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 | ||
2f8b5444 | 11 | #include <linux/bvec.h> |
5db11c21 | 12 | #include "protocol.h" |
575e9461 MM |
13 | #include "orangefs-kernel.h" |
14 | #include "orangefs-bufmap.h" | |
5db11c21 MM |
15 | |
16 | static int read_one_page(struct page *page) | |
17 | { | |
5db11c21 MM |
18 | int ret; |
19 | int max_block; | |
20 | ssize_t bytes_read = 0; | |
21 | struct inode *inode = page->mapping->host; | |
09cbfeaf KS |
22 | const __u32 blocksize = PAGE_SIZE; /* inode->i_blksize */ |
23 | const __u32 blockbits = PAGE_SHIFT; /* inode->i_blkbits */ | |
74f68fce AV |
24 | struct iov_iter to; |
25 | struct bio_vec bv = {.bv_page = page, .bv_len = PAGE_SIZE}; | |
26 | ||
27 | iov_iter_bvec(&to, ITER_BVEC | READ, &bv, 1, PAGE_SIZE); | |
5db11c21 MM |
28 | |
29 | gossip_debug(GOSSIP_INODE_DEBUG, | |
8bb8aefd | 30 | "orangefs_readpage called with page %p\n", |
5db11c21 | 31 | page); |
5db11c21 MM |
32 | |
33 | max_block = ((inode->i_size / blocksize) + 1); | |
34 | ||
35 | if (page->index < max_block) { | |
36 | loff_t blockptr_offset = (((loff_t) page->index) << blockbits); | |
37 | ||
8bb8aefd YL |
38 | bytes_read = orangefs_inode_read(inode, |
39 | &to, | |
40 | &blockptr_offset, | |
41 | inode->i_size); | |
5db11c21 | 42 | } |
74f68fce AV |
43 | /* this will only zero remaining unread portions of the page data */ |
44 | iov_iter_zero(~0U, &to); | |
5db11c21 MM |
45 | /* takes care of potential aliasing */ |
46 | flush_dcache_page(page); | |
47 | if (bytes_read < 0) { | |
48 | ret = bytes_read; | |
49 | SetPageError(page); | |
50 | } else { | |
51 | SetPageUptodate(page); | |
52 | if (PageError(page)) | |
53 | ClearPageError(page); | |
54 | ret = 0; | |
55 | } | |
5db11c21 MM |
56 | /* unlock the page after the ->readpage() routine completes */ |
57 | unlock_page(page); | |
58 | return ret; | |
59 | } | |
60 | ||
8bb8aefd | 61 | static int orangefs_readpage(struct file *file, struct page *page) |
5db11c21 MM |
62 | { |
63 | return read_one_page(page); | |
64 | } | |
65 | ||
8bb8aefd | 66 | static int orangefs_readpages(struct file *file, |
5db11c21 MM |
67 | struct address_space *mapping, |
68 | struct list_head *pages, | |
69 | unsigned nr_pages) | |
70 | { | |
71 | int page_idx; | |
72 | int ret; | |
73 | ||
8bb8aefd | 74 | gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_readpages called\n"); |
5db11c21 MM |
75 | |
76 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { | |
77 | struct page *page; | |
78 | ||
79 | page = list_entry(pages->prev, struct page, lru); | |
80 | list_del(&page->lru); | |
81 | if (!add_to_page_cache(page, | |
82 | mapping, | |
83 | page->index, | |
8a5c743e | 84 | readahead_gfp_mask(mapping))) { |
5db11c21 MM |
85 | ret = read_one_page(page); |
86 | gossip_debug(GOSSIP_INODE_DEBUG, | |
87 | "failure adding page to cache, read_one_page returned: %d\n", | |
88 | ret); | |
89 | } else { | |
09cbfeaf | 90 | put_page(page); |
5db11c21 MM |
91 | } |
92 | } | |
93 | BUG_ON(!list_empty(pages)); | |
94 | return 0; | |
95 | } | |
96 | ||
8bb8aefd | 97 | static void orangefs_invalidatepage(struct page *page, |
5db11c21 MM |
98 | unsigned int offset, |
99 | unsigned int length) | |
100 | { | |
101 | gossip_debug(GOSSIP_INODE_DEBUG, | |
8bb8aefd | 102 | "orangefs_invalidatepage called on page %p " |
5db11c21 MM |
103 | "(offset is %u)\n", |
104 | page, | |
105 | offset); | |
106 | ||
107 | ClearPageUptodate(page); | |
108 | ClearPageMappedToDisk(page); | |
109 | return; | |
110 | ||
111 | } | |
112 | ||
8bb8aefd | 113 | static int orangefs_releasepage(struct page *page, gfp_t foo) |
5db11c21 MM |
114 | { |
115 | gossip_debug(GOSSIP_INODE_DEBUG, | |
8bb8aefd | 116 | "orangefs_releasepage called on page %p\n", |
5db11c21 MM |
117 | page); |
118 | return 0; | |
119 | } | |
120 | ||
121 | /* | |
122 | * Having a direct_IO entry point in the address_space_operations | |
123 | * struct causes the kernel to allows us to use O_DIRECT on | |
124 | * open. Nothing will ever call this thing, but in the future we | |
125 | * will need to be able to use O_DIRECT on open in order to support | |
126 | * AIO. Modeled after NFS, they do this too. | |
127 | */ | |
3903f150 MM |
128 | |
129 | static ssize_t orangefs_direct_IO(struct kiocb *iocb, | |
130 | struct iov_iter *iter) | |
131 | { | |
132 | gossip_debug(GOSSIP_INODE_DEBUG, | |
f66debf1 AV |
133 | "orangefs_direct_IO: %pD\n", |
134 | iocb->ki_filp); | |
3903f150 MM |
135 | |
136 | return -EINVAL; | |
137 | } | |
5db11c21 | 138 | |
8bb8aefd YL |
139 | struct backing_dev_info orangefs_backing_dev_info = { |
140 | .name = "orangefs", | |
5db11c21 MM |
141 | .ra_pages = 0, |
142 | .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, | |
143 | }; | |
144 | ||
8bb8aefd YL |
145 | /** ORANGEFS2 implementation of address space operations */ |
146 | const struct address_space_operations orangefs_address_operations = { | |
147 | .readpage = orangefs_readpage, | |
148 | .readpages = orangefs_readpages, | |
149 | .invalidatepage = orangefs_invalidatepage, | |
150 | .releasepage = orangefs_releasepage, | |
3903f150 | 151 | .direct_IO = orangefs_direct_IO, |
5db11c21 MM |
152 | }; |
153 | ||
8bb8aefd | 154 | static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr) |
5db11c21 | 155 | { |
8bb8aefd YL |
156 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
157 | struct orangefs_kernel_op_s *new_op; | |
fecd86aa | 158 | loff_t orig_size; |
5db11c21 MM |
159 | int ret = -EINVAL; |
160 | ||
161 | gossip_debug(GOSSIP_INODE_DEBUG, | |
162 | "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", | |
163 | __func__, | |
164 | get_khandle_from_ino(inode), | |
8bb8aefd YL |
165 | &orangefs_inode->refn.khandle, |
166 | orangefs_inode->refn.fs_id, | |
5db11c21 MM |
167 | iattr->ia_size); |
168 | ||
fecd86aa MB |
169 | /* Ensure that we have a up to date size, so we know if it changed. */ |
170 | ret = orangefs_inode_getattr(inode, 0, 1); | |
171 | if (ret == -ESTALE) | |
172 | ret = -EIO; | |
173 | if (ret) { | |
174 | gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n", | |
175 | __func__, ret); | |
176 | return ret; | |
177 | } | |
178 | orig_size = i_size_read(inode); | |
179 | ||
5db11c21 MM |
180 | truncate_setsize(inode, iattr->ia_size); |
181 | ||
8bb8aefd | 182 | new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE); |
5db11c21 MM |
183 | if (!new_op) |
184 | return -ENOMEM; | |
185 | ||
8bb8aefd | 186 | new_op->upcall.req.truncate.refn = orangefs_inode->refn; |
5db11c21 MM |
187 | new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; |
188 | ||
189 | ret = service_operation(new_op, __func__, | |
190 | get_interruptible_flag(inode)); | |
191 | ||
192 | /* | |
193 | * the truncate has no downcall members to retrieve, but | |
194 | * the status value tells us if it went through ok or not | |
195 | */ | |
196 | gossip_debug(GOSSIP_INODE_DEBUG, | |
8bb8aefd | 197 | "orangefs: orangefs_truncate got return value of %d\n", |
5db11c21 MM |
198 | ret); |
199 | ||
200 | op_release(new_op); | |
201 | ||
202 | if (ret != 0) | |
203 | return ret; | |
204 | ||
f83140c1 | 205 | if (orig_size != i_size_read(inode)) |
5db11c21 | 206 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; |
5db11c21 MM |
207 | |
208 | return ret; | |
209 | } | |
210 | ||
211 | /* | |
212 | * Change attributes of an object referenced by dentry. | |
213 | */ | |
8bb8aefd | 214 | int orangefs_setattr(struct dentry *dentry, struct iattr *iattr) |
5db11c21 MM |
215 | { |
216 | int ret = -EINVAL; | |
217 | struct inode *inode = dentry->d_inode; | |
218 | ||
219 | gossip_debug(GOSSIP_INODE_DEBUG, | |
f66debf1 AV |
220 | "orangefs_setattr: called on %pd\n", |
221 | dentry); | |
5db11c21 | 222 | |
31051c85 | 223 | ret = setattr_prepare(dentry, iattr); |
5db11c21 MM |
224 | if (ret) |
225 | goto out; | |
226 | ||
227 | if ((iattr->ia_valid & ATTR_SIZE) && | |
228 | iattr->ia_size != i_size_read(inode)) { | |
8bb8aefd | 229 | ret = orangefs_setattr_size(inode, iattr); |
5db11c21 MM |
230 | if (ret) |
231 | goto out; | |
232 | } | |
233 | ||
234 | setattr_copy(inode, iattr); | |
235 | mark_inode_dirty(inode); | |
236 | ||
8bb8aefd | 237 | ret = orangefs_inode_setattr(inode, iattr); |
5db11c21 | 238 | gossip_debug(GOSSIP_INODE_DEBUG, |
8bb8aefd | 239 | "orangefs_setattr: inode_setattr returned %d\n", |
5db11c21 MM |
240 | ret); |
241 | ||
242 | if (!ret && (iattr->ia_valid & ATTR_MODE)) | |
243 | /* change mod on a file that has ACLs */ | |
244 | ret = posix_acl_chmod(inode, inode->i_mode); | |
245 | ||
246 | out: | |
8bb8aefd | 247 | gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", ret); |
5db11c21 MM |
248 | return ret; |
249 | } | |
250 | ||
251 | /* | |
252 | * Obtain attributes of an object given a dentry | |
253 | */ | |
8bb8aefd | 254 | int orangefs_getattr(struct vfsmount *mnt, |
5db11c21 MM |
255 | struct dentry *dentry, |
256 | struct kstat *kstat) | |
257 | { | |
258 | int ret = -ENOENT; | |
259 | struct inode *inode = dentry->d_inode; | |
8bb8aefd | 260 | struct orangefs_inode_s *orangefs_inode = NULL; |
5db11c21 MM |
261 | |
262 | gossip_debug(GOSSIP_INODE_DEBUG, | |
f66debf1 AV |
263 | "orangefs_getattr: called on %pd\n", |
264 | dentry); | |
5db11c21 | 265 | |
71680c18 | 266 | ret = orangefs_inode_getattr(inode, 0, 0); |
5db11c21 MM |
267 | if (ret == 0) { |
268 | generic_fillattr(inode, kstat); | |
a7d3e78a | 269 | |
5db11c21 | 270 | /* override block size reported to stat */ |
8bb8aefd YL |
271 | orangefs_inode = ORANGEFS_I(inode); |
272 | kstat->blksize = orangefs_inode->blksize; | |
5db11c21 MM |
273 | } |
274 | return ret; | |
275 | } | |
276 | ||
933287da MB |
277 | int orangefs_permission(struct inode *inode, int mask) |
278 | { | |
279 | int ret; | |
280 | ||
281 | if (mask & MAY_NOT_BLOCK) | |
282 | return -ECHILD; | |
283 | ||
284 | gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__); | |
285 | ||
286 | /* Make sure the permission (and other common attrs) are up to date. */ | |
8f24928d | 287 | ret = orangefs_inode_getattr(inode, 0, 0); |
933287da MB |
288 | if (ret < 0) |
289 | return ret; | |
290 | ||
291 | return generic_permission(inode, mask); | |
292 | } | |
293 | ||
8bb8aefd | 294 | /* ORANGEDS2 implementation of VFS inode operations for files */ |
6f3fc107 | 295 | const struct inode_operations orangefs_file_inode_operations = { |
8bb8aefd YL |
296 | .get_acl = orangefs_get_acl, |
297 | .set_acl = orangefs_set_acl, | |
298 | .setattr = orangefs_setattr, | |
299 | .getattr = orangefs_getattr, | |
8bb8aefd | 300 | .listxattr = orangefs_listxattr, |
933287da | 301 | .permission = orangefs_permission, |
5db11c21 MM |
302 | }; |
303 | ||
8bb8aefd | 304 | static int orangefs_init_iops(struct inode *inode) |
5db11c21 | 305 | { |
8bb8aefd | 306 | inode->i_mapping->a_ops = &orangefs_address_operations; |
5db11c21 MM |
307 | |
308 | switch (inode->i_mode & S_IFMT) { | |
309 | case S_IFREG: | |
8bb8aefd YL |
310 | inode->i_op = &orangefs_file_inode_operations; |
311 | inode->i_fop = &orangefs_file_operations; | |
09cbfeaf | 312 | inode->i_blkbits = PAGE_SHIFT; |
5db11c21 MM |
313 | break; |
314 | case S_IFLNK: | |
8bb8aefd | 315 | inode->i_op = &orangefs_symlink_inode_operations; |
5db11c21 MM |
316 | break; |
317 | case S_IFDIR: | |
8bb8aefd YL |
318 | inode->i_op = &orangefs_dir_inode_operations; |
319 | inode->i_fop = &orangefs_dir_operations; | |
5db11c21 MM |
320 | break; |
321 | default: | |
322 | gossip_debug(GOSSIP_INODE_DEBUG, | |
323 | "%s: unsupported mode\n", | |
324 | __func__); | |
325 | return -EINVAL; | |
326 | } | |
327 | ||
328 | return 0; | |
329 | } | |
330 | ||
331 | /* | |
8bb8aefd | 332 | * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type |
5db11c21 MM |
333 | * that will be used as a hash-index from where the handle will |
334 | * be searched for in the VFS hash table of inodes. | |
335 | */ | |
8bb8aefd | 336 | static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref) |
5db11c21 MM |
337 | { |
338 | if (!ref) | |
339 | return 0; | |
8bb8aefd | 340 | return orangefs_khandle_to_ino(&(ref->khandle)); |
5db11c21 MM |
341 | } |
342 | ||
343 | /* | |
344 | * Called to set up an inode from iget5_locked. | |
345 | */ | |
8bb8aefd | 346 | static int orangefs_set_inode(struct inode *inode, void *data) |
5db11c21 | 347 | { |
8bb8aefd | 348 | struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; |
a4c680a0 MB |
349 | ORANGEFS_I(inode)->refn.fs_id = ref->fs_id; |
350 | ORANGEFS_I(inode)->refn.khandle = ref->khandle; | |
5db11c21 MM |
351 | return 0; |
352 | } | |
353 | ||
354 | /* | |
355 | * Called to determine if handles match. | |
356 | */ | |
8bb8aefd | 357 | static int orangefs_test_inode(struct inode *inode, void *data) |
5db11c21 | 358 | { |
8bb8aefd YL |
359 | struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; |
360 | struct orangefs_inode_s *orangefs_inode = NULL; | |
5db11c21 | 361 | |
8bb8aefd YL |
362 | orangefs_inode = ORANGEFS_I(inode); |
363 | return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle)) | |
364 | && orangefs_inode->refn.fs_id == ref->fs_id); | |
5db11c21 MM |
365 | } |
366 | ||
367 | /* | |
8bb8aefd | 368 | * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS |
5db11c21 MM |
369 | * file handle. |
370 | * | |
371 | * @sb: the file system super block instance. | |
8bb8aefd | 372 | * @ref: The ORANGEFS object for which we are trying to locate an inode structure. |
5db11c21 | 373 | */ |
8bb8aefd | 374 | struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref) |
5db11c21 MM |
375 | { |
376 | struct inode *inode = NULL; | |
377 | unsigned long hash; | |
378 | int error; | |
379 | ||
8bb8aefd YL |
380 | hash = orangefs_handle_hash(ref); |
381 | inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref); | |
5db11c21 MM |
382 | if (!inode || !(inode->i_state & I_NEW)) |
383 | return inode; | |
384 | ||
71680c18 | 385 | error = orangefs_inode_getattr(inode, 1, 1); |
5db11c21 MM |
386 | if (error) { |
387 | iget_failed(inode); | |
388 | return ERR_PTR(error); | |
389 | } | |
390 | ||
391 | inode->i_ino = hash; /* needed for stat etc */ | |
8bb8aefd | 392 | orangefs_init_iops(inode); |
5db11c21 MM |
393 | unlock_new_inode(inode); |
394 | ||
395 | gossip_debug(GOSSIP_INODE_DEBUG, | |
396 | "iget handle %pU, fsid %d hash %ld i_ino %lu\n", | |
397 | &ref->khandle, | |
398 | ref->fs_id, | |
399 | hash, | |
400 | inode->i_ino); | |
401 | ||
402 | return inode; | |
403 | } | |
404 | ||
405 | /* | |
406 | * Allocate an inode for a newly created file and insert it into the inode hash. | |
407 | */ | |
8bb8aefd YL |
408 | struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir, |
409 | int mode, dev_t dev, struct orangefs_object_kref *ref) | |
5db11c21 | 410 | { |
8bb8aefd | 411 | unsigned long hash = orangefs_handle_hash(ref); |
5db11c21 MM |
412 | struct inode *inode; |
413 | int error; | |
414 | ||
415 | gossip_debug(GOSSIP_INODE_DEBUG, | |
5253487e MM |
416 | "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", |
417 | __func__, | |
5db11c21 MM |
418 | sb, |
419 | MAJOR(dev), | |
420 | MINOR(dev), | |
421 | mode); | |
422 | ||
423 | inode = new_inode(sb); | |
424 | if (!inode) | |
425 | return NULL; | |
426 | ||
8bb8aefd | 427 | orangefs_set_inode(inode, ref); |
5db11c21 MM |
428 | inode->i_ino = hash; /* needed for stat etc */ |
429 | ||
71680c18 | 430 | error = orangefs_inode_getattr(inode, 1, 1); |
5db11c21 MM |
431 | if (error) |
432 | goto out_iput; | |
433 | ||
8bb8aefd | 434 | orangefs_init_iops(inode); |
5db11c21 MM |
435 | |
436 | inode->i_mode = mode; | |
437 | inode->i_uid = current_fsuid(); | |
438 | inode->i_gid = current_fsgid(); | |
078cd827 | 439 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
09cbfeaf | 440 | inode->i_size = PAGE_SIZE; |
5db11c21 MM |
441 | inode->i_rdev = dev; |
442 | ||
8bb8aefd | 443 | error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref); |
5db11c21 MM |
444 | if (error < 0) |
445 | goto out_iput; | |
446 | ||
447 | gossip_debug(GOSSIP_INODE_DEBUG, | |
448 | "Initializing ACL's for inode %pU\n", | |
449 | get_khandle_from_ino(inode)); | |
8bb8aefd | 450 | orangefs_init_acl(inode, dir); |
5db11c21 MM |
451 | return inode; |
452 | ||
453 | out_iput: | |
454 | iput(inode); | |
455 | return ERR_PTR(error); | |
456 | } |