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 | { | |
5db11c21 MM |
17 | int ret; |
18 | int max_block; | |
19 | ssize_t bytes_read = 0; | |
20 | struct inode *inode = page->mapping->host; | |
21 | const __u32 blocksize = PAGE_CACHE_SIZE; /* inode->i_blksize */ | |
22 | const __u32 blockbits = PAGE_CACHE_SHIFT; /* inode->i_blkbits */ | |
74f68fce AV |
23 | struct iov_iter to; |
24 | struct bio_vec bv = {.bv_page = page, .bv_len = PAGE_SIZE}; | |
25 | ||
26 | iov_iter_bvec(&to, ITER_BVEC | READ, &bv, 1, PAGE_SIZE); | |
5db11c21 MM |
27 | |
28 | gossip_debug(GOSSIP_INODE_DEBUG, | |
29 | "pvfs2_readpage called with page %p\n", | |
30 | page); | |
5db11c21 MM |
31 | |
32 | max_block = ((inode->i_size / blocksize) + 1); | |
33 | ||
34 | if (page->index < max_block) { | |
35 | loff_t blockptr_offset = (((loff_t) page->index) << blockbits); | |
36 | ||
37 | bytes_read = pvfs2_inode_read(inode, | |
74f68fce | 38 | &to, |
5db11c21 MM |
39 | &blockptr_offset, |
40 | inode->i_size); | |
41 | } | |
74f68fce AV |
42 | /* this will only zero remaining unread portions of the page data */ |
43 | iov_iter_zero(~0U, &to); | |
5db11c21 MM |
44 | /* takes care of potential aliasing */ |
45 | flush_dcache_page(page); | |
46 | if (bytes_read < 0) { | |
47 | ret = bytes_read; | |
48 | SetPageError(page); | |
49 | } else { | |
50 | SetPageUptodate(page); | |
51 | if (PageError(page)) | |
52 | ClearPageError(page); | |
53 | ret = 0; | |
54 | } | |
5db11c21 MM |
55 | /* unlock the page after the ->readpage() routine completes */ |
56 | unlock_page(page); | |
57 | return ret; | |
58 | } | |
59 | ||
60 | static int pvfs2_readpage(struct file *file, struct page *page) | |
61 | { | |
62 | return read_one_page(page); | |
63 | } | |
64 | ||
65 | static int pvfs2_readpages(struct file *file, | |
66 | struct address_space *mapping, | |
67 | struct list_head *pages, | |
68 | unsigned nr_pages) | |
69 | { | |
70 | int page_idx; | |
71 | int ret; | |
72 | ||
73 | gossip_debug(GOSSIP_INODE_DEBUG, "pvfs2_readpages called\n"); | |
74 | ||
75 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
76 | struct page *page; | |
77 | ||
78 | page = list_entry(pages->prev, struct page, lru); | |
79 | list_del(&page->lru); | |
80 | if (!add_to_page_cache(page, | |
81 | mapping, | |
82 | page->index, | |
83 | GFP_KERNEL)) { | |
84 | ret = read_one_page(page); | |
85 | gossip_debug(GOSSIP_INODE_DEBUG, | |
86 | "failure adding page to cache, read_one_page returned: %d\n", | |
87 | ret); | |
88 | } else { | |
89 | page_cache_release(page); | |
90 | } | |
91 | } | |
92 | BUG_ON(!list_empty(pages)); | |
93 | return 0; | |
94 | } | |
95 | ||
96 | static void pvfs2_invalidatepage(struct page *page, | |
97 | unsigned int offset, | |
98 | unsigned int length) | |
99 | { | |
100 | gossip_debug(GOSSIP_INODE_DEBUG, | |
101 | "pvfs2_invalidatepage called on page %p " | |
102 | "(offset is %u)\n", | |
103 | page, | |
104 | offset); | |
105 | ||
106 | ClearPageUptodate(page); | |
107 | ClearPageMappedToDisk(page); | |
108 | return; | |
109 | ||
110 | } | |
111 | ||
112 | static int pvfs2_releasepage(struct page *page, gfp_t foo) | |
113 | { | |
114 | gossip_debug(GOSSIP_INODE_DEBUG, | |
115 | "pvfs2_releasepage called on page %p\n", | |
116 | page); | |
117 | return 0; | |
118 | } | |
119 | ||
120 | /* | |
121 | * Having a direct_IO entry point in the address_space_operations | |
122 | * struct causes the kernel to allows us to use O_DIRECT on | |
123 | * open. Nothing will ever call this thing, but in the future we | |
124 | * will need to be able to use O_DIRECT on open in order to support | |
125 | * AIO. Modeled after NFS, they do this too. | |
126 | */ | |
127 | /* | |
54804949 MM |
128 | * static ssize_t pvfs2_direct_IO(int rw, |
129 | * struct kiocb *iocb, | |
130 | * struct iov_iter *iter, | |
131 | * loff_t offset) | |
132 | *{ | |
133 | * gossip_debug(GOSSIP_INODE_DEBUG, | |
134 | * "pvfs2_direct_IO: %s\n", | |
135 | * iocb->ki_filp->f_path.dentry->d_name.name); | |
136 | * | |
137 | * return -EINVAL; | |
138 | *} | |
139 | */ | |
5db11c21 MM |
140 | |
141 | struct backing_dev_info pvfs2_backing_dev_info = { | |
142 | .name = "pvfs2", | |
143 | .ra_pages = 0, | |
144 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, | |
145 | }; | |
146 | ||
147 | /** PVFS2 implementation of address space operations */ | |
148 | const struct address_space_operations pvfs2_address_operations = { | |
149 | .readpage = pvfs2_readpage, | |
150 | .readpages = pvfs2_readpages, | |
151 | .invalidatepage = pvfs2_invalidatepage, | |
152 | .releasepage = pvfs2_releasepage, | |
153 | /* .direct_IO = pvfs2_direct_IO */ | |
154 | }; | |
155 | ||
156 | static int pvfs2_setattr_size(struct inode *inode, struct iattr *iattr) | |
157 | { | |
158 | struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode); | |
159 | struct pvfs2_kernel_op_s *new_op; | |
160 | loff_t orig_size = i_size_read(inode); | |
161 | int ret = -EINVAL; | |
162 | ||
163 | gossip_debug(GOSSIP_INODE_DEBUG, | |
164 | "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", | |
165 | __func__, | |
166 | get_khandle_from_ino(inode), | |
167 | &pvfs2_inode->refn.khandle, | |
168 | pvfs2_inode->refn.fs_id, | |
169 | iattr->ia_size); | |
170 | ||
171 | truncate_setsize(inode, iattr->ia_size); | |
172 | ||
173 | new_op = op_alloc(PVFS2_VFS_OP_TRUNCATE); | |
174 | if (!new_op) | |
175 | return -ENOMEM; | |
176 | ||
177 | new_op->upcall.req.truncate.refn = pvfs2_inode->refn; | |
178 | new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; | |
179 | ||
180 | ret = service_operation(new_op, __func__, | |
181 | get_interruptible_flag(inode)); | |
182 | ||
183 | /* | |
184 | * the truncate has no downcall members to retrieve, but | |
185 | * the status value tells us if it went through ok or not | |
186 | */ | |
187 | gossip_debug(GOSSIP_INODE_DEBUG, | |
188 | "pvfs2: pvfs2_truncate got return value of %d\n", | |
189 | ret); | |
190 | ||
191 | op_release(new_op); | |
192 | ||
193 | if (ret != 0) | |
194 | return ret; | |
195 | ||
196 | /* | |
197 | * Only change the c/mtime if we are changing the size or we are | |
198 | * explicitly asked to change it. This handles the semantic difference | |
199 | * between truncate() and ftruncate() as implemented in the VFS. | |
200 | * | |
201 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
202 | * special case where we need to update the times despite not having | |
203 | * these flags set. For all other operations the VFS set these flags | |
204 | * explicitly if it wants a timestamp update. | |
205 | */ | |
206 | if (orig_size != i_size_read(inode) && | |
207 | !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { | |
208 | iattr->ia_ctime = iattr->ia_mtime = | |
209 | current_fs_time(inode->i_sb); | |
210 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; | |
211 | } | |
212 | ||
213 | return ret; | |
214 | } | |
215 | ||
216 | /* | |
217 | * Change attributes of an object referenced by dentry. | |
218 | */ | |
219 | int pvfs2_setattr(struct dentry *dentry, struct iattr *iattr) | |
220 | { | |
221 | int ret = -EINVAL; | |
222 | struct inode *inode = dentry->d_inode; | |
223 | ||
224 | gossip_debug(GOSSIP_INODE_DEBUG, | |
225 | "pvfs2_setattr: called on %s\n", | |
226 | dentry->d_name.name); | |
227 | ||
228 | ret = inode_change_ok(inode, iattr); | |
229 | if (ret) | |
230 | goto out; | |
231 | ||
232 | if ((iattr->ia_valid & ATTR_SIZE) && | |
233 | iattr->ia_size != i_size_read(inode)) { | |
234 | ret = pvfs2_setattr_size(inode, iattr); | |
235 | if (ret) | |
236 | goto out; | |
237 | } | |
238 | ||
239 | setattr_copy(inode, iattr); | |
240 | mark_inode_dirty(inode); | |
241 | ||
242 | ret = pvfs2_inode_setattr(inode, iattr); | |
243 | gossip_debug(GOSSIP_INODE_DEBUG, | |
244 | "pvfs2_setattr: inode_setattr returned %d\n", | |
245 | ret); | |
246 | ||
247 | if (!ret && (iattr->ia_valid & ATTR_MODE)) | |
248 | /* change mod on a file that has ACLs */ | |
249 | ret = posix_acl_chmod(inode, inode->i_mode); | |
250 | ||
251 | out: | |
252 | gossip_debug(GOSSIP_INODE_DEBUG, "pvfs2_setattr: returning %d\n", ret); | |
253 | return ret; | |
254 | } | |
255 | ||
256 | /* | |
257 | * Obtain attributes of an object given a dentry | |
258 | */ | |
259 | int pvfs2_getattr(struct vfsmount *mnt, | |
260 | struct dentry *dentry, | |
261 | struct kstat *kstat) | |
262 | { | |
263 | int ret = -ENOENT; | |
264 | struct inode *inode = dentry->d_inode; | |
265 | struct pvfs2_inode_s *pvfs2_inode = NULL; | |
266 | ||
267 | gossip_debug(GOSSIP_INODE_DEBUG, | |
268 | "pvfs2_getattr: called on %s\n", | |
269 | dentry->d_name.name); | |
270 | ||
271 | /* | |
272 | * Similar to the above comment, a getattr also expects that all | |
273 | * fields/attributes of the inode would be refreshed. So again, we | |
274 | * dont have too much of a choice but refresh all the attributes. | |
275 | */ | |
276 | ret = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); | |
277 | if (ret == 0) { | |
278 | generic_fillattr(inode, kstat); | |
279 | /* override block size reported to stat */ | |
280 | pvfs2_inode = PVFS2_I(inode); | |
281 | kstat->blksize = pvfs2_inode->blksize; | |
282 | } else { | |
283 | /* assume an I/O error and flag inode as bad */ | |
284 | gossip_debug(GOSSIP_INODE_DEBUG, | |
285 | "%s:%s:%d calling make bad inode\n", | |
286 | __FILE__, | |
287 | __func__, | |
288 | __LINE__); | |
289 | pvfs2_make_bad_inode(inode); | |
290 | } | |
291 | return ret; | |
292 | } | |
293 | ||
294 | /* PVFS2 implementation of VFS inode operations for files */ | |
295 | struct inode_operations pvfs2_file_inode_operations = { | |
296 | .get_acl = pvfs2_get_acl, | |
297 | .set_acl = pvfs2_set_acl, | |
298 | .setattr = pvfs2_setattr, | |
299 | .getattr = pvfs2_getattr, | |
300 | .setxattr = generic_setxattr, | |
301 | .getxattr = generic_getxattr, | |
302 | .listxattr = pvfs2_listxattr, | |
303 | .removexattr = generic_removexattr, | |
304 | }; | |
305 | ||
306 | static int pvfs2_init_iops(struct inode *inode) | |
307 | { | |
308 | inode->i_mapping->a_ops = &pvfs2_address_operations; | |
309 | ||
310 | switch (inode->i_mode & S_IFMT) { | |
311 | case S_IFREG: | |
312 | inode->i_op = &pvfs2_file_inode_operations; | |
313 | inode->i_fop = &pvfs2_file_operations; | |
314 | inode->i_blkbits = PAGE_CACHE_SHIFT; | |
315 | break; | |
316 | case S_IFLNK: | |
317 | inode->i_op = &pvfs2_symlink_inode_operations; | |
318 | break; | |
319 | case S_IFDIR: | |
320 | inode->i_op = &pvfs2_dir_inode_operations; | |
321 | inode->i_fop = &pvfs2_dir_operations; | |
322 | break; | |
323 | default: | |
324 | gossip_debug(GOSSIP_INODE_DEBUG, | |
325 | "%s: unsupported mode\n", | |
326 | __func__); | |
327 | return -EINVAL; | |
328 | } | |
329 | ||
330 | return 0; | |
331 | } | |
332 | ||
333 | /* | |
334 | * Given a PVFS2 object identifier (fsid, handle), convert it into a ino_t type | |
335 | * that will be used as a hash-index from where the handle will | |
336 | * be searched for in the VFS hash table of inodes. | |
337 | */ | |
338 | static inline ino_t pvfs2_handle_hash(struct pvfs2_object_kref *ref) | |
339 | { | |
340 | if (!ref) | |
341 | return 0; | |
342 | return pvfs2_khandle_to_ino(&(ref->khandle)); | |
343 | } | |
344 | ||
345 | /* | |
346 | * Called to set up an inode from iget5_locked. | |
347 | */ | |
348 | static int pvfs2_set_inode(struct inode *inode, void *data) | |
349 | { | |
350 | struct pvfs2_object_kref *ref = (struct pvfs2_object_kref *) data; | |
351 | struct pvfs2_inode_s *pvfs2_inode = NULL; | |
352 | ||
353 | /* Make sure that we have sane parameters */ | |
354 | if (!data || !inode) | |
355 | return 0; | |
356 | pvfs2_inode = PVFS2_I(inode); | |
357 | if (!pvfs2_inode) | |
358 | return 0; | |
359 | pvfs2_inode->refn.fs_id = ref->fs_id; | |
360 | pvfs2_inode->refn.khandle = ref->khandle; | |
361 | return 0; | |
362 | } | |
363 | ||
364 | /* | |
365 | * Called to determine if handles match. | |
366 | */ | |
367 | static int pvfs2_test_inode(struct inode *inode, void *data) | |
368 | { | |
369 | struct pvfs2_object_kref *ref = (struct pvfs2_object_kref *) data; | |
370 | struct pvfs2_inode_s *pvfs2_inode = NULL; | |
371 | ||
372 | pvfs2_inode = PVFS2_I(inode); | |
373 | return (!PVFS_khandle_cmp(&(pvfs2_inode->refn.khandle), &(ref->khandle)) | |
374 | && pvfs2_inode->refn.fs_id == ref->fs_id); | |
375 | } | |
376 | ||
377 | /* | |
378 | * Front-end to lookup the inode-cache maintained by the VFS using the PVFS2 | |
379 | * file handle. | |
380 | * | |
381 | * @sb: the file system super block instance. | |
382 | * @ref: The PVFS2 object for which we are trying to locate an inode structure. | |
383 | */ | |
384 | struct inode *pvfs2_iget(struct super_block *sb, struct pvfs2_object_kref *ref) | |
385 | { | |
386 | struct inode *inode = NULL; | |
387 | unsigned long hash; | |
388 | int error; | |
389 | ||
390 | hash = pvfs2_handle_hash(ref); | |
391 | inode = iget5_locked(sb, hash, pvfs2_test_inode, pvfs2_set_inode, ref); | |
392 | if (!inode || !(inode->i_state & I_NEW)) | |
393 | return inode; | |
394 | ||
395 | error = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); | |
396 | if (error) { | |
397 | iget_failed(inode); | |
398 | return ERR_PTR(error); | |
399 | } | |
400 | ||
401 | inode->i_ino = hash; /* needed for stat etc */ | |
402 | pvfs2_init_iops(inode); | |
403 | unlock_new_inode(inode); | |
404 | ||
405 | gossip_debug(GOSSIP_INODE_DEBUG, | |
406 | "iget handle %pU, fsid %d hash %ld i_ino %lu\n", | |
407 | &ref->khandle, | |
408 | ref->fs_id, | |
409 | hash, | |
410 | inode->i_ino); | |
411 | ||
412 | return inode; | |
413 | } | |
414 | ||
415 | /* | |
416 | * Allocate an inode for a newly created file and insert it into the inode hash. | |
417 | */ | |
418 | struct inode *pvfs2_new_inode(struct super_block *sb, struct inode *dir, | |
419 | int mode, dev_t dev, struct pvfs2_object_kref *ref) | |
420 | { | |
421 | unsigned long hash = pvfs2_handle_hash(ref); | |
422 | struct inode *inode; | |
423 | int error; | |
424 | ||
425 | gossip_debug(GOSSIP_INODE_DEBUG, | |
426 | "pvfs2_get_custom_inode_common: called\n" | |
427 | "(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", | |
428 | sb, | |
429 | MAJOR(dev), | |
430 | MINOR(dev), | |
431 | mode); | |
432 | ||
433 | inode = new_inode(sb); | |
434 | if (!inode) | |
435 | return NULL; | |
436 | ||
437 | pvfs2_set_inode(inode, ref); | |
438 | inode->i_ino = hash; /* needed for stat etc */ | |
439 | ||
440 | error = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_ALL_NOHINT); | |
441 | if (error) | |
442 | goto out_iput; | |
443 | ||
444 | pvfs2_init_iops(inode); | |
445 | ||
446 | inode->i_mode = mode; | |
447 | inode->i_uid = current_fsuid(); | |
448 | inode->i_gid = current_fsgid(); | |
449 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
450 | inode->i_size = PAGE_CACHE_SIZE; | |
451 | inode->i_rdev = dev; | |
452 | ||
453 | error = insert_inode_locked4(inode, hash, pvfs2_test_inode, ref); | |
454 | if (error < 0) | |
455 | goto out_iput; | |
456 | ||
457 | gossip_debug(GOSSIP_INODE_DEBUG, | |
458 | "Initializing ACL's for inode %pU\n", | |
459 | get_khandle_from_ino(inode)); | |
460 | pvfs2_init_acl(inode, dir); | |
461 | return inode; | |
462 | ||
463 | out_iput: | |
464 | iput(inode); | |
465 | return ERR_PTR(error); | |
466 | } |