_congestion_kb=%u", fsopt->congestion_kb);
559 	if (fsopt->caps_max)
560 		seq_printf(m, ",caps_max=%d", fsopt->caps_max);
561 	if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT)
562 		seq_printf(m, ",caps_wanted_delay_min=%u",
563 			 fsopt->caps_wanted_delay_min);
564 	if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT)
565 		seq_printf(m, ",caps_wanted_delay_max=%u",
566 			   fsopt->caps_wanted_delay_max);
567 	if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT)
568 		seq_printf(m, ",readdir_max_entries=%u", fsopt->max_readdir);
569 	if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT)
570 		seq_printf(m, ",readdir_max_bytes=%u", fsopt->max_readdir_bytes);
571 	if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT))
572 		seq_show_option(m, "snapdirname", fsopt->snapdir_name);
573 
574 	return 0;
575 }
576 
577 /*
578  * handle any mon messages the standard library doesn't understand.
579  * return error if we don't either.
580  */
581 static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg)
582 {
583 	struct ceph_fs_client *fsc = client->private;
584 	int type = le16_to_cpu(msg->hdr.type);
585 
586 	switch (type) {
587 	case CEPH_MSG_MDS_MAP:
588 		ceph_mdsc_handle_mdsmap(fsc->mdsc, msg);
589 		return 0;
590 	case CEPH_MSG_FS_MAP_USER:
591 		ceph_mdsc_handle_fsmap(fsc->mdsc, msg);
592 		return 0;
593 	default:
594 		return -1;
595 	}
596 }
597 
598 /*
599  * create a new fs client
600  *
601  * Success or not, this function consumes @fsopt and @opt.
602  */
603 static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt,
604 					struct ceph_options *opt)
605 {
606 	struct ceph_fs_client *fsc;
607 	int page_count;
608 	size_t size;
609 	int err;
610 
611 	fsc = kzalloc(sizeof(*fsc), GFP_KERNEL);
612 	if (!fsc) {
613 		err = -ENOMEM;
614 		goto fail;
615 	}
616 
617 	fsc->client = ceph_create_client(opt, fsc);
618 	if (IS_ERR(fsc->client)) {
619 		err = PTR_ERR(fsc->client);
620 		goto fail;
621 	}
622 	opt = NULL; /* fsc->client now owns this */
623 
624 	fsc->client->extra_mon_dispatch = extra_mon_dispatch;
625 	ceph_set_opt(fsc->client, ABORT_ON_FULL);
626 
627 	if (!fsopt->mds_namespace) {
628 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
629 				   0, true);
630 	} else {
631 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_FSMAP,
632 				   0, false);
633 	}
634 
635 	fsc->mount_options = fsopt;
636 
637 	fsc->sb = NULL;
638 	fsc->mount_state = CEPH_MOUNT_MOUNTING;
639 	fsc->filp_gen = 1;
640 
641 	atomic_long_set(&fsc->writeback_count, 0);
642 
643 	err = -ENOMEM;
644 	/*
645 	 * The number of concurrent works can be high but they don't need
646 	 * to be processed in parallel, limit concurrency.
647 	 */
648 	fsc->inode_wq = alloc_workqueue("ceph-inode", WQ_UNBOUND, 0);
649 	if (!fsc->inode_wq)
650 		goto fail_client;
651 	fsc->cap_wq = alloc_workqueue("ceph-cap", 0, 1);
652 	if (!fsc->cap_wq)
653 		goto fail_inode_wq;
654 
655 	/* set up mempools */
656 	err = -ENOMEM;
657 	page_count = fsc->mount_options->wsize >> PAGE_SHIFT;
658 	size = sizeof (struct page *) * (page_count ? page_count : 1);
659 	fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size);
660 	if (!fsc->wb_pagevec_pool)
661 		goto fail_cap_wq;
662 
663 	return fsc;
664 
665 fail_cap_wq:
666 	destroy_workqueue(fsc->cap_wq);
667 fail_inode_wq:
668 	destroy_workqueue(fsc->inode_wq);
669 fail_client:
670 	ceph_destroy_client(fsc->client);
671 fail:
672 	kfree(fsc);
673 	if (opt)
674 		ceph_destroy_options(opt);
675 	destroy_mount_options(fsopt);
676 	return ERR_PTR(err);
677 }
678 
679 static void flush_fs_workqueues(struct ceph_fs_client *fsc)
680 {
681 	flush_workqueue(fsc->inode_wq);
682 	flush_workqueue(fsc->cap_wq);
683 }
684 
685 static void destroy_fs_client(struct ceph_fs_client *fsc)
686 {
687 	dout("destroy_fs_client %p\n", fsc);
688 
689 	ceph_mdsc_destroy(fsc);
690 	destroy_workqueue(fsc->inode_wq);
691 	destroy_workqueue(fsc->cap_wq);
692 
693 	mempool_destroy(fsc->wb_pagevec_pool);
694 
695 	destroy_mount_options(fsc->mount_options);
696 
697 	ceph_destroy_client(fsc->client);
698 
699 	kfree(fsc);
700 	dout("destroy_fs_client %p done\n", fsc);
701 }
702 
703 /*
704  * caches
705  */
706 struct kmem_cache *ceph_inode_cachep;
707 struct kmem_cache *ceph_cap_cachep;
708 struct kmem_cache *ceph_cap_flush_cachep;
709 struct kmem_cache *ceph_dentry_cachep;
710 struct kmem_cache *ceph_file_cachep;
711 struct kmem_cache *ceph_dir_file_cachep;
712 
713 static void ceph_inode_init_once(void *foo)
714 {
715 	struct ceph_inode_info *ci = foo;
716 	inode_init_once(&ci->vfs_inode);
717 }
718 
719 static int __init init_caches(void)
720 {
721 	int error = -ENOMEM;
722 
723 	ceph_inode_cachep = kmem_cache_create("ceph_inode_info",
724 				      sizeof(struct ceph_inode_info),
725 				      __alignof__(struct ceph_inode_info),
726 				      SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD|
727 				      SLAB_ACCOUNT, ceph_inode_init_once);
728 	if (!ceph_inode_cachep)
729 		return -ENOMEM;
730 
731 	ceph_cap_cachep = KMEM_CACHE(ceph_cap, SLAB_MEM_SPREAD);
732 	if (!ceph_cap_cachep)
733 		goto bad_cap;
734 	ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush,
735 					   SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
736 	if (!ceph_cap_flush_cachep)
737 		goto bad_cap_flush;
738 
739 	ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info,
740 					SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD);
741 	if (!ceph_dentry_cachep)
742 		goto bad_dentry;
743 
744 	ceph_file_cachep = KMEM_CACHE(ceph_file_info, SLAB_MEM_SPREAD);
745 	if (!ceph_file_cachep)
746 		goto bad_file;
747 
748 	ceph_dir_file_cachep = KMEM_CACHE(ceph_dir_file_info, SLAB_MEM_SPREAD);
749 	if (!ceph_dir_file_cachep)
750 		goto bad_dir_file;
751 
752 	error = ceph_fscache_register();
753 	if (error)
754 		goto bad_fscache;
755 
756 	return 0;
757 
758 bad_fscache:
759 	kmem_cache_destroy(ceph_dir_file_cachep);
760 bad_dir_file:
761 	kmem_cache_destroy(ceph_file_cachep);
762 bad_file:
763 	kmem_cache_destroy(ceph_dentry_cachep);
764 bad_dentry:
765 	kmem_cache_destroy(ceph_cap_flush_cachep);
766 bad_cap_flush:
767 	kmem_cache_destroy(ceph_cap_cachep);
768 bad_cap:
769 	kmem_cache_destroy(ceph_inode_cachep);
770 	return error;
771 }
772 
773 static void destroy_caches(void)
774 {
775 	/*
776 	 * Make sure all delayed rcu free inodes are flushed before we
777 	 * destroy cache.
778 	 */
779 	rcu_barrier();
780 
781 	kmem_cache_destroy(ceph_inode_cachep);
782 	kmem_cache_destroy(ceph_cap_cachep);
783 	kmem_cache_destroy(ceph_cap_flush_cachep);
784 	kmem_cache_destroy(ceph_dentry_cachep);
785 	kmem_cache_destroy(ceph_file_cachep);
786 	kmem_cache_destroy(ceph_dir_file_cachep);
787 
788 	ceph_fscache_unregister();
789 }
790 
791 
792 /*
793  * ceph_umount_begin - initiate forced umount.  Tear down down the
794  * mount, skipping steps that may hang while waiting for server(s).
795  */
796 static void ceph_umount_begin(struct super_block *sb)
797 {
798 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
799 
800 	dout("ceph_umount_begin - starting forced umount\n");
801 	if (!fsc)
802 		return;
803 	fsc->mount_state = CEPH_MOUNT_SHUTDOWN;
804 	ceph_osdc_abort_requests(&fsc->client->osdc, -EIO);
805 	ceph_mdsc_force_umount(fsc->mdsc);
806 	fsc->filp_gen++; // invalidate open files
807 }
808 
809 static const struct super_operations ceph_super_ops = {
810 	.alloc_inode	= ceph_alloc_inode,
811 	.free_inode	= ceph_free_inode,
812 	.write_inode    = ceph_write_inode,
813 	.drop_inode	= generic_delete_inode,
814 	.evict_inode	= ceph_evict_inode,
815 	.sync_fs        = ceph_sync_fs,
816 	.put_super	= ceph_put_super,
817 	.show_options   = ceph_show_options,
818 	.statfs		= ceph_statfs,
819 	.umount_begin   = ceph_umount_begin,
820 };
821 
822 /*
823  * Bootstrap mount by opening the root directory.  Note the mount
824  * @started time from caller, and time out if this takes too long.
825  */
826 static struct dentry *open_root_dentry(struct ceph_fs_client *fsc,
827 				       const char *path,
828 				       unsigned long started)
829 {
830 	struct ceph_mds_client *mdsc = fsc->mdsc;
831 	struct ceph_mds_request *req = NULL;
832 	int err;
833 	struct dentry *root;
834 
835 	/* open dir */
836 	dout("open_root_inode opening '%s'\n", path);
837 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
838 	if (IS_ERR(req))
839 		return ERR_CAST(req);
840 	req->r_path1 = kstrdup(path, GFP_NOFS);
841 	if (!req->r_path1) {
842 		root = ERR_PTR(-ENOMEM);
843 		goto out;
844 	}
845 
846 	req->r_ino1.ino = CEPH_INO_ROOT;
847 	req->r_ino1.snap = CEPH_NOSNAP;
848 	req->r_started = started;
849 	req->r_timeout = fsc->client->options->mount_timeout;
850 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
851 	req->r_num_caps = 2;
852 	err = ceph_mdsc_do_request(mdsc, NULL, req);
853 	if (err == 0) {
854 		struct inode *inode = req->r_target_inode;
855 		req->r_target_inode = NULL;
856 		dout("open_root_inode success\n");
857 		root = d_make_root(inode);
858 		if (!root) {
859 			root = ERR_PTR(-ENOMEM);
860 			goto out;
861 		}
862 		dout("open_root_inode success, root dentry is %p\n", root);
863 	} else {
864 		root = ERR_PTR(err);
865 	}
866 out:
867 	ceph_mdsc_put_request(req);
868 	return root;
869 }
870 
871 
872 
873 
874 /*
875  * mount: join the ceph cluster, and open root directory.
876  */
877 static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc,
878 				      struct fs_context *fc)
879 {
880 	int err;
881 	unsigned long started = jiffies;  /* note the start time */
882 	struct dentry *root;
883 
884 	dout("mount start %p\n", fsc);
885 	mutex_lock(&fsc->client->mount_mutex);
886 
887 	if (!fsc->sb->s_root) {
888 		const char *path;
889 		err = __ceph_open_session(fsc->client, started);
890 		if (err < 0)
891 			goto out;
892 
893 		/* setup fscache */
894 		if (fsc->mount_options->flags & CEPH_MOUNT_OPT_FSCACHE) {
895 			err = ceph_fscache_register_fs(fsc, fc);
896 			if (err < 0)
897 				goto out;
898 		}
899 
900 		if (!fsc->mount_options->server_path) {
901 			path = "";
902 			dout("mount opening path \\t\n");
903 		} else {
904 			path = fsc->mount_options->server_path + 1;
905 			dout("mount opening path %s\n", path);
906 		}
907 
908 		ceph_fs_debugfs_init(fsc);
909 
910 		root = open_root_dentry(fsc, path, started);
911 		if (IS_ERR(root)) {
912 			err = PTR_ERR(root);
913 			goto out;
914 		}
915 		fsc->sb->s_root = dget(root);
916 	} else {
917 		root = dget(fsc->sb->s_root);
918 	}
919 
920 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
921 	dout("mount success\n");
922 	mutex_unlock(&fsc->client->mount_mutex);
923 	return root;
924 
925 out:
926 	mutex_unlock(&fsc->client->mount_mutex);
927 	return ERR_PTR(err);
928 }
929 
930 static int ceph_set_super(struct super_block *s, struct fs_context *fc)
931 {
932 	struct ceph_fs_client *fsc = s->s_fs_info;
933 	int ret;
934 
935 	dout("set_super %p\n", s);
936 
937 	s->s_maxbytes = MAX_LFS_FILESIZE;
938 
939 	s->s_xattr = ceph_xattr_handlers;
940 	fsc->sb = s;
941 	fsc->max_file_size = 1ULL << 40; /* temp value until we get mdsmap */
942 
943 	s->s_op = &ceph_super_ops;
944 	s->s_d_op = &ceph_dentry_ops;
945 	s->s_export_op = &ceph_export_ops;
946 
947 	s->s_time_gran = 1;
948 	s->s_time_min = 0;
949 	s->s_time_max = U32_MAX;
950 
951 	ret = set_anon_super_fc(s, fc);
952 	if (ret != 0)
953 		fsc->sb = NULL;
954 	return ret;
955 }
956 
957 /*
958  * share superblock if same fs AND options
959  */
960 static int ceph_compare_super(struct super_block *sb, struct fs_context *fc)
961 {
962 	struct ceph_fs_client *new = fc->s_fs_info;
963 	struct ceph_mount_options *fsopt = new->mount_options;
964 	struct ceph_options *opt = new->client->options;
965 	struct ceph_fs_client *other = ceph_sb_to_client(sb);
966 
967 	dout("ceph_compare_super %p\n", sb);
968 
969 	if (compare_mount_options(fsopt, opt, other)) {
970 		dout("monitor(s)/mount options don't match\n");
971 		return 0;
972 	}
973 	if ((opt->flags & CEPH_OPT_FSID) &&
974 	    ceph_fsid_compare(&opt->fsid, &other->client->fsid)) {
975 		dout("fsid doesn't match\n");
976 		return 0;
977 	}
978 	if (fc->sb_flags != (sb->s_flags & ~SB_BORN)) {
979 		dout("flags differ\n");
980 		return 0;
981 	}
982 	return 1;
983 }
984 
985 /*
986  * construct our own bdi so we can control readahead, etc.
987  */
988 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
989 
990 static int ceph_setup_bdi(struct super_block *sb, struct ceph_fs_client *fsc)
991 {
992 	int err;
993 
994 	err = super_setup_bdi_name(sb, "ceph-%ld",
995 				   atomic_long_inc_return(&bdi_seq));
996 	if (err)
997 		return err;
998 
999 	/* set ra_pages based on rasize mount option? */
1000 	sb->s_bdi->ra_pages = fsc->mount_options->rasize >> PAGE_SHIFT;
1001 
1002 	/* set io_pages based on max osd read size */
1003 	sb->s_bdi->io_pages = fsc->mount_options->rsize >> PAGE_SHIFT;
1004 
1005 	return 0;
1006 }
1007 
1008 static int ceph_get_tree(struct fs_context *fc)
1009 {
1010 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1011 	struct super_block *sb;
1012 	struct ceph_fs_client *fsc;
1013 	struct dentry *res;
1014 	int (*compare_super)(struct super_block *, struct fs_context *) =
1015 		ceph_compare_super;
1016 	int err;
1017 
1018 	dout("ceph_get_tree\n");
1019 
1020 	if (!fc->source)
1021 		return invalf(fc, "ceph: No source");
1022 
1023 #ifdef CONFIG_CEPH_FS_POSIX_ACL
1024 	fc->sb_flags |= SB_POSIXACL;
1025 #endif
1026 
1027 	/* create client (which we may/may not use) */
1028 	fsc = create_fs_client(pctx->opts, pctx->copts);
1029 	pctx->opts = NULL;
1030 	pctx->copts = NULL;
1031 	if (IS_ERR(fsc)) {
1032 		err = PTR_ERR(fsc);
1033 		goto out_final;
1034 	}
1035 
1036 	err = ceph_mdsc_init(fsc);
1037 	if (err < 0)
1038 		goto out;
1039 
1040 	if (ceph_test_opt(fsc->client, NOSHARE))
1041 		compare_super = NULL;
1042 
1043 	fc->s_fs_info = fsc;
1044 	sb = sget_fc(fc, compare_super, ceph_set_super);
1045 	fc->s_fs_info = NULL;
1046 	if (IS_ERR(sb)) {
1047 		err = PTR_ERR(sb);
1048 		goto out;
1049 	}
1050 
1051 	if (ceph_sb_to_client(sb) != fsc) {
1052 		destroy_fs_client(fsc);
1053 		fsc = ceph_sb_to_client(sb);
1054 		dout("get_sb got existing client %p\n", fsc);
1055 	} else {
1056 		dout("get_sb using new client %p\n", fsc);
1057 		err = ceph_setup_bdi(sb, fsc);
1058 		if (err < 0)
1059 			goto out_splat;
1060 	}
1061 
1062 	res = ceph_real_mount(fsc, fc);
1063 	if (IS_ERR(res)) {
1064 		err = PTR_ERR(res);
1065 		goto out_splat;
1066 	}
1067 	dout("root %p inode %p ino %llx.%llx\n", res,
1068 	     d_inode(res), ceph_vinop(d_inode(res)));
1069 	fc->root = fsc->sb->s_root;
1070 	return 0;
1071 
1072 out_splat:
1073 	ceph_mdsc_close_sessions(fsc->mdsc);
1074 	deactivate_locked_super(sb);
1075 	goto out_final;
1076 
1077 out:
1078 	destroy_fs_client(fsc);
1079 out_final:
1080 	dout("ceph_get_tree fail %d\n", err);
1081 	return err;
1082 }
1083 
1084 static void ceph_free_fc(struct fs_context *fc)
1085 {
1086 	struct ceph_parse_opts_ctx *pctx = fc->fs_private;
1087 
1088 	if (pctx) {
1089 		destroy_mount_options(pctx->opts);
1090 		ceph_destroy_options(pctx->copts);
1091 		kfree(pctx);
1092 	}
1093 }
1094 
1095 static int ceph_reconfigure_fc(struct fs_context *fc)
1096 {
1097 	sync_filesystem(fc->root->d_sb);
1098 	return 0;
1099 }
1100 
1101 static const struct fs_context_operations ceph_context_ops = {
1102 	.free		= ceph_free_fc,
1103 	.parse_param	= ceph_parse_mount_param,
1104 	.get_tree	= ceph_get_tree,
1105 	.reconfigure	= ceph_reconfigure_fc,
1106 };
1107 
1108 /*
1109  * Set up the filesystem mount context.
1110  */
1111 static int ceph_init_fs_context(struct fs_context *fc)
1112 {
1113 	struct ceph_parse_opts_ctx *pctx;
1114 	struct ceph_mount_options *fsopt;
1115 
1116 	pctx = kzalloc(sizeof(*pctx), GFP_KERNEL);
1117 	if (!pctx)
1118 		return -ENOMEM;
1119 
1120 	pctx->copts = ceph_alloc_options();
1121 	if (!pctx->copts)
1122 		goto nomem;
1123 
1124 	pctx->opts = kzalloc(sizeof(*pctx->opts), GFP_KERNEL);
1125 	if (!pctx->opts)
1126 		goto nomem;
1127 
1128 	fsopt = pctx->opts;
1129 	fsopt->flags = CEPH_MOUNT_OPT_DEFAULT;
1130 
1131 	fsopt->wsize = CEPH_MAX_WRITE_SIZE;
1132 	fsopt->rsize = CEPH_MAX_READ_SIZE;
1133 	fsopt->rasize = CEPH_RASIZE_DEFAULT;
1134 	fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL);
1135 	if (!fsopt->snapdir_name)
1136 		goto nomem;
1137 
1138 	fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT;
1139 	fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT;
1140 	fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT;
1141 	fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT;
1142 	fsopt->congestion_kb = default_congestion_kb();
1143 
1144 	fc->fs_private = pctx;
1145 	fc->ops = &ceph_context_ops;
1146 	return 0;
1147 
1148 nomem:
1149 	destroy_mount_options(pctx->opts);
1150 	ceph_destroy_options(pctx->copts);
1151 	kfree(pctx);
1152 	return -ENOMEM;
1153 }
1154 
1155 static void ceph_kill_sb(struct super_block *s)
1156 {
1157 	struct ceph_fs_client *fsc = ceph_sb_to_client(s);
1158 	dev_t dev = s->s_dev;
1159 
1160 	dout("kill_sb %p\n", s);
1161 
1162 	ceph_mdsc_pre_umount(fsc->mdsc);
1163 	flush_fs_workqueues(fsc);
1164 
1165 	generic_shutdown_super(s);
1166 
1167 	fsc->client->extra_mon_dispatch = NULL;
1168 	ceph_fs_debugfs_cleanup(fsc);
1169 
1170 	ceph_fscache_unregister_fs(fsc);
1171 
1172 	destroy_fs_client(fsc);
1173 	free_anon_bdev(dev);
1174 }
1175 
1176 static struct file_system_type ceph_fs_type = {
1177 	.owner		= THIS_MODULE,
1178 	.name		= "ceph",
1179 	.init_fs_context = ceph_init_fs_context,
1180 	.kill_sb	= ceph_kill_sb,
1181 	.fs_flags	= FS_RENAME_DOES_D_MOVE,
1182 };
1183 MODULE_ALIAS_FS("ceph");
1184 
1185 int ceph_force_reconnect(struct super_block *sb)
1186 {
1187 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1188 	int err = 0;
1189 
1190 	ceph_umount_begin(sb);
1191 
1192 	/* Make sure all page caches get invalidated.
1193 	 * see remove_session_caps_cb() */
1194 	flush_workqueue(fsc->inode_wq);
1195 
1196 	/* In case that we were blacklisted. This also reset
1197 	 * all mon/osd connections */
1198 	ceph_reset_client_addr(fsc->client);
1199 
1200 	ceph_osdc_clear_abort_err(&fsc->client->osdc);
1201 
1202 	fsc->blacklisted = false;
1203 	fsc->mount_state = CEPH_MOUNT_MOUNTED;
1204 
1205 	if (sb->s_root) {
1206 		err = __ceph_do_getattr(d_inode(sb->s_root), NULL,
1207 					CEPH_STAT_CAP_INODE, true);
1208 	}
1209 	return err;
1210 }
1211 
1212 static int __init init_ceph(void)
1213 {
1214 	int ret = init_caches();
1215 	if (ret)
1216 		goto out;
1217 
1218 	ceph_flock_init();
1219 	ret = register_filesystem(&ceph_fs_type);
1220 	if (ret)
1221 		goto out_caches;
1222 
1223 	pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL);
1224 
1225 	return 0;
1226 
1227 out_caches:
1228 	destroy_caches();
1229 out:
1230 	return ret;
1231 }
1232 
1233 static void __exit exit_ceph(void)
1234 {
1235 	dout("exit_ceph\n");
1236 	unregister_filesystem(&ceph_fs_type);
1237 	destroy_caches();
1238 }
1239 
1240 module_init(init_ceph);
1241 module_exit(exit_ceph);
1242 
1243 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
1244 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
1245 MODULE_AUTHOR("Patience Warnick <patience@newdream.net>");
1246 MODULE_DESCRIPTION("Ceph filesystem for Linux");
1247 MODULE_LICENSE("GPL");
1248