Commit | Line | Data |
---|---|---|
8df08c89 MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * dlmfs.c | |
5 | * | |
6 | * Code which implements the kernel side of a minimal userspace | |
7 | * interface to our DLM. This file handles the virtual file system | |
8 | * used for communication with userspace. Credit should go to ramfs, | |
9 | * which was a template for the fs side of this module. | |
10 | * | |
11 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. | |
12 | * | |
13 | * This program is free software; you can redistribute it and/or | |
14 | * modify it under the terms of the GNU General Public | |
15 | * License as published by the Free Software Foundation; either | |
16 | * version 2 of the License, or (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
21 | * General Public License for more details. | |
22 | * | |
23 | * You should have received a copy of the GNU General Public | |
24 | * License along with this program; if not, write to the | |
25 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
26 | * Boston, MA 021110-1307, USA. | |
27 | */ | |
28 | ||
29 | /* Simple VFS hooks based on: */ | |
30 | /* | |
31 | * Resizable simple ram filesystem for Linux. | |
32 | * | |
33 | * Copyright (C) 2000 Linus Torvalds. | |
34 | * 2000 Transmeta Corp. | |
35 | */ | |
36 | ||
37 | #include <linux/module.h> | |
38 | #include <linux/fs.h> | |
39 | #include <linux/pagemap.h> | |
40 | #include <linux/types.h> | |
41 | #include <linux/slab.h> | |
42 | #include <linux/highmem.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/string.h> | |
8df08c89 | 45 | #include <linux/backing-dev.h> |
65b6f340 | 46 | #include <linux/poll.h> |
8df08c89 MF |
47 | |
48 | #include <asm/uaccess.h> | |
49 | ||
0016eedc | 50 | #include "stackglue.h" |
8df08c89 | 51 | #include "userdlm.h" |
8df08c89 MF |
52 | #include "dlmfsver.h" |
53 | ||
54 | #define MLOG_MASK_PREFIX ML_DLMFS | |
55 | #include "cluster/masklog.h" | |
56 | ||
d24fbcda | 57 | |
ee9b6d61 | 58 | static const struct super_operations dlmfs_ops; |
00977a59 | 59 | static const struct file_operations dlmfs_file_operations; |
92e1d5be AV |
60 | static const struct inode_operations dlmfs_dir_inode_operations; |
61 | static const struct inode_operations dlmfs_root_inode_operations; | |
62 | static const struct inode_operations dlmfs_file_inode_operations; | |
e18b890b | 63 | static struct kmem_cache *dlmfs_inode_cache; |
8df08c89 MF |
64 | |
65 | struct workqueue_struct *user_dlm_worker; | |
66 | ||
d24fbcda | 67 | |
14a437c2 JB |
68 | |
69 | /* | |
70 | * These are the ABI capabilities of dlmfs. | |
71 | * | |
72 | * Over time, dlmfs has added some features that were not part of the | |
73 | * initial ABI. Unfortunately, some of these features are not detectable | |
74 | * via standard usage. For example, Linux's default poll always returns | |
75 | * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs | |
76 | * added poll support. Instead, we provide this list of new capabilities. | |
77 | * | |
78 | * Capabilities is a read-only attribute. We do it as a module parameter | |
79 | * so we can discover it whether dlmfs is built in, loaded, or even not | |
80 | * loaded. | |
81 | * | |
82 | * The ABI features are local to this machine's dlmfs mount. This is | |
83 | * distinct from the locking protocol, which is concerned with inter-node | |
84 | * interaction. | |
65b6f340 JB |
85 | * |
86 | * Capabilities: | |
87 | * - bast : POLLIN against the file descriptor of a held lock | |
88 | * signifies a bast fired on the lock. | |
14a437c2 | 89 | */ |
cbe0e331 | 90 | #define DLMFS_CAPABILITIES "bast stackglue" |
14a437c2 JB |
91 | extern int param_set_dlmfs_capabilities(const char *val, |
92 | struct kernel_param *kp) | |
93 | { | |
94 | printk(KERN_ERR "%s: readonly parameter\n", kp->name); | |
95 | return -EINVAL; | |
96 | } | |
97 | static int param_get_dlmfs_capabilities(char *buffer, | |
98 | struct kernel_param *kp) | |
99 | { | |
100 | return strlcpy(buffer, DLMFS_CAPABILITIES, | |
101 | strlen(DLMFS_CAPABILITIES) + 1); | |
102 | } | |
103 | module_param_call(capabilities, param_set_dlmfs_capabilities, | |
104 | param_get_dlmfs_capabilities, NULL, 0444); | |
105 | MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES); | |
106 | ||
107 | ||
8df08c89 MF |
108 | /* |
109 | * decodes a set of open flags into a valid lock level and a set of flags. | |
110 | * returns < 0 if we have invalid flags | |
111 | * flags which mean something to us: | |
112 | * O_RDONLY -> PRMODE level | |
113 | * O_WRONLY -> EXMODE level | |
114 | * | |
efd647f7 | 115 | * O_NONBLOCK -> NOQUEUE |
8df08c89 MF |
116 | */ |
117 | static int dlmfs_decode_open_flags(int open_flags, | |
118 | int *level, | |
119 | int *flags) | |
120 | { | |
121 | if (open_flags & (O_WRONLY|O_RDWR)) | |
efd647f7 | 122 | *level = DLM_LOCK_EX; |
8df08c89 | 123 | else |
efd647f7 | 124 | *level = DLM_LOCK_PR; |
8df08c89 MF |
125 | |
126 | *flags = 0; | |
127 | if (open_flags & O_NONBLOCK) | |
efd647f7 | 128 | *flags |= DLM_LKF_NOQUEUE; |
8df08c89 MF |
129 | |
130 | return 0; | |
131 | } | |
132 | ||
133 | static int dlmfs_file_open(struct inode *inode, | |
134 | struct file *file) | |
135 | { | |
136 | int status, level, flags; | |
137 | struct dlmfs_filp_private *fp = NULL; | |
138 | struct dlmfs_inode_private *ip; | |
139 | ||
140 | if (S_ISDIR(inode->i_mode)) | |
141 | BUG(); | |
142 | ||
143 | mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino, | |
144 | file->f_flags); | |
145 | ||
146 | status = dlmfs_decode_open_flags(file->f_flags, &level, &flags); | |
147 | if (status < 0) | |
148 | goto bail; | |
149 | ||
150 | /* We don't want to honor O_APPEND at read/write time as it | |
151 | * doesn't make sense for LVB writes. */ | |
152 | file->f_flags &= ~O_APPEND; | |
153 | ||
ad8100e0 | 154 | fp = kmalloc(sizeof(*fp), GFP_NOFS); |
8df08c89 MF |
155 | if (!fp) { |
156 | status = -ENOMEM; | |
157 | goto bail; | |
158 | } | |
159 | fp->fp_lock_level = level; | |
160 | ||
161 | ip = DLMFS_I(inode); | |
162 | ||
163 | status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags); | |
164 | if (status < 0) { | |
165 | /* this is a strange error to return here but I want | |
166 | * to be able userspace to be able to distinguish a | |
167 | * valid lock request from one that simply couldn't be | |
168 | * granted. */ | |
efd647f7 | 169 | if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN) |
8df08c89 MF |
170 | status = -ETXTBSY; |
171 | kfree(fp); | |
172 | goto bail; | |
173 | } | |
174 | ||
175 | file->private_data = fp; | |
176 | bail: | |
177 | return status; | |
178 | } | |
179 | ||
180 | static int dlmfs_file_release(struct inode *inode, | |
181 | struct file *file) | |
182 | { | |
183 | int level, status; | |
184 | struct dlmfs_inode_private *ip = DLMFS_I(inode); | |
33fa1d90 | 185 | struct dlmfs_filp_private *fp = file->private_data; |
8df08c89 MF |
186 | |
187 | if (S_ISDIR(inode->i_mode)) | |
188 | BUG(); | |
189 | ||
190 | mlog(0, "close called on inode %lu\n", inode->i_ino); | |
191 | ||
192 | status = 0; | |
193 | if (fp) { | |
194 | level = fp->fp_lock_level; | |
efd647f7 | 195 | if (level != DLM_LOCK_IV) |
8df08c89 MF |
196 | user_dlm_cluster_unlock(&ip->ip_lockres, level); |
197 | ||
198 | kfree(fp); | |
199 | file->private_data = NULL; | |
200 | } | |
201 | ||
202 | return 0; | |
203 | } | |
204 | ||
e8fce482 JB |
205 | /* |
206 | * We do ->setattr() just to override size changes. Our size is the size | |
207 | * of the LVB and nothing else. | |
208 | */ | |
209 | static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr) | |
210 | { | |
211 | int error; | |
212 | struct inode *inode = dentry->d_inode; | |
213 | ||
214 | attr->ia_valid &= ~ATTR_SIZE; | |
215 | error = inode_change_ok(inode, attr); | |
1025774c CH |
216 | if (error) |
217 | return error; | |
e8fce482 | 218 | |
1025774c CH |
219 | setattr_copy(inode, attr); |
220 | mark_inode_dirty(inode); | |
221 | return 0; | |
e8fce482 JB |
222 | } |
223 | ||
65b6f340 JB |
224 | static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait) |
225 | { | |
226 | int event = 0; | |
227 | struct inode *inode = file->f_path.dentry->d_inode; | |
228 | struct dlmfs_inode_private *ip = DLMFS_I(inode); | |
229 | ||
230 | poll_wait(file, &ip->ip_lockres.l_event, wait); | |
231 | ||
232 | spin_lock(&ip->ip_lockres.l_lock); | |
233 | if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED) | |
234 | event = POLLIN | POLLRDNORM; | |
235 | spin_unlock(&ip->ip_lockres.l_lock); | |
236 | ||
237 | return event; | |
238 | } | |
239 | ||
8df08c89 MF |
240 | static ssize_t dlmfs_file_read(struct file *filp, |
241 | char __user *buf, | |
242 | size_t count, | |
243 | loff_t *ppos) | |
244 | { | |
245 | int bytes_left; | |
0016eedc | 246 | ssize_t readlen, got; |
8df08c89 | 247 | char *lvb_buf; |
d28c9174 | 248 | struct inode *inode = filp->f_path.dentry->d_inode; |
8df08c89 MF |
249 | |
250 | mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", | |
251 | inode->i_ino, count, *ppos); | |
252 | ||
253 | if (*ppos >= i_size_read(inode)) | |
254 | return 0; | |
255 | ||
256 | if (!count) | |
257 | return 0; | |
258 | ||
259 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
260 | return -EFAULT; | |
261 | ||
262 | /* don't read past the lvb */ | |
263 | if ((count + *ppos) > i_size_read(inode)) | |
264 | readlen = i_size_read(inode) - *ppos; | |
265 | else | |
a36d515c | 266 | readlen = count; |
8df08c89 | 267 | |
ad8100e0 | 268 | lvb_buf = kmalloc(readlen, GFP_NOFS); |
8df08c89 MF |
269 | if (!lvb_buf) |
270 | return -ENOMEM; | |
271 | ||
0016eedc JB |
272 | got = user_dlm_read_lvb(inode, lvb_buf, readlen); |
273 | if (got) { | |
274 | BUG_ON(got != readlen); | |
275 | bytes_left = __copy_to_user(buf, lvb_buf, readlen); | |
276 | readlen -= bytes_left; | |
277 | } else | |
278 | readlen = 0; | |
8df08c89 MF |
279 | |
280 | kfree(lvb_buf); | |
281 | ||
282 | *ppos = *ppos + readlen; | |
283 | ||
284 | mlog(0, "read %zd bytes\n", readlen); | |
285 | return readlen; | |
286 | } | |
287 | ||
288 | static ssize_t dlmfs_file_write(struct file *filp, | |
289 | const char __user *buf, | |
290 | size_t count, | |
291 | loff_t *ppos) | |
292 | { | |
293 | int bytes_left; | |
294 | ssize_t writelen; | |
295 | char *lvb_buf; | |
d28c9174 | 296 | struct inode *inode = filp->f_path.dentry->d_inode; |
8df08c89 MF |
297 | |
298 | mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", | |
299 | inode->i_ino, count, *ppos); | |
300 | ||
301 | if (*ppos >= i_size_read(inode)) | |
302 | return -ENOSPC; | |
303 | ||
304 | if (!count) | |
305 | return 0; | |
306 | ||
307 | if (!access_ok(VERIFY_READ, buf, count)) | |
308 | return -EFAULT; | |
309 | ||
310 | /* don't write past the lvb */ | |
311 | if ((count + *ppos) > i_size_read(inode)) | |
312 | writelen = i_size_read(inode) - *ppos; | |
313 | else | |
314 | writelen = count - *ppos; | |
315 | ||
ad8100e0 | 316 | lvb_buf = kmalloc(writelen, GFP_NOFS); |
8df08c89 MF |
317 | if (!lvb_buf) |
318 | return -ENOMEM; | |
319 | ||
320 | bytes_left = copy_from_user(lvb_buf, buf, writelen); | |
321 | writelen -= bytes_left; | |
322 | if (writelen) | |
323 | user_dlm_write_lvb(inode, lvb_buf, writelen); | |
324 | ||
325 | kfree(lvb_buf); | |
326 | ||
327 | *ppos = *ppos + writelen; | |
328 | mlog(0, "wrote %zd bytes\n", writelen); | |
329 | return writelen; | |
330 | } | |
331 | ||
51cc5068 | 332 | static void dlmfs_init_once(void *foo) |
8df08c89 MF |
333 | { |
334 | struct dlmfs_inode_private *ip = | |
335 | (struct dlmfs_inode_private *) foo; | |
336 | ||
0016eedc | 337 | ip->ip_conn = NULL; |
a35afb83 | 338 | ip->ip_parent = NULL; |
8df08c89 | 339 | |
a35afb83 | 340 | inode_init_once(&ip->ip_vfs_inode); |
8df08c89 MF |
341 | } |
342 | ||
343 | static struct inode *dlmfs_alloc_inode(struct super_block *sb) | |
344 | { | |
345 | struct dlmfs_inode_private *ip; | |
346 | ||
e6b4f8da | 347 | ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS); |
8df08c89 MF |
348 | if (!ip) |
349 | return NULL; | |
350 | ||
351 | return &ip->ip_vfs_inode; | |
352 | } | |
353 | ||
354 | static void dlmfs_destroy_inode(struct inode *inode) | |
355 | { | |
356 | kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode)); | |
357 | } | |
358 | ||
b57922d9 | 359 | static void dlmfs_evict_inode(struct inode *inode) |
8df08c89 MF |
360 | { |
361 | int status; | |
362 | struct dlmfs_inode_private *ip; | |
363 | ||
b57922d9 | 364 | end_writeback(inode); |
8df08c89 MF |
365 | |
366 | mlog(0, "inode %lu\n", inode->i_ino); | |
367 | ||
368 | ip = DLMFS_I(inode); | |
369 | ||
370 | if (S_ISREG(inode->i_mode)) { | |
371 | status = user_dlm_destroy_lock(&ip->ip_lockres); | |
372 | if (status < 0) | |
373 | mlog_errno(status); | |
374 | iput(ip->ip_parent); | |
375 | goto clear_fields; | |
376 | } | |
377 | ||
0016eedc | 378 | mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn); |
8df08c89 MF |
379 | /* we must be a directory. If required, lets unregister the |
380 | * dlm context now. */ | |
0016eedc JB |
381 | if (ip->ip_conn) |
382 | user_dlm_unregister(ip->ip_conn); | |
8df08c89 MF |
383 | clear_fields: |
384 | ip->ip_parent = NULL; | |
0016eedc | 385 | ip->ip_conn = NULL; |
8df08c89 MF |
386 | } |
387 | ||
388 | static struct backing_dev_info dlmfs_backing_dev_info = { | |
d993831f | 389 | .name = "ocfs2-dlmfs", |
8df08c89 | 390 | .ra_pages = 0, /* No readahead */ |
e4ad08fe | 391 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
8df08c89 MF |
392 | }; |
393 | ||
394 | static struct inode *dlmfs_get_root_inode(struct super_block *sb) | |
395 | { | |
396 | struct inode *inode = new_inode(sb); | |
397 | int mode = S_IFDIR | 0755; | |
398 | struct dlmfs_inode_private *ip; | |
399 | ||
400 | if (inode) { | |
401 | ip = DLMFS_I(inode); | |
402 | ||
85fe4025 | 403 | inode->i_ino = get_next_ino(); |
8df08c89 | 404 | inode->i_mode = mode; |
b19c2a3b DH |
405 | inode->i_uid = current_fsuid(); |
406 | inode->i_gid = current_fsgid(); | |
8df08c89 MF |
407 | inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; |
408 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
d8c76e6f | 409 | inc_nlink(inode); |
8df08c89 MF |
410 | |
411 | inode->i_fop = &simple_dir_operations; | |
412 | inode->i_op = &dlmfs_root_inode_operations; | |
413 | } | |
414 | ||
415 | return inode; | |
416 | } | |
417 | ||
418 | static struct inode *dlmfs_get_inode(struct inode *parent, | |
419 | struct dentry *dentry, | |
420 | int mode) | |
421 | { | |
422 | struct super_block *sb = parent->i_sb; | |
423 | struct inode * inode = new_inode(sb); | |
424 | struct dlmfs_inode_private *ip; | |
425 | ||
426 | if (!inode) | |
427 | return NULL; | |
428 | ||
85fe4025 | 429 | inode->i_ino = get_next_ino(); |
8df08c89 | 430 | inode->i_mode = mode; |
b19c2a3b DH |
431 | inode->i_uid = current_fsuid(); |
432 | inode->i_gid = current_fsgid(); | |
8df08c89 MF |
433 | inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; |
434 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
435 | ||
436 | ip = DLMFS_I(inode); | |
0016eedc | 437 | ip->ip_conn = DLMFS_I(parent)->ip_conn; |
8df08c89 MF |
438 | |
439 | switch (mode & S_IFMT) { | |
440 | default: | |
441 | /* for now we don't support anything other than | |
442 | * directories and regular files. */ | |
443 | BUG(); | |
444 | break; | |
445 | case S_IFREG: | |
446 | inode->i_op = &dlmfs_file_inode_operations; | |
447 | inode->i_fop = &dlmfs_file_operations; | |
448 | ||
449 | i_size_write(inode, DLM_LVB_LEN); | |
450 | ||
451 | user_dlm_lock_res_init(&ip->ip_lockres, dentry); | |
452 | ||
453 | /* released at clear_inode time, this insures that we | |
454 | * get to drop the dlm reference on each lock *before* | |
455 | * we call the unregister code for releasing parent | |
456 | * directories. */ | |
457 | ip->ip_parent = igrab(parent); | |
458 | BUG_ON(!ip->ip_parent); | |
459 | break; | |
460 | case S_IFDIR: | |
461 | inode->i_op = &dlmfs_dir_inode_operations; | |
462 | inode->i_fop = &simple_dir_operations; | |
463 | ||
464 | /* directory inodes start off with i_nlink == | |
465 | * 2 (for "." entry) */ | |
d8c76e6f | 466 | inc_nlink(inode); |
8df08c89 MF |
467 | break; |
468 | } | |
469 | ||
470 | if (parent->i_mode & S_ISGID) { | |
471 | inode->i_gid = parent->i_gid; | |
472 | if (S_ISDIR(mode)) | |
473 | inode->i_mode |= S_ISGID; | |
474 | } | |
475 | ||
476 | return inode; | |
477 | } | |
478 | ||
479 | /* | |
480 | * File creation. Allocate an inode, and we're done.. | |
481 | */ | |
482 | /* SMP-safe */ | |
483 | static int dlmfs_mkdir(struct inode * dir, | |
484 | struct dentry * dentry, | |
485 | int mode) | |
486 | { | |
487 | int status; | |
488 | struct inode *inode = NULL; | |
489 | struct qstr *domain = &dentry->d_name; | |
490 | struct dlmfs_inode_private *ip; | |
0016eedc | 491 | struct ocfs2_cluster_connection *conn; |
8df08c89 MF |
492 | |
493 | mlog(0, "mkdir %.*s\n", domain->len, domain->name); | |
494 | ||
495 | /* verify that we have a proper domain */ | |
0016eedc | 496 | if (domain->len >= GROUP_NAME_MAX) { |
8df08c89 MF |
497 | status = -EINVAL; |
498 | mlog(ML_ERROR, "invalid domain name for directory.\n"); | |
499 | goto bail; | |
500 | } | |
501 | ||
502 | inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR); | |
503 | if (!inode) { | |
504 | status = -ENOMEM; | |
505 | mlog_errno(status); | |
506 | goto bail; | |
507 | } | |
508 | ||
509 | ip = DLMFS_I(inode); | |
510 | ||
0016eedc JB |
511 | conn = user_dlm_register(domain); |
512 | if (IS_ERR(conn)) { | |
513 | status = PTR_ERR(conn); | |
8df08c89 MF |
514 | mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n", |
515 | status, domain->len, domain->name); | |
516 | goto bail; | |
517 | } | |
0016eedc | 518 | ip->ip_conn = conn; |
8df08c89 | 519 | |
d8c76e6f | 520 | inc_nlink(dir); |
8df08c89 MF |
521 | d_instantiate(dentry, inode); |
522 | dget(dentry); /* Extra count - pin the dentry in core */ | |
523 | ||
524 | status = 0; | |
525 | bail: | |
526 | if (status < 0) | |
527 | iput(inode); | |
528 | return status; | |
529 | } | |
530 | ||
531 | static int dlmfs_create(struct inode *dir, | |
532 | struct dentry *dentry, | |
533 | int mode, | |
534 | struct nameidata *nd) | |
535 | { | |
536 | int status = 0; | |
537 | struct inode *inode; | |
538 | struct qstr *name = &dentry->d_name; | |
539 | ||
540 | mlog(0, "create %.*s\n", name->len, name->name); | |
541 | ||
542 | /* verify name is valid and doesn't contain any dlm reserved | |
543 | * characters */ | |
544 | if (name->len >= USER_DLM_LOCK_ID_MAX_LEN || | |
545 | name->name[0] == '$') { | |
546 | status = -EINVAL; | |
547 | mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len, | |
548 | name->name); | |
549 | goto bail; | |
550 | } | |
551 | ||
552 | inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG); | |
553 | if (!inode) { | |
554 | status = -ENOMEM; | |
555 | mlog_errno(status); | |
556 | goto bail; | |
557 | } | |
558 | ||
559 | d_instantiate(dentry, inode); | |
560 | dget(dentry); /* Extra count - pin the dentry in core */ | |
561 | bail: | |
562 | return status; | |
563 | } | |
564 | ||
565 | static int dlmfs_unlink(struct inode *dir, | |
566 | struct dentry *dentry) | |
567 | { | |
568 | int status; | |
569 | struct inode *inode = dentry->d_inode; | |
570 | ||
571 | mlog(0, "unlink inode %lu\n", inode->i_ino); | |
572 | ||
573 | /* if there are no current holders, or none that are waiting | |
574 | * to acquire a lock, this basically destroys our lockres. */ | |
575 | status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres); | |
576 | if (status < 0) { | |
577 | mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n", | |
578 | dentry->d_name.len, dentry->d_name.name, status); | |
579 | goto bail; | |
580 | } | |
581 | status = simple_unlink(dir, dentry); | |
582 | bail: | |
583 | return status; | |
584 | } | |
585 | ||
586 | static int dlmfs_fill_super(struct super_block * sb, | |
587 | void * data, | |
588 | int silent) | |
589 | { | |
590 | struct inode * inode; | |
591 | struct dentry * root; | |
592 | ||
593 | sb->s_maxbytes = MAX_LFS_FILESIZE; | |
594 | sb->s_blocksize = PAGE_CACHE_SIZE; | |
595 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; | |
596 | sb->s_magic = DLMFS_MAGIC; | |
597 | sb->s_op = &dlmfs_ops; | |
598 | inode = dlmfs_get_root_inode(sb); | |
599 | if (!inode) | |
600 | return -ENOMEM; | |
601 | ||
602 | root = d_alloc_root(inode); | |
603 | if (!root) { | |
604 | iput(inode); | |
605 | return -ENOMEM; | |
606 | } | |
607 | sb->s_root = root; | |
608 | return 0; | |
609 | } | |
610 | ||
00977a59 | 611 | static const struct file_operations dlmfs_file_operations = { |
8df08c89 MF |
612 | .open = dlmfs_file_open, |
613 | .release = dlmfs_file_release, | |
65b6f340 | 614 | .poll = dlmfs_file_poll, |
8df08c89 MF |
615 | .read = dlmfs_file_read, |
616 | .write = dlmfs_file_write, | |
6038f373 | 617 | .llseek = default_llseek, |
8df08c89 MF |
618 | }; |
619 | ||
92e1d5be | 620 | static const struct inode_operations dlmfs_dir_inode_operations = { |
8df08c89 MF |
621 | .create = dlmfs_create, |
622 | .lookup = simple_lookup, | |
623 | .unlink = dlmfs_unlink, | |
624 | }; | |
625 | ||
626 | /* this way we can restrict mkdir to only the toplevel of the fs. */ | |
92e1d5be | 627 | static const struct inode_operations dlmfs_root_inode_operations = { |
8df08c89 MF |
628 | .lookup = simple_lookup, |
629 | .mkdir = dlmfs_mkdir, | |
630 | .rmdir = simple_rmdir, | |
631 | }; | |
632 | ||
ee9b6d61 | 633 | static const struct super_operations dlmfs_ops = { |
8df08c89 MF |
634 | .statfs = simple_statfs, |
635 | .alloc_inode = dlmfs_alloc_inode, | |
636 | .destroy_inode = dlmfs_destroy_inode, | |
b57922d9 | 637 | .evict_inode = dlmfs_evict_inode, |
8df08c89 MF |
638 | .drop_inode = generic_delete_inode, |
639 | }; | |
640 | ||
92e1d5be | 641 | static const struct inode_operations dlmfs_file_inode_operations = { |
8df08c89 | 642 | .getattr = simple_getattr, |
e8fce482 | 643 | .setattr = dlmfs_file_setattr, |
8df08c89 MF |
644 | }; |
645 | ||
3c26ff6e AV |
646 | static struct dentry *dlmfs_mount(struct file_system_type *fs_type, |
647 | int flags, const char *dev_name, void *data) | |
8df08c89 | 648 | { |
3c26ff6e | 649 | return mount_nodev(fs_type, flags, data, dlmfs_fill_super); |
8df08c89 MF |
650 | } |
651 | ||
652 | static struct file_system_type dlmfs_fs_type = { | |
653 | .owner = THIS_MODULE, | |
654 | .name = "ocfs2_dlmfs", | |
3c26ff6e | 655 | .mount = dlmfs_mount, |
8df08c89 MF |
656 | .kill_sb = kill_litter_super, |
657 | }; | |
658 | ||
659 | static int __init init_dlmfs_fs(void) | |
660 | { | |
661 | int status; | |
662 | int cleanup_inode = 0, cleanup_worker = 0; | |
663 | ||
664 | dlmfs_print_version(); | |
665 | ||
e0bf68dd PZ |
666 | status = bdi_init(&dlmfs_backing_dev_info); |
667 | if (status) | |
668 | return status; | |
669 | ||
8df08c89 MF |
670 | dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache", |
671 | sizeof(struct dlmfs_inode_private), | |
fffb60f9 PJ |
672 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
673 | SLAB_MEM_SPREAD), | |
20c2df83 | 674 | dlmfs_init_once); |
07d9a395 CL |
675 | if (!dlmfs_inode_cache) { |
676 | status = -ENOMEM; | |
e0bf68dd | 677 | goto bail; |
07d9a395 | 678 | } |
8df08c89 MF |
679 | cleanup_inode = 1; |
680 | ||
681 | user_dlm_worker = create_singlethread_workqueue("user_dlm"); | |
682 | if (!user_dlm_worker) { | |
683 | status = -ENOMEM; | |
684 | goto bail; | |
685 | } | |
686 | cleanup_worker = 1; | |
687 | ||
0016eedc | 688 | user_dlm_set_locking_protocol(); |
8df08c89 MF |
689 | status = register_filesystem(&dlmfs_fs_type); |
690 | bail: | |
691 | if (status) { | |
692 | if (cleanup_inode) | |
693 | kmem_cache_destroy(dlmfs_inode_cache); | |
694 | if (cleanup_worker) | |
695 | destroy_workqueue(user_dlm_worker); | |
e0bf68dd | 696 | bdi_destroy(&dlmfs_backing_dev_info); |
8df08c89 MF |
697 | } else |
698 | printk("OCFS2 User DLM kernel interface loaded\n"); | |
699 | return status; | |
700 | } | |
701 | ||
702 | static void __exit exit_dlmfs_fs(void) | |
703 | { | |
704 | unregister_filesystem(&dlmfs_fs_type); | |
705 | ||
706 | flush_workqueue(user_dlm_worker); | |
707 | destroy_workqueue(user_dlm_worker); | |
708 | ||
1a1d92c1 | 709 | kmem_cache_destroy(dlmfs_inode_cache); |
e0bf68dd PZ |
710 | |
711 | bdi_destroy(&dlmfs_backing_dev_info); | |
8df08c89 MF |
712 | } |
713 | ||
714 | MODULE_AUTHOR("Oracle"); | |
715 | MODULE_LICENSE("GPL"); | |
716 | ||
717 | module_init(init_dlmfs_fs) | |
718 | module_exit(exit_dlmfs_fs) |