]>
Commit | Line | Data |
---|---|---|
16725b9d | 1 | |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
16725b9d SW |
3 | |
4 | #include <linux/backing-dev.h> | |
c309f0ab | 5 | #include <linux/ctype.h> |
16725b9d SW |
6 | #include <linux/fs.h> |
7 | #include <linux/inet.h> | |
8 | #include <linux/in6.h> | |
9 | #include <linux/module.h> | |
10 | #include <linux/mount.h> | |
11 | #include <linux/parser.h> | |
16725b9d SW |
12 | #include <linux/sched.h> |
13 | #include <linux/seq_file.h> | |
5a0e3ad6 | 14 | #include <linux/slab.h> |
16725b9d SW |
15 | #include <linux/statfs.h> |
16 | #include <linux/string.h> | |
16725b9d | 17 | |
16725b9d | 18 | #include "super.h" |
3d14c5d2 | 19 | #include "mds_client.h" |
99ccbd22 | 20 | #include "cache.h" |
3d14c5d2 | 21 | |
1fe60e51 | 22 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
23 | #include <linux/ceph/decode.h> |
24 | #include <linux/ceph/mon_client.h> | |
25 | #include <linux/ceph/auth.h> | |
26 | #include <linux/ceph/debugfs.h> | |
16725b9d SW |
27 | |
28 | /* | |
29 | * Ceph superblock operations | |
30 | * | |
31 | * Handle the basics of mounting, unmounting. | |
32 | */ | |
33 | ||
16725b9d SW |
34 | /* |
35 | * super ops | |
36 | */ | |
37 | static void ceph_put_super(struct super_block *s) | |
38 | { | |
3d14c5d2 | 39 | struct ceph_fs_client *fsc = ceph_sb_to_client(s); |
16725b9d SW |
40 | |
41 | dout("put_super\n"); | |
3d14c5d2 | 42 | ceph_mdsc_close_sessions(fsc->mdsc); |
16725b9d SW |
43 | } |
44 | ||
45 | static int ceph_statfs(struct dentry *dentry, struct kstatfs *buf) | |
46 | { | |
2b0143b5 | 47 | struct ceph_fs_client *fsc = ceph_inode_to_client(d_inode(dentry)); |
3d14c5d2 | 48 | struct ceph_monmap *monmap = fsc->client->monc.monmap; |
16725b9d SW |
49 | struct ceph_statfs st; |
50 | u64 fsid; | |
51 | int err; | |
52 | ||
53 | dout("statfs\n"); | |
3d14c5d2 | 54 | err = ceph_monc_do_statfs(&fsc->client->monc, &st); |
16725b9d SW |
55 | if (err < 0) |
56 | return err; | |
57 | ||
58 | /* fill in kstatfs */ | |
59 | buf->f_type = CEPH_SUPER_MAGIC; /* ?? */ | |
60 | ||
61 | /* | |
62 | * express utilization in terms of large blocks to avoid | |
63 | * overflow on 32-bit machines. | |
92a49fb0 SW |
64 | * |
65 | * NOTE: for the time being, we make bsize == frsize to humor | |
66 | * not-yet-ancient versions of glibc that are broken. | |
67 | * Someday, we will probably want to report a real block | |
68 | * size... whatever that may mean for a network file system! | |
16725b9d SW |
69 | */ |
70 | buf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
92a49fb0 | 71 | buf->f_frsize = 1 << CEPH_BLOCK_SHIFT; |
16725b9d | 72 | buf->f_blocks = le64_to_cpu(st.kb) >> (CEPH_BLOCK_SHIFT-10); |
8f04d422 | 73 | buf->f_bfree = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10); |
16725b9d SW |
74 | buf->f_bavail = le64_to_cpu(st.kb_avail) >> (CEPH_BLOCK_SHIFT-10); |
75 | ||
76 | buf->f_files = le64_to_cpu(st.num_objects); | |
77 | buf->f_ffree = -1; | |
558d3499 | 78 | buf->f_namelen = NAME_MAX; |
16725b9d SW |
79 | |
80 | /* leave fsid little-endian, regardless of host endianness */ | |
81 | fsid = *(u64 *)(&monmap->fsid) ^ *((u64 *)&monmap->fsid + 1); | |
82 | buf->f_fsid.val[0] = fsid & 0xffffffff; | |
83 | buf->f_fsid.val[1] = fsid >> 32; | |
84 | ||
85 | return 0; | |
86 | } | |
87 | ||
88 | ||
2d9c98ae | 89 | static int ceph_sync_fs(struct super_block *sb, int wait) |
16725b9d | 90 | { |
3d14c5d2 | 91 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
2d9c98ae SW |
92 | |
93 | if (!wait) { | |
94 | dout("sync_fs (non-blocking)\n"); | |
3d14c5d2 | 95 | ceph_flush_dirty_caps(fsc->mdsc); |
2d9c98ae SW |
96 | dout("sync_fs (non-blocking) done\n"); |
97 | return 0; | |
98 | } | |
99 | ||
100 | dout("sync_fs (blocking)\n"); | |
3d14c5d2 YS |
101 | ceph_osdc_sync(&fsc->client->osdc); |
102 | ceph_mdsc_sync(fsc->mdsc); | |
2d9c98ae | 103 | dout("sync_fs (blocking) done\n"); |
16725b9d SW |
104 | return 0; |
105 | } | |
106 | ||
16725b9d SW |
107 | /* |
108 | * mount options | |
109 | */ | |
110 | enum { | |
235a0982 | 111 | Opt_mds_namespace, |
16725b9d SW |
112 | Opt_wsize, |
113 | Opt_rsize, | |
83817e35 | 114 | Opt_rasize, |
16725b9d SW |
115 | Opt_caps_wanted_delay_min, |
116 | Opt_caps_wanted_delay_max, | |
6e19a16e | 117 | Opt_cap_release_safety, |
16725b9d | 118 | Opt_readdir_max_entries, |
23804d91 | 119 | Opt_readdir_max_bytes, |
2baba250 | 120 | Opt_congestion_kb, |
e53c2fe0 | 121 | Opt_last_int, |
16725b9d SW |
122 | /* int args above */ |
123 | Opt_snapdirname, | |
e53c2fe0 | 124 | Opt_last_string, |
16725b9d | 125 | /* string args above */ |
16725b9d SW |
126 | Opt_dirstat, |
127 | Opt_nodirstat, | |
128 | Opt_rbytes, | |
129 | Opt_norbytes, | |
cffaba15 | 130 | Opt_asyncreaddir, |
16725b9d | 131 | Opt_noasyncreaddir, |
a40dc6cc SW |
132 | Opt_dcache, |
133 | Opt_nodcache, | |
ad1fee96 | 134 | Opt_ino32, |
cffaba15 | 135 | Opt_noino32, |
99ccbd22 | 136 | Opt_fscache, |
45195e42 | 137 | Opt_nofscache, |
10183a69 YZ |
138 | Opt_poolperm, |
139 | Opt_nopoolperm, | |
45195e42 SW |
140 | #ifdef CONFIG_CEPH_FS_POSIX_ACL |
141 | Opt_acl, | |
142 | #endif | |
10183a69 | 143 | Opt_noacl, |
16725b9d SW |
144 | }; |
145 | ||
3d14c5d2 | 146 | static match_table_t fsopt_tokens = { |
235a0982 | 147 | {Opt_mds_namespace, "mds_namespace=%d"}, |
16725b9d SW |
148 | {Opt_wsize, "wsize=%d"}, |
149 | {Opt_rsize, "rsize=%d"}, | |
83817e35 | 150 | {Opt_rasize, "rasize=%d"}, |
16725b9d SW |
151 | {Opt_caps_wanted_delay_min, "caps_wanted_delay_min=%d"}, |
152 | {Opt_caps_wanted_delay_max, "caps_wanted_delay_max=%d"}, | |
6e19a16e | 153 | {Opt_cap_release_safety, "cap_release_safety=%d"}, |
16725b9d | 154 | {Opt_readdir_max_entries, "readdir_max_entries=%d"}, |
23804d91 | 155 | {Opt_readdir_max_bytes, "readdir_max_bytes=%d"}, |
2baba250 | 156 | {Opt_congestion_kb, "write_congestion_kb=%d"}, |
16725b9d SW |
157 | /* int args above */ |
158 | {Opt_snapdirname, "snapdirname=%s"}, | |
16725b9d | 159 | /* string args above */ |
16725b9d SW |
160 | {Opt_dirstat, "dirstat"}, |
161 | {Opt_nodirstat, "nodirstat"}, | |
162 | {Opt_rbytes, "rbytes"}, | |
163 | {Opt_norbytes, "norbytes"}, | |
cffaba15 | 164 | {Opt_asyncreaddir, "asyncreaddir"}, |
16725b9d | 165 | {Opt_noasyncreaddir, "noasyncreaddir"}, |
a40dc6cc SW |
166 | {Opt_dcache, "dcache"}, |
167 | {Opt_nodcache, "nodcache"}, | |
ad1fee96 | 168 | {Opt_ino32, "ino32"}, |
cffaba15 | 169 | {Opt_noino32, "noino32"}, |
99ccbd22 MT |
170 | {Opt_fscache, "fsc"}, |
171 | {Opt_nofscache, "nofsc"}, | |
10183a69 YZ |
172 | {Opt_poolperm, "poolperm"}, |
173 | {Opt_nopoolperm, "nopoolperm"}, | |
45195e42 SW |
174 | #ifdef CONFIG_CEPH_FS_POSIX_ACL |
175 | {Opt_acl, "acl"}, | |
176 | #endif | |
177 | {Opt_noacl, "noacl"}, | |
16725b9d SW |
178 | {-1, NULL} |
179 | }; | |
180 | ||
3d14c5d2 | 181 | static int parse_fsopt_token(char *c, void *private) |
c309f0ab | 182 | { |
3d14c5d2 YS |
183 | struct ceph_mount_options *fsopt = private; |
184 | substring_t argstr[MAX_OPT_ARGS]; | |
185 | int token, intval, ret; | |
186 | ||
187 | token = match_token((char *)c, fsopt_tokens, argstr); | |
188 | if (token < 0) | |
189 | return -EINVAL; | |
190 | ||
191 | if (token < Opt_last_int) { | |
192 | ret = match_int(&argstr[0], &intval); | |
193 | if (ret < 0) { | |
194 | pr_err("bad mount option arg (not int) " | |
195 | "at '%s'\n", c); | |
196 | return ret; | |
c309f0ab | 197 | } |
3d14c5d2 YS |
198 | dout("got int token %d val %d\n", token, intval); |
199 | } else if (token > Opt_last_int && token < Opt_last_string) { | |
200 | dout("got string token %d val %s\n", token, | |
201 | argstr[0].from); | |
202 | } else { | |
203 | dout("got token %d\n", token); | |
c309f0ab SW |
204 | } |
205 | ||
3d14c5d2 YS |
206 | switch (token) { |
207 | case Opt_snapdirname: | |
208 | kfree(fsopt->snapdir_name); | |
209 | fsopt->snapdir_name = kstrndup(argstr[0].from, | |
210 | argstr[0].to-argstr[0].from, | |
211 | GFP_KERNEL); | |
212 | if (!fsopt->snapdir_name) | |
213 | return -ENOMEM; | |
214 | break; | |
215 | ||
216 | /* misc */ | |
235a0982 YZ |
217 | case Opt_mds_namespace: |
218 | fsopt->mds_namespace = intval; | |
219 | break; | |
3d14c5d2 YS |
220 | case Opt_wsize: |
221 | fsopt->wsize = intval; | |
222 | break; | |
223 | case Opt_rsize: | |
224 | fsopt->rsize = intval; | |
225 | break; | |
83817e35 SW |
226 | case Opt_rasize: |
227 | fsopt->rasize = intval; | |
228 | break; | |
3d14c5d2 YS |
229 | case Opt_caps_wanted_delay_min: |
230 | fsopt->caps_wanted_delay_min = intval; | |
231 | break; | |
232 | case Opt_caps_wanted_delay_max: | |
233 | fsopt->caps_wanted_delay_max = intval; | |
234 | break; | |
235 | case Opt_readdir_max_entries: | |
236 | fsopt->max_readdir = intval; | |
237 | break; | |
238 | case Opt_readdir_max_bytes: | |
239 | fsopt->max_readdir_bytes = intval; | |
240 | break; | |
241 | case Opt_congestion_kb: | |
242 | fsopt->congestion_kb = intval; | |
243 | break; | |
244 | case Opt_dirstat: | |
245 | fsopt->flags |= CEPH_MOUNT_OPT_DIRSTAT; | |
246 | break; | |
247 | case Opt_nodirstat: | |
248 | fsopt->flags &= ~CEPH_MOUNT_OPT_DIRSTAT; | |
249 | break; | |
250 | case Opt_rbytes: | |
251 | fsopt->flags |= CEPH_MOUNT_OPT_RBYTES; | |
252 | break; | |
253 | case Opt_norbytes: | |
254 | fsopt->flags &= ~CEPH_MOUNT_OPT_RBYTES; | |
255 | break; | |
cffaba15 AE |
256 | case Opt_asyncreaddir: |
257 | fsopt->flags &= ~CEPH_MOUNT_OPT_NOASYNCREADDIR; | |
258 | break; | |
3d14c5d2 YS |
259 | case Opt_noasyncreaddir: |
260 | fsopt->flags |= CEPH_MOUNT_OPT_NOASYNCREADDIR; | |
261 | break; | |
a40dc6cc SW |
262 | case Opt_dcache: |
263 | fsopt->flags |= CEPH_MOUNT_OPT_DCACHE; | |
264 | break; | |
265 | case Opt_nodcache: | |
266 | fsopt->flags &= ~CEPH_MOUNT_OPT_DCACHE; | |
267 | break; | |
ad1fee96 YS |
268 | case Opt_ino32: |
269 | fsopt->flags |= CEPH_MOUNT_OPT_INO32; | |
270 | break; | |
cffaba15 AE |
271 | case Opt_noino32: |
272 | fsopt->flags &= ~CEPH_MOUNT_OPT_INO32; | |
273 | break; | |
99ccbd22 MT |
274 | case Opt_fscache: |
275 | fsopt->flags |= CEPH_MOUNT_OPT_FSCACHE; | |
276 | break; | |
277 | case Opt_nofscache: | |
278 | fsopt->flags &= ~CEPH_MOUNT_OPT_FSCACHE; | |
279 | break; | |
10183a69 YZ |
280 | case Opt_poolperm: |
281 | fsopt->flags &= ~CEPH_MOUNT_OPT_NOPOOLPERM; | |
282 | printk ("pool perm"); | |
283 | break; | |
284 | case Opt_nopoolperm: | |
285 | fsopt->flags |= CEPH_MOUNT_OPT_NOPOOLPERM; | |
286 | break; | |
45195e42 SW |
287 | #ifdef CONFIG_CEPH_FS_POSIX_ACL |
288 | case Opt_acl: | |
289 | fsopt->sb_flags |= MS_POSIXACL; | |
290 | break; | |
291 | #endif | |
292 | case Opt_noacl: | |
293 | fsopt->sb_flags &= ~MS_POSIXACL; | |
294 | break; | |
3d14c5d2 YS |
295 | default: |
296 | BUG_ON(token); | |
297 | } | |
298 | return 0; | |
c309f0ab | 299 | } |
16725b9d | 300 | |
3d14c5d2 | 301 | static void destroy_mount_options(struct ceph_mount_options *args) |
16725b9d | 302 | { |
3d14c5d2 YS |
303 | dout("destroy_mount_options %p\n", args); |
304 | kfree(args->snapdir_name); | |
305 | kfree(args); | |
306 | } | |
16725b9d | 307 | |
3d14c5d2 YS |
308 | static int strcmp_null(const char *s1, const char *s2) |
309 | { | |
310 | if (!s1 && !s2) | |
311 | return 0; | |
312 | if (s1 && !s2) | |
313 | return -1; | |
314 | if (!s1 && s2) | |
315 | return 1; | |
316 | return strcmp(s1, s2); | |
317 | } | |
16725b9d | 318 | |
3d14c5d2 YS |
319 | static int compare_mount_options(struct ceph_mount_options *new_fsopt, |
320 | struct ceph_options *new_opt, | |
321 | struct ceph_fs_client *fsc) | |
322 | { | |
323 | struct ceph_mount_options *fsopt1 = new_fsopt; | |
324 | struct ceph_mount_options *fsopt2 = fsc->mount_options; | |
325 | int ofs = offsetof(struct ceph_mount_options, snapdir_name); | |
326 | int ret; | |
16725b9d | 327 | |
3d14c5d2 YS |
328 | ret = memcmp(fsopt1, fsopt2, ofs); |
329 | if (ret) | |
330 | return ret; | |
331 | ||
332 | ret = strcmp_null(fsopt1->snapdir_name, fsopt2->snapdir_name); | |
333 | if (ret) | |
334 | return ret; | |
335 | ||
336 | return ceph_compare_options(new_opt, fsc->client); | |
337 | } | |
338 | ||
339 | static int parse_mount_options(struct ceph_mount_options **pfsopt, | |
340 | struct ceph_options **popt, | |
341 | int flags, char *options, | |
342 | const char *dev_name, | |
343 | const char **path) | |
344 | { | |
345 | struct ceph_mount_options *fsopt; | |
346 | const char *dev_name_end; | |
c98f533c AE |
347 | int err; |
348 | ||
349 | if (!dev_name || !*dev_name) | |
350 | return -EINVAL; | |
3d14c5d2 YS |
351 | |
352 | fsopt = kzalloc(sizeof(*fsopt), GFP_KERNEL); | |
353 | if (!fsopt) | |
354 | return -ENOMEM; | |
355 | ||
356 | dout("parse_mount_options %p, dev_name '%s'\n", fsopt, dev_name); | |
357 | ||
80db8bea NW |
358 | fsopt->sb_flags = flags; |
359 | fsopt->flags = CEPH_MOUNT_OPT_DEFAULT; | |
3d14c5d2 | 360 | |
80db8bea NW |
361 | fsopt->rsize = CEPH_RSIZE_DEFAULT; |
362 | fsopt->rasize = CEPH_RASIZE_DEFAULT; | |
363 | fsopt->snapdir_name = kstrdup(CEPH_SNAPDIRNAME_DEFAULT, GFP_KERNEL); | |
a149bb9a SK |
364 | if (!fsopt->snapdir_name) { |
365 | err = -ENOMEM; | |
366 | goto out; | |
367 | } | |
368 | ||
50aac4fe SW |
369 | fsopt->caps_wanted_delay_min = CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT; |
370 | fsopt->caps_wanted_delay_max = CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT; | |
80db8bea NW |
371 | fsopt->cap_release_safety = CEPH_CAP_RELEASE_SAFETY_DEFAULT; |
372 | fsopt->max_readdir = CEPH_MAX_READDIR_DEFAULT; | |
373 | fsopt->max_readdir_bytes = CEPH_MAX_READDIR_BYTES_DEFAULT; | |
374 | fsopt->congestion_kb = default_congestion_kb(); | |
235a0982 | 375 | fsopt->mds_namespace = CEPH_FS_CLUSTER_ID_NONE; |
80db8bea | 376 | |
c98f533c AE |
377 | /* |
378 | * Distinguish the server list from the path in "dev_name". | |
379 | * Internally we do not include the leading '/' in the path. | |
380 | * | |
381 | * "dev_name" will look like: | |
382 | * <server_spec>[,<server_spec>...]:[<path>] | |
383 | * where | |
384 | * <server_spec> is <ip>[:<port>] | |
385 | * <path> is optional, but if present must begin with '/' | |
386 | */ | |
387 | dev_name_end = strchr(dev_name, '/'); | |
388 | if (dev_name_end) { | |
389 | /* skip over leading '/' for path */ | |
390 | *path = dev_name_end + 1; | |
391 | } else { | |
392 | /* path is empty */ | |
393 | dev_name_end = dev_name + strlen(dev_name); | |
394 | *path = dev_name_end; | |
395 | } | |
80db8bea | 396 | err = -EINVAL; |
c98f533c | 397 | dev_name_end--; /* back up to ':' separator */ |
54464296 | 398 | if (dev_name_end < dev_name || *dev_name_end != ':') { |
c98f533c | 399 | pr_err("device name is missing path (no : separator in %s)\n", |
80db8bea NW |
400 | dev_name); |
401 | goto out; | |
402 | } | |
3d14c5d2 | 403 | dout("device name '%.*s'\n", (int)(dev_name_end - dev_name), dev_name); |
16725b9d SW |
404 | dout("server path '%s'\n", *path); |
405 | ||
ee57741c | 406 | *popt = ceph_parse_options(options, dev_name, dev_name_end, |
3d14c5d2 | 407 | parse_fsopt_token, (void *)fsopt); |
ee57741c AE |
408 | if (IS_ERR(*popt)) { |
409 | err = PTR_ERR(*popt); | |
3d14c5d2 | 410 | goto out; |
ee57741c | 411 | } |
3d14c5d2 YS |
412 | |
413 | /* success */ | |
414 | *pfsopt = fsopt; | |
415 | return 0; | |
16725b9d | 416 | |
7b813c46 | 417 | out: |
3d14c5d2 YS |
418 | destroy_mount_options(fsopt); |
419 | return err; | |
16725b9d SW |
420 | } |
421 | ||
3d14c5d2 YS |
422 | /** |
423 | * ceph_show_options - Show mount options in /proc/mounts | |
424 | * @m: seq_file to write to | |
34c80b1d | 425 | * @root: root of that (sub)tree |
3d14c5d2 | 426 | */ |
34c80b1d | 427 | static int ceph_show_options(struct seq_file *m, struct dentry *root) |
16725b9d | 428 | { |
34c80b1d | 429 | struct ceph_fs_client *fsc = ceph_sb_to_client(root->d_sb); |
3d14c5d2 | 430 | struct ceph_mount_options *fsopt = fsc->mount_options; |
ff40f9ae ID |
431 | size_t pos; |
432 | int ret; | |
433 | ||
434 | /* a comma between MNT/MS and client options */ | |
435 | seq_putc(m, ','); | |
436 | pos = m->count; | |
437 | ||
438 | ret = ceph_print_client_options(m, fsc->client); | |
439 | if (ret) | |
440 | return ret; | |
441 | ||
442 | /* retract our comma if no client options */ | |
443 | if (m->count == pos) | |
444 | m->count--; | |
3d14c5d2 YS |
445 | |
446 | if (fsopt->flags & CEPH_MOUNT_OPT_DIRSTAT) | |
447 | seq_puts(m, ",dirstat"); | |
133e9156 YZ |
448 | if ((fsopt->flags & CEPH_MOUNT_OPT_RBYTES)) |
449 | seq_puts(m, ",rbytes"); | |
3d14c5d2 YS |
450 | if (fsopt->flags & CEPH_MOUNT_OPT_NOASYNCREADDIR) |
451 | seq_puts(m, ",noasyncreaddir"); | |
ff7eeb82 | 452 | if ((fsopt->flags & CEPH_MOUNT_OPT_DCACHE) == 0) |
a40dc6cc | 453 | seq_puts(m, ",nodcache"); |
99ccbd22 MT |
454 | if (fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) |
455 | seq_puts(m, ",fsc"); | |
10183a69 YZ |
456 | if (fsopt->flags & CEPH_MOUNT_OPT_NOPOOLPERM) |
457 | seq_puts(m, ",nopoolperm"); | |
3d14c5d2 | 458 | |
45195e42 SW |
459 | #ifdef CONFIG_CEPH_FS_POSIX_ACL |
460 | if (fsopt->sb_flags & MS_POSIXACL) | |
461 | seq_puts(m, ",acl"); | |
462 | else | |
463 | seq_puts(m, ",noacl"); | |
464 | #endif | |
465 | ||
235a0982 YZ |
466 | if (fsopt->mds_namespace != CEPH_FS_CLUSTER_ID_NONE) |
467 | seq_printf(m, ",mds_namespace=%d", fsopt->mds_namespace); | |
3d14c5d2 YS |
468 | if (fsopt->wsize) |
469 | seq_printf(m, ",wsize=%d", fsopt->wsize); | |
80456f86 | 470 | if (fsopt->rsize != CEPH_RSIZE_DEFAULT) |
3d14c5d2 | 471 | seq_printf(m, ",rsize=%d", fsopt->rsize); |
83817e35 | 472 | if (fsopt->rasize != CEPH_RASIZE_DEFAULT) |
2151937d | 473 | seq_printf(m, ",rasize=%d", fsopt->rasize); |
3d14c5d2 YS |
474 | if (fsopt->congestion_kb != default_congestion_kb()) |
475 | seq_printf(m, ",write_congestion_kb=%d", fsopt->congestion_kb); | |
476 | if (fsopt->caps_wanted_delay_min != CEPH_CAPS_WANTED_DELAY_MIN_DEFAULT) | |
477 | seq_printf(m, ",caps_wanted_delay_min=%d", | |
478 | fsopt->caps_wanted_delay_min); | |
479 | if (fsopt->caps_wanted_delay_max != CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT) | |
480 | seq_printf(m, ",caps_wanted_delay_max=%d", | |
481 | fsopt->caps_wanted_delay_max); | |
482 | if (fsopt->cap_release_safety != CEPH_CAP_RELEASE_SAFETY_DEFAULT) | |
483 | seq_printf(m, ",cap_release_safety=%d", | |
484 | fsopt->cap_release_safety); | |
485 | if (fsopt->max_readdir != CEPH_MAX_READDIR_DEFAULT) | |
486 | seq_printf(m, ",readdir_max_entries=%d", fsopt->max_readdir); | |
487 | if (fsopt->max_readdir_bytes != CEPH_MAX_READDIR_BYTES_DEFAULT) | |
488 | seq_printf(m, ",readdir_max_bytes=%d", fsopt->max_readdir_bytes); | |
489 | if (strcmp(fsopt->snapdir_name, CEPH_SNAPDIRNAME_DEFAULT)) | |
a068acf2 | 490 | seq_show_option(m, "snapdirname", fsopt->snapdir_name); |
ff40f9ae | 491 | |
3d14c5d2 | 492 | return 0; |
16725b9d SW |
493 | } |
494 | ||
495 | /* | |
3d14c5d2 YS |
496 | * handle any mon messages the standard library doesn't understand. |
497 | * return error if we don't either. | |
16725b9d | 498 | */ |
3d14c5d2 | 499 | static int extra_mon_dispatch(struct ceph_client *client, struct ceph_msg *msg) |
16725b9d | 500 | { |
3d14c5d2 YS |
501 | struct ceph_fs_client *fsc = client->private; |
502 | int type = le16_to_cpu(msg->hdr.type); | |
503 | ||
504 | switch (type) { | |
505 | case CEPH_MSG_MDS_MAP: | |
506 | ceph_mdsc_handle_map(fsc->mdsc, msg); | |
507 | return 0; | |
508 | ||
509 | default: | |
510 | return -1; | |
511 | } | |
512 | } | |
513 | ||
514 | /* | |
515 | * create a new fs client | |
516 | */ | |
0c6d4b4e | 517 | static struct ceph_fs_client *create_fs_client(struct ceph_mount_options *fsopt, |
3d14c5d2 YS |
518 | struct ceph_options *opt) |
519 | { | |
520 | struct ceph_fs_client *fsc; | |
12b4629a | 521 | const u64 supported_features = |
d463a43d YZ |
522 | CEPH_FEATURE_FLOCK | CEPH_FEATURE_DIRLAYOUTHASH | |
523 | CEPH_FEATURE_MDSENC | CEPH_FEATURE_MDS_INLINE_DATA; | |
12b4629a | 524 | const u64 required_features = 0; |
3bf53337 AE |
525 | int page_count; |
526 | size_t size; | |
16725b9d SW |
527 | int err = -ENOMEM; |
528 | ||
3d14c5d2 YS |
529 | fsc = kzalloc(sizeof(*fsc), GFP_KERNEL); |
530 | if (!fsc) | |
16725b9d SW |
531 | return ERR_PTR(-ENOMEM); |
532 | ||
6ab00d46 SW |
533 | fsc->client = ceph_create_client(opt, fsc, supported_features, |
534 | required_features); | |
3d14c5d2 YS |
535 | if (IS_ERR(fsc->client)) { |
536 | err = PTR_ERR(fsc->client); | |
537 | goto fail; | |
538 | } | |
539 | fsc->client->extra_mon_dispatch = extra_mon_dispatch; | |
235a0982 | 540 | fsc->client->monc.fs_cluster_id = fsopt->mds_namespace; |
82dcabad | 541 | ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP, 0, true); |
16725b9d | 542 | |
3d14c5d2 | 543 | fsc->mount_options = fsopt; |
16725b9d | 544 | |
3d14c5d2 YS |
545 | fsc->sb = NULL; |
546 | fsc->mount_state = CEPH_MOUNT_MOUNTING; | |
16725b9d | 547 | |
3d14c5d2 | 548 | atomic_long_set(&fsc->writeback_count, 0); |
16725b9d | 549 | |
3d14c5d2 | 550 | err = bdi_init(&fsc->backing_dev_info); |
859e7b14 | 551 | if (err < 0) |
3d14c5d2 | 552 | goto fail_client; |
859e7b14 | 553 | |
16725b9d | 554 | err = -ENOMEM; |
01e6acc4 TH |
555 | /* |
556 | * The number of concurrent works can be high but they don't need | |
557 | * to be processed in parallel, limit concurrency. | |
558 | */ | |
559 | fsc->wb_wq = alloc_workqueue("ceph-writeback", 0, 1); | |
3d14c5d2 | 560 | if (fsc->wb_wq == NULL) |
859e7b14 | 561 | goto fail_bdi; |
01e6acc4 | 562 | fsc->pg_inv_wq = alloc_workqueue("ceph-pg-invalid", 0, 1); |
3d14c5d2 | 563 | if (fsc->pg_inv_wq == NULL) |
16725b9d | 564 | goto fail_wb_wq; |
01e6acc4 | 565 | fsc->trunc_wq = alloc_workqueue("ceph-trunc", 0, 1); |
3d14c5d2 | 566 | if (fsc->trunc_wq == NULL) |
16725b9d SW |
567 | goto fail_pg_inv_wq; |
568 | ||
b9bfb93c SW |
569 | /* set up mempools */ |
570 | err = -ENOMEM; | |
09cbfeaf | 571 | page_count = fsc->mount_options->wsize >> PAGE_SHIFT; |
3bf53337 AE |
572 | size = sizeof (struct page *) * (page_count ? page_count : 1); |
573 | fsc->wb_pagevec_pool = mempool_create_kmalloc_pool(10, size); | |
3d14c5d2 | 574 | if (!fsc->wb_pagevec_pool) |
b9bfb93c SW |
575 | goto fail_trunc_wq; |
576 | ||
99ccbd22 MT |
577 | /* setup fscache */ |
578 | if ((fsopt->flags & CEPH_MOUNT_OPT_FSCACHE) && | |
579 | (ceph_fscache_register_fs(fsc) != 0)) | |
580 | goto fail_fscache; | |
581 | ||
85ccce43 | 582 | /* caps */ |
3d14c5d2 YS |
583 | fsc->min_caps = fsopt->max_readdir; |
584 | ||
585 | return fsc; | |
b9bfb93c | 586 | |
99ccbd22 MT |
587 | fail_fscache: |
588 | ceph_fscache_unregister_fs(fsc); | |
16725b9d | 589 | fail_trunc_wq: |
3d14c5d2 | 590 | destroy_workqueue(fsc->trunc_wq); |
16725b9d | 591 | fail_pg_inv_wq: |
3d14c5d2 | 592 | destroy_workqueue(fsc->pg_inv_wq); |
16725b9d | 593 | fail_wb_wq: |
3d14c5d2 | 594 | destroy_workqueue(fsc->wb_wq); |
859e7b14 | 595 | fail_bdi: |
3d14c5d2 YS |
596 | bdi_destroy(&fsc->backing_dev_info); |
597 | fail_client: | |
598 | ceph_destroy_client(fsc->client); | |
16725b9d | 599 | fail: |
3d14c5d2 | 600 | kfree(fsc); |
16725b9d SW |
601 | return ERR_PTR(err); |
602 | } | |
603 | ||
0c6d4b4e | 604 | static void destroy_fs_client(struct ceph_fs_client *fsc) |
16725b9d | 605 | { |
3d14c5d2 | 606 | dout("destroy_fs_client %p\n", fsc); |
16725b9d | 607 | |
99ccbd22 MT |
608 | ceph_fscache_unregister_fs(fsc); |
609 | ||
3d14c5d2 YS |
610 | destroy_workqueue(fsc->wb_wq); |
611 | destroy_workqueue(fsc->pg_inv_wq); | |
612 | destroy_workqueue(fsc->trunc_wq); | |
16725b9d | 613 | |
3d14c5d2 | 614 | bdi_destroy(&fsc->backing_dev_info); |
a922d38f | 615 | |
3d14c5d2 | 616 | mempool_destroy(fsc->wb_pagevec_pool); |
16725b9d | 617 | |
3d14c5d2 | 618 | destroy_mount_options(fsc->mount_options); |
5dfc589a | 619 | |
3d14c5d2 | 620 | ceph_fs_debugfs_cleanup(fsc); |
16725b9d | 621 | |
3d14c5d2 | 622 | ceph_destroy_client(fsc->client); |
16725b9d | 623 | |
3d14c5d2 YS |
624 | kfree(fsc); |
625 | dout("destroy_fs_client %p done\n", fsc); | |
16725b9d SW |
626 | } |
627 | ||
0743304d | 628 | /* |
3d14c5d2 | 629 | * caches |
0743304d | 630 | */ |
3d14c5d2 YS |
631 | struct kmem_cache *ceph_inode_cachep; |
632 | struct kmem_cache *ceph_cap_cachep; | |
f66fd9f0 | 633 | struct kmem_cache *ceph_cap_flush_cachep; |
3d14c5d2 YS |
634 | struct kmem_cache *ceph_dentry_cachep; |
635 | struct kmem_cache *ceph_file_cachep; | |
636 | ||
637 | static void ceph_inode_init_once(void *foo) | |
0743304d | 638 | { |
3d14c5d2 YS |
639 | struct ceph_inode_info *ci = foo; |
640 | inode_init_once(&ci->vfs_inode); | |
641 | } | |
642 | ||
643 | static int __init init_caches(void) | |
644 | { | |
99ccbd22 MT |
645 | int error = -ENOMEM; |
646 | ||
3d14c5d2 YS |
647 | ceph_inode_cachep = kmem_cache_create("ceph_inode_info", |
648 | sizeof(struct ceph_inode_info), | |
649 | __alignof__(struct ceph_inode_info), | |
5d097056 VD |
650 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD| |
651 | SLAB_ACCOUNT, ceph_inode_init_once); | |
3d14c5d2 YS |
652 | if (ceph_inode_cachep == NULL) |
653 | return -ENOMEM; | |
654 | ||
655 | ceph_cap_cachep = KMEM_CACHE(ceph_cap, | |
656 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD); | |
657 | if (ceph_cap_cachep == NULL) | |
658 | goto bad_cap; | |
f66fd9f0 YZ |
659 | ceph_cap_flush_cachep = KMEM_CACHE(ceph_cap_flush, |
660 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD); | |
661 | if (ceph_cap_flush_cachep == NULL) | |
662 | goto bad_cap_flush; | |
3d14c5d2 YS |
663 | |
664 | ceph_dentry_cachep = KMEM_CACHE(ceph_dentry_info, | |
665 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD); | |
666 | if (ceph_dentry_cachep == NULL) | |
667 | goto bad_dentry; | |
668 | ||
669 | ceph_file_cachep = KMEM_CACHE(ceph_file_info, | |
670 | SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD); | |
671 | if (ceph_file_cachep == NULL) | |
672 | goto bad_file; | |
673 | ||
99ccbd22 MT |
674 | if ((error = ceph_fscache_register())) |
675 | goto bad_file; | |
3d14c5d2 | 676 | |
99ccbd22 | 677 | return 0; |
3d14c5d2 YS |
678 | bad_file: |
679 | kmem_cache_destroy(ceph_dentry_cachep); | |
680 | bad_dentry: | |
f66fd9f0 YZ |
681 | kmem_cache_destroy(ceph_cap_flush_cachep); |
682 | bad_cap_flush: | |
3d14c5d2 YS |
683 | kmem_cache_destroy(ceph_cap_cachep); |
684 | bad_cap: | |
685 | kmem_cache_destroy(ceph_inode_cachep); | |
99ccbd22 | 686 | return error; |
0743304d SW |
687 | } |
688 | ||
3d14c5d2 YS |
689 | static void destroy_caches(void) |
690 | { | |
8c0a8537 KS |
691 | /* |
692 | * Make sure all delayed rcu free inodes are flushed before we | |
693 | * destroy cache. | |
694 | */ | |
695 | rcu_barrier(); | |
99ccbd22 | 696 | |
3d14c5d2 YS |
697 | kmem_cache_destroy(ceph_inode_cachep); |
698 | kmem_cache_destroy(ceph_cap_cachep); | |
f66fd9f0 | 699 | kmem_cache_destroy(ceph_cap_flush_cachep); |
3d14c5d2 YS |
700 | kmem_cache_destroy(ceph_dentry_cachep); |
701 | kmem_cache_destroy(ceph_file_cachep); | |
99ccbd22 MT |
702 | |
703 | ceph_fscache_unregister(); | |
3d14c5d2 YS |
704 | } |
705 | ||
706 | ||
16725b9d | 707 | /* |
3d14c5d2 YS |
708 | * ceph_umount_begin - initiate forced umount. Tear down down the |
709 | * mount, skipping steps that may hang while waiting for server(s). | |
16725b9d | 710 | */ |
3d14c5d2 | 711 | static void ceph_umount_begin(struct super_block *sb) |
16725b9d | 712 | { |
3d14c5d2 YS |
713 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
714 | ||
715 | dout("ceph_umount_begin - starting forced umount\n"); | |
716 | if (!fsc) | |
717 | return; | |
718 | fsc->mount_state = CEPH_MOUNT_SHUTDOWN; | |
48fec5d0 | 719 | ceph_mdsc_force_umount(fsc->mdsc); |
3d14c5d2 | 720 | return; |
16725b9d SW |
721 | } |
722 | ||
3d14c5d2 YS |
723 | static const struct super_operations ceph_super_ops = { |
724 | .alloc_inode = ceph_alloc_inode, | |
725 | .destroy_inode = ceph_destroy_inode, | |
726 | .write_inode = ceph_write_inode, | |
9f12bd11 | 727 | .drop_inode = ceph_drop_inode, |
3d14c5d2 YS |
728 | .sync_fs = ceph_sync_fs, |
729 | .put_super = ceph_put_super, | |
730 | .show_options = ceph_show_options, | |
731 | .statfs = ceph_statfs, | |
732 | .umount_begin = ceph_umount_begin, | |
733 | }; | |
734 | ||
16725b9d SW |
735 | /* |
736 | * Bootstrap mount by opening the root directory. Note the mount | |
737 | * @started time from caller, and time out if this takes too long. | |
738 | */ | |
3d14c5d2 | 739 | static struct dentry *open_root_dentry(struct ceph_fs_client *fsc, |
16725b9d SW |
740 | const char *path, |
741 | unsigned long started) | |
742 | { | |
3d14c5d2 | 743 | struct ceph_mds_client *mdsc = fsc->mdsc; |
16725b9d SW |
744 | struct ceph_mds_request *req = NULL; |
745 | int err; | |
746 | struct dentry *root; | |
747 | ||
748 | /* open dir */ | |
749 | dout("open_root_inode opening '%s'\n", path); | |
750 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); | |
751 | if (IS_ERR(req)) | |
7e34bc52 | 752 | return ERR_CAST(req); |
16725b9d | 753 | req->r_path1 = kstrdup(path, GFP_NOFS); |
a149bb9a SK |
754 | if (!req->r_path1) { |
755 | root = ERR_PTR(-ENOMEM); | |
756 | goto out; | |
757 | } | |
758 | ||
16725b9d SW |
759 | req->r_ino1.ino = CEPH_INO_ROOT; |
760 | req->r_ino1.snap = CEPH_NOSNAP; | |
761 | req->r_started = started; | |
a319bf56 | 762 | req->r_timeout = fsc->client->options->mount_timeout; |
16725b9d SW |
763 | req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); |
764 | req->r_num_caps = 2; | |
765 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
766 | if (err == 0) { | |
3c5184ef AV |
767 | struct inode *inode = req->r_target_inode; |
768 | req->r_target_inode = NULL; | |
16725b9d | 769 | dout("open_root_inode success\n"); |
3c5184ef | 770 | if (ceph_ino(inode) == CEPH_INO_ROOT && |
774ac21d | 771 | fsc->sb->s_root == NULL) { |
48fde701 | 772 | root = d_make_root(inode); |
3c5184ef | 773 | if (!root) { |
3c5184ef AV |
774 | root = ERR_PTR(-ENOMEM); |
775 | goto out; | |
776 | } | |
774ac21d | 777 | } else { |
1a0a397e | 778 | root = d_obtain_root(inode); |
774ac21d | 779 | } |
d46cfba5 | 780 | ceph_init_dentry(root); |
16725b9d SW |
781 | dout("open_root_inode success, root dentry is %p\n", root); |
782 | } else { | |
783 | root = ERR_PTR(err); | |
784 | } | |
3c5184ef | 785 | out: |
16725b9d SW |
786 | ceph_mdsc_put_request(req); |
787 | return root; | |
788 | } | |
789 | ||
3d14c5d2 YS |
790 | |
791 | ||
792 | ||
16725b9d SW |
793 | /* |
794 | * mount: join the ceph cluster, and open root directory. | |
795 | */ | |
a7f9fb20 | 796 | static struct dentry *ceph_real_mount(struct ceph_fs_client *fsc, |
16725b9d SW |
797 | const char *path) |
798 | { | |
16725b9d | 799 | int err; |
16725b9d SW |
800 | unsigned long started = jiffies; /* note the start time */ |
801 | struct dentry *root; | |
3d14c5d2 | 802 | int first = 0; /* first vfsmount for this super_block */ |
16725b9d | 803 | |
132ca7e1 | 804 | dout("mount start %p\n", fsc); |
3d14c5d2 | 805 | mutex_lock(&fsc->client->mount_mutex); |
16725b9d | 806 | |
132ca7e1 YZ |
807 | if (!fsc->sb->s_root) { |
808 | err = __ceph_open_session(fsc->client, started); | |
809 | if (err < 0) | |
810 | goto out; | |
16725b9d | 811 | |
132ca7e1 YZ |
812 | dout("mount opening root\n"); |
813 | root = open_root_dentry(fsc, "", started); | |
814 | if (IS_ERR(root)) { | |
815 | err = PTR_ERR(root); | |
816 | goto out; | |
817 | } | |
3d14c5d2 YS |
818 | fsc->sb->s_root = root; |
819 | first = 1; | |
820 | ||
821 | err = ceph_fs_debugfs_init(fsc); | |
822 | if (err < 0) | |
823 | goto fail; | |
824 | } | |
16725b9d SW |
825 | |
826 | if (path[0] == 0) { | |
132ca7e1 | 827 | root = fsc->sb->s_root; |
16725b9d SW |
828 | dget(root); |
829 | } else { | |
830 | dout("mount opening base mountpoint\n"); | |
3d14c5d2 | 831 | root = open_root_dentry(fsc, path, started); |
16725b9d SW |
832 | if (IS_ERR(root)) { |
833 | err = PTR_ERR(root); | |
3d14c5d2 | 834 | goto fail; |
16725b9d SW |
835 | } |
836 | } | |
837 | ||
3d14c5d2 | 838 | fsc->mount_state = CEPH_MOUNT_MOUNTED; |
16725b9d | 839 | dout("mount success\n"); |
a7f9fb20 AV |
840 | mutex_unlock(&fsc->client->mount_mutex); |
841 | return root; | |
16725b9d | 842 | |
3d14c5d2 YS |
843 | fail: |
844 | if (first) { | |
845 | dput(fsc->sb->s_root); | |
846 | fsc->sb->s_root = NULL; | |
847 | } | |
132ca7e1 YZ |
848 | out: |
849 | mutex_unlock(&fsc->client->mount_mutex); | |
850 | return ERR_PTR(err); | |
16725b9d SW |
851 | } |
852 | ||
853 | static int ceph_set_super(struct super_block *s, void *data) | |
854 | { | |
3d14c5d2 | 855 | struct ceph_fs_client *fsc = data; |
16725b9d SW |
856 | int ret; |
857 | ||
858 | dout("set_super %p data %p\n", s, data); | |
859 | ||
3d14c5d2 | 860 | s->s_flags = fsc->mount_options->sb_flags; |
16725b9d SW |
861 | s->s_maxbytes = 1ULL << 40; /* temp value until we get mdsmap */ |
862 | ||
7221fe4c | 863 | s->s_xattr = ceph_xattr_handlers; |
3d14c5d2 YS |
864 | s->s_fs_info = fsc; |
865 | fsc->sb = s; | |
16725b9d SW |
866 | |
867 | s->s_op = &ceph_super_ops; | |
868 | s->s_export_op = &ceph_export_ops; | |
869 | ||
870 | s->s_time_gran = 1000; /* 1000 ns == 1 us */ | |
871 | ||
872 | ret = set_anon_super(s, NULL); /* what is that second arg for? */ | |
873 | if (ret != 0) | |
874 | goto fail; | |
875 | ||
876 | return ret; | |
877 | ||
878 | fail: | |
879 | s->s_fs_info = NULL; | |
3d14c5d2 | 880 | fsc->sb = NULL; |
16725b9d SW |
881 | return ret; |
882 | } | |
883 | ||
884 | /* | |
885 | * share superblock if same fs AND options | |
886 | */ | |
887 | static int ceph_compare_super(struct super_block *sb, void *data) | |
888 | { | |
3d14c5d2 YS |
889 | struct ceph_fs_client *new = data; |
890 | struct ceph_mount_options *fsopt = new->mount_options; | |
891 | struct ceph_options *opt = new->client->options; | |
892 | struct ceph_fs_client *other = ceph_sb_to_client(sb); | |
16725b9d SW |
893 | |
894 | dout("ceph_compare_super %p\n", sb); | |
3d14c5d2 YS |
895 | |
896 | if (compare_mount_options(fsopt, opt, other)) { | |
897 | dout("monitor(s)/mount options don't match\n"); | |
898 | return 0; | |
16725b9d | 899 | } |
3d14c5d2 YS |
900 | if ((opt->flags & CEPH_OPT_FSID) && |
901 | ceph_fsid_compare(&opt->fsid, &other->client->fsid)) { | |
902 | dout("fsid doesn't match\n"); | |
903 | return 0; | |
904 | } | |
905 | if (fsopt->sb_flags != other->mount_options->sb_flags) { | |
16725b9d SW |
906 | dout("flags differ\n"); |
907 | return 0; | |
908 | } | |
909 | return 1; | |
910 | } | |
911 | ||
912 | /* | |
913 | * construct our own bdi so we can control readahead, etc. | |
914 | */ | |
00d5643e | 915 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
31e0cf8f | 916 | |
3d14c5d2 YS |
917 | static int ceph_register_bdi(struct super_block *sb, |
918 | struct ceph_fs_client *fsc) | |
16725b9d SW |
919 | { |
920 | int err; | |
921 | ||
83817e35 | 922 | /* set ra_pages based on rasize mount option? */ |
09cbfeaf | 923 | if (fsc->mount_options->rasize >= PAGE_SIZE) |
3d14c5d2 | 924 | fsc->backing_dev_info.ra_pages = |
09cbfeaf | 925 | (fsc->mount_options->rasize + PAGE_SIZE - 1) |
16725b9d | 926 | >> PAGE_SHIFT; |
e9852227 YS |
927 | else |
928 | fsc->backing_dev_info.ra_pages = | |
09cbfeaf | 929 | VM_MAX_READAHEAD * 1024 / PAGE_SIZE; |
e9852227 | 930 | |
d2cc4dde | 931 | err = bdi_register(&fsc->backing_dev_info, NULL, "ceph-%ld", |
31e0cf8f | 932 | atomic_long_inc_return(&bdi_seq)); |
5dfc589a | 933 | if (!err) |
3d14c5d2 | 934 | sb->s_bdi = &fsc->backing_dev_info; |
16725b9d SW |
935 | return err; |
936 | } | |
937 | ||
a7f9fb20 AV |
938 | static struct dentry *ceph_mount(struct file_system_type *fs_type, |
939 | int flags, const char *dev_name, void *data) | |
16725b9d SW |
940 | { |
941 | struct super_block *sb; | |
3d14c5d2 | 942 | struct ceph_fs_client *fsc; |
a7f9fb20 | 943 | struct dentry *res; |
16725b9d SW |
944 | int err; |
945 | int (*compare_super)(struct super_block *, void *) = ceph_compare_super; | |
6a18be16 | 946 | const char *path = NULL; |
3d14c5d2 YS |
947 | struct ceph_mount_options *fsopt = NULL; |
948 | struct ceph_options *opt = NULL; | |
16725b9d | 949 | |
a7f9fb20 | 950 | dout("ceph_mount\n"); |
45195e42 SW |
951 | |
952 | #ifdef CONFIG_CEPH_FS_POSIX_ACL | |
953 | flags |= MS_POSIXACL; | |
954 | #endif | |
3d14c5d2 | 955 | err = parse_mount_options(&fsopt, &opt, flags, data, dev_name, &path); |
a7f9fb20 AV |
956 | if (err < 0) { |
957 | res = ERR_PTR(err); | |
6b805185 | 958 | goto out_final; |
a7f9fb20 | 959 | } |
16725b9d SW |
960 | |
961 | /* create client (which we may/may not use) */ | |
3d14c5d2 YS |
962 | fsc = create_fs_client(fsopt, opt); |
963 | if (IS_ERR(fsc)) { | |
a7f9fb20 | 964 | res = ERR_CAST(fsc); |
259a187a NW |
965 | destroy_mount_options(fsopt); |
966 | ceph_destroy_options(opt); | |
6b805185 SW |
967 | goto out_final; |
968 | } | |
16725b9d | 969 | |
3d14c5d2 | 970 | err = ceph_mdsc_init(fsc); |
a7f9fb20 AV |
971 | if (err < 0) { |
972 | res = ERR_PTR(err); | |
3d14c5d2 | 973 | goto out; |
a7f9fb20 | 974 | } |
3d14c5d2 YS |
975 | |
976 | if (ceph_test_opt(fsc->client, NOSHARE)) | |
16725b9d | 977 | compare_super = NULL; |
9249e17f | 978 | sb = sget(fs_type, compare_super, ceph_set_super, flags, fsc); |
16725b9d | 979 | if (IS_ERR(sb)) { |
a7f9fb20 | 980 | res = ERR_CAST(sb); |
16725b9d SW |
981 | goto out; |
982 | } | |
983 | ||
3d14c5d2 YS |
984 | if (ceph_sb_to_client(sb) != fsc) { |
985 | ceph_mdsc_destroy(fsc); | |
986 | destroy_fs_client(fsc); | |
987 | fsc = ceph_sb_to_client(sb); | |
988 | dout("get_sb got existing client %p\n", fsc); | |
16725b9d | 989 | } else { |
3d14c5d2 YS |
990 | dout("get_sb using new client %p\n", fsc); |
991 | err = ceph_register_bdi(sb, fsc); | |
a7f9fb20 AV |
992 | if (err < 0) { |
993 | res = ERR_PTR(err); | |
16725b9d | 994 | goto out_splat; |
a7f9fb20 | 995 | } |
16725b9d SW |
996 | } |
997 | ||
a7f9fb20 AV |
998 | res = ceph_real_mount(fsc, path); |
999 | if (IS_ERR(res)) | |
16725b9d | 1000 | goto out_splat; |
a7f9fb20 | 1001 | dout("root %p inode %p ino %llx.%llx\n", res, |
2b0143b5 | 1002 | d_inode(res), ceph_vinop(d_inode(res))); |
a7f9fb20 | 1003 | return res; |
16725b9d SW |
1004 | |
1005 | out_splat: | |
3d14c5d2 | 1006 | ceph_mdsc_close_sessions(fsc->mdsc); |
3981f2e2 | 1007 | deactivate_locked_super(sb); |
16725b9d SW |
1008 | goto out_final; |
1009 | ||
1010 | out: | |
3d14c5d2 YS |
1011 | ceph_mdsc_destroy(fsc); |
1012 | destroy_fs_client(fsc); | |
16725b9d | 1013 | out_final: |
a7f9fb20 AV |
1014 | dout("ceph_mount fail %ld\n", PTR_ERR(res)); |
1015 | return res; | |
16725b9d SW |
1016 | } |
1017 | ||
1018 | static void ceph_kill_sb(struct super_block *s) | |
1019 | { | |
3d14c5d2 | 1020 | struct ceph_fs_client *fsc = ceph_sb_to_client(s); |
e4d27509 CH |
1021 | dev_t dev = s->s_dev; |
1022 | ||
16725b9d | 1023 | dout("kill_sb %p\n", s); |
e4d27509 | 1024 | |
3d14c5d2 | 1025 | ceph_mdsc_pre_umount(fsc->mdsc); |
e4d27509 | 1026 | generic_shutdown_super(s); |
3d14c5d2 | 1027 | ceph_mdsc_destroy(fsc); |
e4d27509 | 1028 | |
3d14c5d2 | 1029 | destroy_fs_client(fsc); |
e4d27509 | 1030 | free_anon_bdev(dev); |
16725b9d SW |
1031 | } |
1032 | ||
1033 | static struct file_system_type ceph_fs_type = { | |
1034 | .owner = THIS_MODULE, | |
1035 | .name = "ceph", | |
a7f9fb20 | 1036 | .mount = ceph_mount, |
16725b9d SW |
1037 | .kill_sb = ceph_kill_sb, |
1038 | .fs_flags = FS_RENAME_DOES_D_MOVE, | |
1039 | }; | |
7f78e035 | 1040 | MODULE_ALIAS_FS("ceph"); |
16725b9d | 1041 | |
16725b9d SW |
1042 | static int __init init_ceph(void) |
1043 | { | |
3d14c5d2 | 1044 | int ret = init_caches(); |
16725b9d | 1045 | if (ret) |
3d14c5d2 | 1046 | goto out; |
16725b9d | 1047 | |
eb13e832 | 1048 | ceph_flock_init(); |
3ce6cd12 | 1049 | ceph_xattr_init(); |
16725b9d SW |
1050 | ret = register_filesystem(&ceph_fs_type); |
1051 | if (ret) | |
34b759b4 | 1052 | goto out_xattr; |
16725b9d | 1053 | |
3d14c5d2 YS |
1054 | pr_info("loaded (mds proto %d)\n", CEPH_MDSC_PROTOCOL); |
1055 | ||
16725b9d SW |
1056 | return 0; |
1057 | ||
97c85a82 | 1058 | out_xattr: |
3ce6cd12 | 1059 | ceph_xattr_exit(); |
16725b9d | 1060 | destroy_caches(); |
16725b9d SW |
1061 | out: |
1062 | return ret; | |
1063 | } | |
1064 | ||
1065 | static void __exit exit_ceph(void) | |
1066 | { | |
1067 | dout("exit_ceph\n"); | |
1068 | unregister_filesystem(&ceph_fs_type); | |
3ce6cd12 | 1069 | ceph_xattr_exit(); |
16725b9d | 1070 | destroy_caches(); |
16725b9d SW |
1071 | } |
1072 | ||
1073 | module_init(init_ceph); | |
1074 | module_exit(exit_ceph); | |
1075 | ||
1076 | MODULE_AUTHOR("Sage Weil <sage@newdream.net>"); | |
1077 | MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>"); | |
1078 | MODULE_AUTHOR("Patience Warnick <patience@newdream.net>"); | |
1079 | MODULE_DESCRIPTION("Ceph filesystem for Linux"); | |
1080 | MODULE_LICENSE("GPL"); |