xref: /openbmc/linux/fs/orangefs/inode.c (revision fa7f32422ea1ac276b45b96a540ed5981caaa61f)
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 "orangefs-kernel.h"
13 #include "orangefs-bufmap.h"
14 
15 static int read_one_page(struct page *page)
16 {
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_SIZE;	/* inode->i_blksize */
22 	const __u32 blockbits = PAGE_SHIFT;	/* inode->i_blkbits */
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);
27 
28 	gossip_debug(GOSSIP_INODE_DEBUG,
29 		    "orangefs_readpage called with page %p\n",
30 		     page);
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 = orangefs_inode_read(inode,
38 						 &to,
39 						 &blockptr_offset,
40 						 inode->i_size);
41 	}
42 	/* this will only zero remaining unread portions of the page data */
43 	iov_iter_zero(~0U, &to);
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 	}
55 	/* unlock the page after the ->readpage() routine completes */
56 	unlock_page(page);
57 	return ret;
58 }
59 
60 static int orangefs_readpage(struct file *file, struct page *page)
61 {
62 	return read_one_page(page);
63 }
64 
65 static int orangefs_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, "orangefs_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 				       readahead_gfp_mask(mapping))) {
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 			put_page(page);
90 	      }
91 	}
92 	BUG_ON(!list_empty(pages));
93 	return 0;
94 }
95 
96 static void orangefs_invalidatepage(struct page *page,
97 				 unsigned int offset,
98 				 unsigned int length)
99 {
100 	gossip_debug(GOSSIP_INODE_DEBUG,
101 		     "orangefs_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 orangefs_releasepage(struct page *page, gfp_t foo)
113 {
114 	gossip_debug(GOSSIP_INODE_DEBUG,
115 		     "orangefs_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 
128 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
129 				  struct iov_iter *iter)
130 {
131 	gossip_debug(GOSSIP_INODE_DEBUG,
132 		     "orangefs_direct_IO: %pD\n",
133 		     iocb->ki_filp);
134 
135 	return -EINVAL;
136 }
137 
138 struct backing_dev_info orangefs_backing_dev_info = {
139 	.name = "orangefs",
140 	.ra_pages = 0,
141 	.capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
142 };
143 
144 /** ORANGEFS2 implementation of address space operations */
145 const struct address_space_operations orangefs_address_operations = {
146 	.readpage = orangefs_readpage,
147 	.readpages = orangefs_readpages,
148 	.invalidatepage = orangefs_invalidatepage,
149 	.releasepage = orangefs_releasepage,
150 	.direct_IO = orangefs_direct_IO,
151 };
152 
153 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
154 {
155 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
156 	struct orangefs_kernel_op_s *new_op;
157 	loff_t orig_size;
158 	int ret = -EINVAL;
159 
160 	gossip_debug(GOSSIP_INODE_DEBUG,
161 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
162 		     __func__,
163 		     get_khandle_from_ino(inode),
164 		     &orangefs_inode->refn.khandle,
165 		     orangefs_inode->refn.fs_id,
166 		     iattr->ia_size);
167 
168 	/* Ensure that we have a up to date size, so we know if it changed. */
169 	ret = orangefs_inode_getattr(inode, 0, 1);
170 	if (ret == -ESTALE)
171 		ret = -EIO;
172 	if (ret) {
173 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
174 		    __func__, ret);
175 		return ret;
176 	}
177 	orig_size = i_size_read(inode);
178 
179 	truncate_setsize(inode, iattr->ia_size);
180 
181 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
182 	if (!new_op)
183 		return -ENOMEM;
184 
185 	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
186 	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
187 
188 	ret = service_operation(new_op, __func__,
189 				get_interruptible_flag(inode));
190 
191 	/*
192 	 * the truncate has no downcall members to retrieve, but
193 	 * the status value tells us if it went through ok or not
194 	 */
195 	gossip_debug(GOSSIP_INODE_DEBUG,
196 		     "orangefs: orangefs_truncate got return value of %d\n",
197 		     ret);
198 
199 	op_release(new_op);
200 
201 	if (ret != 0)
202 		return ret;
203 
204 	if (orig_size != i_size_read(inode))
205 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
206 
207 	return ret;
208 }
209 
210 /*
211  * Change attributes of an object referenced by dentry.
212  */
213 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
214 {
215 	int ret = -EINVAL;
216 	struct inode *inode = dentry->d_inode;
217 
218 	gossip_debug(GOSSIP_INODE_DEBUG,
219 		     "orangefs_setattr: called on %pd\n",
220 		     dentry);
221 
222 	ret = setattr_prepare(dentry, iattr);
223 	if (ret)
224 		goto out;
225 
226 	if ((iattr->ia_valid & ATTR_SIZE) &&
227 	    iattr->ia_size != i_size_read(inode)) {
228 		ret = orangefs_setattr_size(inode, iattr);
229 		if (ret)
230 			goto out;
231 	}
232 
233 	setattr_copy(inode, iattr);
234 	mark_inode_dirty(inode);
235 
236 	ret = orangefs_inode_setattr(inode, iattr);
237 	gossip_debug(GOSSIP_INODE_DEBUG,
238 		     "orangefs_setattr: inode_setattr returned %d\n",
239 		     ret);
240 
241 	if (!ret && (iattr->ia_valid & ATTR_MODE))
242 		/* change mod on a file that has ACLs */
243 		ret = posix_acl_chmod(inode, inode->i_mode);
244 
245 out:
246 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", ret);
247 	return ret;
248 }
249 
250 /*
251  * Obtain attributes of an object given a dentry
252  */
253 int orangefs_getattr(struct vfsmount *mnt,
254 		  struct dentry *dentry,
255 		  struct kstat *kstat)
256 {
257 	int ret = -ENOENT;
258 	struct inode *inode = dentry->d_inode;
259 	struct orangefs_inode_s *orangefs_inode = NULL;
260 
261 	gossip_debug(GOSSIP_INODE_DEBUG,
262 		     "orangefs_getattr: called on %pd\n",
263 		     dentry);
264 
265 	ret = orangefs_inode_getattr(inode, 0, 0);
266 	if (ret == 0) {
267 		generic_fillattr(inode, kstat);
268 
269 		/* override block size reported to stat */
270 		orangefs_inode = ORANGEFS_I(inode);
271 		kstat->blksize = orangefs_inode->blksize;
272 	}
273 	return ret;
274 }
275 
276 int orangefs_permission(struct inode *inode, int mask)
277 {
278 	int ret;
279 
280 	if (mask & MAY_NOT_BLOCK)
281 		return -ECHILD;
282 
283 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
284 
285 	/* Make sure the permission (and other common attrs) are up to date. */
286 	ret = orangefs_inode_getattr(inode, 0, 0);
287 	if (ret < 0)
288 		return ret;
289 
290 	return generic_permission(inode, mask);
291 }
292 
293 /* ORANGEDS2 implementation of VFS inode operations for files */
294 const struct inode_operations orangefs_file_inode_operations = {
295 	.get_acl = orangefs_get_acl,
296 	.set_acl = orangefs_set_acl,
297 	.setattr = orangefs_setattr,
298 	.getattr = orangefs_getattr,
299 	.listxattr = orangefs_listxattr,
300 	.permission = orangefs_permission,
301 };
302 
303 static int orangefs_init_iops(struct inode *inode)
304 {
305 	inode->i_mapping->a_ops = &orangefs_address_operations;
306 
307 	switch (inode->i_mode & S_IFMT) {
308 	case S_IFREG:
309 		inode->i_op = &orangefs_file_inode_operations;
310 		inode->i_fop = &orangefs_file_operations;
311 		inode->i_blkbits = PAGE_SHIFT;
312 		break;
313 	case S_IFLNK:
314 		inode->i_op = &orangefs_symlink_inode_operations;
315 		break;
316 	case S_IFDIR:
317 		inode->i_op = &orangefs_dir_inode_operations;
318 		inode->i_fop = &orangefs_dir_operations;
319 		break;
320 	default:
321 		gossip_debug(GOSSIP_INODE_DEBUG,
322 			     "%s: unsupported mode\n",
323 			     __func__);
324 		return -EINVAL;
325 	}
326 
327 	return 0;
328 }
329 
330 /*
331  * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
332  * that will be used as a hash-index from where the handle will
333  * be searched for in the VFS hash table of inodes.
334  */
335 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
336 {
337 	if (!ref)
338 		return 0;
339 	return orangefs_khandle_to_ino(&(ref->khandle));
340 }
341 
342 /*
343  * Called to set up an inode from iget5_locked.
344  */
345 static int orangefs_set_inode(struct inode *inode, void *data)
346 {
347 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
348 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
349 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
350 	return 0;
351 }
352 
353 /*
354  * Called to determine if handles match.
355  */
356 static int orangefs_test_inode(struct inode *inode, void *data)
357 {
358 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
359 	struct orangefs_inode_s *orangefs_inode = NULL;
360 
361 	orangefs_inode = ORANGEFS_I(inode);
362 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle))
363 		&& orangefs_inode->refn.fs_id == ref->fs_id);
364 }
365 
366 /*
367  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
368  * file handle.
369  *
370  * @sb: the file system super block instance.
371  * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
372  */
373 struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref)
374 {
375 	struct inode *inode = NULL;
376 	unsigned long hash;
377 	int error;
378 
379 	hash = orangefs_handle_hash(ref);
380 	inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref);
381 	if (!inode || !(inode->i_state & I_NEW))
382 		return inode;
383 
384 	error = orangefs_inode_getattr(inode, 1, 1);
385 	if (error) {
386 		iget_failed(inode);
387 		return ERR_PTR(error);
388 	}
389 
390 	inode->i_ino = hash;	/* needed for stat etc */
391 	orangefs_init_iops(inode);
392 	unlock_new_inode(inode);
393 
394 	gossip_debug(GOSSIP_INODE_DEBUG,
395 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
396 		     &ref->khandle,
397 		     ref->fs_id,
398 		     hash,
399 		     inode->i_ino);
400 
401 	return inode;
402 }
403 
404 /*
405  * Allocate an inode for a newly created file and insert it into the inode hash.
406  */
407 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
408 		int mode, dev_t dev, struct orangefs_object_kref *ref)
409 {
410 	unsigned long hash = orangefs_handle_hash(ref);
411 	struct inode *inode;
412 	int error;
413 
414 	gossip_debug(GOSSIP_INODE_DEBUG,
415 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
416 		     __func__,
417 		     sb,
418 		     MAJOR(dev),
419 		     MINOR(dev),
420 		     mode);
421 
422 	inode = new_inode(sb);
423 	if (!inode)
424 		return NULL;
425 
426 	orangefs_set_inode(inode, ref);
427 	inode->i_ino = hash;	/* needed for stat etc */
428 
429 	error = orangefs_inode_getattr(inode, 1, 1);
430 	if (error)
431 		goto out_iput;
432 
433 	orangefs_init_iops(inode);
434 
435 	inode->i_mode = mode;
436 	inode->i_uid = current_fsuid();
437 	inode->i_gid = current_fsgid();
438 	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
439 	inode->i_size = PAGE_SIZE;
440 	inode->i_rdev = dev;
441 
442 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
443 	if (error < 0)
444 		goto out_iput;
445 
446 	gossip_debug(GOSSIP_INODE_DEBUG,
447 		     "Initializing ACL's for inode %pU\n",
448 		     get_khandle_from_ino(inode));
449 	orangefs_init_acl(inode, dir);
450 	return inode;
451 
452 out_iput:
453 	iput(inode);
454 	return ERR_PTR(error);
455 }
456