1 // SPDX-License-Identifier: MIT
2 /*
3 * VirtualBox Guest Shared Folders support: Regular file inode and file ops.
4 *
5 * Copyright (C) 2006-2018 Oracle Corporation
6 */
7
8 #include <linux/mm.h>
9 #include <linux/page-flags.h>
10 #include <linux/pagemap.h>
11 #include <linux/highmem.h>
12 #include <linux/sizes.h>
13 #include "vfsmod.h"
14
15 struct vboxsf_handle {
16 u64 handle;
17 u32 root;
18 u32 access_flags;
19 struct kref refcount;
20 struct list_head head;
21 };
22
23 struct vboxsf_handle *vboxsf_create_sf_handle(struct inode *inode,
24 u64 handle, u32 access_flags)
25 {
26 struct vboxsf_inode *sf_i = VBOXSF_I(inode);
27 struct vboxsf_handle *sf_handle;
28
29 sf_handle = kmalloc(sizeof(*sf_handle), GFP_KERNEL);
30 if (!sf_handle)
31 return ERR_PTR(-ENOMEM);
32
33 /* the host may have given us different attr then requested */
34 sf_i->force_restat = 1;
35
36 /* init our handle struct and add it to the inode's handles list */
37 sf_handle->handle = handle;
38 sf_handle->root = VBOXSF_SBI(inode->i_sb)->root;
39 sf_handle->access_flags = access_flags;
40 kref_init(&sf_handle->refcount);
41
42 mutex_lock(&sf_i->handle_list_mutex);
43 list_add(&sf_handle->head, &sf_i->handle_list);
44 mutex_unlock(&sf_i->handle_list_mutex);
45
46 return sf_handle;
47 }
48
49 static int vboxsf_file_open(struct inode *inode, struct file *file)
50 {
51 struct vboxsf_sbi *sbi = VBOXSF_SBI(inode->i_sb);
52 struct shfl_createparms params = {};
53 struct vboxsf_handle *sf_handle;
54 u32 access_flags = 0;
55 int err;
56
57 /*
58 * We check the value of params.handle afterwards to find out if
59 * the call succeeded or failed, as the API does not seem to cleanly
60 * distinguish error and informational messages.
61 *
62 * Furthermore, we must set params.handle to SHFL_HANDLE_NIL to
63 * make the shared folders host service use our mode parameter.
64 */
65 params.handle = SHFL_HANDLE_NIL;
66 if (file->f_flags & O_CREAT) {
67 params.create_flags |= SHFL_CF_ACT_CREATE_IF_NEW;
68 /*
69 * We ignore O_EXCL, as the Linux kernel seems to call create
70 * beforehand itself, so O_EXCL should always fail.
71 */
72 if (file->f_flags & O_TRUNC)
73 params.create_flags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
74 else
75 params.create_flags |= SHFL_CF_ACT_OPEN_IF_EXISTS;
76 } else {
77 params.create_flags |= SHFL_CF_ACT_FAIL_IF_NEW;
78 if (file->f_flags & O_TRUNC)
79 params.create_flags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
80 }
81
82 switch (file->f_flags & O_ACCMODE) {
83 case O_RDONLY:
84 access_flags |= SHFL_CF_ACCESS_READ;
85 break;
86
87 case O_WRONLY:
88 access_flags |= SHFL_CF_ACCESS_WRITE;
89 break;
90
91 case O_RDWR:
92 access_flags |= SHFL_CF_ACCESS_READWRITE;
93 break;
94
95 default:
96 WARN_ON(1);
97 }
98
99 if (file->f_flags & O_APPEND)
100 access_flags |= SHFL_CF_ACCESS_APPEND;
101
102 params.create_flags |= access_flags;
103 params.info.attr.mode = inode->i_mode;
104
105 err = vboxsf_create_at_dentry(file_dentry(file), ¶ms);
106 if (err == 0 && params.handle == SHFL_HANDLE_NIL)
107 err = (params.result == SHFL_FILE_EXISTS) ? -EEXIST : -ENOENT;
108 if (err)
109 return err;
110
111 sf_handle = vboxsf_create_sf_handle(inode, params.handle, access_flags);
112 if (IS_ERR(sf_handle)) {
113 vboxsf_close(sbi->root, params.handle);
114 return PTR_ERR(sf_handle);
115 }
116
117 file->private_data = sf_handle;
118 return 0;
119 }
120
121 static void vboxsf_handle_release(struct kref *refcount)
122 {
123 struct vboxsf_handle *sf_handle =
124 container_of(refcount, struct vboxsf_handle, refcount);
125
126 vboxsf_close(sf_handle->root, sf_handle->handle);
127 kfree(sf_handle);
128 }
129
130 void vboxsf_release_sf_handle(struct inode *inode, struct vboxsf_handle *sf_handle)
131 {
132 struct vboxsf_inode *sf_i = VBOXSF_I(inode);
133
134 mutex_lock(&sf_i->handle_list_mutex);
135 list_del(&sf_handle->head);
136 mutex_unlock(&sf_i->handle_list_mutex);
137
138 kref_put(&sf_handle->refcount, vboxsf_handle_release);
139 }
140
141 static int vboxsf_file_release(struct inode *inode, struct file *file)
142 {
143 /*
144 * When a file is closed on our (the guest) side, we want any subsequent
145 * accesses done on the host side to see all changes done from our side.
146 */
147 filemap_write_and_wait(inode->i_mapping);
148
149 vboxsf_release_sf_handle(inode, file->private_data);
150 return 0;
151 }
152
153 /*
154 * Write back dirty pages now, because there may not be any suitable
155 * open files later
156 */
157 static void vboxsf_vma_close(struct vm_area_struct *vma)
158 {
159 filemap_write_and_wait(vma->vm_file->f_mapping);
160 }
161
162 static const struct vm_operations_struct vboxsf_file_vm_ops = {
163 .close = vboxsf_vma_close,
164 .fault = filemap_fault,
165 .map_pages = filemap_map_pages,
166 };
167
168 static int vboxsf_file_mmap(struct file *file, struct vm_area_struct *vma)
169 {
170 int err;
171
172 err = generic_file_mmap(file, vma);
173 if (!err)
174 vma->vm_ops = &vboxsf_file_vm_ops;
175
176 return err;
177 }
178
179 /*
180 * Note that since we are accessing files on the host's filesystem, files
181 * may always be changed underneath us by the host!
182 *
183 * The vboxsf API between the guest and the host does not offer any functions
184 * to deal with this. There is no inode-generation to check for changes, no
185 * events / callback on changes and no way to lock files.
186 *
187 * To avoid returning stale data when a file gets *opened* on our (the guest)
188 * side, we do a "stat" on the host side, then compare the mtime with the
189 * last known mtime and invalidate the page-cache if they differ.
190 * This is done from vboxsf_inode_revalidate().
191 *
192 * When reads are done through the read_iter fop, it is possible to do
193 * further cache revalidation then, there are 3 options to deal with this:
194 *
195 * 1) Rely solely on the revalidation done at open time
196 * 2) Do another "stat" and compare mtime again. Unfortunately the vboxsf
197 * host API does not allow stat on handles, so we would need to use
198 * file->f_path.dentry and the stat will then fail if the file was unlinked
199 * or renamed (and there is no thing like NFS' silly-rename). So we get:
200 * 2a) "stat" and compare mtime, on stat failure invalidate the cache
201 * 2b) "stat" and compare mtime, on stat failure do nothing
202 * 3) Simply always call invalidate_inode_pages2_range on the range of the read
203 *
204 * Currently we are keeping things KISS and using option 1. this allows
205 * directly using generic_file_read_iter without wrapping it.
206 *
207 * This means that only data written on the host side before open() on
208 * the guest side is guaranteed to be seen by the guest. If necessary
209 * we may provide other read-cache strategies in the future and make this
210 * configurable through a mount option.
211 */
212 const struct file_operations vboxsf_reg_fops = {
213 .llseek = generic_file_llseek,
214 .read_iter = generic_file_read_iter,
215 .write_iter = generic_file_write_iter,
216 .mmap = vboxsf_file_mmap,
217 .open = vboxsf_file_open,
218 .release = vboxsf_file_release,
219 .fsync = noop_fsync,
220 .splice_read = filemap_splice_read,
221 .setlease = simple_nosetlease,
222 };
223
224 const struct inode_operations vboxsf_reg_iops = {
225 .getattr = vboxsf_getattr,
226 .setattr = vboxsf_setattr
227 };
228
229 static int vboxsf_read_folio(struct file *file, struct folio *folio)
230 {
231 struct vboxsf_handle *sf_handle = file->private_data;
232 loff_t off = folio_pos(folio);
233 u32 nread = PAGE_SIZE;
234 u8 *buf;
235 int err;
236
237 buf = kmap_local_folio(folio, 0);
238
239 err = vboxsf_read(sf_handle->root, sf_handle->handle, off, &nread, buf);
240 buf = folio_zero_tail(folio, nread, buf + nread);
241
242 kunmap_local(buf);
243 folio_end_read(folio, err == 0);
244 return err;
245 }
246
247 static struct vboxsf_handle *vboxsf_get_write_handle(struct vboxsf_inode *sf_i)
248 {
249 struct vboxsf_handle *h, *sf_handle = NULL;
250
251 mutex_lock(&sf_i->handle_list_mutex);
252 list_for_each_entry(h, &sf_i->handle_list, head) {
253 if (h->access_flags == SHFL_CF_ACCESS_WRITE ||
254 h->access_flags == SHFL_CF_ACCESS_READWRITE) {
255 kref_get(&h->refcount);
256 sf_handle = h;
257 break;
258 }
259 }
260 mutex_unlock(&sf_i->handle_list_mutex);
261
262 return sf_handle;
263 }
264
265 static int vboxsf_writepage(struct page *page, struct writeback_control *wbc)
266 {
267 struct inode *inode = page->mapping->host;
268 struct vboxsf_inode *sf_i = VBOXSF_I(inode);
269 struct vboxsf_handle *sf_handle;
270 loff_t off = page_offset(page);
271 loff_t size = i_size_read(inode);
272 u32 nwrite = PAGE_SIZE;
273 u8 *buf;
274 int err;
275
276 if (off + PAGE_SIZE > size)
277 nwrite = size & ~PAGE_MASK;
278
279 sf_handle = vboxsf_get_write_handle(sf_i);
280 if (!sf_handle)
281 return -EBADF;
282
283 buf = kmap(page);
284 err = vboxsf_write(sf_handle->root, sf_handle->handle,
285 off, &nwrite, buf);
286 kunmap(page);
287
288 kref_put(&sf_handle->refcount, vboxsf_handle_release);
289
290 if (err == 0) {
291 /* mtime changed */
292 sf_i->force_restat = 1;
293 } else {
294 ClearPageUptodate(page);
295 }
296
297 unlock_page(page);
298 return err;
299 }
300
301 static int vboxsf_write_end(struct file *file, struct address_space *mapping,
302 loff_t pos, unsigned int len, unsigned int copied,
303 struct page *page, void *fsdata)
304 {
305 struct inode *inode = mapping->host;
306 struct vboxsf_handle *sf_handle = file->private_data;
307 unsigned int from = pos & ~PAGE_MASK;
308 u32 nwritten = len;
309 u8 *buf;
310 int err;
311
312 /* zero the stale part of the page if we did a short copy */
313 if (!PageUptodate(page) && copied < len)
314 zero_user(page, from + copied, len - copied);
315
316 buf = kmap(page);
317 err = vboxsf_write(sf_handle->root, sf_handle->handle,
318 pos, &nwritten, buf + from);
319 kunmap(page);
320
321 if (err) {
322 nwritten = 0;
323 goto out;
324 }
325
326 /* mtime changed */
327 VBOXSF_I(inode)->force_restat = 1;
328
329 if (!PageUptodate(page) && nwritten == PAGE_SIZE)
330 SetPageUptodate(page);
331
332 pos += nwritten;
333 if (pos > inode->i_size)
334 i_size_write(inode, pos);
335
336 out:
337 unlock_page(page);
338 put_page(page);
339
340 return nwritten;
341 }
342
343 /*
344 * Note simple_write_begin does not read the page from disk on partial writes
345 * this is ok since vboxsf_write_end only writes the written parts of the
346 * page and it does not call SetPageUptodate for partial writes.
347 */
348 const struct address_space_operations vboxsf_reg_aops = {
349 .read_folio = vboxsf_read_folio,
350 .writepage = vboxsf_writepage,
351 .dirty_folio = filemap_dirty_folio,
352 .write_begin = simple_write_begin,
353 .write_end = vboxsf_write_end,
354 };
355
356 static const char *vboxsf_get_link(struct dentry *dentry, struct inode *inode,
357 struct delayed_call *done)
358 {
359 struct vboxsf_sbi *sbi = VBOXSF_SBI(inode->i_sb);
360 struct shfl_string *path;
361 char *link;
362 int err;
363
364 if (!dentry)
365 return ERR_PTR(-ECHILD);
366
367 path = vboxsf_path_from_dentry(sbi, dentry);
368 if (IS_ERR(path))
369 return ERR_CAST(path);
370
371 link = kzalloc(PATH_MAX, GFP_KERNEL);
372 if (!link) {
373 __putname(path);
374 return ERR_PTR(-ENOMEM);
375 }
376
377 err = vboxsf_readlink(sbi->root, path, PATH_MAX, link);
378 __putname(path);
379 if (err) {
380 kfree(link);
381 return ERR_PTR(err);
382 }
383
384 set_delayed_call(done, kfree_link, link);
385 return link;
386 }
387
388 const struct inode_operations vboxsf_lnk_iops = {
389 .get_link = vboxsf_get_link
390 };
391
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.