1 // SPDX-License-Identifier: MIT
3 * VirtualBox Guest Shared Folders support: Regular file inode and file ops.
5 * Copyright (C) 2006-2018 Oracle Corporation
9 #include <linux/page-flags.h>
10 #include <linux/pagemap.h>
11 #include <linux/highmem.h>
12 #include <linux/sizes.h>
15 struct vboxsf_handle {
20 struct list_head head;
23 struct vboxsf_handle *vboxsf_create_sf_handle(struct inode *inode,
24 u64 handle, u32 access_flags)
26 struct vboxsf_inode *sf_i = VBOXSF_I(inode);
27 struct vboxsf_handle *sf_handle;
29 sf_handle = kmalloc(sizeof(*sf_handle), GFP_KERNEL);
31 return ERR_PTR(-ENOMEM);
33 /* the host may have given us different attr then requested */
34 sf_i->force_restat = 1;
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);
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);
49 static int vboxsf_file_open(struct inode *inode, struct file *file)
51 struct vboxsf_sbi *sbi = VBOXSF_SBI(inode->i_sb);
52 struct shfl_createparms params = {};
53 struct vboxsf_handle *sf_handle;
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.
62 * Furthermore, we must set params.handle to SHFL_HANDLE_NIL to
63 * make the shared folders host service use our mode parameter.
65 params.handle = SHFL_HANDLE_NIL;
66 if (file->f_flags & O_CREAT) {
67 params.create_flags |= SHFL_CF_ACT_CREATE_IF_NEW;
69 * We ignore O_EXCL, as the Linux kernel seems to call create
70 * beforehand itself, so O_EXCL should always fail.
72 if (file->f_flags & O_TRUNC)
73 params.create_flags |= SHFL_CF_ACT_OVERWRITE_IF_EXISTS;
75 params.create_flags |= SHFL_CF_ACT_OPEN_IF_EXISTS;
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;
82 switch (file->f_flags & O_ACCMODE) {
84 access_flags |= SHFL_CF_ACCESS_READ;
88 access_flags |= SHFL_CF_ACCESS_WRITE;
92 access_flags |= SHFL_CF_ACCESS_READWRITE;
99 if (file->f_flags & O_APPEND)
100 access_flags |= SHFL_CF_ACCESS_APPEND;
102 params.create_flags |= access_flags;
103 params.info.attr.mode = inode->i_mode;
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;
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);
117 file->private_data = sf_handle;
121 static void vboxsf_handle_release(struct kref *refcount)
123 struct vboxsf_handle *sf_handle =
124 container_of(refcount, struct vboxsf_handle, refcount);
126 vboxsf_close(sf_handle->root, sf_handle->handle);
130 void vboxsf_release_sf_handle(struct inode *inode, struct vboxsf_handle *sf_handle)
132 struct vboxsf_inode *sf_i = VBOXSF_I(inode);
134 mutex_lock(&sf_i->handle_list_mutex);
135 list_del(&sf_handle->head);
136 mutex_unlock(&sf_i->handle_list_mutex);
138 kref_put(&sf_handle->refcount, vboxsf_handle_release);
141 static int vboxsf_file_release(struct inode *inode, struct file *file)
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.
147 filemap_write_and_wait(inode->i_mapping);
149 vboxsf_release_sf_handle(inode, file->private_data);
154 * Write back dirty pages now, because there may not be any suitable
157 static void vboxsf_vma_close(struct vm_area_struct *vma)
159 filemap_write_and_wait(vma->vm_file->f_mapping);
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,
168 static int vboxsf_file_mmap(struct file *file, struct vm_area_struct *vma)
172 err = generic_file_mmap(file, vma);
174 vma->vm_ops = &vboxsf_file_vm_ops;
180 * Note that since we are accessing files on the host's filesystem, files
181 * may always be changed underneath us by the host!
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.
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().
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:
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
204 * Currently we are keeping things KISS and using option 1. this allows
205 * directly using generic_file_read_iter without wrapping it.
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.
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,
220 .splice_read = generic_file_splice_read,
223 const struct inode_operations vboxsf_reg_iops = {
224 .getattr = vboxsf_getattr,
225 .setattr = vboxsf_setattr
228 static int vboxsf_read_folio(struct file *file, struct folio *folio)
230 struct page *page = &folio->page;
231 struct vboxsf_handle *sf_handle = file->private_data;
232 loff_t off = page_offset(page);
233 u32 nread = PAGE_SIZE;
239 err = vboxsf_read(sf_handle->root, sf_handle->handle, off, &nread, buf);
241 memset(&buf[nread], 0, PAGE_SIZE - nread);
242 flush_dcache_page(page);
243 SetPageUptodate(page);
253 static struct vboxsf_handle *vboxsf_get_write_handle(struct vboxsf_inode *sf_i)
255 struct vboxsf_handle *h, *sf_handle = NULL;
257 mutex_lock(&sf_i->handle_list_mutex);
258 list_for_each_entry(h, &sf_i->handle_list, head) {
259 if (h->access_flags == SHFL_CF_ACCESS_WRITE ||
260 h->access_flags == SHFL_CF_ACCESS_READWRITE) {
261 kref_get(&h->refcount);
266 mutex_unlock(&sf_i->handle_list_mutex);
271 static int vboxsf_writepage(struct page *page, struct writeback_control *wbc)
273 struct inode *inode = page->mapping->host;
274 struct vboxsf_inode *sf_i = VBOXSF_I(inode);
275 struct vboxsf_handle *sf_handle;
276 loff_t off = page_offset(page);
277 loff_t size = i_size_read(inode);
278 u32 nwrite = PAGE_SIZE;
282 if (off + PAGE_SIZE > size)
283 nwrite = size & ~PAGE_MASK;
285 sf_handle = vboxsf_get_write_handle(sf_i);
290 err = vboxsf_write(sf_handle->root, sf_handle->handle,
294 kref_put(&sf_handle->refcount, vboxsf_handle_release);
297 ClearPageError(page);
299 sf_i->force_restat = 1;
301 ClearPageUptodate(page);
308 static int vboxsf_write_end(struct file *file, struct address_space *mapping,
309 loff_t pos, unsigned int len, unsigned int copied,
310 struct page *page, void *fsdata)
312 struct inode *inode = mapping->host;
313 struct vboxsf_handle *sf_handle = file->private_data;
314 unsigned int from = pos & ~PAGE_MASK;
319 /* zero the stale part of the page if we did a short copy */
320 if (!PageUptodate(page) && copied < len)
321 zero_user(page, from + copied, len - copied);
324 err = vboxsf_write(sf_handle->root, sf_handle->handle,
325 pos, &nwritten, buf + from);
334 VBOXSF_I(inode)->force_restat = 1;
336 if (!PageUptodate(page) && nwritten == PAGE_SIZE)
337 SetPageUptodate(page);
340 if (pos > inode->i_size)
341 i_size_write(inode, pos);
351 * Note simple_write_begin does not read the page from disk on partial writes
352 * this is ok since vboxsf_write_end only writes the written parts of the
353 * page and it does not call SetPageUptodate for partial writes.
355 const struct address_space_operations vboxsf_reg_aops = {
356 .read_folio = vboxsf_read_folio,
357 .writepage = vboxsf_writepage,
358 .dirty_folio = filemap_dirty_folio,
359 .write_begin = simple_write_begin,
360 .write_end = vboxsf_write_end,
363 static const char *vboxsf_get_link(struct dentry *dentry, struct inode *inode,
364 struct delayed_call *done)
366 struct vboxsf_sbi *sbi = VBOXSF_SBI(inode->i_sb);
367 struct shfl_string *path;
372 return ERR_PTR(-ECHILD);
374 path = vboxsf_path_from_dentry(sbi, dentry);
376 return ERR_CAST(path);
378 link = kzalloc(PATH_MAX, GFP_KERNEL);
381 return ERR_PTR(-ENOMEM);
384 err = vboxsf_readlink(sbi->root, path, PATH_MAX, link);
391 set_delayed_call(done, kfree_link, link);
395 const struct inode_operations vboxsf_lnk_iops = {
396 .get_link = vboxsf_get_link