blk-mq: really fix plug list flushing for nomerge queues
[linux-2.6-block.git] / fs / orangefs / inode.c
CommitLineData
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"
575e9461
MM
12#include "orangefs-kernel.h"
13#include "orangefs-bufmap.h"
5db11c21
MM
14
15static 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;
09cbfeaf
KS
21 const __u32 blocksize = PAGE_SIZE; /* inode->i_blksize */
22 const __u32 blockbits = PAGE_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,
8bb8aefd 29 "orangefs_readpage called with page %p\n",
5db11c21 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
8bb8aefd
YL
37 bytes_read = orangefs_inode_read(inode,
38 &to,
39 &blockptr_offset,
40 inode->i_size);
5db11c21 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
8bb8aefd 60static int orangefs_readpage(struct file *file, struct page *page)
5db11c21
MM
61{
62 return read_one_page(page);
63}
64
8bb8aefd 65static int orangefs_readpages(struct file *file,
5db11c21
MM
66 struct address_space *mapping,
67 struct list_head *pages,
68 unsigned nr_pages)
69{
70 int page_idx;
71 int ret;
72
8bb8aefd 73 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_readpages called\n");
5db11c21
MM
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 {
09cbfeaf 89 put_page(page);
5db11c21
MM
90 }
91 }
92 BUG_ON(!list_empty(pages));
93 return 0;
94}
95
8bb8aefd 96static void orangefs_invalidatepage(struct page *page,
5db11c21
MM
97 unsigned int offset,
98 unsigned int length)
99{
100 gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 101 "orangefs_invalidatepage called on page %p "
5db11c21
MM
102 "(offset is %u)\n",
103 page,
104 offset);
105
106 ClearPageUptodate(page);
107 ClearPageMappedToDisk(page);
108 return;
109
110}
111
8bb8aefd 112static int orangefs_releasepage(struct page *page, gfp_t foo)
5db11c21
MM
113{
114 gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 115 "orangefs_releasepage called on page %p\n",
5db11c21
MM
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/*
575e9461 128 * static ssize_t orangefs_direct_IO(int rw,
54804949
MM
129 * struct kiocb *iocb,
130 * struct iov_iter *iter,
131 * loff_t offset)
132 *{
133 * gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 134 * "orangefs_direct_IO: %s\n",
54804949
MM
135 * iocb->ki_filp->f_path.dentry->d_name.name);
136 *
137 * return -EINVAL;
138 *}
139 */
5db11c21 140
8bb8aefd
YL
141struct backing_dev_info orangefs_backing_dev_info = {
142 .name = "orangefs",
5db11c21
MM
143 .ra_pages = 0,
144 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
145};
146
8bb8aefd
YL
147/** ORANGEFS2 implementation of address space operations */
148const struct address_space_operations orangefs_address_operations = {
149 .readpage = orangefs_readpage,
150 .readpages = orangefs_readpages,
151 .invalidatepage = orangefs_invalidatepage,
152 .releasepage = orangefs_releasepage,
575e9461 153/* .direct_IO = orangefs_direct_IO */
5db11c21
MM
154};
155
8bb8aefd 156static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
5db11c21 157{
8bb8aefd
YL
158 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
159 struct orangefs_kernel_op_s *new_op;
fecd86aa 160 loff_t orig_size;
5db11c21
MM
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),
8bb8aefd
YL
167 &orangefs_inode->refn.khandle,
168 orangefs_inode->refn.fs_id,
5db11c21
MM
169 iattr->ia_size);
170
fecd86aa
MB
171 /* Ensure that we have a up to date size, so we know if it changed. */
172 ret = orangefs_inode_getattr(inode, 0, 1);
173 if (ret == -ESTALE)
174 ret = -EIO;
175 if (ret) {
176 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
177 __func__, ret);
178 return ret;
179 }
180 orig_size = i_size_read(inode);
181
5db11c21
MM
182 truncate_setsize(inode, iattr->ia_size);
183
8bb8aefd 184 new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
5db11c21
MM
185 if (!new_op)
186 return -ENOMEM;
187
8bb8aefd 188 new_op->upcall.req.truncate.refn = orangefs_inode->refn;
5db11c21
MM
189 new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
190
191 ret = service_operation(new_op, __func__,
192 get_interruptible_flag(inode));
193
194 /*
195 * the truncate has no downcall members to retrieve, but
196 * the status value tells us if it went through ok or not
197 */
198 gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 199 "orangefs: orangefs_truncate got return value of %d\n",
5db11c21
MM
200 ret);
201
202 op_release(new_op);
203
204 if (ret != 0)
205 return ret;
206
f83140c1 207 if (orig_size != i_size_read(inode))
5db11c21 208 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
5db11c21
MM
209
210 return ret;
211}
212
213/*
214 * Change attributes of an object referenced by dentry.
215 */
8bb8aefd 216int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
5db11c21
MM
217{
218 int ret = -EINVAL;
219 struct inode *inode = dentry->d_inode;
220
221 gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 222 "orangefs_setattr: called on %s\n",
5db11c21
MM
223 dentry->d_name.name);
224
225 ret = inode_change_ok(inode, iattr);
226 if (ret)
227 goto out;
228
229 if ((iattr->ia_valid & ATTR_SIZE) &&
230 iattr->ia_size != i_size_read(inode)) {
8bb8aefd 231 ret = orangefs_setattr_size(inode, iattr);
5db11c21
MM
232 if (ret)
233 goto out;
234 }
235
236 setattr_copy(inode, iattr);
237 mark_inode_dirty(inode);
238
8bb8aefd 239 ret = orangefs_inode_setattr(inode, iattr);
5db11c21 240 gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 241 "orangefs_setattr: inode_setattr returned %d\n",
5db11c21
MM
242 ret);
243
244 if (!ret && (iattr->ia_valid & ATTR_MODE))
245 /* change mod on a file that has ACLs */
246 ret = posix_acl_chmod(inode, inode->i_mode);
247
248out:
8bb8aefd 249 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", ret);
5db11c21
MM
250 return ret;
251}
252
253/*
254 * Obtain attributes of an object given a dentry
255 */
8bb8aefd 256int orangefs_getattr(struct vfsmount *mnt,
5db11c21
MM
257 struct dentry *dentry,
258 struct kstat *kstat)
259{
260 int ret = -ENOENT;
261 struct inode *inode = dentry->d_inode;
8bb8aefd 262 struct orangefs_inode_s *orangefs_inode = NULL;
5db11c21
MM
263
264 gossip_debug(GOSSIP_INODE_DEBUG,
8bb8aefd 265 "orangefs_getattr: called on %s\n",
5db11c21
MM
266 dentry->d_name.name);
267
8f24928d 268 ret = orangefs_inode_getattr(inode, 0, 1);
5db11c21
MM
269 if (ret == 0) {
270 generic_fillattr(inode, kstat);
a7d3e78a 271
5db11c21 272 /* override block size reported to stat */
8bb8aefd
YL
273 orangefs_inode = ORANGEFS_I(inode);
274 kstat->blksize = orangefs_inode->blksize;
5db11c21
MM
275 }
276 return ret;
277}
278
933287da
MB
279int orangefs_permission(struct inode *inode, int mask)
280{
281 int ret;
282
283 if (mask & MAY_NOT_BLOCK)
284 return -ECHILD;
285
286 gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
287
288 /* Make sure the permission (and other common attrs) are up to date. */
8f24928d 289 ret = orangefs_inode_getattr(inode, 0, 0);
933287da
MB
290 if (ret < 0)
291 return ret;
292
293 return generic_permission(inode, mask);
294}
295
8bb8aefd
YL
296/* ORANGEDS2 implementation of VFS inode operations for files */
297struct inode_operations orangefs_file_inode_operations = {
298 .get_acl = orangefs_get_acl,
299 .set_acl = orangefs_set_acl,
300 .setattr = orangefs_setattr,
301 .getattr = orangefs_getattr,
5db11c21
MM
302 .setxattr = generic_setxattr,
303 .getxattr = generic_getxattr,
8bb8aefd 304 .listxattr = orangefs_listxattr,
5db11c21 305 .removexattr = generic_removexattr,
933287da 306 .permission = orangefs_permission,
5db11c21
MM
307};
308
8bb8aefd 309static int orangefs_init_iops(struct inode *inode)
5db11c21 310{
8bb8aefd 311 inode->i_mapping->a_ops = &orangefs_address_operations;
5db11c21
MM
312
313 switch (inode->i_mode & S_IFMT) {
314 case S_IFREG:
8bb8aefd
YL
315 inode->i_op = &orangefs_file_inode_operations;
316 inode->i_fop = &orangefs_file_operations;
09cbfeaf 317 inode->i_blkbits = PAGE_SHIFT;
5db11c21
MM
318 break;
319 case S_IFLNK:
8bb8aefd 320 inode->i_op = &orangefs_symlink_inode_operations;
5db11c21
MM
321 break;
322 case S_IFDIR:
8bb8aefd
YL
323 inode->i_op = &orangefs_dir_inode_operations;
324 inode->i_fop = &orangefs_dir_operations;
5db11c21
MM
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/*
8bb8aefd 337 * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
5db11c21
MM
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 */
8bb8aefd 341static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
5db11c21
MM
342{
343 if (!ref)
344 return 0;
8bb8aefd 345 return orangefs_khandle_to_ino(&(ref->khandle));
5db11c21
MM
346}
347
348/*
349 * Called to set up an inode from iget5_locked.
350 */
8bb8aefd 351static int orangefs_set_inode(struct inode *inode, void *data)
5db11c21 352{
8bb8aefd 353 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
a4c680a0
MB
354 ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
355 ORANGEFS_I(inode)->refn.khandle = ref->khandle;
5db11c21
MM
356 return 0;
357}
358
359/*
360 * Called to determine if handles match.
361 */
8bb8aefd 362static int orangefs_test_inode(struct inode *inode, void *data)
5db11c21 363{
8bb8aefd
YL
364 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
365 struct orangefs_inode_s *orangefs_inode = NULL;
5db11c21 366
8bb8aefd
YL
367 orangefs_inode = ORANGEFS_I(inode);
368 return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle))
369 && orangefs_inode->refn.fs_id == ref->fs_id);
5db11c21
MM
370}
371
372/*
8bb8aefd 373 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
5db11c21
MM
374 * file handle.
375 *
376 * @sb: the file system super block instance.
8bb8aefd 377 * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
5db11c21 378 */
8bb8aefd 379struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref)
5db11c21
MM
380{
381 struct inode *inode = NULL;
382 unsigned long hash;
383 int error;
384
8bb8aefd
YL
385 hash = orangefs_handle_hash(ref);
386 inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref);
5db11c21
MM
387 if (!inode || !(inode->i_state & I_NEW))
388 return inode;
389
075cca50 390 error = orangefs_inode_getattr(inode, 1, 0);
5db11c21
MM
391 if (error) {
392 iget_failed(inode);
393 return ERR_PTR(error);
394 }
395
396 inode->i_ino = hash; /* needed for stat etc */
8bb8aefd 397 orangefs_init_iops(inode);
5db11c21
MM
398 unlock_new_inode(inode);
399
400 gossip_debug(GOSSIP_INODE_DEBUG,
401 "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
402 &ref->khandle,
403 ref->fs_id,
404 hash,
405 inode->i_ino);
406
407 return inode;
408}
409
410/*
411 * Allocate an inode for a newly created file and insert it into the inode hash.
412 */
8bb8aefd
YL
413struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
414 int mode, dev_t dev, struct orangefs_object_kref *ref)
5db11c21 415{
8bb8aefd 416 unsigned long hash = orangefs_handle_hash(ref);
5db11c21
MM
417 struct inode *inode;
418 int error;
419
420 gossip_debug(GOSSIP_INODE_DEBUG,
5253487e
MM
421 "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
422 __func__,
5db11c21
MM
423 sb,
424 MAJOR(dev),
425 MINOR(dev),
426 mode);
427
428 inode = new_inode(sb);
429 if (!inode)
430 return NULL;
431
8bb8aefd 432 orangefs_set_inode(inode, ref);
5db11c21
MM
433 inode->i_ino = hash; /* needed for stat etc */
434
075cca50 435 error = orangefs_inode_getattr(inode, 1, 0);
5db11c21
MM
436 if (error)
437 goto out_iput;
438
8bb8aefd 439 orangefs_init_iops(inode);
5db11c21
MM
440
441 inode->i_mode = mode;
442 inode->i_uid = current_fsuid();
443 inode->i_gid = current_fsgid();
444 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
09cbfeaf 445 inode->i_size = PAGE_SIZE;
5db11c21
MM
446 inode->i_rdev = dev;
447
8bb8aefd 448 error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
5db11c21
MM
449 if (error < 0)
450 goto out_iput;
451
452 gossip_debug(GOSSIP_INODE_DEBUG,
453 "Initializing ACL's for inode %pU\n",
454 get_khandle_from_ino(inode));
8bb8aefd 455 orangefs_init_acl(inode, dir);
5db11c21
MM
456 return inode;
457
458out_iput:
459 iput(inode);
460 return ERR_PTR(error);
461}