Commit | Line | Data |
---|---|---|
5db11c21 MM |
1 | /* |
2 | * (C) 2001 Clemson University and The University of Chicago | |
3 | * | |
4 | * See COPYING in top-level directory. | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Linux VFS inode operations. | |
9 | */ | |
10 | ||
11 | #include "protocol.h" | |
12 | #include "pvfs2-kernel.h" | |
13 | #include "pvfs2-bufmap.h" | |
14 | ||
15 | static int read_one_page(struct page *page) | |
16 | { | |
17 | void *page_data; | |
18 | int ret; | |
19 | int max_block; | |
20 | ssize_t bytes_read = 0; | |
21 | struct inode *inode = page->mapping->host; | |
22 | const __u32 blocksize = PAGE_CACHE_SIZE; /* inode->i_blksize */ | |
23 | const __u32 blockbits = PAGE_CACHE_SHIFT; /* inode->i_blkbits */ | |
24 | ||
25 | gossip_debug(GOSSIP_INODE_DEBUG, | |
26 | "pvfs2_readpage called with page %p\n", | |
27 | page); | |
28 | page_data = pvfs2_kmap(page); | |
29 | ||
30 | max_block = ((inode->i_size / blocksize) + 1); | |
31 | ||
32 | if (page->index < max_block) { | |
33 | loff_t blockptr_offset = (((loff_t) page->index) << blockbits); | |
34 | ||
35 | bytes_read = pvfs2_inode_read(inode, | |
84d02150 | 36 | (char __user *) page_data, |
5db11c21 MM |
37 | blocksize, |
38 | &blockptr_offset, | |
39 | inode->i_size); | |
40 | } | |
41 | /* only zero remaining unread portions of the page data */ | |
42 | if (bytes_read > 0) | |
43 | memset(page_data + bytes_read, 0, blocksize - bytes_read); | |
44 | else | |
45 | memset(page_data, 0, blocksize); | |
46 | /* takes care of potential aliasing */ | |
47 | flush_dcache_page(page); | |
48 | if (bytes_read < 0) { | |
49 | ret = bytes_read; | |
50 | SetPageError(page); | |
51 | } else { | |
52 | SetPageUptodate(page); | |
53 | if (PageError(page)) | |
54 | ClearPageError(page); | |
55 | ret = 0; | |
56 | } | |
57 | pvfs2_kunmap(page); | |
58 | /* unlock the page after the ->readpage() routine completes */ | |
59 | unlock_page(page); | |
60 | return ret; | |
61 | } | |
62 | ||
63 | static int pvfs2_readpage(struct file *file, struct page *page) | |
64 | { | |
65 | return read_one_page(page); | |
66 | } | |
67 | ||
68 | static int pvfs2_readpages(struct file *file, | |
69 | struct address_space *mapping, | |
70 | struct list_head *pages, | |
71 | unsigned nr_pages) | |
72 | { | |
73 | int page_idx; | |
74 | int ret; | |
75 | ||
76 | gossip_debug(GOSSIP_INODE_DEBUG, "pvfs2_readpages called\n"); | |
77 | ||
78 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
79 | struct page *page; | |
80 | ||
81 | page = list_entry(pages->prev, struct page, lru); | |
82 | list_del(&page->lru); | |
83 | if (!add_to_page_cache(page, | |
84 | mapping, | |
85 | page->index, | |
86 | GFP_KERNEL)) { | |
87 | ret = read_one_page(page); | |
88 | gossip_debug(GOSSIP_INODE_DEBUG, | |
89 | "failure adding page to cache, read_one_page returned: %d\n", | |
90 | ret); | |
91 | } else { | |
92 | page_cache_release(page); | |
93 | } | |
94 | } | |
95 | BUG_ON(!list_empty(pages)); | |
96 | return 0; | |
97 | } | |
98 | ||
99 | static void pvfs2_invalidatepage(struct page *page, | |
100 | unsigned int offset, | |
101 | unsigned int length) | |
102 | { | |
103 | gossip_debug(GOSSIP_INODE_DEBUG, | |
104 | "pvfs2_invalidatepage called on page %p " | |
105 | "(offset is %u)\n", | |
106 | page, | |
107 | offset); | |
108 | ||
109 | ClearPageUptodate(page); | |
110 | ClearPageMappedToDisk(page); | |
111 | return; | |
112 | ||
113 | } | |
114 | ||
115 | static int pvfs2_releasepage(struct page *page, gfp_t foo) | |
116 | { | |
117 | gossip_debug(GOSSIP_INODE_DEBUG, | |
118 | "pvfs2_releasepage called on page %p\n", | |
119 | page); | |
120 | return 0; | |
121 | } | |
122 | ||
123 | /* | |
124 | * Having a direct_IO entry point in the address_space_operations | |
125 | * struct causes the kernel to allows us to use O_DIRECT on | |
126 | * open. Nothing will ever call this thing, but in the future we | |
127 | * will need to be able to use O_DIRECT on open in order to support | |
128 | * AIO. Modeled after NFS, they do this too. | |
129 | */ | |
130 | /* | |
131 | static ssize_t pvfs2_direct_IO(int rw, | |
132 | struct kiocb *iocb, | |
133 | struct iov_iter *iter, | |
134 | loff_t offset) | |
135 | { | |
136 | gossip_debug(GOSSIP_INODE_DEBUG, | |
137 | "pvfs2_direct_IO: %s\n", | |
138 | iocb->ki_filp->f_path.dentry->d_name.name); | |
139 | ||
140 | return -EINVAL; | |
141 | } | |
142 | */ | |
143 | ||
144 | struct backing_dev_info pvfs2_backing_dev_info = { | |
145 | .name = "pvfs2", | |
146 | .ra_pages = 0, | |
147 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, | |
148 | }; | |
149 | ||
150 | /** PVFS2 implementation of address space operations */ | |
151 | const struct address_space_operations pvfs2_address_operations = { | |
152 | .readpage = pvfs2_readpage, | |
153 | .readpages = pvfs2_readpages, | |
154 | .invalidatepage = pvfs2_invalidatepage, | |
155 | .releasepage = pvfs2_releasepage, | |
156 | /* .direct_IO = pvfs2_direct_IO */ | |
157 | }; | |
158 | ||
159 | static int pvfs2_setattr_size(struct inode *inode, struct iattr *iattr) | |
160 | { | |
161 | struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); | |
162 | struct pvfs2_kernel_op_s *new_op; | |
163 | loff_t orig_size = i_size_read(inode); | |
164 | int ret = -EINVAL; | |
165 | ||
166 | gossip_debug(GOSSIP_INODE_DEBUG, | |
167 | "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", | |
168 | __func__, | |
169 | get_khandle_from_ino(inode), | |
170 | &pvfs2_inode->refn.khandle, | |
171 | pvfs2_inode->refn.fs_id, | |
172 | iattr->ia_size); | |
173 | ||
174 | truncate_setsize(inode, iattr->ia_size); | |
175 | ||
176 | new_op = op_alloc(PVFS2_VFS_OP_TRUNCATE); | |
177 | if (!new_op) | |
178 | return -ENOMEM; | |
179 | ||
180 | new_op->upcall.req.truncate.refn = pvfs2_inode->refn; | |
181 | new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; | |
182 | ||
183 | ret = service_operation(new_op, __func__, | |
184 | get_interruptible_flag(inode)); | |
185 | ||
186 | /* | |
187 | * the truncate has no downcall members to retrieve, but | |
188 | * the status value tells us if it went through ok or not | |
189 | */ | |
190 | gossip_debug(GOSSIP_INODE_DEBUG, | |
191 | "pvfs2: pvfs2_truncate got return value of %d\n", | |
192 | ret); | |
193 | ||
194 | op_release(new_op); | |
195 | ||
196 | if (ret != 0) | |
197 | return ret; | |
198 | ||
199 | /* | |
200 | * Only change the c/mtime if we are changing the size or we are | |
201 | * explicitly asked to change it. This handles the semantic difference | |
202 | * between truncate() and ftruncate() as implemented in the VFS. | |
203 | * | |
204 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
205 | * special case where we need to update the times despite not having | |
206 | * these flags set. For all other operations the VFS set these flags | |
207 | * explicitly if it wants a timestamp update. | |
208 | */ | |
209 | if (orig_size != i_size_read(inode) && | |
210 | !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { | |
211 | iattr->ia_ctime = iattr->ia_mtime = | |
212 | current_fs_time(inode->i_sb); | |
213 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; | |
214 | } | |
215 | ||
216 | return ret; | |
217 | } | |
218 | ||
219 | /* | |
220 | * Change attributes of an object referenced by dentry. | |
221 | */ | |
222 | int pvfs2_setattr(struct dentry *dentry, struct iattr *iattr) | |
223 | { | |
224 | int ret = -EINVAL; | |
225 | struct inode *inode = dentry->d_inode; | |
226 | ||
227 | gossip_debug(GOSSIP_INODE_DEBUG, | |
228 | "pvfs2_setattr: called on %s\n", | |
229 | dentry->d_name.name); | |
230 | ||
231 | ret = inode_change_ok(inode, iattr); | |
232 | if (ret) | |
233 | goto out; | |
234 | ||
235 | if ((iattr->ia_valid & ATTR_SIZE) && | |
236 | iattr->ia_size != i_size_read(inode)) { | |
237 | ret = pvfs2_setattr_size(inode, iattr); | |
238 | if (ret) | |
239 | goto out; | |
240 | } | |
241 | ||
242 | setattr_copy(inode, iattr); | |
243 | mark_inode_dirty(inode); | |
244 | ||
245 | ret = pvfs2_inode_setattr(inode, iattr); | |
246 | gossip_debug(GOSSIP_INODE_DEBUG, | |
247 | "pvfs2_setattr: inode_setattr returned %d\n", | |
248 | ret); | |
249 | ||
250 | if (!ret && (iattr->ia_valid & ATTR_MODE)) | |
251 | /* change mod on a file that has ACLs */ | |
252 | ret = posix_acl_chmod(inode, inode->i_mode); | |
253 | ||
254 | out: | |
255 | gossip_debug(GOSSIP_INODE_DEBUG, "pvfs2_setattr: returning %d\n", ret); | |
256 | return ret; | |
257 | } | |
258 | ||
259 | /* | |
260 | * Obtain attributes of an object given a dentry | |
261 | */ | |
262 | int pvfs2_getattr(struct vfsmount *mnt, | |
263 | struct dentry *dentry, | |
264 | struct kstat *kstat) | |
265 | { | |
266 | int ret = -ENOENT; | |
267 | struct inode *inode = dentry->d_inode; | |
268 | struct pvfs2_inode_s *pvfs2_inode = NULL; | |
269 | ||
270 | gossip_debug(GOSSIP_INODE_DEBUG, | |
271 | "pvfs2_getattr: called on %s\n", | |
272 | dentry->d_name.name); | |
273 | ||
274 | /* | |
275 | * Similar to the above comment, a getattr also expects that all | |
276 | * fields/attributes of the inode would be refreshed. So again, we | |
277 | * dont have too much of a choice but refresh all the attributes. | |
278 | */ | |
279 | ret = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); | |
280 | if (ret == 0) { | |
281 | generic_fillattr(inode, kstat); | |
282 | /* override block size reported to stat */ | |
283 | pvfs2_inode = PVFS2_I(inode); | |
284 | kstat->blksize = pvfs2_inode->blksize; | |
285 | } else { | |
286 | /* assume an I/O error and flag inode as bad */ | |
287 | gossip_debug(GOSSIP_INODE_DEBUG, | |
288 | "%s:%s:%d calling make bad inode\n", | |
289 | __FILE__, | |
290 | __func__, | |
291 | __LINE__); | |
292 | pvfs2_make_bad_inode(inode); | |
293 | } | |
294 | return ret; | |
295 | } | |
296 | ||
297 | /* PVFS2 implementation of VFS inode operations for files */ | |
298 | struct inode_operations pvfs2_file_inode_operations = { | |
299 | .get_acl = pvfs2_get_acl, | |
300 | .set_acl = pvfs2_set_acl, | |
301 | .setattr = pvfs2_setattr, | |
302 | .getattr = pvfs2_getattr, | |
303 | .setxattr = generic_setxattr, | |
304 | .getxattr = generic_getxattr, | |
305 | .listxattr = pvfs2_listxattr, | |
306 | .removexattr = generic_removexattr, | |
307 | }; | |
308 | ||
309 | static int pvfs2_init_iops(struct inode *inode) | |
310 | { | |
311 | inode->i_mapping->a_ops = &pvfs2_address_operations; | |
312 | ||
313 | switch (inode->i_mode & S_IFMT) { | |
314 | case S_IFREG: | |
315 | inode->i_op = &pvfs2_file_inode_operations; | |
316 | inode->i_fop = &pvfs2_file_operations; | |
317 | inode->i_blkbits = PAGE_CACHE_SHIFT; | |
318 | break; | |
319 | case S_IFLNK: | |
320 | inode->i_op = &pvfs2_symlink_inode_operations; | |
321 | break; | |
322 | case S_IFDIR: | |
323 | inode->i_op = &pvfs2_dir_inode_operations; | |
324 | inode->i_fop = &pvfs2_dir_operations; | |
325 | break; | |
326 | default: | |
327 | gossip_debug(GOSSIP_INODE_DEBUG, | |
328 | "%s: unsupported mode\n", | |
329 | __func__); | |
330 | return -EINVAL; | |
331 | } | |
332 | ||
333 | return 0; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Given a PVFS2 object identifier (fsid, handle), convert it into a ino_t type | |
338 | * that will be used as a hash-index from where the handle will | |
339 | * be searched for in the VFS hash table of inodes. | |
340 | */ | |
341 | static inline ino_t pvfs2_handle_hash(struct pvfs2_object_kref *ref) | |
342 | { | |
343 | if (!ref) | |
344 | return 0; | |
345 | return pvfs2_khandle_to_ino(&(ref->khandle)); | |
346 | } | |
347 | ||
348 | /* | |
349 | * Called to set up an inode from iget5_locked. | |
350 | */ | |
351 | static int pvfs2_set_inode(struct inode *inode, void *data) | |
352 | { | |
353 | struct pvfs2_object_kref *ref = (struct pvfs2_object_kref *) data; | |
354 | struct pvfs2_inode_s *pvfs2_inode = NULL; | |
355 | ||
356 | /* Make sure that we have sane parameters */ | |
357 | if (!data || !inode) | |
358 | return 0; | |
359 | pvfs2_inode = PVFS2_I(inode); | |
360 | if (!pvfs2_inode) | |
361 | return 0; | |
362 | pvfs2_inode->refn.fs_id = ref->fs_id; | |
363 | pvfs2_inode->refn.khandle = ref->khandle; | |
364 | return 0; | |
365 | } | |
366 | ||
367 | /* | |
368 | * Called to determine if handles match. | |
369 | */ | |
370 | static int pvfs2_test_inode(struct inode *inode, void *data) | |
371 | { | |
372 | struct pvfs2_object_kref *ref = (struct pvfs2_object_kref *) data; | |
373 | struct pvfs2_inode_s *pvfs2_inode = NULL; | |
374 | ||
375 | pvfs2_inode = PVFS2_I(inode); | |
376 | return (!PVFS_khandle_cmp(&(pvfs2_inode->refn.khandle), &(ref->khandle)) | |
377 | && pvfs2_inode->refn.fs_id == ref->fs_id); | |
378 | } | |
379 | ||
380 | /* | |
381 | * Front-end to lookup the inode-cache maintained by the VFS using the PVFS2 | |
382 | * file handle. | |
383 | * | |
384 | * @sb: the file system super block instance. | |
385 | * @ref: The PVFS2 object for which we are trying to locate an inode structure. | |
386 | */ | |
387 | struct inode *pvfs2_iget(struct super_block *sb, struct pvfs2_object_kref *ref) | |
388 | { | |
389 | struct inode *inode = NULL; | |
390 | unsigned long hash; | |
391 | int error; | |
392 | ||
393 | hash = pvfs2_handle_hash(ref); | |
394 | inode = iget5_locked(sb, hash, pvfs2_test_inode, pvfs2_set_inode, ref); | |
395 | if (!inode || !(inode->i_state & I_NEW)) | |
396 | return inode; | |
397 | ||
398 | error = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); | |
399 | if (error) { | |
400 | iget_failed(inode); | |
401 | return ERR_PTR(error); | |
402 | } | |
403 | ||
404 | inode->i_ino = hash; /* needed for stat etc */ | |
405 | pvfs2_init_iops(inode); | |
406 | unlock_new_inode(inode); | |
407 | ||
408 | gossip_debug(GOSSIP_INODE_DEBUG, | |
409 | "iget handle %pU, fsid %d hash %ld i_ino %lu\n", | |
410 | &ref->khandle, | |
411 | ref->fs_id, | |
412 | hash, | |
413 | inode->i_ino); | |
414 | ||
415 | return inode; | |
416 | } | |
417 | ||
418 | /* | |
419 | * Allocate an inode for a newly created file and insert it into the inode hash. | |
420 | */ | |
421 | struct inode *pvfs2_new_inode(struct super_block *sb, struct inode *dir, | |
422 | int mode, dev_t dev, struct pvfs2_object_kref *ref) | |
423 | { | |
424 | unsigned long hash = pvfs2_handle_hash(ref); | |
425 | struct inode *inode; | |
426 | int error; | |
427 | ||
428 | gossip_debug(GOSSIP_INODE_DEBUG, | |
429 | "pvfs2_get_custom_inode_common: called\n" | |
430 | "(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", | |
431 | sb, | |
432 | MAJOR(dev), | |
433 | MINOR(dev), | |
434 | mode); | |
435 | ||
436 | inode = new_inode(sb); | |
437 | if (!inode) | |
438 | return NULL; | |
439 | ||
440 | pvfs2_set_inode(inode, ref); | |
441 | inode->i_ino = hash; /* needed for stat etc */ | |
442 | ||
443 | error = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); | |
444 | if (error) | |
445 | goto out_iput; | |
446 | ||
447 | pvfs2_init_iops(inode); | |
448 | ||
449 | inode->i_mode = mode; | |
450 | inode->i_uid = current_fsuid(); | |
451 | inode->i_gid = current_fsgid(); | |
452 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
453 | inode->i_size = PAGE_CACHE_SIZE; | |
454 | inode->i_rdev = dev; | |
455 | ||
456 | error = insert_inode_locked4(inode, hash, pvfs2_test_inode, ref); | |
457 | if (error < 0) | |
458 | goto out_iput; | |
459 | ||
460 | gossip_debug(GOSSIP_INODE_DEBUG, | |
461 | "Initializing ACL's for inode %pU\n", | |
462 | get_khandle_from_ino(inode)); | |
463 | pvfs2_init_acl(inode, dir); | |
464 | return inode; | |
465 | ||
466 | out_iput: | |
467 | iput(inode); | |
468 | return ERR_PTR(error); | |
469 | } |