xref: /openbmc/linux/fs/orangefs/inode.c (revision 57ee11ea)
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 <linux/bvec.h>
12 #include "protocol.h"
13 #include "orangefs-kernel.h"
14 #include "orangefs-bufmap.h"
15 
16 static int read_one_page(struct page *page)
17 {
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_SIZE;	/* inode->i_blksize */
23 	const __u32 blockbits = PAGE_SHIFT;	/* inode->i_blkbits */
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);
28 
29 	gossip_debug(GOSSIP_INODE_DEBUG,
30 		    "orangefs_readpage called with page %p\n",
31 		     page);
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 
38 		bytes_read = orangefs_inode_read(inode,
39 						 &to,
40 						 &blockptr_offset,
41 						 inode->i_size);
42 	}
43 	/* this will only zero remaining unread portions of the page data */
44 	iov_iter_zero(~0U, &to);
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 	}
56 	/* unlock the page after the ->readpage() routine completes */
57 	unlock_page(page);
58 	return ret;
59 }
60 
61 static int orangefs_readpage(struct file *file, struct page *page)
62 {
63 	return read_one_page(page);
64 }
65 
66 static int orangefs_readpages(struct file *file,
67 			   struct address_space *mapping,
68 			   struct list_head *pages,
69 			   unsigned nr_pages)
70 {
71 	int page_idx;
72 	int ret;
73 
74 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_readpages called\n");
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,
84 				       readahead_gfp_mask(mapping))) {
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 {
90 			put_page(page);
91 	      }
92 	}
93 	BUG_ON(!list_empty(pages));
94 	return 0;
95 }
96 
97 static void orangefs_invalidatepage(struct page *page,
98 				 unsigned int offset,
99 				 unsigned int length)
100 {
101 	gossip_debug(GOSSIP_INODE_DEBUG,
102 		     "orangefs_invalidatepage called on page %p "
103 		     "(offset is %u)\n",
104 		     page,
105 		     offset);
106 
107 	ClearPageUptodate(page);
108 	ClearPageMappedToDisk(page);
109 	return;
110 
111 }
112 
113 static int orangefs_releasepage(struct page *page, gfp_t foo)
114 {
115 	gossip_debug(GOSSIP_INODE_DEBUG,
116 		     "orangefs_releasepage called on page %p\n",
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  */
128 
129 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
130 				  struct iov_iter *iter)
131 {
132 	gossip_debug(GOSSIP_INODE_DEBUG,
133 		     "orangefs_direct_IO: %pD\n",
134 		     iocb->ki_filp);
135 
136 	return -EINVAL;
137 }
138 
139 /** ORANGEFS2 implementation of address space operations */
140 const struct address_space_operations orangefs_address_operations = {
141 	.readpage = orangefs_readpage,
142 	.readpages = orangefs_readpages,
143 	.invalidatepage = orangefs_invalidatepage,
144 	.releasepage = orangefs_releasepage,
145 	.direct_IO = orangefs_direct_IO,
146 };
147 
148 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
149 {
150 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
151 	struct orangefs_kernel_op_s *new_op;
152 	loff_t orig_size;
153 	int ret = -EINVAL;
154 
155 	gossip_debug(GOSSIP_INODE_DEBUG,
156 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
157 		     __func__,
158 		     get_khandle_from_ino(inode),
159 		     &orangefs_inode->refn.khandle,
160 		     orangefs_inode->refn.fs_id,
161 		     iattr->ia_size);
162 
163 	/* Ensure that we have a up to date size, so we know if it changed. */
164 	ret = orangefs_inode_getattr(inode, 0, 1, STATX_SIZE);
165 	if (ret == -ESTALE)
166 		ret = -EIO;
167 	if (ret) {
168 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
169 		    __func__, ret);
170 		return ret;
171 	}
172 	orig_size = i_size_read(inode);
173 
174 	truncate_setsize(inode, iattr->ia_size);
175 
176 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
177 	if (!new_op)
178 		return -ENOMEM;
179 
180 	new_op->upcall.req.truncate.refn = orangefs_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 		     "orangefs: orangefs_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 	if (orig_size != i_size_read(inode))
200 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
201 
202 	return ret;
203 }
204 
205 /*
206  * Change attributes of an object referenced by dentry.
207  */
208 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
209 {
210 	int ret = -EINVAL;
211 	struct inode *inode = dentry->d_inode;
212 
213 	gossip_debug(GOSSIP_INODE_DEBUG,
214 		     "orangefs_setattr: called on %pd\n",
215 		     dentry);
216 
217 	ret = setattr_prepare(dentry, iattr);
218 	if (ret)
219 		goto out;
220 
221 	if (iattr->ia_valid & ATTR_SIZE) {
222 		ret = orangefs_setattr_size(inode, iattr);
223 		if (ret)
224 			goto out;
225 	}
226 
227 	setattr_copy(inode, iattr);
228 	mark_inode_dirty(inode);
229 
230 	ret = orangefs_inode_setattr(inode, iattr);
231 	gossip_debug(GOSSIP_INODE_DEBUG,
232 		     "orangefs_setattr: inode_setattr returned %d\n",
233 		     ret);
234 
235 	if (!ret && (iattr->ia_valid & ATTR_MODE))
236 		/* change mod on a file that has ACLs */
237 		ret = posix_acl_chmod(inode, inode->i_mode);
238 
239 out:
240 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", ret);
241 	return ret;
242 }
243 
244 /*
245  * Obtain attributes of an object given a dentry
246  */
247 int orangefs_getattr(const struct path *path, struct kstat *stat,
248 		     u32 request_mask, unsigned int flags)
249 {
250 	int ret = -ENOENT;
251 	struct inode *inode = path->dentry->d_inode;
252 	struct orangefs_inode_s *orangefs_inode = NULL;
253 
254 	gossip_debug(GOSSIP_INODE_DEBUG,
255 		     "orangefs_getattr: called on %pd\n",
256 		     path->dentry);
257 
258 	ret = orangefs_inode_getattr(inode, 0, 0, request_mask);
259 	if (ret == 0) {
260 		generic_fillattr(inode, stat);
261 
262 		/* override block size reported to stat */
263 		orangefs_inode = ORANGEFS_I(inode);
264 		stat->blksize = orangefs_inode->blksize;
265 
266 		if (request_mask & STATX_SIZE)
267 			stat->result_mask = STATX_BASIC_STATS;
268 		else
269 			stat->result_mask = STATX_BASIC_STATS &
270 			    ~STATX_SIZE;
271 	}
272 	return ret;
273 }
274 
275 int orangefs_permission(struct inode *inode, int mask)
276 {
277 	int ret;
278 
279 	if (mask & MAY_NOT_BLOCK)
280 		return -ECHILD;
281 
282 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
283 
284 	/* Make sure the permission (and other common attrs) are up to date. */
285 	ret = orangefs_inode_getattr(inode, 0, 0, STATX_MODE);
286 	if (ret < 0)
287 		return ret;
288 
289 	return generic_permission(inode, mask);
290 }
291 
292 /* ORANGEDS2 implementation of VFS inode operations for files */
293 const struct inode_operations orangefs_file_inode_operations = {
294 	.get_acl = orangefs_get_acl,
295 	.set_acl = orangefs_set_acl,
296 	.setattr = orangefs_setattr,
297 	.getattr = orangefs_getattr,
298 	.listxattr = orangefs_listxattr,
299 	.permission = orangefs_permission,
300 };
301 
302 static int orangefs_init_iops(struct inode *inode)
303 {
304 	inode->i_mapping->a_ops = &orangefs_address_operations;
305 
306 	switch (inode->i_mode & S_IFMT) {
307 	case S_IFREG:
308 		inode->i_op = &orangefs_file_inode_operations;
309 		inode->i_fop = &orangefs_file_operations;
310 		inode->i_blkbits = PAGE_SHIFT;
311 		break;
312 	case S_IFLNK:
313 		inode->i_op = &orangefs_symlink_inode_operations;
314 		break;
315 	case S_IFDIR:
316 		inode->i_op = &orangefs_dir_inode_operations;
317 		inode->i_fop = &orangefs_dir_operations;
318 		break;
319 	default:
320 		gossip_debug(GOSSIP_INODE_DEBUG,
321 			     "%s: unsupported mode\n",
322 			     __func__);
323 		return -EINVAL;
324 	}
325 
326 	return 0;
327 }
328 
329 /*
330  * Given a ORANGEFS object identifier (fsid, handle), convert it into a ino_t type
331  * that will be used as a hash-index from where the handle will
332  * be searched for in the VFS hash table of inodes.
333  */
334 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
335 {
336 	if (!ref)
337 		return 0;
338 	return orangefs_khandle_to_ino(&(ref->khandle));
339 }
340 
341 /*
342  * Called to set up an inode from iget5_locked.
343  */
344 static int orangefs_set_inode(struct inode *inode, void *data)
345 {
346 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
347 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
348 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
349 	return 0;
350 }
351 
352 /*
353  * Called to determine if handles match.
354  */
355 static int orangefs_test_inode(struct inode *inode, void *data)
356 {
357 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
358 	struct orangefs_inode_s *orangefs_inode = NULL;
359 
360 	orangefs_inode = ORANGEFS_I(inode);
361 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), &(ref->khandle))
362 		&& orangefs_inode->refn.fs_id == ref->fs_id);
363 }
364 
365 /*
366  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
367  * file handle.
368  *
369  * @sb: the file system super block instance.
370  * @ref: The ORANGEFS object for which we are trying to locate an inode structure.
371  */
372 struct inode *orangefs_iget(struct super_block *sb, struct orangefs_object_kref *ref)
373 {
374 	struct inode *inode = NULL;
375 	unsigned long hash;
376 	int error;
377 
378 	hash = orangefs_handle_hash(ref);
379 	inode = iget5_locked(sb, hash, orangefs_test_inode, orangefs_set_inode, ref);
380 	if (!inode || !(inode->i_state & I_NEW))
381 		return inode;
382 
383 	error = orangefs_inode_getattr(inode, 1, 1, STATX_ALL);
384 	if (error) {
385 		iget_failed(inode);
386 		return ERR_PTR(error);
387 	}
388 
389 	inode->i_ino = hash;	/* needed for stat etc */
390 	orangefs_init_iops(inode);
391 	unlock_new_inode(inode);
392 
393 	gossip_debug(GOSSIP_INODE_DEBUG,
394 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
395 		     &ref->khandle,
396 		     ref->fs_id,
397 		     hash,
398 		     inode->i_ino);
399 
400 	return inode;
401 }
402 
403 /*
404  * Allocate an inode for a newly created file and insert it into the inode hash.
405  */
406 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
407 		int mode, dev_t dev, struct orangefs_object_kref *ref)
408 {
409 	unsigned long hash = orangefs_handle_hash(ref);
410 	struct inode *inode;
411 	int error;
412 
413 	gossip_debug(GOSSIP_INODE_DEBUG,
414 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
415 		     __func__,
416 		     sb,
417 		     MAJOR(dev),
418 		     MINOR(dev),
419 		     mode);
420 
421 	inode = new_inode(sb);
422 	if (!inode)
423 		return NULL;
424 
425 	orangefs_set_inode(inode, ref);
426 	inode->i_ino = hash;	/* needed for stat etc */
427 
428 	error = orangefs_inode_getattr(inode, 1, 1, STATX_ALL);
429 	if (error)
430 		goto out_iput;
431 
432 	orangefs_init_iops(inode);
433 
434 	inode->i_mode = mode;
435 	inode->i_uid = current_fsuid();
436 	inode->i_gid = current_fsgid();
437 	inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
438 	inode->i_size = PAGE_SIZE;
439 	inode->i_rdev = dev;
440 
441 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
442 	if (error < 0)
443 		goto out_iput;
444 
445 	gossip_debug(GOSSIP_INODE_DEBUG,
446 		     "Initializing ACL's for inode %pU\n",
447 		     get_khandle_from_ino(inode));
448 	orangefs_init_acl(inode, dir);
449 	return inode;
450 
451 out_iput:
452 	iput(inode);
453 	return ERR_PTR(error);
454 }
455